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
.nonlinear
);
204 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
205 flush_dcache_mmap_unlock(mapping
);
209 * Unlink a file-based vm structure from its interval 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)
357 * vma has some anon_vma assigned, and is already inserted on that
358 * anon_vma's interval trees.
360 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
361 * vma must be removed from the anon_vma's interval trees using
362 * anon_vma_interval_tree_pre_update_vma().
364 * After the update, the vma will be reinserted using
365 * anon_vma_interval_tree_post_update_vma().
367 * The entire update must be protected by exclusive mmap_sem and by
368 * the root anon_vma's mutex.
371 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
373 struct anon_vma_chain
*avc
;
375 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
376 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
380 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
382 struct anon_vma_chain
*avc
;
384 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
385 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
388 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
389 unsigned long end
, struct vm_area_struct
**pprev
,
390 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
392 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
394 __rb_link
= &mm
->mm_rb
.rb_node
;
395 rb_prev
= __rb_parent
= NULL
;
398 struct vm_area_struct
*vma_tmp
;
400 __rb_parent
= *__rb_link
;
401 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
403 if (vma_tmp
->vm_end
> addr
) {
404 /* Fail if an existing vma overlaps the area */
405 if (vma_tmp
->vm_start
< end
)
407 __rb_link
= &__rb_parent
->rb_left
;
409 rb_prev
= __rb_parent
;
410 __rb_link
= &__rb_parent
->rb_right
;
416 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
417 *rb_link
= __rb_link
;
418 *rb_parent
= __rb_parent
;
422 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
423 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
425 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
426 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
429 static void __vma_link_file(struct vm_area_struct
*vma
)
435 struct address_space
*mapping
= file
->f_mapping
;
437 if (vma
->vm_flags
& VM_DENYWRITE
)
438 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
439 if (vma
->vm_flags
& VM_SHARED
)
440 mapping
->i_mmap_writable
++;
442 flush_dcache_mmap_lock(mapping
);
443 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
444 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
446 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
447 flush_dcache_mmap_unlock(mapping
);
452 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
453 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
454 struct rb_node
*rb_parent
)
456 __vma_link_list(mm
, vma
, prev
, rb_parent
);
457 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
460 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
461 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
462 struct rb_node
*rb_parent
)
464 struct address_space
*mapping
= NULL
;
467 mapping
= vma
->vm_file
->f_mapping
;
470 mutex_lock(&mapping
->i_mmap_mutex
);
472 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
473 __vma_link_file(vma
);
476 mutex_unlock(&mapping
->i_mmap_mutex
);
483 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
484 * mm's list and rbtree. It has already been inserted into the interval tree.
486 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
488 struct vm_area_struct
*prev
;
489 struct rb_node
**rb_link
, *rb_parent
;
491 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
492 &prev
, &rb_link
, &rb_parent
))
494 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
499 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
500 struct vm_area_struct
*prev
)
502 struct vm_area_struct
*next
= vma
->vm_next
;
504 prev
->vm_next
= next
;
506 next
->vm_prev
= prev
;
507 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
508 if (mm
->mmap_cache
== vma
)
509 mm
->mmap_cache
= prev
;
513 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
514 * is already present in an i_mmap tree without adjusting the tree.
515 * The following helper function should be used when such adjustments
516 * are necessary. The "insert" vma (if any) is to be inserted
517 * before we drop the necessary locks.
519 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
520 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
522 struct mm_struct
*mm
= vma
->vm_mm
;
523 struct vm_area_struct
*next
= vma
->vm_next
;
524 struct vm_area_struct
*importer
= NULL
;
525 struct address_space
*mapping
= NULL
;
526 struct rb_root
*root
= NULL
;
527 struct anon_vma
*anon_vma
= NULL
;
528 struct file
*file
= vma
->vm_file
;
529 long adjust_next
= 0;
532 if (next
&& !insert
) {
533 struct vm_area_struct
*exporter
= NULL
;
535 if (end
>= next
->vm_end
) {
537 * vma expands, overlapping all the next, and
538 * perhaps the one after too (mprotect case 6).
540 again
: remove_next
= 1 + (end
> next
->vm_end
);
544 } else if (end
> next
->vm_start
) {
546 * vma expands, overlapping part of the next:
547 * mprotect case 5 shifting the boundary up.
549 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
552 } else if (end
< vma
->vm_end
) {
554 * vma shrinks, and !insert tells it's not
555 * split_vma inserting another: so it must be
556 * mprotect case 4 shifting the boundary down.
558 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
564 * Easily overlooked: when mprotect shifts the boundary,
565 * make sure the expanding vma has anon_vma set if the
566 * shrinking vma had, to cover any anon pages imported.
568 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
569 if (anon_vma_clone(importer
, exporter
))
571 importer
->anon_vma
= exporter
->anon_vma
;
576 mapping
= file
->f_mapping
;
577 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
578 root
= &mapping
->i_mmap
;
579 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
582 uprobe_munmap(next
, next
->vm_start
,
586 mutex_lock(&mapping
->i_mmap_mutex
);
589 * Put into interval tree now, so instantiated pages
590 * are visible to arm/parisc __flush_dcache_page
591 * throughout; but we cannot insert into address
592 * space until vma start or end is updated.
594 __vma_link_file(insert
);
598 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
600 anon_vma
= vma
->anon_vma
;
601 if (!anon_vma
&& adjust_next
)
602 anon_vma
= next
->anon_vma
;
604 VM_BUG_ON(adjust_next
&& next
->anon_vma
&&
605 anon_vma
!= next
->anon_vma
);
606 anon_vma_lock(anon_vma
);
607 anon_vma_interval_tree_pre_update_vma(vma
);
609 anon_vma_interval_tree_pre_update_vma(next
);
613 flush_dcache_mmap_lock(mapping
);
614 vma_interval_tree_remove(vma
, root
);
616 vma_interval_tree_remove(next
, root
);
619 vma
->vm_start
= start
;
621 vma
->vm_pgoff
= pgoff
;
623 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
624 next
->vm_pgoff
+= adjust_next
;
629 vma_interval_tree_insert(next
, root
);
630 vma_interval_tree_insert(vma
, root
);
631 flush_dcache_mmap_unlock(mapping
);
636 * vma_merge has merged next into vma, and needs
637 * us to remove next before dropping the locks.
639 __vma_unlink(mm
, next
, vma
);
641 __remove_shared_vm_struct(next
, file
, mapping
);
644 * split_vma has split insert from vma, and needs
645 * us to insert it before dropping the locks
646 * (it may either follow vma or precede it).
648 __insert_vm_struct(mm
, insert
);
652 anon_vma_interval_tree_post_update_vma(vma
);
654 anon_vma_interval_tree_post_update_vma(next
);
655 anon_vma_unlock(anon_vma
);
658 mutex_unlock(&mapping
->i_mmap_mutex
);
669 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
673 anon_vma_merge(vma
, next
);
675 mpol_put(vma_policy(next
));
676 kmem_cache_free(vm_area_cachep
, next
);
678 * In mprotect's case 6 (see comments on vma_merge),
679 * we must remove another next too. It would clutter
680 * up the code too much to do both in one go.
682 if (remove_next
== 2) {
696 * If the vma has a ->close operation then the driver probably needs to release
697 * per-vma resources, so we don't attempt to merge those.
699 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
700 struct file
*file
, unsigned long vm_flags
)
702 if (vma
->vm_flags
^ vm_flags
)
704 if (vma
->vm_file
!= file
)
706 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
711 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
712 struct anon_vma
*anon_vma2
,
713 struct vm_area_struct
*vma
)
716 * The list_is_singular() test is to avoid merging VMA cloned from
717 * parents. This can improve scalability caused by anon_vma lock.
719 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
720 list_is_singular(&vma
->anon_vma_chain
)))
722 return anon_vma1
== anon_vma2
;
726 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
727 * in front of (at a lower virtual address and file offset than) the vma.
729 * We cannot merge two vmas if they have differently assigned (non-NULL)
730 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
732 * We don't check here for the merged mmap wrapping around the end of pagecache
733 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
734 * wrap, nor mmaps which cover the final page at index -1UL.
737 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
738 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
740 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
741 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
742 if (vma
->vm_pgoff
== vm_pgoff
)
749 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
750 * beyond (at a higher virtual address and file offset than) the vma.
752 * We cannot merge two vmas if they have differently assigned (non-NULL)
753 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
756 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
757 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
759 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
760 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
762 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
763 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
770 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
771 * whether that can be merged with its predecessor or its successor.
772 * Or both (it neatly fills a hole).
774 * In most cases - when called for mmap, brk or mremap - [addr,end) is
775 * certain not to be mapped by the time vma_merge is called; but when
776 * called for mprotect, it is certain to be already mapped (either at
777 * an offset within prev, or at the start of next), and the flags of
778 * this area are about to be changed to vm_flags - and the no-change
779 * case has already been eliminated.
781 * The following mprotect cases have to be considered, where AAAA is
782 * the area passed down from mprotect_fixup, never extending beyond one
783 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
785 * AAAA AAAA AAAA AAAA
786 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
787 * cannot merge might become might become might become
788 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
789 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
790 * mremap move: PPPPNNNNNNNN 8
792 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
793 * might become case 1 below case 2 below case 3 below
795 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
796 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
798 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
799 struct vm_area_struct
*prev
, unsigned long addr
,
800 unsigned long end
, unsigned long vm_flags
,
801 struct anon_vma
*anon_vma
, struct file
*file
,
802 pgoff_t pgoff
, struct mempolicy
*policy
)
804 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
805 struct vm_area_struct
*area
, *next
;
809 * We later require that vma->vm_flags == vm_flags,
810 * so this tests vma->vm_flags & VM_SPECIAL, too.
812 if (vm_flags
& VM_SPECIAL
)
816 next
= prev
->vm_next
;
820 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
821 next
= next
->vm_next
;
824 * Can it merge with the predecessor?
826 if (prev
&& prev
->vm_end
== addr
&&
827 mpol_equal(vma_policy(prev
), policy
) &&
828 can_vma_merge_after(prev
, vm_flags
,
829 anon_vma
, file
, pgoff
)) {
831 * OK, it can. Can we now merge in the successor as well?
833 if (next
&& end
== next
->vm_start
&&
834 mpol_equal(policy
, vma_policy(next
)) &&
835 can_vma_merge_before(next
, vm_flags
,
836 anon_vma
, file
, pgoff
+pglen
) &&
837 is_mergeable_anon_vma(prev
->anon_vma
,
838 next
->anon_vma
, NULL
)) {
840 err
= vma_adjust(prev
, prev
->vm_start
,
841 next
->vm_end
, prev
->vm_pgoff
, NULL
);
842 } else /* cases 2, 5, 7 */
843 err
= vma_adjust(prev
, prev
->vm_start
,
844 end
, prev
->vm_pgoff
, NULL
);
847 khugepaged_enter_vma_merge(prev
);
852 * Can this new request be merged in front of next?
854 if (next
&& end
== next
->vm_start
&&
855 mpol_equal(policy
, vma_policy(next
)) &&
856 can_vma_merge_before(next
, vm_flags
,
857 anon_vma
, file
, pgoff
+pglen
)) {
858 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
859 err
= vma_adjust(prev
, prev
->vm_start
,
860 addr
, prev
->vm_pgoff
, NULL
);
861 else /* cases 3, 8 */
862 err
= vma_adjust(area
, addr
, next
->vm_end
,
863 next
->vm_pgoff
- pglen
, NULL
);
866 khugepaged_enter_vma_merge(area
);
874 * Rough compatbility check to quickly see if it's even worth looking
875 * at sharing an anon_vma.
877 * They need to have the same vm_file, and the flags can only differ
878 * in things that mprotect may change.
880 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
881 * we can merge the two vma's. For example, we refuse to merge a vma if
882 * there is a vm_ops->close() function, because that indicates that the
883 * driver is doing some kind of reference counting. But that doesn't
884 * really matter for the anon_vma sharing case.
886 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
888 return a
->vm_end
== b
->vm_start
&&
889 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
890 a
->vm_file
== b
->vm_file
&&
891 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
892 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
896 * Do some basic sanity checking to see if we can re-use the anon_vma
897 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
898 * the same as 'old', the other will be the new one that is trying
899 * to share the anon_vma.
901 * NOTE! This runs with mm_sem held for reading, so it is possible that
902 * the anon_vma of 'old' is concurrently in the process of being set up
903 * by another page fault trying to merge _that_. But that's ok: if it
904 * is being set up, that automatically means that it will be a singleton
905 * acceptable for merging, so we can do all of this optimistically. But
906 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
908 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
909 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
910 * is to return an anon_vma that is "complex" due to having gone through
913 * We also make sure that the two vma's are compatible (adjacent,
914 * and with the same memory policies). That's all stable, even with just
915 * a read lock on the mm_sem.
917 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
919 if (anon_vma_compatible(a
, b
)) {
920 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
922 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
929 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
930 * neighbouring vmas for a suitable anon_vma, before it goes off
931 * to allocate a new anon_vma. It checks because a repetitive
932 * sequence of mprotects and faults may otherwise lead to distinct
933 * anon_vmas being allocated, preventing vma merge in subsequent
936 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
938 struct anon_vma
*anon_vma
;
939 struct vm_area_struct
*near
;
945 anon_vma
= reusable_anon_vma(near
, vma
, near
);
953 anon_vma
= reusable_anon_vma(near
, near
, vma
);
958 * There's no absolute need to look only at touching neighbours:
959 * we could search further afield for "compatible" anon_vmas.
960 * But it would probably just be a waste of time searching,
961 * or lead to too many vmas hanging off the same anon_vma.
962 * We're trying to allow mprotect remerging later on,
963 * not trying to minimize memory used for anon_vmas.
968 #ifdef CONFIG_PROC_FS
969 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
970 struct file
*file
, long pages
)
972 const unsigned long stack_flags
973 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
975 mm
->total_vm
+= pages
;
978 mm
->shared_vm
+= pages
;
979 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
980 mm
->exec_vm
+= pages
;
981 } else if (flags
& stack_flags
)
982 mm
->stack_vm
+= pages
;
984 #endif /* CONFIG_PROC_FS */
987 * If a hint addr is less than mmap_min_addr change hint to be as
988 * low as possible but still greater than mmap_min_addr
990 static inline unsigned long round_hint_to_min(unsigned long hint
)
993 if (((void *)hint
!= NULL
) &&
994 (hint
< mmap_min_addr
))
995 return PAGE_ALIGN(mmap_min_addr
);
1000 * The caller must hold down_write(¤t->mm->mmap_sem).
1003 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
1004 unsigned long len
, unsigned long prot
,
1005 unsigned long flags
, unsigned long pgoff
)
1007 struct mm_struct
* mm
= current
->mm
;
1008 struct inode
*inode
;
1009 vm_flags_t vm_flags
;
1012 * Does the application expect PROT_READ to imply PROT_EXEC?
1014 * (the exception is when the underlying filesystem is noexec
1015 * mounted, in which case we dont add PROT_EXEC.)
1017 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1018 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
1024 if (!(flags
& MAP_FIXED
))
1025 addr
= round_hint_to_min(addr
);
1027 /* Careful about overflows.. */
1028 len
= PAGE_ALIGN(len
);
1032 /* offset overflow? */
1033 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1036 /* Too many mappings? */
1037 if (mm
->map_count
> sysctl_max_map_count
)
1040 /* Obtain the address to map to. we verify (or select) it and ensure
1041 * that it represents a valid section of the address space.
1043 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1044 if (addr
& ~PAGE_MASK
)
1047 /* Do simple checking here so the lower-level routines won't have
1048 * to. we assume access permissions have been handled by the open
1049 * of the memory object, so we don't do any here.
1051 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1052 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1054 if (flags
& MAP_LOCKED
)
1055 if (!can_do_mlock())
1058 /* mlock MCL_FUTURE? */
1059 if (vm_flags
& VM_LOCKED
) {
1060 unsigned long locked
, lock_limit
;
1061 locked
= len
>> PAGE_SHIFT
;
1062 locked
+= mm
->locked_vm
;
1063 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1064 lock_limit
>>= PAGE_SHIFT
;
1065 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1069 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1072 switch (flags
& MAP_TYPE
) {
1074 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1078 * Make sure we don't allow writing to an append-only
1081 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1085 * Make sure there are no mandatory locks on the file.
1087 if (locks_verify_locked(inode
))
1090 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1091 if (!(file
->f_mode
& FMODE_WRITE
))
1092 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1096 if (!(file
->f_mode
& FMODE_READ
))
1098 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1099 if (vm_flags
& VM_EXEC
)
1101 vm_flags
&= ~VM_MAYEXEC
;
1104 if (!file
->f_op
|| !file
->f_op
->mmap
)
1112 switch (flags
& MAP_TYPE
) {
1118 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1122 * Set pgoff according to addr for anon_vma.
1124 pgoff
= addr
>> PAGE_SHIFT
;
1131 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1134 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1135 unsigned long, prot
, unsigned long, flags
,
1136 unsigned long, fd
, unsigned long, pgoff
)
1138 struct file
*file
= NULL
;
1139 unsigned long retval
= -EBADF
;
1141 if (!(flags
& MAP_ANONYMOUS
)) {
1142 audit_mmap_fd(fd
, flags
);
1143 if (unlikely(flags
& MAP_HUGETLB
))
1148 } else if (flags
& MAP_HUGETLB
) {
1149 struct user_struct
*user
= NULL
;
1151 * VM_NORESERVE is used because the reservations will be
1152 * taken when vm_ops->mmap() is called
1153 * A dummy user value is used because we are not locking
1154 * memory so no accounting is necessary
1156 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, addr
, len
,
1157 VM_NORESERVE
, &user
,
1158 HUGETLB_ANONHUGE_INODE
);
1160 return PTR_ERR(file
);
1163 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1165 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1172 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1173 struct mmap_arg_struct
{
1177 unsigned long flags
;
1179 unsigned long offset
;
1182 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1184 struct mmap_arg_struct a
;
1186 if (copy_from_user(&a
, arg
, sizeof(a
)))
1188 if (a
.offset
& ~PAGE_MASK
)
1191 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1192 a
.offset
>> PAGE_SHIFT
);
1194 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1197 * Some shared mappigns will want the pages marked read-only
1198 * to track write events. If so, we'll downgrade vm_page_prot
1199 * to the private version (using protection_map[] without the
1202 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1204 vm_flags_t vm_flags
= vma
->vm_flags
;
1206 /* If it was private or non-writable, the write bit is already clear */
1207 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1210 /* The backer wishes to know when pages are first written to? */
1211 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1214 /* The open routine did something to the protections already? */
1215 if (pgprot_val(vma
->vm_page_prot
) !=
1216 pgprot_val(vm_get_page_prot(vm_flags
)))
1219 /* Specialty mapping? */
1220 if (vm_flags
& VM_PFNMAP
)
1223 /* Can the mapping track the dirty pages? */
1224 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1225 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1229 * We account for memory if it's a private writeable mapping,
1230 * not hugepages and VM_NORESERVE wasn't set.
1232 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1235 * hugetlb has its own accounting separate from the core VM
1236 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1238 if (file
&& is_file_hugepages(file
))
1241 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1244 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1245 unsigned long len
, unsigned long flags
,
1246 vm_flags_t vm_flags
, unsigned long pgoff
)
1248 struct mm_struct
*mm
= current
->mm
;
1249 struct vm_area_struct
*vma
, *prev
;
1250 int correct_wcount
= 0;
1252 struct rb_node
**rb_link
, *rb_parent
;
1253 unsigned long charged
= 0;
1254 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1256 /* Clear old maps */
1259 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
1260 if (do_munmap(mm
, addr
, len
))
1265 /* Check against address space limit. */
1266 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1270 * Set 'VM_NORESERVE' if we should not account for the
1271 * memory use of this mapping.
1273 if ((flags
& MAP_NORESERVE
)) {
1274 /* We honor MAP_NORESERVE if allowed to overcommit */
1275 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1276 vm_flags
|= VM_NORESERVE
;
1278 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1279 if (file
&& is_file_hugepages(file
))
1280 vm_flags
|= VM_NORESERVE
;
1284 * Private writable mapping: check memory availability
1286 if (accountable_mapping(file
, vm_flags
)) {
1287 charged
= len
>> PAGE_SHIFT
;
1288 if (security_vm_enough_memory_mm(mm
, charged
))
1290 vm_flags
|= VM_ACCOUNT
;
1294 * Can we just expand an old mapping?
1296 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1301 * Determine the object being mapped and call the appropriate
1302 * specific mapper. the address has already been validated, but
1303 * not unmapped, but the maps are removed from the list.
1305 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1312 vma
->vm_start
= addr
;
1313 vma
->vm_end
= addr
+ len
;
1314 vma
->vm_flags
= vm_flags
;
1315 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1316 vma
->vm_pgoff
= pgoff
;
1317 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1319 error
= -EINVAL
; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1322 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1324 if (vm_flags
& VM_DENYWRITE
) {
1325 error
= deny_write_access(file
);
1330 vma
->vm_file
= get_file(file
);
1331 error
= file
->f_op
->mmap(file
, vma
);
1333 goto unmap_and_free_vma
;
1335 /* Can addr have changed??
1337 * Answer: Yes, several device drivers can do it in their
1338 * f_op->mmap method. -DaveM
1340 addr
= vma
->vm_start
;
1341 pgoff
= vma
->vm_pgoff
;
1342 vm_flags
= vma
->vm_flags
;
1343 } else if (vm_flags
& VM_SHARED
) {
1344 if (unlikely(vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
)))
1346 error
= shmem_zero_setup(vma
);
1351 if (vma_wants_writenotify(vma
)) {
1352 pgprot_t pprot
= vma
->vm_page_prot
;
1354 /* Can vma->vm_page_prot have changed??
1356 * Answer: Yes, drivers may have changed it in their
1357 * f_op->mmap method.
1359 * Ensures that vmas marked as uncached stay that way.
1361 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1362 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1363 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1366 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1367 file
= vma
->vm_file
;
1369 /* Once vma denies write, undo our temporary denial count */
1371 atomic_inc(&inode
->i_writecount
);
1373 perf_event_mmap(vma
);
1375 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1376 if (vm_flags
& VM_LOCKED
) {
1377 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1378 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1379 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1380 make_pages_present(addr
, addr
+ len
);
1389 atomic_inc(&inode
->i_writecount
);
1390 vma
->vm_file
= NULL
;
1393 /* Undo any partial mapping done by a device driver. */
1394 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1397 kmem_cache_free(vm_area_cachep
, vma
);
1400 vm_unacct_memory(charged
);
1404 /* Get an address range which is currently unmapped.
1405 * For shmat() with addr=0.
1407 * Ugly calling convention alert:
1408 * Return value with the low bits set means error value,
1410 * if (ret & ~PAGE_MASK)
1413 * This function "knows" that -ENOMEM has the bits set.
1415 #ifndef HAVE_ARCH_UNMAPPED_AREA
1417 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1418 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1420 struct mm_struct
*mm
= current
->mm
;
1421 struct vm_area_struct
*vma
;
1422 unsigned long start_addr
;
1424 if (len
> TASK_SIZE
)
1427 if (flags
& MAP_FIXED
)
1431 addr
= PAGE_ALIGN(addr
);
1432 vma
= find_vma(mm
, addr
);
1433 if (TASK_SIZE
- len
>= addr
&&
1434 (!vma
|| addr
+ len
<= vma
->vm_start
))
1437 if (len
> mm
->cached_hole_size
) {
1438 start_addr
= addr
= mm
->free_area_cache
;
1440 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1441 mm
->cached_hole_size
= 0;
1445 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1446 /* At this point: (!vma || addr < vma->vm_end). */
1447 if (TASK_SIZE
- len
< addr
) {
1449 * Start a new search - just in case we missed
1452 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1453 addr
= TASK_UNMAPPED_BASE
;
1455 mm
->cached_hole_size
= 0;
1460 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1462 * Remember the place where we stopped the search:
1464 mm
->free_area_cache
= addr
+ len
;
1467 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1468 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1474 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1477 * Is this a new hole at the lowest possible address?
1479 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1480 mm
->free_area_cache
= addr
;
1484 * This mmap-allocator allocates new areas top-down from below the
1485 * stack's low limit (the base):
1487 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1489 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1490 const unsigned long len
, const unsigned long pgoff
,
1491 const unsigned long flags
)
1493 struct vm_area_struct
*vma
;
1494 struct mm_struct
*mm
= current
->mm
;
1495 unsigned long addr
= addr0
, start_addr
;
1497 /* requested length too big for entire address space */
1498 if (len
> TASK_SIZE
)
1501 if (flags
& MAP_FIXED
)
1504 /* requesting a specific address */
1506 addr
= PAGE_ALIGN(addr
);
1507 vma
= find_vma(mm
, addr
);
1508 if (TASK_SIZE
- len
>= addr
&&
1509 (!vma
|| addr
+ len
<= vma
->vm_start
))
1513 /* check if free_area_cache is useful for us */
1514 if (len
<= mm
->cached_hole_size
) {
1515 mm
->cached_hole_size
= 0;
1516 mm
->free_area_cache
= mm
->mmap_base
;
1520 /* either no address requested or can't fit in requested address hole */
1521 start_addr
= addr
= mm
->free_area_cache
;
1529 * Lookup failure means no vma is above this address,
1530 * else if new region fits below vma->vm_start,
1531 * return with success:
1533 vma
= find_vma(mm
, addr
);
1534 if (!vma
|| addr
+len
<= vma
->vm_start
)
1535 /* remember the address as a hint for next time */
1536 return (mm
->free_area_cache
= addr
);
1538 /* remember the largest hole we saw so far */
1539 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1540 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1542 /* try just below the current vma->vm_start */
1543 addr
= vma
->vm_start
-len
;
1544 } while (len
< vma
->vm_start
);
1548 * if hint left us with no space for the requested
1549 * mapping then try again:
1551 * Note: this is different with the case of bottomup
1552 * which does the fully line-search, but we use find_vma
1553 * here that causes some holes skipped.
1555 if (start_addr
!= mm
->mmap_base
) {
1556 mm
->free_area_cache
= mm
->mmap_base
;
1557 mm
->cached_hole_size
= 0;
1562 * A failed mmap() very likely causes application failure,
1563 * so fall back to the bottom-up function here. This scenario
1564 * can happen with large stack limits and large mmap()
1567 mm
->cached_hole_size
= ~0UL;
1568 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1569 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1571 * Restore the topdown base:
1573 mm
->free_area_cache
= mm
->mmap_base
;
1574 mm
->cached_hole_size
= ~0UL;
1580 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1583 * Is this a new hole at the highest possible address?
1585 if (addr
> mm
->free_area_cache
)
1586 mm
->free_area_cache
= addr
;
1588 /* dont allow allocations above current base */
1589 if (mm
->free_area_cache
> mm
->mmap_base
)
1590 mm
->free_area_cache
= mm
->mmap_base
;
1594 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1595 unsigned long pgoff
, unsigned long flags
)
1597 unsigned long (*get_area
)(struct file
*, unsigned long,
1598 unsigned long, unsigned long, unsigned long);
1600 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1604 /* Careful about overflows.. */
1605 if (len
> TASK_SIZE
)
1608 get_area
= current
->mm
->get_unmapped_area
;
1609 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1610 get_area
= file
->f_op
->get_unmapped_area
;
1611 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1612 if (IS_ERR_VALUE(addr
))
1615 if (addr
> TASK_SIZE
- len
)
1617 if (addr
& ~PAGE_MASK
)
1620 addr
= arch_rebalance_pgtables(addr
, len
);
1621 error
= security_mmap_addr(addr
);
1622 return error
? error
: addr
;
1625 EXPORT_SYMBOL(get_unmapped_area
);
1627 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1628 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1630 struct vm_area_struct
*vma
= NULL
;
1632 if (WARN_ON_ONCE(!mm
)) /* Remove this in linux-3.6 */
1635 /* Check the cache first. */
1636 /* (Cache hit rate is typically around 35%.) */
1637 vma
= mm
->mmap_cache
;
1638 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1639 struct rb_node
*rb_node
;
1641 rb_node
= mm
->mm_rb
.rb_node
;
1645 struct vm_area_struct
*vma_tmp
;
1647 vma_tmp
= rb_entry(rb_node
,
1648 struct vm_area_struct
, vm_rb
);
1650 if (vma_tmp
->vm_end
> addr
) {
1652 if (vma_tmp
->vm_start
<= addr
)
1654 rb_node
= rb_node
->rb_left
;
1656 rb_node
= rb_node
->rb_right
;
1659 mm
->mmap_cache
= vma
;
1664 EXPORT_SYMBOL(find_vma
);
1667 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1669 struct vm_area_struct
*
1670 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1671 struct vm_area_struct
**pprev
)
1673 struct vm_area_struct
*vma
;
1675 vma
= find_vma(mm
, addr
);
1677 *pprev
= vma
->vm_prev
;
1679 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
1682 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1683 rb_node
= rb_node
->rb_right
;
1690 * Verify that the stack growth is acceptable and
1691 * update accounting. This is shared with both the
1692 * grow-up and grow-down cases.
1694 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1696 struct mm_struct
*mm
= vma
->vm_mm
;
1697 struct rlimit
*rlim
= current
->signal
->rlim
;
1698 unsigned long new_start
;
1700 /* address space limit tests */
1701 if (!may_expand_vm(mm
, grow
))
1704 /* Stack limit test */
1705 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1708 /* mlock limit tests */
1709 if (vma
->vm_flags
& VM_LOCKED
) {
1710 unsigned long locked
;
1711 unsigned long limit
;
1712 locked
= mm
->locked_vm
+ grow
;
1713 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1714 limit
>>= PAGE_SHIFT
;
1715 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1719 /* Check to ensure the stack will not grow into a hugetlb-only region */
1720 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1722 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1726 * Overcommit.. This must be the final test, as it will
1727 * update security statistics.
1729 if (security_vm_enough_memory_mm(mm
, grow
))
1732 /* Ok, everything looks good - let it rip */
1733 if (vma
->vm_flags
& VM_LOCKED
)
1734 mm
->locked_vm
+= grow
;
1735 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1739 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1741 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1742 * vma is the last one with address > vma->vm_end. Have to extend vma.
1744 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1748 if (!(vma
->vm_flags
& VM_GROWSUP
))
1752 * We must make sure the anon_vma is allocated
1753 * so that the anon_vma locking is not a noop.
1755 if (unlikely(anon_vma_prepare(vma
)))
1757 vma_lock_anon_vma(vma
);
1760 * vma->vm_start/vm_end cannot change under us because the caller
1761 * is required to hold the mmap_sem in read mode. We need the
1762 * anon_vma lock to serialize against concurrent expand_stacks.
1763 * Also guard against wrapping around to address 0.
1765 if (address
< PAGE_ALIGN(address
+4))
1766 address
= PAGE_ALIGN(address
+4);
1768 vma_unlock_anon_vma(vma
);
1773 /* Somebody else might have raced and expanded it already */
1774 if (address
> vma
->vm_end
) {
1775 unsigned long size
, grow
;
1777 size
= address
- vma
->vm_start
;
1778 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1781 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1782 error
= acct_stack_growth(vma
, size
, grow
);
1784 anon_vma_interval_tree_pre_update_vma(vma
);
1785 vma
->vm_end
= address
;
1786 anon_vma_interval_tree_post_update_vma(vma
);
1787 perf_event_mmap(vma
);
1791 vma_unlock_anon_vma(vma
);
1792 khugepaged_enter_vma_merge(vma
);
1795 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1798 * vma is the first one with address < vma->vm_start. Have to extend vma.
1800 int expand_downwards(struct vm_area_struct
*vma
,
1801 unsigned long address
)
1806 * We must make sure the anon_vma is allocated
1807 * so that the anon_vma locking is not a noop.
1809 if (unlikely(anon_vma_prepare(vma
)))
1812 address
&= PAGE_MASK
;
1813 error
= security_mmap_addr(address
);
1817 vma_lock_anon_vma(vma
);
1820 * vma->vm_start/vm_end cannot change under us because the caller
1821 * is required to hold the mmap_sem in read mode. We need the
1822 * anon_vma lock to serialize against concurrent expand_stacks.
1825 /* Somebody else might have raced and expanded it already */
1826 if (address
< vma
->vm_start
) {
1827 unsigned long size
, grow
;
1829 size
= vma
->vm_end
- address
;
1830 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1833 if (grow
<= vma
->vm_pgoff
) {
1834 error
= acct_stack_growth(vma
, size
, grow
);
1836 anon_vma_interval_tree_pre_update_vma(vma
);
1837 vma
->vm_start
= address
;
1838 vma
->vm_pgoff
-= grow
;
1839 anon_vma_interval_tree_post_update_vma(vma
);
1840 perf_event_mmap(vma
);
1844 vma_unlock_anon_vma(vma
);
1845 khugepaged_enter_vma_merge(vma
);
1849 #ifdef CONFIG_STACK_GROWSUP
1850 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1852 return expand_upwards(vma
, address
);
1855 struct vm_area_struct
*
1856 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1858 struct vm_area_struct
*vma
, *prev
;
1861 vma
= find_vma_prev(mm
, addr
, &prev
);
1862 if (vma
&& (vma
->vm_start
<= addr
))
1864 if (!prev
|| expand_stack(prev
, addr
))
1866 if (prev
->vm_flags
& VM_LOCKED
) {
1867 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1872 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1874 return expand_downwards(vma
, address
);
1877 struct vm_area_struct
*
1878 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1880 struct vm_area_struct
* vma
;
1881 unsigned long start
;
1884 vma
= find_vma(mm
,addr
);
1887 if (vma
->vm_start
<= addr
)
1889 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1891 start
= vma
->vm_start
;
1892 if (expand_stack(vma
, addr
))
1894 if (vma
->vm_flags
& VM_LOCKED
) {
1895 mlock_vma_pages_range(vma
, addr
, start
);
1902 * Ok - we have the memory areas we should free on the vma list,
1903 * so release them, and do the vma updates.
1905 * Called with the mm semaphore held.
1907 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1909 unsigned long nr_accounted
= 0;
1911 /* Update high watermark before we lower total_vm */
1912 update_hiwater_vm(mm
);
1914 long nrpages
= vma_pages(vma
);
1916 if (vma
->vm_flags
& VM_ACCOUNT
)
1917 nr_accounted
+= nrpages
;
1918 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1919 vma
= remove_vma(vma
);
1921 vm_unacct_memory(nr_accounted
);
1926 * Get rid of page table information in the indicated region.
1928 * Called with the mm semaphore held.
1930 static void unmap_region(struct mm_struct
*mm
,
1931 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1932 unsigned long start
, unsigned long end
)
1934 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1935 struct mmu_gather tlb
;
1938 tlb_gather_mmu(&tlb
, mm
, 0);
1939 update_hiwater_rss(mm
);
1940 unmap_vmas(&tlb
, vma
, start
, end
);
1941 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1942 next
? next
->vm_start
: 0);
1943 tlb_finish_mmu(&tlb
, start
, end
);
1947 * Create a list of vma's touched by the unmap, removing them from the mm's
1948 * vma list as we go..
1951 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1952 struct vm_area_struct
*prev
, unsigned long end
)
1954 struct vm_area_struct
**insertion_point
;
1955 struct vm_area_struct
*tail_vma
= NULL
;
1958 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1959 vma
->vm_prev
= NULL
;
1961 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1965 } while (vma
&& vma
->vm_start
< end
);
1966 *insertion_point
= vma
;
1968 vma
->vm_prev
= prev
;
1969 tail_vma
->vm_next
= NULL
;
1970 if (mm
->unmap_area
== arch_unmap_area
)
1971 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1973 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1974 mm
->unmap_area(mm
, addr
);
1975 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1979 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1980 * munmap path where it doesn't make sense to fail.
1982 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1983 unsigned long addr
, int new_below
)
1985 struct mempolicy
*pol
;
1986 struct vm_area_struct
*new;
1989 if (is_vm_hugetlb_page(vma
) && (addr
&
1990 ~(huge_page_mask(hstate_vma(vma
)))))
1993 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1997 /* most fields are the same, copy all, and then fixup */
2000 INIT_LIST_HEAD(&new->anon_vma_chain
);
2005 new->vm_start
= addr
;
2006 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2009 pol
= mpol_dup(vma_policy(vma
));
2014 vma_set_policy(new, pol
);
2016 if (anon_vma_clone(new, vma
))
2020 get_file(new->vm_file
);
2022 if (new->vm_ops
&& new->vm_ops
->open
)
2023 new->vm_ops
->open(new);
2026 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2027 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2029 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2035 /* Clean everything up if vma_adjust failed. */
2036 if (new->vm_ops
&& new->vm_ops
->close
)
2037 new->vm_ops
->close(new);
2040 unlink_anon_vmas(new);
2044 kmem_cache_free(vm_area_cachep
, new);
2050 * Split a vma into two pieces at address 'addr', a new vma is allocated
2051 * either for the first part or the tail.
2053 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2054 unsigned long addr
, int new_below
)
2056 if (mm
->map_count
>= sysctl_max_map_count
)
2059 return __split_vma(mm
, vma
, addr
, new_below
);
2062 /* Munmap is split into 2 main parts -- this part which finds
2063 * what needs doing, and the areas themselves, which do the
2064 * work. This now handles partial unmappings.
2065 * Jeremy Fitzhardinge <jeremy@goop.org>
2067 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2070 struct vm_area_struct
*vma
, *prev
, *last
;
2072 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2075 if ((len
= PAGE_ALIGN(len
)) == 0)
2078 /* Find the first overlapping VMA */
2079 vma
= find_vma(mm
, start
);
2082 prev
= vma
->vm_prev
;
2083 /* we have start < vma->vm_end */
2085 /* if it doesn't overlap, we have nothing.. */
2087 if (vma
->vm_start
>= end
)
2091 * If we need to split any vma, do it now to save pain later.
2093 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2094 * unmapped vm_area_struct will remain in use: so lower split_vma
2095 * places tmp vma above, and higher split_vma places tmp vma below.
2097 if (start
> vma
->vm_start
) {
2101 * Make sure that map_count on return from munmap() will
2102 * not exceed its limit; but let map_count go just above
2103 * its limit temporarily, to help free resources as expected.
2105 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2108 error
= __split_vma(mm
, vma
, start
, 0);
2114 /* Does it split the last one? */
2115 last
= find_vma(mm
, end
);
2116 if (last
&& end
> last
->vm_start
) {
2117 int error
= __split_vma(mm
, last
, end
, 1);
2121 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2124 * unlock any mlock()ed ranges before detaching vmas
2126 if (mm
->locked_vm
) {
2127 struct vm_area_struct
*tmp
= vma
;
2128 while (tmp
&& tmp
->vm_start
< end
) {
2129 if (tmp
->vm_flags
& VM_LOCKED
) {
2130 mm
->locked_vm
-= vma_pages(tmp
);
2131 munlock_vma_pages_all(tmp
);
2138 * Remove the vma's, and unmap the actual pages
2140 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2141 unmap_region(mm
, vma
, prev
, start
, end
);
2143 /* Fix up all other VM information */
2144 remove_vma_list(mm
, vma
);
2149 int vm_munmap(unsigned long start
, size_t len
)
2152 struct mm_struct
*mm
= current
->mm
;
2154 down_write(&mm
->mmap_sem
);
2155 ret
= do_munmap(mm
, start
, len
);
2156 up_write(&mm
->mmap_sem
);
2159 EXPORT_SYMBOL(vm_munmap
);
2161 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2163 profile_munmap(addr
);
2164 return vm_munmap(addr
, len
);
2167 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2169 #ifdef CONFIG_DEBUG_VM
2170 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2172 up_read(&mm
->mmap_sem
);
2178 * this is really a simplified "do_mmap". it only handles
2179 * anonymous maps. eventually we may be able to do some
2180 * brk-specific accounting here.
2182 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2184 struct mm_struct
* mm
= current
->mm
;
2185 struct vm_area_struct
* vma
, * prev
;
2186 unsigned long flags
;
2187 struct rb_node
** rb_link
, * rb_parent
;
2188 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2191 len
= PAGE_ALIGN(len
);
2195 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2197 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2198 if (error
& ~PAGE_MASK
)
2204 if (mm
->def_flags
& VM_LOCKED
) {
2205 unsigned long locked
, lock_limit
;
2206 locked
= len
>> PAGE_SHIFT
;
2207 locked
+= mm
->locked_vm
;
2208 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2209 lock_limit
>>= PAGE_SHIFT
;
2210 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2215 * mm->mmap_sem is required to protect against another thread
2216 * changing the mappings in case we sleep.
2218 verify_mm_writelocked(mm
);
2221 * Clear old maps. this also does some error checking for us
2224 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
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_mm(mm
, 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 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2275 struct mm_struct
*mm
= current
->mm
;
2278 down_write(&mm
->mmap_sem
);
2279 ret
= do_brk(addr
, len
);
2280 up_write(&mm
->mmap_sem
);
2283 EXPORT_SYMBOL(vm_brk
);
2285 /* Release all mmaps. */
2286 void exit_mmap(struct mm_struct
*mm
)
2288 struct mmu_gather tlb
;
2289 struct vm_area_struct
*vma
;
2290 unsigned long nr_accounted
= 0;
2292 /* mm's last user has gone, and its about to be pulled down */
2293 mmu_notifier_release(mm
);
2295 if (mm
->locked_vm
) {
2298 if (vma
->vm_flags
& VM_LOCKED
)
2299 munlock_vma_pages_all(vma
);
2307 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2312 tlb_gather_mmu(&tlb
, mm
, 1);
2313 /* update_hiwater_rss(mm) here? but nobody should be looking */
2314 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2315 unmap_vmas(&tlb
, vma
, 0, -1);
2317 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2318 tlb_finish_mmu(&tlb
, 0, -1);
2321 * Walk the list again, actually closing and freeing it,
2322 * with preemption enabled, without holding any MM locks.
2325 if (vma
->vm_flags
& VM_ACCOUNT
)
2326 nr_accounted
+= vma_pages(vma
);
2327 vma
= remove_vma(vma
);
2329 vm_unacct_memory(nr_accounted
);
2331 WARN_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2334 /* Insert vm structure into process list sorted by address
2335 * and into the inode's i_mmap tree. If vm_file is non-NULL
2336 * then i_mmap_mutex is taken here.
2338 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2340 struct vm_area_struct
*prev
;
2341 struct rb_node
**rb_link
, *rb_parent
;
2344 * The vm_pgoff of a purely anonymous vma should be irrelevant
2345 * until its first write fault, when page's anon_vma and index
2346 * are set. But now set the vm_pgoff it will almost certainly
2347 * end up with (unless mremap moves it elsewhere before that
2348 * first wfault), so /proc/pid/maps tells a consistent story.
2350 * By setting it to reflect the virtual start address of the
2351 * vma, merges and splits can happen in a seamless way, just
2352 * using the existing file pgoff checks and manipulations.
2353 * Similarly in do_mmap_pgoff and in do_brk.
2355 if (!vma
->vm_file
) {
2356 BUG_ON(vma
->anon_vma
);
2357 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2359 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2360 &prev
, &rb_link
, &rb_parent
))
2362 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2363 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2366 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2371 * Copy the vma structure to a new location in the same mm,
2372 * prior to moving page table entries, to effect an mremap move.
2374 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2375 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2377 struct vm_area_struct
*vma
= *vmap
;
2378 unsigned long vma_start
= vma
->vm_start
;
2379 struct mm_struct
*mm
= vma
->vm_mm
;
2380 struct vm_area_struct
*new_vma
, *prev
;
2381 struct rb_node
**rb_link
, *rb_parent
;
2382 struct mempolicy
*pol
;
2383 bool faulted_in_anon_vma
= true;
2386 * If anonymous vma has not yet been faulted, update new pgoff
2387 * to match new location, to increase its chance of merging.
2389 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2390 pgoff
= addr
>> PAGE_SHIFT
;
2391 faulted_in_anon_vma
= false;
2394 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2395 return NULL
; /* should never get here */
2396 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2397 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2400 * Source vma may have been merged into new_vma
2402 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2403 vma_start
< new_vma
->vm_end
)) {
2405 * The only way we can get a vma_merge with
2406 * self during an mremap is if the vma hasn't
2407 * been faulted in yet and we were allowed to
2408 * reset the dst vma->vm_pgoff to the
2409 * destination address of the mremap to allow
2410 * the merge to happen. mremap must change the
2411 * vm_pgoff linearity between src and dst vmas
2412 * (in turn preventing a vma_merge) to be
2413 * safe. It is only safe to keep the vm_pgoff
2414 * linear if there are no pages mapped yet.
2416 VM_BUG_ON(faulted_in_anon_vma
);
2420 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2423 pol
= mpol_dup(vma_policy(vma
));
2426 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2427 if (anon_vma_clone(new_vma
, vma
))
2428 goto out_free_mempol
;
2429 vma_set_policy(new_vma
, pol
);
2430 new_vma
->vm_start
= addr
;
2431 new_vma
->vm_end
= addr
+ len
;
2432 new_vma
->vm_pgoff
= pgoff
;
2433 if (new_vma
->vm_file
)
2434 get_file(new_vma
->vm_file
);
2435 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2436 new_vma
->vm_ops
->open(new_vma
);
2437 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2445 kmem_cache_free(vm_area_cachep
, new_vma
);
2450 * Return true if the calling process may expand its vm space by the passed
2453 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2455 unsigned long cur
= mm
->total_vm
; /* pages */
2458 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2460 if (cur
+ npages
> lim
)
2466 static int special_mapping_fault(struct vm_area_struct
*vma
,
2467 struct vm_fault
*vmf
)
2470 struct page
**pages
;
2473 * special mappings have no vm_file, and in that case, the mm
2474 * uses vm_pgoff internally. So we have to subtract it from here.
2475 * We are allowed to do this because we are the mm; do not copy
2476 * this code into drivers!
2478 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2480 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2484 struct page
*page
= *pages
;
2490 return VM_FAULT_SIGBUS
;
2494 * Having a close hook prevents vma merging regardless of flags.
2496 static void special_mapping_close(struct vm_area_struct
*vma
)
2500 static const struct vm_operations_struct special_mapping_vmops
= {
2501 .close
= special_mapping_close
,
2502 .fault
= special_mapping_fault
,
2506 * Called with mm->mmap_sem held for writing.
2507 * Insert a new vma covering the given region, with the given flags.
2508 * Its pages are supplied by the given array of struct page *.
2509 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2510 * The region past the last page supplied will always produce SIGBUS.
2511 * The array pointer and the pages it points to are assumed to stay alive
2512 * for as long as this mapping might exist.
2514 int install_special_mapping(struct mm_struct
*mm
,
2515 unsigned long addr
, unsigned long len
,
2516 unsigned long vm_flags
, struct page
**pages
)
2519 struct vm_area_struct
*vma
;
2521 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2522 if (unlikely(vma
== NULL
))
2525 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2527 vma
->vm_start
= addr
;
2528 vma
->vm_end
= addr
+ len
;
2530 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2531 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2533 vma
->vm_ops
= &special_mapping_vmops
;
2534 vma
->vm_private_data
= pages
;
2536 ret
= insert_vm_struct(mm
, vma
);
2540 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2542 perf_event_mmap(vma
);
2547 kmem_cache_free(vm_area_cachep
, vma
);
2551 static DEFINE_MUTEX(mm_all_locks_mutex
);
2553 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2555 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
2557 * The LSB of head.next can't change from under us
2558 * because we hold the mm_all_locks_mutex.
2560 mutex_lock_nest_lock(&anon_vma
->root
->mutex
, &mm
->mmap_sem
);
2562 * We can safely modify head.next after taking the
2563 * anon_vma->root->mutex. If some other vma in this mm shares
2564 * the same anon_vma we won't take it again.
2566 * No need of atomic instructions here, head.next
2567 * can't change from under us thanks to the
2568 * anon_vma->root->mutex.
2570 if (__test_and_set_bit(0, (unsigned long *)
2571 &anon_vma
->root
->rb_root
.rb_node
))
2576 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2578 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2580 * AS_MM_ALL_LOCKS can't change from under us because
2581 * we hold the mm_all_locks_mutex.
2583 * Operations on ->flags have to be atomic because
2584 * even if AS_MM_ALL_LOCKS is stable thanks to the
2585 * mm_all_locks_mutex, there may be other cpus
2586 * changing other bitflags in parallel to us.
2588 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2590 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
2595 * This operation locks against the VM for all pte/vma/mm related
2596 * operations that could ever happen on a certain mm. This includes
2597 * vmtruncate, try_to_unmap, and all page faults.
2599 * The caller must take the mmap_sem in write mode before calling
2600 * mm_take_all_locks(). The caller isn't allowed to release the
2601 * mmap_sem until mm_drop_all_locks() returns.
2603 * mmap_sem in write mode is required in order to block all operations
2604 * that could modify pagetables and free pages without need of
2605 * altering the vma layout (for example populate_range() with
2606 * nonlinear vmas). It's also needed in write mode to avoid new
2607 * anon_vmas to be associated with existing vmas.
2609 * A single task can't take more than one mm_take_all_locks() in a row
2610 * or it would deadlock.
2612 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
2613 * mapping->flags avoid to take the same lock twice, if more than one
2614 * vma in this mm is backed by the same anon_vma or address_space.
2616 * We can take all the locks in random order because the VM code
2617 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2618 * takes more than one of them in a row. Secondly we're protected
2619 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2621 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2622 * that may have to take thousand of locks.
2624 * mm_take_all_locks() can fail if it's interrupted by signals.
2626 int mm_take_all_locks(struct mm_struct
*mm
)
2628 struct vm_area_struct
*vma
;
2629 struct anon_vma_chain
*avc
;
2631 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2633 mutex_lock(&mm_all_locks_mutex
);
2635 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2636 if (signal_pending(current
))
2638 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2639 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2642 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2643 if (signal_pending(current
))
2646 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2647 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2653 mm_drop_all_locks(mm
);
2657 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2659 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
2661 * The LSB of head.next can't change to 0 from under
2662 * us because we hold the mm_all_locks_mutex.
2664 * We must however clear the bitflag before unlocking
2665 * the vma so the users using the anon_vma->rb_root will
2666 * never see our bitflag.
2668 * No need of atomic instructions here, head.next
2669 * can't change from under us until we release the
2670 * anon_vma->root->mutex.
2672 if (!__test_and_clear_bit(0, (unsigned long *)
2673 &anon_vma
->root
->rb_root
.rb_node
))
2675 anon_vma_unlock(anon_vma
);
2679 static void vm_unlock_mapping(struct address_space
*mapping
)
2681 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2683 * AS_MM_ALL_LOCKS can't change to 0 from under us
2684 * because we hold the mm_all_locks_mutex.
2686 mutex_unlock(&mapping
->i_mmap_mutex
);
2687 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2694 * The mmap_sem cannot be released by the caller until
2695 * mm_drop_all_locks() returns.
2697 void mm_drop_all_locks(struct mm_struct
*mm
)
2699 struct vm_area_struct
*vma
;
2700 struct anon_vma_chain
*avc
;
2702 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2703 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2705 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2707 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2708 vm_unlock_anon_vma(avc
->anon_vma
);
2709 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2710 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2713 mutex_unlock(&mm_all_locks_mutex
);
2717 * initialise the VMA slab
2719 void __init
mmap_init(void)
2723 ret
= percpu_counter_init(&vm_committed_as
, 0);