6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #include <linux/slab.h>
10 #include <linux/backing-dev.h>
12 #include <linux/shm.h>
13 #include <linux/mman.h>
14 #include <linux/pagemap.h>
15 #include <linux/swap.h>
16 #include <linux/syscalls.h>
17 #include <linux/capability.h>
18 #include <linux/init.h>
19 #include <linux/file.h>
21 #include <linux/personality.h>
22 #include <linux/security.h>
23 #include <linux/hugetlb.h>
24 #include <linux/profile.h>
25 #include <linux/export.h>
26 #include <linux/mount.h>
27 #include <linux/mempolicy.h>
28 #include <linux/rmap.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/perf_event.h>
31 #include <linux/audit.h>
32 #include <linux/khugepaged.h>
33 #include <linux/uprobes.h>
35 #include <asm/uaccess.h>
36 #include <asm/cacheflush.h>
38 #include <asm/mmu_context.h>
42 #ifndef arch_mmap_check
43 #define arch_mmap_check(addr, len, flags) (0)
46 #ifndef arch_rebalance_pgtables
47 #define arch_rebalance_pgtables(addr, len) (addr)
50 static void unmap_region(struct mm_struct
*mm
,
51 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
52 unsigned long start
, unsigned long end
);
55 * WARNING: the debugging will use recursive algorithms so never enable this
56 * unless you know what you are doing.
60 /* description of effects of mapping type and prot in current implementation.
61 * this is due to the limited x86 page protection hardware. The expected
62 * behavior is in parens:
65 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
66 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (yes) yes w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (copy) copy w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
75 pgprot_t protection_map
[16] = {
76 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
77 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
80 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
82 return __pgprot(pgprot_val(protection_map
[vm_flags
&
83 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
84 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
86 EXPORT_SYMBOL(vm_get_page_prot
);
88 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
89 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
90 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
93 * other variables. It can be updated by several CPUs frequently.
95 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
98 * Check that a process has enough memory to allocate a new virtual
99 * mapping. 0 means there is enough memory for the allocation to
100 * succeed and -ENOMEM implies there is not.
102 * We currently support three overcommit policies, which are set via the
103 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
106 * Additional code 2002 Jul 20 by Robert Love.
108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
110 * Note this is a helper function intended to be used by LSMs which
111 * wish to use this logic.
113 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
115 unsigned long free
, allowed
;
117 vm_acct_memory(pages
);
120 * Sometimes we want to use more memory than we have
122 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
125 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
126 free
= global_page_state(NR_FREE_PAGES
);
127 free
+= global_page_state(NR_FILE_PAGES
);
130 * shmem pages shouldn't be counted as free in this
131 * case, they can't be purged, only swapped out, and
132 * that won't affect the overall amount of available
133 * memory in the system.
135 free
-= global_page_state(NR_SHMEM
);
137 free
+= nr_swap_pages
;
140 * Any slabs which are created with the
141 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
142 * which are reclaimable, under pressure. The dentry
143 * cache and most inode caches should fall into this
145 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
148 * Leave reserved pages. The pages are not for anonymous pages.
150 if (free
<= totalreserve_pages
)
153 free
-= totalreserve_pages
;
156 * Leave the last 3% for root
167 allowed
= (totalram_pages
- hugetlb_total_pages())
168 * sysctl_overcommit_ratio
/ 100;
170 * Leave the last 3% for root
173 allowed
-= allowed
/ 32;
174 allowed
+= total_swap_pages
;
176 /* Don't let a single process grow too big:
177 leave 3% of the size of this process for other processes */
179 allowed
-= mm
->total_vm
/ 32;
181 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
184 vm_unacct_memory(pages
);
190 * Requires inode->i_mapping->i_mmap_mutex
192 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
193 struct file
*file
, struct address_space
*mapping
)
195 if (vma
->vm_flags
& VM_DENYWRITE
)
196 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
197 if (vma
->vm_flags
& VM_SHARED
)
198 mapping
->i_mmap_writable
--;
200 flush_dcache_mmap_lock(mapping
);
201 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
202 list_del_init(&vma
->shared
.vm_set
.list
);
204 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
205 flush_dcache_mmap_unlock(mapping
);
209 * Unlink a file-based vm structure from its prio_tree, to hide
210 * vma from rmap and vmtruncate before freeing its page tables.
212 void unlink_file_vma(struct vm_area_struct
*vma
)
214 struct file
*file
= vma
->vm_file
;
217 struct address_space
*mapping
= file
->f_mapping
;
218 mutex_lock(&mapping
->i_mmap_mutex
);
219 __remove_shared_vm_struct(vma
, file
, mapping
);
220 mutex_unlock(&mapping
->i_mmap_mutex
);
225 * Close a vm structure and free it, returning the next.
227 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
229 struct vm_area_struct
*next
= vma
->vm_next
;
232 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
233 vma
->vm_ops
->close(vma
);
236 if (vma
->vm_flags
& VM_EXECUTABLE
)
237 removed_exe_file_vma(vma
->vm_mm
);
239 mpol_put(vma_policy(vma
));
240 kmem_cache_free(vm_area_cachep
, vma
);
244 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
246 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
248 unsigned long rlim
, retval
;
249 unsigned long newbrk
, oldbrk
;
250 struct mm_struct
*mm
= current
->mm
;
251 unsigned long min_brk
;
253 down_write(&mm
->mmap_sem
);
255 #ifdef CONFIG_COMPAT_BRK
257 * CONFIG_COMPAT_BRK can still be overridden by setting
258 * randomize_va_space to 2, which will still cause mm->start_brk
259 * to be arbitrarily shifted
261 if (current
->brk_randomized
)
262 min_brk
= mm
->start_brk
;
264 min_brk
= mm
->end_data
;
266 min_brk
= mm
->start_brk
;
272 * Check against rlimit here. If this check is done later after the test
273 * of oldbrk with newbrk then it can escape the test and let the data
274 * segment grow beyond its set limit the in case where the limit is
275 * not page aligned -Ram Gupta
277 rlim
= rlimit(RLIMIT_DATA
);
278 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
279 (mm
->end_data
- mm
->start_data
) > rlim
)
282 newbrk
= PAGE_ALIGN(brk
);
283 oldbrk
= PAGE_ALIGN(mm
->brk
);
284 if (oldbrk
== newbrk
)
287 /* Always allow shrinking brk. */
288 if (brk
<= mm
->brk
) {
289 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
294 /* Check against existing mmap mappings. */
295 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
298 /* Ok, looks good - let it rip. */
299 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
305 up_write(&mm
->mmap_sem
);
310 static int browse_rb(struct rb_root
*root
)
313 struct rb_node
*nd
, *pn
= NULL
;
314 unsigned long prev
= 0, pend
= 0;
316 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
317 struct vm_area_struct
*vma
;
318 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
319 if (vma
->vm_start
< prev
)
320 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
321 if (vma
->vm_start
< pend
)
322 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
323 if (vma
->vm_start
> vma
->vm_end
)
324 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
327 prev
= vma
->vm_start
;
331 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
335 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
339 void validate_mm(struct mm_struct
*mm
)
343 struct vm_area_struct
*tmp
= mm
->mmap
;
348 if (i
!= mm
->map_count
)
349 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
350 i
= browse_rb(&mm
->mm_rb
);
351 if (i
!= mm
->map_count
)
352 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
356 #define validate_mm(mm) do { } while (0)
359 static struct vm_area_struct
*
360 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
361 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
362 struct rb_node
** rb_parent
)
364 struct vm_area_struct
* vma
;
365 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
367 __rb_link
= &mm
->mm_rb
.rb_node
;
368 rb_prev
= __rb_parent
= NULL
;
372 struct vm_area_struct
*vma_tmp
;
374 __rb_parent
= *__rb_link
;
375 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
377 if (vma_tmp
->vm_end
> addr
) {
379 if (vma_tmp
->vm_start
<= addr
)
381 __rb_link
= &__rb_parent
->rb_left
;
383 rb_prev
= __rb_parent
;
384 __rb_link
= &__rb_parent
->rb_right
;
390 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
391 *rb_link
= __rb_link
;
392 *rb_parent
= __rb_parent
;
396 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
397 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
399 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
400 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
403 static void __vma_link_file(struct vm_area_struct
*vma
)
409 struct address_space
*mapping
= file
->f_mapping
;
411 if (vma
->vm_flags
& VM_DENYWRITE
)
412 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
413 if (vma
->vm_flags
& VM_SHARED
)
414 mapping
->i_mmap_writable
++;
416 flush_dcache_mmap_lock(mapping
);
417 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
418 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
420 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
421 flush_dcache_mmap_unlock(mapping
);
426 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
427 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
428 struct rb_node
*rb_parent
)
430 __vma_link_list(mm
, vma
, prev
, rb_parent
);
431 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
434 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
435 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
436 struct rb_node
*rb_parent
)
438 struct address_space
*mapping
= NULL
;
441 mapping
= vma
->vm_file
->f_mapping
;
444 mutex_lock(&mapping
->i_mmap_mutex
);
446 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
447 __vma_link_file(vma
);
450 mutex_unlock(&mapping
->i_mmap_mutex
);
457 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
458 * mm's list and rbtree. It has already been inserted into the prio_tree.
460 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
462 struct vm_area_struct
*__vma
, *prev
;
463 struct rb_node
**rb_link
, *rb_parent
;
465 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
466 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
467 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
472 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
473 struct vm_area_struct
*prev
)
475 struct vm_area_struct
*next
= vma
->vm_next
;
477 prev
->vm_next
= next
;
479 next
->vm_prev
= prev
;
480 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
481 if (mm
->mmap_cache
== vma
)
482 mm
->mmap_cache
= prev
;
486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
487 * is already present in an i_mmap tree without adjusting the tree.
488 * The following helper function should be used when such adjustments
489 * are necessary. The "insert" vma (if any) is to be inserted
490 * before we drop the necessary locks.
492 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
493 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
495 struct mm_struct
*mm
= vma
->vm_mm
;
496 struct vm_area_struct
*next
= vma
->vm_next
;
497 struct vm_area_struct
*importer
= NULL
;
498 struct address_space
*mapping
= NULL
;
499 struct prio_tree_root
*root
= NULL
;
500 struct anon_vma
*anon_vma
= NULL
;
501 struct file
*file
= vma
->vm_file
;
502 long adjust_next
= 0;
505 if (next
&& !insert
) {
506 struct vm_area_struct
*exporter
= NULL
;
508 if (end
>= next
->vm_end
) {
510 * vma expands, overlapping all the next, and
511 * perhaps the one after too (mprotect case 6).
513 again
: remove_next
= 1 + (end
> next
->vm_end
);
517 } else if (end
> next
->vm_start
) {
519 * vma expands, overlapping part of the next:
520 * mprotect case 5 shifting the boundary up.
522 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
525 } else if (end
< vma
->vm_end
) {
527 * vma shrinks, and !insert tells it's not
528 * split_vma inserting another: so it must be
529 * mprotect case 4 shifting the boundary down.
531 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
537 * Easily overlooked: when mprotect shifts the boundary,
538 * make sure the expanding vma has anon_vma set if the
539 * shrinking vma had, to cover any anon pages imported.
541 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
542 if (anon_vma_clone(importer
, exporter
))
544 importer
->anon_vma
= exporter
->anon_vma
;
549 mapping
= file
->f_mapping
;
550 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
551 root
= &mapping
->i_mmap
;
552 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
555 uprobe_munmap(next
, next
->vm_start
,
559 mutex_lock(&mapping
->i_mmap_mutex
);
562 * Put into prio_tree now, so instantiated pages
563 * are visible to arm/parisc __flush_dcache_page
564 * throughout; but we cannot insert into address
565 * space until vma start or end is updated.
567 __vma_link_file(insert
);
571 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
574 * When changing only vma->vm_end, we don't really need anon_vma
575 * lock. This is a fairly rare case by itself, but the anon_vma
576 * lock may be shared between many sibling processes. Skipping
577 * the lock for brk adjustments makes a difference sometimes.
579 if (vma
->anon_vma
&& (importer
|| start
!= vma
->vm_start
)) {
580 anon_vma
= vma
->anon_vma
;
581 anon_vma_lock(anon_vma
);
585 flush_dcache_mmap_lock(mapping
);
586 vma_prio_tree_remove(vma
, root
);
588 vma_prio_tree_remove(next
, root
);
591 vma
->vm_start
= start
;
593 vma
->vm_pgoff
= pgoff
;
595 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
596 next
->vm_pgoff
+= adjust_next
;
601 vma_prio_tree_insert(next
, root
);
602 vma_prio_tree_insert(vma
, root
);
603 flush_dcache_mmap_unlock(mapping
);
608 * vma_merge has merged next into vma, and needs
609 * us to remove next before dropping the locks.
611 __vma_unlink(mm
, next
, vma
);
613 __remove_shared_vm_struct(next
, file
, mapping
);
616 * split_vma has split insert from vma, and needs
617 * us to insert it before dropping the locks
618 * (it may either follow vma or precede it).
620 __insert_vm_struct(mm
, insert
);
624 anon_vma_unlock(anon_vma
);
626 mutex_unlock(&mapping
->i_mmap_mutex
);
637 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
639 if (next
->vm_flags
& VM_EXECUTABLE
)
640 removed_exe_file_vma(mm
);
643 anon_vma_merge(vma
, next
);
645 mpol_put(vma_policy(next
));
646 kmem_cache_free(vm_area_cachep
, next
);
648 * In mprotect's case 6 (see comments on vma_merge),
649 * we must remove another next too. It would clutter
650 * up the code too much to do both in one go.
652 if (remove_next
== 2) {
666 * If the vma has a ->close operation then the driver probably needs to release
667 * per-vma resources, so we don't attempt to merge those.
669 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
670 struct file
*file
, unsigned long vm_flags
)
672 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
673 if ((vma
->vm_flags
^ vm_flags
) & ~VM_CAN_NONLINEAR
)
675 if (vma
->vm_file
!= file
)
677 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
682 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
683 struct anon_vma
*anon_vma2
,
684 struct vm_area_struct
*vma
)
687 * The list_is_singular() test is to avoid merging VMA cloned from
688 * parents. This can improve scalability caused by anon_vma lock.
690 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
691 list_is_singular(&vma
->anon_vma_chain
)))
693 return anon_vma1
== anon_vma2
;
697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
698 * in front of (at a lower virtual address and file offset than) the vma.
700 * We cannot merge two vmas if they have differently assigned (non-NULL)
701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
703 * We don't check here for the merged mmap wrapping around the end of pagecache
704 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
705 * wrap, nor mmaps which cover the final page at index -1UL.
708 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
709 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
711 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
712 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
713 if (vma
->vm_pgoff
== vm_pgoff
)
720 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
721 * beyond (at a higher virtual address and file offset than) the vma.
723 * We cannot merge two vmas if they have differently assigned (non-NULL)
724 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
727 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
728 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
730 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
731 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
733 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
734 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
741 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
742 * whether that can be merged with its predecessor or its successor.
743 * Or both (it neatly fills a hole).
745 * In most cases - when called for mmap, brk or mremap - [addr,end) is
746 * certain not to be mapped by the time vma_merge is called; but when
747 * called for mprotect, it is certain to be already mapped (either at
748 * an offset within prev, or at the start of next), and the flags of
749 * this area are about to be changed to vm_flags - and the no-change
750 * case has already been eliminated.
752 * The following mprotect cases have to be considered, where AAAA is
753 * the area passed down from mprotect_fixup, never extending beyond one
754 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
756 * AAAA AAAA AAAA AAAA
757 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
758 * cannot merge might become might become might become
759 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
760 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
761 * mremap move: PPPPNNNNNNNN 8
763 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
764 * might become case 1 below case 2 below case 3 below
766 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
767 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
769 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
770 struct vm_area_struct
*prev
, unsigned long addr
,
771 unsigned long end
, unsigned long vm_flags
,
772 struct anon_vma
*anon_vma
, struct file
*file
,
773 pgoff_t pgoff
, struct mempolicy
*policy
)
775 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
776 struct vm_area_struct
*area
, *next
;
780 * We later require that vma->vm_flags == vm_flags,
781 * so this tests vma->vm_flags & VM_SPECIAL, too.
783 if (vm_flags
& VM_SPECIAL
)
787 next
= prev
->vm_next
;
791 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
792 next
= next
->vm_next
;
795 * Can it merge with the predecessor?
797 if (prev
&& prev
->vm_end
== addr
&&
798 mpol_equal(vma_policy(prev
), policy
) &&
799 can_vma_merge_after(prev
, vm_flags
,
800 anon_vma
, file
, pgoff
)) {
802 * OK, it can. Can we now merge in the successor as well?
804 if (next
&& end
== next
->vm_start
&&
805 mpol_equal(policy
, vma_policy(next
)) &&
806 can_vma_merge_before(next
, vm_flags
,
807 anon_vma
, file
, pgoff
+pglen
) &&
808 is_mergeable_anon_vma(prev
->anon_vma
,
809 next
->anon_vma
, NULL
)) {
811 err
= vma_adjust(prev
, prev
->vm_start
,
812 next
->vm_end
, prev
->vm_pgoff
, NULL
);
813 } else /* cases 2, 5, 7 */
814 err
= vma_adjust(prev
, prev
->vm_start
,
815 end
, prev
->vm_pgoff
, NULL
);
818 khugepaged_enter_vma_merge(prev
);
823 * Can this new request be merged in front of next?
825 if (next
&& end
== next
->vm_start
&&
826 mpol_equal(policy
, vma_policy(next
)) &&
827 can_vma_merge_before(next
, vm_flags
,
828 anon_vma
, file
, pgoff
+pglen
)) {
829 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
830 err
= vma_adjust(prev
, prev
->vm_start
,
831 addr
, prev
->vm_pgoff
, NULL
);
832 else /* cases 3, 8 */
833 err
= vma_adjust(area
, addr
, next
->vm_end
,
834 next
->vm_pgoff
- pglen
, NULL
);
837 khugepaged_enter_vma_merge(area
);
845 * Rough compatbility check to quickly see if it's even worth looking
846 * at sharing an anon_vma.
848 * They need to have the same vm_file, and the flags can only differ
849 * in things that mprotect may change.
851 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
852 * we can merge the two vma's. For example, we refuse to merge a vma if
853 * there is a vm_ops->close() function, because that indicates that the
854 * driver is doing some kind of reference counting. But that doesn't
855 * really matter for the anon_vma sharing case.
857 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
859 return a
->vm_end
== b
->vm_start
&&
860 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
861 a
->vm_file
== b
->vm_file
&&
862 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
863 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
867 * Do some basic sanity checking to see if we can re-use the anon_vma
868 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
869 * the same as 'old', the other will be the new one that is trying
870 * to share the anon_vma.
872 * NOTE! This runs with mm_sem held for reading, so it is possible that
873 * the anon_vma of 'old' is concurrently in the process of being set up
874 * by another page fault trying to merge _that_. But that's ok: if it
875 * is being set up, that automatically means that it will be a singleton
876 * acceptable for merging, so we can do all of this optimistically. But
877 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
879 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
880 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
881 * is to return an anon_vma that is "complex" due to having gone through
884 * We also make sure that the two vma's are compatible (adjacent,
885 * and with the same memory policies). That's all stable, even with just
886 * a read lock on the mm_sem.
888 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
890 if (anon_vma_compatible(a
, b
)) {
891 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
893 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
900 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
901 * neighbouring vmas for a suitable anon_vma, before it goes off
902 * to allocate a new anon_vma. It checks because a repetitive
903 * sequence of mprotects and faults may otherwise lead to distinct
904 * anon_vmas being allocated, preventing vma merge in subsequent
907 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
909 struct anon_vma
*anon_vma
;
910 struct vm_area_struct
*near
;
916 anon_vma
= reusable_anon_vma(near
, vma
, near
);
924 anon_vma
= reusable_anon_vma(near
, near
, vma
);
929 * There's no absolute need to look only at touching neighbours:
930 * we could search further afield for "compatible" anon_vmas.
931 * But it would probably just be a waste of time searching,
932 * or lead to too many vmas hanging off the same anon_vma.
933 * We're trying to allow mprotect remerging later on,
934 * not trying to minimize memory used for anon_vmas.
939 #ifdef CONFIG_PROC_FS
940 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
941 struct file
*file
, long pages
)
943 const unsigned long stack_flags
944 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
947 mm
->shared_vm
+= pages
;
948 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
949 mm
->exec_vm
+= pages
;
950 } else if (flags
& stack_flags
)
951 mm
->stack_vm
+= pages
;
952 if (flags
& (VM_RESERVED
|VM_IO
))
953 mm
->reserved_vm
+= pages
;
955 #endif /* CONFIG_PROC_FS */
958 * If a hint addr is less than mmap_min_addr change hint to be as
959 * low as possible but still greater than mmap_min_addr
961 static inline unsigned long round_hint_to_min(unsigned long hint
)
964 if (((void *)hint
!= NULL
) &&
965 (hint
< mmap_min_addr
))
966 return PAGE_ALIGN(mmap_min_addr
);
971 * The caller must hold down_write(¤t->mm->mmap_sem).
974 static unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
975 unsigned long len
, unsigned long prot
,
976 unsigned long flags
, unsigned long pgoff
)
978 struct mm_struct
* mm
= current
->mm
;
984 * Does the application expect PROT_READ to imply PROT_EXEC?
986 * (the exception is when the underlying filesystem is noexec
987 * mounted, in which case we dont add PROT_EXEC.)
989 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
990 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
996 if (!(flags
& MAP_FIXED
))
997 addr
= round_hint_to_min(addr
);
999 /* Careful about overflows.. */
1000 len
= PAGE_ALIGN(len
);
1004 /* offset overflow? */
1005 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1008 /* Too many mappings? */
1009 if (mm
->map_count
> sysctl_max_map_count
)
1012 /* Obtain the address to map to. we verify (or select) it and ensure
1013 * that it represents a valid section of the address space.
1015 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1016 if (addr
& ~PAGE_MASK
)
1019 /* Do simple checking here so the lower-level routines won't have
1020 * to. we assume access permissions have been handled by the open
1021 * of the memory object, so we don't do any here.
1023 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1024 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1026 if (flags
& MAP_LOCKED
)
1027 if (!can_do_mlock())
1030 /* mlock MCL_FUTURE? */
1031 if (vm_flags
& VM_LOCKED
) {
1032 unsigned long locked
, lock_limit
;
1033 locked
= len
>> PAGE_SHIFT
;
1034 locked
+= mm
->locked_vm
;
1035 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1036 lock_limit
>>= PAGE_SHIFT
;
1037 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1041 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1044 switch (flags
& MAP_TYPE
) {
1046 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1050 * Make sure we don't allow writing to an append-only
1053 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1057 * Make sure there are no mandatory locks on the file.
1059 if (locks_verify_locked(inode
))
1062 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1063 if (!(file
->f_mode
& FMODE_WRITE
))
1064 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1068 if (!(file
->f_mode
& FMODE_READ
))
1070 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1071 if (vm_flags
& VM_EXEC
)
1073 vm_flags
&= ~VM_MAYEXEC
;
1076 if (!file
->f_op
|| !file
->f_op
->mmap
)
1084 switch (flags
& MAP_TYPE
) {
1090 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1094 * Set pgoff according to addr for anon_vma.
1096 pgoff
= addr
>> PAGE_SHIFT
;
1103 error
= security_mmap_addr(addr
);
1107 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1110 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1111 unsigned long len
, unsigned long prot
,
1112 unsigned long flag
, unsigned long offset
)
1114 if (unlikely(offset
+ PAGE_ALIGN(len
) < offset
))
1116 if (unlikely(offset
& ~PAGE_MASK
))
1118 return do_mmap_pgoff(file
, addr
, len
, prot
, flag
, offset
>> PAGE_SHIFT
);
1121 unsigned long vm_mmap(struct file
*file
, unsigned long addr
,
1122 unsigned long len
, unsigned long prot
,
1123 unsigned long flag
, unsigned long offset
)
1126 struct mm_struct
*mm
= current
->mm
;
1128 ret
= security_mmap_file(file
, prot
, flag
);
1130 down_write(&mm
->mmap_sem
);
1131 ret
= do_mmap(file
, addr
, len
, prot
, flag
, offset
);
1132 up_write(&mm
->mmap_sem
);
1136 EXPORT_SYMBOL(vm_mmap
);
1138 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1139 unsigned long, prot
, unsigned long, flags
,
1140 unsigned long, fd
, unsigned long, pgoff
)
1142 struct file
*file
= NULL
;
1143 unsigned long retval
= -EBADF
;
1145 if (!(flags
& MAP_ANONYMOUS
)) {
1146 audit_mmap_fd(fd
, flags
);
1147 if (unlikely(flags
& MAP_HUGETLB
))
1152 } else if (flags
& MAP_HUGETLB
) {
1153 struct user_struct
*user
= NULL
;
1155 * VM_NORESERVE is used because the reservations will be
1156 * taken when vm_ops->mmap() is called
1157 * A dummy user value is used because we are not locking
1158 * memory so no accounting is necessary
1160 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, addr
, len
,
1161 VM_NORESERVE
, &user
,
1162 HUGETLB_ANONHUGE_INODE
);
1164 return PTR_ERR(file
);
1167 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1169 retval
= security_mmap_file(file
, prot
, flags
);
1171 down_write(¤t
->mm
->mmap_sem
);
1172 retval
= do_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1173 up_write(¤t
->mm
->mmap_sem
);
1182 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1183 struct mmap_arg_struct
{
1187 unsigned long flags
;
1189 unsigned long offset
;
1192 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1194 struct mmap_arg_struct a
;
1196 if (copy_from_user(&a
, arg
, sizeof(a
)))
1198 if (a
.offset
& ~PAGE_MASK
)
1201 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1202 a
.offset
>> PAGE_SHIFT
);
1204 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1207 * Some shared mappigns will want the pages marked read-only
1208 * to track write events. If so, we'll downgrade vm_page_prot
1209 * to the private version (using protection_map[] without the
1212 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1214 vm_flags_t vm_flags
= vma
->vm_flags
;
1216 /* If it was private or non-writable, the write bit is already clear */
1217 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1220 /* The backer wishes to know when pages are first written to? */
1221 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1224 /* The open routine did something to the protections already? */
1225 if (pgprot_val(vma
->vm_page_prot
) !=
1226 pgprot_val(vm_get_page_prot(vm_flags
)))
1229 /* Specialty mapping? */
1230 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1233 /* Can the mapping track the dirty pages? */
1234 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1235 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1239 * We account for memory if it's a private writeable mapping,
1240 * not hugepages and VM_NORESERVE wasn't set.
1242 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1245 * hugetlb has its own accounting separate from the core VM
1246 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1248 if (file
&& is_file_hugepages(file
))
1251 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1254 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1255 unsigned long len
, unsigned long flags
,
1256 vm_flags_t vm_flags
, unsigned long pgoff
)
1258 struct mm_struct
*mm
= current
->mm
;
1259 struct vm_area_struct
*vma
, *prev
;
1260 int correct_wcount
= 0;
1262 struct rb_node
**rb_link
, *rb_parent
;
1263 unsigned long charged
= 0;
1264 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1266 /* Clear old maps */
1269 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1270 if (vma
&& vma
->vm_start
< addr
+ len
) {
1271 if (do_munmap(mm
, addr
, len
))
1276 /* Check against address space limit. */
1277 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1281 * Set 'VM_NORESERVE' if we should not account for the
1282 * memory use of this mapping.
1284 if ((flags
& MAP_NORESERVE
)) {
1285 /* We honor MAP_NORESERVE if allowed to overcommit */
1286 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1287 vm_flags
|= VM_NORESERVE
;
1289 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1290 if (file
&& is_file_hugepages(file
))
1291 vm_flags
|= VM_NORESERVE
;
1295 * Private writable mapping: check memory availability
1297 if (accountable_mapping(file
, vm_flags
)) {
1298 charged
= len
>> PAGE_SHIFT
;
1299 if (security_vm_enough_memory_mm(mm
, charged
))
1301 vm_flags
|= VM_ACCOUNT
;
1305 * Can we just expand an old mapping?
1307 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1312 * Determine the object being mapped and call the appropriate
1313 * specific mapper. the address has already been validated, but
1314 * not unmapped, but the maps are removed from the list.
1316 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1323 vma
->vm_start
= addr
;
1324 vma
->vm_end
= addr
+ len
;
1325 vma
->vm_flags
= vm_flags
;
1326 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1327 vma
->vm_pgoff
= pgoff
;
1328 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1330 error
= -EINVAL
; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1333 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1335 if (vm_flags
& VM_DENYWRITE
) {
1336 error
= deny_write_access(file
);
1341 vma
->vm_file
= file
;
1343 error
= file
->f_op
->mmap(file
, vma
);
1345 goto unmap_and_free_vma
;
1346 if (vm_flags
& VM_EXECUTABLE
)
1347 added_exe_file_vma(mm
);
1349 /* Can addr have changed??
1351 * Answer: Yes, several device drivers can do it in their
1352 * f_op->mmap method. -DaveM
1354 addr
= vma
->vm_start
;
1355 pgoff
= vma
->vm_pgoff
;
1356 vm_flags
= vma
->vm_flags
;
1357 } else if (vm_flags
& VM_SHARED
) {
1358 if (unlikely(vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
)))
1360 error
= shmem_zero_setup(vma
);
1365 if (vma_wants_writenotify(vma
)) {
1366 pgprot_t pprot
= vma
->vm_page_prot
;
1368 /* Can vma->vm_page_prot have changed??
1370 * Answer: Yes, drivers may have changed it in their
1371 * f_op->mmap method.
1373 * Ensures that vmas marked as uncached stay that way.
1375 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1376 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1377 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1380 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1381 file
= vma
->vm_file
;
1383 /* Once vma denies write, undo our temporary denial count */
1385 atomic_inc(&inode
->i_writecount
);
1387 perf_event_mmap(vma
);
1389 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1390 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1391 if (vm_flags
& VM_LOCKED
) {
1392 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1393 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1394 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1395 make_pages_present(addr
, addr
+ len
);
1397 if (file
&& uprobe_mmap(vma
))
1398 /* matching probes but cannot insert */
1399 goto unmap_and_free_vma
;
1405 atomic_inc(&inode
->i_writecount
);
1406 vma
->vm_file
= NULL
;
1409 /* Undo any partial mapping done by a device driver. */
1410 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1413 kmem_cache_free(vm_area_cachep
, vma
);
1416 vm_unacct_memory(charged
);
1420 /* Get an address range which is currently unmapped.
1421 * For shmat() with addr=0.
1423 * Ugly calling convention alert:
1424 * Return value with the low bits set means error value,
1426 * if (ret & ~PAGE_MASK)
1429 * This function "knows" that -ENOMEM has the bits set.
1431 #ifndef HAVE_ARCH_UNMAPPED_AREA
1433 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1434 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1436 struct mm_struct
*mm
= current
->mm
;
1437 struct vm_area_struct
*vma
;
1438 unsigned long start_addr
;
1440 if (len
> TASK_SIZE
)
1443 if (flags
& MAP_FIXED
)
1447 addr
= PAGE_ALIGN(addr
);
1448 vma
= find_vma(mm
, addr
);
1449 if (TASK_SIZE
- len
>= addr
&&
1450 (!vma
|| addr
+ len
<= vma
->vm_start
))
1453 if (len
> mm
->cached_hole_size
) {
1454 start_addr
= addr
= mm
->free_area_cache
;
1456 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1457 mm
->cached_hole_size
= 0;
1461 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1462 /* At this point: (!vma || addr < vma->vm_end). */
1463 if (TASK_SIZE
- len
< addr
) {
1465 * Start a new search - just in case we missed
1468 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1469 addr
= TASK_UNMAPPED_BASE
;
1471 mm
->cached_hole_size
= 0;
1476 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1478 * Remember the place where we stopped the search:
1480 mm
->free_area_cache
= addr
+ len
;
1483 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1484 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1490 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1493 * Is this a new hole at the lowest possible address?
1495 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1496 mm
->free_area_cache
= addr
;
1500 * This mmap-allocator allocates new areas top-down from below the
1501 * stack's low limit (the base):
1503 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1505 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1506 const unsigned long len
, const unsigned long pgoff
,
1507 const unsigned long flags
)
1509 struct vm_area_struct
*vma
;
1510 struct mm_struct
*mm
= current
->mm
;
1511 unsigned long addr
= addr0
, start_addr
;
1513 /* requested length too big for entire address space */
1514 if (len
> TASK_SIZE
)
1517 if (flags
& MAP_FIXED
)
1520 /* requesting a specific address */
1522 addr
= PAGE_ALIGN(addr
);
1523 vma
= find_vma(mm
, addr
);
1524 if (TASK_SIZE
- len
>= addr
&&
1525 (!vma
|| addr
+ len
<= vma
->vm_start
))
1529 /* check if free_area_cache is useful for us */
1530 if (len
<= mm
->cached_hole_size
) {
1531 mm
->cached_hole_size
= 0;
1532 mm
->free_area_cache
= mm
->mmap_base
;
1536 /* either no address requested or can't fit in requested address hole */
1537 start_addr
= addr
= mm
->free_area_cache
;
1545 * Lookup failure means no vma is above this address,
1546 * else if new region fits below vma->vm_start,
1547 * return with success:
1549 vma
= find_vma(mm
, addr
);
1550 if (!vma
|| addr
+len
<= vma
->vm_start
)
1551 /* remember the address as a hint for next time */
1552 return (mm
->free_area_cache
= addr
);
1554 /* remember the largest hole we saw so far */
1555 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1556 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1558 /* try just below the current vma->vm_start */
1559 addr
= vma
->vm_start
-len
;
1560 } while (len
< vma
->vm_start
);
1564 * if hint left us with no space for the requested
1565 * mapping then try again:
1567 * Note: this is different with the case of bottomup
1568 * which does the fully line-search, but we use find_vma
1569 * here that causes some holes skipped.
1571 if (start_addr
!= mm
->mmap_base
) {
1572 mm
->free_area_cache
= mm
->mmap_base
;
1573 mm
->cached_hole_size
= 0;
1578 * A failed mmap() very likely causes application failure,
1579 * so fall back to the bottom-up function here. This scenario
1580 * can happen with large stack limits and large mmap()
1583 mm
->cached_hole_size
= ~0UL;
1584 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1585 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1587 * Restore the topdown base:
1589 mm
->free_area_cache
= mm
->mmap_base
;
1590 mm
->cached_hole_size
= ~0UL;
1596 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1599 * Is this a new hole at the highest possible address?
1601 if (addr
> mm
->free_area_cache
)
1602 mm
->free_area_cache
= addr
;
1604 /* dont allow allocations above current base */
1605 if (mm
->free_area_cache
> mm
->mmap_base
)
1606 mm
->free_area_cache
= mm
->mmap_base
;
1610 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1611 unsigned long pgoff
, unsigned long flags
)
1613 unsigned long (*get_area
)(struct file
*, unsigned long,
1614 unsigned long, unsigned long, unsigned long);
1616 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1620 /* Careful about overflows.. */
1621 if (len
> TASK_SIZE
)
1624 get_area
= current
->mm
->get_unmapped_area
;
1625 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1626 get_area
= file
->f_op
->get_unmapped_area
;
1627 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1628 if (IS_ERR_VALUE(addr
))
1631 if (addr
> TASK_SIZE
- len
)
1633 if (addr
& ~PAGE_MASK
)
1636 return arch_rebalance_pgtables(addr
, len
);
1639 EXPORT_SYMBOL(get_unmapped_area
);
1641 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1642 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1644 struct vm_area_struct
*vma
= NULL
;
1647 /* Check the cache first. */
1648 /* (Cache hit rate is typically around 35%.) */
1649 vma
= mm
->mmap_cache
;
1650 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1651 struct rb_node
* rb_node
;
1653 rb_node
= mm
->mm_rb
.rb_node
;
1657 struct vm_area_struct
* vma_tmp
;
1659 vma_tmp
= rb_entry(rb_node
,
1660 struct vm_area_struct
, vm_rb
);
1662 if (vma_tmp
->vm_end
> addr
) {
1664 if (vma_tmp
->vm_start
<= addr
)
1666 rb_node
= rb_node
->rb_left
;
1668 rb_node
= rb_node
->rb_right
;
1671 mm
->mmap_cache
= vma
;
1677 EXPORT_SYMBOL(find_vma
);
1680 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1682 struct vm_area_struct
*
1683 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1684 struct vm_area_struct
**pprev
)
1686 struct vm_area_struct
*vma
;
1688 vma
= find_vma(mm
, addr
);
1690 *pprev
= vma
->vm_prev
;
1692 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
1695 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1696 rb_node
= rb_node
->rb_right
;
1703 * Verify that the stack growth is acceptable and
1704 * update accounting. This is shared with both the
1705 * grow-up and grow-down cases.
1707 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1709 struct mm_struct
*mm
= vma
->vm_mm
;
1710 struct rlimit
*rlim
= current
->signal
->rlim
;
1711 unsigned long new_start
;
1713 /* address space limit tests */
1714 if (!may_expand_vm(mm
, grow
))
1717 /* Stack limit test */
1718 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1721 /* mlock limit tests */
1722 if (vma
->vm_flags
& VM_LOCKED
) {
1723 unsigned long locked
;
1724 unsigned long limit
;
1725 locked
= mm
->locked_vm
+ grow
;
1726 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1727 limit
>>= PAGE_SHIFT
;
1728 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1732 /* Check to ensure the stack will not grow into a hugetlb-only region */
1733 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1735 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1739 * Overcommit.. This must be the final test, as it will
1740 * update security statistics.
1742 if (security_vm_enough_memory_mm(mm
, grow
))
1745 /* Ok, everything looks good - let it rip */
1746 mm
->total_vm
+= grow
;
1747 if (vma
->vm_flags
& VM_LOCKED
)
1748 mm
->locked_vm
+= grow
;
1749 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1753 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1755 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1756 * vma is the last one with address > vma->vm_end. Have to extend vma.
1758 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1762 if (!(vma
->vm_flags
& VM_GROWSUP
))
1766 * We must make sure the anon_vma is allocated
1767 * so that the anon_vma locking is not a noop.
1769 if (unlikely(anon_vma_prepare(vma
)))
1771 vma_lock_anon_vma(vma
);
1774 * vma->vm_start/vm_end cannot change under us because the caller
1775 * is required to hold the mmap_sem in read mode. We need the
1776 * anon_vma lock to serialize against concurrent expand_stacks.
1777 * Also guard against wrapping around to address 0.
1779 if (address
< PAGE_ALIGN(address
+4))
1780 address
= PAGE_ALIGN(address
+4);
1782 vma_unlock_anon_vma(vma
);
1787 /* Somebody else might have raced and expanded it already */
1788 if (address
> vma
->vm_end
) {
1789 unsigned long size
, grow
;
1791 size
= address
- vma
->vm_start
;
1792 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1795 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1796 error
= acct_stack_growth(vma
, size
, grow
);
1798 vma
->vm_end
= address
;
1799 perf_event_mmap(vma
);
1803 vma_unlock_anon_vma(vma
);
1804 khugepaged_enter_vma_merge(vma
);
1807 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1810 * vma is the first one with address < vma->vm_start. Have to extend vma.
1812 int expand_downwards(struct vm_area_struct
*vma
,
1813 unsigned long address
)
1818 * We must make sure the anon_vma is allocated
1819 * so that the anon_vma locking is not a noop.
1821 if (unlikely(anon_vma_prepare(vma
)))
1824 address
&= PAGE_MASK
;
1825 error
= security_mmap_addr(address
);
1829 vma_lock_anon_vma(vma
);
1832 * vma->vm_start/vm_end cannot change under us because the caller
1833 * is required to hold the mmap_sem in read mode. We need the
1834 * anon_vma lock to serialize against concurrent expand_stacks.
1837 /* Somebody else might have raced and expanded it already */
1838 if (address
< vma
->vm_start
) {
1839 unsigned long size
, grow
;
1841 size
= vma
->vm_end
- address
;
1842 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1845 if (grow
<= vma
->vm_pgoff
) {
1846 error
= acct_stack_growth(vma
, size
, grow
);
1848 vma
->vm_start
= address
;
1849 vma
->vm_pgoff
-= grow
;
1850 perf_event_mmap(vma
);
1854 vma_unlock_anon_vma(vma
);
1855 khugepaged_enter_vma_merge(vma
);
1859 #ifdef CONFIG_STACK_GROWSUP
1860 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1862 return expand_upwards(vma
, address
);
1865 struct vm_area_struct
*
1866 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1868 struct vm_area_struct
*vma
, *prev
;
1871 vma
= find_vma_prev(mm
, addr
, &prev
);
1872 if (vma
&& (vma
->vm_start
<= addr
))
1874 if (!prev
|| expand_stack(prev
, addr
))
1876 if (prev
->vm_flags
& VM_LOCKED
) {
1877 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1882 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1884 return expand_downwards(vma
, address
);
1887 struct vm_area_struct
*
1888 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1890 struct vm_area_struct
* vma
;
1891 unsigned long start
;
1894 vma
= find_vma(mm
,addr
);
1897 if (vma
->vm_start
<= addr
)
1899 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1901 start
= vma
->vm_start
;
1902 if (expand_stack(vma
, addr
))
1904 if (vma
->vm_flags
& VM_LOCKED
) {
1905 mlock_vma_pages_range(vma
, addr
, start
);
1912 * Ok - we have the memory areas we should free on the vma list,
1913 * so release them, and do the vma updates.
1915 * Called with the mm semaphore held.
1917 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1919 unsigned long nr_accounted
= 0;
1921 /* Update high watermark before we lower total_vm */
1922 update_hiwater_vm(mm
);
1924 long nrpages
= vma_pages(vma
);
1926 if (vma
->vm_flags
& VM_ACCOUNT
)
1927 nr_accounted
+= nrpages
;
1928 mm
->total_vm
-= nrpages
;
1929 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1930 vma
= remove_vma(vma
);
1932 vm_unacct_memory(nr_accounted
);
1937 * Get rid of page table information in the indicated region.
1939 * Called with the mm semaphore held.
1941 static void unmap_region(struct mm_struct
*mm
,
1942 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1943 unsigned long start
, unsigned long end
)
1945 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1946 struct mmu_gather tlb
;
1949 tlb_gather_mmu(&tlb
, mm
, 0);
1950 update_hiwater_rss(mm
);
1951 unmap_vmas(&tlb
, vma
, start
, end
);
1952 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1953 next
? next
->vm_start
: 0);
1954 tlb_finish_mmu(&tlb
, start
, end
);
1958 * Create a list of vma's touched by the unmap, removing them from the mm's
1959 * vma list as we go..
1962 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1963 struct vm_area_struct
*prev
, unsigned long end
)
1965 struct vm_area_struct
**insertion_point
;
1966 struct vm_area_struct
*tail_vma
= NULL
;
1969 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1970 vma
->vm_prev
= NULL
;
1972 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1976 } while (vma
&& vma
->vm_start
< end
);
1977 *insertion_point
= vma
;
1979 vma
->vm_prev
= prev
;
1980 tail_vma
->vm_next
= NULL
;
1981 if (mm
->unmap_area
== arch_unmap_area
)
1982 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1984 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1985 mm
->unmap_area(mm
, addr
);
1986 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1990 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1991 * munmap path where it doesn't make sense to fail.
1993 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1994 unsigned long addr
, int new_below
)
1996 struct mempolicy
*pol
;
1997 struct vm_area_struct
*new;
2000 if (is_vm_hugetlb_page(vma
) && (addr
&
2001 ~(huge_page_mask(hstate_vma(vma
)))))
2004 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2008 /* most fields are the same, copy all, and then fixup */
2011 INIT_LIST_HEAD(&new->anon_vma_chain
);
2016 new->vm_start
= addr
;
2017 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2020 pol
= mpol_dup(vma_policy(vma
));
2025 vma_set_policy(new, pol
);
2027 if (anon_vma_clone(new, vma
))
2031 get_file(new->vm_file
);
2032 if (vma
->vm_flags
& VM_EXECUTABLE
)
2033 added_exe_file_vma(mm
);
2036 if (new->vm_ops
&& new->vm_ops
->open
)
2037 new->vm_ops
->open(new);
2040 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2041 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2043 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2049 /* Clean everything up if vma_adjust failed. */
2050 if (new->vm_ops
&& new->vm_ops
->close
)
2051 new->vm_ops
->close(new);
2053 if (vma
->vm_flags
& VM_EXECUTABLE
)
2054 removed_exe_file_vma(mm
);
2057 unlink_anon_vmas(new);
2061 kmem_cache_free(vm_area_cachep
, new);
2067 * Split a vma into two pieces at address 'addr', a new vma is allocated
2068 * either for the first part or the tail.
2070 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2071 unsigned long addr
, int new_below
)
2073 if (mm
->map_count
>= sysctl_max_map_count
)
2076 return __split_vma(mm
, vma
, addr
, new_below
);
2079 /* Munmap is split into 2 main parts -- this part which finds
2080 * what needs doing, and the areas themselves, which do the
2081 * work. This now handles partial unmappings.
2082 * Jeremy Fitzhardinge <jeremy@goop.org>
2084 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2087 struct vm_area_struct
*vma
, *prev
, *last
;
2089 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2092 if ((len
= PAGE_ALIGN(len
)) == 0)
2095 /* Find the first overlapping VMA */
2096 vma
= find_vma(mm
, start
);
2099 prev
= vma
->vm_prev
;
2100 /* we have start < vma->vm_end */
2102 /* if it doesn't overlap, we have nothing.. */
2104 if (vma
->vm_start
>= end
)
2108 * If we need to split any vma, do it now to save pain later.
2110 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2111 * unmapped vm_area_struct will remain in use: so lower split_vma
2112 * places tmp vma above, and higher split_vma places tmp vma below.
2114 if (start
> vma
->vm_start
) {
2118 * Make sure that map_count on return from munmap() will
2119 * not exceed its limit; but let map_count go just above
2120 * its limit temporarily, to help free resources as expected.
2122 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2125 error
= __split_vma(mm
, vma
, start
, 0);
2131 /* Does it split the last one? */
2132 last
= find_vma(mm
, end
);
2133 if (last
&& end
> last
->vm_start
) {
2134 int error
= __split_vma(mm
, last
, end
, 1);
2138 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2141 * unlock any mlock()ed ranges before detaching vmas
2143 if (mm
->locked_vm
) {
2144 struct vm_area_struct
*tmp
= vma
;
2145 while (tmp
&& tmp
->vm_start
< end
) {
2146 if (tmp
->vm_flags
& VM_LOCKED
) {
2147 mm
->locked_vm
-= vma_pages(tmp
);
2148 munlock_vma_pages_all(tmp
);
2155 * Remove the vma's, and unmap the actual pages
2157 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2158 unmap_region(mm
, vma
, prev
, start
, end
);
2160 /* Fix up all other VM information */
2161 remove_vma_list(mm
, vma
);
2165 EXPORT_SYMBOL(do_munmap
);
2167 int vm_munmap(unsigned long start
, size_t len
)
2170 struct mm_struct
*mm
= current
->mm
;
2172 down_write(&mm
->mmap_sem
);
2173 ret
= do_munmap(mm
, start
, len
);
2174 up_write(&mm
->mmap_sem
);
2177 EXPORT_SYMBOL(vm_munmap
);
2179 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2181 profile_munmap(addr
);
2182 return vm_munmap(addr
, len
);
2185 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2187 #ifdef CONFIG_DEBUG_VM
2188 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2190 up_read(&mm
->mmap_sem
);
2196 * this is really a simplified "do_mmap". it only handles
2197 * anonymous maps. eventually we may be able to do some
2198 * brk-specific accounting here.
2200 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2202 struct mm_struct
* mm
= current
->mm
;
2203 struct vm_area_struct
* vma
, * prev
;
2204 unsigned long flags
;
2205 struct rb_node
** rb_link
, * rb_parent
;
2206 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2209 len
= PAGE_ALIGN(len
);
2213 error
= security_mmap_addr(addr
);
2217 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2219 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2220 if (error
& ~PAGE_MASK
)
2226 if (mm
->def_flags
& VM_LOCKED
) {
2227 unsigned long locked
, lock_limit
;
2228 locked
= len
>> PAGE_SHIFT
;
2229 locked
+= mm
->locked_vm
;
2230 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2231 lock_limit
>>= PAGE_SHIFT
;
2232 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2237 * mm->mmap_sem is required to protect against another thread
2238 * changing the mappings in case we sleep.
2240 verify_mm_writelocked(mm
);
2243 * Clear old maps. this also does some error checking for us
2246 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2247 if (vma
&& vma
->vm_start
< addr
+ len
) {
2248 if (do_munmap(mm
, addr
, len
))
2253 /* Check against address space limits *after* clearing old maps... */
2254 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2257 if (mm
->map_count
> sysctl_max_map_count
)
2260 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2263 /* Can we just expand an old private anonymous mapping? */
2264 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2265 NULL
, NULL
, pgoff
, NULL
);
2270 * create a vma struct for an anonymous mapping
2272 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2274 vm_unacct_memory(len
>> PAGE_SHIFT
);
2278 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2280 vma
->vm_start
= addr
;
2281 vma
->vm_end
= addr
+ len
;
2282 vma
->vm_pgoff
= pgoff
;
2283 vma
->vm_flags
= flags
;
2284 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2285 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2287 perf_event_mmap(vma
);
2288 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2289 if (flags
& VM_LOCKED
) {
2290 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2291 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2296 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2298 struct mm_struct
*mm
= current
->mm
;
2301 down_write(&mm
->mmap_sem
);
2302 ret
= do_brk(addr
, len
);
2303 up_write(&mm
->mmap_sem
);
2306 EXPORT_SYMBOL(vm_brk
);
2308 /* Release all mmaps. */
2309 void exit_mmap(struct mm_struct
*mm
)
2311 struct mmu_gather tlb
;
2312 struct vm_area_struct
*vma
;
2313 unsigned long nr_accounted
= 0;
2315 /* mm's last user has gone, and its about to be pulled down */
2316 mmu_notifier_release(mm
);
2318 if (mm
->locked_vm
) {
2321 if (vma
->vm_flags
& VM_LOCKED
)
2322 munlock_vma_pages_all(vma
);
2330 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2335 tlb_gather_mmu(&tlb
, mm
, 1);
2336 /* update_hiwater_rss(mm) here? but nobody should be looking */
2337 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2338 unmap_vmas(&tlb
, vma
, 0, -1);
2340 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2341 tlb_finish_mmu(&tlb
, 0, -1);
2344 * Walk the list again, actually closing and freeing it,
2345 * with preemption enabled, without holding any MM locks.
2348 if (vma
->vm_flags
& VM_ACCOUNT
)
2349 nr_accounted
+= vma_pages(vma
);
2350 vma
= remove_vma(vma
);
2352 vm_unacct_memory(nr_accounted
);
2354 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2357 /* Insert vm structure into process list sorted by address
2358 * and into the inode's i_mmap tree. If vm_file is non-NULL
2359 * then i_mmap_mutex is taken here.
2361 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2363 struct vm_area_struct
* __vma
, * prev
;
2364 struct rb_node
** rb_link
, * rb_parent
;
2367 * The vm_pgoff of a purely anonymous vma should be irrelevant
2368 * until its first write fault, when page's anon_vma and index
2369 * are set. But now set the vm_pgoff it will almost certainly
2370 * end up with (unless mremap moves it elsewhere before that
2371 * first wfault), so /proc/pid/maps tells a consistent story.
2373 * By setting it to reflect the virtual start address of the
2374 * vma, merges and splits can happen in a seamless way, just
2375 * using the existing file pgoff checks and manipulations.
2376 * Similarly in do_mmap_pgoff and in do_brk.
2378 if (!vma
->vm_file
) {
2379 BUG_ON(vma
->anon_vma
);
2380 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2382 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2383 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2385 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2386 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2389 if (vma
->vm_file
&& uprobe_mmap(vma
))
2392 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2397 * Copy the vma structure to a new location in the same mm,
2398 * prior to moving page table entries, to effect an mremap move.
2400 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2401 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2403 struct vm_area_struct
*vma
= *vmap
;
2404 unsigned long vma_start
= vma
->vm_start
;
2405 struct mm_struct
*mm
= vma
->vm_mm
;
2406 struct vm_area_struct
*new_vma
, *prev
;
2407 struct rb_node
**rb_link
, *rb_parent
;
2408 struct mempolicy
*pol
;
2409 bool faulted_in_anon_vma
= true;
2412 * If anonymous vma has not yet been faulted, update new pgoff
2413 * to match new location, to increase its chance of merging.
2415 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2416 pgoff
= addr
>> PAGE_SHIFT
;
2417 faulted_in_anon_vma
= false;
2420 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2421 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2422 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2425 * Source vma may have been merged into new_vma
2427 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2428 vma_start
< new_vma
->vm_end
)) {
2430 * The only way we can get a vma_merge with
2431 * self during an mremap is if the vma hasn't
2432 * been faulted in yet and we were allowed to
2433 * reset the dst vma->vm_pgoff to the
2434 * destination address of the mremap to allow
2435 * the merge to happen. mremap must change the
2436 * vm_pgoff linearity between src and dst vmas
2437 * (in turn preventing a vma_merge) to be
2438 * safe. It is only safe to keep the vm_pgoff
2439 * linear if there are no pages mapped yet.
2441 VM_BUG_ON(faulted_in_anon_vma
);
2444 anon_vma_moveto_tail(new_vma
);
2446 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2449 pol
= mpol_dup(vma_policy(vma
));
2452 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2453 if (anon_vma_clone(new_vma
, vma
))
2454 goto out_free_mempol
;
2455 vma_set_policy(new_vma
, pol
);
2456 new_vma
->vm_start
= addr
;
2457 new_vma
->vm_end
= addr
+ len
;
2458 new_vma
->vm_pgoff
= pgoff
;
2459 if (new_vma
->vm_file
) {
2460 get_file(new_vma
->vm_file
);
2462 if (uprobe_mmap(new_vma
))
2463 goto out_free_mempol
;
2465 if (vma
->vm_flags
& VM_EXECUTABLE
)
2466 added_exe_file_vma(mm
);
2468 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2469 new_vma
->vm_ops
->open(new_vma
);
2470 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2478 kmem_cache_free(vm_area_cachep
, new_vma
);
2483 * Return true if the calling process may expand its vm space by the passed
2486 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2488 unsigned long cur
= mm
->total_vm
; /* pages */
2491 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2493 if (cur
+ npages
> lim
)
2499 static int special_mapping_fault(struct vm_area_struct
*vma
,
2500 struct vm_fault
*vmf
)
2503 struct page
**pages
;
2506 * special mappings have no vm_file, and in that case, the mm
2507 * uses vm_pgoff internally. So we have to subtract it from here.
2508 * We are allowed to do this because we are the mm; do not copy
2509 * this code into drivers!
2511 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2513 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2517 struct page
*page
= *pages
;
2523 return VM_FAULT_SIGBUS
;
2527 * Having a close hook prevents vma merging regardless of flags.
2529 static void special_mapping_close(struct vm_area_struct
*vma
)
2533 static const struct vm_operations_struct special_mapping_vmops
= {
2534 .close
= special_mapping_close
,
2535 .fault
= special_mapping_fault
,
2539 * Called with mm->mmap_sem held for writing.
2540 * Insert a new vma covering the given region, with the given flags.
2541 * Its pages are supplied by the given array of struct page *.
2542 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2543 * The region past the last page supplied will always produce SIGBUS.
2544 * The array pointer and the pages it points to are assumed to stay alive
2545 * for as long as this mapping might exist.
2547 int install_special_mapping(struct mm_struct
*mm
,
2548 unsigned long addr
, unsigned long len
,
2549 unsigned long vm_flags
, struct page
**pages
)
2552 struct vm_area_struct
*vma
;
2554 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2555 if (unlikely(vma
== NULL
))
2558 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2560 vma
->vm_start
= addr
;
2561 vma
->vm_end
= addr
+ len
;
2563 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2564 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2566 vma
->vm_ops
= &special_mapping_vmops
;
2567 vma
->vm_private_data
= pages
;
2569 ret
= security_mmap_addr(vma
->vm_start
);
2573 ret
= insert_vm_struct(mm
, vma
);
2577 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2579 perf_event_mmap(vma
);
2584 kmem_cache_free(vm_area_cachep
, vma
);
2588 static DEFINE_MUTEX(mm_all_locks_mutex
);
2590 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2592 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2594 * The LSB of head.next can't change from under us
2595 * because we hold the mm_all_locks_mutex.
2597 mutex_lock_nest_lock(&anon_vma
->root
->mutex
, &mm
->mmap_sem
);
2599 * We can safely modify head.next after taking the
2600 * anon_vma->root->mutex. If some other vma in this mm shares
2601 * the same anon_vma we won't take it again.
2603 * No need of atomic instructions here, head.next
2604 * can't change from under us thanks to the
2605 * anon_vma->root->mutex.
2607 if (__test_and_set_bit(0, (unsigned long *)
2608 &anon_vma
->root
->head
.next
))
2613 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2615 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2617 * AS_MM_ALL_LOCKS can't change from under us because
2618 * we hold the mm_all_locks_mutex.
2620 * Operations on ->flags have to be atomic because
2621 * even if AS_MM_ALL_LOCKS is stable thanks to the
2622 * mm_all_locks_mutex, there may be other cpus
2623 * changing other bitflags in parallel to us.
2625 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2627 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
2632 * This operation locks against the VM for all pte/vma/mm related
2633 * operations that could ever happen on a certain mm. This includes
2634 * vmtruncate, try_to_unmap, and all page faults.
2636 * The caller must take the mmap_sem in write mode before calling
2637 * mm_take_all_locks(). The caller isn't allowed to release the
2638 * mmap_sem until mm_drop_all_locks() returns.
2640 * mmap_sem in write mode is required in order to block all operations
2641 * that could modify pagetables and free pages without need of
2642 * altering the vma layout (for example populate_range() with
2643 * nonlinear vmas). It's also needed in write mode to avoid new
2644 * anon_vmas to be associated with existing vmas.
2646 * A single task can't take more than one mm_take_all_locks() in a row
2647 * or it would deadlock.
2649 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2650 * mapping->flags avoid to take the same lock twice, if more than one
2651 * vma in this mm is backed by the same anon_vma or address_space.
2653 * We can take all the locks in random order because the VM code
2654 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2655 * takes more than one of them in a row. Secondly we're protected
2656 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2658 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2659 * that may have to take thousand of locks.
2661 * mm_take_all_locks() can fail if it's interrupted by signals.
2663 int mm_take_all_locks(struct mm_struct
*mm
)
2665 struct vm_area_struct
*vma
;
2666 struct anon_vma_chain
*avc
;
2668 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2670 mutex_lock(&mm_all_locks_mutex
);
2672 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2673 if (signal_pending(current
))
2675 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2676 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2679 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2680 if (signal_pending(current
))
2683 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2684 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2690 mm_drop_all_locks(mm
);
2694 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2696 if (test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2698 * The LSB of head.next can't change to 0 from under
2699 * us because we hold the mm_all_locks_mutex.
2701 * We must however clear the bitflag before unlocking
2702 * the vma so the users using the anon_vma->head will
2703 * never see our bitflag.
2705 * No need of atomic instructions here, head.next
2706 * can't change from under us until we release the
2707 * anon_vma->root->mutex.
2709 if (!__test_and_clear_bit(0, (unsigned long *)
2710 &anon_vma
->root
->head
.next
))
2712 anon_vma_unlock(anon_vma
);
2716 static void vm_unlock_mapping(struct address_space
*mapping
)
2718 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2720 * AS_MM_ALL_LOCKS can't change to 0 from under us
2721 * because we hold the mm_all_locks_mutex.
2723 mutex_unlock(&mapping
->i_mmap_mutex
);
2724 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2731 * The mmap_sem cannot be released by the caller until
2732 * mm_drop_all_locks() returns.
2734 void mm_drop_all_locks(struct mm_struct
*mm
)
2736 struct vm_area_struct
*vma
;
2737 struct anon_vma_chain
*avc
;
2739 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2740 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2742 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2744 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2745 vm_unlock_anon_vma(avc
->anon_vma
);
2746 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2747 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2750 mutex_unlock(&mm_all_locks_mutex
);
2754 * initialise the VMA slab
2756 void __init
mmap_init(void)
2760 ret
= percpu_counter_init(&vm_committed_as
, 0);