6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/profile.h>
29 #include <linux/export.h>
30 #include <linux/mount.h>
31 #include <linux/mempolicy.h>
32 #include <linux/rmap.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/mmdebug.h>
35 #include <linux/perf_event.h>
36 #include <linux/audit.h>
37 #include <linux/khugepaged.h>
38 #include <linux/uprobes.h>
39 #include <linux/rbtree_augmented.h>
40 #include <linux/sched/sysctl.h>
41 #include <linux/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
45 #include <asm/uaccess.h>
46 #include <asm/cacheflush.h>
48 #include <asm/mmu_context.h>
52 #ifndef arch_mmap_check
53 #define arch_mmap_check(addr, len, flags) (0)
56 #ifndef arch_rebalance_pgtables
57 #define arch_rebalance_pgtables(addr, len) (addr)
60 static void unmap_region(struct mm_struct
*mm
,
61 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
62 unsigned long start
, unsigned long end
);
64 /* description of effects of mapping type and prot in current implementation.
65 * this is due to the limited x86 page protection hardware. The expected
66 * behavior is in parens:
69 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
70 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (yes) yes w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
74 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
75 * w: (no) no w: (no) no w: (copy) copy w: (no) no
76 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
79 pgprot_t protection_map
[16] = {
80 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
81 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
84 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
86 return __pgprot(pgprot_val(protection_map
[vm_flags
&
87 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
88 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
90 EXPORT_SYMBOL(vm_get_page_prot
);
92 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
94 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
97 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
98 void vma_set_page_prot(struct vm_area_struct
*vma
)
100 unsigned long vm_flags
= vma
->vm_flags
;
102 vma
->vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
103 if (vma_wants_writenotify(vma
)) {
104 vm_flags
&= ~VM_SHARED
;
105 vma
->vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
,
111 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
112 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
113 unsigned long sysctl_overcommit_kbytes __read_mostly
;
114 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
115 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
116 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
118 * Make sure vm_committed_as in one cacheline and not cacheline shared with
119 * other variables. It can be updated by several CPUs frequently.
121 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
124 * The global memory commitment made in the system can be a metric
125 * that can be used to drive ballooning decisions when Linux is hosted
126 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
127 * balancing memory across competing virtual machines that are hosted.
128 * Several metrics drive this policy engine including the guest reported
131 unsigned long vm_memory_committed(void)
133 return percpu_counter_read_positive(&vm_committed_as
);
135 EXPORT_SYMBOL_GPL(vm_memory_committed
);
138 * Check that a process has enough memory to allocate a new virtual
139 * mapping. 0 means there is enough memory for the allocation to
140 * succeed and -ENOMEM implies there is not.
142 * We currently support three overcommit policies, which are set via the
143 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
145 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
146 * Additional code 2002 Jul 20 by Robert Love.
148 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
150 * Note this is a helper function intended to be used by LSMs which
151 * wish to use this logic.
153 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
155 unsigned long free
, allowed
, reserve
;
157 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as
) <
158 -(s64
)vm_committed_as_batch
* num_online_cpus(),
159 "memory commitment underflow");
161 vm_acct_memory(pages
);
164 * Sometimes we want to use more memory than we have
166 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
169 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
170 free
= global_page_state(NR_FREE_PAGES
);
171 free
+= global_page_state(NR_FILE_PAGES
);
174 * shmem pages shouldn't be counted as free in this
175 * case, they can't be purged, only swapped out, and
176 * that won't affect the overall amount of available
177 * memory in the system.
179 free
-= global_page_state(NR_SHMEM
);
181 free
+= get_nr_swap_pages();
184 * Any slabs which are created with the
185 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
186 * which are reclaimable, under pressure. The dentry
187 * cache and most inode caches should fall into this
189 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
192 * Leave reserved pages. The pages are not for anonymous pages.
194 if (free
<= totalreserve_pages
)
197 free
-= totalreserve_pages
;
200 * Reserve some for root
203 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
211 allowed
= vm_commit_limit();
213 * Reserve some for root
216 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
219 * Don't let a single process grow so big a user can't recover
222 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
223 allowed
-= min(mm
->total_vm
/ 32, reserve
);
226 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
229 vm_unacct_memory(pages
);
235 * Requires inode->i_mapping->i_mmap_mutex
237 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
238 struct file
*file
, struct address_space
*mapping
)
240 if (vma
->vm_flags
& VM_DENYWRITE
)
241 atomic_inc(&file_inode(file
)->i_writecount
);
242 if (vma
->vm_flags
& VM_SHARED
)
243 mapping_unmap_writable(mapping
);
245 flush_dcache_mmap_lock(mapping
);
246 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
247 list_del_init(&vma
->shared
.nonlinear
);
249 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
250 flush_dcache_mmap_unlock(mapping
);
254 * Unlink a file-based vm structure from its interval tree, to hide
255 * vma from rmap and vmtruncate before freeing its page tables.
257 void unlink_file_vma(struct vm_area_struct
*vma
)
259 struct file
*file
= vma
->vm_file
;
262 struct address_space
*mapping
= file
->f_mapping
;
263 mutex_lock(&mapping
->i_mmap_mutex
);
264 __remove_shared_vm_struct(vma
, file
, mapping
);
265 mutex_unlock(&mapping
->i_mmap_mutex
);
270 * Close a vm structure and free it, returning the next.
272 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
274 struct vm_area_struct
*next
= vma
->vm_next
;
277 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
278 vma
->vm_ops
->close(vma
);
281 mpol_put(vma_policy(vma
));
282 kmem_cache_free(vm_area_cachep
, vma
);
286 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
288 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
290 unsigned long retval
;
291 unsigned long newbrk
, oldbrk
;
292 struct mm_struct
*mm
= current
->mm
;
293 unsigned long min_brk
;
296 down_write(&mm
->mmap_sem
);
298 #ifdef CONFIG_COMPAT_BRK
300 * CONFIG_COMPAT_BRK can still be overridden by setting
301 * randomize_va_space to 2, which will still cause mm->start_brk
302 * to be arbitrarily shifted
304 if (current
->brk_randomized
)
305 min_brk
= mm
->start_brk
;
307 min_brk
= mm
->end_data
;
309 min_brk
= mm
->start_brk
;
315 * Check against rlimit here. If this check is done later after the test
316 * of oldbrk with newbrk then it can escape the test and let the data
317 * segment grow beyond its set limit the in case where the limit is
318 * not page aligned -Ram Gupta
320 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
321 mm
->end_data
, mm
->start_data
))
324 newbrk
= PAGE_ALIGN(brk
);
325 oldbrk
= PAGE_ALIGN(mm
->brk
);
326 if (oldbrk
== newbrk
)
329 /* Always allow shrinking brk. */
330 if (brk
<= mm
->brk
) {
331 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
336 /* Check against existing mmap mappings. */
337 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
340 /* Ok, looks good - let it rip. */
341 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
346 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
347 up_write(&mm
->mmap_sem
);
349 mm_populate(oldbrk
, newbrk
- oldbrk
);
354 up_write(&mm
->mmap_sem
);
358 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
360 unsigned long max
, subtree_gap
;
363 max
-= vma
->vm_prev
->vm_end
;
364 if (vma
->vm_rb
.rb_left
) {
365 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
366 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
367 if (subtree_gap
> max
)
370 if (vma
->vm_rb
.rb_right
) {
371 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
372 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
373 if (subtree_gap
> max
)
379 #ifdef CONFIG_DEBUG_VM_RB
380 static int browse_rb(struct rb_root
*root
)
382 int i
= 0, j
, bug
= 0;
383 struct rb_node
*nd
, *pn
= NULL
;
384 unsigned long prev
= 0, pend
= 0;
386 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
387 struct vm_area_struct
*vma
;
388 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
389 if (vma
->vm_start
< prev
) {
390 pr_emerg("vm_start %lx < prev %lx\n",
391 vma
->vm_start
, prev
);
394 if (vma
->vm_start
< pend
) {
395 pr_emerg("vm_start %lx < pend %lx\n",
396 vma
->vm_start
, pend
);
399 if (vma
->vm_start
> vma
->vm_end
) {
400 pr_emerg("vm_start %lx > vm_end %lx\n",
401 vma
->vm_start
, vma
->vm_end
);
404 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
405 pr_emerg("free gap %lx, correct %lx\n",
407 vma_compute_subtree_gap(vma
));
412 prev
= vma
->vm_start
;
416 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
419 pr_emerg("backwards %d, forwards %d\n", j
, i
);
425 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
429 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
430 struct vm_area_struct
*vma
;
431 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
432 VM_BUG_ON_VMA(vma
!= ignore
&&
433 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
438 static void validate_mm(struct mm_struct
*mm
)
442 unsigned long highest_address
= 0;
443 struct vm_area_struct
*vma
= mm
->mmap
;
446 struct anon_vma_chain
*avc
;
448 vma_lock_anon_vma(vma
);
449 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
450 anon_vma_interval_tree_verify(avc
);
451 vma_unlock_anon_vma(vma
);
452 highest_address
= vma
->vm_end
;
456 if (i
!= mm
->map_count
) {
457 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
460 if (highest_address
!= mm
->highest_vm_end
) {
461 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
462 mm
->highest_vm_end
, highest_address
);
465 i
= browse_rb(&mm
->mm_rb
);
466 if (i
!= mm
->map_count
) {
468 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
471 VM_BUG_ON_MM(bug
, mm
);
474 #define validate_mm_rb(root, ignore) do { } while (0)
475 #define validate_mm(mm) do { } while (0)
478 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
479 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
482 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
483 * vma->vm_prev->vm_end values changed, without modifying the vma's position
486 static void vma_gap_update(struct vm_area_struct
*vma
)
489 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
490 * function that does exacltly what we want.
492 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
495 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
496 struct rb_root
*root
)
498 /* All rb_subtree_gap values must be consistent prior to insertion */
499 validate_mm_rb(root
, NULL
);
501 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
504 static void vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
507 * All rb_subtree_gap values must be consistent prior to erase,
508 * with the possible exception of the vma being erased.
510 validate_mm_rb(root
, vma
);
513 * Note rb_erase_augmented is a fairly large inline function,
514 * so make sure we instantiate it only once with our desired
515 * augmented rbtree callbacks.
517 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
521 * vma has some anon_vma assigned, and is already inserted on that
522 * anon_vma's interval trees.
524 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
525 * vma must be removed from the anon_vma's interval trees using
526 * anon_vma_interval_tree_pre_update_vma().
528 * After the update, the vma will be reinserted using
529 * anon_vma_interval_tree_post_update_vma().
531 * The entire update must be protected by exclusive mmap_sem and by
532 * the root anon_vma's mutex.
535 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
537 struct anon_vma_chain
*avc
;
539 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
540 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
544 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
546 struct anon_vma_chain
*avc
;
548 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
549 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
552 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
553 unsigned long end
, struct vm_area_struct
**pprev
,
554 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
556 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
558 __rb_link
= &mm
->mm_rb
.rb_node
;
559 rb_prev
= __rb_parent
= NULL
;
562 struct vm_area_struct
*vma_tmp
;
564 __rb_parent
= *__rb_link
;
565 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
567 if (vma_tmp
->vm_end
> addr
) {
568 /* Fail if an existing vma overlaps the area */
569 if (vma_tmp
->vm_start
< end
)
571 __rb_link
= &__rb_parent
->rb_left
;
573 rb_prev
= __rb_parent
;
574 __rb_link
= &__rb_parent
->rb_right
;
580 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
581 *rb_link
= __rb_link
;
582 *rb_parent
= __rb_parent
;
586 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
587 unsigned long addr
, unsigned long end
)
589 unsigned long nr_pages
= 0;
590 struct vm_area_struct
*vma
;
592 /* Find first overlaping mapping */
593 vma
= find_vma_intersection(mm
, addr
, end
);
597 nr_pages
= (min(end
, vma
->vm_end
) -
598 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
600 /* Iterate over the rest of the overlaps */
601 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
602 unsigned long overlap_len
;
604 if (vma
->vm_start
> end
)
607 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
608 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
614 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
615 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
617 /* Update tracking information for the gap following the new vma. */
619 vma_gap_update(vma
->vm_next
);
621 mm
->highest_vm_end
= vma
->vm_end
;
624 * vma->vm_prev wasn't known when we followed the rbtree to find the
625 * correct insertion point for that vma. As a result, we could not
626 * update the vma vm_rb parents rb_subtree_gap values on the way down.
627 * So, we first insert the vma with a zero rb_subtree_gap value
628 * (to be consistent with what we did on the way down), and then
629 * immediately update the gap to the correct value. Finally we
630 * rebalance the rbtree after all augmented values have been set.
632 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
633 vma
->rb_subtree_gap
= 0;
635 vma_rb_insert(vma
, &mm
->mm_rb
);
638 static void __vma_link_file(struct vm_area_struct
*vma
)
644 struct address_space
*mapping
= file
->f_mapping
;
646 if (vma
->vm_flags
& VM_DENYWRITE
)
647 atomic_dec(&file_inode(file
)->i_writecount
);
648 if (vma
->vm_flags
& VM_SHARED
)
649 atomic_inc(&mapping
->i_mmap_writable
);
651 flush_dcache_mmap_lock(mapping
);
652 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
653 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
655 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
656 flush_dcache_mmap_unlock(mapping
);
661 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
662 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
663 struct rb_node
*rb_parent
)
665 __vma_link_list(mm
, vma
, prev
, rb_parent
);
666 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
669 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
670 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
671 struct rb_node
*rb_parent
)
673 struct address_space
*mapping
= NULL
;
676 mapping
= vma
->vm_file
->f_mapping
;
677 mutex_lock(&mapping
->i_mmap_mutex
);
680 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
681 __vma_link_file(vma
);
684 mutex_unlock(&mapping
->i_mmap_mutex
);
691 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
692 * mm's list and rbtree. It has already been inserted into the interval tree.
694 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
696 struct vm_area_struct
*prev
;
697 struct rb_node
**rb_link
, *rb_parent
;
699 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
700 &prev
, &rb_link
, &rb_parent
))
702 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
707 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
708 struct vm_area_struct
*prev
)
710 struct vm_area_struct
*next
;
712 vma_rb_erase(vma
, &mm
->mm_rb
);
713 prev
->vm_next
= next
= vma
->vm_next
;
715 next
->vm_prev
= prev
;
718 vmacache_invalidate(mm
);
722 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
723 * is already present in an i_mmap tree without adjusting the tree.
724 * The following helper function should be used when such adjustments
725 * are necessary. The "insert" vma (if any) is to be inserted
726 * before we drop the necessary locks.
728 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
729 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
731 struct mm_struct
*mm
= vma
->vm_mm
;
732 struct vm_area_struct
*next
= vma
->vm_next
;
733 struct vm_area_struct
*importer
= NULL
;
734 struct address_space
*mapping
= NULL
;
735 struct rb_root
*root
= NULL
;
736 struct anon_vma
*anon_vma
= NULL
;
737 struct file
*file
= vma
->vm_file
;
738 bool start_changed
= false, end_changed
= false;
739 long adjust_next
= 0;
742 if (next
&& !insert
) {
743 struct vm_area_struct
*exporter
= NULL
;
745 if (end
>= next
->vm_end
) {
747 * vma expands, overlapping all the next, and
748 * perhaps the one after too (mprotect case 6).
750 again
: remove_next
= 1 + (end
> next
->vm_end
);
754 } else if (end
> next
->vm_start
) {
756 * vma expands, overlapping part of the next:
757 * mprotect case 5 shifting the boundary up.
759 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
762 } else if (end
< vma
->vm_end
) {
764 * vma shrinks, and !insert tells it's not
765 * split_vma inserting another: so it must be
766 * mprotect case 4 shifting the boundary down.
768 adjust_next
= -((vma
->vm_end
- end
) >> PAGE_SHIFT
);
774 * Easily overlooked: when mprotect shifts the boundary,
775 * make sure the expanding vma has anon_vma set if the
776 * shrinking vma had, to cover any anon pages imported.
778 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
781 error
= anon_vma_clone(importer
, exporter
);
784 importer
->anon_vma
= exporter
->anon_vma
;
789 mapping
= file
->f_mapping
;
790 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
791 root
= &mapping
->i_mmap
;
792 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
795 uprobe_munmap(next
, next
->vm_start
,
799 mutex_lock(&mapping
->i_mmap_mutex
);
802 * Put into interval tree now, so instantiated pages
803 * are visible to arm/parisc __flush_dcache_page
804 * throughout; but we cannot insert into address
805 * space until vma start or end is updated.
807 __vma_link_file(insert
);
811 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
813 anon_vma
= vma
->anon_vma
;
814 if (!anon_vma
&& adjust_next
)
815 anon_vma
= next
->anon_vma
;
817 VM_BUG_ON_VMA(adjust_next
&& next
->anon_vma
&&
818 anon_vma
!= next
->anon_vma
, next
);
819 anon_vma_lock_write(anon_vma
);
820 anon_vma_interval_tree_pre_update_vma(vma
);
822 anon_vma_interval_tree_pre_update_vma(next
);
826 flush_dcache_mmap_lock(mapping
);
827 vma_interval_tree_remove(vma
, root
);
829 vma_interval_tree_remove(next
, root
);
832 if (start
!= vma
->vm_start
) {
833 vma
->vm_start
= start
;
834 start_changed
= true;
836 if (end
!= vma
->vm_end
) {
840 vma
->vm_pgoff
= pgoff
;
842 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
843 next
->vm_pgoff
+= adjust_next
;
848 vma_interval_tree_insert(next
, root
);
849 vma_interval_tree_insert(vma
, root
);
850 flush_dcache_mmap_unlock(mapping
);
855 * vma_merge has merged next into vma, and needs
856 * us to remove next before dropping the locks.
858 __vma_unlink(mm
, next
, vma
);
860 __remove_shared_vm_struct(next
, file
, mapping
);
863 * split_vma has split insert from vma, and needs
864 * us to insert it before dropping the locks
865 * (it may either follow vma or precede it).
867 __insert_vm_struct(mm
, insert
);
873 mm
->highest_vm_end
= end
;
874 else if (!adjust_next
)
875 vma_gap_update(next
);
880 anon_vma_interval_tree_post_update_vma(vma
);
882 anon_vma_interval_tree_post_update_vma(next
);
883 anon_vma_unlock_write(anon_vma
);
886 mutex_unlock(&mapping
->i_mmap_mutex
);
897 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
901 anon_vma_merge(vma
, next
);
903 mpol_put(vma_policy(next
));
904 kmem_cache_free(vm_area_cachep
, next
);
906 * In mprotect's case 6 (see comments on vma_merge),
907 * we must remove another next too. It would clutter
908 * up the code too much to do both in one go.
911 if (remove_next
== 2)
914 vma_gap_update(next
);
916 mm
->highest_vm_end
= end
;
927 * If the vma has a ->close operation then the driver probably needs to release
928 * per-vma resources, so we don't attempt to merge those.
930 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
931 struct file
*file
, unsigned long vm_flags
)
934 * VM_SOFTDIRTY should not prevent from VMA merging, if we
935 * match the flags but dirty bit -- the caller should mark
936 * merged VMA as dirty. If dirty bit won't be excluded from
937 * comparison, we increase pressue on the memory system forcing
938 * the kernel to generate new VMAs when old one could be
941 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
943 if (vma
->vm_file
!= file
)
945 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
950 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
951 struct anon_vma
*anon_vma2
,
952 struct vm_area_struct
*vma
)
955 * The list_is_singular() test is to avoid merging VMA cloned from
956 * parents. This can improve scalability caused by anon_vma lock.
958 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
959 list_is_singular(&vma
->anon_vma_chain
)))
961 return anon_vma1
== anon_vma2
;
965 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
966 * in front of (at a lower virtual address and file offset than) the vma.
968 * We cannot merge two vmas if they have differently assigned (non-NULL)
969 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
971 * We don't check here for the merged mmap wrapping around the end of pagecache
972 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
973 * wrap, nor mmaps which cover the final page at index -1UL.
976 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
977 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
979 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
980 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
981 if (vma
->vm_pgoff
== vm_pgoff
)
988 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
989 * beyond (at a higher virtual address and file offset than) the vma.
991 * We cannot merge two vmas if they have differently assigned (non-NULL)
992 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
995 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
996 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
998 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
999 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1001 vm_pglen
= vma_pages(vma
);
1002 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1009 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1010 * whether that can be merged with its predecessor or its successor.
1011 * Or both (it neatly fills a hole).
1013 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1014 * certain not to be mapped by the time vma_merge is called; but when
1015 * called for mprotect, it is certain to be already mapped (either at
1016 * an offset within prev, or at the start of next), and the flags of
1017 * this area are about to be changed to vm_flags - and the no-change
1018 * case has already been eliminated.
1020 * The following mprotect cases have to be considered, where AAAA is
1021 * the area passed down from mprotect_fixup, never extending beyond one
1022 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1024 * AAAA AAAA AAAA AAAA
1025 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1026 * cannot merge might become might become might become
1027 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1028 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1029 * mremap move: PPPPNNNNNNNN 8
1031 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1032 * might become case 1 below case 2 below case 3 below
1034 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1035 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1037 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1038 struct vm_area_struct
*prev
, unsigned long addr
,
1039 unsigned long end
, unsigned long vm_flags
,
1040 struct anon_vma
*anon_vma
, struct file
*file
,
1041 pgoff_t pgoff
, struct mempolicy
*policy
)
1043 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1044 struct vm_area_struct
*area
, *next
;
1048 * We later require that vma->vm_flags == vm_flags,
1049 * so this tests vma->vm_flags & VM_SPECIAL, too.
1051 if (vm_flags
& VM_SPECIAL
)
1055 next
= prev
->vm_next
;
1059 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
1060 next
= next
->vm_next
;
1063 * Can it merge with the predecessor?
1065 if (prev
&& prev
->vm_end
== addr
&&
1066 mpol_equal(vma_policy(prev
), policy
) &&
1067 can_vma_merge_after(prev
, vm_flags
,
1068 anon_vma
, file
, pgoff
)) {
1070 * OK, it can. Can we now merge in the successor as well?
1072 if (next
&& end
== next
->vm_start
&&
1073 mpol_equal(policy
, vma_policy(next
)) &&
1074 can_vma_merge_before(next
, vm_flags
,
1075 anon_vma
, file
, pgoff
+pglen
) &&
1076 is_mergeable_anon_vma(prev
->anon_vma
,
1077 next
->anon_vma
, NULL
)) {
1079 err
= vma_adjust(prev
, prev
->vm_start
,
1080 next
->vm_end
, prev
->vm_pgoff
, NULL
);
1081 } else /* cases 2, 5, 7 */
1082 err
= vma_adjust(prev
, prev
->vm_start
,
1083 end
, prev
->vm_pgoff
, NULL
);
1086 khugepaged_enter_vma_merge(prev
, vm_flags
);
1091 * Can this new request be merged in front of next?
1093 if (next
&& end
== next
->vm_start
&&
1094 mpol_equal(policy
, vma_policy(next
)) &&
1095 can_vma_merge_before(next
, vm_flags
,
1096 anon_vma
, file
, pgoff
+pglen
)) {
1097 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1098 err
= vma_adjust(prev
, prev
->vm_start
,
1099 addr
, prev
->vm_pgoff
, NULL
);
1100 else /* cases 3, 8 */
1101 err
= vma_adjust(area
, addr
, next
->vm_end
,
1102 next
->vm_pgoff
- pglen
, NULL
);
1105 khugepaged_enter_vma_merge(area
, vm_flags
);
1113 * Rough compatbility check to quickly see if it's even worth looking
1114 * at sharing an anon_vma.
1116 * They need to have the same vm_file, and the flags can only differ
1117 * in things that mprotect may change.
1119 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1120 * we can merge the two vma's. For example, we refuse to merge a vma if
1121 * there is a vm_ops->close() function, because that indicates that the
1122 * driver is doing some kind of reference counting. But that doesn't
1123 * really matter for the anon_vma sharing case.
1125 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1127 return a
->vm_end
== b
->vm_start
&&
1128 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1129 a
->vm_file
== b
->vm_file
&&
1130 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
|VM_SOFTDIRTY
)) &&
1131 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1135 * Do some basic sanity checking to see if we can re-use the anon_vma
1136 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1137 * the same as 'old', the other will be the new one that is trying
1138 * to share the anon_vma.
1140 * NOTE! This runs with mm_sem held for reading, so it is possible that
1141 * the anon_vma of 'old' is concurrently in the process of being set up
1142 * by another page fault trying to merge _that_. But that's ok: if it
1143 * is being set up, that automatically means that it will be a singleton
1144 * acceptable for merging, so we can do all of this optimistically. But
1145 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
1147 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1148 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1149 * is to return an anon_vma that is "complex" due to having gone through
1152 * We also make sure that the two vma's are compatible (adjacent,
1153 * and with the same memory policies). That's all stable, even with just
1154 * a read lock on the mm_sem.
1156 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1158 if (anon_vma_compatible(a
, b
)) {
1159 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
1161 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1168 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1169 * neighbouring vmas for a suitable anon_vma, before it goes off
1170 * to allocate a new anon_vma. It checks because a repetitive
1171 * sequence of mprotects and faults may otherwise lead to distinct
1172 * anon_vmas being allocated, preventing vma merge in subsequent
1175 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1177 struct anon_vma
*anon_vma
;
1178 struct vm_area_struct
*near
;
1180 near
= vma
->vm_next
;
1184 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1188 near
= vma
->vm_prev
;
1192 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1197 * There's no absolute need to look only at touching neighbours:
1198 * we could search further afield for "compatible" anon_vmas.
1199 * But it would probably just be a waste of time searching,
1200 * or lead to too many vmas hanging off the same anon_vma.
1201 * We're trying to allow mprotect remerging later on,
1202 * not trying to minimize memory used for anon_vmas.
1207 #ifdef CONFIG_PROC_FS
1208 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
1209 struct file
*file
, long pages
)
1211 const unsigned long stack_flags
1212 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
1214 mm
->total_vm
+= pages
;
1217 mm
->shared_vm
+= pages
;
1218 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
1219 mm
->exec_vm
+= pages
;
1220 } else if (flags
& stack_flags
)
1221 mm
->stack_vm
+= pages
;
1223 #endif /* CONFIG_PROC_FS */
1226 * If a hint addr is less than mmap_min_addr change hint to be as
1227 * low as possible but still greater than mmap_min_addr
1229 static inline unsigned long round_hint_to_min(unsigned long hint
)
1232 if (((void *)hint
!= NULL
) &&
1233 (hint
< mmap_min_addr
))
1234 return PAGE_ALIGN(mmap_min_addr
);
1238 static inline int mlock_future_check(struct mm_struct
*mm
,
1239 unsigned long flags
,
1242 unsigned long locked
, lock_limit
;
1244 /* mlock MCL_FUTURE? */
1245 if (flags
& VM_LOCKED
) {
1246 locked
= len
>> PAGE_SHIFT
;
1247 locked
+= mm
->locked_vm
;
1248 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1249 lock_limit
>>= PAGE_SHIFT
;
1250 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1257 * The caller must hold down_write(¤t->mm->mmap_sem).
1260 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
1261 unsigned long len
, unsigned long prot
,
1262 unsigned long flags
, unsigned long pgoff
,
1263 unsigned long *populate
)
1265 struct mm_struct
*mm
= current
->mm
;
1266 vm_flags_t vm_flags
;
1271 * Does the application expect PROT_READ to imply PROT_EXEC?
1273 * (the exception is when the underlying filesystem is noexec
1274 * mounted, in which case we dont add PROT_EXEC.)
1276 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1277 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
1283 if (!(flags
& MAP_FIXED
))
1284 addr
= round_hint_to_min(addr
);
1286 /* Careful about overflows.. */
1287 len
= PAGE_ALIGN(len
);
1291 /* offset overflow? */
1292 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1295 /* Too many mappings? */
1296 if (mm
->map_count
> sysctl_max_map_count
)
1299 /* Obtain the address to map to. we verify (or select) it and ensure
1300 * that it represents a valid section of the address space.
1302 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1303 if (addr
& ~PAGE_MASK
)
1306 /* Do simple checking here so the lower-level routines won't have
1307 * to. we assume access permissions have been handled by the open
1308 * of the memory object, so we don't do any here.
1310 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1311 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1313 if (flags
& MAP_LOCKED
)
1314 if (!can_do_mlock())
1317 if (mlock_future_check(mm
, vm_flags
, len
))
1321 struct inode
*inode
= file_inode(file
);
1323 switch (flags
& MAP_TYPE
) {
1325 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1329 * Make sure we don't allow writing to an append-only
1332 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1336 * Make sure there are no mandatory locks on the file.
1338 if (locks_verify_locked(file
))
1341 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1342 if (!(file
->f_mode
& FMODE_WRITE
))
1343 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1347 if (!(file
->f_mode
& FMODE_READ
))
1349 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1350 if (vm_flags
& VM_EXEC
)
1352 vm_flags
&= ~VM_MAYEXEC
;
1355 if (!file
->f_op
->mmap
)
1357 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1365 switch (flags
& MAP_TYPE
) {
1367 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1373 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1377 * Set pgoff according to addr for anon_vma.
1379 pgoff
= addr
>> PAGE_SHIFT
;
1387 * Set 'VM_NORESERVE' if we should not account for the
1388 * memory use of this mapping.
1390 if (flags
& MAP_NORESERVE
) {
1391 /* We honor MAP_NORESERVE if allowed to overcommit */
1392 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1393 vm_flags
|= VM_NORESERVE
;
1395 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1396 if (file
&& is_file_hugepages(file
))
1397 vm_flags
|= VM_NORESERVE
;
1400 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
);
1401 if (!IS_ERR_VALUE(addr
) &&
1402 ((vm_flags
& VM_LOCKED
) ||
1403 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1408 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1409 unsigned long, prot
, unsigned long, flags
,
1410 unsigned long, fd
, unsigned long, pgoff
)
1412 struct file
*file
= NULL
;
1413 unsigned long retval
= -EBADF
;
1415 if (!(flags
& MAP_ANONYMOUS
)) {
1416 audit_mmap_fd(fd
, flags
);
1420 if (is_file_hugepages(file
))
1421 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1423 if (unlikely(flags
& MAP_HUGETLB
&& !is_file_hugepages(file
)))
1425 } else if (flags
& MAP_HUGETLB
) {
1426 struct user_struct
*user
= NULL
;
1429 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & SHM_HUGE_MASK
);
1433 len
= ALIGN(len
, huge_page_size(hs
));
1435 * VM_NORESERVE is used because the reservations will be
1436 * taken when vm_ops->mmap() is called
1437 * A dummy user value is used because we are not locking
1438 * memory so no accounting is necessary
1440 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1442 &user
, HUGETLB_ANONHUGE_INODE
,
1443 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1445 return PTR_ERR(file
);
1448 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1450 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1458 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1459 struct mmap_arg_struct
{
1463 unsigned long flags
;
1465 unsigned long offset
;
1468 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1470 struct mmap_arg_struct a
;
1472 if (copy_from_user(&a
, arg
, sizeof(a
)))
1474 if (a
.offset
& ~PAGE_MASK
)
1477 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1478 a
.offset
>> PAGE_SHIFT
);
1480 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1483 * Some shared mappigns will want the pages marked read-only
1484 * to track write events. If so, we'll downgrade vm_page_prot
1485 * to the private version (using protection_map[] without the
1488 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1490 vm_flags_t vm_flags
= vma
->vm_flags
;
1492 /* If it was private or non-writable, the write bit is already clear */
1493 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1496 /* The backer wishes to know when pages are first written to? */
1497 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1500 /* The open routine did something to the protections that pgprot_modify
1501 * won't preserve? */
1502 if (pgprot_val(vma
->vm_page_prot
) !=
1503 pgprot_val(vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
)))
1506 /* Do we need to track softdirty? */
1507 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1510 /* Specialty mapping? */
1511 if (vm_flags
& VM_PFNMAP
)
1514 /* Can the mapping track the dirty pages? */
1515 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1516 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1520 * We account for memory if it's a private writeable mapping,
1521 * not hugepages and VM_NORESERVE wasn't set.
1523 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1526 * hugetlb has its own accounting separate from the core VM
1527 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1529 if (file
&& is_file_hugepages(file
))
1532 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1535 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1536 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
)
1538 struct mm_struct
*mm
= current
->mm
;
1539 struct vm_area_struct
*vma
, *prev
;
1541 struct rb_node
**rb_link
, *rb_parent
;
1542 unsigned long charged
= 0;
1544 /* Check against address space limit. */
1545 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
)) {
1546 unsigned long nr_pages
;
1549 * MAP_FIXED may remove pages of mappings that intersects with
1550 * requested mapping. Account for the pages it would unmap.
1552 if (!(vm_flags
& MAP_FIXED
))
1555 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1557 if (!may_expand_vm(mm
, (len
>> PAGE_SHIFT
) - nr_pages
))
1561 /* Clear old maps */
1564 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
1565 if (do_munmap(mm
, addr
, len
))
1571 * Private writable mapping: check memory availability
1573 if (accountable_mapping(file
, vm_flags
)) {
1574 charged
= len
>> PAGE_SHIFT
;
1575 if (security_vm_enough_memory_mm(mm
, charged
))
1577 vm_flags
|= VM_ACCOUNT
;
1581 * Can we just expand an old mapping?
1583 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1588 * Determine the object being mapped and call the appropriate
1589 * specific mapper. the address has already been validated, but
1590 * not unmapped, but the maps are removed from the list.
1592 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1599 vma
->vm_start
= addr
;
1600 vma
->vm_end
= addr
+ len
;
1601 vma
->vm_flags
= vm_flags
;
1602 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1603 vma
->vm_pgoff
= pgoff
;
1604 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1607 if (vm_flags
& VM_DENYWRITE
) {
1608 error
= deny_write_access(file
);
1612 if (vm_flags
& VM_SHARED
) {
1613 error
= mapping_map_writable(file
->f_mapping
);
1615 goto allow_write_and_free_vma
;
1618 /* ->mmap() can change vma->vm_file, but must guarantee that
1619 * vma_link() below can deny write-access if VM_DENYWRITE is set
1620 * and map writably if VM_SHARED is set. This usually means the
1621 * new file must not have been exposed to user-space, yet.
1623 vma
->vm_file
= get_file(file
);
1624 error
= file
->f_op
->mmap(file
, vma
);
1626 goto unmap_and_free_vma
;
1628 /* Can addr have changed??
1630 * Answer: Yes, several device drivers can do it in their
1631 * f_op->mmap method. -DaveM
1632 * Bug: If addr is changed, prev, rb_link, rb_parent should
1633 * be updated for vma_link()
1635 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1637 addr
= vma
->vm_start
;
1638 vm_flags
= vma
->vm_flags
;
1639 } else if (vm_flags
& VM_SHARED
) {
1640 error
= shmem_zero_setup(vma
);
1645 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1646 /* Once vma denies write, undo our temporary denial count */
1648 if (vm_flags
& VM_SHARED
)
1649 mapping_unmap_writable(file
->f_mapping
);
1650 if (vm_flags
& VM_DENYWRITE
)
1651 allow_write_access(file
);
1653 file
= vma
->vm_file
;
1655 perf_event_mmap(vma
);
1657 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1658 if (vm_flags
& VM_LOCKED
) {
1659 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1660 vma
== get_gate_vma(current
->mm
)))
1661 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1663 vma
->vm_flags
&= ~VM_LOCKED
;
1670 * New (or expanded) vma always get soft dirty status.
1671 * Otherwise user-space soft-dirty page tracker won't
1672 * be able to distinguish situation when vma area unmapped,
1673 * then new mapped in-place (which must be aimed as
1674 * a completely new data area).
1676 vma
->vm_flags
|= VM_SOFTDIRTY
;
1678 vma_set_page_prot(vma
);
1683 vma
->vm_file
= NULL
;
1686 /* Undo any partial mapping done by a device driver. */
1687 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1689 if (vm_flags
& VM_SHARED
)
1690 mapping_unmap_writable(file
->f_mapping
);
1691 allow_write_and_free_vma
:
1692 if (vm_flags
& VM_DENYWRITE
)
1693 allow_write_access(file
);
1695 kmem_cache_free(vm_area_cachep
, vma
);
1698 vm_unacct_memory(charged
);
1702 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1705 * We implement the search by looking for an rbtree node that
1706 * immediately follows a suitable gap. That is,
1707 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1708 * - gap_end = vma->vm_start >= info->low_limit + length;
1709 * - gap_end - gap_start >= length
1712 struct mm_struct
*mm
= current
->mm
;
1713 struct vm_area_struct
*vma
;
1714 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1716 /* Adjust search length to account for worst case alignment overhead */
1717 length
= info
->length
+ info
->align_mask
;
1718 if (length
< info
->length
)
1721 /* Adjust search limits by the desired length */
1722 if (info
->high_limit
< length
)
1724 high_limit
= info
->high_limit
- length
;
1726 if (info
->low_limit
> high_limit
)
1728 low_limit
= info
->low_limit
+ length
;
1730 /* Check if rbtree root looks promising */
1731 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1733 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1734 if (vma
->rb_subtree_gap
< length
)
1738 /* Visit left subtree if it looks promising */
1739 gap_end
= vma
->vm_start
;
1740 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1741 struct vm_area_struct
*left
=
1742 rb_entry(vma
->vm_rb
.rb_left
,
1743 struct vm_area_struct
, vm_rb
);
1744 if (left
->rb_subtree_gap
>= length
) {
1750 gap_start
= vma
->vm_prev
? vma
->vm_prev
->vm_end
: 0;
1752 /* Check if current node has a suitable gap */
1753 if (gap_start
> high_limit
)
1755 if (gap_end
>= low_limit
&& gap_end
- gap_start
>= length
)
1758 /* Visit right subtree if it looks promising */
1759 if (vma
->vm_rb
.rb_right
) {
1760 struct vm_area_struct
*right
=
1761 rb_entry(vma
->vm_rb
.rb_right
,
1762 struct vm_area_struct
, vm_rb
);
1763 if (right
->rb_subtree_gap
>= length
) {
1769 /* Go back up the rbtree to find next candidate node */
1771 struct rb_node
*prev
= &vma
->vm_rb
;
1772 if (!rb_parent(prev
))
1774 vma
= rb_entry(rb_parent(prev
),
1775 struct vm_area_struct
, vm_rb
);
1776 if (prev
== vma
->vm_rb
.rb_left
) {
1777 gap_start
= vma
->vm_prev
->vm_end
;
1778 gap_end
= vma
->vm_start
;
1785 /* Check highest gap, which does not precede any rbtree node */
1786 gap_start
= mm
->highest_vm_end
;
1787 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1788 if (gap_start
> high_limit
)
1792 /* We found a suitable gap. Clip it with the original low_limit. */
1793 if (gap_start
< info
->low_limit
)
1794 gap_start
= info
->low_limit
;
1796 /* Adjust gap address to the desired alignment */
1797 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1799 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1800 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1804 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1806 struct mm_struct
*mm
= current
->mm
;
1807 struct vm_area_struct
*vma
;
1808 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1810 /* Adjust search length to account for worst case alignment overhead */
1811 length
= info
->length
+ info
->align_mask
;
1812 if (length
< info
->length
)
1816 * Adjust search limits by the desired length.
1817 * See implementation comment at top of unmapped_area().
1819 gap_end
= info
->high_limit
;
1820 if (gap_end
< length
)
1822 high_limit
= gap_end
- length
;
1824 if (info
->low_limit
> high_limit
)
1826 low_limit
= info
->low_limit
+ length
;
1828 /* Check highest gap, which does not precede any rbtree node */
1829 gap_start
= mm
->highest_vm_end
;
1830 if (gap_start
<= high_limit
)
1833 /* Check if rbtree root looks promising */
1834 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1836 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1837 if (vma
->rb_subtree_gap
< length
)
1841 /* Visit right subtree if it looks promising */
1842 gap_start
= vma
->vm_prev
? vma
->vm_prev
->vm_end
: 0;
1843 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1844 struct vm_area_struct
*right
=
1845 rb_entry(vma
->vm_rb
.rb_right
,
1846 struct vm_area_struct
, vm_rb
);
1847 if (right
->rb_subtree_gap
>= length
) {
1854 /* Check if current node has a suitable gap */
1855 gap_end
= vma
->vm_start
;
1856 if (gap_end
< low_limit
)
1858 if (gap_start
<= high_limit
&& gap_end
- gap_start
>= length
)
1861 /* Visit left subtree if it looks promising */
1862 if (vma
->vm_rb
.rb_left
) {
1863 struct vm_area_struct
*left
=
1864 rb_entry(vma
->vm_rb
.rb_left
,
1865 struct vm_area_struct
, vm_rb
);
1866 if (left
->rb_subtree_gap
>= length
) {
1872 /* Go back up the rbtree to find next candidate node */
1874 struct rb_node
*prev
= &vma
->vm_rb
;
1875 if (!rb_parent(prev
))
1877 vma
= rb_entry(rb_parent(prev
),
1878 struct vm_area_struct
, vm_rb
);
1879 if (prev
== vma
->vm_rb
.rb_right
) {
1880 gap_start
= vma
->vm_prev
?
1881 vma
->vm_prev
->vm_end
: 0;
1888 /* We found a suitable gap. Clip it with the original high_limit. */
1889 if (gap_end
> info
->high_limit
)
1890 gap_end
= info
->high_limit
;
1893 /* Compute highest gap address at the desired alignment */
1894 gap_end
-= info
->length
;
1895 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
1897 VM_BUG_ON(gap_end
< info
->low_limit
);
1898 VM_BUG_ON(gap_end
< gap_start
);
1902 /* Get an address range which is currently unmapped.
1903 * For shmat() with addr=0.
1905 * Ugly calling convention alert:
1906 * Return value with the low bits set means error value,
1908 * if (ret & ~PAGE_MASK)
1911 * This function "knows" that -ENOMEM has the bits set.
1913 #ifndef HAVE_ARCH_UNMAPPED_AREA
1915 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1916 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1918 struct mm_struct
*mm
= current
->mm
;
1919 struct vm_area_struct
*vma
;
1920 struct vm_unmapped_area_info info
;
1922 if (len
> TASK_SIZE
- mmap_min_addr
)
1925 if (flags
& MAP_FIXED
)
1929 addr
= PAGE_ALIGN(addr
);
1930 vma
= find_vma(mm
, addr
);
1931 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1932 (!vma
|| addr
+ len
<= vma
->vm_start
))
1938 info
.low_limit
= mm
->mmap_base
;
1939 info
.high_limit
= TASK_SIZE
;
1940 info
.align_mask
= 0;
1941 return vm_unmapped_area(&info
);
1946 * This mmap-allocator allocates new areas top-down from below the
1947 * stack's low limit (the base):
1949 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1951 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1952 const unsigned long len
, const unsigned long pgoff
,
1953 const unsigned long flags
)
1955 struct vm_area_struct
*vma
;
1956 struct mm_struct
*mm
= current
->mm
;
1957 unsigned long addr
= addr0
;
1958 struct vm_unmapped_area_info info
;
1960 /* requested length too big for entire address space */
1961 if (len
> TASK_SIZE
- mmap_min_addr
)
1964 if (flags
& MAP_FIXED
)
1967 /* requesting a specific address */
1969 addr
= PAGE_ALIGN(addr
);
1970 vma
= find_vma(mm
, addr
);
1971 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1972 (!vma
|| addr
+ len
<= vma
->vm_start
))
1976 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1978 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
1979 info
.high_limit
= mm
->mmap_base
;
1980 info
.align_mask
= 0;
1981 addr
= vm_unmapped_area(&info
);
1984 * A failed mmap() very likely causes application failure,
1985 * so fall back to the bottom-up function here. This scenario
1986 * can happen with large stack limits and large mmap()
1989 if (addr
& ~PAGE_MASK
) {
1990 VM_BUG_ON(addr
!= -ENOMEM
);
1992 info
.low_limit
= TASK_UNMAPPED_BASE
;
1993 info
.high_limit
= TASK_SIZE
;
1994 addr
= vm_unmapped_area(&info
);
2002 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2003 unsigned long pgoff
, unsigned long flags
)
2005 unsigned long (*get_area
)(struct file
*, unsigned long,
2006 unsigned long, unsigned long, unsigned long);
2008 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2012 /* Careful about overflows.. */
2013 if (len
> TASK_SIZE
)
2016 get_area
= current
->mm
->get_unmapped_area
;
2017 if (file
&& file
->f_op
->get_unmapped_area
)
2018 get_area
= file
->f_op
->get_unmapped_area
;
2019 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2020 if (IS_ERR_VALUE(addr
))
2023 if (addr
> TASK_SIZE
- len
)
2025 if (addr
& ~PAGE_MASK
)
2028 addr
= arch_rebalance_pgtables(addr
, len
);
2029 error
= security_mmap_addr(addr
);
2030 return error
? error
: addr
;
2033 EXPORT_SYMBOL(get_unmapped_area
);
2035 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2036 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2038 struct rb_node
*rb_node
;
2039 struct vm_area_struct
*vma
;
2041 /* Check the cache first. */
2042 vma
= vmacache_find(mm
, addr
);
2046 rb_node
= mm
->mm_rb
.rb_node
;
2050 struct vm_area_struct
*tmp
;
2052 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2054 if (tmp
->vm_end
> addr
) {
2056 if (tmp
->vm_start
<= addr
)
2058 rb_node
= rb_node
->rb_left
;
2060 rb_node
= rb_node
->rb_right
;
2064 vmacache_update(addr
, vma
);
2068 EXPORT_SYMBOL(find_vma
);
2071 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2073 struct vm_area_struct
*
2074 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2075 struct vm_area_struct
**pprev
)
2077 struct vm_area_struct
*vma
;
2079 vma
= find_vma(mm
, addr
);
2081 *pprev
= vma
->vm_prev
;
2083 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2086 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2087 rb_node
= rb_node
->rb_right
;
2094 * Verify that the stack growth is acceptable and
2095 * update accounting. This is shared with both the
2096 * grow-up and grow-down cases.
2098 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
2100 struct mm_struct
*mm
= vma
->vm_mm
;
2101 struct rlimit
*rlim
= current
->signal
->rlim
;
2102 unsigned long new_start
;
2104 /* address space limit tests */
2105 if (!may_expand_vm(mm
, grow
))
2108 /* Stack limit test */
2109 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
2112 /* mlock limit tests */
2113 if (vma
->vm_flags
& VM_LOCKED
) {
2114 unsigned long locked
;
2115 unsigned long limit
;
2116 locked
= mm
->locked_vm
+ grow
;
2117 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
2118 limit
>>= PAGE_SHIFT
;
2119 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2123 /* Check to ensure the stack will not grow into a hugetlb-only region */
2124 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2126 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2130 * Overcommit.. This must be the final test, as it will
2131 * update security statistics.
2133 if (security_vm_enough_memory_mm(mm
, grow
))
2136 /* Ok, everything looks good - let it rip */
2137 if (vma
->vm_flags
& VM_LOCKED
)
2138 mm
->locked_vm
+= grow
;
2139 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
2143 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2145 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2146 * vma is the last one with address > vma->vm_end. Have to extend vma.
2148 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2152 if (!(vma
->vm_flags
& VM_GROWSUP
))
2156 * We must make sure the anon_vma is allocated
2157 * so that the anon_vma locking is not a noop.
2159 if (unlikely(anon_vma_prepare(vma
)))
2161 vma_lock_anon_vma(vma
);
2164 * vma->vm_start/vm_end cannot change under us because the caller
2165 * is required to hold the mmap_sem in read mode. We need the
2166 * anon_vma lock to serialize against concurrent expand_stacks.
2167 * Also guard against wrapping around to address 0.
2169 if (address
< PAGE_ALIGN(address
+4))
2170 address
= PAGE_ALIGN(address
+4);
2172 vma_unlock_anon_vma(vma
);
2177 /* Somebody else might have raced and expanded it already */
2178 if (address
> vma
->vm_end
) {
2179 unsigned long size
, grow
;
2181 size
= address
- vma
->vm_start
;
2182 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2185 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2186 error
= acct_stack_growth(vma
, size
, grow
);
2189 * vma_gap_update() doesn't support concurrent
2190 * updates, but we only hold a shared mmap_sem
2191 * lock here, so we need to protect against
2192 * concurrent vma expansions.
2193 * vma_lock_anon_vma() doesn't help here, as
2194 * we don't guarantee that all growable vmas
2195 * in a mm share the same root anon vma.
2196 * So, we reuse mm->page_table_lock to guard
2197 * against concurrent vma expansions.
2199 spin_lock(&vma
->vm_mm
->page_table_lock
);
2200 anon_vma_interval_tree_pre_update_vma(vma
);
2201 vma
->vm_end
= address
;
2202 anon_vma_interval_tree_post_update_vma(vma
);
2204 vma_gap_update(vma
->vm_next
);
2206 vma
->vm_mm
->highest_vm_end
= address
;
2207 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2209 perf_event_mmap(vma
);
2213 vma_unlock_anon_vma(vma
);
2214 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2215 validate_mm(vma
->vm_mm
);
2218 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2221 * vma is the first one with address < vma->vm_start. Have to extend vma.
2223 int expand_downwards(struct vm_area_struct
*vma
,
2224 unsigned long address
)
2229 * We must make sure the anon_vma is allocated
2230 * so that the anon_vma locking is not a noop.
2232 if (unlikely(anon_vma_prepare(vma
)))
2235 address
&= PAGE_MASK
;
2236 error
= security_mmap_addr(address
);
2240 vma_lock_anon_vma(vma
);
2243 * vma->vm_start/vm_end cannot change under us because the caller
2244 * is required to hold the mmap_sem in read mode. We need the
2245 * anon_vma lock to serialize against concurrent expand_stacks.
2248 /* Somebody else might have raced and expanded it already */
2249 if (address
< vma
->vm_start
) {
2250 unsigned long size
, grow
;
2252 size
= vma
->vm_end
- address
;
2253 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2256 if (grow
<= vma
->vm_pgoff
) {
2257 error
= acct_stack_growth(vma
, size
, grow
);
2260 * vma_gap_update() doesn't support concurrent
2261 * updates, but we only hold a shared mmap_sem
2262 * lock here, so we need to protect against
2263 * concurrent vma expansions.
2264 * vma_lock_anon_vma() doesn't help here, as
2265 * we don't guarantee that all growable vmas
2266 * in a mm share the same root anon vma.
2267 * So, we reuse mm->page_table_lock to guard
2268 * against concurrent vma expansions.
2270 spin_lock(&vma
->vm_mm
->page_table_lock
);
2271 anon_vma_interval_tree_pre_update_vma(vma
);
2272 vma
->vm_start
= address
;
2273 vma
->vm_pgoff
-= grow
;
2274 anon_vma_interval_tree_post_update_vma(vma
);
2275 vma_gap_update(vma
);
2276 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2278 perf_event_mmap(vma
);
2282 vma_unlock_anon_vma(vma
);
2283 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2284 validate_mm(vma
->vm_mm
);
2289 * Note how expand_stack() refuses to expand the stack all the way to
2290 * abut the next virtual mapping, *unless* that mapping itself is also
2291 * a stack mapping. We want to leave room for a guard page, after all
2292 * (the guard page itself is not added here, that is done by the
2293 * actual page faulting logic)
2295 * This matches the behavior of the guard page logic (see mm/memory.c:
2296 * check_stack_guard_page()), which only allows the guard page to be
2297 * removed under these circumstances.
2299 #ifdef CONFIG_STACK_GROWSUP
2300 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2302 struct vm_area_struct
*next
;
2304 address
&= PAGE_MASK
;
2305 next
= vma
->vm_next
;
2306 if (next
&& next
->vm_start
== address
+ PAGE_SIZE
) {
2307 if (!(next
->vm_flags
& VM_GROWSUP
))
2310 return expand_upwards(vma
, address
);
2313 struct vm_area_struct
*
2314 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2316 struct vm_area_struct
*vma
, *prev
;
2319 vma
= find_vma_prev(mm
, addr
, &prev
);
2320 if (vma
&& (vma
->vm_start
<= addr
))
2322 if (!prev
|| expand_stack(prev
, addr
))
2324 if (prev
->vm_flags
& VM_LOCKED
)
2325 __mlock_vma_pages_range(prev
, addr
, prev
->vm_end
, NULL
);
2329 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2331 struct vm_area_struct
*prev
;
2333 address
&= PAGE_MASK
;
2334 prev
= vma
->vm_prev
;
2335 if (prev
&& prev
->vm_end
== address
) {
2336 if (!(prev
->vm_flags
& VM_GROWSDOWN
))
2339 return expand_downwards(vma
, address
);
2342 struct vm_area_struct
*
2343 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2345 struct vm_area_struct
*vma
;
2346 unsigned long start
;
2349 vma
= find_vma(mm
, addr
);
2352 if (vma
->vm_start
<= addr
)
2354 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2356 start
= vma
->vm_start
;
2357 if (expand_stack(vma
, addr
))
2359 if (vma
->vm_flags
& VM_LOCKED
)
2360 __mlock_vma_pages_range(vma
, addr
, start
, NULL
);
2366 * Ok - we have the memory areas we should free on the vma list,
2367 * so release them, and do the vma updates.
2369 * Called with the mm semaphore held.
2371 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2373 unsigned long nr_accounted
= 0;
2375 /* Update high watermark before we lower total_vm */
2376 update_hiwater_vm(mm
);
2378 long nrpages
= vma_pages(vma
);
2380 if (vma
->vm_flags
& VM_ACCOUNT
)
2381 nr_accounted
+= nrpages
;
2382 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
2383 vma
= remove_vma(vma
);
2385 vm_unacct_memory(nr_accounted
);
2390 * Get rid of page table information in the indicated region.
2392 * Called with the mm semaphore held.
2394 static void unmap_region(struct mm_struct
*mm
,
2395 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2396 unsigned long start
, unsigned long end
)
2398 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2399 struct mmu_gather tlb
;
2402 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2403 update_hiwater_rss(mm
);
2404 unmap_vmas(&tlb
, vma
, start
, end
);
2405 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2406 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2407 tlb_finish_mmu(&tlb
, start
, end
);
2411 * Create a list of vma's touched by the unmap, removing them from the mm's
2412 * vma list as we go..
2415 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2416 struct vm_area_struct
*prev
, unsigned long end
)
2418 struct vm_area_struct
**insertion_point
;
2419 struct vm_area_struct
*tail_vma
= NULL
;
2421 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2422 vma
->vm_prev
= NULL
;
2424 vma_rb_erase(vma
, &mm
->mm_rb
);
2428 } while (vma
&& vma
->vm_start
< end
);
2429 *insertion_point
= vma
;
2431 vma
->vm_prev
= prev
;
2432 vma_gap_update(vma
);
2434 mm
->highest_vm_end
= prev
? prev
->vm_end
: 0;
2435 tail_vma
->vm_next
= NULL
;
2437 /* Kill the cache */
2438 vmacache_invalidate(mm
);
2442 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2443 * munmap path where it doesn't make sense to fail.
2445 static int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2446 unsigned long addr
, int new_below
)
2448 struct vm_area_struct
*new;
2451 if (is_vm_hugetlb_page(vma
) && (addr
&
2452 ~(huge_page_mask(hstate_vma(vma
)))))
2455 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2459 /* most fields are the same, copy all, and then fixup */
2462 INIT_LIST_HEAD(&new->anon_vma_chain
);
2467 new->vm_start
= addr
;
2468 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2471 err
= vma_dup_policy(vma
, new);
2475 err
= anon_vma_clone(new, vma
);
2480 get_file(new->vm_file
);
2482 if (new->vm_ops
&& new->vm_ops
->open
)
2483 new->vm_ops
->open(new);
2486 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2487 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2489 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2495 /* Clean everything up if vma_adjust failed. */
2496 if (new->vm_ops
&& new->vm_ops
->close
)
2497 new->vm_ops
->close(new);
2500 unlink_anon_vmas(new);
2502 mpol_put(vma_policy(new));
2504 kmem_cache_free(vm_area_cachep
, new);
2510 * Split a vma into two pieces at address 'addr', a new vma is allocated
2511 * either for the first part or the tail.
2513 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2514 unsigned long addr
, int new_below
)
2516 if (mm
->map_count
>= sysctl_max_map_count
)
2519 return __split_vma(mm
, vma
, addr
, new_below
);
2522 /* Munmap is split into 2 main parts -- this part which finds
2523 * what needs doing, and the areas themselves, which do the
2524 * work. This now handles partial unmappings.
2525 * Jeremy Fitzhardinge <jeremy@goop.org>
2527 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2530 struct vm_area_struct
*vma
, *prev
, *last
;
2532 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2535 len
= PAGE_ALIGN(len
);
2539 /* Find the first overlapping VMA */
2540 vma
= find_vma(mm
, start
);
2543 prev
= vma
->vm_prev
;
2544 /* we have start < vma->vm_end */
2546 /* if it doesn't overlap, we have nothing.. */
2548 if (vma
->vm_start
>= end
)
2552 * If we need to split any vma, do it now to save pain later.
2554 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2555 * unmapped vm_area_struct will remain in use: so lower split_vma
2556 * places tmp vma above, and higher split_vma places tmp vma below.
2558 if (start
> vma
->vm_start
) {
2562 * Make sure that map_count on return from munmap() will
2563 * not exceed its limit; but let map_count go just above
2564 * its limit temporarily, to help free resources as expected.
2566 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2569 error
= __split_vma(mm
, vma
, start
, 0);
2575 /* Does it split the last one? */
2576 last
= find_vma(mm
, end
);
2577 if (last
&& end
> last
->vm_start
) {
2578 int error
= __split_vma(mm
, last
, end
, 1);
2582 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2585 * unlock any mlock()ed ranges before detaching vmas
2587 if (mm
->locked_vm
) {
2588 struct vm_area_struct
*tmp
= vma
;
2589 while (tmp
&& tmp
->vm_start
< end
) {
2590 if (tmp
->vm_flags
& VM_LOCKED
) {
2591 mm
->locked_vm
-= vma_pages(tmp
);
2592 munlock_vma_pages_all(tmp
);
2599 * Remove the vma's, and unmap the actual pages
2601 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2602 unmap_region(mm
, vma
, prev
, start
, end
);
2604 /* Fix up all other VM information */
2605 remove_vma_list(mm
, vma
);
2610 int vm_munmap(unsigned long start
, size_t len
)
2613 struct mm_struct
*mm
= current
->mm
;
2615 down_write(&mm
->mmap_sem
);
2616 ret
= do_munmap(mm
, start
, len
);
2617 up_write(&mm
->mmap_sem
);
2620 EXPORT_SYMBOL(vm_munmap
);
2622 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2624 profile_munmap(addr
);
2625 return vm_munmap(addr
, len
);
2628 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2630 #ifdef CONFIG_DEBUG_VM
2631 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2633 up_read(&mm
->mmap_sem
);
2639 * this is really a simplified "do_mmap". it only handles
2640 * anonymous maps. eventually we may be able to do some
2641 * brk-specific accounting here.
2643 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2645 struct mm_struct
*mm
= current
->mm
;
2646 struct vm_area_struct
*vma
, *prev
;
2647 unsigned long flags
;
2648 struct rb_node
**rb_link
, *rb_parent
;
2649 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2652 len
= PAGE_ALIGN(len
);
2656 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2658 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2659 if (error
& ~PAGE_MASK
)
2662 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
2667 * mm->mmap_sem is required to protect against another thread
2668 * changing the mappings in case we sleep.
2670 verify_mm_writelocked(mm
);
2673 * Clear old maps. this also does some error checking for us
2676 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
)) {
2677 if (do_munmap(mm
, addr
, len
))
2682 /* Check against address space limits *after* clearing old maps... */
2683 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2686 if (mm
->map_count
> sysctl_max_map_count
)
2689 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2692 /* Can we just expand an old private anonymous mapping? */
2693 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2694 NULL
, NULL
, pgoff
, NULL
);
2699 * create a vma struct for an anonymous mapping
2701 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2703 vm_unacct_memory(len
>> PAGE_SHIFT
);
2707 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2709 vma
->vm_start
= addr
;
2710 vma
->vm_end
= addr
+ len
;
2711 vma
->vm_pgoff
= pgoff
;
2712 vma
->vm_flags
= flags
;
2713 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2714 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2716 perf_event_mmap(vma
);
2717 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2718 if (flags
& VM_LOCKED
)
2719 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2720 vma
->vm_flags
|= VM_SOFTDIRTY
;
2724 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2726 struct mm_struct
*mm
= current
->mm
;
2730 down_write(&mm
->mmap_sem
);
2731 ret
= do_brk(addr
, len
);
2732 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
2733 up_write(&mm
->mmap_sem
);
2735 mm_populate(addr
, len
);
2738 EXPORT_SYMBOL(vm_brk
);
2740 /* Release all mmaps. */
2741 void exit_mmap(struct mm_struct
*mm
)
2743 struct mmu_gather tlb
;
2744 struct vm_area_struct
*vma
;
2745 unsigned long nr_accounted
= 0;
2747 /* mm's last user has gone, and its about to be pulled down */
2748 mmu_notifier_release(mm
);
2750 if (mm
->locked_vm
) {
2753 if (vma
->vm_flags
& VM_LOCKED
)
2754 munlock_vma_pages_all(vma
);
2762 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2767 tlb_gather_mmu(&tlb
, mm
, 0, -1);
2768 /* update_hiwater_rss(mm) here? but nobody should be looking */
2769 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2770 unmap_vmas(&tlb
, vma
, 0, -1);
2772 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
2773 tlb_finish_mmu(&tlb
, 0, -1);
2776 * Walk the list again, actually closing and freeing it,
2777 * with preemption enabled, without holding any MM locks.
2780 if (vma
->vm_flags
& VM_ACCOUNT
)
2781 nr_accounted
+= vma_pages(vma
);
2782 vma
= remove_vma(vma
);
2784 vm_unacct_memory(nr_accounted
);
2786 WARN_ON(atomic_long_read(&mm
->nr_ptes
) >
2787 (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2790 /* Insert vm structure into process list sorted by address
2791 * and into the inode's i_mmap tree. If vm_file is non-NULL
2792 * then i_mmap_mutex is taken here.
2794 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2796 struct vm_area_struct
*prev
;
2797 struct rb_node
**rb_link
, *rb_parent
;
2800 * The vm_pgoff of a purely anonymous vma should be irrelevant
2801 * until its first write fault, when page's anon_vma and index
2802 * are set. But now set the vm_pgoff it will almost certainly
2803 * end up with (unless mremap moves it elsewhere before that
2804 * first wfault), so /proc/pid/maps tells a consistent story.
2806 * By setting it to reflect the virtual start address of the
2807 * vma, merges and splits can happen in a seamless way, just
2808 * using the existing file pgoff checks and manipulations.
2809 * Similarly in do_mmap_pgoff and in do_brk.
2811 if (!vma
->vm_file
) {
2812 BUG_ON(vma
->anon_vma
);
2813 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2815 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2816 &prev
, &rb_link
, &rb_parent
))
2818 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2819 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2822 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2827 * Copy the vma structure to a new location in the same mm,
2828 * prior to moving page table entries, to effect an mremap move.
2830 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2831 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
2832 bool *need_rmap_locks
)
2834 struct vm_area_struct
*vma
= *vmap
;
2835 unsigned long vma_start
= vma
->vm_start
;
2836 struct mm_struct
*mm
= vma
->vm_mm
;
2837 struct vm_area_struct
*new_vma
, *prev
;
2838 struct rb_node
**rb_link
, *rb_parent
;
2839 bool faulted_in_anon_vma
= true;
2842 * If anonymous vma has not yet been faulted, update new pgoff
2843 * to match new location, to increase its chance of merging.
2845 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2846 pgoff
= addr
>> PAGE_SHIFT
;
2847 faulted_in_anon_vma
= false;
2850 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2851 return NULL
; /* should never get here */
2852 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2853 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2856 * Source vma may have been merged into new_vma
2858 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2859 vma_start
< new_vma
->vm_end
)) {
2861 * The only way we can get a vma_merge with
2862 * self during an mremap is if the vma hasn't
2863 * been faulted in yet and we were allowed to
2864 * reset the dst vma->vm_pgoff to the
2865 * destination address of the mremap to allow
2866 * the merge to happen. mremap must change the
2867 * vm_pgoff linearity between src and dst vmas
2868 * (in turn preventing a vma_merge) to be
2869 * safe. It is only safe to keep the vm_pgoff
2870 * linear if there are no pages mapped yet.
2872 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
2873 *vmap
= vma
= new_vma
;
2875 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
2877 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2880 new_vma
->vm_start
= addr
;
2881 new_vma
->vm_end
= addr
+ len
;
2882 new_vma
->vm_pgoff
= pgoff
;
2883 if (vma_dup_policy(vma
, new_vma
))
2885 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2886 if (anon_vma_clone(new_vma
, vma
))
2887 goto out_free_mempol
;
2888 if (new_vma
->vm_file
)
2889 get_file(new_vma
->vm_file
);
2890 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2891 new_vma
->vm_ops
->open(new_vma
);
2892 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2893 *need_rmap_locks
= false;
2899 mpol_put(vma_policy(new_vma
));
2901 kmem_cache_free(vm_area_cachep
, new_vma
);
2906 * Return true if the calling process may expand its vm space by the passed
2909 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2911 unsigned long cur
= mm
->total_vm
; /* pages */
2914 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2916 if (cur
+ npages
> lim
)
2921 static int special_mapping_fault(struct vm_area_struct
*vma
,
2922 struct vm_fault
*vmf
);
2925 * Having a close hook prevents vma merging regardless of flags.
2927 static void special_mapping_close(struct vm_area_struct
*vma
)
2931 static const char *special_mapping_name(struct vm_area_struct
*vma
)
2933 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
2936 static const struct vm_operations_struct special_mapping_vmops
= {
2937 .close
= special_mapping_close
,
2938 .fault
= special_mapping_fault
,
2939 .name
= special_mapping_name
,
2942 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
2943 .close
= special_mapping_close
,
2944 .fault
= special_mapping_fault
,
2947 static int special_mapping_fault(struct vm_area_struct
*vma
,
2948 struct vm_fault
*vmf
)
2951 struct page
**pages
;
2954 * special mappings have no vm_file, and in that case, the mm
2955 * uses vm_pgoff internally. So we have to subtract it from here.
2956 * We are allowed to do this because we are the mm; do not copy
2957 * this code into drivers!
2959 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2961 if (vma
->vm_ops
== &legacy_special_mapping_vmops
)
2962 pages
= vma
->vm_private_data
;
2964 pages
= ((struct vm_special_mapping
*)vma
->vm_private_data
)->
2967 for (; pgoff
&& *pages
; ++pages
)
2971 struct page
*page
= *pages
;
2977 return VM_FAULT_SIGBUS
;
2980 static struct vm_area_struct
*__install_special_mapping(
2981 struct mm_struct
*mm
,
2982 unsigned long addr
, unsigned long len
,
2983 unsigned long vm_flags
, const struct vm_operations_struct
*ops
,
2987 struct vm_area_struct
*vma
;
2989 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2990 if (unlikely(vma
== NULL
))
2991 return ERR_PTR(-ENOMEM
);
2993 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2995 vma
->vm_start
= addr
;
2996 vma
->vm_end
= addr
+ len
;
2998 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
2999 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3002 vma
->vm_private_data
= priv
;
3004 ret
= insert_vm_struct(mm
, vma
);
3008 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3010 perf_event_mmap(vma
);
3015 kmem_cache_free(vm_area_cachep
, vma
);
3016 return ERR_PTR(ret
);
3020 * Called with mm->mmap_sem held for writing.
3021 * Insert a new vma covering the given region, with the given flags.
3022 * Its pages are supplied by the given array of struct page *.
3023 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3024 * The region past the last page supplied will always produce SIGBUS.
3025 * The array pointer and the pages it points to are assumed to stay alive
3026 * for as long as this mapping might exist.
3028 struct vm_area_struct
*_install_special_mapping(
3029 struct mm_struct
*mm
,
3030 unsigned long addr
, unsigned long len
,
3031 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3033 return __install_special_mapping(mm
, addr
, len
, vm_flags
,
3034 &special_mapping_vmops
, (void *)spec
);
3037 int install_special_mapping(struct mm_struct
*mm
,
3038 unsigned long addr
, unsigned long len
,
3039 unsigned long vm_flags
, struct page
**pages
)
3041 struct vm_area_struct
*vma
= __install_special_mapping(
3042 mm
, addr
, len
, vm_flags
, &legacy_special_mapping_vmops
,
3045 return PTR_ERR_OR_ZERO(vma
);
3048 static DEFINE_MUTEX(mm_all_locks_mutex
);
3050 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3052 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3054 * The LSB of head.next can't change from under us
3055 * because we hold the mm_all_locks_mutex.
3057 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3059 * We can safely modify head.next after taking the
3060 * anon_vma->root->rwsem. If some other vma in this mm shares
3061 * the same anon_vma we won't take it again.
3063 * No need of atomic instructions here, head.next
3064 * can't change from under us thanks to the
3065 * anon_vma->root->rwsem.
3067 if (__test_and_set_bit(0, (unsigned long *)
3068 &anon_vma
->root
->rb_root
.rb_node
))
3073 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3075 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3077 * AS_MM_ALL_LOCKS can't change from under us because
3078 * we hold the mm_all_locks_mutex.
3080 * Operations on ->flags have to be atomic because
3081 * even if AS_MM_ALL_LOCKS is stable thanks to the
3082 * mm_all_locks_mutex, there may be other cpus
3083 * changing other bitflags in parallel to us.
3085 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3087 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
3092 * This operation locks against the VM for all pte/vma/mm related
3093 * operations that could ever happen on a certain mm. This includes
3094 * vmtruncate, try_to_unmap, and all page faults.
3096 * The caller must take the mmap_sem in write mode before calling
3097 * mm_take_all_locks(). The caller isn't allowed to release the
3098 * mmap_sem until mm_drop_all_locks() returns.
3100 * mmap_sem in write mode is required in order to block all operations
3101 * that could modify pagetables and free pages without need of
3102 * altering the vma layout (for example populate_range() with
3103 * nonlinear vmas). It's also needed in write mode to avoid new
3104 * anon_vmas to be associated with existing vmas.
3106 * A single task can't take more than one mm_take_all_locks() in a row
3107 * or it would deadlock.
3109 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3110 * mapping->flags avoid to take the same lock twice, if more than one
3111 * vma in this mm is backed by the same anon_vma or address_space.
3113 * We can take all the locks in random order because the VM code
3114 * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never
3115 * takes more than one of them in a row. Secondly we're protected
3116 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3118 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3119 * that may have to take thousand of locks.
3121 * mm_take_all_locks() can fail if it's interrupted by signals.
3123 int mm_take_all_locks(struct mm_struct
*mm
)
3125 struct vm_area_struct
*vma
;
3126 struct anon_vma_chain
*avc
;
3128 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3130 mutex_lock(&mm_all_locks_mutex
);
3132 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3133 if (signal_pending(current
))
3135 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3136 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3139 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3140 if (signal_pending(current
))
3143 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3144 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3150 mm_drop_all_locks(mm
);
3154 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3156 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3158 * The LSB of head.next can't change to 0 from under
3159 * us because we hold the mm_all_locks_mutex.
3161 * We must however clear the bitflag before unlocking
3162 * the vma so the users using the anon_vma->rb_root will
3163 * never see our bitflag.
3165 * No need of atomic instructions here, head.next
3166 * can't change from under us until we release the
3167 * anon_vma->root->rwsem.
3169 if (!__test_and_clear_bit(0, (unsigned long *)
3170 &anon_vma
->root
->rb_root
.rb_node
))
3172 anon_vma_unlock_write(anon_vma
);
3176 static void vm_unlock_mapping(struct address_space
*mapping
)
3178 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3180 * AS_MM_ALL_LOCKS can't change to 0 from under us
3181 * because we hold the mm_all_locks_mutex.
3183 mutex_unlock(&mapping
->i_mmap_mutex
);
3184 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3191 * The mmap_sem cannot be released by the caller until
3192 * mm_drop_all_locks() returns.
3194 void mm_drop_all_locks(struct mm_struct
*mm
)
3196 struct vm_area_struct
*vma
;
3197 struct anon_vma_chain
*avc
;
3199 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3200 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3202 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3204 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3205 vm_unlock_anon_vma(avc
->anon_vma
);
3206 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3207 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3210 mutex_unlock(&mm_all_locks_mutex
);
3214 * initialise the VMA slab
3216 void __init
mmap_init(void)
3220 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3225 * Initialise sysctl_user_reserve_kbytes.
3227 * This is intended to prevent a user from starting a single memory hogging
3228 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3231 * The default value is min(3% of free memory, 128MB)
3232 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3234 static int init_user_reserve(void)
3236 unsigned long free_kbytes
;
3238 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3240 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3243 subsys_initcall(init_user_reserve
);
3246 * Initialise sysctl_admin_reserve_kbytes.
3248 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3249 * to log in and kill a memory hogging process.
3251 * Systems with more than 256MB will reserve 8MB, enough to recover
3252 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3253 * only reserve 3% of free pages by default.
3255 static int init_admin_reserve(void)
3257 unsigned long free_kbytes
;
3259 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3261 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3264 subsys_initcall(init_admin_reserve
);
3267 * Reinititalise user and admin reserves if memory is added or removed.
3269 * The default user reserve max is 128MB, and the default max for the
3270 * admin reserve is 8MB. These are usually, but not always, enough to
3271 * enable recovery from a memory hogging process using login/sshd, a shell,
3272 * and tools like top. It may make sense to increase or even disable the
3273 * reserve depending on the existence of swap or variations in the recovery
3274 * tools. So, the admin may have changed them.
3276 * If memory is added and the reserves have been eliminated or increased above
3277 * the default max, then we'll trust the admin.
3279 * If memory is removed and there isn't enough free memory, then we
3280 * need to reset the reserves.
3282 * Otherwise keep the reserve set by the admin.
3284 static int reserve_mem_notifier(struct notifier_block
*nb
,
3285 unsigned long action
, void *data
)
3287 unsigned long tmp
, free_kbytes
;
3291 /* Default max is 128MB. Leave alone if modified by operator. */
3292 tmp
= sysctl_user_reserve_kbytes
;
3293 if (0 < tmp
&& tmp
< (1UL << 17))
3294 init_user_reserve();
3296 /* Default max is 8MB. Leave alone if modified by operator. */
3297 tmp
= sysctl_admin_reserve_kbytes
;
3298 if (0 < tmp
&& tmp
< (1UL << 13))
3299 init_admin_reserve();
3303 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3305 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3306 init_user_reserve();
3307 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3308 sysctl_user_reserve_kbytes
);
3311 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3312 init_admin_reserve();
3313 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3314 sysctl_admin_reserve_kbytes
);
3323 static struct notifier_block reserve_mem_nb
= {
3324 .notifier_call
= reserve_mem_notifier
,
3327 static int __meminit
init_reserve_notifier(void)
3329 if (register_hotmemory_notifier(&reserve_mem_nb
))
3330 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3334 subsys_initcall(init_reserve_notifier
);