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)
356 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
357 unsigned long end
, struct vm_area_struct
**pprev
,
358 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
360 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
362 __rb_link
= &mm
->mm_rb
.rb_node
;
363 rb_prev
= __rb_parent
= NULL
;
366 struct vm_area_struct
*vma_tmp
;
368 __rb_parent
= *__rb_link
;
369 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
371 if (vma_tmp
->vm_end
> addr
) {
372 /* Fail if an existing vma overlaps the area */
373 if (vma_tmp
->vm_start
< end
)
375 __rb_link
= &__rb_parent
->rb_left
;
377 rb_prev
= __rb_parent
;
378 __rb_link
= &__rb_parent
->rb_right
;
384 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
385 *rb_link
= __rb_link
;
386 *rb_parent
= __rb_parent
;
390 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
391 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
393 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
394 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
397 static void __vma_link_file(struct vm_area_struct
*vma
)
403 struct address_space
*mapping
= file
->f_mapping
;
405 if (vma
->vm_flags
& VM_DENYWRITE
)
406 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
407 if (vma
->vm_flags
& VM_SHARED
)
408 mapping
->i_mmap_writable
++;
410 flush_dcache_mmap_lock(mapping
);
411 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
412 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
414 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
415 flush_dcache_mmap_unlock(mapping
);
420 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
421 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
422 struct rb_node
*rb_parent
)
424 __vma_link_list(mm
, vma
, prev
, rb_parent
);
425 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
428 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
429 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
430 struct rb_node
*rb_parent
)
432 struct address_space
*mapping
= NULL
;
435 mapping
= vma
->vm_file
->f_mapping
;
438 mutex_lock(&mapping
->i_mmap_mutex
);
440 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
441 __vma_link_file(vma
);
444 mutex_unlock(&mapping
->i_mmap_mutex
);
451 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
452 * mm's list and rbtree. It has already been inserted into the interval tree.
454 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
456 struct vm_area_struct
*prev
;
457 struct rb_node
**rb_link
, *rb_parent
;
459 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
460 &prev
, &rb_link
, &rb_parent
))
462 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
467 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
468 struct vm_area_struct
*prev
)
470 struct vm_area_struct
*next
= vma
->vm_next
;
472 prev
->vm_next
= next
;
474 next
->vm_prev
= prev
;
475 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
476 if (mm
->mmap_cache
== vma
)
477 mm
->mmap_cache
= prev
;
481 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
482 * is already present in an i_mmap tree without adjusting the tree.
483 * The following helper function should be used when such adjustments
484 * are necessary. The "insert" vma (if any) is to be inserted
485 * before we drop the necessary locks.
487 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
488 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
490 struct mm_struct
*mm
= vma
->vm_mm
;
491 struct vm_area_struct
*next
= vma
->vm_next
;
492 struct vm_area_struct
*importer
= NULL
;
493 struct address_space
*mapping
= NULL
;
494 struct rb_root
*root
= NULL
;
495 struct anon_vma
*anon_vma
= NULL
;
496 struct file
*file
= vma
->vm_file
;
497 long adjust_next
= 0;
500 if (next
&& !insert
) {
501 struct vm_area_struct
*exporter
= NULL
;
503 if (end
>= next
->vm_end
) {
505 * vma expands, overlapping all the next, and
506 * perhaps the one after too (mprotect case 6).
508 again
: remove_next
= 1 + (end
> next
->vm_end
);
512 } else if (end
> next
->vm_start
) {
514 * vma expands, overlapping part of the next:
515 * mprotect case 5 shifting the boundary up.
517 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
520 } else if (end
< vma
->vm_end
) {
522 * vma shrinks, and !insert tells it's not
523 * split_vma inserting another: so it must be
524 * mprotect case 4 shifting the boundary down.
526 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
532 * Easily overlooked: when mprotect shifts the boundary,
533 * make sure the expanding vma has anon_vma set if the
534 * shrinking vma had, to cover any anon pages imported.
536 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
537 if (anon_vma_clone(importer
, exporter
))
539 importer
->anon_vma
= exporter
->anon_vma
;
544 mapping
= file
->f_mapping
;
545 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
546 root
= &mapping
->i_mmap
;
547 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
550 uprobe_munmap(next
, next
->vm_start
,
554 mutex_lock(&mapping
->i_mmap_mutex
);
557 * Put into interval tree now, so instantiated pages
558 * are visible to arm/parisc __flush_dcache_page
559 * throughout; but we cannot insert into address
560 * space until vma start or end is updated.
562 __vma_link_file(insert
);
566 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
569 * When changing only vma->vm_end, we don't really need anon_vma
570 * lock. This is a fairly rare case by itself, but the anon_vma
571 * lock may be shared between many sibling processes. Skipping
572 * the lock for brk adjustments makes a difference sometimes.
574 if (vma
->anon_vma
&& (importer
|| start
!= vma
->vm_start
)) {
575 anon_vma
= vma
->anon_vma
;
576 VM_BUG_ON(adjust_next
&& next
->anon_vma
&&
577 anon_vma
!= next
->anon_vma
);
578 } else if (adjust_next
&& next
->anon_vma
)
579 anon_vma
= next
->anon_vma
;
581 anon_vma_lock(anon_vma
);
584 flush_dcache_mmap_lock(mapping
);
585 vma_interval_tree_remove(vma
, root
);
587 vma_interval_tree_remove(next
, root
);
590 vma
->vm_start
= start
;
592 vma
->vm_pgoff
= pgoff
;
594 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
595 next
->vm_pgoff
+= adjust_next
;
600 vma_interval_tree_insert(next
, root
);
601 vma_interval_tree_insert(vma
, root
);
602 flush_dcache_mmap_unlock(mapping
);
607 * vma_merge has merged next into vma, and needs
608 * us to remove next before dropping the locks.
610 __vma_unlink(mm
, next
, vma
);
612 __remove_shared_vm_struct(next
, file
, mapping
);
615 * split_vma has split insert from vma, and needs
616 * us to insert it before dropping the locks
617 * (it may either follow vma or precede it).
619 __insert_vm_struct(mm
, insert
);
623 anon_vma_unlock(anon_vma
);
625 mutex_unlock(&mapping
->i_mmap_mutex
);
636 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
640 anon_vma_merge(vma
, next
);
642 mpol_put(vma_policy(next
));
643 kmem_cache_free(vm_area_cachep
, next
);
645 * In mprotect's case 6 (see comments on vma_merge),
646 * we must remove another next too. It would clutter
647 * up the code too much to do both in one go.
649 if (remove_next
== 2) {
663 * If the vma has a ->close operation then the driver probably needs to release
664 * per-vma resources, so we don't attempt to merge those.
666 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
667 struct file
*file
, unsigned long vm_flags
)
669 if (vma
->vm_flags
^ vm_flags
)
671 if (vma
->vm_file
!= file
)
673 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
678 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
679 struct anon_vma
*anon_vma2
,
680 struct vm_area_struct
*vma
)
683 * The list_is_singular() test is to avoid merging VMA cloned from
684 * parents. This can improve scalability caused by anon_vma lock.
686 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
687 list_is_singular(&vma
->anon_vma_chain
)))
689 return anon_vma1
== anon_vma2
;
693 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
694 * in front of (at a lower virtual address and file offset than) the vma.
696 * We cannot merge two vmas if they have differently assigned (non-NULL)
697 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
699 * We don't check here for the merged mmap wrapping around the end of pagecache
700 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
701 * wrap, nor mmaps which cover the final page at index -1UL.
704 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
705 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
707 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
708 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
709 if (vma
->vm_pgoff
== vm_pgoff
)
716 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
717 * beyond (at a higher virtual address and file offset than) the vma.
719 * We cannot merge two vmas if they have differently assigned (non-NULL)
720 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
723 can_vma_merge_after(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
, vma
)) {
729 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
730 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
737 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
738 * whether that can be merged with its predecessor or its successor.
739 * Or both (it neatly fills a hole).
741 * In most cases - when called for mmap, brk or mremap - [addr,end) is
742 * certain not to be mapped by the time vma_merge is called; but when
743 * called for mprotect, it is certain to be already mapped (either at
744 * an offset within prev, or at the start of next), and the flags of
745 * this area are about to be changed to vm_flags - and the no-change
746 * case has already been eliminated.
748 * The following mprotect cases have to be considered, where AAAA is
749 * the area passed down from mprotect_fixup, never extending beyond one
750 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
752 * AAAA AAAA AAAA AAAA
753 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
754 * cannot merge might become might become might become
755 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
756 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
757 * mremap move: PPPPNNNNNNNN 8
759 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
760 * might become case 1 below case 2 below case 3 below
762 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
763 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
765 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
766 struct vm_area_struct
*prev
, unsigned long addr
,
767 unsigned long end
, unsigned long vm_flags
,
768 struct anon_vma
*anon_vma
, struct file
*file
,
769 pgoff_t pgoff
, struct mempolicy
*policy
)
771 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
772 struct vm_area_struct
*area
, *next
;
776 * We later require that vma->vm_flags == vm_flags,
777 * so this tests vma->vm_flags & VM_SPECIAL, too.
779 if (vm_flags
& VM_SPECIAL
)
783 next
= prev
->vm_next
;
787 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
788 next
= next
->vm_next
;
791 * Can it merge with the predecessor?
793 if (prev
&& prev
->vm_end
== addr
&&
794 mpol_equal(vma_policy(prev
), policy
) &&
795 can_vma_merge_after(prev
, vm_flags
,
796 anon_vma
, file
, pgoff
)) {
798 * OK, it can. Can we now merge in the successor as well?
800 if (next
&& end
== next
->vm_start
&&
801 mpol_equal(policy
, vma_policy(next
)) &&
802 can_vma_merge_before(next
, vm_flags
,
803 anon_vma
, file
, pgoff
+pglen
) &&
804 is_mergeable_anon_vma(prev
->anon_vma
,
805 next
->anon_vma
, NULL
)) {
807 err
= vma_adjust(prev
, prev
->vm_start
,
808 next
->vm_end
, prev
->vm_pgoff
, NULL
);
809 } else /* cases 2, 5, 7 */
810 err
= vma_adjust(prev
, prev
->vm_start
,
811 end
, prev
->vm_pgoff
, NULL
);
814 khugepaged_enter_vma_merge(prev
);
819 * Can this new request be merged in front of next?
821 if (next
&& end
== next
->vm_start
&&
822 mpol_equal(policy
, vma_policy(next
)) &&
823 can_vma_merge_before(next
, vm_flags
,
824 anon_vma
, file
, pgoff
+pglen
)) {
825 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
826 err
= vma_adjust(prev
, prev
->vm_start
,
827 addr
, prev
->vm_pgoff
, NULL
);
828 else /* cases 3, 8 */
829 err
= vma_adjust(area
, addr
, next
->vm_end
,
830 next
->vm_pgoff
- pglen
, NULL
);
833 khugepaged_enter_vma_merge(area
);
841 * Rough compatbility check to quickly see if it's even worth looking
842 * at sharing an anon_vma.
844 * They need to have the same vm_file, and the flags can only differ
845 * in things that mprotect may change.
847 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
848 * we can merge the two vma's. For example, we refuse to merge a vma if
849 * there is a vm_ops->close() function, because that indicates that the
850 * driver is doing some kind of reference counting. But that doesn't
851 * really matter for the anon_vma sharing case.
853 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
855 return a
->vm_end
== b
->vm_start
&&
856 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
857 a
->vm_file
== b
->vm_file
&&
858 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
859 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
863 * Do some basic sanity checking to see if we can re-use the anon_vma
864 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
865 * the same as 'old', the other will be the new one that is trying
866 * to share the anon_vma.
868 * NOTE! This runs with mm_sem held for reading, so it is possible that
869 * the anon_vma of 'old' is concurrently in the process of being set up
870 * by another page fault trying to merge _that_. But that's ok: if it
871 * is being set up, that automatically means that it will be a singleton
872 * acceptable for merging, so we can do all of this optimistically. But
873 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
875 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
876 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
877 * is to return an anon_vma that is "complex" due to having gone through
880 * We also make sure that the two vma's are compatible (adjacent,
881 * and with the same memory policies). That's all stable, even with just
882 * a read lock on the mm_sem.
884 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
886 if (anon_vma_compatible(a
, b
)) {
887 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
889 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
896 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
897 * neighbouring vmas for a suitable anon_vma, before it goes off
898 * to allocate a new anon_vma. It checks because a repetitive
899 * sequence of mprotects and faults may otherwise lead to distinct
900 * anon_vmas being allocated, preventing vma merge in subsequent
903 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
905 struct anon_vma
*anon_vma
;
906 struct vm_area_struct
*near
;
912 anon_vma
= reusable_anon_vma(near
, vma
, near
);
920 anon_vma
= reusable_anon_vma(near
, near
, vma
);
925 * There's no absolute need to look only at touching neighbours:
926 * we could search further afield for "compatible" anon_vmas.
927 * But it would probably just be a waste of time searching,
928 * or lead to too many vmas hanging off the same anon_vma.
929 * We're trying to allow mprotect remerging later on,
930 * not trying to minimize memory used for anon_vmas.
935 #ifdef CONFIG_PROC_FS
936 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
937 struct file
*file
, long pages
)
939 const unsigned long stack_flags
940 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
942 mm
->total_vm
+= pages
;
945 mm
->shared_vm
+= pages
;
946 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
947 mm
->exec_vm
+= pages
;
948 } else if (flags
& stack_flags
)
949 mm
->stack_vm
+= pages
;
951 #endif /* CONFIG_PROC_FS */
954 * If a hint addr is less than mmap_min_addr change hint to be as
955 * low as possible but still greater than mmap_min_addr
957 static inline unsigned long round_hint_to_min(unsigned long hint
)
960 if (((void *)hint
!= NULL
) &&
961 (hint
< mmap_min_addr
))
962 return PAGE_ALIGN(mmap_min_addr
);
967 * The caller must hold down_write(¤t->mm->mmap_sem).
970 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
971 unsigned long len
, unsigned long prot
,
972 unsigned long flags
, unsigned long pgoff
)
974 struct mm_struct
* mm
= current
->mm
;
979 * Does the application expect PROT_READ to imply PROT_EXEC?
981 * (the exception is when the underlying filesystem is noexec
982 * mounted, in which case we dont add PROT_EXEC.)
984 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
985 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
991 if (!(flags
& MAP_FIXED
))
992 addr
= round_hint_to_min(addr
);
994 /* Careful about overflows.. */
995 len
= PAGE_ALIGN(len
);
999 /* offset overflow? */
1000 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1003 /* Too many mappings? */
1004 if (mm
->map_count
> sysctl_max_map_count
)
1007 /* Obtain the address to map to. we verify (or select) it and ensure
1008 * that it represents a valid section of the address space.
1010 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1011 if (addr
& ~PAGE_MASK
)
1014 /* Do simple checking here so the lower-level routines won't have
1015 * to. we assume access permissions have been handled by the open
1016 * of the memory object, so we don't do any here.
1018 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1019 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1021 if (flags
& MAP_LOCKED
)
1022 if (!can_do_mlock())
1025 /* mlock MCL_FUTURE? */
1026 if (vm_flags
& VM_LOCKED
) {
1027 unsigned long locked
, lock_limit
;
1028 locked
= len
>> PAGE_SHIFT
;
1029 locked
+= mm
->locked_vm
;
1030 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1031 lock_limit
>>= PAGE_SHIFT
;
1032 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1036 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1039 switch (flags
& MAP_TYPE
) {
1041 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1045 * Make sure we don't allow writing to an append-only
1048 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1052 * Make sure there are no mandatory locks on the file.
1054 if (locks_verify_locked(inode
))
1057 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1058 if (!(file
->f_mode
& FMODE_WRITE
))
1059 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1063 if (!(file
->f_mode
& FMODE_READ
))
1065 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1066 if (vm_flags
& VM_EXEC
)
1068 vm_flags
&= ~VM_MAYEXEC
;
1071 if (!file
->f_op
|| !file
->f_op
->mmap
)
1079 switch (flags
& MAP_TYPE
) {
1085 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1089 * Set pgoff according to addr for anon_vma.
1091 pgoff
= addr
>> PAGE_SHIFT
;
1098 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1101 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1102 unsigned long, prot
, unsigned long, flags
,
1103 unsigned long, fd
, unsigned long, pgoff
)
1105 struct file
*file
= NULL
;
1106 unsigned long retval
= -EBADF
;
1108 if (!(flags
& MAP_ANONYMOUS
)) {
1109 audit_mmap_fd(fd
, flags
);
1110 if (unlikely(flags
& MAP_HUGETLB
))
1115 } else if (flags
& MAP_HUGETLB
) {
1116 struct user_struct
*user
= NULL
;
1118 * VM_NORESERVE is used because the reservations will be
1119 * taken when vm_ops->mmap() is called
1120 * A dummy user value is used because we are not locking
1121 * memory so no accounting is necessary
1123 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, addr
, len
,
1124 VM_NORESERVE
, &user
,
1125 HUGETLB_ANONHUGE_INODE
);
1127 return PTR_ERR(file
);
1130 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1132 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1139 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1140 struct mmap_arg_struct
{
1144 unsigned long flags
;
1146 unsigned long offset
;
1149 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1151 struct mmap_arg_struct a
;
1153 if (copy_from_user(&a
, arg
, sizeof(a
)))
1155 if (a
.offset
& ~PAGE_MASK
)
1158 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1159 a
.offset
>> PAGE_SHIFT
);
1161 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1164 * Some shared mappigns will want the pages marked read-only
1165 * to track write events. If so, we'll downgrade vm_page_prot
1166 * to the private version (using protection_map[] without the
1169 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1171 vm_flags_t vm_flags
= vma
->vm_flags
;
1173 /* If it was private or non-writable, the write bit is already clear */
1174 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1177 /* The backer wishes to know when pages are first written to? */
1178 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1181 /* The open routine did something to the protections already? */
1182 if (pgprot_val(vma
->vm_page_prot
) !=
1183 pgprot_val(vm_get_page_prot(vm_flags
)))
1186 /* Specialty mapping? */
1187 if (vm_flags
& VM_PFNMAP
)
1190 /* Can the mapping track the dirty pages? */
1191 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1192 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1196 * We account for memory if it's a private writeable mapping,
1197 * not hugepages and VM_NORESERVE wasn't set.
1199 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1202 * hugetlb has its own accounting separate from the core VM
1203 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1205 if (file
&& is_file_hugepages(file
))
1208 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1211 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1212 unsigned long len
, unsigned long flags
,
1213 vm_flags_t vm_flags
, unsigned long pgoff
)
1215 struct mm_struct
*mm
= current
->mm
;
1216 struct vm_area_struct
*vma
, *prev
;
1217 int correct_wcount
= 0;
1219 struct rb_node
**rb_link
, *rb_parent
;
1220 unsigned long charged
= 0;
1221 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1223 /* Clear old maps */
1226 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
1227 if (do_munmap(mm
, addr
, len
))
1232 /* Check against address space limit. */
1233 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1237 * Set 'VM_NORESERVE' if we should not account for the
1238 * memory use of this mapping.
1240 if ((flags
& MAP_NORESERVE
)) {
1241 /* We honor MAP_NORESERVE if allowed to overcommit */
1242 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1243 vm_flags
|= VM_NORESERVE
;
1245 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1246 if (file
&& is_file_hugepages(file
))
1247 vm_flags
|= VM_NORESERVE
;
1251 * Private writable mapping: check memory availability
1253 if (accountable_mapping(file
, vm_flags
)) {
1254 charged
= len
>> PAGE_SHIFT
;
1255 if (security_vm_enough_memory_mm(mm
, charged
))
1257 vm_flags
|= VM_ACCOUNT
;
1261 * Can we just expand an old mapping?
1263 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1268 * Determine the object being mapped and call the appropriate
1269 * specific mapper. the address has already been validated, but
1270 * not unmapped, but the maps are removed from the list.
1272 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1279 vma
->vm_start
= addr
;
1280 vma
->vm_end
= addr
+ len
;
1281 vma
->vm_flags
= vm_flags
;
1282 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1283 vma
->vm_pgoff
= pgoff
;
1284 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1286 error
= -EINVAL
; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1289 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1291 if (vm_flags
& VM_DENYWRITE
) {
1292 error
= deny_write_access(file
);
1297 vma
->vm_file
= get_file(file
);
1298 error
= file
->f_op
->mmap(file
, vma
);
1300 goto unmap_and_free_vma
;
1302 /* Can addr have changed??
1304 * Answer: Yes, several device drivers can do it in their
1305 * f_op->mmap method. -DaveM
1307 addr
= vma
->vm_start
;
1308 pgoff
= vma
->vm_pgoff
;
1309 vm_flags
= vma
->vm_flags
;
1310 } else if (vm_flags
& VM_SHARED
) {
1311 if (unlikely(vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
)))
1313 error
= shmem_zero_setup(vma
);
1318 if (vma_wants_writenotify(vma
)) {
1319 pgprot_t pprot
= vma
->vm_page_prot
;
1321 /* Can vma->vm_page_prot have changed??
1323 * Answer: Yes, drivers may have changed it in their
1324 * f_op->mmap method.
1326 * Ensures that vmas marked as uncached stay that way.
1328 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1329 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1330 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1333 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1334 file
= vma
->vm_file
;
1336 /* Once vma denies write, undo our temporary denial count */
1338 atomic_inc(&inode
->i_writecount
);
1340 perf_event_mmap(vma
);
1342 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1343 if (vm_flags
& VM_LOCKED
) {
1344 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1345 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1346 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1347 make_pages_present(addr
, addr
+ len
);
1356 atomic_inc(&inode
->i_writecount
);
1357 vma
->vm_file
= NULL
;
1360 /* Undo any partial mapping done by a device driver. */
1361 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1364 kmem_cache_free(vm_area_cachep
, vma
);
1367 vm_unacct_memory(charged
);
1371 /* Get an address range which is currently unmapped.
1372 * For shmat() with addr=0.
1374 * Ugly calling convention alert:
1375 * Return value with the low bits set means error value,
1377 * if (ret & ~PAGE_MASK)
1380 * This function "knows" that -ENOMEM has the bits set.
1382 #ifndef HAVE_ARCH_UNMAPPED_AREA
1384 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1385 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1387 struct mm_struct
*mm
= current
->mm
;
1388 struct vm_area_struct
*vma
;
1389 unsigned long start_addr
;
1391 if (len
> TASK_SIZE
)
1394 if (flags
& MAP_FIXED
)
1398 addr
= PAGE_ALIGN(addr
);
1399 vma
= find_vma(mm
, addr
);
1400 if (TASK_SIZE
- len
>= addr
&&
1401 (!vma
|| addr
+ len
<= vma
->vm_start
))
1404 if (len
> mm
->cached_hole_size
) {
1405 start_addr
= addr
= mm
->free_area_cache
;
1407 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1408 mm
->cached_hole_size
= 0;
1412 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1413 /* At this point: (!vma || addr < vma->vm_end). */
1414 if (TASK_SIZE
- len
< addr
) {
1416 * Start a new search - just in case we missed
1419 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1420 addr
= TASK_UNMAPPED_BASE
;
1422 mm
->cached_hole_size
= 0;
1427 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1429 * Remember the place where we stopped the search:
1431 mm
->free_area_cache
= addr
+ len
;
1434 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1435 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1441 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1444 * Is this a new hole at the lowest possible address?
1446 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1447 mm
->free_area_cache
= addr
;
1451 * This mmap-allocator allocates new areas top-down from below the
1452 * stack's low limit (the base):
1454 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1456 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1457 const unsigned long len
, const unsigned long pgoff
,
1458 const unsigned long flags
)
1460 struct vm_area_struct
*vma
;
1461 struct mm_struct
*mm
= current
->mm
;
1462 unsigned long addr
= addr0
, start_addr
;
1464 /* requested length too big for entire address space */
1465 if (len
> TASK_SIZE
)
1468 if (flags
& MAP_FIXED
)
1471 /* requesting a specific address */
1473 addr
= PAGE_ALIGN(addr
);
1474 vma
= find_vma(mm
, addr
);
1475 if (TASK_SIZE
- len
>= addr
&&
1476 (!vma
|| addr
+ len
<= vma
->vm_start
))
1480 /* check if free_area_cache is useful for us */
1481 if (len
<= mm
->cached_hole_size
) {
1482 mm
->cached_hole_size
= 0;
1483 mm
->free_area_cache
= mm
->mmap_base
;
1487 /* either no address requested or can't fit in requested address hole */
1488 start_addr
= addr
= mm
->free_area_cache
;
1496 * Lookup failure means no vma is above this address,
1497 * else if new region fits below vma->vm_start,
1498 * return with success:
1500 vma
= find_vma(mm
, addr
);
1501 if (!vma
|| addr
+len
<= vma
->vm_start
)
1502 /* remember the address as a hint for next time */
1503 return (mm
->free_area_cache
= addr
);
1505 /* remember the largest hole we saw so far */
1506 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1507 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1509 /* try just below the current vma->vm_start */
1510 addr
= vma
->vm_start
-len
;
1511 } while (len
< vma
->vm_start
);
1515 * if hint left us with no space for the requested
1516 * mapping then try again:
1518 * Note: this is different with the case of bottomup
1519 * which does the fully line-search, but we use find_vma
1520 * here that causes some holes skipped.
1522 if (start_addr
!= mm
->mmap_base
) {
1523 mm
->free_area_cache
= mm
->mmap_base
;
1524 mm
->cached_hole_size
= 0;
1529 * A failed mmap() very likely causes application failure,
1530 * so fall back to the bottom-up function here. This scenario
1531 * can happen with large stack limits and large mmap()
1534 mm
->cached_hole_size
= ~0UL;
1535 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1536 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1538 * Restore the topdown base:
1540 mm
->free_area_cache
= mm
->mmap_base
;
1541 mm
->cached_hole_size
= ~0UL;
1547 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1550 * Is this a new hole at the highest possible address?
1552 if (addr
> mm
->free_area_cache
)
1553 mm
->free_area_cache
= addr
;
1555 /* dont allow allocations above current base */
1556 if (mm
->free_area_cache
> mm
->mmap_base
)
1557 mm
->free_area_cache
= mm
->mmap_base
;
1561 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1562 unsigned long pgoff
, unsigned long flags
)
1564 unsigned long (*get_area
)(struct file
*, unsigned long,
1565 unsigned long, unsigned long, unsigned long);
1567 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1571 /* Careful about overflows.. */
1572 if (len
> TASK_SIZE
)
1575 get_area
= current
->mm
->get_unmapped_area
;
1576 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1577 get_area
= file
->f_op
->get_unmapped_area
;
1578 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1579 if (IS_ERR_VALUE(addr
))
1582 if (addr
> TASK_SIZE
- len
)
1584 if (addr
& ~PAGE_MASK
)
1587 addr
= arch_rebalance_pgtables(addr
, len
);
1588 error
= security_mmap_addr(addr
);
1589 return error
? error
: addr
;
1592 EXPORT_SYMBOL(get_unmapped_area
);
1594 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1595 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1597 struct vm_area_struct
*vma
= NULL
;
1599 if (WARN_ON_ONCE(!mm
)) /* Remove this in linux-3.6 */
1602 /* Check the cache first. */
1603 /* (Cache hit rate is typically around 35%.) */
1604 vma
= mm
->mmap_cache
;
1605 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1606 struct rb_node
*rb_node
;
1608 rb_node
= mm
->mm_rb
.rb_node
;
1612 struct vm_area_struct
*vma_tmp
;
1614 vma_tmp
= rb_entry(rb_node
,
1615 struct vm_area_struct
, vm_rb
);
1617 if (vma_tmp
->vm_end
> addr
) {
1619 if (vma_tmp
->vm_start
<= addr
)
1621 rb_node
= rb_node
->rb_left
;
1623 rb_node
= rb_node
->rb_right
;
1626 mm
->mmap_cache
= vma
;
1631 EXPORT_SYMBOL(find_vma
);
1634 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1636 struct vm_area_struct
*
1637 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1638 struct vm_area_struct
**pprev
)
1640 struct vm_area_struct
*vma
;
1642 vma
= find_vma(mm
, addr
);
1644 *pprev
= vma
->vm_prev
;
1646 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
1649 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1650 rb_node
= rb_node
->rb_right
;
1657 * Verify that the stack growth is acceptable and
1658 * update accounting. This is shared with both the
1659 * grow-up and grow-down cases.
1661 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1663 struct mm_struct
*mm
= vma
->vm_mm
;
1664 struct rlimit
*rlim
= current
->signal
->rlim
;
1665 unsigned long new_start
;
1667 /* address space limit tests */
1668 if (!may_expand_vm(mm
, grow
))
1671 /* Stack limit test */
1672 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1675 /* mlock limit tests */
1676 if (vma
->vm_flags
& VM_LOCKED
) {
1677 unsigned long locked
;
1678 unsigned long limit
;
1679 locked
= mm
->locked_vm
+ grow
;
1680 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1681 limit
>>= PAGE_SHIFT
;
1682 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1686 /* Check to ensure the stack will not grow into a hugetlb-only region */
1687 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1689 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1693 * Overcommit.. This must be the final test, as it will
1694 * update security statistics.
1696 if (security_vm_enough_memory_mm(mm
, grow
))
1699 /* Ok, everything looks good - let it rip */
1700 if (vma
->vm_flags
& VM_LOCKED
)
1701 mm
->locked_vm
+= grow
;
1702 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1706 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1708 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1709 * vma is the last one with address > vma->vm_end. Have to extend vma.
1711 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1715 if (!(vma
->vm_flags
& VM_GROWSUP
))
1719 * We must make sure the anon_vma is allocated
1720 * so that the anon_vma locking is not a noop.
1722 if (unlikely(anon_vma_prepare(vma
)))
1724 vma_lock_anon_vma(vma
);
1727 * vma->vm_start/vm_end cannot change under us because the caller
1728 * is required to hold the mmap_sem in read mode. We need the
1729 * anon_vma lock to serialize against concurrent expand_stacks.
1730 * Also guard against wrapping around to address 0.
1732 if (address
< PAGE_ALIGN(address
+4))
1733 address
= PAGE_ALIGN(address
+4);
1735 vma_unlock_anon_vma(vma
);
1740 /* Somebody else might have raced and expanded it already */
1741 if (address
> vma
->vm_end
) {
1742 unsigned long size
, grow
;
1744 size
= address
- vma
->vm_start
;
1745 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1748 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1749 error
= acct_stack_growth(vma
, size
, grow
);
1751 vma
->vm_end
= address
;
1752 perf_event_mmap(vma
);
1756 vma_unlock_anon_vma(vma
);
1757 khugepaged_enter_vma_merge(vma
);
1760 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1763 * vma is the first one with address < vma->vm_start. Have to extend vma.
1765 int expand_downwards(struct vm_area_struct
*vma
,
1766 unsigned long address
)
1771 * We must make sure the anon_vma is allocated
1772 * so that the anon_vma locking is not a noop.
1774 if (unlikely(anon_vma_prepare(vma
)))
1777 address
&= PAGE_MASK
;
1778 error
= security_mmap_addr(address
);
1782 vma_lock_anon_vma(vma
);
1785 * vma->vm_start/vm_end cannot change under us because the caller
1786 * is required to hold the mmap_sem in read mode. We need the
1787 * anon_vma lock to serialize against concurrent expand_stacks.
1790 /* Somebody else might have raced and expanded it already */
1791 if (address
< vma
->vm_start
) {
1792 unsigned long size
, grow
;
1794 size
= vma
->vm_end
- address
;
1795 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1798 if (grow
<= vma
->vm_pgoff
) {
1799 error
= acct_stack_growth(vma
, size
, grow
);
1801 vma
->vm_start
= address
;
1802 vma
->vm_pgoff
-= grow
;
1803 perf_event_mmap(vma
);
1807 vma_unlock_anon_vma(vma
);
1808 khugepaged_enter_vma_merge(vma
);
1812 #ifdef CONFIG_STACK_GROWSUP
1813 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1815 return expand_upwards(vma
, address
);
1818 struct vm_area_struct
*
1819 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1821 struct vm_area_struct
*vma
, *prev
;
1824 vma
= find_vma_prev(mm
, addr
, &prev
);
1825 if (vma
&& (vma
->vm_start
<= addr
))
1827 if (!prev
|| expand_stack(prev
, addr
))
1829 if (prev
->vm_flags
& VM_LOCKED
) {
1830 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1835 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1837 return expand_downwards(vma
, address
);
1840 struct vm_area_struct
*
1841 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1843 struct vm_area_struct
* vma
;
1844 unsigned long start
;
1847 vma
= find_vma(mm
,addr
);
1850 if (vma
->vm_start
<= addr
)
1852 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1854 start
= vma
->vm_start
;
1855 if (expand_stack(vma
, addr
))
1857 if (vma
->vm_flags
& VM_LOCKED
) {
1858 mlock_vma_pages_range(vma
, addr
, start
);
1865 * Ok - we have the memory areas we should free on the vma list,
1866 * so release them, and do the vma updates.
1868 * Called with the mm semaphore held.
1870 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1872 unsigned long nr_accounted
= 0;
1874 /* Update high watermark before we lower total_vm */
1875 update_hiwater_vm(mm
);
1877 long nrpages
= vma_pages(vma
);
1879 if (vma
->vm_flags
& VM_ACCOUNT
)
1880 nr_accounted
+= nrpages
;
1881 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1882 vma
= remove_vma(vma
);
1884 vm_unacct_memory(nr_accounted
);
1889 * Get rid of page table information in the indicated region.
1891 * Called with the mm semaphore held.
1893 static void unmap_region(struct mm_struct
*mm
,
1894 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1895 unsigned long start
, unsigned long end
)
1897 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1898 struct mmu_gather tlb
;
1901 tlb_gather_mmu(&tlb
, mm
, 0);
1902 update_hiwater_rss(mm
);
1903 unmap_vmas(&tlb
, vma
, start
, end
);
1904 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1905 next
? next
->vm_start
: 0);
1906 tlb_finish_mmu(&tlb
, start
, end
);
1910 * Create a list of vma's touched by the unmap, removing them from the mm's
1911 * vma list as we go..
1914 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1915 struct vm_area_struct
*prev
, unsigned long end
)
1917 struct vm_area_struct
**insertion_point
;
1918 struct vm_area_struct
*tail_vma
= NULL
;
1921 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1922 vma
->vm_prev
= NULL
;
1924 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1928 } while (vma
&& vma
->vm_start
< end
);
1929 *insertion_point
= vma
;
1931 vma
->vm_prev
= prev
;
1932 tail_vma
->vm_next
= NULL
;
1933 if (mm
->unmap_area
== arch_unmap_area
)
1934 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1936 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1937 mm
->unmap_area(mm
, addr
);
1938 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1942 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1943 * munmap path where it doesn't make sense to fail.
1945 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1946 unsigned long addr
, int new_below
)
1948 struct mempolicy
*pol
;
1949 struct vm_area_struct
*new;
1952 if (is_vm_hugetlb_page(vma
) && (addr
&
1953 ~(huge_page_mask(hstate_vma(vma
)))))
1956 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1960 /* most fields are the same, copy all, and then fixup */
1963 INIT_LIST_HEAD(&new->anon_vma_chain
);
1968 new->vm_start
= addr
;
1969 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1972 pol
= mpol_dup(vma_policy(vma
));
1977 vma_set_policy(new, pol
);
1979 if (anon_vma_clone(new, vma
))
1983 get_file(new->vm_file
);
1985 if (new->vm_ops
&& new->vm_ops
->open
)
1986 new->vm_ops
->open(new);
1989 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1990 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1992 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1998 /* Clean everything up if vma_adjust failed. */
1999 if (new->vm_ops
&& new->vm_ops
->close
)
2000 new->vm_ops
->close(new);
2003 unlink_anon_vmas(new);
2007 kmem_cache_free(vm_area_cachep
, new);
2013 * Split a vma into two pieces at address 'addr', a new vma is allocated
2014 * either for the first part or the tail.
2016 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2017 unsigned long addr
, int new_below
)
2019 if (mm
->map_count
>= sysctl_max_map_count
)
2022 return __split_vma(mm
, vma
, addr
, new_below
);
2025 /* Munmap is split into 2 main parts -- this part which finds
2026 * what needs doing, and the areas themselves, which do the
2027 * work. This now handles partial unmappings.
2028 * Jeremy Fitzhardinge <jeremy@goop.org>
2030 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2033 struct vm_area_struct
*vma
, *prev
, *last
;
2035 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2038 if ((len
= PAGE_ALIGN(len
)) == 0)
2041 /* Find the first overlapping VMA */
2042 vma
= find_vma(mm
, start
);
2045 prev
= vma
->vm_prev
;
2046 /* we have start < vma->vm_end */
2048 /* if it doesn't overlap, we have nothing.. */
2050 if (vma
->vm_start
>= end
)
2054 * If we need to split any vma, do it now to save pain later.
2056 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2057 * unmapped vm_area_struct will remain in use: so lower split_vma
2058 * places tmp vma above, and higher split_vma places tmp vma below.
2060 if (start
> vma
->vm_start
) {
2064 * Make sure that map_count on return from munmap() will
2065 * not exceed its limit; but let map_count go just above
2066 * its limit temporarily, to help free resources as expected.
2068 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2071 error
= __split_vma(mm
, vma
, start
, 0);
2077 /* Does it split the last one? */
2078 last
= find_vma(mm
, end
);
2079 if (last
&& end
> last
->vm_start
) {
2080 int error
= __split_vma(mm
, last
, end
, 1);
2084 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2087 * unlock any mlock()ed ranges before detaching vmas
2089 if (mm
->locked_vm
) {
2090 struct vm_area_struct
*tmp
= vma
;
2091 while (tmp
&& tmp
->vm_start
< end
) {
2092 if (tmp
->vm_flags
& VM_LOCKED
) {
2093 mm
->locked_vm
-= vma_pages(tmp
);
2094 munlock_vma_pages_all(tmp
);
2101 * Remove the vma's, and unmap the actual pages
2103 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2104 unmap_region(mm
, vma
, prev
, start
, end
);
2106 /* Fix up all other VM information */
2107 remove_vma_list(mm
, vma
);
2112 int vm_munmap(unsigned long start
, size_t len
)
2115 struct mm_struct
*mm
= current
->mm
;
2117 down_write(&mm
->mmap_sem
);
2118 ret
= do_munmap(mm
, start
, len
);
2119 up_write(&mm
->mmap_sem
);
2122 EXPORT_SYMBOL(vm_munmap
);
2124 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2126 profile_munmap(addr
);
2127 return vm_munmap(addr
, len
);
2130 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2132 #ifdef CONFIG_DEBUG_VM
2133 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2135 up_read(&mm
->mmap_sem
);
2141 * this is really a simplified "do_mmap". it only handles
2142 * anonymous maps. eventually we may be able to do some
2143 * brk-specific accounting here.
2145 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2147 struct mm_struct
* mm
= current
->mm
;
2148 struct vm_area_struct
* vma
, * prev
;
2149 unsigned long flags
;
2150 struct rb_node
** rb_link
, * rb_parent
;
2151 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2154 len
= PAGE_ALIGN(len
);
2158 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2160 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2161 if (error
& ~PAGE_MASK
)
2167 if (mm
->def_flags
& VM_LOCKED
) {
2168 unsigned long locked
, lock_limit
;
2169 locked
= len
>> PAGE_SHIFT
;
2170 locked
+= mm
->locked_vm
;
2171 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2172 lock_limit
>>= PAGE_SHIFT
;
2173 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2178 * mm->mmap_sem is required to protect against another thread
2179 * changing the mappings in case we sleep.
2181 verify_mm_writelocked(mm
);
2184 * Clear old maps. this also does some error checking for us
2187 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
2188 if (do_munmap(mm
, addr
, len
))
2193 /* Check against address space limits *after* clearing old maps... */
2194 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2197 if (mm
->map_count
> sysctl_max_map_count
)
2200 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2203 /* Can we just expand an old private anonymous mapping? */
2204 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2205 NULL
, NULL
, pgoff
, NULL
);
2210 * create a vma struct for an anonymous mapping
2212 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2214 vm_unacct_memory(len
>> PAGE_SHIFT
);
2218 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2220 vma
->vm_start
= addr
;
2221 vma
->vm_end
= addr
+ len
;
2222 vma
->vm_pgoff
= pgoff
;
2223 vma
->vm_flags
= flags
;
2224 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2225 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2227 perf_event_mmap(vma
);
2228 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2229 if (flags
& VM_LOCKED
) {
2230 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2231 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2236 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2238 struct mm_struct
*mm
= current
->mm
;
2241 down_write(&mm
->mmap_sem
);
2242 ret
= do_brk(addr
, len
);
2243 up_write(&mm
->mmap_sem
);
2246 EXPORT_SYMBOL(vm_brk
);
2248 /* Release all mmaps. */
2249 void exit_mmap(struct mm_struct
*mm
)
2251 struct mmu_gather tlb
;
2252 struct vm_area_struct
*vma
;
2253 unsigned long nr_accounted
= 0;
2255 /* mm's last user has gone, and its about to be pulled down */
2256 mmu_notifier_release(mm
);
2258 if (mm
->locked_vm
) {
2261 if (vma
->vm_flags
& VM_LOCKED
)
2262 munlock_vma_pages_all(vma
);
2270 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2275 tlb_gather_mmu(&tlb
, mm
, 1);
2276 /* update_hiwater_rss(mm) here? but nobody should be looking */
2277 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2278 unmap_vmas(&tlb
, vma
, 0, -1);
2280 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2281 tlb_finish_mmu(&tlb
, 0, -1);
2284 * Walk the list again, actually closing and freeing it,
2285 * with preemption enabled, without holding any MM locks.
2288 if (vma
->vm_flags
& VM_ACCOUNT
)
2289 nr_accounted
+= vma_pages(vma
);
2290 vma
= remove_vma(vma
);
2292 vm_unacct_memory(nr_accounted
);
2294 WARN_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2297 /* Insert vm structure into process list sorted by address
2298 * and into the inode's i_mmap tree. If vm_file is non-NULL
2299 * then i_mmap_mutex is taken here.
2301 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2303 struct vm_area_struct
*prev
;
2304 struct rb_node
**rb_link
, *rb_parent
;
2307 * The vm_pgoff of a purely anonymous vma should be irrelevant
2308 * until its first write fault, when page's anon_vma and index
2309 * are set. But now set the vm_pgoff it will almost certainly
2310 * end up with (unless mremap moves it elsewhere before that
2311 * first wfault), so /proc/pid/maps tells a consistent story.
2313 * By setting it to reflect the virtual start address of the
2314 * vma, merges and splits can happen in a seamless way, just
2315 * using the existing file pgoff checks and manipulations.
2316 * Similarly in do_mmap_pgoff and in do_brk.
2318 if (!vma
->vm_file
) {
2319 BUG_ON(vma
->anon_vma
);
2320 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2322 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2323 &prev
, &rb_link
, &rb_parent
))
2325 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2326 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2329 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2334 * Copy the vma structure to a new location in the same mm,
2335 * prior to moving page table entries, to effect an mremap move.
2337 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2338 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2340 struct vm_area_struct
*vma
= *vmap
;
2341 unsigned long vma_start
= vma
->vm_start
;
2342 struct mm_struct
*mm
= vma
->vm_mm
;
2343 struct vm_area_struct
*new_vma
, *prev
;
2344 struct rb_node
**rb_link
, *rb_parent
;
2345 struct mempolicy
*pol
;
2346 bool faulted_in_anon_vma
= true;
2349 * If anonymous vma has not yet been faulted, update new pgoff
2350 * to match new location, to increase its chance of merging.
2352 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2353 pgoff
= addr
>> PAGE_SHIFT
;
2354 faulted_in_anon_vma
= false;
2357 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2358 return NULL
; /* should never get here */
2359 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2360 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2363 * Source vma may have been merged into new_vma
2365 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2366 vma_start
< new_vma
->vm_end
)) {
2368 * The only way we can get a vma_merge with
2369 * self during an mremap is if the vma hasn't
2370 * been faulted in yet and we were allowed to
2371 * reset the dst vma->vm_pgoff to the
2372 * destination address of the mremap to allow
2373 * the merge to happen. mremap must change the
2374 * vm_pgoff linearity between src and dst vmas
2375 * (in turn preventing a vma_merge) to be
2376 * safe. It is only safe to keep the vm_pgoff
2377 * linear if there are no pages mapped yet.
2379 VM_BUG_ON(faulted_in_anon_vma
);
2382 anon_vma_moveto_tail(new_vma
);
2384 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2387 pol
= mpol_dup(vma_policy(vma
));
2390 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2391 if (anon_vma_clone(new_vma
, vma
))
2392 goto out_free_mempol
;
2393 vma_set_policy(new_vma
, pol
);
2394 new_vma
->vm_start
= addr
;
2395 new_vma
->vm_end
= addr
+ len
;
2396 new_vma
->vm_pgoff
= pgoff
;
2397 if (new_vma
->vm_file
)
2398 get_file(new_vma
->vm_file
);
2399 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2400 new_vma
->vm_ops
->open(new_vma
);
2401 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2409 kmem_cache_free(vm_area_cachep
, new_vma
);
2414 * Return true if the calling process may expand its vm space by the passed
2417 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2419 unsigned long cur
= mm
->total_vm
; /* pages */
2422 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2424 if (cur
+ npages
> lim
)
2430 static int special_mapping_fault(struct vm_area_struct
*vma
,
2431 struct vm_fault
*vmf
)
2434 struct page
**pages
;
2437 * special mappings have no vm_file, and in that case, the mm
2438 * uses vm_pgoff internally. So we have to subtract it from here.
2439 * We are allowed to do this because we are the mm; do not copy
2440 * this code into drivers!
2442 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2444 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2448 struct page
*page
= *pages
;
2454 return VM_FAULT_SIGBUS
;
2458 * Having a close hook prevents vma merging regardless of flags.
2460 static void special_mapping_close(struct vm_area_struct
*vma
)
2464 static const struct vm_operations_struct special_mapping_vmops
= {
2465 .close
= special_mapping_close
,
2466 .fault
= special_mapping_fault
,
2470 * Called with mm->mmap_sem held for writing.
2471 * Insert a new vma covering the given region, with the given flags.
2472 * Its pages are supplied by the given array of struct page *.
2473 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2474 * The region past the last page supplied will always produce SIGBUS.
2475 * The array pointer and the pages it points to are assumed to stay alive
2476 * for as long as this mapping might exist.
2478 int install_special_mapping(struct mm_struct
*mm
,
2479 unsigned long addr
, unsigned long len
,
2480 unsigned long vm_flags
, struct page
**pages
)
2483 struct vm_area_struct
*vma
;
2485 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2486 if (unlikely(vma
== NULL
))
2489 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2491 vma
->vm_start
= addr
;
2492 vma
->vm_end
= addr
+ len
;
2494 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2495 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2497 vma
->vm_ops
= &special_mapping_vmops
;
2498 vma
->vm_private_data
= pages
;
2500 ret
= insert_vm_struct(mm
, vma
);
2504 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2506 perf_event_mmap(vma
);
2511 kmem_cache_free(vm_area_cachep
, vma
);
2515 static DEFINE_MUTEX(mm_all_locks_mutex
);
2517 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2519 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2521 * The LSB of head.next can't change from under us
2522 * because we hold the mm_all_locks_mutex.
2524 mutex_lock_nest_lock(&anon_vma
->root
->mutex
, &mm
->mmap_sem
);
2526 * We can safely modify head.next after taking the
2527 * anon_vma->root->mutex. If some other vma in this mm shares
2528 * the same anon_vma we won't take it again.
2530 * No need of atomic instructions here, head.next
2531 * can't change from under us thanks to the
2532 * anon_vma->root->mutex.
2534 if (__test_and_set_bit(0, (unsigned long *)
2535 &anon_vma
->root
->head
.next
))
2540 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2542 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2544 * AS_MM_ALL_LOCKS can't change from under us because
2545 * we hold the mm_all_locks_mutex.
2547 * Operations on ->flags have to be atomic because
2548 * even if AS_MM_ALL_LOCKS is stable thanks to the
2549 * mm_all_locks_mutex, there may be other cpus
2550 * changing other bitflags in parallel to us.
2552 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2554 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
2559 * This operation locks against the VM for all pte/vma/mm related
2560 * operations that could ever happen on a certain mm. This includes
2561 * vmtruncate, try_to_unmap, and all page faults.
2563 * The caller must take the mmap_sem in write mode before calling
2564 * mm_take_all_locks(). The caller isn't allowed to release the
2565 * mmap_sem until mm_drop_all_locks() returns.
2567 * mmap_sem in write mode is required in order to block all operations
2568 * that could modify pagetables and free pages without need of
2569 * altering the vma layout (for example populate_range() with
2570 * nonlinear vmas). It's also needed in write mode to avoid new
2571 * anon_vmas to be associated with existing vmas.
2573 * A single task can't take more than one mm_take_all_locks() in a row
2574 * or it would deadlock.
2576 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2577 * mapping->flags avoid to take the same lock twice, if more than one
2578 * vma in this mm is backed by the same anon_vma or address_space.
2580 * We can take all the locks in random order because the VM code
2581 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2582 * takes more than one of them in a row. Secondly we're protected
2583 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2585 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2586 * that may have to take thousand of locks.
2588 * mm_take_all_locks() can fail if it's interrupted by signals.
2590 int mm_take_all_locks(struct mm_struct
*mm
)
2592 struct vm_area_struct
*vma
;
2593 struct anon_vma_chain
*avc
;
2595 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2597 mutex_lock(&mm_all_locks_mutex
);
2599 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2600 if (signal_pending(current
))
2602 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2603 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2606 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2607 if (signal_pending(current
))
2610 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2611 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2617 mm_drop_all_locks(mm
);
2621 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2623 if (test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2625 * The LSB of head.next can't change to 0 from under
2626 * us because we hold the mm_all_locks_mutex.
2628 * We must however clear the bitflag before unlocking
2629 * the vma so the users using the anon_vma->head will
2630 * never see our bitflag.
2632 * No need of atomic instructions here, head.next
2633 * can't change from under us until we release the
2634 * anon_vma->root->mutex.
2636 if (!__test_and_clear_bit(0, (unsigned long *)
2637 &anon_vma
->root
->head
.next
))
2639 anon_vma_unlock(anon_vma
);
2643 static void vm_unlock_mapping(struct address_space
*mapping
)
2645 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2647 * AS_MM_ALL_LOCKS can't change to 0 from under us
2648 * because we hold the mm_all_locks_mutex.
2650 mutex_unlock(&mapping
->i_mmap_mutex
);
2651 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2658 * The mmap_sem cannot be released by the caller until
2659 * mm_drop_all_locks() returns.
2661 void mm_drop_all_locks(struct mm_struct
*mm
)
2663 struct vm_area_struct
*vma
;
2664 struct anon_vma_chain
*avc
;
2666 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2667 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2669 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2671 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2672 vm_unlock_anon_vma(avc
->anon_vma
);
2673 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2674 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2677 mutex_unlock(&mm_all_locks_mutex
);
2681 * initialise the VMA slab
2683 void __init
mmap_init(void)
2687 ret
= percpu_counter_init(&vm_committed_as
, 0);