1 // SPDX-License-Identifier: GPL-2.0-only
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/backing-dev.h>
16 #include <linux/vmacache.h>
17 #include <linux/shm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/syscalls.h>
22 #include <linux/capability.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/hugetlb.h>
29 #include <linux/shmem_fs.h>
30 #include <linux/profile.h>
31 #include <linux/export.h>
32 #include <linux/mount.h>
33 #include <linux/mempolicy.h>
34 #include <linux/rmap.h>
35 #include <linux/mmu_notifier.h>
36 #include <linux/mmdebug.h>
37 #include <linux/perf_event.h>
38 #include <linux/audit.h>
39 #include <linux/khugepaged.h>
40 #include <linux/uprobes.h>
41 #include <linux/rbtree_augmented.h>
42 #include <linux/notifier.h>
43 #include <linux/memory.h>
44 #include <linux/printk.h>
45 #include <linux/userfaultfd_k.h>
46 #include <linux/moduleparam.h>
47 #include <linux/pkeys.h>
48 #include <linux/oom.h>
49 #include <linux/sched/mm.h>
51 #include <linux/uaccess.h>
52 #include <asm/cacheflush.h>
54 #include <asm/mmu_context.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/mmap.h>
61 #ifndef arch_mmap_check
62 #define arch_mmap_check(addr, len, flags) (0)
65 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
66 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
67 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
68 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
70 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
71 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
72 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
73 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
76 static bool ignore_rlimit_data
;
77 core_param(ignore_rlimit_data
, ignore_rlimit_data
, bool, 0644);
79 static void unmap_region(struct mm_struct
*mm
,
80 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
81 unsigned long start
, unsigned long end
);
83 /* description of effects of mapping type and prot in current implementation.
84 * this is due to the limited x86 page protection hardware. The expected
85 * behavior is in parens:
88 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
89 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
90 * w: (no) no w: (no) no w: (yes) yes w: (no) no
91 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
94 * w: (no) no w: (no) no w: (copy) copy w: (no) no
95 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
97 pgprot_t protection_map
[16] __ro_after_init
= {
98 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
99 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
102 #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
103 static inline pgprot_t
arch_filter_pgprot(pgprot_t prot
)
109 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
111 pgprot_t ret
= __pgprot(pgprot_val(protection_map
[vm_flags
&
112 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
113 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
115 return arch_filter_pgprot(ret
);
117 EXPORT_SYMBOL(vm_get_page_prot
);
119 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
121 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
124 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
125 void vma_set_page_prot(struct vm_area_struct
*vma
)
127 unsigned long vm_flags
= vma
->vm_flags
;
128 pgprot_t vm_page_prot
;
130 vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
131 if (vma_wants_writenotify(vma
, vm_page_prot
)) {
132 vm_flags
&= ~VM_SHARED
;
133 vm_page_prot
= vm_pgprot_modify(vm_page_prot
, vm_flags
);
135 /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
136 WRITE_ONCE(vma
->vm_page_prot
, vm_page_prot
);
140 * Requires inode->i_mapping->i_mmap_rwsem
142 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
143 struct file
*file
, struct address_space
*mapping
)
145 if (vma
->vm_flags
& VM_DENYWRITE
)
146 allow_write_access(file
);
147 if (vma
->vm_flags
& VM_SHARED
)
148 mapping_unmap_writable(mapping
);
150 flush_dcache_mmap_lock(mapping
);
151 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
152 flush_dcache_mmap_unlock(mapping
);
156 * Unlink a file-based vm structure from its interval tree, to hide
157 * vma from rmap and vmtruncate before freeing its page tables.
159 void unlink_file_vma(struct vm_area_struct
*vma
)
161 struct file
*file
= vma
->vm_file
;
164 struct address_space
*mapping
= file
->f_mapping
;
165 i_mmap_lock_write(mapping
);
166 __remove_shared_vm_struct(vma
, file
, mapping
);
167 i_mmap_unlock_write(mapping
);
172 * Close a vm structure and free it, returning the next.
174 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
176 struct vm_area_struct
*next
= vma
->vm_next
;
179 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
180 vma
->vm_ops
->close(vma
);
183 mpol_put(vma_policy(vma
));
188 static int do_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
,
189 struct list_head
*uf
);
190 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
192 unsigned long retval
;
193 unsigned long newbrk
, oldbrk
, origbrk
;
194 struct mm_struct
*mm
= current
->mm
;
195 struct vm_area_struct
*next
;
196 unsigned long min_brk
;
198 bool downgraded
= false;
201 if (mmap_write_lock_killable(mm
))
206 #ifdef CONFIG_COMPAT_BRK
208 * CONFIG_COMPAT_BRK can still be overridden by setting
209 * randomize_va_space to 2, which will still cause mm->start_brk
210 * to be arbitrarily shifted
212 if (current
->brk_randomized
)
213 min_brk
= mm
->start_brk
;
215 min_brk
= mm
->end_data
;
217 min_brk
= mm
->start_brk
;
223 * Check against rlimit here. If this check is done later after the test
224 * of oldbrk with newbrk then it can escape the test and let the data
225 * segment grow beyond its set limit the in case where the limit is
226 * not page aligned -Ram Gupta
228 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
229 mm
->end_data
, mm
->start_data
))
232 newbrk
= PAGE_ALIGN(brk
);
233 oldbrk
= PAGE_ALIGN(mm
->brk
);
234 if (oldbrk
== newbrk
) {
240 * Always allow shrinking brk.
241 * __do_munmap() may downgrade mmap_lock to read.
243 if (brk
<= mm
->brk
) {
247 * mm->brk must to be protected by write mmap_lock so update it
248 * before downgrading mmap_lock. When __do_munmap() fails,
249 * mm->brk will be restored from origbrk.
252 ret
= __do_munmap(mm
, newbrk
, oldbrk
-newbrk
, &uf
, true);
256 } else if (ret
== 1) {
262 /* Check against existing mmap mappings. */
263 next
= find_vma(mm
, oldbrk
);
264 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
267 /* Ok, looks good - let it rip. */
268 if (do_brk_flags(oldbrk
, newbrk
-oldbrk
, 0, &uf
) < 0)
273 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
275 mmap_read_unlock(mm
);
277 mmap_write_unlock(mm
);
278 userfaultfd_unmap_complete(mm
, &uf
);
280 mm_populate(oldbrk
, newbrk
- oldbrk
);
285 mmap_write_unlock(mm
);
289 static inline unsigned long vma_compute_gap(struct vm_area_struct
*vma
)
291 unsigned long gap
, prev_end
;
294 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
295 * allow two stack_guard_gaps between them here, and when choosing
296 * an unmapped area; whereas when expanding we only require one.
297 * That's a little inconsistent, but keeps the code here simpler.
299 gap
= vm_start_gap(vma
);
301 prev_end
= vm_end_gap(vma
->vm_prev
);
310 #ifdef CONFIG_DEBUG_VM_RB
311 static unsigned long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
313 unsigned long max
= vma_compute_gap(vma
), subtree_gap
;
314 if (vma
->vm_rb
.rb_left
) {
315 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
316 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
317 if (subtree_gap
> max
)
320 if (vma
->vm_rb
.rb_right
) {
321 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
322 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
323 if (subtree_gap
> max
)
329 static int browse_rb(struct mm_struct
*mm
)
331 struct rb_root
*root
= &mm
->mm_rb
;
332 int i
= 0, j
, bug
= 0;
333 struct rb_node
*nd
, *pn
= NULL
;
334 unsigned long prev
= 0, pend
= 0;
336 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
337 struct vm_area_struct
*vma
;
338 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
339 if (vma
->vm_start
< prev
) {
340 pr_emerg("vm_start %lx < prev %lx\n",
341 vma
->vm_start
, prev
);
344 if (vma
->vm_start
< pend
) {
345 pr_emerg("vm_start %lx < pend %lx\n",
346 vma
->vm_start
, pend
);
349 if (vma
->vm_start
> vma
->vm_end
) {
350 pr_emerg("vm_start %lx > vm_end %lx\n",
351 vma
->vm_start
, vma
->vm_end
);
354 spin_lock(&mm
->page_table_lock
);
355 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
356 pr_emerg("free gap %lx, correct %lx\n",
358 vma_compute_subtree_gap(vma
));
361 spin_unlock(&mm
->page_table_lock
);
364 prev
= vma
->vm_start
;
368 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
371 pr_emerg("backwards %d, forwards %d\n", j
, i
);
377 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
381 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
382 struct vm_area_struct
*vma
;
383 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
384 VM_BUG_ON_VMA(vma
!= ignore
&&
385 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
390 static void validate_mm(struct mm_struct
*mm
)
394 unsigned long highest_address
= 0;
395 struct vm_area_struct
*vma
= mm
->mmap
;
398 struct anon_vma
*anon_vma
= vma
->anon_vma
;
399 struct anon_vma_chain
*avc
;
402 anon_vma_lock_read(anon_vma
);
403 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
404 anon_vma_interval_tree_verify(avc
);
405 anon_vma_unlock_read(anon_vma
);
408 highest_address
= vm_end_gap(vma
);
412 if (i
!= mm
->map_count
) {
413 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
416 if (highest_address
!= mm
->highest_vm_end
) {
417 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
418 mm
->highest_vm_end
, highest_address
);
422 if (i
!= mm
->map_count
) {
424 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
427 VM_BUG_ON_MM(bug
, mm
);
430 #define validate_mm_rb(root, ignore) do { } while (0)
431 #define validate_mm(mm) do { } while (0)
434 RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks
,
435 struct vm_area_struct
, vm_rb
,
436 unsigned long, rb_subtree_gap
, vma_compute_gap
)
439 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
440 * vma->vm_prev->vm_end values changed, without modifying the vma's position
443 static void vma_gap_update(struct vm_area_struct
*vma
)
446 * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created
447 * a callback function that does exactly what we want.
449 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
452 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
453 struct rb_root
*root
)
455 /* All rb_subtree_gap values must be consistent prior to insertion */
456 validate_mm_rb(root
, NULL
);
458 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
461 static void __vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
464 * Note rb_erase_augmented is a fairly large inline function,
465 * so make sure we instantiate it only once with our desired
466 * augmented rbtree callbacks.
468 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
471 static __always_inline
void vma_rb_erase_ignore(struct vm_area_struct
*vma
,
472 struct rb_root
*root
,
473 struct vm_area_struct
*ignore
)
476 * All rb_subtree_gap values must be consistent prior to erase,
477 * with the possible exception of
479 * a. the "next" vma being erased if next->vm_start was reduced in
480 * __vma_adjust() -> __vma_unlink()
481 * b. the vma being erased in detach_vmas_to_be_unmapped() ->
484 validate_mm_rb(root
, ignore
);
486 __vma_rb_erase(vma
, root
);
489 static __always_inline
void vma_rb_erase(struct vm_area_struct
*vma
,
490 struct rb_root
*root
)
492 vma_rb_erase_ignore(vma
, root
, vma
);
496 * vma has some anon_vma assigned, and is already inserted on that
497 * anon_vma's interval trees.
499 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
500 * vma must be removed from the anon_vma's interval trees using
501 * anon_vma_interval_tree_pre_update_vma().
503 * After the update, the vma will be reinserted using
504 * anon_vma_interval_tree_post_update_vma().
506 * The entire update must be protected by exclusive mmap_lock and by
507 * the root anon_vma's mutex.
510 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
512 struct anon_vma_chain
*avc
;
514 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
515 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
519 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
521 struct anon_vma_chain
*avc
;
523 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
524 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
527 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
528 unsigned long end
, struct vm_area_struct
**pprev
,
529 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
531 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
533 __rb_link
= &mm
->mm_rb
.rb_node
;
534 rb_prev
= __rb_parent
= NULL
;
537 struct vm_area_struct
*vma_tmp
;
539 __rb_parent
= *__rb_link
;
540 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
542 if (vma_tmp
->vm_end
> addr
) {
543 /* Fail if an existing vma overlaps the area */
544 if (vma_tmp
->vm_start
< end
)
546 __rb_link
= &__rb_parent
->rb_left
;
548 rb_prev
= __rb_parent
;
549 __rb_link
= &__rb_parent
->rb_right
;
555 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
556 *rb_link
= __rb_link
;
557 *rb_parent
= __rb_parent
;
562 * vma_next() - Get the next VMA.
563 * @mm: The mm_struct.
564 * @vma: The current vma.
566 * If @vma is NULL, return the first vma in the mm.
568 * Returns: The next VMA after @vma.
570 static inline struct vm_area_struct
*vma_next(struct mm_struct
*mm
,
571 struct vm_area_struct
*vma
)
580 * munmap_vma_range() - munmap VMAs that overlap a range.
582 * @start: The start of the range.
583 * @len: The length of the range.
584 * @pprev: pointer to the pointer that will be set to previous vm_area_struct
585 * @rb_link: the rb_node
586 * @rb_parent: the parent rb_node
588 * Find all the vm_area_struct that overlap from @start to
589 * @end and munmap them. Set @pprev to the previous vm_area_struct.
591 * Returns: -ENOMEM on munmap failure or 0 on success.
594 munmap_vma_range(struct mm_struct
*mm
, unsigned long start
, unsigned long len
,
595 struct vm_area_struct
**pprev
, struct rb_node
***link
,
596 struct rb_node
**parent
, struct list_head
*uf
)
599 while (find_vma_links(mm
, start
, start
+ len
, pprev
, link
, parent
))
600 if (do_munmap(mm
, start
, len
, uf
))
605 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
606 unsigned long addr
, unsigned long end
)
608 unsigned long nr_pages
= 0;
609 struct vm_area_struct
*vma
;
611 /* Find first overlaping mapping */
612 vma
= find_vma_intersection(mm
, addr
, end
);
616 nr_pages
= (min(end
, vma
->vm_end
) -
617 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
619 /* Iterate over the rest of the overlaps */
620 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
621 unsigned long overlap_len
;
623 if (vma
->vm_start
> end
)
626 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
627 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
633 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
634 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
636 /* Update tracking information for the gap following the new vma. */
638 vma_gap_update(vma
->vm_next
);
640 mm
->highest_vm_end
= vm_end_gap(vma
);
643 * vma->vm_prev wasn't known when we followed the rbtree to find the
644 * correct insertion point for that vma. As a result, we could not
645 * update the vma vm_rb parents rb_subtree_gap values on the way down.
646 * So, we first insert the vma with a zero rb_subtree_gap value
647 * (to be consistent with what we did on the way down), and then
648 * immediately update the gap to the correct value. Finally we
649 * rebalance the rbtree after all augmented values have been set.
651 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
652 vma
->rb_subtree_gap
= 0;
654 vma_rb_insert(vma
, &mm
->mm_rb
);
657 static void __vma_link_file(struct vm_area_struct
*vma
)
663 struct address_space
*mapping
= file
->f_mapping
;
665 if (vma
->vm_flags
& VM_DENYWRITE
)
666 put_write_access(file_inode(file
));
667 if (vma
->vm_flags
& VM_SHARED
)
668 mapping_allow_writable(mapping
);
670 flush_dcache_mmap_lock(mapping
);
671 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
672 flush_dcache_mmap_unlock(mapping
);
677 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
678 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
679 struct rb_node
*rb_parent
)
681 __vma_link_list(mm
, vma
, prev
);
682 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
685 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
686 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
687 struct rb_node
*rb_parent
)
689 struct address_space
*mapping
= NULL
;
692 mapping
= vma
->vm_file
->f_mapping
;
693 i_mmap_lock_write(mapping
);
696 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
697 __vma_link_file(vma
);
700 i_mmap_unlock_write(mapping
);
707 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
708 * mm's list and rbtree. It has already been inserted into the interval tree.
710 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
712 struct vm_area_struct
*prev
;
713 struct rb_node
**rb_link
, *rb_parent
;
715 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
716 &prev
, &rb_link
, &rb_parent
))
718 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
722 static __always_inline
void __vma_unlink(struct mm_struct
*mm
,
723 struct vm_area_struct
*vma
,
724 struct vm_area_struct
*ignore
)
726 vma_rb_erase_ignore(vma
, &mm
->mm_rb
, ignore
);
727 __vma_unlink_list(mm
, vma
);
729 vmacache_invalidate(mm
);
733 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
734 * is already present in an i_mmap tree without adjusting the tree.
735 * The following helper function should be used when such adjustments
736 * are necessary. The "insert" vma (if any) is to be inserted
737 * before we drop the necessary locks.
739 int __vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
740 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
,
741 struct vm_area_struct
*expand
)
743 struct mm_struct
*mm
= vma
->vm_mm
;
744 struct vm_area_struct
*next
= vma
->vm_next
, *orig_vma
= vma
;
745 struct address_space
*mapping
= NULL
;
746 struct rb_root_cached
*root
= NULL
;
747 struct anon_vma
*anon_vma
= NULL
;
748 struct file
*file
= vma
->vm_file
;
749 bool start_changed
= false, end_changed
= false;
750 long adjust_next
= 0;
753 if (next
&& !insert
) {
754 struct vm_area_struct
*exporter
= NULL
, *importer
= NULL
;
756 if (end
>= next
->vm_end
) {
758 * vma expands, overlapping all the next, and
759 * perhaps the one after too (mprotect case 6).
760 * The only other cases that gets here are
761 * case 1, case 7 and case 8.
763 if (next
== expand
) {
765 * The only case where we don't expand "vma"
766 * and we expand "next" instead is case 8.
768 VM_WARN_ON(end
!= next
->vm_end
);
770 * remove_next == 3 means we're
771 * removing "vma" and that to do so we
772 * swapped "vma" and "next".
775 VM_WARN_ON(file
!= next
->vm_file
);
778 VM_WARN_ON(expand
!= vma
);
780 * case 1, 6, 7, remove_next == 2 is case 6,
781 * remove_next == 1 is case 1 or 7.
783 remove_next
= 1 + (end
> next
->vm_end
);
784 VM_WARN_ON(remove_next
== 2 &&
785 end
!= next
->vm_next
->vm_end
);
786 /* trim end to next, for case 6 first pass */
794 * If next doesn't have anon_vma, import from vma after
795 * next, if the vma overlaps with it.
797 if (remove_next
== 2 && !next
->anon_vma
)
798 exporter
= next
->vm_next
;
800 } else if (end
> next
->vm_start
) {
802 * vma expands, overlapping part of the next:
803 * mprotect case 5 shifting the boundary up.
805 adjust_next
= (end
- next
->vm_start
);
808 VM_WARN_ON(expand
!= importer
);
809 } else if (end
< vma
->vm_end
) {
811 * vma shrinks, and !insert tells it's not
812 * split_vma inserting another: so it must be
813 * mprotect case 4 shifting the boundary down.
815 adjust_next
= -(vma
->vm_end
- end
);
818 VM_WARN_ON(expand
!= importer
);
822 * Easily overlooked: when mprotect shifts the boundary,
823 * make sure the expanding vma has anon_vma set if the
824 * shrinking vma had, to cover any anon pages imported.
826 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
829 importer
->anon_vma
= exporter
->anon_vma
;
830 error
= anon_vma_clone(importer
, exporter
);
836 vma_adjust_trans_huge(orig_vma
, start
, end
, adjust_next
);
839 mapping
= file
->f_mapping
;
840 root
= &mapping
->i_mmap
;
841 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
844 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
846 i_mmap_lock_write(mapping
);
849 * Put into interval tree now, so instantiated pages
850 * are visible to arm/parisc __flush_dcache_page
851 * throughout; but we cannot insert into address
852 * space until vma start or end is updated.
854 __vma_link_file(insert
);
858 anon_vma
= vma
->anon_vma
;
859 if (!anon_vma
&& adjust_next
)
860 anon_vma
= next
->anon_vma
;
862 VM_WARN_ON(adjust_next
&& next
->anon_vma
&&
863 anon_vma
!= next
->anon_vma
);
864 anon_vma_lock_write(anon_vma
);
865 anon_vma_interval_tree_pre_update_vma(vma
);
867 anon_vma_interval_tree_pre_update_vma(next
);
871 flush_dcache_mmap_lock(mapping
);
872 vma_interval_tree_remove(vma
, root
);
874 vma_interval_tree_remove(next
, root
);
877 if (start
!= vma
->vm_start
) {
878 vma
->vm_start
= start
;
879 start_changed
= true;
881 if (end
!= vma
->vm_end
) {
885 vma
->vm_pgoff
= pgoff
;
887 next
->vm_start
+= adjust_next
;
888 next
->vm_pgoff
+= adjust_next
>> PAGE_SHIFT
;
893 vma_interval_tree_insert(next
, root
);
894 vma_interval_tree_insert(vma
, root
);
895 flush_dcache_mmap_unlock(mapping
);
900 * vma_merge has merged next into vma, and needs
901 * us to remove next before dropping the locks.
903 if (remove_next
!= 3)
904 __vma_unlink(mm
, next
, next
);
907 * vma is not before next if they've been
910 * pre-swap() next->vm_start was reduced so
911 * tell validate_mm_rb to ignore pre-swap()
912 * "next" (which is stored in post-swap()
915 __vma_unlink(mm
, next
, vma
);
917 __remove_shared_vm_struct(next
, file
, mapping
);
920 * split_vma has split insert from vma, and needs
921 * us to insert it before dropping the locks
922 * (it may either follow vma or precede it).
924 __insert_vm_struct(mm
, insert
);
930 mm
->highest_vm_end
= vm_end_gap(vma
);
931 else if (!adjust_next
)
932 vma_gap_update(next
);
937 anon_vma_interval_tree_post_update_vma(vma
);
939 anon_vma_interval_tree_post_update_vma(next
);
940 anon_vma_unlock_write(anon_vma
);
944 i_mmap_unlock_write(mapping
);
953 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
957 anon_vma_merge(vma
, next
);
959 mpol_put(vma_policy(next
));
962 * In mprotect's case 6 (see comments on vma_merge),
963 * we must remove another next too. It would clutter
964 * up the code too much to do both in one go.
966 if (remove_next
!= 3) {
968 * If "next" was removed and vma->vm_end was
969 * expanded (up) over it, in turn
970 * "next->vm_prev->vm_end" changed and the
971 * "vma->vm_next" gap must be updated.
976 * For the scope of the comment "next" and
977 * "vma" considered pre-swap(): if "vma" was
978 * removed, next->vm_start was expanded (down)
979 * over it and the "next" gap must be updated.
980 * Because of the swap() the post-swap() "vma"
981 * actually points to pre-swap() "next"
982 * (post-swap() "next" as opposed is now a
987 if (remove_next
== 2) {
993 vma_gap_update(next
);
996 * If remove_next == 2 we obviously can't
999 * If remove_next == 3 we can't reach this
1000 * path because pre-swap() next is always not
1001 * NULL. pre-swap() "next" is not being
1002 * removed and its next->vm_end is not altered
1003 * (and furthermore "end" already matches
1004 * next->vm_end in remove_next == 3).
1006 * We reach this only in the remove_next == 1
1007 * case if the "next" vma that was removed was
1008 * the highest vma of the mm. However in such
1009 * case next->vm_end == "end" and the extended
1010 * "vma" has vma->vm_end == next->vm_end so
1011 * mm->highest_vm_end doesn't need any update
1012 * in remove_next == 1 case.
1014 VM_WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
1018 uprobe_mmap(insert
);
1026 * If the vma has a ->close operation then the driver probably needs to release
1027 * per-vma resources, so we don't attempt to merge those.
1029 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
1030 struct file
*file
, unsigned long vm_flags
,
1031 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1034 * VM_SOFTDIRTY should not prevent from VMA merging, if we
1035 * match the flags but dirty bit -- the caller should mark
1036 * merged VMA as dirty. If dirty bit won't be excluded from
1037 * comparison, we increase pressure on the memory system forcing
1038 * the kernel to generate new VMAs when old one could be
1041 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
1043 if (vma
->vm_file
!= file
)
1045 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
1047 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
1052 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
1053 struct anon_vma
*anon_vma2
,
1054 struct vm_area_struct
*vma
)
1057 * The list_is_singular() test is to avoid merging VMA cloned from
1058 * parents. This can improve scalability caused by anon_vma lock.
1060 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
1061 list_is_singular(&vma
->anon_vma_chain
)))
1063 return anon_vma1
== anon_vma2
;
1067 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1068 * in front of (at a lower virtual address and file offset than) the vma.
1070 * We cannot merge two vmas if they have differently assigned (non-NULL)
1071 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1073 * We don't check here for the merged mmap wrapping around the end of pagecache
1074 * indices (16TB on ia32) because do_mmap() does not permit mmap's which
1075 * wrap, nor mmaps which cover the final page at index -1UL.
1078 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1079 struct anon_vma
*anon_vma
, struct file
*file
,
1081 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1083 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1084 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1085 if (vma
->vm_pgoff
== vm_pgoff
)
1092 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1093 * beyond (at a higher virtual address and file offset than) the vma.
1095 * We cannot merge two vmas if they have differently assigned (non-NULL)
1096 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1099 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1100 struct anon_vma
*anon_vma
, struct file
*file
,
1102 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1104 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1105 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1107 vm_pglen
= vma_pages(vma
);
1108 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1115 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1116 * whether that can be merged with its predecessor or its successor.
1117 * Or both (it neatly fills a hole).
1119 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1120 * certain not to be mapped by the time vma_merge is called; but when
1121 * called for mprotect, it is certain to be already mapped (either at
1122 * an offset within prev, or at the start of next), and the flags of
1123 * this area are about to be changed to vm_flags - and the no-change
1124 * case has already been eliminated.
1126 * The following mprotect cases have to be considered, where AAAA is
1127 * the area passed down from mprotect_fixup, never extending beyond one
1128 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1131 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN
1132 * cannot merge might become might become
1133 * PPNNNNNNNNNN PPPPPPPPPPNN
1134 * mmap, brk or case 4 below case 5 below
1137 * PPPP NNNN PPPPNNNNXXXX
1138 * might become might become
1139 * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or
1140 * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or
1141 * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8
1143 * It is important for case 8 that the vma NNNN overlapping the
1144 * region AAAA is never going to extended over XXXX. Instead XXXX must
1145 * be extended in region AAAA and NNNN must be removed. This way in
1146 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1147 * rmap_locks, the properties of the merged vma will be already
1148 * correct for the whole merged range. Some of those properties like
1149 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1150 * be correct for the whole merged range immediately after the
1151 * rmap_locks are released. Otherwise if XXXX would be removed and
1152 * NNNN would be extended over the XXXX range, remove_migration_ptes
1153 * or other rmap walkers (if working on addresses beyond the "end"
1154 * parameter) may establish ptes with the wrong permissions of NNNN
1155 * instead of the right permissions of XXXX.
1157 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1158 struct vm_area_struct
*prev
, unsigned long addr
,
1159 unsigned long end
, unsigned long vm_flags
,
1160 struct anon_vma
*anon_vma
, struct file
*file
,
1161 pgoff_t pgoff
, struct mempolicy
*policy
,
1162 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1164 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1165 struct vm_area_struct
*area
, *next
;
1169 * We later require that vma->vm_flags == vm_flags,
1170 * so this tests vma->vm_flags & VM_SPECIAL, too.
1172 if (vm_flags
& VM_SPECIAL
)
1175 next
= vma_next(mm
, prev
);
1177 if (area
&& area
->vm_end
== end
) /* cases 6, 7, 8 */
1178 next
= next
->vm_next
;
1180 /* verify some invariant that must be enforced by the caller */
1181 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
1182 VM_WARN_ON(area
&& end
> area
->vm_end
);
1183 VM_WARN_ON(addr
>= end
);
1186 * Can it merge with the predecessor?
1188 if (prev
&& prev
->vm_end
== addr
&&
1189 mpol_equal(vma_policy(prev
), policy
) &&
1190 can_vma_merge_after(prev
, vm_flags
,
1191 anon_vma
, file
, pgoff
,
1192 vm_userfaultfd_ctx
)) {
1194 * OK, it can. Can we now merge in the successor as well?
1196 if (next
&& end
== next
->vm_start
&&
1197 mpol_equal(policy
, vma_policy(next
)) &&
1198 can_vma_merge_before(next
, vm_flags
,
1201 vm_userfaultfd_ctx
) &&
1202 is_mergeable_anon_vma(prev
->anon_vma
,
1203 next
->anon_vma
, NULL
)) {
1205 err
= __vma_adjust(prev
, prev
->vm_start
,
1206 next
->vm_end
, prev
->vm_pgoff
, NULL
,
1208 } else /* cases 2, 5, 7 */
1209 err
= __vma_adjust(prev
, prev
->vm_start
,
1210 end
, prev
->vm_pgoff
, NULL
, prev
);
1213 khugepaged_enter_vma_merge(prev
, vm_flags
);
1218 * Can this new request be merged in front of next?
1220 if (next
&& end
== next
->vm_start
&&
1221 mpol_equal(policy
, vma_policy(next
)) &&
1222 can_vma_merge_before(next
, vm_flags
,
1223 anon_vma
, file
, pgoff
+pglen
,
1224 vm_userfaultfd_ctx
)) {
1225 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1226 err
= __vma_adjust(prev
, prev
->vm_start
,
1227 addr
, prev
->vm_pgoff
, NULL
, next
);
1228 else { /* cases 3, 8 */
1229 err
= __vma_adjust(area
, addr
, next
->vm_end
,
1230 next
->vm_pgoff
- pglen
, NULL
, next
);
1232 * In case 3 area is already equal to next and
1233 * this is a noop, but in case 8 "area" has
1234 * been removed and next was expanded over it.
1240 khugepaged_enter_vma_merge(area
, vm_flags
);
1248 * Rough compatibility check to quickly see if it's even worth looking
1249 * at sharing an anon_vma.
1251 * They need to have the same vm_file, and the flags can only differ
1252 * in things that mprotect may change.
1254 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1255 * we can merge the two vma's. For example, we refuse to merge a vma if
1256 * there is a vm_ops->close() function, because that indicates that the
1257 * driver is doing some kind of reference counting. But that doesn't
1258 * really matter for the anon_vma sharing case.
1260 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1262 return a
->vm_end
== b
->vm_start
&&
1263 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1264 a
->vm_file
== b
->vm_file
&&
1265 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_ACCESS_FLAGS
| VM_SOFTDIRTY
)) &&
1266 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1270 * Do some basic sanity checking to see if we can re-use the anon_vma
1271 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1272 * the same as 'old', the other will be the new one that is trying
1273 * to share the anon_vma.
1275 * NOTE! This runs with mm_sem held for reading, so it is possible that
1276 * the anon_vma of 'old' is concurrently in the process of being set up
1277 * by another page fault trying to merge _that_. But that's ok: if it
1278 * is being set up, that automatically means that it will be a singleton
1279 * acceptable for merging, so we can do all of this optimistically. But
1280 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1282 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1283 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1284 * is to return an anon_vma that is "complex" due to having gone through
1287 * We also make sure that the two vma's are compatible (adjacent,
1288 * and with the same memory policies). That's all stable, even with just
1289 * a read lock on the mm_sem.
1291 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1293 if (anon_vma_compatible(a
, b
)) {
1294 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1296 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1303 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1304 * neighbouring vmas for a suitable anon_vma, before it goes off
1305 * to allocate a new anon_vma. It checks because a repetitive
1306 * sequence of mprotects and faults may otherwise lead to distinct
1307 * anon_vmas being allocated, preventing vma merge in subsequent
1310 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1312 struct anon_vma
*anon_vma
= NULL
;
1314 /* Try next first. */
1316 anon_vma
= reusable_anon_vma(vma
->vm_next
, vma
, vma
->vm_next
);
1321 /* Try prev next. */
1323 anon_vma
= reusable_anon_vma(vma
->vm_prev
, vma
->vm_prev
, vma
);
1326 * We might reach here with anon_vma == NULL if we can't find
1327 * any reusable anon_vma.
1328 * There's no absolute need to look only at touching neighbours:
1329 * we could search further afield for "compatible" anon_vmas.
1330 * But it would probably just be a waste of time searching,
1331 * or lead to too many vmas hanging off the same anon_vma.
1332 * We're trying to allow mprotect remerging later on,
1333 * not trying to minimize memory used for anon_vmas.
1339 * If a hint addr is less than mmap_min_addr change hint to be as
1340 * low as possible but still greater than mmap_min_addr
1342 static inline unsigned long round_hint_to_min(unsigned long hint
)
1345 if (((void *)hint
!= NULL
) &&
1346 (hint
< mmap_min_addr
))
1347 return PAGE_ALIGN(mmap_min_addr
);
1351 static inline int mlock_future_check(struct mm_struct
*mm
,
1352 unsigned long flags
,
1355 unsigned long locked
, lock_limit
;
1357 /* mlock MCL_FUTURE? */
1358 if (flags
& VM_LOCKED
) {
1359 locked
= len
>> PAGE_SHIFT
;
1360 locked
+= mm
->locked_vm
;
1361 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1362 lock_limit
>>= PAGE_SHIFT
;
1363 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1369 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1371 if (S_ISREG(inode
->i_mode
))
1372 return MAX_LFS_FILESIZE
;
1374 if (S_ISBLK(inode
->i_mode
))
1375 return MAX_LFS_FILESIZE
;
1377 if (S_ISSOCK(inode
->i_mode
))
1378 return MAX_LFS_FILESIZE
;
1380 /* Special "we do even unsigned file positions" case */
1381 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1384 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1388 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1389 unsigned long pgoff
, unsigned long len
)
1391 u64 maxsize
= file_mmap_size_max(file
, inode
);
1393 if (maxsize
&& len
> maxsize
)
1396 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1402 * The caller must write-lock current->mm->mmap_lock.
1404 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1405 unsigned long len
, unsigned long prot
,
1406 unsigned long flags
, unsigned long pgoff
,
1407 unsigned long *populate
, struct list_head
*uf
)
1409 struct mm_struct
*mm
= current
->mm
;
1410 vm_flags_t vm_flags
;
1419 * Does the application expect PROT_READ to imply PROT_EXEC?
1421 * (the exception is when the underlying filesystem is noexec
1422 * mounted, in which case we dont add PROT_EXEC.)
1424 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1425 if (!(file
&& path_noexec(&file
->f_path
)))
1428 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1429 if (flags
& MAP_FIXED_NOREPLACE
)
1432 if (!(flags
& MAP_FIXED
))
1433 addr
= round_hint_to_min(addr
);
1435 /* Careful about overflows.. */
1436 len
= PAGE_ALIGN(len
);
1440 /* offset overflow? */
1441 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1444 /* Too many mappings? */
1445 if (mm
->map_count
> sysctl_max_map_count
)
1448 /* Obtain the address to map to. we verify (or select) it and ensure
1449 * that it represents a valid section of the address space.
1451 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1452 if (IS_ERR_VALUE(addr
))
1455 if (flags
& MAP_FIXED_NOREPLACE
) {
1456 struct vm_area_struct
*vma
= find_vma(mm
, addr
);
1458 if (vma
&& vma
->vm_start
< addr
+ len
)
1462 if (prot
== PROT_EXEC
) {
1463 pkey
= execute_only_pkey(mm
);
1468 /* Do simple checking here so the lower-level routines won't have
1469 * to. we assume access permissions have been handled by the open
1470 * of the memory object, so we don't do any here.
1472 vm_flags
= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1473 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1475 if (flags
& MAP_LOCKED
)
1476 if (!can_do_mlock())
1479 if (mlock_future_check(mm
, vm_flags
, len
))
1483 struct inode
*inode
= file_inode(file
);
1484 unsigned long flags_mask
;
1486 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1489 flags_mask
= LEGACY_MAP_MASK
| file
->f_op
->mmap_supported_flags
;
1491 switch (flags
& MAP_TYPE
) {
1494 * Force use of MAP_SHARED_VALIDATE with non-legacy
1495 * flags. E.g. MAP_SYNC is dangerous to use with
1496 * MAP_SHARED as you don't know which consistency model
1497 * you will get. We silently ignore unsupported flags
1498 * with MAP_SHARED to preserve backward compatibility.
1500 flags
&= LEGACY_MAP_MASK
;
1502 case MAP_SHARED_VALIDATE
:
1503 if (flags
& ~flags_mask
)
1505 if (prot
& PROT_WRITE
) {
1506 if (!(file
->f_mode
& FMODE_WRITE
))
1508 if (IS_SWAPFILE(file
->f_mapping
->host
))
1513 * Make sure we don't allow writing to an append-only
1516 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1520 * Make sure there are no mandatory locks on the file.
1522 if (locks_verify_locked(file
))
1525 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1526 if (!(file
->f_mode
& FMODE_WRITE
))
1527 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1530 if (!(file
->f_mode
& FMODE_READ
))
1532 if (path_noexec(&file
->f_path
)) {
1533 if (vm_flags
& VM_EXEC
)
1535 vm_flags
&= ~VM_MAYEXEC
;
1538 if (!file
->f_op
->mmap
)
1540 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1548 switch (flags
& MAP_TYPE
) {
1550 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1556 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1560 * Set pgoff according to addr for anon_vma.
1562 pgoff
= addr
>> PAGE_SHIFT
;
1570 * Set 'VM_NORESERVE' if we should not account for the
1571 * memory use of this mapping.
1573 if (flags
& MAP_NORESERVE
) {
1574 /* We honor MAP_NORESERVE if allowed to overcommit */
1575 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1576 vm_flags
|= VM_NORESERVE
;
1578 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1579 if (file
&& is_file_hugepages(file
))
1580 vm_flags
|= VM_NORESERVE
;
1583 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1584 if (!IS_ERR_VALUE(addr
) &&
1585 ((vm_flags
& VM_LOCKED
) ||
1586 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1591 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1592 unsigned long prot
, unsigned long flags
,
1593 unsigned long fd
, unsigned long pgoff
)
1595 struct file
*file
= NULL
;
1596 unsigned long retval
;
1598 if (!(flags
& MAP_ANONYMOUS
)) {
1599 audit_mmap_fd(fd
, flags
);
1603 if (is_file_hugepages(file
)) {
1604 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1605 } else if (unlikely(flags
& MAP_HUGETLB
)) {
1609 } else if (flags
& MAP_HUGETLB
) {
1610 struct user_struct
*user
= NULL
;
1613 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1617 len
= ALIGN(len
, huge_page_size(hs
));
1619 * VM_NORESERVE is used because the reservations will be
1620 * taken when vm_ops->mmap() is called
1621 * A dummy user value is used because we are not locking
1622 * memory so no accounting is necessary
1624 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1626 &user
, HUGETLB_ANONHUGE_INODE
,
1627 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1629 return PTR_ERR(file
);
1632 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1634 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1641 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1642 unsigned long, prot
, unsigned long, flags
,
1643 unsigned long, fd
, unsigned long, pgoff
)
1645 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1648 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1649 struct mmap_arg_struct
{
1653 unsigned long flags
;
1655 unsigned long offset
;
1658 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1660 struct mmap_arg_struct a
;
1662 if (copy_from_user(&a
, arg
, sizeof(a
)))
1664 if (offset_in_page(a
.offset
))
1667 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1668 a
.offset
>> PAGE_SHIFT
);
1670 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1673 * Some shared mappings will want the pages marked read-only
1674 * to track write events. If so, we'll downgrade vm_page_prot
1675 * to the private version (using protection_map[] without the
1678 int vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1680 vm_flags_t vm_flags
= vma
->vm_flags
;
1681 const struct vm_operations_struct
*vm_ops
= vma
->vm_ops
;
1683 /* If it was private or non-writable, the write bit is already clear */
1684 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1687 /* The backer wishes to know when pages are first written to? */
1688 if (vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
))
1691 /* The open routine did something to the protections that pgprot_modify
1692 * won't preserve? */
1693 if (pgprot_val(vm_page_prot
) !=
1694 pgprot_val(vm_pgprot_modify(vm_page_prot
, vm_flags
)))
1697 /* Do we need to track softdirty? */
1698 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1701 /* Specialty mapping? */
1702 if (vm_flags
& VM_PFNMAP
)
1705 /* Can the mapping track the dirty pages? */
1706 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1707 mapping_can_writeback(vma
->vm_file
->f_mapping
);
1711 * We account for memory if it's a private writeable mapping,
1712 * not hugepages and VM_NORESERVE wasn't set.
1714 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1717 * hugetlb has its own accounting separate from the core VM
1718 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1720 if (file
&& is_file_hugepages(file
))
1723 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1726 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1727 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
1728 struct list_head
*uf
)
1730 struct mm_struct
*mm
= current
->mm
;
1731 struct vm_area_struct
*vma
, *prev
, *merge
;
1733 struct rb_node
**rb_link
, *rb_parent
;
1734 unsigned long charged
= 0;
1736 /* Check against address space limit. */
1737 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
1738 unsigned long nr_pages
;
1741 * MAP_FIXED may remove pages of mappings that intersects with
1742 * requested mapping. Account for the pages it would unmap.
1744 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1746 if (!may_expand_vm(mm
, vm_flags
,
1747 (len
>> PAGE_SHIFT
) - nr_pages
))
1751 /* Clear old maps, set up prev, rb_link, rb_parent, and uf */
1752 if (munmap_vma_range(mm
, addr
, len
, &prev
, &rb_link
, &rb_parent
, uf
))
1755 * Private writable mapping: check memory availability
1757 if (accountable_mapping(file
, vm_flags
)) {
1758 charged
= len
>> PAGE_SHIFT
;
1759 if (security_vm_enough_memory_mm(mm
, charged
))
1761 vm_flags
|= VM_ACCOUNT
;
1765 * Can we just expand an old mapping?
1767 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1768 NULL
, file
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
1773 * Determine the object being mapped and call the appropriate
1774 * specific mapper. the address has already been validated, but
1775 * not unmapped, but the maps are removed from the list.
1777 vma
= vm_area_alloc(mm
);
1783 vma
->vm_start
= addr
;
1784 vma
->vm_end
= addr
+ len
;
1785 vma
->vm_flags
= vm_flags
;
1786 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1787 vma
->vm_pgoff
= pgoff
;
1790 if (vm_flags
& VM_DENYWRITE
) {
1791 error
= deny_write_access(file
);
1795 if (vm_flags
& VM_SHARED
) {
1796 error
= mapping_map_writable(file
->f_mapping
);
1798 goto allow_write_and_free_vma
;
1801 /* ->mmap() can change vma->vm_file, but must guarantee that
1802 * vma_link() below can deny write-access if VM_DENYWRITE is set
1803 * and map writably if VM_SHARED is set. This usually means the
1804 * new file must not have been exposed to user-space, yet.
1806 vma
->vm_file
= get_file(file
);
1807 error
= call_mmap(file
, vma
);
1809 goto unmap_and_free_vma
;
1811 /* If vm_flags changed after call_mmap(), we should try merge vma again
1812 * as we may succeed this time.
1814 if (unlikely(vm_flags
!= vma
->vm_flags
&& prev
)) {
1815 merge
= vma_merge(mm
, prev
, vma
->vm_start
, vma
->vm_end
, vma
->vm_flags
,
1816 NULL
, vma
->vm_file
, vma
->vm_pgoff
, NULL
, NULL_VM_UFFD_CTX
);
1818 /* ->mmap() can change vma->vm_file and fput the original file. So
1819 * fput the vma->vm_file here or we would add an extra fput for file
1820 * and cause general protection fault ultimately.
1825 /* Update vm_flags and possible addr to pick up the change. We don't
1826 * warn here if addr changed as the vma is not linked by vma_link().
1828 addr
= vma
->vm_start
;
1829 vm_flags
= vma
->vm_flags
;
1830 goto unmap_writable
;
1834 /* Can addr have changed??
1836 * Answer: Yes, several device drivers can do it in their
1837 * f_op->mmap method. -DaveM
1838 * Bug: If addr is changed, prev, rb_link, rb_parent should
1839 * be updated for vma_link()
1841 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1843 addr
= vma
->vm_start
;
1844 vm_flags
= vma
->vm_flags
;
1845 } else if (vm_flags
& VM_SHARED
) {
1846 error
= shmem_zero_setup(vma
);
1850 vma_set_anonymous(vma
);
1853 /* Allow architectures to sanity-check the vm_flags */
1854 if (!arch_validate_flags(vma
->vm_flags
)) {
1857 goto unmap_and_free_vma
;
1862 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1863 /* Once vma denies write, undo our temporary denial count */
1866 if (vm_flags
& VM_SHARED
)
1867 mapping_unmap_writable(file
->f_mapping
);
1868 if (vm_flags
& VM_DENYWRITE
)
1869 allow_write_access(file
);
1871 file
= vma
->vm_file
;
1873 perf_event_mmap(vma
);
1875 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
1876 if (vm_flags
& VM_LOCKED
) {
1877 if ((vm_flags
& VM_SPECIAL
) || vma_is_dax(vma
) ||
1878 is_vm_hugetlb_page(vma
) ||
1879 vma
== get_gate_vma(current
->mm
))
1880 vma
->vm_flags
&= VM_LOCKED_CLEAR_MASK
;
1882 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1889 * New (or expanded) vma always get soft dirty status.
1890 * Otherwise user-space soft-dirty page tracker won't
1891 * be able to distinguish situation when vma area unmapped,
1892 * then new mapped in-place (which must be aimed as
1893 * a completely new data area).
1895 vma
->vm_flags
|= VM_SOFTDIRTY
;
1897 vma_set_page_prot(vma
);
1902 vma
->vm_file
= NULL
;
1905 /* Undo any partial mapping done by a device driver. */
1906 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1908 if (vm_flags
& VM_SHARED
)
1909 mapping_unmap_writable(file
->f_mapping
);
1910 allow_write_and_free_vma
:
1911 if (vm_flags
& VM_DENYWRITE
)
1912 allow_write_access(file
);
1917 vm_unacct_memory(charged
);
1921 static unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1924 * We implement the search by looking for an rbtree node that
1925 * immediately follows a suitable gap. That is,
1926 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1927 * - gap_end = vma->vm_start >= info->low_limit + length;
1928 * - gap_end - gap_start >= length
1931 struct mm_struct
*mm
= current
->mm
;
1932 struct vm_area_struct
*vma
;
1933 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1935 /* Adjust search length to account for worst case alignment overhead */
1936 length
= info
->length
+ info
->align_mask
;
1937 if (length
< info
->length
)
1940 /* Adjust search limits by the desired length */
1941 if (info
->high_limit
< length
)
1943 high_limit
= info
->high_limit
- length
;
1945 if (info
->low_limit
> high_limit
)
1947 low_limit
= info
->low_limit
+ length
;
1949 /* Check if rbtree root looks promising */
1950 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1952 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1953 if (vma
->rb_subtree_gap
< length
)
1957 /* Visit left subtree if it looks promising */
1958 gap_end
= vm_start_gap(vma
);
1959 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1960 struct vm_area_struct
*left
=
1961 rb_entry(vma
->vm_rb
.rb_left
,
1962 struct vm_area_struct
, vm_rb
);
1963 if (left
->rb_subtree_gap
>= length
) {
1969 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1971 /* Check if current node has a suitable gap */
1972 if (gap_start
> high_limit
)
1974 if (gap_end
>= low_limit
&&
1975 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1978 /* Visit right subtree if it looks promising */
1979 if (vma
->vm_rb
.rb_right
) {
1980 struct vm_area_struct
*right
=
1981 rb_entry(vma
->vm_rb
.rb_right
,
1982 struct vm_area_struct
, vm_rb
);
1983 if (right
->rb_subtree_gap
>= length
) {
1989 /* Go back up the rbtree to find next candidate node */
1991 struct rb_node
*prev
= &vma
->vm_rb
;
1992 if (!rb_parent(prev
))
1994 vma
= rb_entry(rb_parent(prev
),
1995 struct vm_area_struct
, vm_rb
);
1996 if (prev
== vma
->vm_rb
.rb_left
) {
1997 gap_start
= vm_end_gap(vma
->vm_prev
);
1998 gap_end
= vm_start_gap(vma
);
2005 /* Check highest gap, which does not precede any rbtree node */
2006 gap_start
= mm
->highest_vm_end
;
2007 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
2008 if (gap_start
> high_limit
)
2012 /* We found a suitable gap. Clip it with the original low_limit. */
2013 if (gap_start
< info
->low_limit
)
2014 gap_start
= info
->low_limit
;
2016 /* Adjust gap address to the desired alignment */
2017 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
2019 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
2020 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
2024 static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
2026 struct mm_struct
*mm
= current
->mm
;
2027 struct vm_area_struct
*vma
;
2028 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
2030 /* Adjust search length to account for worst case alignment overhead */
2031 length
= info
->length
+ info
->align_mask
;
2032 if (length
< info
->length
)
2036 * Adjust search limits by the desired length.
2037 * See implementation comment at top of unmapped_area().
2039 gap_end
= info
->high_limit
;
2040 if (gap_end
< length
)
2042 high_limit
= gap_end
- length
;
2044 if (info
->low_limit
> high_limit
)
2046 low_limit
= info
->low_limit
+ length
;
2048 /* Check highest gap, which does not precede any rbtree node */
2049 gap_start
= mm
->highest_vm_end
;
2050 if (gap_start
<= high_limit
)
2053 /* Check if rbtree root looks promising */
2054 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
2056 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
2057 if (vma
->rb_subtree_gap
< length
)
2061 /* Visit right subtree if it looks promising */
2062 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
2063 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
2064 struct vm_area_struct
*right
=
2065 rb_entry(vma
->vm_rb
.rb_right
,
2066 struct vm_area_struct
, vm_rb
);
2067 if (right
->rb_subtree_gap
>= length
) {
2074 /* Check if current node has a suitable gap */
2075 gap_end
= vm_start_gap(vma
);
2076 if (gap_end
< low_limit
)
2078 if (gap_start
<= high_limit
&&
2079 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
2082 /* Visit left subtree if it looks promising */
2083 if (vma
->vm_rb
.rb_left
) {
2084 struct vm_area_struct
*left
=
2085 rb_entry(vma
->vm_rb
.rb_left
,
2086 struct vm_area_struct
, vm_rb
);
2087 if (left
->rb_subtree_gap
>= length
) {
2093 /* Go back up the rbtree to find next candidate node */
2095 struct rb_node
*prev
= &vma
->vm_rb
;
2096 if (!rb_parent(prev
))
2098 vma
= rb_entry(rb_parent(prev
),
2099 struct vm_area_struct
, vm_rb
);
2100 if (prev
== vma
->vm_rb
.rb_right
) {
2101 gap_start
= vma
->vm_prev
?
2102 vm_end_gap(vma
->vm_prev
) : 0;
2109 /* We found a suitable gap. Clip it with the original high_limit. */
2110 if (gap_end
> info
->high_limit
)
2111 gap_end
= info
->high_limit
;
2114 /* Compute highest gap address at the desired alignment */
2115 gap_end
-= info
->length
;
2116 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
2118 VM_BUG_ON(gap_end
< info
->low_limit
);
2119 VM_BUG_ON(gap_end
< gap_start
);
2124 * Search for an unmapped address range.
2126 * We are looking for a range that:
2127 * - does not intersect with any VMA;
2128 * - is contained within the [low_limit, high_limit) interval;
2129 * - is at least the desired size.
2130 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
2132 unsigned long vm_unmapped_area(struct vm_unmapped_area_info
*info
)
2136 if (info
->flags
& VM_UNMAPPED_AREA_TOPDOWN
)
2137 addr
= unmapped_area_topdown(info
);
2139 addr
= unmapped_area(info
);
2141 trace_vm_unmapped_area(addr
, info
);
2145 #ifndef arch_get_mmap_end
2146 #define arch_get_mmap_end(addr) (TASK_SIZE)
2149 #ifndef arch_get_mmap_base
2150 #define arch_get_mmap_base(addr, base) (base)
2153 /* Get an address range which is currently unmapped.
2154 * For shmat() with addr=0.
2156 * Ugly calling convention alert:
2157 * Return value with the low bits set means error value,
2159 * if (ret & ~PAGE_MASK)
2162 * This function "knows" that -ENOMEM has the bits set.
2164 #ifndef HAVE_ARCH_UNMAPPED_AREA
2166 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
2167 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
2169 struct mm_struct
*mm
= current
->mm
;
2170 struct vm_area_struct
*vma
, *prev
;
2171 struct vm_unmapped_area_info info
;
2172 const unsigned long mmap_end
= arch_get_mmap_end(addr
);
2174 if (len
> mmap_end
- mmap_min_addr
)
2177 if (flags
& MAP_FIXED
)
2181 addr
= PAGE_ALIGN(addr
);
2182 vma
= find_vma_prev(mm
, addr
, &prev
);
2183 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
2184 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2185 (!prev
|| addr
>= vm_end_gap(prev
)))
2191 info
.low_limit
= mm
->mmap_base
;
2192 info
.high_limit
= mmap_end
;
2193 info
.align_mask
= 0;
2194 info
.align_offset
= 0;
2195 return vm_unmapped_area(&info
);
2200 * This mmap-allocator allocates new areas top-down from below the
2201 * stack's low limit (the base):
2203 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2205 arch_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
2206 unsigned long len
, unsigned long pgoff
,
2207 unsigned long flags
)
2209 struct vm_area_struct
*vma
, *prev
;
2210 struct mm_struct
*mm
= current
->mm
;
2211 struct vm_unmapped_area_info info
;
2212 const unsigned long mmap_end
= arch_get_mmap_end(addr
);
2214 /* requested length too big for entire address space */
2215 if (len
> mmap_end
- mmap_min_addr
)
2218 if (flags
& MAP_FIXED
)
2221 /* requesting a specific address */
2223 addr
= PAGE_ALIGN(addr
);
2224 vma
= find_vma_prev(mm
, addr
, &prev
);
2225 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
2226 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2227 (!prev
|| addr
>= vm_end_gap(prev
)))
2231 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
2233 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
2234 info
.high_limit
= arch_get_mmap_base(addr
, mm
->mmap_base
);
2235 info
.align_mask
= 0;
2236 info
.align_offset
= 0;
2237 addr
= vm_unmapped_area(&info
);
2240 * A failed mmap() very likely causes application failure,
2241 * so fall back to the bottom-up function here. This scenario
2242 * can happen with large stack limits and large mmap()
2245 if (offset_in_page(addr
)) {
2246 VM_BUG_ON(addr
!= -ENOMEM
);
2248 info
.low_limit
= TASK_UNMAPPED_BASE
;
2249 info
.high_limit
= mmap_end
;
2250 addr
= vm_unmapped_area(&info
);
2258 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2259 unsigned long pgoff
, unsigned long flags
)
2261 unsigned long (*get_area
)(struct file
*, unsigned long,
2262 unsigned long, unsigned long, unsigned long);
2264 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2268 /* Careful about overflows.. */
2269 if (len
> TASK_SIZE
)
2272 get_area
= current
->mm
->get_unmapped_area
;
2274 if (file
->f_op
->get_unmapped_area
)
2275 get_area
= file
->f_op
->get_unmapped_area
;
2276 } else if (flags
& MAP_SHARED
) {
2278 * mmap_region() will call shmem_zero_setup() to create a file,
2279 * so use shmem's get_unmapped_area in case it can be huge.
2280 * do_mmap() will clear pgoff, so match alignment.
2283 get_area
= shmem_get_unmapped_area
;
2286 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2287 if (IS_ERR_VALUE(addr
))
2290 if (addr
> TASK_SIZE
- len
)
2292 if (offset_in_page(addr
))
2295 error
= security_mmap_addr(addr
);
2296 return error
? error
: addr
;
2299 EXPORT_SYMBOL(get_unmapped_area
);
2301 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2302 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2304 struct rb_node
*rb_node
;
2305 struct vm_area_struct
*vma
;
2307 /* Check the cache first. */
2308 vma
= vmacache_find(mm
, addr
);
2312 rb_node
= mm
->mm_rb
.rb_node
;
2315 struct vm_area_struct
*tmp
;
2317 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2319 if (tmp
->vm_end
> addr
) {
2321 if (tmp
->vm_start
<= addr
)
2323 rb_node
= rb_node
->rb_left
;
2325 rb_node
= rb_node
->rb_right
;
2329 vmacache_update(addr
, vma
);
2333 EXPORT_SYMBOL(find_vma
);
2336 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2338 struct vm_area_struct
*
2339 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2340 struct vm_area_struct
**pprev
)
2342 struct vm_area_struct
*vma
;
2344 vma
= find_vma(mm
, addr
);
2346 *pprev
= vma
->vm_prev
;
2348 struct rb_node
*rb_node
= rb_last(&mm
->mm_rb
);
2350 *pprev
= rb_node
? rb_entry(rb_node
, struct vm_area_struct
, vm_rb
) : NULL
;
2356 * Verify that the stack growth is acceptable and
2357 * update accounting. This is shared with both the
2358 * grow-up and grow-down cases.
2360 static int acct_stack_growth(struct vm_area_struct
*vma
,
2361 unsigned long size
, unsigned long grow
)
2363 struct mm_struct
*mm
= vma
->vm_mm
;
2364 unsigned long new_start
;
2366 /* address space limit tests */
2367 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
2370 /* Stack limit test */
2371 if (size
> rlimit(RLIMIT_STACK
))
2374 /* mlock limit tests */
2375 if (vma
->vm_flags
& VM_LOCKED
) {
2376 unsigned long locked
;
2377 unsigned long limit
;
2378 locked
= mm
->locked_vm
+ grow
;
2379 limit
= rlimit(RLIMIT_MEMLOCK
);
2380 limit
>>= PAGE_SHIFT
;
2381 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2385 /* Check to ensure the stack will not grow into a hugetlb-only region */
2386 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2388 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2392 * Overcommit.. This must be the final test, as it will
2393 * update security statistics.
2395 if (security_vm_enough_memory_mm(mm
, grow
))
2401 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2403 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2404 * vma is the last one with address > vma->vm_end. Have to extend vma.
2406 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2408 struct mm_struct
*mm
= vma
->vm_mm
;
2409 struct vm_area_struct
*next
;
2410 unsigned long gap_addr
;
2413 if (!(vma
->vm_flags
& VM_GROWSUP
))
2416 /* Guard against exceeding limits of the address space. */
2417 address
&= PAGE_MASK
;
2418 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2420 address
+= PAGE_SIZE
;
2422 /* Enforce stack_guard_gap */
2423 gap_addr
= address
+ stack_guard_gap
;
2425 /* Guard against overflow */
2426 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2427 gap_addr
= TASK_SIZE
;
2429 next
= vma
->vm_next
;
2430 if (next
&& next
->vm_start
< gap_addr
&& vma_is_accessible(next
)) {
2431 if (!(next
->vm_flags
& VM_GROWSUP
))
2433 /* Check that both stack segments have the same anon_vma? */
2436 /* We must make sure the anon_vma is allocated. */
2437 if (unlikely(anon_vma_prepare(vma
)))
2441 * vma->vm_start/vm_end cannot change under us because the caller
2442 * is required to hold the mmap_lock in read mode. We need the
2443 * anon_vma lock to serialize against concurrent expand_stacks.
2445 anon_vma_lock_write(vma
->anon_vma
);
2447 /* Somebody else might have raced and expanded it already */
2448 if (address
> vma
->vm_end
) {
2449 unsigned long size
, grow
;
2451 size
= address
- vma
->vm_start
;
2452 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2455 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2456 error
= acct_stack_growth(vma
, size
, grow
);
2459 * vma_gap_update() doesn't support concurrent
2460 * updates, but we only hold a shared mmap_lock
2461 * lock here, so we need to protect against
2462 * concurrent vma expansions.
2463 * anon_vma_lock_write() doesn't help here, as
2464 * we don't guarantee that all growable vmas
2465 * in a mm share the same root anon vma.
2466 * So, we reuse mm->page_table_lock to guard
2467 * against concurrent vma expansions.
2469 spin_lock(&mm
->page_table_lock
);
2470 if (vma
->vm_flags
& VM_LOCKED
)
2471 mm
->locked_vm
+= grow
;
2472 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2473 anon_vma_interval_tree_pre_update_vma(vma
);
2474 vma
->vm_end
= address
;
2475 anon_vma_interval_tree_post_update_vma(vma
);
2477 vma_gap_update(vma
->vm_next
);
2479 mm
->highest_vm_end
= vm_end_gap(vma
);
2480 spin_unlock(&mm
->page_table_lock
);
2482 perf_event_mmap(vma
);
2486 anon_vma_unlock_write(vma
->anon_vma
);
2487 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2491 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2494 * vma is the first one with address < vma->vm_start. Have to extend vma.
2496 int expand_downwards(struct vm_area_struct
*vma
,
2497 unsigned long address
)
2499 struct mm_struct
*mm
= vma
->vm_mm
;
2500 struct vm_area_struct
*prev
;
2503 address
&= PAGE_MASK
;
2504 if (address
< mmap_min_addr
)
2507 /* Enforce stack_guard_gap */
2508 prev
= vma
->vm_prev
;
2509 /* Check that both stack segments have the same anon_vma? */
2510 if (prev
&& !(prev
->vm_flags
& VM_GROWSDOWN
) &&
2511 vma_is_accessible(prev
)) {
2512 if (address
- prev
->vm_end
< stack_guard_gap
)
2516 /* We must make sure the anon_vma is allocated. */
2517 if (unlikely(anon_vma_prepare(vma
)))
2521 * vma->vm_start/vm_end cannot change under us because the caller
2522 * is required to hold the mmap_lock in read mode. We need the
2523 * anon_vma lock to serialize against concurrent expand_stacks.
2525 anon_vma_lock_write(vma
->anon_vma
);
2527 /* Somebody else might have raced and expanded it already */
2528 if (address
< vma
->vm_start
) {
2529 unsigned long size
, grow
;
2531 size
= vma
->vm_end
- address
;
2532 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2535 if (grow
<= vma
->vm_pgoff
) {
2536 error
= acct_stack_growth(vma
, size
, grow
);
2539 * vma_gap_update() doesn't support concurrent
2540 * updates, but we only hold a shared mmap_lock
2541 * lock here, so we need to protect against
2542 * concurrent vma expansions.
2543 * anon_vma_lock_write() doesn't help here, as
2544 * we don't guarantee that all growable vmas
2545 * in a mm share the same root anon vma.
2546 * So, we reuse mm->page_table_lock to guard
2547 * against concurrent vma expansions.
2549 spin_lock(&mm
->page_table_lock
);
2550 if (vma
->vm_flags
& VM_LOCKED
)
2551 mm
->locked_vm
+= grow
;
2552 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2553 anon_vma_interval_tree_pre_update_vma(vma
);
2554 vma
->vm_start
= address
;
2555 vma
->vm_pgoff
-= grow
;
2556 anon_vma_interval_tree_post_update_vma(vma
);
2557 vma_gap_update(vma
);
2558 spin_unlock(&mm
->page_table_lock
);
2560 perf_event_mmap(vma
);
2564 anon_vma_unlock_write(vma
->anon_vma
);
2565 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2570 /* enforced gap between the expanding stack and other mappings. */
2571 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2573 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2578 val
= simple_strtoul(p
, &endptr
, 10);
2580 stack_guard_gap
= val
<< PAGE_SHIFT
;
2584 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2586 #ifdef CONFIG_STACK_GROWSUP
2587 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2589 return expand_upwards(vma
, address
);
2592 struct vm_area_struct
*
2593 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2595 struct vm_area_struct
*vma
, *prev
;
2598 vma
= find_vma_prev(mm
, addr
, &prev
);
2599 if (vma
&& (vma
->vm_start
<= addr
))
2601 /* don't alter vm_end if the coredump is running */
2602 if (!prev
|| expand_stack(prev
, addr
))
2604 if (prev
->vm_flags
& VM_LOCKED
)
2605 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2609 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2611 return expand_downwards(vma
, address
);
2614 struct vm_area_struct
*
2615 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2617 struct vm_area_struct
*vma
;
2618 unsigned long start
;
2621 vma
= find_vma(mm
, addr
);
2624 if (vma
->vm_start
<= addr
)
2626 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2628 start
= vma
->vm_start
;
2629 if (expand_stack(vma
, addr
))
2631 if (vma
->vm_flags
& VM_LOCKED
)
2632 populate_vma_page_range(vma
, addr
, start
, NULL
);
2637 EXPORT_SYMBOL_GPL(find_extend_vma
);
2640 * Ok - we have the memory areas we should free on the vma list,
2641 * so release them, and do the vma updates.
2643 * Called with the mm semaphore held.
2645 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2647 unsigned long nr_accounted
= 0;
2649 /* Update high watermark before we lower total_vm */
2650 update_hiwater_vm(mm
);
2652 long nrpages
= vma_pages(vma
);
2654 if (vma
->vm_flags
& VM_ACCOUNT
)
2655 nr_accounted
+= nrpages
;
2656 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2657 vma
= remove_vma(vma
);
2659 vm_unacct_memory(nr_accounted
);
2664 * Get rid of page table information in the indicated region.
2666 * Called with the mm semaphore held.
2668 static void unmap_region(struct mm_struct
*mm
,
2669 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2670 unsigned long start
, unsigned long end
)
2672 struct vm_area_struct
*next
= vma_next(mm
, prev
);
2673 struct mmu_gather tlb
;
2676 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2677 update_hiwater_rss(mm
);
2678 unmap_vmas(&tlb
, vma
, start
, end
);
2679 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2680 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2681 tlb_finish_mmu(&tlb
, start
, end
);
2685 * Create a list of vma's touched by the unmap, removing them from the mm's
2686 * vma list as we go..
2689 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2690 struct vm_area_struct
*prev
, unsigned long end
)
2692 struct vm_area_struct
**insertion_point
;
2693 struct vm_area_struct
*tail_vma
= NULL
;
2695 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2696 vma
->vm_prev
= NULL
;
2698 vma_rb_erase(vma
, &mm
->mm_rb
);
2702 } while (vma
&& vma
->vm_start
< end
);
2703 *insertion_point
= vma
;
2705 vma
->vm_prev
= prev
;
2706 vma_gap_update(vma
);
2708 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2709 tail_vma
->vm_next
= NULL
;
2711 /* Kill the cache */
2712 vmacache_invalidate(mm
);
2715 * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2716 * VM_GROWSUP VMA. Such VMAs can change their size under
2717 * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2719 if (vma
&& (vma
->vm_flags
& VM_GROWSDOWN
))
2721 if (prev
&& (prev
->vm_flags
& VM_GROWSUP
))
2727 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2728 * has already been checked or doesn't make sense to fail.
2730 int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2731 unsigned long addr
, int new_below
)
2733 struct vm_area_struct
*new;
2736 if (vma
->vm_ops
&& vma
->vm_ops
->split
) {
2737 err
= vma
->vm_ops
->split(vma
, addr
);
2742 new = vm_area_dup(vma
);
2749 new->vm_start
= addr
;
2750 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2753 err
= vma_dup_policy(vma
, new);
2757 err
= anon_vma_clone(new, vma
);
2762 get_file(new->vm_file
);
2764 if (new->vm_ops
&& new->vm_ops
->open
)
2765 new->vm_ops
->open(new);
2768 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2769 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2771 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2777 /* Clean everything up if vma_adjust failed. */
2778 if (new->vm_ops
&& new->vm_ops
->close
)
2779 new->vm_ops
->close(new);
2782 unlink_anon_vmas(new);
2784 mpol_put(vma_policy(new));
2791 * Split a vma into two pieces at address 'addr', a new vma is allocated
2792 * either for the first part or the tail.
2794 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2795 unsigned long addr
, int new_below
)
2797 if (mm
->map_count
>= sysctl_max_map_count
)
2800 return __split_vma(mm
, vma
, addr
, new_below
);
2803 /* Munmap is split into 2 main parts -- this part which finds
2804 * what needs doing, and the areas themselves, which do the
2805 * work. This now handles partial unmappings.
2806 * Jeremy Fitzhardinge <jeremy@goop.org>
2808 int __do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2809 struct list_head
*uf
, bool downgrade
)
2812 struct vm_area_struct
*vma
, *prev
, *last
;
2814 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2817 len
= PAGE_ALIGN(len
);
2823 * arch_unmap() might do unmaps itself. It must be called
2824 * and finish any rbtree manipulation before this code
2825 * runs and also starts to manipulate the rbtree.
2827 arch_unmap(mm
, start
, end
);
2829 /* Find the first overlapping VMA */
2830 vma
= find_vma(mm
, start
);
2833 prev
= vma
->vm_prev
;
2834 /* we have start < vma->vm_end */
2836 /* if it doesn't overlap, we have nothing.. */
2837 if (vma
->vm_start
>= end
)
2841 * If we need to split any vma, do it now to save pain later.
2843 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2844 * unmapped vm_area_struct will remain in use: so lower split_vma
2845 * places tmp vma above, and higher split_vma places tmp vma below.
2847 if (start
> vma
->vm_start
) {
2851 * Make sure that map_count on return from munmap() will
2852 * not exceed its limit; but let map_count go just above
2853 * its limit temporarily, to help free resources as expected.
2855 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2858 error
= __split_vma(mm
, vma
, start
, 0);
2864 /* Does it split the last one? */
2865 last
= find_vma(mm
, end
);
2866 if (last
&& end
> last
->vm_start
) {
2867 int error
= __split_vma(mm
, last
, end
, 1);
2871 vma
= vma_next(mm
, prev
);
2875 * If userfaultfd_unmap_prep returns an error the vmas
2876 * will remain splitted, but userland will get a
2877 * highly unexpected error anyway. This is no
2878 * different than the case where the first of the two
2879 * __split_vma fails, but we don't undo the first
2880 * split, despite we could. This is unlikely enough
2881 * failure that it's not worth optimizing it for.
2883 int error
= userfaultfd_unmap_prep(vma
, start
, end
, uf
);
2889 * unlock any mlock()ed ranges before detaching vmas
2891 if (mm
->locked_vm
) {
2892 struct vm_area_struct
*tmp
= vma
;
2893 while (tmp
&& tmp
->vm_start
< end
) {
2894 if (tmp
->vm_flags
& VM_LOCKED
) {
2895 mm
->locked_vm
-= vma_pages(tmp
);
2896 munlock_vma_pages_all(tmp
);
2903 /* Detach vmas from rbtree */
2904 if (!detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
))
2908 mmap_write_downgrade(mm
);
2910 unmap_region(mm
, vma
, prev
, start
, end
);
2912 /* Fix up all other VM information */
2913 remove_vma_list(mm
, vma
);
2915 return downgrade
? 1 : 0;
2918 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2919 struct list_head
*uf
)
2921 return __do_munmap(mm
, start
, len
, uf
, false);
2924 static int __vm_munmap(unsigned long start
, size_t len
, bool downgrade
)
2927 struct mm_struct
*mm
= current
->mm
;
2930 if (mmap_write_lock_killable(mm
))
2933 ret
= __do_munmap(mm
, start
, len
, &uf
, downgrade
);
2935 * Returning 1 indicates mmap_lock is downgraded.
2936 * But 1 is not legal return value of vm_munmap() and munmap(), reset
2937 * it to 0 before return.
2940 mmap_read_unlock(mm
);
2943 mmap_write_unlock(mm
);
2945 userfaultfd_unmap_complete(mm
, &uf
);
2949 int vm_munmap(unsigned long start
, size_t len
)
2951 return __vm_munmap(start
, len
, false);
2953 EXPORT_SYMBOL(vm_munmap
);
2955 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2957 addr
= untagged_addr(addr
);
2958 profile_munmap(addr
);
2959 return __vm_munmap(addr
, len
, true);
2964 * Emulation of deprecated remap_file_pages() syscall.
2966 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2967 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2970 struct mm_struct
*mm
= current
->mm
;
2971 struct vm_area_struct
*vma
;
2972 unsigned long populate
= 0;
2973 unsigned long ret
= -EINVAL
;
2976 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
2977 current
->comm
, current
->pid
);
2981 start
= start
& PAGE_MASK
;
2982 size
= size
& PAGE_MASK
;
2984 if (start
+ size
<= start
)
2987 /* Does pgoff wrap? */
2988 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2991 if (mmap_write_lock_killable(mm
))
2994 vma
= find_vma(mm
, start
);
2996 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2999 if (start
< vma
->vm_start
)
3002 if (start
+ size
> vma
->vm_end
) {
3003 struct vm_area_struct
*next
;
3005 for (next
= vma
->vm_next
; next
; next
= next
->vm_next
) {
3006 /* hole between vmas ? */
3007 if (next
->vm_start
!= next
->vm_prev
->vm_end
)
3010 if (next
->vm_file
!= vma
->vm_file
)
3013 if (next
->vm_flags
!= vma
->vm_flags
)
3016 if (start
+ size
<= next
->vm_end
)
3024 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
3025 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
3026 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
3028 flags
&= MAP_NONBLOCK
;
3029 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
3030 if (vma
->vm_flags
& VM_LOCKED
) {
3031 struct vm_area_struct
*tmp
;
3032 flags
|= MAP_LOCKED
;
3034 /* drop PG_Mlocked flag for over-mapped range */
3035 for (tmp
= vma
; tmp
->vm_start
>= start
+ size
;
3036 tmp
= tmp
->vm_next
) {
3038 * Split pmd and munlock page on the border
3041 vma_adjust_trans_huge(tmp
, start
, start
+ size
, 0);
3043 munlock_vma_pages_range(tmp
,
3044 max(tmp
->vm_start
, start
),
3045 min(tmp
->vm_end
, start
+ size
));
3049 file
= get_file(vma
->vm_file
);
3050 ret
= do_mmap(vma
->vm_file
, start
, size
,
3051 prot
, flags
, pgoff
, &populate
, NULL
);
3054 mmap_write_unlock(mm
);
3056 mm_populate(ret
, populate
);
3057 if (!IS_ERR_VALUE(ret
))
3063 * this is really a simplified "do_mmap". it only handles
3064 * anonymous maps. eventually we may be able to do some
3065 * brk-specific accounting here.
3067 static int do_brk_flags(unsigned long addr
, unsigned long len
, unsigned long flags
, struct list_head
*uf
)
3069 struct mm_struct
*mm
= current
->mm
;
3070 struct vm_area_struct
*vma
, *prev
;
3071 struct rb_node
**rb_link
, *rb_parent
;
3072 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
3074 unsigned long mapped_addr
;
3076 /* Until we need other flags, refuse anything except VM_EXEC. */
3077 if ((flags
& (~VM_EXEC
)) != 0)
3079 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
3081 mapped_addr
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
3082 if (IS_ERR_VALUE(mapped_addr
))
3085 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
3089 /* Clear old maps, set up prev, rb_link, rb_parent, and uf */
3090 if (munmap_vma_range(mm
, addr
, len
, &prev
, &rb_link
, &rb_parent
, uf
))
3093 /* Check against address space limits *after* clearing old maps... */
3094 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
3097 if (mm
->map_count
> sysctl_max_map_count
)
3100 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
3103 /* Can we just expand an old private anonymous mapping? */
3104 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
3105 NULL
, NULL
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
3110 * create a vma struct for an anonymous mapping
3112 vma
= vm_area_alloc(mm
);
3114 vm_unacct_memory(len
>> PAGE_SHIFT
);
3118 vma_set_anonymous(vma
);
3119 vma
->vm_start
= addr
;
3120 vma
->vm_end
= addr
+ len
;
3121 vma
->vm_pgoff
= pgoff
;
3122 vma
->vm_flags
= flags
;
3123 vma
->vm_page_prot
= vm_get_page_prot(flags
);
3124 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3126 perf_event_mmap(vma
);
3127 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3128 mm
->data_vm
+= len
>> PAGE_SHIFT
;
3129 if (flags
& VM_LOCKED
)
3130 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
3131 vma
->vm_flags
|= VM_SOFTDIRTY
;
3135 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
3137 struct mm_struct
*mm
= current
->mm
;
3143 len
= PAGE_ALIGN(request
);
3149 if (mmap_write_lock_killable(mm
))
3152 ret
= do_brk_flags(addr
, len
, flags
, &uf
);
3153 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3154 mmap_write_unlock(mm
);
3155 userfaultfd_unmap_complete(mm
, &uf
);
3156 if (populate
&& !ret
)
3157 mm_populate(addr
, len
);
3160 EXPORT_SYMBOL(vm_brk_flags
);
3162 int vm_brk(unsigned long addr
, unsigned long len
)
3164 return vm_brk_flags(addr
, len
, 0);
3166 EXPORT_SYMBOL(vm_brk
);
3168 /* Release all mmaps. */
3169 void exit_mmap(struct mm_struct
*mm
)
3171 struct mmu_gather tlb
;
3172 struct vm_area_struct
*vma
;
3173 unsigned long nr_accounted
= 0;
3175 /* mm's last user has gone, and its about to be pulled down */
3176 mmu_notifier_release(mm
);
3178 if (unlikely(mm_is_oom_victim(mm
))) {
3180 * Manually reap the mm to free as much memory as possible.
3181 * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3182 * this mm from further consideration. Taking mm->mmap_lock for
3183 * write after setting MMF_OOM_SKIP will guarantee that the oom
3184 * reaper will not run on this mm again after mmap_lock is
3187 * Nothing can be holding mm->mmap_lock here and the above call
3188 * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3189 * __oom_reap_task_mm() will not block.
3191 * This needs to be done before calling munlock_vma_pages_all(),
3192 * which clears VM_LOCKED, otherwise the oom reaper cannot
3195 (void)__oom_reap_task_mm(mm
);
3197 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3198 mmap_write_lock(mm
);
3199 mmap_write_unlock(mm
);
3202 if (mm
->locked_vm
) {
3205 if (vma
->vm_flags
& VM_LOCKED
)
3206 munlock_vma_pages_all(vma
);
3214 if (!vma
) /* Can happen if dup_mmap() received an OOM */
3219 tlb_gather_mmu(&tlb
, mm
, 0, -1);
3220 /* update_hiwater_rss(mm) here? but nobody should be looking */
3221 /* Use -1 here to ensure all VMAs in the mm are unmapped */
3222 unmap_vmas(&tlb
, vma
, 0, -1);
3223 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
3224 tlb_finish_mmu(&tlb
, 0, -1);
3227 * Walk the list again, actually closing and freeing it,
3228 * with preemption enabled, without holding any MM locks.
3231 if (vma
->vm_flags
& VM_ACCOUNT
)
3232 nr_accounted
+= vma_pages(vma
);
3233 vma
= remove_vma(vma
);
3236 vm_unacct_memory(nr_accounted
);
3239 /* Insert vm structure into process list sorted by address
3240 * and into the inode's i_mmap tree. If vm_file is non-NULL
3241 * then i_mmap_rwsem is taken here.
3243 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3245 struct vm_area_struct
*prev
;
3246 struct rb_node
**rb_link
, *rb_parent
;
3248 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
3249 &prev
, &rb_link
, &rb_parent
))
3251 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3252 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
3256 * The vm_pgoff of a purely anonymous vma should be irrelevant
3257 * until its first write fault, when page's anon_vma and index
3258 * are set. But now set the vm_pgoff it will almost certainly
3259 * end up with (unless mremap moves it elsewhere before that
3260 * first wfault), so /proc/pid/maps tells a consistent story.
3262 * By setting it to reflect the virtual start address of the
3263 * vma, merges and splits can happen in a seamless way, just
3264 * using the existing file pgoff checks and manipulations.
3265 * Similarly in do_mmap and in do_brk_flags.
3267 if (vma_is_anonymous(vma
)) {
3268 BUG_ON(vma
->anon_vma
);
3269 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3272 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3277 * Copy the vma structure to a new location in the same mm,
3278 * prior to moving page table entries, to effect an mremap move.
3280 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3281 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3282 bool *need_rmap_locks
)
3284 struct vm_area_struct
*vma
= *vmap
;
3285 unsigned long vma_start
= vma
->vm_start
;
3286 struct mm_struct
*mm
= vma
->vm_mm
;
3287 struct vm_area_struct
*new_vma
, *prev
;
3288 struct rb_node
**rb_link
, *rb_parent
;
3289 bool faulted_in_anon_vma
= true;
3292 * If anonymous vma has not yet been faulted, update new pgoff
3293 * to match new location, to increase its chance of merging.
3295 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3296 pgoff
= addr
>> PAGE_SHIFT
;
3297 faulted_in_anon_vma
= false;
3300 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
3301 return NULL
; /* should never get here */
3302 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
3303 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
3304 vma
->vm_userfaultfd_ctx
);
3307 * Source vma may have been merged into new_vma
3309 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3310 vma_start
< new_vma
->vm_end
)) {
3312 * The only way we can get a vma_merge with
3313 * self during an mremap is if the vma hasn't
3314 * been faulted in yet and we were allowed to
3315 * reset the dst vma->vm_pgoff to the
3316 * destination address of the mremap to allow
3317 * the merge to happen. mremap must change the
3318 * vm_pgoff linearity between src and dst vmas
3319 * (in turn preventing a vma_merge) to be
3320 * safe. It is only safe to keep the vm_pgoff
3321 * linear if there are no pages mapped yet.
3323 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3324 *vmap
= vma
= new_vma
;
3326 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3328 new_vma
= vm_area_dup(vma
);
3331 new_vma
->vm_start
= addr
;
3332 new_vma
->vm_end
= addr
+ len
;
3333 new_vma
->vm_pgoff
= pgoff
;
3334 if (vma_dup_policy(vma
, new_vma
))
3336 if (anon_vma_clone(new_vma
, vma
))
3337 goto out_free_mempol
;
3338 if (new_vma
->vm_file
)
3339 get_file(new_vma
->vm_file
);
3340 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3341 new_vma
->vm_ops
->open(new_vma
);
3342 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
3343 *need_rmap_locks
= false;
3348 mpol_put(vma_policy(new_vma
));
3350 vm_area_free(new_vma
);
3356 * Return true if the calling process may expand its vm space by the passed
3359 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3361 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3364 if (is_data_mapping(flags
) &&
3365 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3366 /* Workaround for Valgrind */
3367 if (rlimit(RLIMIT_DATA
) == 0 &&
3368 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3371 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3372 current
->comm
, current
->pid
,
3373 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3374 rlimit(RLIMIT_DATA
),
3375 ignore_rlimit_data
? "" : " or use boot option ignore_rlimit_data");
3377 if (!ignore_rlimit_data
)
3384 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3386 mm
->total_vm
+= npages
;
3388 if (is_exec_mapping(flags
))
3389 mm
->exec_vm
+= npages
;
3390 else if (is_stack_mapping(flags
))
3391 mm
->stack_vm
+= npages
;
3392 else if (is_data_mapping(flags
))
3393 mm
->data_vm
+= npages
;
3396 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
);
3399 * Having a close hook prevents vma merging regardless of flags.
3401 static void special_mapping_close(struct vm_area_struct
*vma
)
3405 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3407 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3410 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3412 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3414 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3418 return sm
->mremap(sm
, new_vma
);
3423 static const struct vm_operations_struct special_mapping_vmops
= {
3424 .close
= special_mapping_close
,
3425 .fault
= special_mapping_fault
,
3426 .mremap
= special_mapping_mremap
,
3427 .name
= special_mapping_name
,
3428 /* vDSO code relies that VVAR can't be accessed remotely */
3432 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3433 .close
= special_mapping_close
,
3434 .fault
= special_mapping_fault
,
3437 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
)
3439 struct vm_area_struct
*vma
= vmf
->vma
;
3441 struct page
**pages
;
3443 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3444 pages
= vma
->vm_private_data
;
3446 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3449 return sm
->fault(sm
, vmf
->vma
, vmf
);
3454 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3458 struct page
*page
= *pages
;
3464 return VM_FAULT_SIGBUS
;
3467 static struct vm_area_struct
*__install_special_mapping(
3468 struct mm_struct
*mm
,
3469 unsigned long addr
, unsigned long len
,
3470 unsigned long vm_flags
, void *priv
,
3471 const struct vm_operations_struct
*ops
)
3474 struct vm_area_struct
*vma
;
3476 vma
= vm_area_alloc(mm
);
3477 if (unlikely(vma
== NULL
))
3478 return ERR_PTR(-ENOMEM
);
3480 vma
->vm_start
= addr
;
3481 vma
->vm_end
= addr
+ len
;
3483 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
3484 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3487 vma
->vm_private_data
= priv
;
3489 ret
= insert_vm_struct(mm
, vma
);
3493 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3495 perf_event_mmap(vma
);
3501 return ERR_PTR(ret
);
3504 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3505 const struct vm_special_mapping
*sm
)
3507 return vma
->vm_private_data
== sm
&&
3508 (vma
->vm_ops
== &special_mapping_vmops
||
3509 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3513 * Called with mm->mmap_lock held for writing.
3514 * Insert a new vma covering the given region, with the given flags.
3515 * Its pages are supplied by the given array of struct page *.
3516 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3517 * The region past the last page supplied will always produce SIGBUS.
3518 * The array pointer and the pages it points to are assumed to stay alive
3519 * for as long as this mapping might exist.
3521 struct vm_area_struct
*_install_special_mapping(
3522 struct mm_struct
*mm
,
3523 unsigned long addr
, unsigned long len
,
3524 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3526 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3527 &special_mapping_vmops
);
3530 int install_special_mapping(struct mm_struct
*mm
,
3531 unsigned long addr
, unsigned long len
,
3532 unsigned long vm_flags
, struct page
**pages
)
3534 struct vm_area_struct
*vma
= __install_special_mapping(
3535 mm
, addr
, len
, vm_flags
, (void *)pages
,
3536 &legacy_special_mapping_vmops
);
3538 return PTR_ERR_OR_ZERO(vma
);
3541 static DEFINE_MUTEX(mm_all_locks_mutex
);
3543 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3545 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3547 * The LSB of head.next can't change from under us
3548 * because we hold the mm_all_locks_mutex.
3550 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_lock
);
3552 * We can safely modify head.next after taking the
3553 * anon_vma->root->rwsem. If some other vma in this mm shares
3554 * the same anon_vma we won't take it again.
3556 * No need of atomic instructions here, head.next
3557 * can't change from under us thanks to the
3558 * anon_vma->root->rwsem.
3560 if (__test_and_set_bit(0, (unsigned long *)
3561 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3566 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3568 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3570 * AS_MM_ALL_LOCKS can't change from under us because
3571 * we hold the mm_all_locks_mutex.
3573 * Operations on ->flags have to be atomic because
3574 * even if AS_MM_ALL_LOCKS is stable thanks to the
3575 * mm_all_locks_mutex, there may be other cpus
3576 * changing other bitflags in parallel to us.
3578 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3580 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_lock
);
3585 * This operation locks against the VM for all pte/vma/mm related
3586 * operations that could ever happen on a certain mm. This includes
3587 * vmtruncate, try_to_unmap, and all page faults.
3589 * The caller must take the mmap_lock in write mode before calling
3590 * mm_take_all_locks(). The caller isn't allowed to release the
3591 * mmap_lock until mm_drop_all_locks() returns.
3593 * mmap_lock in write mode is required in order to block all operations
3594 * that could modify pagetables and free pages without need of
3595 * altering the vma layout. It's also needed in write mode to avoid new
3596 * anon_vmas to be associated with existing vmas.
3598 * A single task can't take more than one mm_take_all_locks() in a row
3599 * or it would deadlock.
3601 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3602 * mapping->flags avoid to take the same lock twice, if more than one
3603 * vma in this mm is backed by the same anon_vma or address_space.
3605 * We take locks in following order, accordingly to comment at beginning
3607 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3609 * - all i_mmap_rwsem locks;
3610 * - all anon_vma->rwseml
3612 * We can take all locks within these types randomly because the VM code
3613 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3614 * mm_all_locks_mutex.
3616 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3617 * that may have to take thousand of locks.
3619 * mm_take_all_locks() can fail if it's interrupted by signals.
3621 int mm_take_all_locks(struct mm_struct
*mm
)
3623 struct vm_area_struct
*vma
;
3624 struct anon_vma_chain
*avc
;
3626 BUG_ON(mmap_read_trylock(mm
));
3628 mutex_lock(&mm_all_locks_mutex
);
3630 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3631 if (signal_pending(current
))
3633 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3634 is_vm_hugetlb_page(vma
))
3635 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3638 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3639 if (signal_pending(current
))
3641 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3642 !is_vm_hugetlb_page(vma
))
3643 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3646 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3647 if (signal_pending(current
))
3650 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3651 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3657 mm_drop_all_locks(mm
);
3661 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3663 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3665 * The LSB of head.next can't change to 0 from under
3666 * us because we hold the mm_all_locks_mutex.
3668 * We must however clear the bitflag before unlocking
3669 * the vma so the users using the anon_vma->rb_root will
3670 * never see our bitflag.
3672 * No need of atomic instructions here, head.next
3673 * can't change from under us until we release the
3674 * anon_vma->root->rwsem.
3676 if (!__test_and_clear_bit(0, (unsigned long *)
3677 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3679 anon_vma_unlock_write(anon_vma
);
3683 static void vm_unlock_mapping(struct address_space
*mapping
)
3685 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3687 * AS_MM_ALL_LOCKS can't change to 0 from under us
3688 * because we hold the mm_all_locks_mutex.
3690 i_mmap_unlock_write(mapping
);
3691 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3698 * The mmap_lock cannot be released by the caller until
3699 * mm_drop_all_locks() returns.
3701 void mm_drop_all_locks(struct mm_struct
*mm
)
3703 struct vm_area_struct
*vma
;
3704 struct anon_vma_chain
*avc
;
3706 BUG_ON(mmap_read_trylock(mm
));
3707 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3709 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3711 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3712 vm_unlock_anon_vma(avc
->anon_vma
);
3713 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3714 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3717 mutex_unlock(&mm_all_locks_mutex
);
3721 * initialise the percpu counter for VM
3723 void __init
mmap_init(void)
3727 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3732 * Initialise sysctl_user_reserve_kbytes.
3734 * This is intended to prevent a user from starting a single memory hogging
3735 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3738 * The default value is min(3% of free memory, 128MB)
3739 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3741 static int init_user_reserve(void)
3743 unsigned long free_kbytes
;
3745 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3747 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3750 subsys_initcall(init_user_reserve
);
3753 * Initialise sysctl_admin_reserve_kbytes.
3755 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3756 * to log in and kill a memory hogging process.
3758 * Systems with more than 256MB will reserve 8MB, enough to recover
3759 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3760 * only reserve 3% of free pages by default.
3762 static int init_admin_reserve(void)
3764 unsigned long free_kbytes
;
3766 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3768 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3771 subsys_initcall(init_admin_reserve
);
3774 * Reinititalise user and admin reserves if memory is added or removed.
3776 * The default user reserve max is 128MB, and the default max for the
3777 * admin reserve is 8MB. These are usually, but not always, enough to
3778 * enable recovery from a memory hogging process using login/sshd, a shell,
3779 * and tools like top. It may make sense to increase or even disable the
3780 * reserve depending on the existence of swap or variations in the recovery
3781 * tools. So, the admin may have changed them.
3783 * If memory is added and the reserves have been eliminated or increased above
3784 * the default max, then we'll trust the admin.
3786 * If memory is removed and there isn't enough free memory, then we
3787 * need to reset the reserves.
3789 * Otherwise keep the reserve set by the admin.
3791 static int reserve_mem_notifier(struct notifier_block
*nb
,
3792 unsigned long action
, void *data
)
3794 unsigned long tmp
, free_kbytes
;
3798 /* Default max is 128MB. Leave alone if modified by operator. */
3799 tmp
= sysctl_user_reserve_kbytes
;
3800 if (0 < tmp
&& tmp
< (1UL << 17))
3801 init_user_reserve();
3803 /* Default max is 8MB. Leave alone if modified by operator. */
3804 tmp
= sysctl_admin_reserve_kbytes
;
3805 if (0 < tmp
&& tmp
< (1UL << 13))
3806 init_admin_reserve();
3810 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3812 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3813 init_user_reserve();
3814 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3815 sysctl_user_reserve_kbytes
);
3818 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3819 init_admin_reserve();
3820 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3821 sysctl_admin_reserve_kbytes
);
3830 static struct notifier_block reserve_mem_nb
= {
3831 .notifier_call
= reserve_mem_notifier
,
3834 static int __meminit
init_reserve_notifier(void)
3836 if (register_hotmemory_notifier(&reserve_mem_nb
))
3837 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3841 subsys_initcall(init_reserve_notifier
);