6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #include <linux/kernel.h>
10 #include <linux/slab.h>
11 #include <linux/backing-dev.h>
13 #include <linux/vmacache.h>
14 #include <linux/shm.h>
15 #include <linux/mman.h>
16 #include <linux/pagemap.h>
17 #include <linux/swap.h>
18 #include <linux/syscalls.h>
19 #include <linux/capability.h>
20 #include <linux/init.h>
21 #include <linux/file.h>
23 #include <linux/personality.h>
24 #include <linux/security.h>
25 #include <linux/hugetlb.h>
26 #include <linux/profile.h>
27 #include <linux/export.h>
28 #include <linux/mount.h>
29 #include <linux/mempolicy.h>
30 #include <linux/rmap.h>
31 #include <linux/mmu_notifier.h>
32 #include <linux/perf_event.h>
33 #include <linux/audit.h>
34 #include <linux/khugepaged.h>
35 #include <linux/uprobes.h>
36 #include <linux/rbtree_augmented.h>
37 #include <linux/sched/sysctl.h>
38 #include <linux/notifier.h>
39 #include <linux/memory.h>
41 #include <asm/uaccess.h>
42 #include <asm/cacheflush.h>
44 #include <asm/mmu_context.h>
47 #ifdef CONFIG_SDCARD_FS
48 #include "../fs/sdcardfs/sdcardfs.h"
51 #ifndef arch_mmap_check
52 #define arch_mmap_check(addr, len, flags) (0)
55 #ifndef arch_rebalance_pgtables
56 #define arch_rebalance_pgtables(addr, len) (addr)
59 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
60 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
61 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
62 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
64 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
65 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
66 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
67 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
71 static void unmap_region(struct mm_struct
*mm
,
72 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
73 unsigned long start
, unsigned long end
);
75 /* description of effects of mapping type and prot in current implementation.
76 * this is due to the limited x86 page protection hardware. The expected
77 * behavior is in parens:
80 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
81 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
82 * w: (no) no w: (no) no w: (yes) yes w: (no) no
83 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
85 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
86 * w: (no) no w: (no) no w: (copy) copy w: (no) no
87 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
90 pgprot_t protection_map
[16] = {
91 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
92 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
95 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
97 return __pgprot(pgprot_val(protection_map
[vm_flags
&
98 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
99 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
101 EXPORT_SYMBOL(vm_get_page_prot
);
103 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
104 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
105 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
106 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
107 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
109 * Make sure vm_committed_as in one cacheline and not cacheline shared with
110 * other variables. It can be updated by several CPUs frequently.
112 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
115 * The global memory commitment made in the system can be a metric
116 * that can be used to drive ballooning decisions when Linux is hosted
117 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
118 * balancing memory across competing virtual machines that are hosted.
119 * Several metrics drive this policy engine including the guest reported
122 unsigned long vm_memory_committed(void)
124 return percpu_counter_read_positive(&vm_committed_as
);
126 EXPORT_SYMBOL_GPL(vm_memory_committed
);
129 * Check that a process has enough memory to allocate a new virtual
130 * mapping. 0 means there is enough memory for the allocation to
131 * succeed and -ENOMEM implies there is not.
133 * We currently support three overcommit policies, which are set via the
134 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
136 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
137 * Additional code 2002 Jul 20 by Robert Love.
139 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
141 * Note this is a helper function intended to be used by LSMs which
142 * wish to use this logic.
144 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
146 long free
, allowed
, reserve
;
148 vm_acct_memory(pages
);
151 * Sometimes we want to use more memory than we have
153 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
156 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
157 free
= global_page_state(NR_FREE_PAGES
);
158 free
+= global_page_state(NR_FILE_PAGES
);
161 * shmem pages shouldn't be counted as free in this
162 * case, they can't be purged, only swapped out, and
163 * that won't affect the overall amount of available
164 * memory in the system.
166 free
-= global_page_state(NR_SHMEM
);
168 free
+= get_nr_swap_pages();
171 * Any slabs which are created with the
172 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
173 * which are reclaimable, under pressure. The dentry
174 * cache and most inode caches should fall into this
176 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
179 * Leave reserved pages. The pages are not for anonymous pages.
181 if (free
<= totalreserve_pages
)
184 free
-= totalreserve_pages
;
187 * Reserve some for root
190 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
198 allowed
= (totalram_pages
- hugetlb_total_pages())
199 * sysctl_overcommit_ratio
/ 100;
201 * Reserve some for root
204 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
205 allowed
+= total_swap_pages
;
208 * Don't let a single process grow so big a user can't recover
211 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
212 allowed
-= min_t(long, mm
->total_vm
/ 32, reserve
);
215 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
218 vm_unacct_memory(pages
);
224 * Requires inode->i_mapping->i_mmap_mutex
226 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
227 struct file
*file
, struct address_space
*mapping
)
229 if (vma
->vm_flags
& VM_DENYWRITE
)
230 atomic_inc(&file_inode(file
)->i_writecount
);
231 if (vma
->vm_flags
& VM_SHARED
)
232 mapping_unmap_writable(mapping
);
234 flush_dcache_mmap_lock(mapping
);
235 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
236 list_del_init(&vma
->shared
.nonlinear
);
238 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
239 flush_dcache_mmap_unlock(mapping
);
243 * Unlink a file-based vm structure from its interval tree, to hide
244 * vma from rmap and vmtruncate before freeing its page tables.
246 void unlink_file_vma(struct vm_area_struct
*vma
)
248 struct file
*file
= vma
->vm_file
;
251 struct address_space
*mapping
= file
->f_mapping
;
252 mutex_lock(&mapping
->i_mmap_mutex
);
253 __remove_shared_vm_struct(vma
, file
, mapping
);
254 mutex_unlock(&mapping
->i_mmap_mutex
);
259 * Close a vm structure and free it, returning the next.
261 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
263 struct vm_area_struct
*next
= vma
->vm_next
;
266 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
267 vma
->vm_ops
->close(vma
);
270 mpol_put(vma_policy(vma
));
271 kmem_cache_free(vm_area_cachep
, vma
);
275 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
277 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
279 unsigned long rlim
, retval
;
280 unsigned long newbrk
, oldbrk
;
281 struct mm_struct
*mm
= current
->mm
;
282 struct vm_area_struct
*next
;
283 unsigned long min_brk
;
286 down_write(&mm
->mmap_sem
);
288 #ifdef CONFIG_COMPAT_BRK
290 * CONFIG_COMPAT_BRK can still be overridden by setting
291 * randomize_va_space to 2, which will still cause mm->start_brk
292 * to be arbitrarily shifted
294 if (current
->brk_randomized
)
295 min_brk
= mm
->start_brk
;
297 min_brk
= mm
->end_data
;
299 min_brk
= mm
->start_brk
;
305 * Check against rlimit here. If this check is done later after the test
306 * of oldbrk with newbrk then it can escape the test and let the data
307 * segment grow beyond its set limit the in case where the limit is
308 * not page aligned -Ram Gupta
310 rlim
= rlimit(RLIMIT_DATA
);
311 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
312 (mm
->end_data
- mm
->start_data
) > rlim
)
315 newbrk
= PAGE_ALIGN(brk
);
316 oldbrk
= PAGE_ALIGN(mm
->brk
);
317 if (oldbrk
== newbrk
)
320 /* Always allow shrinking brk. */
321 if (brk
<= mm
->brk
) {
322 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
327 /* Check against existing mmap mappings. */
328 next
= find_vma(mm
, oldbrk
);
329 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
332 /* Ok, looks good - let it rip. */
333 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
338 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
339 up_write(&mm
->mmap_sem
);
341 mm_populate(oldbrk
, newbrk
- oldbrk
);
346 up_write(&mm
->mmap_sem
);
350 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
352 unsigned long max
, prev_end
, subtree_gap
;
355 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
356 * allow two stack_guard_gaps between them here, and when choosing
357 * an unmapped area; whereas when expanding we only require one.
358 * That's a little inconsistent, but keeps the code here simpler.
360 max
= vm_start_gap(vma
);
362 prev_end
= vm_end_gap(vma
->vm_prev
);
368 if (vma
->vm_rb
.rb_left
) {
369 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
370 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
371 if (subtree_gap
> max
)
374 if (vma
->vm_rb
.rb_right
) {
375 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
376 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
377 if (subtree_gap
> max
)
383 #ifdef CONFIG_DEBUG_VM_RB
384 static int browse_rb(struct rb_root
*root
)
386 int i
= 0, j
, bug
= 0;
387 struct rb_node
*nd
, *pn
= NULL
;
388 unsigned long prev
= 0, pend
= 0;
390 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
391 struct vm_area_struct
*vma
;
392 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
393 if (vma
->vm_start
< prev
) {
394 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
);
397 if (vma
->vm_start
< pend
) {
398 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
401 if (vma
->vm_start
> vma
->vm_end
) {
402 printk("vm_end %lx < vm_start %lx\n",
403 vma
->vm_end
, vma
->vm_start
);
406 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
407 printk("free gap %lx, correct %lx\n",
409 vma_compute_subtree_gap(vma
));
414 prev
= vma
->vm_start
;
418 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
421 printk("backwards %d, forwards %d\n", j
, i
);
427 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
431 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
432 struct vm_area_struct
*vma
;
433 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
434 BUG_ON(vma
!= ignore
&&
435 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
));
439 void validate_mm(struct mm_struct
*mm
)
443 unsigned long highest_address
= 0;
444 struct vm_area_struct
*vma
= mm
->mmap
;
446 struct anon_vma_chain
*avc
;
447 vma_lock_anon_vma(vma
);
448 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
449 anon_vma_interval_tree_verify(avc
);
450 vma_unlock_anon_vma(vma
);
451 highest_address
= vm_end_gap(vma
);
455 if (i
!= mm
->map_count
) {
456 printk("map_count %d vm_next %d\n", mm
->map_count
, i
);
459 if (highest_address
!= mm
->highest_vm_end
) {
460 printk("mm->highest_vm_end %lx, found %lx\n",
461 mm
->highest_vm_end
, highest_address
);
464 i
= browse_rb(&mm
->mm_rb
);
465 if (i
!= mm
->map_count
) {
466 printk("map_count %d rb %d\n", mm
->map_count
, i
);
472 #define validate_mm_rb(root, ignore) do { } while (0)
473 #define validate_mm(mm) do { } while (0)
476 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
477 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
480 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
481 * vma->vm_prev->vm_end values changed, without modifying the vma's position
484 static void vma_gap_update(struct vm_area_struct
*vma
)
487 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
488 * function that does exacltly what we want.
490 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
493 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
494 struct rb_root
*root
)
496 /* All rb_subtree_gap values must be consistent prior to insertion */
497 validate_mm_rb(root
, NULL
);
499 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
502 static void vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
505 * All rb_subtree_gap values must be consistent prior to erase,
506 * with the possible exception of the vma being erased.
508 validate_mm_rb(root
, vma
);
511 * Note rb_erase_augmented is a fairly large inline function,
512 * so make sure we instantiate it only once with our desired
513 * augmented rbtree callbacks.
515 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
519 * vma has some anon_vma assigned, and is already inserted on that
520 * anon_vma's interval trees.
522 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
523 * vma must be removed from the anon_vma's interval trees using
524 * anon_vma_interval_tree_pre_update_vma().
526 * After the update, the vma will be reinserted using
527 * anon_vma_interval_tree_post_update_vma().
529 * The entire update must be protected by exclusive mmap_sem and by
530 * the root anon_vma's mutex.
533 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
535 struct anon_vma_chain
*avc
;
537 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
538 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
542 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
544 struct anon_vma_chain
*avc
;
546 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
547 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
550 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
551 unsigned long end
, struct vm_area_struct
**pprev
,
552 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
554 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
556 __rb_link
= &mm
->mm_rb
.rb_node
;
557 rb_prev
= __rb_parent
= NULL
;
560 struct vm_area_struct
*vma_tmp
;
562 __rb_parent
= *__rb_link
;
563 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
565 if (vma_tmp
->vm_end
> addr
) {
566 /* Fail if an existing vma overlaps the area */
567 if (vma_tmp
->vm_start
< end
)
569 __rb_link
= &__rb_parent
->rb_left
;
571 rb_prev
= __rb_parent
;
572 __rb_link
= &__rb_parent
->rb_right
;
578 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
579 *rb_link
= __rb_link
;
580 *rb_parent
= __rb_parent
;
584 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
585 unsigned long addr
, unsigned long end
)
587 unsigned long nr_pages
= 0;
588 struct vm_area_struct
*vma
;
590 /* Find first overlaping mapping */
591 vma
= find_vma_intersection(mm
, addr
, end
);
595 nr_pages
= (min(end
, vma
->vm_end
) -
596 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
598 /* Iterate over the rest of the overlaps */
599 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
600 unsigned long overlap_len
;
602 if (vma
->vm_start
> end
)
605 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
606 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
612 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
613 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
615 /* Update tracking information for the gap following the new vma. */
617 vma_gap_update(vma
->vm_next
);
619 mm
->highest_vm_end
= vm_end_gap(vma
);
622 * vma->vm_prev wasn't known when we followed the rbtree to find the
623 * correct insertion point for that vma. As a result, we could not
624 * update the vma vm_rb parents rb_subtree_gap values on the way down.
625 * So, we first insert the vma with a zero rb_subtree_gap value
626 * (to be consistent with what we did on the way down), and then
627 * immediately update the gap to the correct value. Finally we
628 * rebalance the rbtree after all augmented values have been set.
630 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
631 vma
->rb_subtree_gap
= 0;
633 vma_rb_insert(vma
, &mm
->mm_rb
);
636 static void __vma_link_file(struct vm_area_struct
*vma
)
642 struct address_space
*mapping
= file
->f_mapping
;
644 if (vma
->vm_flags
& VM_DENYWRITE
)
645 atomic_dec(&file_inode(file
)->i_writecount
);
646 if (vma
->vm_flags
& VM_SHARED
)
647 atomic_inc(&mapping
->i_mmap_writable
);
649 flush_dcache_mmap_lock(mapping
);
650 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
651 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
653 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
654 flush_dcache_mmap_unlock(mapping
);
659 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
660 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
661 struct rb_node
*rb_parent
)
663 __vma_link_list(mm
, vma
, prev
, rb_parent
);
664 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
667 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
668 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
669 struct rb_node
*rb_parent
)
671 struct address_space
*mapping
= NULL
;
674 mapping
= vma
->vm_file
->f_mapping
;
677 mutex_lock(&mapping
->i_mmap_mutex
);
679 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
680 __vma_link_file(vma
);
683 mutex_unlock(&mapping
->i_mmap_mutex
);
690 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
691 * mm's list and rbtree. It has already been inserted into the interval tree.
693 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
695 struct vm_area_struct
*prev
;
696 struct rb_node
**rb_link
, *rb_parent
;
698 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
699 &prev
, &rb_link
, &rb_parent
))
701 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
706 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
707 struct vm_area_struct
*prev
)
709 struct vm_area_struct
*next
;
711 vma_rb_erase(vma
, &mm
->mm_rb
);
712 prev
->vm_next
= next
= vma
->vm_next
;
714 next
->vm_prev
= prev
;
717 vmacache_invalidate(mm
);
721 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
722 * is already present in an i_mmap tree without adjusting the tree.
723 * The following helper function should be used when such adjustments
724 * are necessary. The "insert" vma (if any) is to be inserted
725 * before we drop the necessary locks.
727 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
728 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
730 struct mm_struct
*mm
= vma
->vm_mm
;
731 struct vm_area_struct
*next
= vma
->vm_next
;
732 struct vm_area_struct
*importer
= NULL
;
733 struct address_space
*mapping
= NULL
;
734 struct rb_root
*root
= NULL
;
735 struct anon_vma
*anon_vma
= NULL
;
736 struct file
*file
= vma
->vm_file
;
737 bool start_changed
= false, end_changed
= false;
738 long adjust_next
= 0;
741 if (next
&& !insert
) {
742 struct vm_area_struct
*exporter
= NULL
;
744 if (end
>= next
->vm_end
) {
746 * vma expands, overlapping all the next, and
747 * perhaps the one after too (mprotect case 6).
749 again
: remove_next
= 1 + (end
> next
->vm_end
);
753 } else if (end
> next
->vm_start
) {
755 * vma expands, overlapping part of the next:
756 * mprotect case 5 shifting the boundary up.
758 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
761 } else if (end
< vma
->vm_end
) {
763 * vma shrinks, and !insert tells it's not
764 * split_vma inserting another: so it must be
765 * mprotect case 4 shifting the boundary down.
767 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
773 * Easily overlooked: when mprotect shifts the boundary,
774 * make sure the expanding vma has anon_vma set if the
775 * shrinking vma had, to cover any anon pages imported.
777 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
778 if (anon_vma_clone(importer
, exporter
))
780 importer
->anon_vma
= exporter
->anon_vma
;
785 mapping
= file
->f_mapping
;
786 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
787 root
= &mapping
->i_mmap
;
788 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
791 uprobe_munmap(next
, next
->vm_start
,
795 mutex_lock(&mapping
->i_mmap_mutex
);
798 * Put into interval tree now, so instantiated pages
799 * are visible to arm/parisc __flush_dcache_page
800 * throughout; but we cannot insert into address
801 * space until vma start or end is updated.
803 __vma_link_file(insert
);
807 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
809 anon_vma
= vma
->anon_vma
;
810 if (!anon_vma
&& adjust_next
)
811 anon_vma
= next
->anon_vma
;
813 VM_BUG_ON(adjust_next
&& next
->anon_vma
&&
814 anon_vma
!= next
->anon_vma
);
815 anon_vma_lock_write(anon_vma
);
816 anon_vma_interval_tree_pre_update_vma(vma
);
818 anon_vma_interval_tree_pre_update_vma(next
);
822 flush_dcache_mmap_lock(mapping
);
823 vma_interval_tree_remove(vma
, root
);
825 vma_interval_tree_remove(next
, root
);
828 if (start
!= vma
->vm_start
) {
829 vma
->vm_start
= start
;
830 start_changed
= true;
832 if (end
!= vma
->vm_end
) {
836 vma
->vm_pgoff
= pgoff
;
838 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
839 next
->vm_pgoff
+= adjust_next
;
844 vma_interval_tree_insert(next
, root
);
845 vma_interval_tree_insert(vma
, root
);
846 flush_dcache_mmap_unlock(mapping
);
851 * vma_merge has merged next into vma, and needs
852 * us to remove next before dropping the locks.
854 __vma_unlink(mm
, next
, vma
);
856 __remove_shared_vm_struct(next
, file
, mapping
);
859 * split_vma has split insert from vma, and needs
860 * us to insert it before dropping the locks
861 * (it may either follow vma or precede it).
863 __insert_vm_struct(mm
, insert
);
869 mm
->highest_vm_end
= vm_end_gap(vma
);
870 else if (!adjust_next
)
871 vma_gap_update(next
);
876 anon_vma_interval_tree_post_update_vma(vma
);
878 anon_vma_interval_tree_post_update_vma(next
);
879 anon_vma_unlock_write(anon_vma
);
882 mutex_unlock(&mapping
->i_mmap_mutex
);
893 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
897 anon_vma_merge(vma
, next
);
899 mpol_put(vma_policy(next
));
900 kmem_cache_free(vm_area_cachep
, next
);
902 * In mprotect's case 6 (see comments on vma_merge),
903 * we must remove another next too. It would clutter
904 * up the code too much to do both in one go.
907 if (remove_next
== 2)
910 vma_gap_update(next
);
912 WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
923 * If the vma has a ->close operation then the driver probably needs to release
924 * per-vma resources, so we don't attempt to merge those.
926 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
927 struct file
*file
, unsigned long vm_flags
,
928 const char __user
*anon_name
)
930 if (vma
->vm_flags
^ vm_flags
)
932 if (vma
->vm_file
!= file
)
934 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
936 if (vma_get_anon_name(vma
) != anon_name
)
941 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
942 struct anon_vma
*anon_vma2
,
943 struct vm_area_struct
*vma
)
946 * The list_is_singular() test is to avoid merging VMA cloned from
947 * parents. This can improve scalability caused by anon_vma lock.
949 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
950 list_is_singular(&vma
->anon_vma_chain
)))
952 return anon_vma1
== anon_vma2
;
956 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
957 * in front of (at a lower virtual address and file offset than) the vma.
959 * We cannot merge two vmas if they have differently assigned (non-NULL)
960 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
962 * We don't check here for the merged mmap wrapping around the end of pagecache
963 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
964 * wrap, nor mmaps which cover the final page at index -1UL.
967 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
968 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
,
969 const char __user
*anon_name
)
971 if (is_mergeable_vma(vma
, file
, vm_flags
, anon_name
) &&
972 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
973 if (vma
->vm_pgoff
== vm_pgoff
)
980 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
981 * beyond (at a higher virtual address and file offset than) the vma.
983 * We cannot merge two vmas if they have differently assigned (non-NULL)
984 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
987 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
988 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
,
989 const char __user
*anon_name
)
991 if (is_mergeable_vma(vma
, file
, vm_flags
, anon_name
) &&
992 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
994 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
995 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1002 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
1003 * figure out whether that can be merged with its predecessor or its
1004 * successor. Or both (it neatly fills a hole).
1006 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1007 * certain not to be mapped by the time vma_merge is called; but when
1008 * called for mprotect, it is certain to be already mapped (either at
1009 * an offset within prev, or at the start of next), and the flags of
1010 * this area are about to be changed to vm_flags - and the no-change
1011 * case has already been eliminated.
1013 * The following mprotect cases have to be considered, where AAAA is
1014 * the area passed down from mprotect_fixup, never extending beyond one
1015 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1017 * AAAA AAAA AAAA AAAA
1018 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1019 * cannot merge might become might become might become
1020 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1021 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1022 * mremap move: PPPPNNNNNNNN 8
1024 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1025 * might become case 1 below case 2 below case 3 below
1027 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1028 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1030 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1031 struct vm_area_struct
*prev
, unsigned long addr
,
1032 unsigned long end
, unsigned long vm_flags
,
1033 struct anon_vma
*anon_vma
, struct file
*file
,
1034 pgoff_t pgoff
, struct mempolicy
*policy
,
1035 const char __user
*anon_name
)
1037 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1038 struct vm_area_struct
*area
, *next
;
1042 * We later require that vma->vm_flags == vm_flags,
1043 * so this tests vma->vm_flags & VM_SPECIAL, too.
1045 if (vm_flags
& VM_SPECIAL
)
1049 next
= prev
->vm_next
;
1053 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
1054 next
= next
->vm_next
;
1057 * Can it merge with the predecessor?
1059 if (prev
&& prev
->vm_end
== addr
&&
1060 mpol_equal(vma_policy(prev
), policy
) &&
1061 can_vma_merge_after(prev
, vm_flags
, anon_vma
,
1062 file
, pgoff
, anon_name
)) {
1064 * OK, it can. Can we now merge in the successor as well?
1066 if (next
&& end
== next
->vm_start
&&
1067 mpol_equal(policy
, vma_policy(next
)) &&
1068 can_vma_merge_before(next
, vm_flags
, anon_vma
,
1069 file
, pgoff
+pglen
, anon_name
) &&
1070 is_mergeable_anon_vma(prev
->anon_vma
,
1071 next
->anon_vma
, NULL
)) {
1073 err
= vma_adjust(prev
, prev
->vm_start
,
1074 next
->vm_end
, prev
->vm_pgoff
, NULL
);
1075 } else /* cases 2, 5, 7 */
1076 err
= vma_adjust(prev
, prev
->vm_start
,
1077 end
, prev
->vm_pgoff
, NULL
);
1080 khugepaged_enter_vma_merge(prev
);
1085 * Can this new request be merged in front of next?
1087 if (next
&& end
== next
->vm_start
&&
1088 mpol_equal(policy
, vma_policy(next
)) &&
1089 can_vma_merge_before(next
, vm_flags
, anon_vma
,
1090 file
, pgoff
+pglen
, anon_name
)) {
1091 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1092 err
= vma_adjust(prev
, prev
->vm_start
,
1093 addr
, prev
->vm_pgoff
, NULL
);
1094 else /* cases 3, 8 */
1095 err
= vma_adjust(area
, addr
, next
->vm_end
,
1096 next
->vm_pgoff
- pglen
, NULL
);
1099 khugepaged_enter_vma_merge(area
);
1107 * Rough compatbility check to quickly see if it's even worth looking
1108 * at sharing an anon_vma.
1110 * They need to have the same vm_file, and the flags can only differ
1111 * in things that mprotect may change.
1113 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1114 * we can merge the two vma's. For example, we refuse to merge a vma if
1115 * there is a vm_ops->close() function, because that indicates that the
1116 * driver is doing some kind of reference counting. But that doesn't
1117 * really matter for the anon_vma sharing case.
1119 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1121 return a
->vm_end
== b
->vm_start
&&
1122 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1123 a
->vm_file
== b
->vm_file
&&
1124 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
1125 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1129 * Do some basic sanity checking to see if we can re-use the anon_vma
1130 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1131 * the same as 'old', the other will be the new one that is trying
1132 * to share the anon_vma.
1134 * NOTE! This runs with mm_sem held for reading, so it is possible that
1135 * the anon_vma of 'old' is concurrently in the process of being set up
1136 * by another page fault trying to merge _that_. But that's ok: if it
1137 * is being set up, that automatically means that it will be a singleton
1138 * acceptable for merging, so we can do all of this optimistically. But
1139 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1141 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1142 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1143 * is to return an anon_vma that is "complex" due to having gone through
1146 * We also make sure that the two vma's are compatible (adjacent,
1147 * and with the same memory policies). That's all stable, even with just
1148 * a read lock on the mm_sem.
1150 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1152 if (anon_vma_compatible(a
, b
)) {
1153 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
1155 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1162 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1163 * neighbouring vmas for a suitable anon_vma, before it goes off
1164 * to allocate a new anon_vma. It checks because a repetitive
1165 * sequence of mprotects and faults may otherwise lead to distinct
1166 * anon_vmas being allocated, preventing vma merge in subsequent
1169 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1171 struct anon_vma
*anon_vma
;
1172 struct vm_area_struct
*near
;
1174 near
= vma
->vm_next
;
1178 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1182 near
= vma
->vm_prev
;
1186 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1191 * There's no absolute need to look only at touching neighbours:
1192 * we could search further afield for "compatible" anon_vmas.
1193 * But it would probably just be a waste of time searching,
1194 * or lead to too many vmas hanging off the same anon_vma.
1195 * We're trying to allow mprotect remerging later on,
1196 * not trying to minimize memory used for anon_vmas.
1201 #ifdef CONFIG_PROC_FS
1202 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
1203 struct file
*file
, long pages
)
1205 const unsigned long stack_flags
1206 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
1208 mm
->total_vm
+= pages
;
1211 mm
->shared_vm
+= pages
;
1212 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
1213 mm
->exec_vm
+= pages
;
1214 } else if (flags
& stack_flags
)
1215 mm
->stack_vm
+= pages
;
1217 #endif /* CONFIG_PROC_FS */
1220 * If a hint addr is less than mmap_min_addr change hint to be as
1221 * low as possible but still greater than mmap_min_addr
1223 static inline unsigned long round_hint_to_min(unsigned long hint
)
1226 if (((void *)hint
!= NULL
) &&
1227 (hint
< mmap_min_addr
))
1228 return PAGE_ALIGN(mmap_min_addr
);
1233 * The caller must hold down_write(¤t->mm->mmap_sem).
1236 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
1237 unsigned long len
, unsigned long prot
,
1238 unsigned long flags
, unsigned long pgoff
,
1239 unsigned long *populate
)
1241 struct mm_struct
* mm
= current
->mm
;
1242 vm_flags_t vm_flags
;
1246 while (file
&& (file
->f_mode
& FMODE_NONMAPPABLE
))
1247 file
= file
->f_op
->get_lower_file(file
);
1250 * Does the application expect PROT_READ to imply PROT_EXEC?
1252 * (the exception is when the underlying filesystem is noexec
1253 * mounted, in which case we dont add PROT_EXEC.)
1255 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1256 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
1262 if (!(flags
& MAP_FIXED
))
1263 addr
= round_hint_to_min(addr
);
1265 /* Careful about overflows.. */
1266 len
= PAGE_ALIGN(len
);
1270 /* offset overflow? */
1271 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1274 /* Too many mappings? */
1275 if (mm
->map_count
> sysctl_max_map_count
)
1278 /* Obtain the address to map to. we verify (or select) it and ensure
1279 * that it represents a valid section of the address space.
1281 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1282 if (addr
& ~PAGE_MASK
)
1285 /* Do simple checking here so the lower-level routines won't have
1286 * to. we assume access permissions have been handled by the open
1287 * of the memory object, so we don't do any here.
1289 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1290 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1292 if (flags
& MAP_LOCKED
)
1293 if (!can_do_mlock())
1296 /* mlock MCL_FUTURE? */
1297 if (vm_flags
& VM_LOCKED
) {
1298 unsigned long locked
, lock_limit
;
1299 locked
= len
>> PAGE_SHIFT
;
1300 locked
+= mm
->locked_vm
;
1301 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1302 lock_limit
>>= PAGE_SHIFT
;
1303 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1308 struct inode
*inode
= file_inode(file
);
1310 switch (flags
& MAP_TYPE
) {
1312 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1316 * Make sure we don't allow writing to an append-only
1319 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1323 * Make sure there are no mandatory locks on the file.
1325 if (locks_verify_locked(inode
))
1328 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1329 if (!(file
->f_mode
& FMODE_WRITE
))
1330 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1334 if (!(file
->f_mode
& FMODE_READ
))
1336 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1337 if (vm_flags
& VM_EXEC
)
1339 vm_flags
&= ~VM_MAYEXEC
;
1342 if (!file
->f_op
|| !file
->f_op
->mmap
)
1344 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1352 switch (flags
& MAP_TYPE
) {
1354 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1360 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1364 * Set pgoff according to addr for anon_vma.
1366 pgoff
= addr
>> PAGE_SHIFT
;
1374 * Set 'VM_NORESERVE' if we should not account for the
1375 * memory use of this mapping.
1377 if (flags
& MAP_NORESERVE
) {
1378 /* We honor MAP_NORESERVE if allowed to overcommit */
1379 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1380 vm_flags
|= VM_NORESERVE
;
1382 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1383 if (file
&& is_file_hugepages(file
))
1384 vm_flags
|= VM_NORESERVE
;
1387 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
);
1388 if (!IS_ERR_VALUE(addr
) &&
1389 ((vm_flags
& VM_LOCKED
) ||
1390 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1395 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1396 unsigned long, prot
, unsigned long, flags
,
1397 unsigned long, fd
, unsigned long, pgoff
)
1399 struct file
*file
= NULL
;
1400 unsigned long retval
= -EBADF
;
1402 if (!(flags
& MAP_ANONYMOUS
)) {
1403 audit_mmap_fd(fd
, flags
);
1404 if (unlikely(flags
& MAP_HUGETLB
))
1409 if (is_file_hugepages(file
))
1410 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1411 } else if (flags
& MAP_HUGETLB
) {
1412 struct user_struct
*user
= NULL
;
1413 struct hstate
*hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) &
1419 len
= ALIGN(len
, huge_page_size(hs
));
1421 * VM_NORESERVE is used because the reservations will be
1422 * taken when vm_ops->mmap() is called
1423 * A dummy user value is used because we are not locking
1424 * memory so no accounting is necessary
1426 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1428 &user
, HUGETLB_ANONHUGE_INODE
,
1429 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1431 return PTR_ERR(file
);
1434 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1436 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1443 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1444 struct mmap_arg_struct
{
1448 unsigned long flags
;
1450 unsigned long offset
;
1453 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1455 struct mmap_arg_struct a
;
1457 if (copy_from_user(&a
, arg
, sizeof(a
)))
1459 if (a
.offset
& ~PAGE_MASK
)
1462 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1463 a
.offset
>> PAGE_SHIFT
);
1465 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1468 * Some shared mappigns will want the pages marked read-only
1469 * to track write events. If so, we'll downgrade vm_page_prot
1470 * to the private version (using protection_map[] without the
1473 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1475 vm_flags_t vm_flags
= vma
->vm_flags
;
1477 /* If it was private or non-writable, the write bit is already clear */
1478 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1481 /* The backer wishes to know when pages are first written to? */
1482 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1485 /* The open routine did something to the protections already? */
1486 if (pgprot_val(vma
->vm_page_prot
) !=
1487 pgprot_val(vm_get_page_prot(vm_flags
)))
1490 /* Specialty mapping? */
1491 if (vm_flags
& VM_PFNMAP
)
1494 /* Can the mapping track the dirty pages? */
1495 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1496 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1500 * We account for memory if it's a private writeable mapping,
1501 * not hugepages and VM_NORESERVE wasn't set.
1503 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1506 * hugetlb has its own accounting separate from the core VM
1507 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1509 if (file
&& is_file_hugepages(file
))
1512 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1515 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1516 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
)
1518 struct mm_struct
*mm
= current
->mm
;
1519 struct vm_area_struct
*vma
, *prev
;
1521 struct rb_node
**rb_link
, *rb_parent
;
1522 unsigned long charged
= 0;
1524 /* Check against address space limit. */
1525 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
)) {
1526 unsigned long nr_pages
;
1529 * MAP_FIXED may remove pages of mappings that intersects with
1530 * requested mapping. Account for the pages it would unmap.
1532 if (!(vm_flags
& MAP_FIXED
))
1535 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1537 if (!may_expand_vm(mm
, (len
>> PAGE_SHIFT
) - nr_pages
))
1541 /* Clear old maps */
1544 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
1545 if (do_munmap(mm
, addr
, len
))
1551 * Private writable mapping: check memory availability
1553 if (accountable_mapping(file
, vm_flags
)) {
1554 charged
= len
>> PAGE_SHIFT
;
1555 if (security_vm_enough_memory_mm(mm
, charged
))
1557 vm_flags
|= VM_ACCOUNT
;
1561 * Can we just expand an old mapping?
1563 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
,
1569 * Determine the object being mapped and call the appropriate
1570 * specific mapper. the address has already been validated, but
1571 * not unmapped, but the maps are removed from the list.
1573 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1580 vma
->vm_start
= addr
;
1581 vma
->vm_end
= addr
+ len
;
1582 vma
->vm_flags
= vm_flags
;
1583 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1584 vma
->vm_pgoff
= pgoff
;
1585 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1588 if (vm_flags
& VM_DENYWRITE
) {
1589 error
= deny_write_access(file
);
1593 if (vm_flags
& VM_SHARED
) {
1594 error
= mapping_map_writable(file
->f_mapping
);
1596 goto allow_write_and_free_vma
;
1599 /* ->mmap() can change vma->vm_file, but must guarantee that
1600 * vma_link() below can deny write-access if VM_DENYWRITE is set
1601 * and map writably if VM_SHARED is set. This usually means the
1602 * new file must not have been exposed to user-space, yet.
1604 vma
->vm_file
= get_file(file
);
1605 error
= file
->f_op
->mmap(file
, vma
);
1607 goto unmap_and_free_vma
;
1609 /* Can addr have changed??
1611 * Answer: Yes, several device drivers can do it in their
1612 * f_op->mmap method. -DaveM
1613 * Bug: If addr is changed, prev, rb_link, rb_parent should
1614 * be updated for vma_link()
1616 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1618 addr
= vma
->vm_start
;
1619 pgoff
= vma
->vm_pgoff
;
1620 vm_flags
= vma
->vm_flags
;
1621 } else if (vm_flags
& VM_SHARED
) {
1622 error
= shmem_zero_setup(vma
);
1627 if (vma_wants_writenotify(vma
)) {
1628 pgprot_t pprot
= vma
->vm_page_prot
;
1630 /* Can vma->vm_page_prot have changed??
1632 * Answer: Yes, drivers may have changed it in their
1633 * f_op->mmap method.
1635 * Ensures that vmas marked as uncached stay that way.
1637 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1638 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1639 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1642 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1643 /* Once vma denies write, undo our temporary denial count */
1645 if (vm_flags
& VM_SHARED
)
1646 mapping_unmap_writable(file
->f_mapping
);
1647 if (vm_flags
& VM_DENYWRITE
)
1648 allow_write_access(file
);
1650 file
= vma
->vm_file
;
1652 perf_event_mmap(vma
);
1654 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1655 if (vm_flags
& VM_LOCKED
) {
1656 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1657 vma
== get_gate_vma(current
->mm
)))
1658 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1660 vma
->vm_flags
&= ~VM_LOCKED
;
1669 vma
->vm_file
= NULL
;
1672 /* Undo any partial mapping done by a device driver. */
1673 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1675 if (vm_flags
& VM_SHARED
)
1676 mapping_unmap_writable(file
->f_mapping
);
1677 allow_write_and_free_vma
:
1678 if (vm_flags
& VM_DENYWRITE
)
1679 allow_write_access(file
);
1681 kmem_cache_free(vm_area_cachep
, vma
);
1684 vm_unacct_memory(charged
);
1688 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1691 * We implement the search by looking for an rbtree node that
1692 * immediately follows a suitable gap. That is,
1693 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1694 * - gap_end = vma->vm_start >= info->low_limit + length;
1695 * - gap_end - gap_start >= length
1698 struct mm_struct
*mm
= current
->mm
;
1699 struct vm_area_struct
*vma
;
1700 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1702 /* Adjust search length to account for worst case alignment overhead */
1703 length
= info
->length
+ info
->align_mask
;
1704 if (length
< info
->length
)
1707 /* Adjust search limits by the desired length */
1708 if (info
->high_limit
< length
)
1710 high_limit
= info
->high_limit
- length
;
1712 if (info
->low_limit
> high_limit
)
1714 low_limit
= info
->low_limit
+ length
;
1716 /* Check if rbtree root looks promising */
1717 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1719 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1720 if (vma
->rb_subtree_gap
< length
)
1724 /* Visit left subtree if it looks promising */
1725 gap_end
= vm_start_gap(vma
);
1726 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1727 struct vm_area_struct
*left
=
1728 rb_entry(vma
->vm_rb
.rb_left
,
1729 struct vm_area_struct
, vm_rb
);
1730 if (left
->rb_subtree_gap
>= length
) {
1736 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1738 /* Check if current node has a suitable gap */
1739 if (gap_start
> high_limit
)
1741 if (gap_end
>= low_limit
&&
1742 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1745 /* Visit right subtree if it looks promising */
1746 if (vma
->vm_rb
.rb_right
) {
1747 struct vm_area_struct
*right
=
1748 rb_entry(vma
->vm_rb
.rb_right
,
1749 struct vm_area_struct
, vm_rb
);
1750 if (right
->rb_subtree_gap
>= length
) {
1756 /* Go back up the rbtree to find next candidate node */
1758 struct rb_node
*prev
= &vma
->vm_rb
;
1759 if (!rb_parent(prev
))
1761 vma
= rb_entry(rb_parent(prev
),
1762 struct vm_area_struct
, vm_rb
);
1763 if (prev
== vma
->vm_rb
.rb_left
) {
1764 gap_start
= vm_end_gap(vma
->vm_prev
);
1765 gap_end
= vm_start_gap(vma
);
1772 /* Check highest gap, which does not precede any rbtree node */
1773 gap_start
= mm
->highest_vm_end
;
1774 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1775 if (gap_start
> high_limit
)
1779 /* We found a suitable gap. Clip it with the original low_limit. */
1780 if (gap_start
< info
->low_limit
)
1781 gap_start
= info
->low_limit
;
1783 /* Adjust gap address to the desired alignment */
1784 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1786 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1787 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1791 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1793 struct mm_struct
*mm
= current
->mm
;
1794 struct vm_area_struct
*vma
;
1795 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1797 /* Adjust search length to account for worst case alignment overhead */
1798 length
= info
->length
+ info
->align_mask
;
1799 if (length
< info
->length
)
1803 * Adjust search limits by the desired length.
1804 * See implementation comment at top of unmapped_area().
1806 gap_end
= info
->high_limit
;
1807 if (gap_end
< length
)
1809 high_limit
= gap_end
- length
;
1811 if (info
->low_limit
> high_limit
)
1813 low_limit
= info
->low_limit
+ length
;
1815 /* Check highest gap, which does not precede any rbtree node */
1816 gap_start
= mm
->highest_vm_end
;
1817 if (gap_start
<= high_limit
)
1820 /* Check if rbtree root looks promising */
1821 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1823 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1824 if (vma
->rb_subtree_gap
< length
)
1828 /* Visit right subtree if it looks promising */
1829 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1830 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1831 struct vm_area_struct
*right
=
1832 rb_entry(vma
->vm_rb
.rb_right
,
1833 struct vm_area_struct
, vm_rb
);
1834 if (right
->rb_subtree_gap
>= length
) {
1841 /* Check if current node has a suitable gap */
1842 gap_end
= vm_start_gap(vma
);
1843 if (gap_end
< low_limit
)
1845 if (gap_start
<= high_limit
&&
1846 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1849 /* Visit left subtree if it looks promising */
1850 if (vma
->vm_rb
.rb_left
) {
1851 struct vm_area_struct
*left
=
1852 rb_entry(vma
->vm_rb
.rb_left
,
1853 struct vm_area_struct
, vm_rb
);
1854 if (left
->rb_subtree_gap
>= length
) {
1860 /* Go back up the rbtree to find next candidate node */
1862 struct rb_node
*prev
= &vma
->vm_rb
;
1863 if (!rb_parent(prev
))
1865 vma
= rb_entry(rb_parent(prev
),
1866 struct vm_area_struct
, vm_rb
);
1867 if (prev
== vma
->vm_rb
.rb_right
) {
1868 gap_start
= vma
->vm_prev
?
1869 vm_end_gap(vma
->vm_prev
) : 0;
1876 /* We found a suitable gap. Clip it with the original high_limit. */
1877 if (gap_end
> info
->high_limit
)
1878 gap_end
= info
->high_limit
;
1881 /* Compute highest gap address at the desired alignment */
1882 gap_end
-= info
->length
;
1883 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
1885 VM_BUG_ON(gap_end
< info
->low_limit
);
1886 VM_BUG_ON(gap_end
< gap_start
);
1890 /* Get an address range which is currently unmapped.
1891 * For shmat() with addr=0.
1893 * Ugly calling convention alert:
1894 * Return value with the low bits set means error value,
1896 * if (ret & ~PAGE_MASK)
1899 * This function "knows" that -ENOMEM has the bits set.
1901 #ifndef HAVE_ARCH_UNMAPPED_AREA
1903 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1904 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1906 struct mm_struct
*mm
= current
->mm
;
1907 struct vm_area_struct
*vma
, *prev
;
1908 struct vm_unmapped_area_info info
;
1910 if (len
> TASK_SIZE
- mmap_min_addr
)
1913 if (flags
& MAP_FIXED
)
1917 addr
= PAGE_ALIGN(addr
);
1918 vma
= find_vma_prev(mm
, addr
, &prev
);
1919 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1920 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1921 (!prev
|| addr
>= vm_end_gap(prev
)))
1927 info
.low_limit
= TASK_UNMAPPED_BASE
;
1928 info
.high_limit
= TASK_SIZE
;
1929 info
.align_mask
= 0;
1930 return vm_unmapped_area(&info
);
1934 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1937 * Is this a new hole at the lowest possible address?
1939 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1940 mm
->free_area_cache
= addr
;
1944 * This mmap-allocator allocates new areas top-down from below the
1945 * stack's low limit (the base):
1947 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1949 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1950 const unsigned long len
, const unsigned long pgoff
,
1951 const unsigned long flags
)
1953 struct vm_area_struct
*vma
, *prev
;
1954 struct mm_struct
*mm
= current
->mm
;
1955 unsigned long addr
= addr0
;
1956 struct vm_unmapped_area_info info
;
1958 /* requested length too big for entire address space */
1959 if (len
> TASK_SIZE
- mmap_min_addr
)
1962 if (flags
& MAP_FIXED
)
1965 /* requesting a specific address */
1967 addr
= PAGE_ALIGN(addr
);
1968 vma
= find_vma_prev(mm
, addr
, &prev
);
1969 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1970 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1971 (!prev
|| addr
>= vm_end_gap(prev
)))
1975 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1977 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
1978 info
.high_limit
= mm
->mmap_base
;
1979 info
.align_mask
= 0;
1980 addr
= vm_unmapped_area(&info
);
1983 * A failed mmap() very likely causes application failure,
1984 * so fall back to the bottom-up function here. This scenario
1985 * can happen with large stack limits and large mmap()
1988 if (addr
& ~PAGE_MASK
) {
1989 VM_BUG_ON(addr
!= -ENOMEM
);
1991 info
.low_limit
= TASK_UNMAPPED_BASE
;
1992 info
.high_limit
= TASK_SIZE
;
1993 addr
= vm_unmapped_area(&info
);
2000 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
2003 * Is this a new hole at the highest possible address?
2005 if (addr
> mm
->free_area_cache
)
2006 mm
->free_area_cache
= addr
;
2008 /* dont allow allocations above current base */
2009 if (mm
->free_area_cache
> mm
->mmap_base
)
2010 mm
->free_area_cache
= mm
->mmap_base
;
2014 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2015 unsigned long pgoff
, unsigned long flags
)
2017 unsigned long (*get_area
)(struct file
*, unsigned long,
2018 unsigned long, unsigned long, unsigned long);
2020 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2024 /* Careful about overflows.. */
2025 if (len
> TASK_SIZE
)
2028 get_area
= current
->mm
->get_unmapped_area
;
2029 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
2030 get_area
= file
->f_op
->get_unmapped_area
;
2031 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2032 if (IS_ERR_VALUE(addr
))
2035 if (addr
> TASK_SIZE
- len
)
2037 if (addr
& ~PAGE_MASK
)
2040 addr
= arch_rebalance_pgtables(addr
, len
);
2041 error
= security_mmap_addr(addr
);
2042 return error
? error
: addr
;
2045 EXPORT_SYMBOL(get_unmapped_area
);
2047 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2048 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2050 struct rb_node
*rb_node
;
2051 struct vm_area_struct
*vma
;
2053 /* Check the cache first. */
2054 vma
= vmacache_find(mm
, addr
);
2058 rb_node
= mm
->mm_rb
.rb_node
;
2062 struct vm_area_struct
*tmp
;
2064 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2066 if (tmp
->vm_end
> addr
) {
2068 if (tmp
->vm_start
<= addr
)
2070 rb_node
= rb_node
->rb_left
;
2072 rb_node
= rb_node
->rb_right
;
2076 vmacache_update(addr
, vma
);
2080 EXPORT_SYMBOL(find_vma
);
2083 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2085 struct vm_area_struct
*
2086 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2087 struct vm_area_struct
**pprev
)
2089 struct vm_area_struct
*vma
;
2091 vma
= find_vma(mm
, addr
);
2093 *pprev
= vma
->vm_prev
;
2095 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2098 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2099 rb_node
= rb_node
->rb_right
;
2106 * Verify that the stack growth is acceptable and
2107 * update accounting. This is shared with both the
2108 * grow-up and grow-down cases.
2110 static int acct_stack_growth(struct vm_area_struct
*vma
,
2111 unsigned long size
, unsigned long grow
)
2113 struct mm_struct
*mm
= vma
->vm_mm
;
2114 struct rlimit
*rlim
= current
->signal
->rlim
;
2115 unsigned long new_start
;
2117 /* address space limit tests */
2118 if (!may_expand_vm(mm
, grow
))
2121 /* Stack limit test */
2122 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
2125 /* mlock limit tests */
2126 if (vma
->vm_flags
& VM_LOCKED
) {
2127 unsigned long locked
;
2128 unsigned long limit
;
2129 locked
= mm
->locked_vm
+ grow
;
2130 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
2131 limit
>>= PAGE_SHIFT
;
2132 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2136 /* Check to ensure the stack will not grow into a hugetlb-only region */
2137 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2139 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2143 * Overcommit.. This must be the final test, as it will
2144 * update security statistics.
2146 if (security_vm_enough_memory_mm(mm
, grow
))
2149 /* Ok, everything looks good - let it rip */
2150 if (vma
->vm_flags
& VM_LOCKED
)
2151 mm
->locked_vm
+= grow
;
2152 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
2156 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2158 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2159 * vma is the last one with address > vma->vm_end. Have to extend vma.
2161 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2163 struct vm_area_struct
*next
;
2164 unsigned long gap_addr
;
2167 if (!(vma
->vm_flags
& VM_GROWSUP
))
2170 /* Guard against exceeding limits of the address space. */
2171 address
&= PAGE_MASK
;
2172 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2174 address
+= PAGE_SIZE
;
2176 /* Enforce stack_guard_gap */
2177 gap_addr
= address
+ stack_guard_gap
;
2179 /* Guard against overflow */
2180 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2181 gap_addr
= TASK_SIZE
;
2183 next
= vma
->vm_next
;
2184 if (next
&& next
->vm_start
< gap_addr
) {
2185 if (!(next
->vm_flags
& VM_GROWSUP
))
2187 /* Check that both stack segments have the same anon_vma? */
2190 /* We must make sure the anon_vma is allocated. */
2191 if (unlikely(anon_vma_prepare(vma
)))
2194 /* Enforce stack_guard_gap */
2195 gap_addr
= address
+ stack_guard_gap
;
2196 if (gap_addr
< address
)
2198 next
= vma
->vm_next
;
2199 if (next
&& next
->vm_start
< gap_addr
) {
2200 if (!(next
->vm_flags
& VM_GROWSUP
))
2202 /* Check that both stack segments have the same anon_vma? */
2205 /* We must make sure the anon_vma is allocated. */
2206 if (unlikely(anon_vma_prepare(vma
)))
2210 * vma->vm_start/vm_end cannot change under us because the caller
2211 * is required to hold the mmap_sem in read mode. We need the
2212 * anon_vma lock to serialize against concurrent expand_stacks.
2214 vma_lock_anon_vma(vma
);
2216 /* Somebody else might have raced and expanded it already */
2217 if (address
> vma
->vm_end
) {
2218 unsigned long size
, grow
;
2220 size
= address
- vma
->vm_start
;
2221 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2224 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2225 error
= acct_stack_growth(vma
, size
, grow
);
2228 * vma_gap_update() doesn't support concurrent
2229 * updates, but we only hold a shared mmap_sem
2230 * lock here, so we need to protect against
2231 * concurrent vma expansions.
2232 * vma_lock_anon_vma() doesn't help here, as
2233 * we don't guarantee that all growable vmas
2234 * in a mm share the same root anon vma.
2235 * So, we reuse mm->page_table_lock to guard
2236 * against concurrent vma expansions.
2238 spin_lock(&vma
->vm_mm
->page_table_lock
);
2239 anon_vma_interval_tree_pre_update_vma(vma
);
2240 vma
->vm_end
= address
;
2241 anon_vma_interval_tree_post_update_vma(vma
);
2243 vma_gap_update(vma
->vm_next
);
2245 vma
->vm_mm
->highest_vm_end
= vm_end_gap(vma
);
2246 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2248 perf_event_mmap(vma
);
2252 vma_unlock_anon_vma(vma
);
2253 khugepaged_enter_vma_merge(vma
);
2254 validate_mm(vma
->vm_mm
);
2257 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2260 * vma is the first one with address < vma->vm_start. Have to extend vma.
2262 int expand_downwards(struct vm_area_struct
*vma
,
2263 unsigned long address
)
2265 struct vm_area_struct
*prev
;
2266 unsigned long gap_addr
;
2269 address
&= PAGE_MASK
;
2270 error
= security_mmap_addr(address
);
2274 /* Enforce stack_guard_gap */
2275 gap_addr
= address
- stack_guard_gap
;
2276 if (gap_addr
> address
)
2278 prev
= vma
->vm_prev
;
2279 if (prev
&& prev
->vm_end
> gap_addr
) {
2280 if (!(prev
->vm_flags
& VM_GROWSDOWN
))
2282 /* Check that both stack segments have the same anon_vma? */
2285 /* We must make sure the anon_vma is allocated. */
2286 if (unlikely(anon_vma_prepare(vma
)))
2290 * vma->vm_start/vm_end cannot change under us because the caller
2291 * is required to hold the mmap_sem in read mode. We need the
2292 * anon_vma lock to serialize against concurrent expand_stacks.
2294 vma_lock_anon_vma(vma
);
2296 /* Somebody else might have raced and expanded it already */
2297 if (address
< vma
->vm_start
) {
2298 unsigned long size
, grow
;
2300 size
= vma
->vm_end
- address
;
2301 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2304 if (grow
<= vma
->vm_pgoff
) {
2305 error
= acct_stack_growth(vma
, size
, grow
);
2308 * vma_gap_update() doesn't support concurrent
2309 * updates, but we only hold a shared mmap_sem
2310 * lock here, so we need to protect against
2311 * concurrent vma expansions.
2312 * vma_lock_anon_vma() doesn't help here, as
2313 * we don't guarantee that all growable vmas
2314 * in a mm share the same root anon vma.
2315 * So, we reuse mm->page_table_lock to guard
2316 * against concurrent vma expansions.
2318 spin_lock(&vma
->vm_mm
->page_table_lock
);
2319 anon_vma_interval_tree_pre_update_vma(vma
);
2320 vma
->vm_start
= address
;
2321 vma
->vm_pgoff
-= grow
;
2322 anon_vma_interval_tree_post_update_vma(vma
);
2323 vma_gap_update(vma
);
2324 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2326 perf_event_mmap(vma
);
2330 vma_unlock_anon_vma(vma
);
2331 khugepaged_enter_vma_merge(vma
);
2332 validate_mm(vma
->vm_mm
);
2336 /* enforced gap between the expanding stack and other mappings. */
2337 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2339 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2344 val
= simple_strtoul(p
, &endptr
, 10);
2346 stack_guard_gap
= val
<< PAGE_SHIFT
;
2350 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2352 #ifdef CONFIG_STACK_GROWSUP
2353 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2355 return expand_upwards(vma
, address
);
2358 struct vm_area_struct
*
2359 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2361 struct vm_area_struct
*vma
, *prev
;
2364 vma
= find_vma_prev(mm
, addr
, &prev
);
2365 if (vma
&& (vma
->vm_start
<= addr
))
2367 if (!prev
|| expand_stack(prev
, addr
))
2369 if (prev
->vm_flags
& VM_LOCKED
)
2370 __mlock_vma_pages_range(prev
, addr
, prev
->vm_end
, NULL
);
2374 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2376 return expand_downwards(vma
, address
);
2379 struct vm_area_struct
*
2380 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
2382 struct vm_area_struct
* vma
;
2383 unsigned long start
;
2386 vma
= find_vma(mm
,addr
);
2389 if (vma
->vm_start
<= addr
)
2391 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2393 start
= vma
->vm_start
;
2394 if (expand_stack(vma
, addr
))
2396 if (vma
->vm_flags
& VM_LOCKED
)
2397 __mlock_vma_pages_range(vma
, addr
, start
, NULL
);
2403 * Ok - we have the memory areas we should free on the vma list,
2404 * so release them, and do the vma updates.
2406 * Called with the mm semaphore held.
2408 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2410 unsigned long nr_accounted
= 0;
2412 /* Update high watermark before we lower total_vm */
2413 update_hiwater_vm(mm
);
2415 long nrpages
= vma_pages(vma
);
2417 if (vma
->vm_flags
& VM_ACCOUNT
)
2418 nr_accounted
+= nrpages
;
2419 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
2420 vma
= remove_vma(vma
);
2422 vm_unacct_memory(nr_accounted
);
2427 * Get rid of page table information in the indicated region.
2429 * Called with the mm semaphore held.
2431 static void unmap_region(struct mm_struct
*mm
,
2432 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2433 unsigned long start
, unsigned long end
)
2435 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2436 struct mmu_gather tlb
;
2439 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2440 update_hiwater_rss(mm
);
2441 unmap_vmas(&tlb
, vma
, start
, end
);
2442 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2443 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2444 tlb_finish_mmu(&tlb
, start
, end
);
2448 * Create a list of vma's touched by the unmap, removing them from the mm's
2449 * vma list as we go..
2452 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2453 struct vm_area_struct
*prev
, unsigned long end
)
2455 struct vm_area_struct
**insertion_point
;
2456 struct vm_area_struct
*tail_vma
= NULL
;
2459 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2460 vma
->vm_prev
= NULL
;
2462 vma_rb_erase(vma
, &mm
->mm_rb
);
2466 } while (vma
&& vma
->vm_start
< end
);
2467 *insertion_point
= vma
;
2469 vma
->vm_prev
= prev
;
2470 vma_gap_update(vma
);
2472 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2473 tail_vma
->vm_next
= NULL
;
2474 if (mm
->unmap_area
== arch_unmap_area
)
2475 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
2477 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
2478 mm
->unmap_area(mm
, addr
);
2480 /* Kill the cache */
2481 vmacache_invalidate(mm
);
2485 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2486 * munmap path where it doesn't make sense to fail.
2488 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
2489 unsigned long addr
, int new_below
)
2491 struct vm_area_struct
*new;
2494 if (is_vm_hugetlb_page(vma
) && (addr
&
2495 ~(huge_page_mask(hstate_vma(vma
)))))
2498 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2502 /* most fields are the same, copy all, and then fixup */
2505 INIT_LIST_HEAD(&new->anon_vma_chain
);
2510 new->vm_start
= addr
;
2511 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2514 err
= vma_dup_policy(vma
, new);
2518 if (anon_vma_clone(new, vma
))
2522 get_file(new->vm_file
);
2524 if (new->vm_ops
&& new->vm_ops
->open
)
2525 new->vm_ops
->open(new);
2528 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2529 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2531 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2537 /* Clean everything up if vma_adjust failed. */
2538 if (new->vm_ops
&& new->vm_ops
->close
)
2539 new->vm_ops
->close(new);
2542 unlink_anon_vmas(new);
2544 mpol_put(vma_policy(new));
2546 kmem_cache_free(vm_area_cachep
, new);
2552 * Split a vma into two pieces at address 'addr', a new vma is allocated
2553 * either for the first part or the tail.
2555 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2556 unsigned long addr
, int new_below
)
2558 if (mm
->map_count
>= sysctl_max_map_count
)
2561 return __split_vma(mm
, vma
, addr
, new_below
);
2564 /* Munmap is split into 2 main parts -- this part which finds
2565 * what needs doing, and the areas themselves, which do the
2566 * work. This now handles partial unmappings.
2567 * Jeremy Fitzhardinge <jeremy@goop.org>
2569 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2572 struct vm_area_struct
*vma
, *prev
, *last
;
2574 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2577 if ((len
= PAGE_ALIGN(len
)) == 0)
2580 /* Find the first overlapping VMA */
2581 vma
= find_vma(mm
, start
);
2584 prev
= vma
->vm_prev
;
2585 /* we have start < vma->vm_end */
2587 /* if it doesn't overlap, we have nothing.. */
2589 if (vma
->vm_start
>= end
)
2593 * If we need to split any vma, do it now to save pain later.
2595 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2596 * unmapped vm_area_struct will remain in use: so lower split_vma
2597 * places tmp vma above, and higher split_vma places tmp vma below.
2599 if (start
> vma
->vm_start
) {
2603 * Make sure that map_count on return from munmap() will
2604 * not exceed its limit; but let map_count go just above
2605 * its limit temporarily, to help free resources as expected.
2607 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2610 error
= __split_vma(mm
, vma
, start
, 0);
2616 /* Does it split the last one? */
2617 last
= find_vma(mm
, end
);
2618 if (last
&& end
> last
->vm_start
) {
2619 int error
= __split_vma(mm
, last
, end
, 1);
2623 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2626 * unlock any mlock()ed ranges before detaching vmas
2628 if (mm
->locked_vm
) {
2629 struct vm_area_struct
*tmp
= vma
;
2630 while (tmp
&& tmp
->vm_start
< end
) {
2631 if (tmp
->vm_flags
& VM_LOCKED
) {
2632 mm
->locked_vm
-= vma_pages(tmp
);
2633 munlock_vma_pages_all(tmp
);
2640 * Remove the vma's, and unmap the actual pages
2642 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2643 unmap_region(mm
, vma
, prev
, start
, end
);
2645 /* Fix up all other VM information */
2646 remove_vma_list(mm
, vma
);
2651 int vm_munmap(unsigned long start
, size_t len
)
2654 struct mm_struct
*mm
= current
->mm
;
2656 down_write(&mm
->mmap_sem
);
2657 ret
= do_munmap(mm
, start
, len
);
2658 up_write(&mm
->mmap_sem
);
2661 EXPORT_SYMBOL(vm_munmap
);
2663 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2665 profile_munmap(addr
);
2666 return vm_munmap(addr
, len
);
2669 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2671 #ifdef CONFIG_DEBUG_VM
2672 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2674 up_read(&mm
->mmap_sem
);
2680 * this is really a simplified "do_mmap". it only handles
2681 * anonymous maps. eventually we may be able to do some
2682 * brk-specific accounting here.
2684 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2686 struct mm_struct
* mm
= current
->mm
;
2687 struct vm_area_struct
* vma
, * prev
;
2688 unsigned long flags
;
2689 struct rb_node
** rb_link
, * rb_parent
;
2690 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2693 len
= PAGE_ALIGN(len
);
2697 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2699 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2700 if (error
& ~PAGE_MASK
)
2706 if (mm
->def_flags
& VM_LOCKED
) {
2707 unsigned long locked
, lock_limit
;
2708 locked
= len
>> PAGE_SHIFT
;
2709 locked
+= mm
->locked_vm
;
2710 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2711 lock_limit
>>= PAGE_SHIFT
;
2712 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2717 * mm->mmap_sem is required to protect against another thread
2718 * changing the mappings in case we sleep.
2720 verify_mm_writelocked(mm
);
2723 * Clear old maps. this also does some error checking for us
2726 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
2727 if (do_munmap(mm
, addr
, len
))
2732 /* Check against address space limits *after* clearing old maps... */
2733 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2736 if (mm
->map_count
> sysctl_max_map_count
)
2739 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2742 /* Can we just expand an old private anonymous mapping? */
2743 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2744 NULL
, NULL
, pgoff
, NULL
, NULL
);
2749 * create a vma struct for an anonymous mapping
2751 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2753 vm_unacct_memory(len
>> PAGE_SHIFT
);
2757 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2759 vma
->vm_start
= addr
;
2760 vma
->vm_end
= addr
+ len
;
2761 vma
->vm_pgoff
= pgoff
;
2762 vma
->vm_flags
= flags
;
2763 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2764 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2766 perf_event_mmap(vma
);
2767 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2768 if (flags
& VM_LOCKED
)
2769 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2773 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2775 struct mm_struct
*mm
= current
->mm
;
2779 down_write(&mm
->mmap_sem
);
2780 ret
= do_brk(addr
, len
);
2781 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
2782 up_write(&mm
->mmap_sem
);
2784 mm_populate(addr
, len
);
2787 EXPORT_SYMBOL(vm_brk
);
2789 /* Release all mmaps. */
2790 void exit_mmap(struct mm_struct
*mm
)
2792 struct mmu_gather tlb
;
2793 struct vm_area_struct
*vma
;
2794 unsigned long nr_accounted
= 0;
2796 /* mm's last user has gone, and its about to be pulled down */
2797 mmu_notifier_release(mm
);
2799 if (mm
->locked_vm
) {
2802 if (vma
->vm_flags
& VM_LOCKED
)
2803 munlock_vma_pages_all(vma
);
2811 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2816 tlb_gather_mmu(&tlb
, mm
, 0, -1);
2817 /* update_hiwater_rss(mm) here? but nobody should be looking */
2818 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2819 unmap_vmas(&tlb
, vma
, 0, -1);
2821 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
2822 tlb_finish_mmu(&tlb
, 0, -1);
2825 * Walk the list again, actually closing and freeing it,
2826 * with preemption enabled, without holding any MM locks.
2829 if (vma
->vm_flags
& VM_ACCOUNT
)
2830 nr_accounted
+= vma_pages(vma
);
2831 vma
= remove_vma(vma
);
2833 vm_unacct_memory(nr_accounted
);
2835 WARN_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2838 /* Insert vm structure into process list sorted by address
2839 * and into the inode's i_mmap tree. If vm_file is non-NULL
2840 * then i_mmap_mutex is taken here.
2842 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2844 struct vm_area_struct
*prev
;
2845 struct rb_node
**rb_link
, *rb_parent
;
2848 * The vm_pgoff of a purely anonymous vma should be irrelevant
2849 * until its first write fault, when page's anon_vma and index
2850 * are set. But now set the vm_pgoff it will almost certainly
2851 * end up with (unless mremap moves it elsewhere before that
2852 * first wfault), so /proc/pid/maps tells a consistent story.
2854 * By setting it to reflect the virtual start address of the
2855 * vma, merges and splits can happen in a seamless way, just
2856 * using the existing file pgoff checks and manipulations.
2857 * Similarly in do_mmap_pgoff and in do_brk.
2859 if (!vma
->vm_file
) {
2860 BUG_ON(vma
->anon_vma
);
2861 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2863 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2864 &prev
, &rb_link
, &rb_parent
))
2866 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2867 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2870 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2875 * Copy the vma structure to a new location in the same mm,
2876 * prior to moving page table entries, to effect an mremap move.
2878 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2879 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
2880 bool *need_rmap_locks
)
2882 struct vm_area_struct
*vma
= *vmap
;
2883 unsigned long vma_start
= vma
->vm_start
;
2884 struct mm_struct
*mm
= vma
->vm_mm
;
2885 struct vm_area_struct
*new_vma
, *prev
;
2886 struct rb_node
**rb_link
, *rb_parent
;
2887 bool faulted_in_anon_vma
= true;
2890 * If anonymous vma has not yet been faulted, update new pgoff
2891 * to match new location, to increase its chance of merging.
2893 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2894 pgoff
= addr
>> PAGE_SHIFT
;
2895 faulted_in_anon_vma
= false;
2898 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2899 return NULL
; /* should never get here */
2900 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2901 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
2902 vma_get_anon_name(vma
));
2905 * Source vma may have been merged into new_vma
2907 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2908 vma_start
< new_vma
->vm_end
)) {
2910 * The only way we can get a vma_merge with
2911 * self during an mremap is if the vma hasn't
2912 * been faulted in yet and we were allowed to
2913 * reset the dst vma->vm_pgoff to the
2914 * destination address of the mremap to allow
2915 * the merge to happen. mremap must change the
2916 * vm_pgoff linearity between src and dst vmas
2917 * (in turn preventing a vma_merge) to be
2918 * safe. It is only safe to keep the vm_pgoff
2919 * linear if there are no pages mapped yet.
2921 VM_BUG_ON(faulted_in_anon_vma
);
2922 *vmap
= vma
= new_vma
;
2924 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
2926 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2929 new_vma
->vm_start
= addr
;
2930 new_vma
->vm_end
= addr
+ len
;
2931 new_vma
->vm_pgoff
= pgoff
;
2932 if (vma_dup_policy(vma
, new_vma
))
2934 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2935 if (anon_vma_clone(new_vma
, vma
))
2936 goto out_free_mempol
;
2937 if (new_vma
->vm_file
)
2938 get_file(new_vma
->vm_file
);
2939 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2940 new_vma
->vm_ops
->open(new_vma
);
2941 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2942 *need_rmap_locks
= false;
2948 mpol_put(vma_policy(new_vma
));
2950 kmem_cache_free(vm_area_cachep
, new_vma
);
2955 * Return true if the calling process may expand its vm space by the passed
2958 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2960 unsigned long cur
= mm
->total_vm
; /* pages */
2963 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2965 if (cur
+ npages
> lim
)
2971 static int special_mapping_fault(struct vm_area_struct
*vma
,
2972 struct vm_fault
*vmf
)
2975 struct page
**pages
;
2978 * special mappings have no vm_file, and in that case, the mm
2979 * uses vm_pgoff internally. So we have to subtract it from here.
2980 * We are allowed to do this because we are the mm; do not copy
2981 * this code into drivers!
2983 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2985 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2989 struct page
*page
= *pages
;
2995 return VM_FAULT_SIGBUS
;
2999 * Having a close hook prevents vma merging regardless of flags.
3001 static void special_mapping_close(struct vm_area_struct
*vma
)
3005 static const struct vm_operations_struct special_mapping_vmops
= {
3006 .close
= special_mapping_close
,
3007 .fault
= special_mapping_fault
,
3011 * Called with mm->mmap_sem held for writing.
3012 * Insert a new vma covering the given region, with the given flags.
3013 * Its pages are supplied by the given array of struct page *.
3014 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3015 * The region past the last page supplied will always produce SIGBUS.
3016 * The array pointer and the pages it points to are assumed to stay alive
3017 * for as long as this mapping might exist.
3019 int install_special_mapping(struct mm_struct
*mm
,
3020 unsigned long addr
, unsigned long len
,
3021 unsigned long vm_flags
, struct page
**pages
)
3024 struct vm_area_struct
*vma
;
3026 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
3027 if (unlikely(vma
== NULL
))
3030 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
3032 vma
->vm_start
= addr
;
3033 vma
->vm_end
= addr
+ len
;
3035 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
3036 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3038 vma
->vm_ops
= &special_mapping_vmops
;
3039 vma
->vm_private_data
= pages
;
3041 ret
= insert_vm_struct(mm
, vma
);
3045 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3047 perf_event_mmap(vma
);
3052 kmem_cache_free(vm_area_cachep
, vma
);
3056 static DEFINE_MUTEX(mm_all_locks_mutex
);
3058 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3060 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3062 * The LSB of head.next can't change from under us
3063 * because we hold the mm_all_locks_mutex.
3065 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3067 * We can safely modify head.next after taking the
3068 * anon_vma->root->rwsem. If some other vma in this mm shares
3069 * the same anon_vma we won't take it again.
3071 * No need of atomic instructions here, head.next
3072 * can't change from under us thanks to the
3073 * anon_vma->root->rwsem.
3075 if (__test_and_set_bit(0, (unsigned long *)
3076 &anon_vma
->root
->rb_root
.rb_node
))
3081 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3083 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3085 * AS_MM_ALL_LOCKS can't change from under us because
3086 * we hold the mm_all_locks_mutex.
3088 * Operations on ->flags have to be atomic because
3089 * even if AS_MM_ALL_LOCKS is stable thanks to the
3090 * mm_all_locks_mutex, there may be other cpus
3091 * changing other bitflags in parallel to us.
3093 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3095 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
3100 * This operation locks against the VM for all pte/vma/mm related
3101 * operations that could ever happen on a certain mm. This includes
3102 * vmtruncate, try_to_unmap, and all page faults.
3104 * The caller must take the mmap_sem in write mode before calling
3105 * mm_take_all_locks(). The caller isn't allowed to release the
3106 * mmap_sem until mm_drop_all_locks() returns.
3108 * mmap_sem in write mode is required in order to block all operations
3109 * that could modify pagetables and free pages without need of
3110 * altering the vma layout (for example populate_range() with
3111 * nonlinear vmas). It's also needed in write mode to avoid new
3112 * anon_vmas to be associated with existing vmas.
3114 * A single task can't take more than one mm_take_all_locks() in a row
3115 * or it would deadlock.
3117 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3118 * mapping->flags avoid to take the same lock twice, if more than one
3119 * vma in this mm is backed by the same anon_vma or address_space.
3121 * We can take all the locks in random order because the VM code
3122 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3123 * takes more than one of them in a row. Secondly we're protected
3124 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3126 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3127 * that may have to take thousand of locks.
3129 * mm_take_all_locks() can fail if it's interrupted by signals.
3131 int mm_take_all_locks(struct mm_struct
*mm
)
3133 struct vm_area_struct
*vma
;
3134 struct anon_vma_chain
*avc
;
3136 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3138 mutex_lock(&mm_all_locks_mutex
);
3140 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3141 if (signal_pending(current
))
3143 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3144 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3147 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3148 if (signal_pending(current
))
3151 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3152 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3158 mm_drop_all_locks(mm
);
3162 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3164 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3166 * The LSB of head.next can't change to 0 from under
3167 * us because we hold the mm_all_locks_mutex.
3169 * We must however clear the bitflag before unlocking
3170 * the vma so the users using the anon_vma->rb_root will
3171 * never see our bitflag.
3173 * No need of atomic instructions here, head.next
3174 * can't change from under us until we release the
3175 * anon_vma->root->rwsem.
3177 if (!__test_and_clear_bit(0, (unsigned long *)
3178 &anon_vma
->root
->rb_root
.rb_node
))
3180 anon_vma_unlock_write(anon_vma
);
3184 static void vm_unlock_mapping(struct address_space
*mapping
)
3186 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3188 * AS_MM_ALL_LOCKS can't change to 0 from under us
3189 * because we hold the mm_all_locks_mutex.
3191 mutex_unlock(&mapping
->i_mmap_mutex
);
3192 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3199 * The mmap_sem cannot be released by the caller until
3200 * mm_drop_all_locks() returns.
3202 void mm_drop_all_locks(struct mm_struct
*mm
)
3204 struct vm_area_struct
*vma
;
3205 struct anon_vma_chain
*avc
;
3207 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3208 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3210 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3212 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3213 vm_unlock_anon_vma(avc
->anon_vma
);
3214 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3215 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3218 mutex_unlock(&mm_all_locks_mutex
);
3222 * initialise the VMA slab
3224 void __init
mmap_init(void)
3228 ret
= percpu_counter_init(&vm_committed_as
, 0);
3233 * Initialise sysctl_user_reserve_kbytes.
3235 * This is intended to prevent a user from starting a single memory hogging
3236 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3239 * The default value is min(3% of free memory, 128MB)
3240 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3242 static int init_user_reserve(void)
3244 unsigned long free_kbytes
;
3246 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3248 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3251 module_init(init_user_reserve
)
3254 * Initialise sysctl_admin_reserve_kbytes.
3256 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3257 * to log in and kill a memory hogging process.
3259 * Systems with more than 256MB will reserve 8MB, enough to recover
3260 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3261 * only reserve 3% of free pages by default.
3263 static int init_admin_reserve(void)
3265 unsigned long free_kbytes
;
3267 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3269 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3272 module_init(init_admin_reserve
)
3275 * Reinititalise user and admin reserves if memory is added or removed.
3277 * The default user reserve max is 128MB, and the default max for the
3278 * admin reserve is 8MB. These are usually, but not always, enough to
3279 * enable recovery from a memory hogging process using login/sshd, a shell,
3280 * and tools like top. It may make sense to increase or even disable the
3281 * reserve depending on the existence of swap or variations in the recovery
3282 * tools. So, the admin may have changed them.
3284 * If memory is added and the reserves have been eliminated or increased above
3285 * the default max, then we'll trust the admin.
3287 * If memory is removed and there isn't enough free memory, then we
3288 * need to reset the reserves.
3290 * Otherwise keep the reserve set by the admin.
3292 static int reserve_mem_notifier(struct notifier_block
*nb
,
3293 unsigned long action
, void *data
)
3295 unsigned long tmp
, free_kbytes
;
3299 /* Default max is 128MB. Leave alone if modified by operator. */
3300 tmp
= sysctl_user_reserve_kbytes
;
3301 if (0 < tmp
&& tmp
< (1UL << 17))
3302 init_user_reserve();
3304 /* Default max is 8MB. Leave alone if modified by operator. */
3305 tmp
= sysctl_admin_reserve_kbytes
;
3306 if (0 < tmp
&& tmp
< (1UL << 13))
3307 init_admin_reserve();
3311 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3313 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3314 init_user_reserve();
3315 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3316 sysctl_user_reserve_kbytes
);
3319 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3320 init_admin_reserve();
3321 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3322 sysctl_admin_reserve_kbytes
);
3331 static struct notifier_block reserve_mem_nb
= {
3332 .notifier_call
= reserve_mem_notifier
,
3335 static int __meminit
init_reserve_notifier(void)
3337 if (register_hotmemory_notifier(&reserve_mem_nb
))
3338 printk("Failed registering memory add/remove notifier for admin reserve");
3342 module_init(init_reserve_notifier
)