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>
44 #include <linux/userfaultfd_k.h>
46 #include <asm/uaccess.h>
47 #include <asm/cacheflush.h>
49 #include <asm/mmu_context.h>
53 #ifndef arch_mmap_check
54 #define arch_mmap_check(addr, len, flags) (0)
57 #ifndef arch_rebalance_pgtables
58 #define arch_rebalance_pgtables(addr, len) (addr)
61 static void unmap_region(struct mm_struct
*mm
,
62 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
63 unsigned long start
, unsigned long end
);
65 /* description of effects of mapping type and prot in current implementation.
66 * this is due to the limited x86 page protection hardware. The expected
67 * behavior is in parens:
70 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
71 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
72 * w: (no) no w: (no) no w: (yes) yes w: (no) no
73 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
75 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
76 * w: (no) no w: (no) no w: (copy) copy w: (no) no
77 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
80 pgprot_t protection_map
[16] = {
81 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
82 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
85 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
87 return __pgprot(pgprot_val(protection_map
[vm_flags
&
88 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
89 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
91 EXPORT_SYMBOL(vm_get_page_prot
);
93 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
95 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
98 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
99 void vma_set_page_prot(struct vm_area_struct
*vma
)
101 unsigned long vm_flags
= vma
->vm_flags
;
103 vma
->vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
104 if (vma_wants_writenotify(vma
)) {
105 vm_flags
&= ~VM_SHARED
;
106 vma
->vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
,
112 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
113 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
114 unsigned long sysctl_overcommit_kbytes __read_mostly
;
115 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
116 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
117 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
119 * Make sure vm_committed_as in one cacheline and not cacheline shared with
120 * other variables. It can be updated by several CPUs frequently.
122 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
125 * The global memory commitment made in the system can be a metric
126 * that can be used to drive ballooning decisions when Linux is hosted
127 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
128 * balancing memory across competing virtual machines that are hosted.
129 * Several metrics drive this policy engine including the guest reported
132 unsigned long vm_memory_committed(void)
134 return percpu_counter_read_positive(&vm_committed_as
);
136 EXPORT_SYMBOL_GPL(vm_memory_committed
);
139 * Check that a process has enough memory to allocate a new virtual
140 * mapping. 0 means there is enough memory for the allocation to
141 * succeed and -ENOMEM implies there is not.
143 * We currently support three overcommit policies, which are set via the
144 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
146 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
147 * Additional code 2002 Jul 20 by Robert Love.
149 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
151 * Note this is a helper function intended to be used by LSMs which
152 * wish to use this logic.
154 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
156 long free
, allowed
, reserve
;
158 VM_WARN_ONCE(percpu_counter_read(&vm_committed_as
) <
159 -(s64
)vm_committed_as_batch
* num_online_cpus(),
160 "memory commitment underflow");
162 vm_acct_memory(pages
);
165 * Sometimes we want to use more memory than we have
167 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
170 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
171 free
= global_page_state(NR_FREE_PAGES
);
172 free
+= global_page_state(NR_FILE_PAGES
);
175 * shmem pages shouldn't be counted as free in this
176 * case, they can't be purged, only swapped out, and
177 * that won't affect the overall amount of available
178 * memory in the system.
180 free
-= global_page_state(NR_SHMEM
);
182 free
+= get_nr_swap_pages();
185 * Any slabs which are created with the
186 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
187 * which are reclaimable, under pressure. The dentry
188 * cache and most inode caches should fall into this
190 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
193 * Leave reserved pages. The pages are not for anonymous pages.
195 if (free
<= totalreserve_pages
)
198 free
-= totalreserve_pages
;
201 * Reserve some for root
204 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
212 allowed
= vm_commit_limit();
214 * Reserve some for root
217 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
220 * Don't let a single process grow so big a user can't recover
223 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
224 allowed
-= min_t(long, mm
->total_vm
/ 32, reserve
);
227 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
230 vm_unacct_memory(pages
);
236 * Requires inode->i_mapping->i_mmap_rwsem
238 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
239 struct file
*file
, struct address_space
*mapping
)
241 if (vma
->vm_flags
& VM_DENYWRITE
)
242 atomic_inc(&file_inode(file
)->i_writecount
);
243 if (vma
->vm_flags
& VM_SHARED
)
244 mapping_unmap_writable(mapping
);
246 flush_dcache_mmap_lock(mapping
);
247 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
248 flush_dcache_mmap_unlock(mapping
);
252 * Unlink a file-based vm structure from its interval tree, to hide
253 * vma from rmap and vmtruncate before freeing its page tables.
255 void unlink_file_vma(struct vm_area_struct
*vma
)
257 struct file
*file
= vma
->vm_file
;
260 struct address_space
*mapping
= file
->f_mapping
;
261 i_mmap_lock_write(mapping
);
262 __remove_shared_vm_struct(vma
, file
, mapping
);
263 i_mmap_unlock_write(mapping
);
268 * Close a vm structure and free it, returning the next.
270 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
272 struct vm_area_struct
*next
= vma
->vm_next
;
275 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
276 vma
->vm_ops
->close(vma
);
279 mpol_put(vma_policy(vma
));
280 kmem_cache_free(vm_area_cachep
, vma
);
284 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
286 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
288 unsigned long retval
;
289 unsigned long newbrk
, oldbrk
;
290 struct mm_struct
*mm
= current
->mm
;
291 unsigned long min_brk
;
294 down_write(&mm
->mmap_sem
);
296 #ifdef CONFIG_COMPAT_BRK
298 * CONFIG_COMPAT_BRK can still be overridden by setting
299 * randomize_va_space to 2, which will still cause mm->start_brk
300 * to be arbitrarily shifted
302 if (current
->brk_randomized
)
303 min_brk
= mm
->start_brk
;
305 min_brk
= mm
->end_data
;
307 min_brk
= mm
->start_brk
;
313 * Check against rlimit here. If this check is done later after the test
314 * of oldbrk with newbrk then it can escape the test and let the data
315 * segment grow beyond its set limit the in case where the limit is
316 * not page aligned -Ram Gupta
318 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
319 mm
->end_data
, mm
->start_data
))
322 newbrk
= PAGE_ALIGN(brk
);
323 oldbrk
= PAGE_ALIGN(mm
->brk
);
324 if (oldbrk
== newbrk
)
327 /* Always allow shrinking brk. */
328 if (brk
<= mm
->brk
) {
329 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
334 /* Check against existing mmap mappings. */
335 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
338 /* Ok, looks good - let it rip. */
339 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
344 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
345 up_write(&mm
->mmap_sem
);
347 mm_populate(oldbrk
, newbrk
- oldbrk
);
352 up_write(&mm
->mmap_sem
);
356 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
358 unsigned long max
, subtree_gap
;
361 max
-= vma
->vm_prev
->vm_end
;
362 if (vma
->vm_rb
.rb_left
) {
363 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
364 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
365 if (subtree_gap
> max
)
368 if (vma
->vm_rb
.rb_right
) {
369 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
370 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
371 if (subtree_gap
> max
)
377 #ifdef CONFIG_DEBUG_VM_RB
378 static int browse_rb(struct rb_root
*root
)
380 int i
= 0, j
, bug
= 0;
381 struct rb_node
*nd
, *pn
= NULL
;
382 unsigned long prev
= 0, pend
= 0;
384 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
385 struct vm_area_struct
*vma
;
386 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
387 if (vma
->vm_start
< prev
) {
388 pr_emerg("vm_start %lx < prev %lx\n",
389 vma
->vm_start
, prev
);
392 if (vma
->vm_start
< pend
) {
393 pr_emerg("vm_start %lx < pend %lx\n",
394 vma
->vm_start
, pend
);
397 if (vma
->vm_start
> vma
->vm_end
) {
398 pr_emerg("vm_start %lx > vm_end %lx\n",
399 vma
->vm_start
, vma
->vm_end
);
402 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
403 pr_emerg("free gap %lx, correct %lx\n",
405 vma_compute_subtree_gap(vma
));
410 prev
= vma
->vm_start
;
414 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
417 pr_emerg("backwards %d, forwards %d\n", j
, i
);
423 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
427 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
428 struct vm_area_struct
*vma
;
429 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
430 VM_BUG_ON_VMA(vma
!= ignore
&&
431 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
436 static void validate_mm(struct mm_struct
*mm
)
440 unsigned long highest_address
= 0;
441 struct vm_area_struct
*vma
= mm
->mmap
;
444 struct anon_vma_chain
*avc
;
446 vma_lock_anon_vma(vma
);
447 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
448 anon_vma_interval_tree_verify(avc
);
449 vma_unlock_anon_vma(vma
);
450 highest_address
= vma
->vm_end
;
454 if (i
!= mm
->map_count
) {
455 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
458 if (highest_address
!= mm
->highest_vm_end
) {
459 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
460 mm
->highest_vm_end
, highest_address
);
463 i
= browse_rb(&mm
->mm_rb
);
464 if (i
!= mm
->map_count
) {
466 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
469 VM_BUG_ON_MM(bug
, mm
);
472 #define validate_mm_rb(root, ignore) do { } while (0)
473 #define validate_mm(mm) do { } while (0)
476 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
477 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
480 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
481 * vma->vm_prev->vm_end values changed, without modifying the vma's position
484 static void vma_gap_update(struct vm_area_struct
*vma
)
487 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
488 * function that does exacltly what we want.
490 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
493 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
494 struct rb_root
*root
)
496 /* All rb_subtree_gap values must be consistent prior to insertion */
497 validate_mm_rb(root
, NULL
);
499 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
502 static void vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
505 * All rb_subtree_gap values must be consistent prior to erase,
506 * with the possible exception of the vma being erased.
508 validate_mm_rb(root
, vma
);
511 * Note rb_erase_augmented is a fairly large inline function,
512 * so make sure we instantiate it only once with our desired
513 * augmented rbtree callbacks.
515 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
519 * vma has some anon_vma assigned, and is already inserted on that
520 * anon_vma's interval trees.
522 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
523 * vma must be removed from the anon_vma's interval trees using
524 * anon_vma_interval_tree_pre_update_vma().
526 * After the update, the vma will be reinserted using
527 * anon_vma_interval_tree_post_update_vma().
529 * The entire update must be protected by exclusive mmap_sem and by
530 * the root anon_vma's mutex.
533 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
535 struct anon_vma_chain
*avc
;
537 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
538 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
542 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
544 struct anon_vma_chain
*avc
;
546 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
547 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
550 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
551 unsigned long end
, struct vm_area_struct
**pprev
,
552 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
554 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
556 __rb_link
= &mm
->mm_rb
.rb_node
;
557 rb_prev
= __rb_parent
= NULL
;
560 struct vm_area_struct
*vma_tmp
;
562 __rb_parent
= *__rb_link
;
563 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
565 if (vma_tmp
->vm_end
> addr
) {
566 /* Fail if an existing vma overlaps the area */
567 if (vma_tmp
->vm_start
< end
)
569 __rb_link
= &__rb_parent
->rb_left
;
571 rb_prev
= __rb_parent
;
572 __rb_link
= &__rb_parent
->rb_right
;
578 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
579 *rb_link
= __rb_link
;
580 *rb_parent
= __rb_parent
;
584 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
585 unsigned long addr
, unsigned long end
)
587 unsigned long nr_pages
= 0;
588 struct vm_area_struct
*vma
;
590 /* Find first overlaping mapping */
591 vma
= find_vma_intersection(mm
, addr
, end
);
595 nr_pages
= (min(end
, vma
->vm_end
) -
596 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
598 /* Iterate over the rest of the overlaps */
599 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
600 unsigned long overlap_len
;
602 if (vma
->vm_start
> end
)
605 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
606 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
612 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
613 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
615 WARN_ONCE(vma
->vm_file
&& !vma
->vm_ops
, "missing vma->vm_ops");
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 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
653 flush_dcache_mmap_unlock(mapping
);
658 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
659 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
660 struct rb_node
*rb_parent
)
662 __vma_link_list(mm
, vma
, prev
, rb_parent
);
663 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
666 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
667 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
668 struct rb_node
*rb_parent
)
670 struct address_space
*mapping
= NULL
;
673 mapping
= vma
->vm_file
->f_mapping
;
674 i_mmap_lock_write(mapping
);
677 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
678 __vma_link_file(vma
);
681 i_mmap_unlock_write(mapping
);
688 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
689 * mm's list and rbtree. It has already been inserted into the interval tree.
691 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
693 struct vm_area_struct
*prev
;
694 struct rb_node
**rb_link
, *rb_parent
;
696 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
697 &prev
, &rb_link
, &rb_parent
))
699 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
704 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
705 struct vm_area_struct
*prev
)
707 struct vm_area_struct
*next
;
709 vma_rb_erase(vma
, &mm
->mm_rb
);
710 prev
->vm_next
= next
= vma
->vm_next
;
712 next
->vm_prev
= prev
;
715 vmacache_invalidate(mm
);
719 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
720 * is already present in an i_mmap tree without adjusting the tree.
721 * The following helper function should be used when such adjustments
722 * are necessary. The "insert" vma (if any) is to be inserted
723 * before we drop the necessary locks.
725 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
726 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
728 struct mm_struct
*mm
= vma
->vm_mm
;
729 struct vm_area_struct
*next
= vma
->vm_next
;
730 struct vm_area_struct
*importer
= NULL
;
731 struct address_space
*mapping
= NULL
;
732 struct rb_root
*root
= NULL
;
733 struct anon_vma
*anon_vma
= NULL
;
734 struct file
*file
= vma
->vm_file
;
735 bool start_changed
= false, end_changed
= false;
736 long adjust_next
= 0;
739 if (next
&& !insert
) {
740 struct vm_area_struct
*exporter
= NULL
;
742 if (end
>= next
->vm_end
) {
744 * vma expands, overlapping all the next, and
745 * perhaps the one after too (mprotect case 6).
747 again
: remove_next
= 1 + (end
> next
->vm_end
);
751 } else if (end
> next
->vm_start
) {
753 * vma expands, overlapping part of the next:
754 * mprotect case 5 shifting the boundary up.
756 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
759 } else if (end
< vma
->vm_end
) {
761 * vma shrinks, and !insert tells it's not
762 * split_vma inserting another: so it must be
763 * mprotect case 4 shifting the boundary down.
765 adjust_next
= -((vma
->vm_end
- end
) >> PAGE_SHIFT
);
771 * Easily overlooked: when mprotect shifts the boundary,
772 * make sure the expanding vma has anon_vma set if the
773 * shrinking vma had, to cover any anon pages imported.
775 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
778 importer
->anon_vma
= exporter
->anon_vma
;
779 error
= anon_vma_clone(importer
, exporter
);
786 mapping
= file
->f_mapping
;
787 root
= &mapping
->i_mmap
;
788 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
791 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
793 i_mmap_lock_write(mapping
);
796 * Put into interval tree now, so instantiated pages
797 * are visible to arm/parisc __flush_dcache_page
798 * throughout; but we cannot insert into address
799 * space until vma start or end is updated.
801 __vma_link_file(insert
);
805 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
807 anon_vma
= vma
->anon_vma
;
808 if (!anon_vma
&& adjust_next
)
809 anon_vma
= next
->anon_vma
;
811 VM_BUG_ON_VMA(adjust_next
&& next
->anon_vma
&&
812 anon_vma
!= next
->anon_vma
, next
);
813 anon_vma_lock_write(anon_vma
);
814 anon_vma_interval_tree_pre_update_vma(vma
);
816 anon_vma_interval_tree_pre_update_vma(next
);
820 flush_dcache_mmap_lock(mapping
);
821 vma_interval_tree_remove(vma
, root
);
823 vma_interval_tree_remove(next
, root
);
826 if (start
!= vma
->vm_start
) {
827 vma
->vm_start
= start
;
828 start_changed
= true;
830 if (end
!= vma
->vm_end
) {
834 vma
->vm_pgoff
= pgoff
;
836 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
837 next
->vm_pgoff
+= adjust_next
;
842 vma_interval_tree_insert(next
, root
);
843 vma_interval_tree_insert(vma
, root
);
844 flush_dcache_mmap_unlock(mapping
);
849 * vma_merge has merged next into vma, and needs
850 * us to remove next before dropping the locks.
852 __vma_unlink(mm
, next
, vma
);
854 __remove_shared_vm_struct(next
, file
, mapping
);
857 * split_vma has split insert from vma, and needs
858 * us to insert it before dropping the locks
859 * (it may either follow vma or precede it).
861 __insert_vm_struct(mm
, insert
);
867 mm
->highest_vm_end
= end
;
868 else if (!adjust_next
)
869 vma_gap_update(next
);
874 anon_vma_interval_tree_post_update_vma(vma
);
876 anon_vma_interval_tree_post_update_vma(next
);
877 anon_vma_unlock_write(anon_vma
);
880 i_mmap_unlock_write(mapping
);
891 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
895 anon_vma_merge(vma
, next
);
897 mpol_put(vma_policy(next
));
898 kmem_cache_free(vm_area_cachep
, next
);
900 * In mprotect's case 6 (see comments on vma_merge),
901 * we must remove another next too. It would clutter
902 * up the code too much to do both in one go.
905 if (remove_next
== 2)
908 vma_gap_update(next
);
910 mm
->highest_vm_end
= end
;
921 * If the vma has a ->close operation then the driver probably needs to release
922 * per-vma resources, so we don't attempt to merge those.
924 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
925 struct file
*file
, unsigned long vm_flags
,
926 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
929 * VM_SOFTDIRTY should not prevent from VMA merging, if we
930 * match the flags but dirty bit -- the caller should mark
931 * merged VMA as dirty. If dirty bit won't be excluded from
932 * comparison, we increase pressue on the memory system forcing
933 * the kernel to generate new VMAs when old one could be
936 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
938 if (vma
->vm_file
!= file
)
940 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
942 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
947 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
948 struct anon_vma
*anon_vma2
,
949 struct vm_area_struct
*vma
)
952 * The list_is_singular() test is to avoid merging VMA cloned from
953 * parents. This can improve scalability caused by anon_vma lock.
955 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
956 list_is_singular(&vma
->anon_vma_chain
)))
958 return anon_vma1
== anon_vma2
;
962 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
963 * in front of (at a lower virtual address and file offset than) the vma.
965 * We cannot merge two vmas if they have differently assigned (non-NULL)
966 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
968 * We don't check here for the merged mmap wrapping around the end of pagecache
969 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
970 * wrap, nor mmaps which cover the final page at index -1UL.
973 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
974 struct anon_vma
*anon_vma
, struct file
*file
,
976 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
978 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
979 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
980 if (vma
->vm_pgoff
== vm_pgoff
)
987 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
988 * beyond (at a higher virtual address and file offset than) the vma.
990 * We cannot merge two vmas if they have differently assigned (non-NULL)
991 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
994 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
995 struct anon_vma
*anon_vma
, struct file
*file
,
997 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
999 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1000 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1002 vm_pglen
= vma_pages(vma
);
1003 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1010 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1011 * whether that can be merged with its predecessor or its successor.
1012 * Or both (it neatly fills a hole).
1014 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1015 * certain not to be mapped by the time vma_merge is called; but when
1016 * called for mprotect, it is certain to be already mapped (either at
1017 * an offset within prev, or at the start of next), and the flags of
1018 * this area are about to be changed to vm_flags - and the no-change
1019 * case has already been eliminated.
1021 * The following mprotect cases have to be considered, where AAAA is
1022 * the area passed down from mprotect_fixup, never extending beyond one
1023 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1025 * AAAA AAAA AAAA AAAA
1026 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1027 * cannot merge might become might become might become
1028 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1029 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1030 * mremap move: PPPPNNNNNNNN 8
1032 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1033 * might become case 1 below case 2 below case 3 below
1035 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1036 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1038 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1039 struct vm_area_struct
*prev
, unsigned long addr
,
1040 unsigned long end
, unsigned long vm_flags
,
1041 struct anon_vma
*anon_vma
, struct file
*file
,
1042 pgoff_t pgoff
, struct mempolicy
*policy
,
1043 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1045 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1046 struct vm_area_struct
*area
, *next
;
1050 * We later require that vma->vm_flags == vm_flags,
1051 * so this tests vma->vm_flags & VM_SPECIAL, too.
1053 if (vm_flags
& VM_SPECIAL
)
1057 next
= prev
->vm_next
;
1061 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
1062 next
= next
->vm_next
;
1065 * Can it merge with the predecessor?
1067 if (prev
&& prev
->vm_end
== addr
&&
1068 mpol_equal(vma_policy(prev
), policy
) &&
1069 can_vma_merge_after(prev
, vm_flags
,
1070 anon_vma
, file
, pgoff
,
1071 vm_userfaultfd_ctx
)) {
1073 * OK, it can. Can we now merge in the successor as well?
1075 if (next
&& end
== next
->vm_start
&&
1076 mpol_equal(policy
, vma_policy(next
)) &&
1077 can_vma_merge_before(next
, vm_flags
,
1080 vm_userfaultfd_ctx
) &&
1081 is_mergeable_anon_vma(prev
->anon_vma
,
1082 next
->anon_vma
, NULL
)) {
1084 err
= vma_adjust(prev
, prev
->vm_start
,
1085 next
->vm_end
, prev
->vm_pgoff
, NULL
);
1086 } else /* cases 2, 5, 7 */
1087 err
= vma_adjust(prev
, prev
->vm_start
,
1088 end
, prev
->vm_pgoff
, NULL
);
1091 khugepaged_enter_vma_merge(prev
, vm_flags
);
1096 * Can this new request be merged in front of next?
1098 if (next
&& end
== next
->vm_start
&&
1099 mpol_equal(policy
, vma_policy(next
)) &&
1100 can_vma_merge_before(next
, vm_flags
,
1101 anon_vma
, file
, pgoff
+pglen
,
1102 vm_userfaultfd_ctx
)) {
1103 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1104 err
= vma_adjust(prev
, prev
->vm_start
,
1105 addr
, prev
->vm_pgoff
, NULL
);
1106 else /* cases 3, 8 */
1107 err
= vma_adjust(area
, addr
, next
->vm_end
,
1108 next
->vm_pgoff
- pglen
, NULL
);
1111 khugepaged_enter_vma_merge(area
, vm_flags
);
1119 * Rough compatbility check to quickly see if it's even worth looking
1120 * at sharing an anon_vma.
1122 * They need to have the same vm_file, and the flags can only differ
1123 * in things that mprotect may change.
1125 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1126 * we can merge the two vma's. For example, we refuse to merge a vma if
1127 * there is a vm_ops->close() function, because that indicates that the
1128 * driver is doing some kind of reference counting. But that doesn't
1129 * really matter for the anon_vma sharing case.
1131 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1133 return a
->vm_end
== b
->vm_start
&&
1134 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1135 a
->vm_file
== b
->vm_file
&&
1136 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
|VM_SOFTDIRTY
)) &&
1137 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1141 * Do some basic sanity checking to see if we can re-use the anon_vma
1142 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1143 * the same as 'old', the other will be the new one that is trying
1144 * to share the anon_vma.
1146 * NOTE! This runs with mm_sem held for reading, so it is possible that
1147 * the anon_vma of 'old' is concurrently in the process of being set up
1148 * by another page fault trying to merge _that_. But that's ok: if it
1149 * is being set up, that automatically means that it will be a singleton
1150 * acceptable for merging, so we can do all of this optimistically. But
1151 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1153 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1154 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1155 * is to return an anon_vma that is "complex" due to having gone through
1158 * We also make sure that the two vma's are compatible (adjacent,
1159 * and with the same memory policies). That's all stable, even with just
1160 * a read lock on the mm_sem.
1162 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1164 if (anon_vma_compatible(a
, b
)) {
1165 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1167 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1174 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1175 * neighbouring vmas for a suitable anon_vma, before it goes off
1176 * to allocate a new anon_vma. It checks because a repetitive
1177 * sequence of mprotects and faults may otherwise lead to distinct
1178 * anon_vmas being allocated, preventing vma merge in subsequent
1181 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1183 struct anon_vma
*anon_vma
;
1184 struct vm_area_struct
*near
;
1186 near
= vma
->vm_next
;
1190 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1194 near
= vma
->vm_prev
;
1198 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1203 * There's no absolute need to look only at touching neighbours:
1204 * we could search further afield for "compatible" anon_vmas.
1205 * But it would probably just be a waste of time searching,
1206 * or lead to too many vmas hanging off the same anon_vma.
1207 * We're trying to allow mprotect remerging later on,
1208 * not trying to minimize memory used for anon_vmas.
1213 #ifdef CONFIG_PROC_FS
1214 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
1215 struct file
*file
, long pages
)
1217 const unsigned long stack_flags
1218 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
1220 mm
->total_vm
+= pages
;
1223 mm
->shared_vm
+= pages
;
1224 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
1225 mm
->exec_vm
+= pages
;
1226 } else if (flags
& stack_flags
)
1227 mm
->stack_vm
+= pages
;
1229 #endif /* CONFIG_PROC_FS */
1232 * If a hint addr is less than mmap_min_addr change hint to be as
1233 * low as possible but still greater than mmap_min_addr
1235 static inline unsigned long round_hint_to_min(unsigned long hint
)
1238 if (((void *)hint
!= NULL
) &&
1239 (hint
< mmap_min_addr
))
1240 return PAGE_ALIGN(mmap_min_addr
);
1244 static inline int mlock_future_check(struct mm_struct
*mm
,
1245 unsigned long flags
,
1248 unsigned long locked
, lock_limit
;
1250 /* mlock MCL_FUTURE? */
1251 if (flags
& VM_LOCKED
) {
1252 locked
= len
>> PAGE_SHIFT
;
1253 locked
+= mm
->locked_vm
;
1254 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1255 lock_limit
>>= PAGE_SHIFT
;
1256 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1263 * The caller must hold down_write(¤t->mm->mmap_sem).
1265 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1266 unsigned long len
, unsigned long prot
,
1267 unsigned long flags
, vm_flags_t vm_flags
,
1268 unsigned long pgoff
, unsigned long *populate
)
1270 struct mm_struct
*mm
= current
->mm
;
1278 * Does the application expect PROT_READ to imply PROT_EXEC?
1280 * (the exception is when the underlying filesystem is noexec
1281 * mounted, in which case we dont add PROT_EXEC.)
1283 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1284 if (!(file
&& path_noexec(&file
->f_path
)))
1287 if (!(flags
& MAP_FIXED
))
1288 addr
= round_hint_to_min(addr
);
1290 /* Careful about overflows.. */
1291 len
= PAGE_ALIGN(len
);
1295 /* offset overflow? */
1296 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1299 /* Too many mappings? */
1300 if (mm
->map_count
> sysctl_max_map_count
)
1303 /* Obtain the address to map to. we verify (or select) it and ensure
1304 * that it represents a valid section of the address space.
1306 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1307 if (addr
& ~PAGE_MASK
)
1310 /* Do simple checking here so the lower-level routines won't have
1311 * to. we assume access permissions have been handled by the open
1312 * of the memory object, so we don't do any here.
1314 vm_flags
|= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1315 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1317 if (flags
& MAP_LOCKED
)
1318 if (!can_do_mlock())
1321 if (mlock_future_check(mm
, vm_flags
, len
))
1325 struct inode
*inode
= file_inode(file
);
1327 switch (flags
& MAP_TYPE
) {
1329 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1333 * Make sure we don't allow writing to an append-only
1336 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1340 * Make sure there are no mandatory locks on the file.
1342 if (locks_verify_locked(file
))
1345 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1346 if (!(file
->f_mode
& FMODE_WRITE
))
1347 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1351 if (!(file
->f_mode
& FMODE_READ
))
1353 if (path_noexec(&file
->f_path
)) {
1354 if (vm_flags
& VM_EXEC
)
1356 vm_flags
&= ~VM_MAYEXEC
;
1359 if (!file
->f_op
->mmap
)
1361 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1369 switch (flags
& MAP_TYPE
) {
1371 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1377 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1381 * Set pgoff according to addr for anon_vma.
1383 pgoff
= addr
>> PAGE_SHIFT
;
1391 * Set 'VM_NORESERVE' if we should not account for the
1392 * memory use of this mapping.
1394 if (flags
& MAP_NORESERVE
) {
1395 /* We honor MAP_NORESERVE if allowed to overcommit */
1396 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1397 vm_flags
|= VM_NORESERVE
;
1399 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1400 if (file
&& is_file_hugepages(file
))
1401 vm_flags
|= VM_NORESERVE
;
1404 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
);
1405 if (!IS_ERR_VALUE(addr
) &&
1406 ((vm_flags
& VM_LOCKED
) ||
1407 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1412 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1413 unsigned long, prot
, unsigned long, flags
,
1414 unsigned long, fd
, unsigned long, pgoff
)
1416 struct file
*file
= NULL
;
1417 unsigned long retval
= -EBADF
;
1419 if (!(flags
& MAP_ANONYMOUS
)) {
1420 audit_mmap_fd(fd
, flags
);
1424 if (is_file_hugepages(file
))
1425 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1427 if (unlikely(flags
& MAP_HUGETLB
&& !is_file_hugepages(file
)))
1429 } else if (flags
& MAP_HUGETLB
) {
1430 struct user_struct
*user
= NULL
;
1433 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & SHM_HUGE_MASK
);
1437 len
= ALIGN(len
, huge_page_size(hs
));
1439 * VM_NORESERVE is used because the reservations will be
1440 * taken when vm_ops->mmap() is called
1441 * A dummy user value is used because we are not locking
1442 * memory so no accounting is necessary
1444 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1446 &user
, HUGETLB_ANONHUGE_INODE
,
1447 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1449 return PTR_ERR(file
);
1452 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1454 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1462 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1463 struct mmap_arg_struct
{
1467 unsigned long flags
;
1469 unsigned long offset
;
1472 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1474 struct mmap_arg_struct a
;
1476 if (copy_from_user(&a
, arg
, sizeof(a
)))
1478 if (a
.offset
& ~PAGE_MASK
)
1481 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1482 a
.offset
>> PAGE_SHIFT
);
1484 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1487 * Some shared mappigns will want the pages marked read-only
1488 * to track write events. If so, we'll downgrade vm_page_prot
1489 * to the private version (using protection_map[] without the
1492 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1494 vm_flags_t vm_flags
= vma
->vm_flags
;
1496 /* If it was private or non-writable, the write bit is already clear */
1497 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1500 /* The backer wishes to know when pages are first written to? */
1501 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1504 /* The open routine did something to the protections that pgprot_modify
1505 * won't preserve? */
1506 if (pgprot_val(vma
->vm_page_prot
) !=
1507 pgprot_val(vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
)))
1510 /* Do we need to track softdirty? */
1511 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1514 /* Specialty mapping? */
1515 if (vm_flags
& VM_PFNMAP
)
1518 /* Can the mapping track the dirty pages? */
1519 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1520 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1524 * We account for memory if it's a private writeable mapping,
1525 * not hugepages and VM_NORESERVE wasn't set.
1527 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1530 * hugetlb has its own accounting separate from the core VM
1531 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1533 if (file
&& is_file_hugepages(file
))
1536 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1539 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1540 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
)
1542 struct mm_struct
*mm
= current
->mm
;
1543 struct vm_area_struct
*vma
, *prev
;
1545 struct rb_node
**rb_link
, *rb_parent
;
1546 unsigned long charged
= 0;
1548 /* Check against address space limit. */
1549 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
)) {
1550 unsigned long nr_pages
;
1553 * MAP_FIXED may remove pages of mappings that intersects with
1554 * requested mapping. Account for the pages it would unmap.
1556 if (!(vm_flags
& MAP_FIXED
))
1559 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1561 if (!may_expand_vm(mm
, (len
>> PAGE_SHIFT
) - nr_pages
))
1565 /* Clear old maps */
1567 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
1569 if (do_munmap(mm
, addr
, len
))
1574 * Private writable mapping: check memory availability
1576 if (accountable_mapping(file
, vm_flags
)) {
1577 charged
= len
>> PAGE_SHIFT
;
1578 if (security_vm_enough_memory_mm(mm
, charged
))
1580 vm_flags
|= VM_ACCOUNT
;
1584 * Can we just expand an old mapping?
1586 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1587 NULL
, file
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
1592 * Determine the object being mapped and call the appropriate
1593 * specific mapper. the address has already been validated, but
1594 * not unmapped, but the maps are removed from the list.
1596 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1603 vma
->vm_start
= addr
;
1604 vma
->vm_end
= addr
+ len
;
1605 vma
->vm_flags
= vm_flags
;
1606 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1607 vma
->vm_pgoff
= pgoff
;
1608 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1611 if (vm_flags
& VM_DENYWRITE
) {
1612 error
= deny_write_access(file
);
1616 if (vm_flags
& VM_SHARED
) {
1617 error
= mapping_map_writable(file
->f_mapping
);
1619 goto allow_write_and_free_vma
;
1622 /* ->mmap() can change vma->vm_file, but must guarantee that
1623 * vma_link() below can deny write-access if VM_DENYWRITE is set
1624 * and map writably if VM_SHARED is set. This usually means the
1625 * new file must not have been exposed to user-space, yet.
1627 vma
->vm_file
= get_file(file
);
1628 error
= file
->f_op
->mmap(file
, vma
);
1630 goto unmap_and_free_vma
;
1632 /* Can addr have changed??
1634 * Answer: Yes, several device drivers can do it in their
1635 * f_op->mmap method. -DaveM
1636 * Bug: If addr is changed, prev, rb_link, rb_parent should
1637 * be updated for vma_link()
1639 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1641 /* All file mapping must have ->vm_ops set */
1643 static const struct vm_operations_struct dummy_ops
= {};
1644 vma
->vm_ops
= &dummy_ops
;
1647 addr
= vma
->vm_start
;
1648 vm_flags
= vma
->vm_flags
;
1649 } else if (vm_flags
& VM_SHARED
) {
1650 error
= shmem_zero_setup(vma
);
1655 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1656 /* Once vma denies write, undo our temporary denial count */
1658 if (vm_flags
& VM_SHARED
)
1659 mapping_unmap_writable(file
->f_mapping
);
1660 if (vm_flags
& VM_DENYWRITE
)
1661 allow_write_access(file
);
1663 file
= vma
->vm_file
;
1665 perf_event_mmap(vma
);
1667 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1668 if (vm_flags
& VM_LOCKED
) {
1669 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1670 vma
== get_gate_vma(current
->mm
)))
1671 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1673 vma
->vm_flags
&= ~VM_LOCKED
;
1680 * New (or expanded) vma always get soft dirty status.
1681 * Otherwise user-space soft-dirty page tracker won't
1682 * be able to distinguish situation when vma area unmapped,
1683 * then new mapped in-place (which must be aimed as
1684 * a completely new data area).
1686 vma
->vm_flags
|= VM_SOFTDIRTY
;
1688 vma_set_page_prot(vma
);
1693 vma
->vm_file
= NULL
;
1696 /* Undo any partial mapping done by a device driver. */
1697 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1699 if (vm_flags
& VM_SHARED
)
1700 mapping_unmap_writable(file
->f_mapping
);
1701 allow_write_and_free_vma
:
1702 if (vm_flags
& VM_DENYWRITE
)
1703 allow_write_access(file
);
1705 kmem_cache_free(vm_area_cachep
, vma
);
1708 vm_unacct_memory(charged
);
1712 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1715 * We implement the search by looking for an rbtree node that
1716 * immediately follows a suitable gap. That is,
1717 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1718 * - gap_end = vma->vm_start >= info->low_limit + length;
1719 * - gap_end - gap_start >= length
1722 struct mm_struct
*mm
= current
->mm
;
1723 struct vm_area_struct
*vma
;
1724 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1726 /* Adjust search length to account for worst case alignment overhead */
1727 length
= info
->length
+ info
->align_mask
;
1728 if (length
< info
->length
)
1731 /* Adjust search limits by the desired length */
1732 if (info
->high_limit
< length
)
1734 high_limit
= info
->high_limit
- length
;
1736 if (info
->low_limit
> high_limit
)
1738 low_limit
= info
->low_limit
+ length
;
1740 /* Check if rbtree root looks promising */
1741 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1743 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1744 if (vma
->rb_subtree_gap
< length
)
1748 /* Visit left subtree if it looks promising */
1749 gap_end
= vma
->vm_start
;
1750 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1751 struct vm_area_struct
*left
=
1752 rb_entry(vma
->vm_rb
.rb_left
,
1753 struct vm_area_struct
, vm_rb
);
1754 if (left
->rb_subtree_gap
>= length
) {
1760 gap_start
= vma
->vm_prev
? vma
->vm_prev
->vm_end
: 0;
1762 /* Check if current node has a suitable gap */
1763 if (gap_start
> high_limit
)
1765 if (gap_end
>= low_limit
&& gap_end
- gap_start
>= length
)
1768 /* Visit right subtree if it looks promising */
1769 if (vma
->vm_rb
.rb_right
) {
1770 struct vm_area_struct
*right
=
1771 rb_entry(vma
->vm_rb
.rb_right
,
1772 struct vm_area_struct
, vm_rb
);
1773 if (right
->rb_subtree_gap
>= length
) {
1779 /* Go back up the rbtree to find next candidate node */
1781 struct rb_node
*prev
= &vma
->vm_rb
;
1782 if (!rb_parent(prev
))
1784 vma
= rb_entry(rb_parent(prev
),
1785 struct vm_area_struct
, vm_rb
);
1786 if (prev
== vma
->vm_rb
.rb_left
) {
1787 gap_start
= vma
->vm_prev
->vm_end
;
1788 gap_end
= vma
->vm_start
;
1795 /* Check highest gap, which does not precede any rbtree node */
1796 gap_start
= mm
->highest_vm_end
;
1797 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1798 if (gap_start
> high_limit
)
1802 /* We found a suitable gap. Clip it with the original low_limit. */
1803 if (gap_start
< info
->low_limit
)
1804 gap_start
= info
->low_limit
;
1806 /* Adjust gap address to the desired alignment */
1807 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1809 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1810 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1814 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1816 struct mm_struct
*mm
= current
->mm
;
1817 struct vm_area_struct
*vma
;
1818 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1820 /* Adjust search length to account for worst case alignment overhead */
1821 length
= info
->length
+ info
->align_mask
;
1822 if (length
< info
->length
)
1826 * Adjust search limits by the desired length.
1827 * See implementation comment at top of unmapped_area().
1829 gap_end
= info
->high_limit
;
1830 if (gap_end
< length
)
1832 high_limit
= gap_end
- length
;
1834 if (info
->low_limit
> high_limit
)
1836 low_limit
= info
->low_limit
+ length
;
1838 /* Check highest gap, which does not precede any rbtree node */
1839 gap_start
= mm
->highest_vm_end
;
1840 if (gap_start
<= high_limit
)
1843 /* Check if rbtree root looks promising */
1844 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1846 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1847 if (vma
->rb_subtree_gap
< length
)
1851 /* Visit right subtree if it looks promising */
1852 gap_start
= vma
->vm_prev
? vma
->vm_prev
->vm_end
: 0;
1853 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1854 struct vm_area_struct
*right
=
1855 rb_entry(vma
->vm_rb
.rb_right
,
1856 struct vm_area_struct
, vm_rb
);
1857 if (right
->rb_subtree_gap
>= length
) {
1864 /* Check if current node has a suitable gap */
1865 gap_end
= vma
->vm_start
;
1866 if (gap_end
< low_limit
)
1868 if (gap_start
<= high_limit
&& gap_end
- gap_start
>= length
)
1871 /* Visit left subtree if it looks promising */
1872 if (vma
->vm_rb
.rb_left
) {
1873 struct vm_area_struct
*left
=
1874 rb_entry(vma
->vm_rb
.rb_left
,
1875 struct vm_area_struct
, vm_rb
);
1876 if (left
->rb_subtree_gap
>= length
) {
1882 /* Go back up the rbtree to find next candidate node */
1884 struct rb_node
*prev
= &vma
->vm_rb
;
1885 if (!rb_parent(prev
))
1887 vma
= rb_entry(rb_parent(prev
),
1888 struct vm_area_struct
, vm_rb
);
1889 if (prev
== vma
->vm_rb
.rb_right
) {
1890 gap_start
= vma
->vm_prev
?
1891 vma
->vm_prev
->vm_end
: 0;
1898 /* We found a suitable gap. Clip it with the original high_limit. */
1899 if (gap_end
> info
->high_limit
)
1900 gap_end
= info
->high_limit
;
1903 /* Compute highest gap address at the desired alignment */
1904 gap_end
-= info
->length
;
1905 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
1907 VM_BUG_ON(gap_end
< info
->low_limit
);
1908 VM_BUG_ON(gap_end
< gap_start
);
1912 /* Get an address range which is currently unmapped.
1913 * For shmat() with addr=0.
1915 * Ugly calling convention alert:
1916 * Return value with the low bits set means error value,
1918 * if (ret & ~PAGE_MASK)
1921 * This function "knows" that -ENOMEM has the bits set.
1923 #ifndef HAVE_ARCH_UNMAPPED_AREA
1925 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1926 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1928 struct mm_struct
*mm
= current
->mm
;
1929 struct vm_area_struct
*vma
;
1930 struct vm_unmapped_area_info info
;
1932 if (len
> TASK_SIZE
- mmap_min_addr
)
1935 if (flags
& MAP_FIXED
)
1939 addr
= PAGE_ALIGN(addr
);
1940 vma
= find_vma(mm
, addr
);
1941 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1942 (!vma
|| addr
+ len
<= vma
->vm_start
))
1948 info
.low_limit
= mm
->mmap_base
;
1949 info
.high_limit
= TASK_SIZE
;
1950 info
.align_mask
= 0;
1951 return vm_unmapped_area(&info
);
1956 * This mmap-allocator allocates new areas top-down from below the
1957 * stack's low limit (the base):
1959 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1961 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1962 const unsigned long len
, const unsigned long pgoff
,
1963 const unsigned long flags
)
1965 struct vm_area_struct
*vma
;
1966 struct mm_struct
*mm
= current
->mm
;
1967 unsigned long addr
= addr0
;
1968 struct vm_unmapped_area_info info
;
1970 /* requested length too big for entire address space */
1971 if (len
> TASK_SIZE
- mmap_min_addr
)
1974 if (flags
& MAP_FIXED
)
1977 /* requesting a specific address */
1979 addr
= PAGE_ALIGN(addr
);
1980 vma
= find_vma(mm
, addr
);
1981 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
1982 (!vma
|| addr
+ len
<= vma
->vm_start
))
1986 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1988 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
1989 info
.high_limit
= mm
->mmap_base
;
1990 info
.align_mask
= 0;
1991 addr
= vm_unmapped_area(&info
);
1994 * A failed mmap() very likely causes application failure,
1995 * so fall back to the bottom-up function here. This scenario
1996 * can happen with large stack limits and large mmap()
1999 if (addr
& ~PAGE_MASK
) {
2000 VM_BUG_ON(addr
!= -ENOMEM
);
2002 info
.low_limit
= TASK_UNMAPPED_BASE
;
2003 info
.high_limit
= TASK_SIZE
;
2004 addr
= vm_unmapped_area(&info
);
2012 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2013 unsigned long pgoff
, unsigned long flags
)
2015 unsigned long (*get_area
)(struct file
*, unsigned long,
2016 unsigned long, unsigned long, unsigned long);
2018 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2022 /* Careful about overflows.. */
2023 if (len
> TASK_SIZE
)
2026 get_area
= current
->mm
->get_unmapped_area
;
2027 if (file
&& file
->f_op
->get_unmapped_area
)
2028 get_area
= file
->f_op
->get_unmapped_area
;
2029 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2030 if (IS_ERR_VALUE(addr
))
2033 if (addr
> TASK_SIZE
- len
)
2035 if (addr
& ~PAGE_MASK
)
2038 addr
= arch_rebalance_pgtables(addr
, len
);
2039 error
= security_mmap_addr(addr
);
2040 return error
? error
: addr
;
2043 EXPORT_SYMBOL(get_unmapped_area
);
2045 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2046 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2048 struct rb_node
*rb_node
;
2049 struct vm_area_struct
*vma
;
2051 /* Check the cache first. */
2052 vma
= vmacache_find(mm
, addr
);
2056 rb_node
= mm
->mm_rb
.rb_node
;
2060 struct vm_area_struct
*tmp
;
2062 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2064 if (tmp
->vm_end
> addr
) {
2066 if (tmp
->vm_start
<= addr
)
2068 rb_node
= rb_node
->rb_left
;
2070 rb_node
= rb_node
->rb_right
;
2074 vmacache_update(addr
, vma
);
2078 EXPORT_SYMBOL(find_vma
);
2081 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2083 struct vm_area_struct
*
2084 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2085 struct vm_area_struct
**pprev
)
2087 struct vm_area_struct
*vma
;
2089 vma
= find_vma(mm
, addr
);
2091 *pprev
= vma
->vm_prev
;
2093 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2096 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2097 rb_node
= rb_node
->rb_right
;
2104 * Verify that the stack growth is acceptable and
2105 * update accounting. This is shared with both the
2106 * grow-up and grow-down cases.
2108 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
2110 struct mm_struct
*mm
= vma
->vm_mm
;
2111 struct rlimit
*rlim
= current
->signal
->rlim
;
2112 unsigned long new_start
, actual_size
;
2114 /* address space limit tests */
2115 if (!may_expand_vm(mm
, grow
))
2118 /* Stack limit test */
2120 if (size
&& (vma
->vm_flags
& (VM_GROWSUP
| VM_GROWSDOWN
)))
2121 actual_size
-= PAGE_SIZE
;
2122 if (actual_size
> READ_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
2125 /* mlock limit tests */
2126 if (vma
->vm_flags
& VM_LOCKED
) {
2127 unsigned long locked
;
2128 unsigned long limit
;
2129 locked
= mm
->locked_vm
+ grow
;
2130 limit
= READ_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
2131 limit
>>= PAGE_SHIFT
;
2132 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2136 /* Check to ensure the stack will not grow into a hugetlb-only region */
2137 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2139 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2143 * Overcommit.. This must be the final test, as it will
2144 * update security statistics.
2146 if (security_vm_enough_memory_mm(mm
, grow
))
2149 /* Ok, everything looks good - let it rip */
2150 if (vma
->vm_flags
& VM_LOCKED
)
2151 mm
->locked_vm
+= grow
;
2152 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
2156 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2158 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2159 * vma is the last one with address > vma->vm_end. Have to extend vma.
2161 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2165 if (!(vma
->vm_flags
& VM_GROWSUP
))
2169 * We must make sure the anon_vma is allocated
2170 * so that the anon_vma locking is not a noop.
2172 if (unlikely(anon_vma_prepare(vma
)))
2174 vma_lock_anon_vma(vma
);
2177 * vma->vm_start/vm_end cannot change under us because the caller
2178 * is required to hold the mmap_sem in read mode. We need the
2179 * anon_vma lock to serialize against concurrent expand_stacks.
2180 * Also guard against wrapping around to address 0.
2182 if (address
< PAGE_ALIGN(address
+4))
2183 address
= PAGE_ALIGN(address
+4);
2185 vma_unlock_anon_vma(vma
);
2190 /* Somebody else might have raced and expanded it already */
2191 if (address
> vma
->vm_end
) {
2192 unsigned long size
, grow
;
2194 size
= address
- vma
->vm_start
;
2195 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2198 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2199 error
= acct_stack_growth(vma
, size
, grow
);
2202 * vma_gap_update() doesn't support concurrent
2203 * updates, but we only hold a shared mmap_sem
2204 * lock here, so we need to protect against
2205 * concurrent vma expansions.
2206 * vma_lock_anon_vma() doesn't help here, as
2207 * we don't guarantee that all growable vmas
2208 * in a mm share the same root anon vma.
2209 * So, we reuse mm->page_table_lock to guard
2210 * against concurrent vma expansions.
2212 spin_lock(&vma
->vm_mm
->page_table_lock
);
2213 anon_vma_interval_tree_pre_update_vma(vma
);
2214 vma
->vm_end
= address
;
2215 anon_vma_interval_tree_post_update_vma(vma
);
2217 vma_gap_update(vma
->vm_next
);
2219 vma
->vm_mm
->highest_vm_end
= address
;
2220 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2222 perf_event_mmap(vma
);
2226 vma_unlock_anon_vma(vma
);
2227 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2228 validate_mm(vma
->vm_mm
);
2231 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2234 * vma is the first one with address < vma->vm_start. Have to extend vma.
2236 int expand_downwards(struct vm_area_struct
*vma
,
2237 unsigned long address
)
2242 * We must make sure the anon_vma is allocated
2243 * so that the anon_vma locking is not a noop.
2245 if (unlikely(anon_vma_prepare(vma
)))
2248 address
&= PAGE_MASK
;
2249 error
= security_mmap_addr(address
);
2253 vma_lock_anon_vma(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.
2261 /* Somebody else might have raced and expanded it already */
2262 if (address
< vma
->vm_start
) {
2263 unsigned long size
, grow
;
2265 size
= vma
->vm_end
- address
;
2266 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2269 if (grow
<= vma
->vm_pgoff
) {
2270 error
= acct_stack_growth(vma
, size
, grow
);
2273 * vma_gap_update() doesn't support concurrent
2274 * updates, but we only hold a shared mmap_sem
2275 * lock here, so we need to protect against
2276 * concurrent vma expansions.
2277 * vma_lock_anon_vma() doesn't help here, as
2278 * we don't guarantee that all growable vmas
2279 * in a mm share the same root anon vma.
2280 * So, we reuse mm->page_table_lock to guard
2281 * against concurrent vma expansions.
2283 spin_lock(&vma
->vm_mm
->page_table_lock
);
2284 anon_vma_interval_tree_pre_update_vma(vma
);
2285 vma
->vm_start
= address
;
2286 vma
->vm_pgoff
-= grow
;
2287 anon_vma_interval_tree_post_update_vma(vma
);
2288 vma_gap_update(vma
);
2289 spin_unlock(&vma
->vm_mm
->page_table_lock
);
2291 perf_event_mmap(vma
);
2295 vma_unlock_anon_vma(vma
);
2296 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2297 validate_mm(vma
->vm_mm
);
2302 * Note how expand_stack() refuses to expand the stack all the way to
2303 * abut the next virtual mapping, *unless* that mapping itself is also
2304 * a stack mapping. We want to leave room for a guard page, after all
2305 * (the guard page itself is not added here, that is done by the
2306 * actual page faulting logic)
2308 * This matches the behavior of the guard page logic (see mm/memory.c:
2309 * check_stack_guard_page()), which only allows the guard page to be
2310 * removed under these circumstances.
2312 #ifdef CONFIG_STACK_GROWSUP
2313 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2315 struct vm_area_struct
*next
;
2317 address
&= PAGE_MASK
;
2318 next
= vma
->vm_next
;
2319 if (next
&& next
->vm_start
== address
+ PAGE_SIZE
) {
2320 if (!(next
->vm_flags
& VM_GROWSUP
))
2323 return expand_upwards(vma
, address
);
2326 struct vm_area_struct
*
2327 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2329 struct vm_area_struct
*vma
, *prev
;
2332 vma
= find_vma_prev(mm
, addr
, &prev
);
2333 if (vma
&& (vma
->vm_start
<= addr
))
2335 if (!prev
|| expand_stack(prev
, addr
))
2337 if (prev
->vm_flags
& VM_LOCKED
)
2338 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2342 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2344 struct vm_area_struct
*prev
;
2346 address
&= PAGE_MASK
;
2347 prev
= vma
->vm_prev
;
2348 if (prev
&& prev
->vm_end
== address
) {
2349 if (!(prev
->vm_flags
& VM_GROWSDOWN
))
2352 return expand_downwards(vma
, address
);
2355 struct vm_area_struct
*
2356 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2358 struct vm_area_struct
*vma
;
2359 unsigned long start
;
2362 vma
= find_vma(mm
, addr
);
2365 if (vma
->vm_start
<= addr
)
2367 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2369 start
= vma
->vm_start
;
2370 if (expand_stack(vma
, addr
))
2372 if (vma
->vm_flags
& VM_LOCKED
)
2373 populate_vma_page_range(vma
, addr
, start
, NULL
);
2378 EXPORT_SYMBOL_GPL(find_extend_vma
);
2381 * Ok - we have the memory areas we should free on the vma list,
2382 * so release them, and do the vma updates.
2384 * Called with the mm semaphore held.
2386 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2388 unsigned long nr_accounted
= 0;
2390 /* Update high watermark before we lower total_vm */
2391 update_hiwater_vm(mm
);
2393 long nrpages
= vma_pages(vma
);
2395 if (vma
->vm_flags
& VM_ACCOUNT
)
2396 nr_accounted
+= nrpages
;
2397 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
2398 vma
= remove_vma(vma
);
2400 vm_unacct_memory(nr_accounted
);
2405 * Get rid of page table information in the indicated region.
2407 * Called with the mm semaphore held.
2409 static void unmap_region(struct mm_struct
*mm
,
2410 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2411 unsigned long start
, unsigned long end
)
2413 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2414 struct mmu_gather tlb
;
2417 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2418 update_hiwater_rss(mm
);
2419 unmap_vmas(&tlb
, vma
, start
, end
);
2420 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2421 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2422 tlb_finish_mmu(&tlb
, start
, end
);
2426 * Create a list of vma's touched by the unmap, removing them from the mm's
2427 * vma list as we go..
2430 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2431 struct vm_area_struct
*prev
, unsigned long end
)
2433 struct vm_area_struct
**insertion_point
;
2434 struct vm_area_struct
*tail_vma
= NULL
;
2436 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2437 vma
->vm_prev
= NULL
;
2439 vma_rb_erase(vma
, &mm
->mm_rb
);
2443 } while (vma
&& vma
->vm_start
< end
);
2444 *insertion_point
= vma
;
2446 vma
->vm_prev
= prev
;
2447 vma_gap_update(vma
);
2449 mm
->highest_vm_end
= prev
? prev
->vm_end
: 0;
2450 tail_vma
->vm_next
= NULL
;
2452 /* Kill the cache */
2453 vmacache_invalidate(mm
);
2457 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2458 * munmap path where it doesn't make sense to fail.
2460 static int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2461 unsigned long addr
, int new_below
)
2463 struct vm_area_struct
*new;
2466 if (is_vm_hugetlb_page(vma
) && (addr
&
2467 ~(huge_page_mask(hstate_vma(vma
)))))
2470 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2474 /* most fields are the same, copy all, and then fixup */
2477 INIT_LIST_HEAD(&new->anon_vma_chain
);
2482 new->vm_start
= addr
;
2483 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2486 err
= vma_dup_policy(vma
, new);
2490 err
= anon_vma_clone(new, vma
);
2495 get_file(new->vm_file
);
2497 if (new->vm_ops
&& new->vm_ops
->open
)
2498 new->vm_ops
->open(new);
2501 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2502 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2504 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2510 /* Clean everything up if vma_adjust failed. */
2511 if (new->vm_ops
&& new->vm_ops
->close
)
2512 new->vm_ops
->close(new);
2515 unlink_anon_vmas(new);
2517 mpol_put(vma_policy(new));
2519 kmem_cache_free(vm_area_cachep
, new);
2524 * Split a vma into two pieces at address 'addr', a new vma is allocated
2525 * either for the first part or the tail.
2527 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2528 unsigned long addr
, int new_below
)
2530 if (mm
->map_count
>= sysctl_max_map_count
)
2533 return __split_vma(mm
, vma
, addr
, new_below
);
2536 /* Munmap is split into 2 main parts -- this part which finds
2537 * what needs doing, and the areas themselves, which do the
2538 * work. This now handles partial unmappings.
2539 * Jeremy Fitzhardinge <jeremy@goop.org>
2541 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2544 struct vm_area_struct
*vma
, *prev
, *last
;
2546 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2549 len
= PAGE_ALIGN(len
);
2553 /* Find the first overlapping VMA */
2554 vma
= find_vma(mm
, start
);
2557 prev
= vma
->vm_prev
;
2558 /* we have start < vma->vm_end */
2560 /* if it doesn't overlap, we have nothing.. */
2562 if (vma
->vm_start
>= end
)
2566 * If we need to split any vma, do it now to save pain later.
2568 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2569 * unmapped vm_area_struct will remain in use: so lower split_vma
2570 * places tmp vma above, and higher split_vma places tmp vma below.
2572 if (start
> vma
->vm_start
) {
2576 * Make sure that map_count on return from munmap() will
2577 * not exceed its limit; but let map_count go just above
2578 * its limit temporarily, to help free resources as expected.
2580 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2583 error
= __split_vma(mm
, vma
, start
, 0);
2589 /* Does it split the last one? */
2590 last
= find_vma(mm
, end
);
2591 if (last
&& end
> last
->vm_start
) {
2592 int error
= __split_vma(mm
, last
, end
, 1);
2596 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2599 * unlock any mlock()ed ranges before detaching vmas
2601 if (mm
->locked_vm
) {
2602 struct vm_area_struct
*tmp
= vma
;
2603 while (tmp
&& tmp
->vm_start
< end
) {
2604 if (tmp
->vm_flags
& VM_LOCKED
) {
2605 mm
->locked_vm
-= vma_pages(tmp
);
2606 munlock_vma_pages_all(tmp
);
2613 * Remove the vma's, and unmap the actual pages
2615 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2616 unmap_region(mm
, vma
, prev
, start
, end
);
2618 arch_unmap(mm
, vma
, start
, end
);
2620 /* Fix up all other VM information */
2621 remove_vma_list(mm
, vma
);
2626 int vm_munmap(unsigned long start
, size_t len
)
2629 struct mm_struct
*mm
= current
->mm
;
2631 down_write(&mm
->mmap_sem
);
2632 ret
= do_munmap(mm
, start
, len
);
2633 up_write(&mm
->mmap_sem
);
2636 EXPORT_SYMBOL(vm_munmap
);
2638 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2640 profile_munmap(addr
);
2641 return vm_munmap(addr
, len
);
2646 * Emulation of deprecated remap_file_pages() syscall.
2648 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2649 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2652 struct mm_struct
*mm
= current
->mm
;
2653 struct vm_area_struct
*vma
;
2654 unsigned long populate
= 0;
2655 unsigned long ret
= -EINVAL
;
2658 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2659 "See Documentation/vm/remap_file_pages.txt.\n",
2660 current
->comm
, current
->pid
);
2664 start
= start
& PAGE_MASK
;
2665 size
= size
& PAGE_MASK
;
2667 if (start
+ size
<= start
)
2670 /* Does pgoff wrap? */
2671 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2674 down_write(&mm
->mmap_sem
);
2675 vma
= find_vma(mm
, start
);
2677 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2680 if (start
< vma
->vm_start
|| start
+ size
> vma
->vm_end
)
2683 if (pgoff
== linear_page_index(vma
, start
)) {
2688 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
2689 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
2690 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
2692 flags
&= MAP_NONBLOCK
;
2693 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
2694 if (vma
->vm_flags
& VM_LOCKED
) {
2695 flags
|= MAP_LOCKED
;
2696 /* drop PG_Mlocked flag for over-mapped range */
2697 munlock_vma_pages_range(vma
, start
, start
+ size
);
2700 file
= get_file(vma
->vm_file
);
2701 ret
= do_mmap_pgoff(vma
->vm_file
, start
, size
,
2702 prot
, flags
, pgoff
, &populate
);
2705 up_write(&mm
->mmap_sem
);
2707 mm_populate(ret
, populate
);
2708 if (!IS_ERR_VALUE(ret
))
2713 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2715 #ifdef CONFIG_DEBUG_VM
2716 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2718 up_read(&mm
->mmap_sem
);
2724 * this is really a simplified "do_mmap". it only handles
2725 * anonymous maps. eventually we may be able to do some
2726 * brk-specific accounting here.
2728 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2730 struct mm_struct
*mm
= current
->mm
;
2731 struct vm_area_struct
*vma
, *prev
;
2732 unsigned long flags
;
2733 struct rb_node
**rb_link
, *rb_parent
;
2734 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2737 len
= PAGE_ALIGN(len
);
2741 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2743 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2744 if (error
& ~PAGE_MASK
)
2747 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
2752 * mm->mmap_sem is required to protect against another thread
2753 * changing the mappings in case we sleep.
2755 verify_mm_writelocked(mm
);
2758 * Clear old maps. this also does some error checking for us
2760 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
2762 if (do_munmap(mm
, addr
, len
))
2766 /* Check against address space limits *after* clearing old maps... */
2767 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2770 if (mm
->map_count
> sysctl_max_map_count
)
2773 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2776 /* Can we just expand an old private anonymous mapping? */
2777 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2778 NULL
, NULL
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
2783 * create a vma struct for an anonymous mapping
2785 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2787 vm_unacct_memory(len
>> PAGE_SHIFT
);
2791 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2793 vma
->vm_start
= addr
;
2794 vma
->vm_end
= addr
+ len
;
2795 vma
->vm_pgoff
= pgoff
;
2796 vma
->vm_flags
= flags
;
2797 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2798 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2800 perf_event_mmap(vma
);
2801 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2802 if (flags
& VM_LOCKED
)
2803 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2804 vma
->vm_flags
|= VM_SOFTDIRTY
;
2808 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2810 struct mm_struct
*mm
= current
->mm
;
2814 down_write(&mm
->mmap_sem
);
2815 ret
= do_brk(addr
, len
);
2816 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
2817 up_write(&mm
->mmap_sem
);
2819 mm_populate(addr
, len
);
2822 EXPORT_SYMBOL(vm_brk
);
2824 /* Release all mmaps. */
2825 void exit_mmap(struct mm_struct
*mm
)
2827 struct mmu_gather tlb
;
2828 struct vm_area_struct
*vma
;
2829 unsigned long nr_accounted
= 0;
2831 /* mm's last user has gone, and its about to be pulled down */
2832 mmu_notifier_release(mm
);
2834 if (mm
->locked_vm
) {
2837 if (vma
->vm_flags
& VM_LOCKED
)
2838 munlock_vma_pages_all(vma
);
2846 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2851 tlb_gather_mmu(&tlb
, mm
, 0, -1);
2852 /* update_hiwater_rss(mm) here? but nobody should be looking */
2853 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2854 unmap_vmas(&tlb
, vma
, 0, -1);
2856 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
2857 tlb_finish_mmu(&tlb
, 0, -1);
2860 * Walk the list again, actually closing and freeing it,
2861 * with preemption enabled, without holding any MM locks.
2864 if (vma
->vm_flags
& VM_ACCOUNT
)
2865 nr_accounted
+= vma_pages(vma
);
2866 vma
= remove_vma(vma
);
2868 vm_unacct_memory(nr_accounted
);
2871 /* Insert vm structure into process list sorted by address
2872 * and into the inode's i_mmap tree. If vm_file is non-NULL
2873 * then i_mmap_rwsem is taken here.
2875 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2877 struct vm_area_struct
*prev
;
2878 struct rb_node
**rb_link
, *rb_parent
;
2880 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
2881 &prev
, &rb_link
, &rb_parent
))
2883 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2884 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2888 * The vm_pgoff of a purely anonymous vma should be irrelevant
2889 * until its first write fault, when page's anon_vma and index
2890 * are set. But now set the vm_pgoff it will almost certainly
2891 * end up with (unless mremap moves it elsewhere before that
2892 * first wfault), so /proc/pid/maps tells a consistent story.
2894 * By setting it to reflect the virtual start address of the
2895 * vma, merges and splits can happen in a seamless way, just
2896 * using the existing file pgoff checks and manipulations.
2897 * Similarly in do_mmap_pgoff and in do_brk.
2899 if (vma_is_anonymous(vma
)) {
2900 BUG_ON(vma
->anon_vma
);
2901 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2904 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2909 * Copy the vma structure to a new location in the same mm,
2910 * prior to moving page table entries, to effect an mremap move.
2912 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2913 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
2914 bool *need_rmap_locks
)
2916 struct vm_area_struct
*vma
= *vmap
;
2917 unsigned long vma_start
= vma
->vm_start
;
2918 struct mm_struct
*mm
= vma
->vm_mm
;
2919 struct vm_area_struct
*new_vma
, *prev
;
2920 struct rb_node
**rb_link
, *rb_parent
;
2921 bool faulted_in_anon_vma
= true;
2924 * If anonymous vma has not yet been faulted, update new pgoff
2925 * to match new location, to increase its chance of merging.
2927 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
2928 pgoff
= addr
>> PAGE_SHIFT
;
2929 faulted_in_anon_vma
= false;
2932 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
2933 return NULL
; /* should never get here */
2934 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2935 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
2936 vma
->vm_userfaultfd_ctx
);
2939 * Source vma may have been merged into new_vma
2941 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2942 vma_start
< new_vma
->vm_end
)) {
2944 * The only way we can get a vma_merge with
2945 * self during an mremap is if the vma hasn't
2946 * been faulted in yet and we were allowed to
2947 * reset the dst vma->vm_pgoff to the
2948 * destination address of the mremap to allow
2949 * the merge to happen. mremap must change the
2950 * vm_pgoff linearity between src and dst vmas
2951 * (in turn preventing a vma_merge) to be
2952 * safe. It is only safe to keep the vm_pgoff
2953 * linear if there are no pages mapped yet.
2955 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
2956 *vmap
= vma
= new_vma
;
2958 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
2960 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2964 new_vma
->vm_start
= addr
;
2965 new_vma
->vm_end
= addr
+ len
;
2966 new_vma
->vm_pgoff
= pgoff
;
2967 if (vma_dup_policy(vma
, new_vma
))
2969 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2970 if (anon_vma_clone(new_vma
, vma
))
2971 goto out_free_mempol
;
2972 if (new_vma
->vm_file
)
2973 get_file(new_vma
->vm_file
);
2974 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2975 new_vma
->vm_ops
->open(new_vma
);
2976 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2977 *need_rmap_locks
= false;
2982 mpol_put(vma_policy(new_vma
));
2984 kmem_cache_free(vm_area_cachep
, new_vma
);
2990 * Return true if the calling process may expand its vm space by the passed
2993 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2995 unsigned long cur
= mm
->total_vm
; /* pages */
2998 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
3000 if (cur
+ npages
> lim
)
3005 static int special_mapping_fault(struct vm_area_struct
*vma
,
3006 struct vm_fault
*vmf
);
3009 * Having a close hook prevents vma merging regardless of flags.
3011 static void special_mapping_close(struct vm_area_struct
*vma
)
3015 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3017 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3020 static const struct vm_operations_struct special_mapping_vmops
= {
3021 .close
= special_mapping_close
,
3022 .fault
= special_mapping_fault
,
3023 .name
= special_mapping_name
,
3026 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3027 .close
= special_mapping_close
,
3028 .fault
= special_mapping_fault
,
3031 static int special_mapping_fault(struct vm_area_struct
*vma
,
3032 struct vm_fault
*vmf
)
3035 struct page
**pages
;
3037 if (vma
->vm_ops
== &legacy_special_mapping_vmops
)
3038 pages
= vma
->vm_private_data
;
3040 pages
= ((struct vm_special_mapping
*)vma
->vm_private_data
)->
3043 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3047 struct page
*page
= *pages
;
3053 return VM_FAULT_SIGBUS
;
3056 static struct vm_area_struct
*__install_special_mapping(
3057 struct mm_struct
*mm
,
3058 unsigned long addr
, unsigned long len
,
3059 unsigned long vm_flags
, const struct vm_operations_struct
*ops
,
3063 struct vm_area_struct
*vma
;
3065 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
3066 if (unlikely(vma
== NULL
))
3067 return ERR_PTR(-ENOMEM
);
3069 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
3071 vma
->vm_start
= addr
;
3072 vma
->vm_end
= addr
+ len
;
3074 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
3075 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3078 vma
->vm_private_data
= priv
;
3080 ret
= insert_vm_struct(mm
, vma
);
3084 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3086 perf_event_mmap(vma
);
3091 kmem_cache_free(vm_area_cachep
, vma
);
3092 return ERR_PTR(ret
);
3096 * Called with mm->mmap_sem held for writing.
3097 * Insert a new vma covering the given region, with the given flags.
3098 * Its pages are supplied by the given array of struct page *.
3099 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3100 * The region past the last page supplied will always produce SIGBUS.
3101 * The array pointer and the pages it points to are assumed to stay alive
3102 * for as long as this mapping might exist.
3104 struct vm_area_struct
*_install_special_mapping(
3105 struct mm_struct
*mm
,
3106 unsigned long addr
, unsigned long len
,
3107 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3109 return __install_special_mapping(mm
, addr
, len
, vm_flags
,
3110 &special_mapping_vmops
, (void *)spec
);
3113 int install_special_mapping(struct mm_struct
*mm
,
3114 unsigned long addr
, unsigned long len
,
3115 unsigned long vm_flags
, struct page
**pages
)
3117 struct vm_area_struct
*vma
= __install_special_mapping(
3118 mm
, addr
, len
, vm_flags
, &legacy_special_mapping_vmops
,
3121 return PTR_ERR_OR_ZERO(vma
);
3124 static DEFINE_MUTEX(mm_all_locks_mutex
);
3126 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3128 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3130 * The LSB of head.next can't change from under us
3131 * because we hold the mm_all_locks_mutex.
3133 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3135 * We can safely modify head.next after taking the
3136 * anon_vma->root->rwsem. If some other vma in this mm shares
3137 * the same anon_vma we won't take it again.
3139 * No need of atomic instructions here, head.next
3140 * can't change from under us thanks to the
3141 * anon_vma->root->rwsem.
3143 if (__test_and_set_bit(0, (unsigned long *)
3144 &anon_vma
->root
->rb_root
.rb_node
))
3149 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3151 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3153 * AS_MM_ALL_LOCKS can't change from under us because
3154 * we hold the mm_all_locks_mutex.
3156 * Operations on ->flags have to be atomic because
3157 * even if AS_MM_ALL_LOCKS is stable thanks to the
3158 * mm_all_locks_mutex, there may be other cpus
3159 * changing other bitflags in parallel to us.
3161 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3163 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_sem
);
3168 * This operation locks against the VM for all pte/vma/mm related
3169 * operations that could ever happen on a certain mm. This includes
3170 * vmtruncate, try_to_unmap, and all page faults.
3172 * The caller must take the mmap_sem in write mode before calling
3173 * mm_take_all_locks(). The caller isn't allowed to release the
3174 * mmap_sem until mm_drop_all_locks() returns.
3176 * mmap_sem in write mode is required in order to block all operations
3177 * that could modify pagetables and free pages without need of
3178 * altering the vma layout. It's also needed in write mode to avoid new
3179 * anon_vmas to be associated with existing vmas.
3181 * A single task can't take more than one mm_take_all_locks() in a row
3182 * or it would deadlock.
3184 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3185 * mapping->flags avoid to take the same lock twice, if more than one
3186 * vma in this mm is backed by the same anon_vma or address_space.
3188 * We can take all the locks in random order because the VM code
3189 * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never
3190 * takes more than one of them in a row. Secondly we're protected
3191 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
3193 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3194 * that may have to take thousand of locks.
3196 * mm_take_all_locks() can fail if it's interrupted by signals.
3198 int mm_take_all_locks(struct mm_struct
*mm
)
3200 struct vm_area_struct
*vma
;
3201 struct anon_vma_chain
*avc
;
3203 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3205 mutex_lock(&mm_all_locks_mutex
);
3207 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3208 if (signal_pending(current
))
3210 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3211 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3214 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3215 if (signal_pending(current
))
3218 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3219 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3225 mm_drop_all_locks(mm
);
3229 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3231 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_node
)) {
3233 * The LSB of head.next can't change to 0 from under
3234 * us because we hold the mm_all_locks_mutex.
3236 * We must however clear the bitflag before unlocking
3237 * the vma so the users using the anon_vma->rb_root will
3238 * never see our bitflag.
3240 * No need of atomic instructions here, head.next
3241 * can't change from under us until we release the
3242 * anon_vma->root->rwsem.
3244 if (!__test_and_clear_bit(0, (unsigned long *)
3245 &anon_vma
->root
->rb_root
.rb_node
))
3247 anon_vma_unlock_write(anon_vma
);
3251 static void vm_unlock_mapping(struct address_space
*mapping
)
3253 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3255 * AS_MM_ALL_LOCKS can't change to 0 from under us
3256 * because we hold the mm_all_locks_mutex.
3258 i_mmap_unlock_write(mapping
);
3259 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3266 * The mmap_sem cannot be released by the caller until
3267 * mm_drop_all_locks() returns.
3269 void mm_drop_all_locks(struct mm_struct
*mm
)
3271 struct vm_area_struct
*vma
;
3272 struct anon_vma_chain
*avc
;
3274 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3275 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3277 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3279 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3280 vm_unlock_anon_vma(avc
->anon_vma
);
3281 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3282 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3285 mutex_unlock(&mm_all_locks_mutex
);
3289 * initialise the VMA slab
3291 void __init
mmap_init(void)
3295 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3300 * Initialise sysctl_user_reserve_kbytes.
3302 * This is intended to prevent a user from starting a single memory hogging
3303 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3306 * The default value is min(3% of free memory, 128MB)
3307 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3309 static int init_user_reserve(void)
3311 unsigned long free_kbytes
;
3313 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3315 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3318 subsys_initcall(init_user_reserve
);
3321 * Initialise sysctl_admin_reserve_kbytes.
3323 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3324 * to log in and kill a memory hogging process.
3326 * Systems with more than 256MB will reserve 8MB, enough to recover
3327 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3328 * only reserve 3% of free pages by default.
3330 static int init_admin_reserve(void)
3332 unsigned long free_kbytes
;
3334 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3336 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3339 subsys_initcall(init_admin_reserve
);
3342 * Reinititalise user and admin reserves if memory is added or removed.
3344 * The default user reserve max is 128MB, and the default max for the
3345 * admin reserve is 8MB. These are usually, but not always, enough to
3346 * enable recovery from a memory hogging process using login/sshd, a shell,
3347 * and tools like top. It may make sense to increase or even disable the
3348 * reserve depending on the existence of swap or variations in the recovery
3349 * tools. So, the admin may have changed them.
3351 * If memory is added and the reserves have been eliminated or increased above
3352 * the default max, then we'll trust the admin.
3354 * If memory is removed and there isn't enough free memory, then we
3355 * need to reset the reserves.
3357 * Otherwise keep the reserve set by the admin.
3359 static int reserve_mem_notifier(struct notifier_block
*nb
,
3360 unsigned long action
, void *data
)
3362 unsigned long tmp
, free_kbytes
;
3366 /* Default max is 128MB. Leave alone if modified by operator. */
3367 tmp
= sysctl_user_reserve_kbytes
;
3368 if (0 < tmp
&& tmp
< (1UL << 17))
3369 init_user_reserve();
3371 /* Default max is 8MB. Leave alone if modified by operator. */
3372 tmp
= sysctl_admin_reserve_kbytes
;
3373 if (0 < tmp
&& tmp
< (1UL << 13))
3374 init_admin_reserve();
3378 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3380 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3381 init_user_reserve();
3382 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3383 sysctl_user_reserve_kbytes
);
3386 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3387 init_admin_reserve();
3388 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3389 sysctl_admin_reserve_kbytes
);
3398 static struct notifier_block reserve_mem_nb
= {
3399 .notifier_call
= reserve_mem_notifier
,
3402 static int __meminit
init_reserve_notifier(void)
3404 if (register_hotmemory_notifier(&reserve_mem_nb
))
3405 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3409 subsys_initcall(init_reserve_notifier
);