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/shm.h>
14 #include <linux/mman.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <linux/syscalls.h>
18 #include <linux/capability.h>
19 #include <linux/init.h>
20 #include <linux/file.h>
22 #include <linux/personality.h>
23 #include <linux/security.h>
24 #include <linux/hugetlb.h>
25 #include <linux/profile.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/mempolicy.h>
29 #include <linux/rmap.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/perf_event.h>
32 #include <linux/audit.h>
33 #include <linux/khugepaged.h>
34 #include <linux/uprobes.h>
35 #include <linux/rbtree_augmented.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/notifier.h>
38 #include <linux/memory.h>
40 #include <asm/uaccess.h>
41 #include <asm/cacheflush.h>
43 #include <asm/mmu_context.h>
47 #ifndef arch_mmap_check
48 #define arch_mmap_check(addr, len, flags) (0)
51 #ifndef arch_rebalance_pgtables
52 #define arch_rebalance_pgtables(addr, len) (addr)
55 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
56 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
57 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
58 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
60 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
61 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
62 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
63 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
66 static void unmap_region(struct mm_struct
*mm
,
67 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
68 unsigned long start
, unsigned long end
);
70 /* description of effects of mapping type and prot in current implementation.
71 * this is due to the limited x86 page protection hardware. The expected
72 * behavior is in parens:
75 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
76 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
77 * w: (no) no w: (no) no w: (yes) yes w: (no) no
78 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
80 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
81 * w: (no) no w: (no) no w: (copy) copy w: (no) no
82 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
85 pgprot_t protection_map
[16] = {
86 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
87 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
90 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
92 return __pgprot(pgprot_val(protection_map
[vm_flags
&
93 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
94 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
96 EXPORT_SYMBOL(vm_get_page_prot
);
98 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
99 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
100 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
101 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
102 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
104 * Make sure vm_committed_as in one cacheline and not cacheline shared with
105 * other variables. It can be updated by several CPUs frequently.
107 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
110 * The global memory commitment made in the system can be a metric
111 * that can be used to drive ballooning decisions when Linux is hosted
112 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
113 * balancing memory across competing virtual machines that are hosted.
114 * Several metrics drive this policy engine including the guest reported
117 unsigned long vm_memory_committed(void)
119 return percpu_counter_read_positive(&vm_committed_as
);
121 EXPORT_SYMBOL_GPL(vm_memory_committed
);
124 * Check that a process has enough memory to allocate a new virtual
125 * mapping. 0 means there is enough memory for the allocation to
126 * succeed and -ENOMEM implies there is not.
128 * We currently support three overcommit policies, which are set via the
129 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
131 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
132 * Additional code 2002 Jul 20 by Robert Love.
134 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
136 * Note this is a helper function intended to be used by LSMs which
137 * wish to use this logic.
139 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
141 long free
, allowed
, reserve
;
143 vm_acct_memory(pages
);
146 * Sometimes we want to use more memory than we have
148 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
151 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
152 free
= global_page_state(NR_FREE_PAGES
);
153 free
+= global_page_state(NR_FILE_PAGES
);
156 * shmem pages shouldn't be counted as free in this
157 * case, they can't be purged, only swapped out, and
158 * that won't affect the overall amount of available
159 * memory in the system.
161 free
-= global_page_state(NR_SHMEM
);
163 free
+= get_nr_swap_pages();
166 * Any slabs which are created with the
167 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
168 * which are reclaimable, under pressure. The dentry
169 * cache and most inode caches should fall into this
171 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
174 * Leave reserved pages. The pages are not for anonymous pages.
176 if (free
<= totalreserve_pages
)
179 free
-= totalreserve_pages
;
182 * Reserve some for root
185 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
193 allowed
= (totalram_pages
- hugetlb_total_pages())
194 * sysctl_overcommit_ratio
/ 100;
196 * Reserve some for root
199 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
200 allowed
+= total_swap_pages
;
203 * Don't let a single process grow so big a user can't recover
206 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
207 allowed
-= min_t(long, mm
->total_vm
/ 32, reserve
);
210 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
213 vm_unacct_memory(pages
);
219 * Requires inode->i_mapping->i_mmap_mutex
221 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
222 struct file
*file
, struct address_space
*mapping
)
224 if (vma
->vm_flags
& VM_DENYWRITE
)
225 atomic_inc(&file_inode(file
)->i_writecount
);
226 if (vma
->vm_flags
& VM_SHARED
)
227 mapping
->i_mmap_writable
--;
229 flush_dcache_mmap_lock(mapping
);
230 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
231 list_del_init(&vma
->shared
.nonlinear
);
233 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
234 flush_dcache_mmap_unlock(mapping
);
238 * Unlink a file-based vm structure from its interval tree, to hide
239 * vma from rmap and vmtruncate before freeing its page tables.
241 void unlink_file_vma(struct vm_area_struct
*vma
)
243 struct file
*file
= vma
->vm_file
;
246 struct address_space
*mapping
= file
->f_mapping
;
247 mutex_lock(&mapping
->i_mmap_mutex
);
248 __remove_shared_vm_struct(vma
, file
, mapping
);
249 mutex_unlock(&mapping
->i_mmap_mutex
);
254 * Close a vm structure and free it, returning the next.
256 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
258 struct vm_area_struct
*next
= vma
->vm_next
;
261 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
262 vma
->vm_ops
->close(vma
);
265 mpol_put(vma_policy(vma
));
266 kmem_cache_free(vm_area_cachep
, vma
);
270 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
272 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
274 unsigned long rlim
, retval
;
275 unsigned long newbrk
, oldbrk
;
276 struct mm_struct
*mm
= current
->mm
;
277 struct vm_area_struct
*next
;
278 unsigned long min_brk
;
281 down_write(&mm
->mmap_sem
);
283 #ifdef CONFIG_COMPAT_BRK
285 * CONFIG_COMPAT_BRK can still be overridden by setting
286 * randomize_va_space to 2, which will still cause mm->start_brk
287 * to be arbitrarily shifted
289 if (current
->brk_randomized
)
290 min_brk
= mm
->start_brk
;
292 min_brk
= mm
->end_data
;
294 min_brk
= mm
->start_brk
;
300 * Check against rlimit here. If this check is done later after the test
301 * of oldbrk with newbrk then it can escape the test and let the data
302 * segment grow beyond its set limit the in case where the limit is
303 * not page aligned -Ram Gupta
305 rlim
= rlimit(RLIMIT_DATA
);
306 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
307 (mm
->end_data
- mm
->start_data
) > rlim
)
310 newbrk
= PAGE_ALIGN(brk
);
311 oldbrk
= PAGE_ALIGN(mm
->brk
);
312 if (oldbrk
== newbrk
)
315 /* Always allow shrinking brk. */
316 if (brk
<= mm
->brk
) {
317 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
322 /* Check against existing mmap mappings. */
323 next
= find_vma(mm
, oldbrk
);
324 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
327 /* Ok, looks good - let it rip. */
328 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
333 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
334 up_write(&mm
->mmap_sem
);
336 mm_populate(oldbrk
, newbrk
- oldbrk
);
341 up_write(&mm
->mmap_sem
);
345 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
347 unsigned long max
, prev_end
, subtree_gap
;
350 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
351 * allow two stack_guard_gaps between them here, and when choosing
352 * an unmapped area; whereas when expanding we only require one.
353 * That's a little inconsistent, but keeps the code here simpler.
355 max
= vm_start_gap(vma
);
357 prev_end
= vm_end_gap(vma
->vm_prev
);
363 if (vma
->vm_rb
.rb_left
) {
364 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
365 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
366 if (subtree_gap
> max
)
369 if (vma
->vm_rb
.rb_right
) {
370 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
371 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
372 if (subtree_gap
> max
)
378 #ifdef CONFIG_DEBUG_VM_RB
379 static int browse_rb(struct rb_root
*root
)
381 int i
= 0, j
, bug
= 0;
382 struct rb_node
*nd
, *pn
= NULL
;
383 unsigned long prev
= 0, pend
= 0;
385 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
386 struct vm_area_struct
*vma
;
387 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
388 if (vma
->vm_start
< prev
) {
389 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
);
392 if (vma
->vm_start
< pend
) {
393 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
396 if (vma
->vm_start
> vma
->vm_end
) {
397 printk("vm_end %lx < vm_start %lx\n",
398 vma
->vm_end
, vma
->vm_start
);
401 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
402 printk("free gap %lx, correct %lx\n",
404 vma_compute_subtree_gap(vma
));
409 prev
= vma
->vm_start
;
413 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
416 printk("backwards %d, forwards %d\n", j
, i
);
422 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
426 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
427 struct vm_area_struct
*vma
;
428 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
429 BUG_ON(vma
!= ignore
&&
430 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
));
434 void validate_mm(struct mm_struct
*mm
)
438 unsigned long highest_address
= 0;
439 struct vm_area_struct
*vma
= mm
->mmap
;
441 struct anon_vma_chain
*avc
;
442 vma_lock_anon_vma(vma
);
443 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
444 anon_vma_interval_tree_verify(avc
);
445 vma_unlock_anon_vma(vma
);
446 highest_address
= vm_end_gap(vma
);
450 if (i
!= mm
->map_count
) {
451 printk("map_count %d vm_next %d\n", mm
->map_count
, i
);
454 if (highest_address
!= mm
->highest_vm_end
) {
455 printk("mm->highest_vm_end %lx, found %lx\n",
456 mm
->highest_vm_end
, highest_address
);
459 i
= browse_rb(&mm
->mm_rb
);
460 if (i
!= mm
->map_count
) {
461 printk("map_count %d rb %d\n", mm
->map_count
, i
);
467 #define validate_mm_rb(root, ignore) do { } while (0)
468 #define validate_mm(mm) do { } while (0)
471 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
472 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
475 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
476 * vma->vm_prev->vm_end values changed, without modifying the vma's position
479 static void vma_gap_update(struct vm_area_struct
*vma
)
482 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
483 * function that does exacltly what we want.
485 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
488 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
489 struct rb_root
*root
)
491 /* All rb_subtree_gap values must be consistent prior to insertion */
492 validate_mm_rb(root
, NULL
);
494 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
497 static void vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
500 * All rb_subtree_gap values must be consistent prior to erase,
501 * with the possible exception of the vma being erased.
503 validate_mm_rb(root
, vma
);
506 * Note rb_erase_augmented is a fairly large inline function,
507 * so make sure we instantiate it only once with our desired
508 * augmented rbtree callbacks.
510 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
514 * vma has some anon_vma assigned, and is already inserted on that
515 * anon_vma's interval trees.
517 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
518 * vma must be removed from the anon_vma's interval trees using
519 * anon_vma_interval_tree_pre_update_vma().
521 * After the update, the vma will be reinserted using
522 * anon_vma_interval_tree_post_update_vma().
524 * The entire update must be protected by exclusive mmap_sem and by
525 * the root anon_vma's mutex.
528 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
530 struct anon_vma_chain
*avc
;
532 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
533 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
537 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
539 struct anon_vma_chain
*avc
;
541 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
542 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
545 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
546 unsigned long end
, struct vm_area_struct
**pprev
,
547 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
549 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
551 __rb_link
= &mm
->mm_rb
.rb_node
;
552 rb_prev
= __rb_parent
= NULL
;
555 struct vm_area_struct
*vma_tmp
;
557 __rb_parent
= *__rb_link
;
558 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
560 if (vma_tmp
->vm_end
> addr
) {
561 /* Fail if an existing vma overlaps the area */
562 if (vma_tmp
->vm_start
< end
)
564 __rb_link
= &__rb_parent
->rb_left
;
566 rb_prev
= __rb_parent
;
567 __rb_link
= &__rb_parent
->rb_right
;
573 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
574 *rb_link
= __rb_link
;
575 *rb_parent
= __rb_parent
;
579 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
580 unsigned long addr
, unsigned long end
)
582 unsigned long nr_pages
= 0;
583 struct vm_area_struct
*vma
;
585 /* Find first overlaping mapping */
586 vma
= find_vma_intersection(mm
, addr
, end
);
590 nr_pages
= (min(end
, vma
->vm_end
) -
591 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
593 /* Iterate over the rest of the overlaps */
594 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
595 unsigned long overlap_len
;
597 if (vma
->vm_start
> end
)
600 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
601 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
607 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
608 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
610 /* Update tracking information for the gap following the new vma. */
612 vma_gap_update(vma
->vm_next
);
614 mm
->highest_vm_end
= vm_end_gap(vma
);
617 * vma->vm_prev wasn't known when we followed the rbtree to find the
618 * correct insertion point for that vma. As a result, we could not
619 * update the vma vm_rb parents rb_subtree_gap values on the way down.
620 * So, we first insert the vma with a zero rb_subtree_gap value
621 * (to be consistent with what we did on the way down), and then
622 * immediately update the gap to the correct value. Finally we
623 * rebalance the rbtree after all augmented values have been set.
625 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
626 vma
->rb_subtree_gap
= 0;
628 vma_rb_insert(vma
, &mm
->mm_rb
);
631 static void __vma_link_file(struct vm_area_struct
*vma
)
637 struct address_space
*mapping
= file
->f_mapping
;
639 if (vma
->vm_flags
& VM_DENYWRITE
)
640 atomic_dec(&file_inode(file
)->i_writecount
);
641 if (vma
->vm_flags
& VM_SHARED
)
642 mapping
->i_mmap_writable
++;
644 flush_dcache_mmap_lock(mapping
);
645 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
646 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
648 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
649 flush_dcache_mmap_unlock(mapping
);
654 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
655 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
656 struct rb_node
*rb_parent
)
658 __vma_link_list(mm
, vma
, prev
, rb_parent
);
659 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
662 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
663 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
664 struct rb_node
*rb_parent
)
666 struct address_space
*mapping
= NULL
;
669 mapping
= vma
->vm_file
->f_mapping
;
672 mutex_lock(&mapping
->i_mmap_mutex
);
674 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
675 __vma_link_file(vma
);
678 mutex_unlock(&mapping
->i_mmap_mutex
);
685 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
686 * mm's list and rbtree. It has already been inserted into the interval tree.
688 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
690 struct vm_area_struct
*prev
;
691 struct rb_node
**rb_link
, *rb_parent
;
693 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
694 &prev
, &rb_link
, &rb_parent
))
696 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
701 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
702 struct vm_area_struct
*prev
)
704 struct vm_area_struct
*next
;
706 vma_rb_erase(vma
, &mm
->mm_rb
);
707 prev
->vm_next
= next
= vma
->vm_next
;
709 next
->vm_prev
= prev
;
710 if (mm
->mmap_cache
== vma
)
711 mm
->mmap_cache
= prev
;
715 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
716 * is already present in an i_mmap tree without adjusting the tree.
717 * The following helper function should be used when such adjustments
718 * are necessary. The "insert" vma (if any) is to be inserted
719 * before we drop the necessary locks.
721 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
722 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
724 struct mm_struct
*mm
= vma
->vm_mm
;
725 struct vm_area_struct
*next
= vma
->vm_next
;
726 struct vm_area_struct
*importer
= NULL
;
727 struct address_space
*mapping
= NULL
;
728 struct rb_root
*root
= NULL
;
729 struct anon_vma
*anon_vma
= NULL
;
730 struct file
*file
= vma
->vm_file
;
731 bool start_changed
= false, end_changed
= false;
732 long adjust_next
= 0;
735 if (next
&& !insert
) {
736 struct vm_area_struct
*exporter
= NULL
;
738 if (end
>= next
->vm_end
) {
740 * vma expands, overlapping all the next, and
741 * perhaps the one after too (mprotect case 6).
743 again
: remove_next
= 1 + (end
> next
->vm_end
);
747 } else if (end
> next
->vm_start
) {
749 * vma expands, overlapping part of the next:
750 * mprotect case 5 shifting the boundary up.
752 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
755 } else if (end
< vma
->vm_end
) {
757 * vma shrinks, and !insert tells it's not
758 * split_vma inserting another: so it must be
759 * mprotect case 4 shifting the boundary down.
761 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
767 * Easily overlooked: when mprotect shifts the boundary,
768 * make sure the expanding vma has anon_vma set if the
769 * shrinking vma had, to cover any anon pages imported.
771 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
772 if (anon_vma_clone(importer
, exporter
))
774 importer
->anon_vma
= exporter
->anon_vma
;
779 mapping
= file
->f_mapping
;
780 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
781 root
= &mapping
->i_mmap
;
782 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
785 uprobe_munmap(next
, next
->vm_start
,
789 mutex_lock(&mapping
->i_mmap_mutex
);
792 * Put into interval tree now, so instantiated pages
793 * are visible to arm/parisc __flush_dcache_page
794 * throughout; but we cannot insert into address
795 * space until vma start or end is updated.
797 __vma_link_file(insert
);
801 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
803 anon_vma
= vma
->anon_vma
;
804 if (!anon_vma
&& adjust_next
)
805 anon_vma
= next
->anon_vma
;
807 VM_BUG_ON(adjust_next
&& next
->anon_vma
&&
808 anon_vma
!= next
->anon_vma
);
809 anon_vma_lock_write(anon_vma
);
810 anon_vma_interval_tree_pre_update_vma(vma
);
812 anon_vma_interval_tree_pre_update_vma(next
);
816 flush_dcache_mmap_lock(mapping
);
817 vma_interval_tree_remove(vma
, root
);
819 vma_interval_tree_remove(next
, root
);
822 if (start
!= vma
->vm_start
) {
823 vma
->vm_start
= start
;
824 start_changed
= true;
826 if (end
!= vma
->vm_end
) {
830 vma
->vm_pgoff
= pgoff
;
832 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
833 next
->vm_pgoff
+= adjust_next
;
838 vma_interval_tree_insert(next
, root
);
839 vma_interval_tree_insert(vma
, root
);
840 flush_dcache_mmap_unlock(mapping
);
845 * vma_merge has merged next into vma, and needs
846 * us to remove next before dropping the locks.
848 __vma_unlink(mm
, next
, vma
);
850 __remove_shared_vm_struct(next
, file
, mapping
);
853 * split_vma has split insert from vma, and needs
854 * us to insert it before dropping the locks
855 * (it may either follow vma or precede it).
857 __insert_vm_struct(mm
, insert
);
863 mm
->highest_vm_end
= vm_end_gap(vma
);
864 else if (!adjust_next
)
865 vma_gap_update(next
);
870 anon_vma_interval_tree_post_update_vma(vma
);
872 anon_vma_interval_tree_post_update_vma(next
);
873 anon_vma_unlock_write(anon_vma
);
876 mutex_unlock(&mapping
->i_mmap_mutex
);
887 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
891 anon_vma_merge(vma
, next
);
893 mpol_put(vma_policy(next
));
894 kmem_cache_free(vm_area_cachep
, next
);
896 * In mprotect's case 6 (see comments on vma_merge),
897 * we must remove another next too. It would clutter
898 * up the code too much to do both in one go.
901 if (remove_next
== 2)
904 vma_gap_update(next
);
906 WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
917 * If the vma has a ->close operation then the driver probably needs to release
918 * per-vma resources, so we don't attempt to merge those.
920 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
921 struct file
*file
, unsigned long vm_flags
,
922 const char __user
*anon_name
)
924 if (vma
->vm_flags
^ vm_flags
)
926 if (vma
->vm_file
!= file
)
928 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
930 if (vma_get_anon_name(vma
) != anon_name
)
935 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
936 struct anon_vma
*anon_vma2
,
937 struct vm_area_struct
*vma
)
940 * The list_is_singular() test is to avoid merging VMA cloned from
941 * parents. This can improve scalability caused by anon_vma lock.
943 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
944 list_is_singular(&vma
->anon_vma_chain
)))
946 return anon_vma1
== anon_vma2
;
950 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
951 * in front of (at a lower virtual address and file offset than) the vma.
953 * We cannot merge two vmas if they have differently assigned (non-NULL)
954 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
956 * We don't check here for the merged mmap wrapping around the end of pagecache
957 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
958 * wrap, nor mmaps which cover the final page at index -1UL.
961 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
962 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
,
963 const char __user
*anon_name
)
965 if (is_mergeable_vma(vma
, file
, vm_flags
, anon_name
) &&
966 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
967 if (vma
->vm_pgoff
== vm_pgoff
)
974 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
975 * beyond (at a higher virtual address and file offset than) the vma.
977 * We cannot merge two vmas if they have differently assigned (non-NULL)
978 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
981 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
982 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
,
983 const char __user
*anon_name
)
985 if (is_mergeable_vma(vma
, file
, vm_flags
, anon_name
) &&
986 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
988 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
989 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
996 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
997 * figure out whether that can be merged with its predecessor or its
998 * successor. Or both (it neatly fills a hole).
1000 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1001 * certain not to be mapped by the time vma_merge is called; but when
1002 * called for mprotect, it is certain to be already mapped (either at
1003 * an offset within prev, or at the start of next), and the flags of
1004 * this area are about to be changed to vm_flags - and the no-change
1005 * case has already been eliminated.
1007 * The following mprotect cases have to be considered, where AAAA is
1008 * the area passed down from mprotect_fixup, never extending beyond one
1009 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1011 * AAAA AAAA AAAA AAAA
1012 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1013 * cannot merge might become might become might become
1014 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1015 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1016 * mremap move: PPPPNNNNNNNN 8
1018 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1019 * might become case 1 below case 2 below case 3 below
1021 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1022 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1024 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1025 struct vm_area_struct
*prev
, unsigned long addr
,
1026 unsigned long end
, unsigned long vm_flags
,
1027 struct anon_vma
*anon_vma
, struct file
*file
,
1028 pgoff_t pgoff
, struct mempolicy
*policy
,
1029 const char __user
*anon_name
)
1031 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1032 struct vm_area_struct
*area
, *next
;
1036 * We later require that vma->vm_flags == vm_flags,
1037 * so this tests vma->vm_flags & VM_SPECIAL, too.
1039 if (vm_flags
& VM_SPECIAL
)
1043 next
= prev
->vm_next
;
1047 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
1048 next
= next
->vm_next
;
1051 * Can it merge with the predecessor?
1053 if (prev
&& prev
->vm_end
== addr
&&
1054 mpol_equal(vma_policy(prev
), policy
) &&
1055 can_vma_merge_after(prev
, vm_flags
, anon_vma
,
1056 file
, pgoff
, anon_name
)) {
1058 * OK, it can. Can we now merge in the successor as well?
1060 if (next
&& end
== next
->vm_start
&&
1061 mpol_equal(policy
, vma_policy(next
)) &&
1062 can_vma_merge_before(next
, vm_flags
, anon_vma
,
1063 file
, pgoff
+pglen
, anon_name
) &&
1064 is_mergeable_anon_vma(prev
->anon_vma
,
1065 next
->anon_vma
, NULL
)) {
1067 err
= vma_adjust(prev
, prev
->vm_start
,
1068 next
->vm_end
, prev
->vm_pgoff
, NULL
);
1069 } else /* cases 2, 5, 7 */
1070 err
= vma_adjust(prev
, prev
->vm_start
,
1071 end
, prev
->vm_pgoff
, NULL
);
1074 khugepaged_enter_vma_merge(prev
);
1079 * Can this new request be merged in front of next?
1081 if (next
&& end
== next
->vm_start
&&
1082 mpol_equal(policy
, vma_policy(next
)) &&
1083 can_vma_merge_before(next
, vm_flags
, anon_vma
,
1084 file
, pgoff
+pglen
, anon_name
)) {
1085 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1086 err
= vma_adjust(prev
, prev
->vm_start
,
1087 addr
, prev
->vm_pgoff
, NULL
);
1088 else /* cases 3, 8 */
1089 err
= vma_adjust(area
, addr
, next
->vm_end
,
1090 next
->vm_pgoff
- pglen
, NULL
);
1093 khugepaged_enter_vma_merge(area
);
1101 * Rough compatbility check to quickly see if it's even worth looking
1102 * at sharing an anon_vma.
1104 * They need to have the same vm_file, and the flags can only differ
1105 * in things that mprotect may change.
1107 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1108 * we can merge the two vma's. For example, we refuse to merge a vma if
1109 * there is a vm_ops->close() function, because that indicates that the
1110 * driver is doing some kind of reference counting. But that doesn't
1111 * really matter for the anon_vma sharing case.
1113 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1115 return a
->vm_end
== b
->vm_start
&&
1116 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1117 a
->vm_file
== b
->vm_file
&&
1118 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
1119 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1123 * Do some basic sanity checking to see if we can re-use the anon_vma
1124 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1125 * the same as 'old', the other will be the new one that is trying
1126 * to share the anon_vma.
1128 * NOTE! This runs with mm_sem held for reading, so it is possible that
1129 * the anon_vma of 'old' is concurrently in the process of being set up
1130 * by another page fault trying to merge _that_. But that's ok: if it
1131 * is being set up, that automatically means that it will be a singleton
1132 * acceptable for merging, so we can do all of this optimistically. But
1133 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1135 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1136 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1137 * is to return an anon_vma that is "complex" due to having gone through
1140 * We also make sure that the two vma's are compatible (adjacent,
1141 * and with the same memory policies). That's all stable, even with just
1142 * a read lock on the mm_sem.
1144 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1146 if (anon_vma_compatible(a
, b
)) {
1147 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
1149 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1156 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1157 * neighbouring vmas for a suitable anon_vma, before it goes off
1158 * to allocate a new anon_vma. It checks because a repetitive
1159 * sequence of mprotects and faults may otherwise lead to distinct
1160 * anon_vmas being allocated, preventing vma merge in subsequent
1163 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1165 struct anon_vma
*anon_vma
;
1166 struct vm_area_struct
*near
;
1168 near
= vma
->vm_next
;
1172 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1176 near
= vma
->vm_prev
;
1180 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1185 * There's no absolute need to look only at touching neighbours:
1186 * we could search further afield for "compatible" anon_vmas.
1187 * But it would probably just be a waste of time searching,
1188 * or lead to too many vmas hanging off the same anon_vma.
1189 * We're trying to allow mprotect remerging later on,
1190 * not trying to minimize memory used for anon_vmas.
1195 #ifdef CONFIG_PROC_FS
1196 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
1197 struct file
*file
, long pages
)
1199 const unsigned long stack_flags
1200 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
1202 mm
->total_vm
+= pages
;
1205 mm
->shared_vm
+= pages
;
1206 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
1207 mm
->exec_vm
+= pages
;
1208 } else if (flags
& stack_flags
)
1209 mm
->stack_vm
+= pages
;
1211 #endif /* CONFIG_PROC_FS */
1214 * If a hint addr is less than mmap_min_addr change hint to be as
1215 * low as possible but still greater than mmap_min_addr
1217 static inline unsigned long round_hint_to_min(unsigned long hint
)
1220 if (((void *)hint
!= NULL
) &&
1221 (hint
< mmap_min_addr
))
1222 return PAGE_ALIGN(mmap_min_addr
);
1227 * The caller must hold down_write(¤t->mm->mmap_sem).
1230 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
1231 unsigned long len
, unsigned long prot
,
1232 unsigned long flags
, unsigned long pgoff
,
1233 unsigned long *populate
)
1235 struct mm_struct
* mm
= current
->mm
;
1236 struct inode
*inode
;
1237 vm_flags_t vm_flags
;
1242 * Does the application expect PROT_READ to imply PROT_EXEC?
1244 * (the exception is when the underlying filesystem is noexec
1245 * mounted, in which case we dont add PROT_EXEC.)
1247 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1248 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
1254 if (!(flags
& MAP_FIXED
))
1255 addr
= round_hint_to_min(addr
);
1257 /* Careful about overflows.. */
1258 len
= PAGE_ALIGN(len
);
1262 /* offset overflow? */
1263 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1266 /* Too many mappings? */
1267 if (mm
->map_count
> sysctl_max_map_count
)
1270 /* Obtain the address to map to. we verify (or select) it and ensure
1271 * that it represents a valid section of the address space.
1273 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1274 if (addr
& ~PAGE_MASK
)
1277 /* Do simple checking here so the lower-level routines won't have
1278 * to. we assume access permissions have been handled by the open
1279 * of the memory object, so we don't do any here.
1281 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1282 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1284 if (flags
& MAP_LOCKED
)
1285 if (!can_do_mlock())
1288 /* mlock MCL_FUTURE? */
1289 if (vm_flags
& VM_LOCKED
) {
1290 unsigned long locked
, lock_limit
;
1291 locked
= len
>> PAGE_SHIFT
;
1292 locked
+= mm
->locked_vm
;
1293 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1294 lock_limit
>>= PAGE_SHIFT
;
1295 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1299 inode
= file
? file_inode(file
) : NULL
;
1302 switch (flags
& MAP_TYPE
) {
1304 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1308 * Make sure we don't allow writing to an append-only
1311 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1315 * Make sure there are no mandatory locks on the file.
1317 if (locks_verify_locked(inode
))
1320 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1321 if (!(file
->f_mode
& FMODE_WRITE
))
1322 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1326 if (!(file
->f_mode
& FMODE_READ
))
1328 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1329 if (vm_flags
& VM_EXEC
)
1331 vm_flags
&= ~VM_MAYEXEC
;
1334 if (!file
->f_op
|| !file
->f_op
->mmap
)
1342 switch (flags
& MAP_TYPE
) {
1348 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1352 * Set pgoff according to addr for anon_vma.
1354 pgoff
= addr
>> PAGE_SHIFT
;
1362 * Set 'VM_NORESERVE' if we should not account for the
1363 * memory use of this mapping.
1365 if (flags
& MAP_NORESERVE
) {
1366 /* We honor MAP_NORESERVE if allowed to overcommit */
1367 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1368 vm_flags
|= VM_NORESERVE
;
1370 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1371 if (file
&& is_file_hugepages(file
))
1372 vm_flags
|= VM_NORESERVE
;
1375 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
);
1376 if (!IS_ERR_VALUE(addr
) &&
1377 ((vm_flags
& VM_LOCKED
) ||
1378 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1383 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1384 unsigned long, prot
, unsigned long, flags
,
1385 unsigned long, fd
, unsigned long, pgoff
)
1387 struct file
*file
= NULL
;
1388 unsigned long retval
= -EBADF
;
1390 if (!(flags
& MAP_ANONYMOUS
)) {
1391 audit_mmap_fd(fd
, flags
);
1392 if (unlikely(flags
& MAP_HUGETLB
))
1397 if (is_file_hugepages(file
))
1398 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1399 } else if (flags
& MAP_HUGETLB
) {
1400 struct user_struct
*user
= NULL
;
1401 struct hstate
*hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) &
1407 len
= ALIGN(len
, huge_page_size(hs
));
1409 * VM_NORESERVE is used because the reservations will be
1410 * taken when vm_ops->mmap() is called
1411 * A dummy user value is used because we are not locking
1412 * memory so no accounting is necessary
1414 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1416 &user
, HUGETLB_ANONHUGE_INODE
,
1417 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1419 return PTR_ERR(file
);
1422 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1424 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1431 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1432 struct mmap_arg_struct
{
1436 unsigned long flags
;
1438 unsigned long offset
;
1441 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1443 struct mmap_arg_struct a
;
1445 if (copy_from_user(&a
, arg
, sizeof(a
)))
1447 if (a
.offset
& ~PAGE_MASK
)
1450 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1451 a
.offset
>> PAGE_SHIFT
);
1453 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1456 * Some shared mappigns will want the pages marked read-only
1457 * to track write events. If so, we'll downgrade vm_page_prot
1458 * to the private version (using protection_map[] without the
1461 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1463 vm_flags_t vm_flags
= vma
->vm_flags
;
1465 /* If it was private or non-writable, the write bit is already clear */
1466 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1469 /* The backer wishes to know when pages are first written to? */
1470 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1473 /* The open routine did something to the protections already? */
1474 if (pgprot_val(vma
->vm_page_prot
) !=
1475 pgprot_val(vm_get_page_prot(vm_flags
)))
1478 /* Specialty mapping? */
1479 if (vm_flags
& VM_PFNMAP
)
1482 /* Can the mapping track the dirty pages? */
1483 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1484 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1488 * We account for memory if it's a private writeable mapping,
1489 * not hugepages and VM_NORESERVE wasn't set.
1491 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1494 * hugetlb has its own accounting separate from the core VM
1495 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1497 if (file
&& is_file_hugepages(file
))
1500 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1503 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1504 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
)
1506 struct mm_struct
*mm
= current
->mm
;
1507 struct vm_area_struct
*vma
, *prev
;
1508 int correct_wcount
= 0;
1510 struct rb_node
**rb_link
, *rb_parent
;
1511 unsigned long charged
= 0;
1512 struct inode
*inode
= file
? file_inode(file
) : NULL
;
1514 /* Check against address space limit. */
1515 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
)) {
1516 unsigned long nr_pages
;
1519 * MAP_FIXED may remove pages of mappings that intersects with
1520 * requested mapping. Account for the pages it would unmap.
1522 if (!(vm_flags
& MAP_FIXED
))
1525 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1527 if (!may_expand_vm(mm
, (len
>> PAGE_SHIFT
) - nr_pages
))
1531 /* Clear old maps */
1534 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
1535 if (do_munmap(mm
, addr
, len
))
1541 * Private writable mapping: check memory availability
1543 if (accountable_mapping(file
, vm_flags
)) {
1544 charged
= len
>> PAGE_SHIFT
;
1545 if (security_vm_enough_memory_mm(mm
, charged
))
1547 vm_flags
|= VM_ACCOUNT
;
1551 * Can we just expand an old mapping?
1553 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
,
1559 * Determine the object being mapped and call the appropriate
1560 * specific mapper. the address has already been validated, but
1561 * not unmapped, but the maps are removed from the list.
1563 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1570 vma
->vm_start
= addr
;
1571 vma
->vm_end
= addr
+ len
;
1572 vma
->vm_flags
= vm_flags
;
1573 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1574 vma
->vm_pgoff
= pgoff
;
1575 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1577 error
= -EINVAL
; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1580 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1582 if (vm_flags
& VM_DENYWRITE
) {
1583 error
= deny_write_access(file
);
1588 vma
->vm_file
= get_file(file
);
1589 error
= file
->f_op
->mmap(file
, vma
);
1591 goto unmap_and_free_vma
;
1593 /* Can addr have changed??
1595 * Answer: Yes, several device drivers can do it in their
1596 * f_op->mmap method. -DaveM
1597 * Bug: If addr is changed, prev, rb_link, rb_parent should
1598 * be updated for vma_link()
1600 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1602 addr
= vma
->vm_start
;
1603 pgoff
= vma
->vm_pgoff
;
1604 vm_flags
= vma
->vm_flags
;
1605 } else if (vm_flags
& VM_SHARED
) {
1606 if (unlikely(vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
)))
1608 error
= shmem_zero_setup(vma
);
1613 if (vma_wants_writenotify(vma
)) {
1614 pgprot_t pprot
= vma
->vm_page_prot
;
1616 /* Can vma->vm_page_prot have changed??
1618 * Answer: Yes, drivers may have changed it in their
1619 * f_op->mmap method.
1621 * Ensures that vmas marked as uncached stay that way.
1623 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1624 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1625 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1628 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1629 file
= vma
->vm_file
;
1631 /* Once vma denies write, undo our temporary denial count */
1633 atomic_inc(&inode
->i_writecount
);
1635 perf_event_mmap(vma
);
1637 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1638 if (vm_flags
& VM_LOCKED
) {
1639 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1640 vma
== get_gate_vma(current
->mm
)))
1641 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1643 vma
->vm_flags
&= ~VM_LOCKED
;
1653 atomic_inc(&inode
->i_writecount
);
1654 vma
->vm_file
= NULL
;
1657 /* Undo any partial mapping done by a device driver. */
1658 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1661 kmem_cache_free(vm_area_cachep
, vma
);
1664 vm_unacct_memory(charged
);
1668 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1671 * We implement the search by looking for an rbtree node that
1672 * immediately follows a suitable gap. That is,
1673 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1674 * - gap_end = vma->vm_start >= info->low_limit + length;
1675 * - gap_end - gap_start >= length
1678 struct mm_struct
*mm
= current
->mm
;
1679 struct vm_area_struct
*vma
;
1680 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1682 /* Adjust search length to account for worst case alignment overhead */
1683 length
= info
->length
+ info
->align_mask
;
1684 if (length
< info
->length
)
1687 /* Adjust search limits by the desired length */
1688 if (info
->high_limit
< length
)
1690 high_limit
= info
->high_limit
- length
;
1692 if (info
->low_limit
> high_limit
)
1694 low_limit
= info
->low_limit
+ length
;
1696 /* Check if rbtree root looks promising */
1697 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1699 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1700 if (vma
->rb_subtree_gap
< length
)
1704 /* Visit left subtree if it looks promising */
1705 gap_end
= vm_start_gap(vma
);
1706 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1707 struct vm_area_struct
*left
=
1708 rb_entry(vma
->vm_rb
.rb_left
,
1709 struct vm_area_struct
, vm_rb
);
1710 if (left
->rb_subtree_gap
>= length
) {
1716 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1718 /* Check if current node has a suitable gap */
1719 if (gap_start
> high_limit
)
1721 if (gap_end
>= low_limit
&&
1722 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1725 /* Visit right subtree if it looks promising */
1726 if (vma
->vm_rb
.rb_right
) {
1727 struct vm_area_struct
*right
=
1728 rb_entry(vma
->vm_rb
.rb_right
,
1729 struct vm_area_struct
, vm_rb
);
1730 if (right
->rb_subtree_gap
>= length
) {
1736 /* Go back up the rbtree to find next candidate node */
1738 struct rb_node
*prev
= &vma
->vm_rb
;
1739 if (!rb_parent(prev
))
1741 vma
= rb_entry(rb_parent(prev
),
1742 struct vm_area_struct
, vm_rb
);
1743 if (prev
== vma
->vm_rb
.rb_left
) {
1744 gap_start
= vm_end_gap(vma
->vm_prev
);
1745 gap_end
= vm_start_gap(vma
);
1752 /* Check highest gap, which does not precede any rbtree node */
1753 gap_start
= mm
->highest_vm_end
;
1754 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1755 if (gap_start
> high_limit
)
1759 /* We found a suitable gap. Clip it with the original low_limit. */
1760 if (gap_start
< info
->low_limit
)
1761 gap_start
= info
->low_limit
;
1763 /* Adjust gap address to the desired alignment */
1764 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1766 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1767 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1771 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1773 struct mm_struct
*mm
= current
->mm
;
1774 struct vm_area_struct
*vma
;
1775 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1777 /* Adjust search length to account for worst case alignment overhead */
1778 length
= info
->length
+ info
->align_mask
;
1779 if (length
< info
->length
)
1783 * Adjust search limits by the desired length.
1784 * See implementation comment at top of unmapped_area().
1786 gap_end
= info
->high_limit
;
1787 if (gap_end
< length
)
1789 high_limit
= gap_end
- length
;
1791 if (info
->low_limit
> high_limit
)
1793 low_limit
= info
->low_limit
+ length
;
1795 /* Check highest gap, which does not precede any rbtree node */
1796 gap_start
= mm
->highest_vm_end
;
1797 if (gap_start
<= high_limit
)
1800 /* Check if rbtree root looks promising */
1801 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1803 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1804 if (vma
->rb_subtree_gap
< length
)
1808 /* Visit right subtree if it looks promising */
1809 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1810 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1811 struct vm_area_struct
*right
=
1812 rb_entry(vma
->vm_rb
.rb_right
,
1813 struct vm_area_struct
, vm_rb
);
1814 if (right
->rb_subtree_gap
>= length
) {
1821 /* Check if current node has a suitable gap */
1822 gap_end
= vm_start_gap(vma
);
1823 if (gap_end
< low_limit
)
1825 if (gap_start
<= high_limit
&&
1826 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1829 /* Visit left subtree if it looks promising */
1830 if (vma
->vm_rb
.rb_left
) {
1831 struct vm_area_struct
*left
=
1832 rb_entry(vma
->vm_rb
.rb_left
,
1833 struct vm_area_struct
, vm_rb
);
1834 if (left
->rb_subtree_gap
>= length
) {
1840 /* Go back up the rbtree to find next candidate node */
1842 struct rb_node
*prev
= &vma
->vm_rb
;
1843 if (!rb_parent(prev
))
1845 vma
= rb_entry(rb_parent(prev
),
1846 struct vm_area_struct
, vm_rb
);
1847 if (prev
== vma
->vm_rb
.rb_right
) {
1848 gap_start
= vma
->vm_prev
?
1849 vm_end_gap(vma
->vm_prev
) : 0;
1856 /* We found a suitable gap. Clip it with the original high_limit. */
1857 if (gap_end
> info
->high_limit
)
1858 gap_end
= info
->high_limit
;
1861 /* Compute highest gap address at the desired alignment */
1862 gap_end
-= info
->length
;
1863 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
1865 VM_BUG_ON(gap_end
< info
->low_limit
);
1866 VM_BUG_ON(gap_end
< gap_start
);
1870 /* Get an address range which is currently unmapped.
1871 * For shmat() with addr=0.
1873 * Ugly calling convention alert:
1874 * Return value with the low bits set means error value,
1876 * if (ret & ~PAGE_MASK)
1879 * This function "knows" that -ENOMEM has the bits set.
1881 #ifndef HAVE_ARCH_UNMAPPED_AREA
1883 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1884 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1886 struct mm_struct
*mm
= current
->mm
;
1887 struct vm_area_struct
*vma
, *prev
;
1888 struct vm_unmapped_area_info info
;
1890 if (len
> TASK_SIZE
- mmap_min_addr
)
1893 if (flags
& MAP_FIXED
)
1897 addr
= PAGE_ALIGN(addr
);
1898 vma
= find_vma_prev(mm
, addr
, &prev
);
1899 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1900 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1901 (!prev
|| addr
>= vm_end_gap(prev
)))
1907 info
.low_limit
= TASK_UNMAPPED_BASE
;
1908 info
.high_limit
= TASK_SIZE
;
1909 info
.align_mask
= 0;
1910 return vm_unmapped_area(&info
);
1914 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1917 * Is this a new hole at the lowest possible address?
1919 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1920 mm
->free_area_cache
= addr
;
1924 * This mmap-allocator allocates new areas top-down from below the
1925 * stack's low limit (the base):
1927 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1929 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1930 const unsigned long len
, const unsigned long pgoff
,
1931 const unsigned long flags
)
1933 struct vm_area_struct
*vma
, *prev
;
1934 struct mm_struct
*mm
= current
->mm
;
1935 unsigned long addr
= addr0
;
1936 struct vm_unmapped_area_info info
;
1938 /* requested length too big for entire address space */
1939 if (len
> TASK_SIZE
- mmap_min_addr
)
1942 if (flags
& MAP_FIXED
)
1945 /* requesting a specific address */
1947 addr
= PAGE_ALIGN(addr
);
1948 vma
= find_vma_prev(mm
, addr
, &prev
);
1949 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1950 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1951 (!prev
|| addr
>= vm_end_gap(prev
)))
1955 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1957 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
1958 info
.high_limit
= mm
->mmap_base
;
1959 info
.align_mask
= 0;
1960 addr
= vm_unmapped_area(&info
);
1963 * A failed mmap() very likely causes application failure,
1964 * so fall back to the bottom-up function here. This scenario
1965 * can happen with large stack limits and large mmap()
1968 if (addr
& ~PAGE_MASK
) {
1969 VM_BUG_ON(addr
!= -ENOMEM
);
1971 info
.low_limit
= TASK_UNMAPPED_BASE
;
1972 info
.high_limit
= TASK_SIZE
;
1973 addr
= vm_unmapped_area(&info
);
1980 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1983 * Is this a new hole at the highest possible address?
1985 if (addr
> mm
->free_area_cache
)
1986 mm
->free_area_cache
= addr
;
1988 /* dont allow allocations above current base */
1989 if (mm
->free_area_cache
> mm
->mmap_base
)
1990 mm
->free_area_cache
= mm
->mmap_base
;
1994 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1995 unsigned long pgoff
, unsigned long flags
)
1997 unsigned long (*get_area
)(struct file
*, unsigned long,
1998 unsigned long, unsigned long, unsigned long);
2000 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2004 /* Careful about overflows.. */
2005 if (len
> TASK_SIZE
)
2008 get_area
= current
->mm
->get_unmapped_area
;
2009 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
2010 get_area
= file
->f_op
->get_unmapped_area
;
2011 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2012 if (IS_ERR_VALUE(addr
))
2015 if (addr
> TASK_SIZE
- len
)
2017 if (addr
& ~PAGE_MASK
)
2020 addr
= arch_rebalance_pgtables(addr
, len
);
2021 error
= security_mmap_addr(addr
);
2022 return error
? error
: addr
;
2025 EXPORT_SYMBOL(get_unmapped_area
);
2027 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2028 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2030 struct vm_area_struct
*vma
= NULL
;
2032 /* Check the cache first. */
2033 /* (Cache hit rate is typically around 35%.) */
2034 vma
= ACCESS_ONCE(mm
->mmap_cache
);
2035 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
2036 struct rb_node
*rb_node
;
2038 rb_node
= mm
->mm_rb
.rb_node
;
2042 struct vm_area_struct
*vma_tmp
;
2044 vma_tmp
= rb_entry(rb_node
,
2045 struct vm_area_struct
, vm_rb
);
2047 if (vma_tmp
->vm_end
> addr
) {
2049 if (vma_tmp
->vm_start
<= addr
)
2051 rb_node
= rb_node
->rb_left
;
2053 rb_node
= rb_node
->rb_right
;
2056 mm
->mmap_cache
= vma
;
2061 EXPORT_SYMBOL(find_vma
);
2064 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2066 struct vm_area_struct
*
2067 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2068 struct vm_area_struct
**pprev
)
2070 struct vm_area_struct
*vma
;
2072 vma
= find_vma(mm
, addr
);
2074 *pprev
= vma
->vm_prev
;
2076 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2079 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2080 rb_node
= rb_node
->rb_right
;
2087 * Verify that the stack growth is acceptable and
2088 * update accounting. This is shared with both the
2089 * grow-up and grow-down cases.
2091 static int acct_stack_growth(struct vm_area_struct
*vma
,
2092 unsigned long size
, unsigned long grow
)
2094 struct mm_struct
*mm
= vma
->vm_mm
;
2095 struct rlimit
*rlim
= current
->signal
->rlim
;
2096 unsigned long new_start
;
2098 /* address space limit tests */
2099 if (!may_expand_vm(mm
, grow
))
2102 /* Stack limit test */
2103 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
2106 /* mlock limit tests */
2107 if (vma
->vm_flags
& VM_LOCKED
) {
2108 unsigned long locked
;
2109 unsigned long limit
;
2110 locked
= mm
->locked_vm
+ grow
;
2111 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
2112 limit
>>= PAGE_SHIFT
;
2113 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2117 /* Check to ensure the stack will not grow into a hugetlb-only region */
2118 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2120 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2124 * Overcommit.. This must be the final test, as it will
2125 * update security statistics.
2127 if (security_vm_enough_memory_mm(mm
, grow
))
2130 /* Ok, everything looks good - let it rip */
2131 if (vma
->vm_flags
& VM_LOCKED
)
2132 mm
->locked_vm
+= grow
;
2133 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
2137 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2139 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2140 * vma is the last one with address > vma->vm_end. Have to extend vma.
2142 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2144 struct vm_area_struct
*next
;
2145 unsigned long gap_addr
;
2148 if (!(vma
->vm_flags
& VM_GROWSUP
))
2151 /* Guard against exceeding limits of the address space. */
2152 address
&= PAGE_MASK
;
2153 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2155 address
+= PAGE_SIZE
;
2157 /* Enforce stack_guard_gap */
2158 gap_addr
= address
+ stack_guard_gap
;
2160 /* Guard against overflow */
2161 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2162 gap_addr
= TASK_SIZE
;
2164 next
= vma
->vm_next
;
2165 if (next
&& next
->vm_start
< gap_addr
) {
2166 if (!(next
->vm_flags
& VM_GROWSUP
))
2168 /* Check that both stack segments have the same anon_vma? */
2171 /* We must make sure the anon_vma is allocated. */
2172 if (unlikely(anon_vma_prepare(vma
)))
2176 * vma->vm_start/vm_end cannot change under us because the caller
2177 * is required to hold the mmap_sem in read mode. We need the
2178 * anon_vma lock to serialize against concurrent expand_stacks.
2180 vma_lock_anon_vma(vma
);
2182 /* Somebody else might have raced and expanded it already */
2183 if (address
> vma
->vm_end
) {
2184 unsigned long size
, grow
;
2186 size
= address
- vma
->vm_start
;
2187 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2190 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2191 error
= acct_stack_growth(vma
, size
, grow
);
2194 * vma_gap_update() doesn't support concurrent
2195 * updates, but we only hold a shared mmap_sem
2196 * lock here, so we need to protect against
2197 * concurrent vma expansions.
2198 * vma_lock_anon_vma() doesn't help here, as
2199 * we don't guarantee that all growable vmas
2200 * in a mm share the same root anon vma.
2201 * So, we reuse mm->page_table_lock to guard
2202 * against concurrent vma expansions.
2204 spin_lock(&vma
->vm_mm
->page_table_lock
);
2205 anon_vma_interval_tree_pre_update_vma(vma
);
2206 vma
->vm_end
= address
;
2207 anon_vma_interval_tree_post_update_vma(vma
);
2209 vma_gap_update(vma
->vm_next
);
2211 vma
->vm_mm
->highest_vm_end
= vm_end_gap(vma
);
2212 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2214 perf_event_mmap(vma
);
2218 vma_unlock_anon_vma(vma
);
2219 khugepaged_enter_vma_merge(vma
);
2220 validate_mm(vma
->vm_mm
);
2223 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2226 * vma is the first one with address < vma->vm_start. Have to extend vma.
2228 int expand_downwards(struct vm_area_struct
*vma
,
2229 unsigned long address
)
2231 struct vm_area_struct
*prev
;
2232 unsigned long gap_addr
;
2235 address
&= PAGE_MASK
;
2236 error
= security_mmap_addr(address
);
2240 /* Enforce stack_guard_gap */
2241 gap_addr
= address
- stack_guard_gap
;
2242 if (gap_addr
> address
)
2244 prev
= vma
->vm_prev
;
2245 if (prev
&& prev
->vm_end
> gap_addr
) {
2246 if (!(prev
->vm_flags
& VM_GROWSDOWN
))
2248 /* Check that both stack segments have the same anon_vma? */
2251 /* We must make sure the anon_vma is allocated. */
2252 if (unlikely(anon_vma_prepare(vma
)))
2256 * vma->vm_start/vm_end cannot change under us because the caller
2257 * is required to hold the mmap_sem in read mode. We need the
2258 * anon_vma lock to serialize against concurrent expand_stacks.
2260 vma_lock_anon_vma(vma
);
2262 /* Somebody else might have raced and expanded it already */
2263 if (address
< vma
->vm_start
) {
2264 unsigned long size
, grow
;
2266 size
= vma
->vm_end
- address
;
2267 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2270 if (grow
<= vma
->vm_pgoff
) {
2271 error
= acct_stack_growth(vma
, size
, grow
);
2274 * vma_gap_update() doesn't support concurrent
2275 * updates, but we only hold a shared mmap_sem
2276 * lock here, so we need to protect against
2277 * concurrent vma expansions.
2278 * vma_lock_anon_vma() doesn't help here, as
2279 * we don't guarantee that all growable vmas
2280 * in a mm share the same root anon vma.
2281 * So, we reuse mm->page_table_lock to guard
2282 * against concurrent vma expansions.
2284 spin_lock(&vma
->vm_mm
->page_table_lock
);
2285 anon_vma_interval_tree_pre_update_vma(vma
);
2286 vma
->vm_start
= address
;
2287 vma
->vm_pgoff
-= grow
;
2288 anon_vma_interval_tree_post_update_vma(vma
);
2289 vma_gap_update(vma
);
2290 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2292 perf_event_mmap(vma
);
2296 vma_unlock_anon_vma(vma
);
2297 khugepaged_enter_vma_merge(vma
);
2298 validate_mm(vma
->vm_mm
);
2302 /* enforced gap between the expanding stack and other mappings. */
2303 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2305 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2310 val
= simple_strtoul(p
, &endptr
, 10);
2312 stack_guard_gap
= val
<< PAGE_SHIFT
;
2316 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2318 #ifdef CONFIG_STACK_GROWSUP
2319 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2321 return expand_upwards(vma
, address
);
2324 struct vm_area_struct
*
2325 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2327 struct vm_area_struct
*vma
, *prev
;
2330 vma
= find_vma_prev(mm
, addr
, &prev
);
2331 if (vma
&& (vma
->vm_start
<= addr
))
2333 if (!prev
|| expand_stack(prev
, addr
))
2335 if (prev
->vm_flags
& VM_LOCKED
)
2336 __mlock_vma_pages_range(prev
, addr
, prev
->vm_end
, NULL
);
2340 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2342 return expand_downwards(vma
, address
);
2345 struct vm_area_struct
*
2346 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
2348 struct vm_area_struct
* vma
;
2349 unsigned long start
;
2352 vma
= find_vma(mm
,addr
);
2355 if (vma
->vm_start
<= addr
)
2357 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2359 start
= vma
->vm_start
;
2360 if (expand_stack(vma
, addr
))
2362 if (vma
->vm_flags
& VM_LOCKED
)
2363 __mlock_vma_pages_range(vma
, addr
, start
, NULL
);
2369 * Ok - we have the memory areas we should free on the vma list,
2370 * so release them, and do the vma updates.
2372 * Called with the mm semaphore held.
2374 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2376 unsigned long nr_accounted
= 0;
2378 /* Update high watermark before we lower total_vm */
2379 update_hiwater_vm(mm
);
2381 long nrpages
= vma_pages(vma
);
2383 if (vma
->vm_flags
& VM_ACCOUNT
)
2384 nr_accounted
+= nrpages
;
2385 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
2386 vma
= remove_vma(vma
);
2388 vm_unacct_memory(nr_accounted
);
2393 * Get rid of page table information in the indicated region.
2395 * Called with the mm semaphore held.
2397 static void unmap_region(struct mm_struct
*mm
,
2398 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2399 unsigned long start
, unsigned long end
)
2401 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2402 struct mmu_gather tlb
;
2405 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2406 update_hiwater_rss(mm
);
2407 unmap_vmas(&tlb
, vma
, start
, end
);
2408 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2409 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2410 tlb_finish_mmu(&tlb
, start
, end
);
2414 * Create a list of vma's touched by the unmap, removing them from the mm's
2415 * vma list as we go..
2418 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2419 struct vm_area_struct
*prev
, unsigned long end
)
2421 struct vm_area_struct
**insertion_point
;
2422 struct vm_area_struct
*tail_vma
= NULL
;
2425 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2426 vma
->vm_prev
= NULL
;
2428 vma_rb_erase(vma
, &mm
->mm_rb
);
2432 } while (vma
&& vma
->vm_start
< end
);
2433 *insertion_point
= vma
;
2435 vma
->vm_prev
= prev
;
2436 vma_gap_update(vma
);
2438 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2439 tail_vma
->vm_next
= NULL
;
2440 if (mm
->unmap_area
== arch_unmap_area
)
2441 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
2443 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
2444 mm
->unmap_area(mm
, addr
);
2445 mm
->mmap_cache
= NULL
; /* Kill the cache. */
2449 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2450 * munmap path where it doesn't make sense to fail.
2452 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
2453 unsigned long addr
, int new_below
)
2455 struct mempolicy
*pol
;
2456 struct vm_area_struct
*new;
2459 if (is_vm_hugetlb_page(vma
) && (addr
&
2460 ~(huge_page_mask(hstate_vma(vma
)))))
2463 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2467 /* most fields are the same, copy all, and then fixup */
2470 INIT_LIST_HEAD(&new->anon_vma_chain
);
2475 new->vm_start
= addr
;
2476 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2479 pol
= mpol_dup(vma_policy(vma
));
2484 vma_set_policy(new, pol
);
2486 if (anon_vma_clone(new, vma
))
2490 get_file(new->vm_file
);
2492 if (new->vm_ops
&& new->vm_ops
->open
)
2493 new->vm_ops
->open(new);
2496 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2497 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2499 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2505 /* Clean everything up if vma_adjust failed. */
2506 if (new->vm_ops
&& new->vm_ops
->close
)
2507 new->vm_ops
->close(new);
2510 unlink_anon_vmas(new);
2514 kmem_cache_free(vm_area_cachep
, new);
2520 * Split a vma into two pieces at address 'addr', a new vma is allocated
2521 * either for the first part or the tail.
2523 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2524 unsigned long addr
, int new_below
)
2526 if (mm
->map_count
>= sysctl_max_map_count
)
2529 return __split_vma(mm
, vma
, addr
, new_below
);
2532 /* Munmap is split into 2 main parts -- this part which finds
2533 * what needs doing, and the areas themselves, which do the
2534 * work. This now handles partial unmappings.
2535 * Jeremy Fitzhardinge <jeremy@goop.org>
2537 #ifdef CONFIG_MTK_EXTMEM
2538 extern bool extmem_in_mspace(struct vm_area_struct
*vma
);
2539 extern void * get_virt_from_mspace(void * pa
);
2540 extern size_t extmem_get_mem_size(unsigned long pgoff
);
2541 extern void extmem_free(void* mem
);
2544 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2548 struct vm_area_struct
*vma
, *prev
, *last
;
2550 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2553 if ((len
= PAGE_ALIGN(len
)) == 0)
2556 /* Find the first overlapping VMA */
2557 vma
= find_vma(mm
, start
);
2563 const char *name
=file
->f_path
.dentry
->d_iname
;
2564 if(name
&& (strstr(name
,"app_process") || strstr(name
,"app_process64") || strstr(name
,"main") || strstr(name
,"Binder_")))
2565 printk("name:%s unmap vm_start %lx end: %lx\n", name
, vma
->vm_start
, vma
->vm_end
);
2569 const char *name
= arch_vma_name(vma
);
2570 if(name
&& (strstr(name
,"app_process") || strstr(name
,"app_process64") || strstr(name
,"main") || strstr(name
,"Binder_")))
2571 printk("name:%s unmap vm_start %lx end: %lx\n", name
, vma
->vm_start
, vma
->vm_end
);
2573 prev
= vma
->vm_prev
;
2574 /* we have start < vma->vm_end */
2576 #ifdef CONFIG_MTK_EXTMEM
2577 /* get correct mmap size if in mspace. */
2578 if (extmem_in_mspace(vma
))
2579 len
= extmem_get_mem_size(vma
->vm_pgoff
);
2582 /* if it doesn't overlap, we have nothing.. */
2584 if (vma
->vm_start
>= end
)
2588 * If we need to split any vma, do it now to save pain later.
2590 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2591 * unmapped vm_area_struct will remain in use: so lower split_vma
2592 * places tmp vma above, and higher split_vma places tmp vma below.
2594 if (start
> vma
->vm_start
) {
2598 * Make sure that map_count on return from munmap() will
2599 * not exceed its limit; but let map_count go just above
2600 * its limit temporarily, to help free resources as expected.
2602 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2605 error
= __split_vma(mm
, vma
, start
, 0);
2611 /* Does it split the last one? */
2612 last
= find_vma(mm
, end
);
2613 if (last
&& end
> last
->vm_start
) {
2614 int error
= __split_vma(mm
, last
, end
, 1);
2618 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2621 * unlock any mlock()ed ranges before detaching vmas
2623 if (mm
->locked_vm
) {
2624 struct vm_area_struct
*tmp
= vma
;
2625 while (tmp
&& tmp
->vm_start
< end
) {
2626 if (tmp
->vm_flags
& VM_LOCKED
) {
2627 mm
->locked_vm
-= vma_pages(tmp
);
2628 munlock_vma_pages_all(tmp
);
2635 * Remove the vma's, and unmap the actual pages
2637 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2638 unmap_region(mm
, vma
, prev
, start
, end
);
2640 /* Fix up all other VM information */
2641 remove_vma_list(mm
, vma
);
2646 int vm_munmap(unsigned long start
, size_t len
)
2649 struct mm_struct
*mm
= current
->mm
;
2651 down_write(&mm
->mmap_sem
);
2652 ret
= do_munmap(mm
, start
, len
);
2653 up_write(&mm
->mmap_sem
);
2656 EXPORT_SYMBOL(vm_munmap
);
2658 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2660 profile_munmap(addr
);
2661 return vm_munmap(addr
, len
);
2664 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2666 #ifdef CONFIG_DEBUG_VM
2667 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2669 up_read(&mm
->mmap_sem
);
2675 * this is really a simplified "do_mmap". it only handles
2676 * anonymous maps. eventually we may be able to do some
2677 * brk-specific accounting here.
2679 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2681 struct mm_struct
* mm
= current
->mm
;
2682 struct vm_area_struct
* vma
, * prev
;
2683 unsigned long flags
;
2684 struct rb_node
** rb_link
, * rb_parent
;
2685 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2688 len
= PAGE_ALIGN(len
);
2692 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2694 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2695 if (error
& ~PAGE_MASK
)
2701 if (mm
->def_flags
& VM_LOCKED
) {
2702 unsigned long locked
, lock_limit
;
2703 locked
= len
>> PAGE_SHIFT
;
2704 locked
+= mm
->locked_vm
;
2705 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2706 lock_limit
>>= PAGE_SHIFT
;
2707 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2712 * mm->mmap_sem is required to protect against another thread
2713 * changing the mappings in case we sleep.
2715 verify_mm_writelocked(mm
);
2718 * Clear old maps. this also does some error checking for us
2721 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
2722 if (do_munmap(mm
, addr
, len
))
2727 /* Check against address space limits *after* clearing old maps... */
2728 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2731 if (mm
->map_count
> sysctl_max_map_count
)
2734 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2737 /* Can we just expand an old private anonymous mapping? */
2738 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2739 NULL
, NULL
, pgoff
, NULL
, NULL
);
2744 * create a vma struct for an anonymous mapping
2746 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2748 vm_unacct_memory(len
>> PAGE_SHIFT
);
2752 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2754 vma
->vm_start
= addr
;
2755 vma
->vm_end
= addr
+ len
;
2756 vma
->vm_pgoff
= pgoff
;
2757 vma
->vm_flags
= flags
;
2758 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2759 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2761 perf_event_mmap(vma
);
2762 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2763 if (flags
& VM_LOCKED
)
2764 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2768 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2770 struct mm_struct
*mm
= current
->mm
;
2774 down_write(&mm
->mmap_sem
);
2775 ret
= do_brk(addr
, len
);
2776 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
2777 up_write(&mm
->mmap_sem
);
2779 mm_populate(addr
, len
);
2782 EXPORT_SYMBOL(vm_brk
);
2784 /* Release all mmaps. */
2785 void exit_mmap(struct mm_struct
*mm
)
2787 struct mmu_gather tlb
;
2788 struct vm_area_struct
*vma
;
2789 unsigned long nr_accounted
= 0;
2791 /* mm's last user has gone, and its about to be pulled down */
2792 mmu_notifier_release(mm
);
2794 if (mm
->locked_vm
) {
2797 if (vma
->vm_flags
& VM_LOCKED
)
2798 munlock_vma_pages_all(vma
);
2806 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2811 tlb_gather_mmu(&tlb
, mm
, 0, -1);
2812 /* update_hiwater_rss(mm) here? but nobody should be looking */
2813 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2814 unmap_vmas(&tlb
, vma
, 0, -1);
2816 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
2817 tlb_finish_mmu(&tlb
, 0, -1);
2820 * Walk the list again, actually closing and freeing it,
2821 * with preemption enabled, without holding any MM locks.
2824 if (vma
->vm_flags
& VM_ACCOUNT
)
2825 nr_accounted
+= vma_pages(vma
);
2826 vma
= remove_vma(vma
);
2828 vm_unacct_memory(nr_accounted
);
2830 WARN_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2833 /* Insert vm structure into process list sorted by address
2834 * and into the inode's i_mmap tree. If vm_file is non-NULL
2835 * then i_mmap_mutex is taken here.
2837 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2839 struct vm_area_struct
*prev
;
2840 struct rb_node
**rb_link
, *rb_parent
;
2843 * The vm_pgoff of a purely anonymous vma should be irrelevant
2844 * until its first write fault, when page's anon_vma and index
2845 * are set. But now set the vm_pgoff it will almost certainly
2846 * end up with (unless mremap moves it elsewhere before that
2847 * first wfault), so /proc/pid/maps tells a consistent story.
2849 * By setting it to reflect the virtual start address of the
2850 * vma, merges and splits can happen in a seamless way, just
2851 * using the existing file pgoff checks and manipulations.
2852 * Similarly in do_mmap_pgoff and in do_brk.
2854 if (!vma
->vm_file
) {
2855 BUG_ON(vma
->anon_vma
);
2856 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2858 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2859 &prev
, &rb_link
, &rb_parent
))
2861 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2862 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2865 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2870 * Copy the vma structure to a new location in the same mm,
2871 * prior to moving page table entries, to effect an mremap move.
2873 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2874 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
2875 bool *need_rmap_locks
)
2877 struct vm_area_struct
*vma
= *vmap
;
2878 unsigned long vma_start
= vma
->vm_start
;
2879 struct mm_struct
*mm
= vma
->vm_mm
;
2880 struct vm_area_struct
*new_vma
, *prev
;
2881 struct rb_node
**rb_link
, *rb_parent
;
2882 struct mempolicy
*pol
;
2883 bool faulted_in_anon_vma
= true;
2886 * If anonymous vma has not yet been faulted, update new pgoff
2887 * to match new location, to increase its chance of merging.
2889 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2890 pgoff
= addr
>> PAGE_SHIFT
;
2891 faulted_in_anon_vma
= false;
2894 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2895 return NULL
; /* should never get here */
2896 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2897 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
2898 vma_get_anon_name(vma
));
2901 * Source vma may have been merged into new_vma
2903 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2904 vma_start
< new_vma
->vm_end
)) {
2906 * The only way we can get a vma_merge with
2907 * self during an mremap is if the vma hasn't
2908 * been faulted in yet and we were allowed to
2909 * reset the dst vma->vm_pgoff to the
2910 * destination address of the mremap to allow
2911 * the merge to happen. mremap must change the
2912 * vm_pgoff linearity between src and dst vmas
2913 * (in turn preventing a vma_merge) to be
2914 * safe. It is only safe to keep the vm_pgoff
2915 * linear if there are no pages mapped yet.
2917 VM_BUG_ON(faulted_in_anon_vma
);
2918 *vmap
= vma
= new_vma
;
2920 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
2922 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2925 new_vma
->vm_start
= addr
;
2926 new_vma
->vm_end
= addr
+ len
;
2927 new_vma
->vm_pgoff
= pgoff
;
2928 pol
= mpol_dup(vma_policy(vma
));
2931 vma_set_policy(new_vma
, pol
);
2932 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2933 if (anon_vma_clone(new_vma
, vma
))
2934 goto out_free_mempol
;
2935 if (new_vma
->vm_file
)
2936 get_file(new_vma
->vm_file
);
2937 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2938 new_vma
->vm_ops
->open(new_vma
);
2939 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2940 *need_rmap_locks
= false;
2948 kmem_cache_free(vm_area_cachep
, new_vma
);
2953 * Return true if the calling process may expand its vm space by the passed
2956 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2958 unsigned long cur
= mm
->total_vm
; /* pages */
2961 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2963 if (cur
+ npages
> lim
)
2969 static int special_mapping_fault(struct vm_area_struct
*vma
,
2970 struct vm_fault
*vmf
)
2973 struct page
**pages
;
2976 * special mappings have no vm_file, and in that case, the mm
2977 * uses vm_pgoff internally. So we have to subtract it from here.
2978 * We are allowed to do this because we are the mm; do not copy
2979 * this code into drivers!
2981 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2983 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2987 struct page
*page
= *pages
;
2993 return VM_FAULT_SIGBUS
;
2997 * Having a close hook prevents vma merging regardless of flags.
2999 static void special_mapping_close(struct vm_area_struct
*vma
)
3003 static const struct vm_operations_struct special_mapping_vmops
= {
3004 .close
= special_mapping_close
,
3005 .fault
= special_mapping_fault
,
3009 * Called with mm->mmap_sem held for writing.
3010 * Insert a new vma covering the given region, with the given flags.
3011 * Its pages are supplied by the given array of struct page *.
3012 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3013 * The region past the last page supplied will always produce SIGBUS.
3014 * The array pointer and the pages it points to are assumed to stay alive
3015 * for as long as this mapping might exist.
3017 int install_special_mapping(struct mm_struct
*mm
,
3018 unsigned long addr
, unsigned long len
,
3019 unsigned long vm_flags
, struct page
**pages
)
3022 struct vm_area_struct
*vma
;
3024 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
3025 if (unlikely(vma
== NULL
))
3028 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
3030 vma
->vm_start
= addr
;
3031 vma
->vm_end
= addr
+ len
;
3033 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
3034 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3036 vma
->vm_ops
= &special_mapping_vmops
;
3037 vma
->vm_private_data
= pages
;
3039 ret
= insert_vm_struct(mm
, vma
);
3043 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3045 perf_event_mmap(vma
);
3050 kmem_cache_free(vm_area_cachep
, vma
);
3054 static DEFINE_MUTEX(mm_all_locks_mutex
);
3056 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3058 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3060 * The LSB of head.next can't change from under us
3061 * because we hold the mm_all_locks_mutex.
3063 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3065 * We can safely modify head.next after taking the
3066 * anon_vma->root->rwsem. If some other vma in this mm shares
3067 * the same anon_vma we won't take it again.
3069 * No need of atomic instructions here, head.next
3070 * can't change from under us thanks to the
3071 * anon_vma->root->rwsem.
3073 if (__test_and_set_bit(0, (unsigned long *)
3074 &anon_vma
->root
->rb_root
.rb_node
))
3079 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3081 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3083 * AS_MM_ALL_LOCKS can't change from under us because
3084 * we hold the mm_all_locks_mutex.
3086 * Operations on ->flags have to be atomic because
3087 * even if AS_MM_ALL_LOCKS is stable thanks to the
3088 * mm_all_locks_mutex, there may be other cpus
3089 * changing other bitflags in parallel to us.
3091 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3093 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
3098 * This operation locks against the VM for all pte/vma/mm related
3099 * operations that could ever happen on a certain mm. This includes
3100 * vmtruncate, try_to_unmap, and all page faults.
3102 * The caller must take the mmap_sem in write mode before calling
3103 * mm_take_all_locks(). The caller isn't allowed to release the
3104 * mmap_sem until mm_drop_all_locks() returns.
3106 * mmap_sem in write mode is required in order to block all operations
3107 * that could modify pagetables and free pages without need of
3108 * altering the vma layout (for example populate_range() with
3109 * nonlinear vmas). It's also needed in write mode to avoid new
3110 * anon_vmas to be associated with existing vmas.
3112 * A single task can't take more than one mm_take_all_locks() in a row
3113 * or it would deadlock.
3115 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3116 * mapping->flags avoid to take the same lock twice, if more than one
3117 * vma in this mm is backed by the same anon_vma or address_space.
3119 * We can take all the locks in random order because the VM code
3120 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3121 * takes more than one of them in a row. Secondly we're protected
3122 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3124 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3125 * that may have to take thousand of locks.
3127 * mm_take_all_locks() can fail if it's interrupted by signals.
3129 int mm_take_all_locks(struct mm_struct
*mm
)
3131 struct vm_area_struct
*vma
;
3132 struct anon_vma_chain
*avc
;
3134 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3136 mutex_lock(&mm_all_locks_mutex
);
3138 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3139 if (signal_pending(current
))
3141 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3142 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3145 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3146 if (signal_pending(current
))
3149 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3150 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3156 mm_drop_all_locks(mm
);
3160 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3162 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3164 * The LSB of head.next can't change to 0 from under
3165 * us because we hold the mm_all_locks_mutex.
3167 * We must however clear the bitflag before unlocking
3168 * the vma so the users using the anon_vma->rb_root will
3169 * never see our bitflag.
3171 * No need of atomic instructions here, head.next
3172 * can't change from under us until we release the
3173 * anon_vma->root->rwsem.
3175 if (!__test_and_clear_bit(0, (unsigned long *)
3176 &anon_vma
->root
->rb_root
.rb_node
))
3178 anon_vma_unlock_write(anon_vma
);
3182 static void vm_unlock_mapping(struct address_space
*mapping
)
3184 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3186 * AS_MM_ALL_LOCKS can't change to 0 from under us
3187 * because we hold the mm_all_locks_mutex.
3189 mutex_unlock(&mapping
->i_mmap_mutex
);
3190 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3197 * The mmap_sem cannot be released by the caller until
3198 * mm_drop_all_locks() returns.
3200 void mm_drop_all_locks(struct mm_struct
*mm
)
3202 struct vm_area_struct
*vma
;
3203 struct anon_vma_chain
*avc
;
3205 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3206 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3208 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3210 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3211 vm_unlock_anon_vma(avc
->anon_vma
);
3212 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3213 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3216 mutex_unlock(&mm_all_locks_mutex
);
3220 * initialise the VMA slab
3222 void __init
mmap_init(void)
3226 ret
= percpu_counter_init(&vm_committed_as
, 0);
3231 * Initialise sysctl_user_reserve_kbytes.
3233 * This is intended to prevent a user from starting a single memory hogging
3234 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3237 * The default value is min(3% of free memory, 128MB)
3238 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3240 static int init_user_reserve(void)
3242 unsigned long free_kbytes
;
3244 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3246 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3249 module_init(init_user_reserve
)
3252 * Initialise sysctl_admin_reserve_kbytes.
3254 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3255 * to log in and kill a memory hogging process.
3257 * Systems with more than 256MB will reserve 8MB, enough to recover
3258 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3259 * only reserve 3% of free pages by default.
3261 static int init_admin_reserve(void)
3263 unsigned long free_kbytes
;
3265 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3267 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3270 module_init(init_admin_reserve
)
3273 * Reinititalise user and admin reserves if memory is added or removed.
3275 * The default user reserve max is 128MB, and the default max for the
3276 * admin reserve is 8MB. These are usually, but not always, enough to
3277 * enable recovery from a memory hogging process using login/sshd, a shell,
3278 * and tools like top. It may make sense to increase or even disable the
3279 * reserve depending on the existence of swap or variations in the recovery
3280 * tools. So, the admin may have changed them.
3282 * If memory is added and the reserves have been eliminated or increased above
3283 * the default max, then we'll trust the admin.
3285 * If memory is removed and there isn't enough free memory, then we
3286 * need to reset the reserves.
3288 * Otherwise keep the reserve set by the admin.
3290 static int reserve_mem_notifier(struct notifier_block
*nb
,
3291 unsigned long action
, void *data
)
3293 unsigned long tmp
, free_kbytes
;
3297 /* Default max is 128MB. Leave alone if modified by operator. */
3298 tmp
= sysctl_user_reserve_kbytes
;
3299 if (0 < tmp
&& tmp
< (1UL << 17))
3300 init_user_reserve();
3302 /* Default max is 8MB. Leave alone if modified by operator. */
3303 tmp
= sysctl_admin_reserve_kbytes
;
3304 if (0 < tmp
&& tmp
< (1UL << 13))
3305 init_admin_reserve();
3309 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3311 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3312 init_user_reserve();
3313 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3314 sysctl_user_reserve_kbytes
);
3317 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3318 init_admin_reserve();
3319 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3320 sysctl_admin_reserve_kbytes
);
3329 static struct notifier_block reserve_mem_nb
= {
3330 .notifier_call
= reserve_mem_notifier
,
3333 static int __meminit
init_reserve_notifier(void)
3335 if (register_hotmemory_notifier(&reserve_mem_nb
))
3336 printk("Failed registering memory add/remove notifier for admin reserve");
3340 module_init(init_reserve_notifier
)