6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/shmem_fs.h>
29 #include <linux/profile.h>
30 #include <linux/export.h>
31 #include <linux/mount.h>
32 #include <linux/mempolicy.h>
33 #include <linux/rmap.h>
34 #include <linux/mmu_notifier.h>
35 #include <linux/mmdebug.h>
36 #include <linux/perf_event.h>
37 #include <linux/audit.h>
38 #include <linux/khugepaged.h>
39 #include <linux/uprobes.h>
40 #include <linux/rbtree_augmented.h>
41 #include <linux/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
44 #include <linux/userfaultfd_k.h>
45 #include <linux/moduleparam.h>
46 #include <linux/pkeys.h>
47 #include <linux/oom.h>
48 #include <linux/sched/mm.h>
50 #include <linux/uaccess.h>
51 #include <asm/cacheflush.h>
53 #include <asm/mmu_context.h>
57 #ifndef arch_mmap_check
58 #define arch_mmap_check(addr, len, flags) (0)
61 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
62 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
63 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
64 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
66 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
67 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
68 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
69 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
72 static bool ignore_rlimit_data
;
73 core_param(ignore_rlimit_data
, ignore_rlimit_data
, bool, 0644);
75 static void unmap_region(struct mm_struct
*mm
,
76 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
77 unsigned long start
, unsigned long end
);
79 /* description of effects of mapping type and prot in current implementation.
80 * this is due to the limited x86 page protection hardware. The expected
81 * behavior is in parens:
84 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
85 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
86 * w: (no) no w: (no) no w: (yes) yes w: (no) no
87 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
89 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
90 * w: (no) no w: (no) no w: (copy) copy w: (no) no
91 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
99 pgprot_t protection_map
[16] __ro_after_init
= {
100 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
101 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
104 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
106 return __pgprot(pgprot_val(protection_map
[vm_flags
&
107 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
108 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
110 EXPORT_SYMBOL(vm_get_page_prot
);
112 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
114 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
117 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
118 void vma_set_page_prot(struct vm_area_struct
*vma
)
120 unsigned long vm_flags
= vma
->vm_flags
;
121 pgprot_t vm_page_prot
;
123 vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
124 if (vma_wants_writenotify(vma
, vm_page_prot
)) {
125 vm_flags
&= ~VM_SHARED
;
126 vm_page_prot
= vm_pgprot_modify(vm_page_prot
, vm_flags
);
128 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
129 WRITE_ONCE(vma
->vm_page_prot
, vm_page_prot
);
133 * Requires inode->i_mapping->i_mmap_rwsem
135 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
136 struct file
*file
, struct address_space
*mapping
)
138 if (vma
->vm_flags
& VM_DENYWRITE
)
139 atomic_inc(&file_inode(file
)->i_writecount
);
140 if (vma
->vm_flags
& VM_SHARED
)
141 mapping_unmap_writable(mapping
);
143 flush_dcache_mmap_lock(mapping
);
144 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
145 flush_dcache_mmap_unlock(mapping
);
149 * Unlink a file-based vm structure from its interval tree, to hide
150 * vma from rmap and vmtruncate before freeing its page tables.
152 void unlink_file_vma(struct vm_area_struct
*vma
)
154 struct file
*file
= vma
->vm_file
;
157 struct address_space
*mapping
= file
->f_mapping
;
158 i_mmap_lock_write(mapping
);
159 __remove_shared_vm_struct(vma
, file
, mapping
);
160 i_mmap_unlock_write(mapping
);
165 * Close a vm structure and free it, returning the next.
167 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
169 struct vm_area_struct
*next
= vma
->vm_next
;
172 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
173 vma
->vm_ops
->close(vma
);
176 mpol_put(vma_policy(vma
));
177 kmem_cache_free(vm_area_cachep
, vma
);
181 static int do_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
,
182 struct list_head
*uf
);
183 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
185 unsigned long retval
;
186 unsigned long newbrk
, oldbrk
;
187 struct mm_struct
*mm
= current
->mm
;
188 struct vm_area_struct
*next
;
189 unsigned long min_brk
;
193 if (down_write_killable(&mm
->mmap_sem
))
196 #ifdef CONFIG_COMPAT_BRK
198 * CONFIG_COMPAT_BRK can still be overridden by setting
199 * randomize_va_space to 2, which will still cause mm->start_brk
200 * to be arbitrarily shifted
202 if (current
->brk_randomized
)
203 min_brk
= mm
->start_brk
;
205 min_brk
= mm
->end_data
;
207 min_brk
= mm
->start_brk
;
213 * Check against rlimit here. If this check is done later after the test
214 * of oldbrk with newbrk then it can escape the test and let the data
215 * segment grow beyond its set limit the in case where the limit is
216 * not page aligned -Ram Gupta
218 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
219 mm
->end_data
, mm
->start_data
))
222 newbrk
= PAGE_ALIGN(brk
);
223 oldbrk
= PAGE_ALIGN(mm
->brk
);
224 if (oldbrk
== newbrk
)
227 /* Always allow shrinking brk. */
228 if (brk
<= mm
->brk
) {
229 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
, &uf
))
234 /* Check against existing mmap mappings. */
235 next
= find_vma(mm
, oldbrk
);
236 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
239 /* Ok, looks good - let it rip. */
240 if (do_brk_flags(oldbrk
, newbrk
-oldbrk
, 0, &uf
) < 0)
245 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
246 up_write(&mm
->mmap_sem
);
247 userfaultfd_unmap_complete(mm
, &uf
);
249 mm_populate(oldbrk
, newbrk
- oldbrk
);
254 up_write(&mm
->mmap_sem
);
258 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
260 unsigned long max
, prev_end
, subtree_gap
;
263 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
264 * allow two stack_guard_gaps between them here, and when choosing
265 * an unmapped area; whereas when expanding we only require one.
266 * That's a little inconsistent, but keeps the code here simpler.
268 max
= vm_start_gap(vma
);
270 prev_end
= vm_end_gap(vma
->vm_prev
);
276 if (vma
->vm_rb
.rb_left
) {
277 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
278 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
279 if (subtree_gap
> max
)
282 if (vma
->vm_rb
.rb_right
) {
283 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
284 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
285 if (subtree_gap
> max
)
291 #ifdef CONFIG_DEBUG_VM_RB
292 static int browse_rb(struct mm_struct
*mm
)
294 struct rb_root
*root
= &mm
->mm_rb
;
295 int i
= 0, j
, bug
= 0;
296 struct rb_node
*nd
, *pn
= NULL
;
297 unsigned long prev
= 0, pend
= 0;
299 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
300 struct vm_area_struct
*vma
;
301 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
302 if (vma
->vm_start
< prev
) {
303 pr_emerg("vm_start %lx < prev %lx\n",
304 vma
->vm_start
, prev
);
307 if (vma
->vm_start
< pend
) {
308 pr_emerg("vm_start %lx < pend %lx\n",
309 vma
->vm_start
, pend
);
312 if (vma
->vm_start
> vma
->vm_end
) {
313 pr_emerg("vm_start %lx > vm_end %lx\n",
314 vma
->vm_start
, vma
->vm_end
);
317 spin_lock(&mm
->page_table_lock
);
318 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
319 pr_emerg("free gap %lx, correct %lx\n",
321 vma_compute_subtree_gap(vma
));
324 spin_unlock(&mm
->page_table_lock
);
327 prev
= vma
->vm_start
;
331 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
334 pr_emerg("backwards %d, forwards %d\n", j
, i
);
340 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
344 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
345 struct vm_area_struct
*vma
;
346 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
347 VM_BUG_ON_VMA(vma
!= ignore
&&
348 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
353 static void validate_mm(struct mm_struct
*mm
)
357 unsigned long highest_address
= 0;
358 struct vm_area_struct
*vma
= mm
->mmap
;
361 struct anon_vma
*anon_vma
= vma
->anon_vma
;
362 struct anon_vma_chain
*avc
;
365 anon_vma_lock_read(anon_vma
);
366 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
367 anon_vma_interval_tree_verify(avc
);
368 anon_vma_unlock_read(anon_vma
);
371 highest_address
= vm_end_gap(vma
);
375 if (i
!= mm
->map_count
) {
376 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
379 if (highest_address
!= mm
->highest_vm_end
) {
380 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
381 mm
->highest_vm_end
, highest_address
);
385 if (i
!= mm
->map_count
) {
387 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
390 VM_BUG_ON_MM(bug
, mm
);
393 #define validate_mm_rb(root, ignore) do { } while (0)
394 #define validate_mm(mm) do { } while (0)
397 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
398 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
401 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
402 * vma->vm_prev->vm_end values changed, without modifying the vma's position
405 static void vma_gap_update(struct vm_area_struct
*vma
)
408 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
409 * function that does exacltly what we want.
411 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
414 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
415 struct rb_root
*root
)
417 /* All rb_subtree_gap values must be consistent prior to insertion */
418 validate_mm_rb(root
, NULL
);
420 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
423 static void __vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
426 * Note rb_erase_augmented is a fairly large inline function,
427 * so make sure we instantiate it only once with our desired
428 * augmented rbtree callbacks.
430 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
433 static __always_inline
void vma_rb_erase_ignore(struct vm_area_struct
*vma
,
434 struct rb_root
*root
,
435 struct vm_area_struct
*ignore
)
438 * All rb_subtree_gap values must be consistent prior to erase,
439 * with the possible exception of the "next" vma being erased if
440 * next->vm_start was reduced.
442 validate_mm_rb(root
, ignore
);
444 __vma_rb_erase(vma
, root
);
447 static __always_inline
void vma_rb_erase(struct vm_area_struct
*vma
,
448 struct rb_root
*root
)
451 * All rb_subtree_gap values must be consistent prior to erase,
452 * with the possible exception of the vma being erased.
454 validate_mm_rb(root
, vma
);
456 __vma_rb_erase(vma
, root
);
460 * vma has some anon_vma assigned, and is already inserted on that
461 * anon_vma's interval trees.
463 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
464 * vma must be removed from the anon_vma's interval trees using
465 * anon_vma_interval_tree_pre_update_vma().
467 * After the update, the vma will be reinserted using
468 * anon_vma_interval_tree_post_update_vma().
470 * The entire update must be protected by exclusive mmap_sem and by
471 * the root anon_vma's mutex.
474 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
476 struct anon_vma_chain
*avc
;
478 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
479 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
483 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
485 struct anon_vma_chain
*avc
;
487 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
488 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
491 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
492 unsigned long end
, struct vm_area_struct
**pprev
,
493 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
495 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
497 __rb_link
= &mm
->mm_rb
.rb_node
;
498 rb_prev
= __rb_parent
= NULL
;
501 struct vm_area_struct
*vma_tmp
;
503 __rb_parent
= *__rb_link
;
504 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
506 if (vma_tmp
->vm_end
> addr
) {
507 /* Fail if an existing vma overlaps the area */
508 if (vma_tmp
->vm_start
< end
)
510 __rb_link
= &__rb_parent
->rb_left
;
512 rb_prev
= __rb_parent
;
513 __rb_link
= &__rb_parent
->rb_right
;
519 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
520 *rb_link
= __rb_link
;
521 *rb_parent
= __rb_parent
;
525 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
526 unsigned long addr
, unsigned long end
)
528 unsigned long nr_pages
= 0;
529 struct vm_area_struct
*vma
;
531 /* Find first overlaping mapping */
532 vma
= find_vma_intersection(mm
, addr
, end
);
536 nr_pages
= (min(end
, vma
->vm_end
) -
537 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
539 /* Iterate over the rest of the overlaps */
540 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
541 unsigned long overlap_len
;
543 if (vma
->vm_start
> end
)
546 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
547 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
553 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
554 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
556 /* Update tracking information for the gap following the new vma. */
558 vma_gap_update(vma
->vm_next
);
560 mm
->highest_vm_end
= vm_end_gap(vma
);
563 * vma->vm_prev wasn't known when we followed the rbtree to find the
564 * correct insertion point for that vma. As a result, we could not
565 * update the vma vm_rb parents rb_subtree_gap values on the way down.
566 * So, we first insert the vma with a zero rb_subtree_gap value
567 * (to be consistent with what we did on the way down), and then
568 * immediately update the gap to the correct value. Finally we
569 * rebalance the rbtree after all augmented values have been set.
571 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
572 vma
->rb_subtree_gap
= 0;
574 vma_rb_insert(vma
, &mm
->mm_rb
);
577 static void __vma_link_file(struct vm_area_struct
*vma
)
583 struct address_space
*mapping
= file
->f_mapping
;
585 if (vma
->vm_flags
& VM_DENYWRITE
)
586 atomic_dec(&file_inode(file
)->i_writecount
);
587 if (vma
->vm_flags
& VM_SHARED
)
588 atomic_inc(&mapping
->i_mmap_writable
);
590 flush_dcache_mmap_lock(mapping
);
591 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
592 flush_dcache_mmap_unlock(mapping
);
597 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
598 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
599 struct rb_node
*rb_parent
)
601 __vma_link_list(mm
, vma
, prev
, rb_parent
);
602 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
605 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
606 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
607 struct rb_node
*rb_parent
)
609 struct address_space
*mapping
= NULL
;
612 mapping
= vma
->vm_file
->f_mapping
;
613 i_mmap_lock_write(mapping
);
616 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
617 __vma_link_file(vma
);
620 i_mmap_unlock_write(mapping
);
627 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
628 * mm's list and rbtree. It has already been inserted into the interval tree.
630 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
632 struct vm_area_struct
*prev
;
633 struct rb_node
**rb_link
, *rb_parent
;
635 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
636 &prev
, &rb_link
, &rb_parent
))
638 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
642 static __always_inline
void __vma_unlink_common(struct mm_struct
*mm
,
643 struct vm_area_struct
*vma
,
644 struct vm_area_struct
*prev
,
646 struct vm_area_struct
*ignore
)
648 struct vm_area_struct
*next
;
650 vma_rb_erase_ignore(vma
, &mm
->mm_rb
, ignore
);
653 prev
->vm_next
= next
;
657 prev
->vm_next
= next
;
662 next
->vm_prev
= prev
;
665 vmacache_invalidate(mm
);
668 static inline void __vma_unlink_prev(struct mm_struct
*mm
,
669 struct vm_area_struct
*vma
,
670 struct vm_area_struct
*prev
)
672 __vma_unlink_common(mm
, vma
, prev
, true, vma
);
676 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
677 * is already present in an i_mmap tree without adjusting the tree.
678 * The following helper function should be used when such adjustments
679 * are necessary. The "insert" vma (if any) is to be inserted
680 * before we drop the necessary locks.
682 int __vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
683 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
,
684 struct vm_area_struct
*expand
)
686 struct mm_struct
*mm
= vma
->vm_mm
;
687 struct vm_area_struct
*next
= vma
->vm_next
, *orig_vma
= vma
;
688 struct address_space
*mapping
= NULL
;
689 struct rb_root_cached
*root
= NULL
;
690 struct anon_vma
*anon_vma
= NULL
;
691 struct file
*file
= vma
->vm_file
;
692 bool start_changed
= false, end_changed
= false;
693 long adjust_next
= 0;
696 if (next
&& !insert
) {
697 struct vm_area_struct
*exporter
= NULL
, *importer
= NULL
;
699 if (end
>= next
->vm_end
) {
701 * vma expands, overlapping all the next, and
702 * perhaps the one after too (mprotect case 6).
703 * The only other cases that gets here are
704 * case 1, case 7 and case 8.
706 if (next
== expand
) {
708 * The only case where we don't expand "vma"
709 * and we expand "next" instead is case 8.
711 VM_WARN_ON(end
!= next
->vm_end
);
713 * remove_next == 3 means we're
714 * removing "vma" and that to do so we
715 * swapped "vma" and "next".
718 VM_WARN_ON(file
!= next
->vm_file
);
721 VM_WARN_ON(expand
!= vma
);
723 * case 1, 6, 7, remove_next == 2 is case 6,
724 * remove_next == 1 is case 1 or 7.
726 remove_next
= 1 + (end
> next
->vm_end
);
727 VM_WARN_ON(remove_next
== 2 &&
728 end
!= next
->vm_next
->vm_end
);
729 VM_WARN_ON(remove_next
== 1 &&
730 end
!= next
->vm_end
);
731 /* trim end to next, for case 6 first pass */
739 * If next doesn't have anon_vma, import from vma after
740 * next, if the vma overlaps with it.
742 if (remove_next
== 2 && !next
->anon_vma
)
743 exporter
= next
->vm_next
;
745 } else if (end
> next
->vm_start
) {
747 * vma expands, overlapping part of the next:
748 * mprotect case 5 shifting the boundary up.
750 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
753 VM_WARN_ON(expand
!= importer
);
754 } else if (end
< vma
->vm_end
) {
756 * vma shrinks, and !insert tells it's not
757 * split_vma inserting another: so it must be
758 * mprotect case 4 shifting the boundary down.
760 adjust_next
= -((vma
->vm_end
- end
) >> PAGE_SHIFT
);
763 VM_WARN_ON(expand
!= importer
);
767 * Easily overlooked: when mprotect shifts the boundary,
768 * make sure the expanding vma has anon_vma set if the
769 * shrinking vma had, to cover any anon pages imported.
771 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
774 importer
->anon_vma
= exporter
->anon_vma
;
775 error
= anon_vma_clone(importer
, exporter
);
781 vma_adjust_trans_huge(orig_vma
, start
, end
, adjust_next
);
784 mapping
= file
->f_mapping
;
785 root
= &mapping
->i_mmap
;
786 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
789 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
791 i_mmap_lock_write(mapping
);
794 * Put into interval tree now, so instantiated pages
795 * are visible to arm/parisc __flush_dcache_page
796 * throughout; but we cannot insert into address
797 * space until vma start or end is updated.
799 __vma_link_file(insert
);
803 anon_vma
= vma
->anon_vma
;
804 if (!anon_vma
&& adjust_next
)
805 anon_vma
= next
->anon_vma
;
807 VM_WARN_ON(adjust_next
&& next
->anon_vma
&&
808 anon_vma
!= next
->anon_vma
);
809 anon_vma_lock_write(anon_vma
);
810 anon_vma_interval_tree_pre_update_vma(vma
);
812 anon_vma_interval_tree_pre_update_vma(next
);
816 flush_dcache_mmap_lock(mapping
);
817 vma_interval_tree_remove(vma
, root
);
819 vma_interval_tree_remove(next
, root
);
822 if (start
!= vma
->vm_start
) {
823 vma
->vm_start
= start
;
824 start_changed
= true;
826 if (end
!= vma
->vm_end
) {
830 vma
->vm_pgoff
= pgoff
;
832 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
833 next
->vm_pgoff
+= adjust_next
;
838 vma_interval_tree_insert(next
, root
);
839 vma_interval_tree_insert(vma
, root
);
840 flush_dcache_mmap_unlock(mapping
);
845 * vma_merge has merged next into vma, and needs
846 * us to remove next before dropping the locks.
848 if (remove_next
!= 3)
849 __vma_unlink_prev(mm
, next
, vma
);
852 * vma is not before next if they've been
855 * pre-swap() next->vm_start was reduced so
856 * tell validate_mm_rb to ignore pre-swap()
857 * "next" (which is stored in post-swap()
860 __vma_unlink_common(mm
, next
, NULL
, false, vma
);
862 __remove_shared_vm_struct(next
, file
, mapping
);
865 * split_vma has split insert from vma, and needs
866 * us to insert it before dropping the locks
867 * (it may either follow vma or precede it).
869 __insert_vm_struct(mm
, insert
);
875 mm
->highest_vm_end
= vm_end_gap(vma
);
876 else if (!adjust_next
)
877 vma_gap_update(next
);
882 anon_vma_interval_tree_post_update_vma(vma
);
884 anon_vma_interval_tree_post_update_vma(next
);
885 anon_vma_unlock_write(anon_vma
);
888 i_mmap_unlock_write(mapping
);
899 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
903 anon_vma_merge(vma
, next
);
905 mpol_put(vma_policy(next
));
906 kmem_cache_free(vm_area_cachep
, next
);
908 * In mprotect's case 6 (see comments on vma_merge),
909 * we must remove another next too. It would clutter
910 * up the code too much to do both in one go.
912 if (remove_next
!= 3) {
914 * If "next" was removed and vma->vm_end was
915 * expanded (up) over it, in turn
916 * "next->vm_prev->vm_end" changed and the
917 * "vma->vm_next" gap must be updated.
922 * For the scope of the comment "next" and
923 * "vma" considered pre-swap(): if "vma" was
924 * removed, next->vm_start was expanded (down)
925 * over it and the "next" gap must be updated.
926 * Because of the swap() the post-swap() "vma"
927 * actually points to pre-swap() "next"
928 * (post-swap() "next" as opposed is now a
933 if (remove_next
== 2) {
939 vma_gap_update(next
);
942 * If remove_next == 2 we obviously can't
945 * If remove_next == 3 we can't reach this
946 * path because pre-swap() next is always not
947 * NULL. pre-swap() "next" is not being
948 * removed and its next->vm_end is not altered
949 * (and furthermore "end" already matches
950 * next->vm_end in remove_next == 3).
952 * We reach this only in the remove_next == 1
953 * case if the "next" vma that was removed was
954 * the highest vma of the mm. However in such
955 * case next->vm_end == "end" and the extended
956 * "vma" has vma->vm_end == next->vm_end so
957 * mm->highest_vm_end doesn't need any update
958 * in remove_next == 1 case.
960 VM_WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
972 * If the vma has a ->close operation then the driver probably needs to release
973 * per-vma resources, so we don't attempt to merge those.
975 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
976 struct file
*file
, unsigned long vm_flags
,
977 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
978 const char __user
*anon_name
)
981 * VM_SOFTDIRTY should not prevent from VMA merging, if we
982 * match the flags but dirty bit -- the caller should mark
983 * merged VMA as dirty. If dirty bit won't be excluded from
984 * comparison, we increase pressue on the memory system forcing
985 * the kernel to generate new VMAs when old one could be
988 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
990 if (vma
->vm_file
!= file
)
992 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
994 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
996 if (vma_get_anon_name(vma
) != anon_name
)
1001 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
1002 struct anon_vma
*anon_vma2
,
1003 struct vm_area_struct
*vma
)
1006 * The list_is_singular() test is to avoid merging VMA cloned from
1007 * parents. This can improve scalability caused by anon_vma lock.
1009 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
1010 list_is_singular(&vma
->anon_vma_chain
)))
1012 return anon_vma1
== anon_vma2
;
1016 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1017 * in front of (at a lower virtual address and file offset than) the vma.
1019 * We cannot merge two vmas if they have differently assigned (non-NULL)
1020 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1022 * We don't check here for the merged mmap wrapping around the end of pagecache
1023 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1024 * wrap, nor mmaps which cover the final page at index -1UL.
1027 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1028 struct anon_vma
*anon_vma
, struct file
*file
,
1030 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
1031 const char __user
*anon_name
)
1033 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
) &&
1034 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1035 if (vma
->vm_pgoff
== vm_pgoff
)
1042 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1043 * beyond (at a higher virtual address and file offset than) the vma.
1045 * We cannot merge two vmas if they have differently assigned (non-NULL)
1046 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1049 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1050 struct anon_vma
*anon_vma
, struct file
*file
,
1052 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
1053 const char __user
*anon_name
)
1055 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
) &&
1056 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1058 vm_pglen
= vma_pages(vma
);
1059 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1066 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
1067 * figure out whether that can be merged with its predecessor or its
1068 * successor. Or both (it neatly fills a hole).
1070 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1071 * certain not to be mapped by the time vma_merge is called; but when
1072 * called for mprotect, it is certain to be already mapped (either at
1073 * an offset within prev, or at the start of next), and the flags of
1074 * this area are about to be changed to vm_flags - and the no-change
1075 * case has already been eliminated.
1077 * The following mprotect cases have to be considered, where AAAA is
1078 * the area passed down from mprotect_fixup, never extending beyond one
1079 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1081 * AAAA AAAA AAAA AAAA
1082 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1083 * cannot merge might become might become might become
1084 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1085 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1086 * mremap move: PPPPXXXXXXXX 8
1088 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1089 * might become case 1 below case 2 below case 3 below
1091 * It is important for case 8 that the the vma NNNN overlapping the
1092 * region AAAA is never going to extended over XXXX. Instead XXXX must
1093 * be extended in region AAAA and NNNN must be removed. This way in
1094 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1095 * rmap_locks, the properties of the merged vma will be already
1096 * correct for the whole merged range. Some of those properties like
1097 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1098 * be correct for the whole merged range immediately after the
1099 * rmap_locks are released. Otherwise if XXXX would be removed and
1100 * NNNN would be extended over the XXXX range, remove_migration_ptes
1101 * or other rmap walkers (if working on addresses beyond the "end"
1102 * parameter) may establish ptes with the wrong permissions of NNNN
1103 * instead of the right permissions of XXXX.
1105 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1106 struct vm_area_struct
*prev
, unsigned long addr
,
1107 unsigned long end
, unsigned long vm_flags
,
1108 struct anon_vma
*anon_vma
, struct file
*file
,
1109 pgoff_t pgoff
, struct mempolicy
*policy
,
1110 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
1111 const char __user
*anon_name
)
1113 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1114 struct vm_area_struct
*area
, *next
;
1118 * We later require that vma->vm_flags == vm_flags,
1119 * so this tests vma->vm_flags & VM_SPECIAL, too.
1121 if (vm_flags
& VM_SPECIAL
)
1125 next
= prev
->vm_next
;
1129 if (area
&& area
->vm_end
== end
) /* cases 6, 7, 8 */
1130 next
= next
->vm_next
;
1132 /* verify some invariant that must be enforced by the caller */
1133 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
1134 VM_WARN_ON(area
&& end
> area
->vm_end
);
1135 VM_WARN_ON(addr
>= end
);
1138 * Can it merge with the predecessor?
1140 if (prev
&& prev
->vm_end
== addr
&&
1141 mpol_equal(vma_policy(prev
), policy
) &&
1142 can_vma_merge_after(prev
, vm_flags
,
1143 anon_vma
, file
, pgoff
,
1147 * OK, it can. Can we now merge in the successor as well?
1149 if (next
&& end
== next
->vm_start
&&
1150 mpol_equal(policy
, vma_policy(next
)) &&
1151 can_vma_merge_before(next
, vm_flags
,
1156 is_mergeable_anon_vma(prev
->anon_vma
,
1157 next
->anon_vma
, NULL
)) {
1159 err
= __vma_adjust(prev
, prev
->vm_start
,
1160 next
->vm_end
, prev
->vm_pgoff
, NULL
,
1162 } else /* cases 2, 5, 7 */
1163 err
= __vma_adjust(prev
, prev
->vm_start
,
1164 end
, prev
->vm_pgoff
, NULL
, prev
);
1167 khugepaged_enter_vma_merge(prev
, vm_flags
);
1172 * Can this new request be merged in front of next?
1174 if (next
&& end
== next
->vm_start
&&
1175 mpol_equal(policy
, vma_policy(next
)) &&
1176 can_vma_merge_before(next
, vm_flags
,
1177 anon_vma
, file
, pgoff
+pglen
,
1180 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1181 err
= __vma_adjust(prev
, prev
->vm_start
,
1182 addr
, prev
->vm_pgoff
, NULL
, next
);
1183 else { /* cases 3, 8 */
1184 err
= __vma_adjust(area
, addr
, next
->vm_end
,
1185 next
->vm_pgoff
- pglen
, NULL
, next
);
1187 * In case 3 area is already equal to next and
1188 * this is a noop, but in case 8 "area" has
1189 * been removed and next was expanded over it.
1195 khugepaged_enter_vma_merge(area
, vm_flags
);
1203 * Rough compatbility check to quickly see if it's even worth looking
1204 * at sharing an anon_vma.
1206 * They need to have the same vm_file, and the flags can only differ
1207 * in things that mprotect may change.
1209 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1210 * we can merge the two vma's. For example, we refuse to merge a vma if
1211 * there is a vm_ops->close() function, because that indicates that the
1212 * driver is doing some kind of reference counting. But that doesn't
1213 * really matter for the anon_vma sharing case.
1215 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1217 return a
->vm_end
== b
->vm_start
&&
1218 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1219 a
->vm_file
== b
->vm_file
&&
1220 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
|VM_SOFTDIRTY
)) &&
1221 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1225 * Do some basic sanity checking to see if we can re-use the anon_vma
1226 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1227 * the same as 'old', the other will be the new one that is trying
1228 * to share the anon_vma.
1230 * NOTE! This runs with mm_sem held for reading, so it is possible that
1231 * the anon_vma of 'old' is concurrently in the process of being set up
1232 * by another page fault trying to merge _that_. But that's ok: if it
1233 * is being set up, that automatically means that it will be a singleton
1234 * acceptable for merging, so we can do all of this optimistically. But
1235 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1237 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1238 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1239 * is to return an anon_vma that is "complex" due to having gone through
1242 * We also make sure that the two vma's are compatible (adjacent,
1243 * and with the same memory policies). That's all stable, even with just
1244 * a read lock on the mm_sem.
1246 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1248 if (anon_vma_compatible(a
, b
)) {
1249 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1251 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1258 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1259 * neighbouring vmas for a suitable anon_vma, before it goes off
1260 * to allocate a new anon_vma. It checks because a repetitive
1261 * sequence of mprotects and faults may otherwise lead to distinct
1262 * anon_vmas being allocated, preventing vma merge in subsequent
1265 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1267 struct anon_vma
*anon_vma
;
1268 struct vm_area_struct
*near
;
1270 near
= vma
->vm_next
;
1274 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1278 near
= vma
->vm_prev
;
1282 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1287 * There's no absolute need to look only at touching neighbours:
1288 * we could search further afield for "compatible" anon_vmas.
1289 * But it would probably just be a waste of time searching,
1290 * or lead to too many vmas hanging off the same anon_vma.
1291 * We're trying to allow mprotect remerging later on,
1292 * not trying to minimize memory used for anon_vmas.
1298 * If a hint addr is less than mmap_min_addr change hint to be as
1299 * low as possible but still greater than mmap_min_addr
1301 static inline unsigned long round_hint_to_min(unsigned long hint
)
1304 if (((void *)hint
!= NULL
) &&
1305 (hint
< mmap_min_addr
))
1306 return PAGE_ALIGN(mmap_min_addr
);
1310 static inline int mlock_future_check(struct mm_struct
*mm
,
1311 unsigned long flags
,
1314 unsigned long locked
, lock_limit
;
1316 /* mlock MCL_FUTURE? */
1317 if (flags
& VM_LOCKED
) {
1318 locked
= len
>> PAGE_SHIFT
;
1319 locked
+= mm
->locked_vm
;
1320 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1321 lock_limit
>>= PAGE_SHIFT
;
1322 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1328 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1330 if (S_ISREG(inode
->i_mode
))
1331 return MAX_LFS_FILESIZE
;
1333 if (S_ISBLK(inode
->i_mode
))
1334 return MAX_LFS_FILESIZE
;
1336 /* Special "we do even unsigned file positions" case */
1337 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1340 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1344 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1345 unsigned long pgoff
, unsigned long len
)
1347 u64 maxsize
= file_mmap_size_max(file
, inode
);
1349 if (maxsize
&& len
> maxsize
)
1352 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1358 * The caller must hold down_write(¤t->mm->mmap_sem).
1360 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1361 unsigned long len
, unsigned long prot
,
1362 unsigned long flags
, vm_flags_t vm_flags
,
1363 unsigned long pgoff
, unsigned long *populate
,
1364 struct list_head
*uf
)
1366 struct mm_struct
*mm
= current
->mm
;
1375 * Does the application expect PROT_READ to imply PROT_EXEC?
1377 * (the exception is when the underlying filesystem is noexec
1378 * mounted, in which case we dont add PROT_EXEC.)
1380 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1381 if (!(file
&& path_noexec(&file
->f_path
)))
1384 if (!(flags
& MAP_FIXED
))
1385 addr
= round_hint_to_min(addr
);
1387 /* Careful about overflows.. */
1388 len
= PAGE_ALIGN(len
);
1392 /* offset overflow? */
1393 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1396 /* Too many mappings? */
1397 if (mm
->map_count
> sysctl_max_map_count
)
1400 /* Obtain the address to map to. we verify (or select) it and ensure
1401 * that it represents a valid section of the address space.
1403 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1404 if (offset_in_page(addr
))
1407 if (prot
== PROT_EXEC
) {
1408 pkey
= execute_only_pkey(mm
);
1413 /* Do simple checking here so the lower-level routines won't have
1414 * to. we assume access permissions have been handled by the open
1415 * of the memory object, so we don't do any here.
1417 vm_flags
|= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1418 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1420 if (flags
& MAP_LOCKED
)
1421 if (!can_do_mlock())
1424 if (mlock_future_check(mm
, vm_flags
, len
))
1428 struct inode
*inode
= file_inode(file
);
1430 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1433 switch (flags
& MAP_TYPE
) {
1435 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1439 * Make sure we don't allow writing to an append-only
1442 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1446 * Make sure there are no mandatory locks on the file.
1448 if (locks_verify_locked(file
))
1451 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1452 if (!(file
->f_mode
& FMODE_WRITE
))
1453 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1457 if (!(file
->f_mode
& FMODE_READ
))
1459 if (path_noexec(&file
->f_path
)) {
1460 if (vm_flags
& VM_EXEC
)
1462 vm_flags
&= ~VM_MAYEXEC
;
1465 if (!file
->f_op
->mmap
)
1467 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1475 switch (flags
& MAP_TYPE
) {
1477 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1483 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1487 * Set pgoff according to addr for anon_vma.
1489 pgoff
= addr
>> PAGE_SHIFT
;
1497 * Set 'VM_NORESERVE' if we should not account for the
1498 * memory use of this mapping.
1500 if (flags
& MAP_NORESERVE
) {
1501 /* We honor MAP_NORESERVE if allowed to overcommit */
1502 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1503 vm_flags
|= VM_NORESERVE
;
1505 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1506 if (file
&& is_file_hugepages(file
))
1507 vm_flags
|= VM_NORESERVE
;
1510 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1511 if (!IS_ERR_VALUE(addr
) &&
1512 ((vm_flags
& VM_LOCKED
) ||
1513 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1518 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1519 unsigned long, prot
, unsigned long, flags
,
1520 unsigned long, fd
, unsigned long, pgoff
)
1522 struct file
*file
= NULL
;
1523 unsigned long retval
;
1525 if (!(flags
& MAP_ANONYMOUS
)) {
1526 audit_mmap_fd(fd
, flags
);
1530 if (is_file_hugepages(file
))
1531 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1533 if (unlikely(flags
& MAP_HUGETLB
&& !is_file_hugepages(file
)))
1535 } else if (flags
& MAP_HUGETLB
) {
1536 struct user_struct
*user
= NULL
;
1539 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1543 len
= ALIGN(len
, huge_page_size(hs
));
1545 * VM_NORESERVE is used because the reservations will be
1546 * taken when vm_ops->mmap() is called
1547 * A dummy user value is used because we are not locking
1548 * memory so no accounting is necessary
1550 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1552 &user
, HUGETLB_ANONHUGE_INODE
,
1553 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1555 return PTR_ERR(file
);
1558 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1560 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1567 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1568 struct mmap_arg_struct
{
1572 unsigned long flags
;
1574 unsigned long offset
;
1577 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1579 struct mmap_arg_struct a
;
1581 if (copy_from_user(&a
, arg
, sizeof(a
)))
1583 if (offset_in_page(a
.offset
))
1586 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1587 a
.offset
>> PAGE_SHIFT
);
1589 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1592 * Some shared mappigns will want the pages marked read-only
1593 * to track write events. If so, we'll downgrade vm_page_prot
1594 * to the private version (using protection_map[] without the
1597 int vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1599 vm_flags_t vm_flags
= vma
->vm_flags
;
1600 const struct vm_operations_struct
*vm_ops
= vma
->vm_ops
;
1602 /* If it was private or non-writable, the write bit is already clear */
1603 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1606 /* The backer wishes to know when pages are first written to? */
1607 if (vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
))
1610 /* The open routine did something to the protections that pgprot_modify
1611 * won't preserve? */
1612 if (pgprot_val(vm_page_prot
) !=
1613 pgprot_val(vm_pgprot_modify(vm_page_prot
, vm_flags
)))
1616 /* Do we need to track softdirty? */
1617 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1620 /* Specialty mapping? */
1621 if (vm_flags
& VM_PFNMAP
)
1624 /* Can the mapping track the dirty pages? */
1625 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1626 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1630 * We account for memory if it's a private writeable mapping,
1631 * not hugepages and VM_NORESERVE wasn't set.
1633 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1636 * hugetlb has its own accounting separate from the core VM
1637 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1639 if (file
&& is_file_hugepages(file
))
1642 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1645 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1646 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
1647 struct list_head
*uf
)
1649 struct mm_struct
*mm
= current
->mm
;
1650 struct vm_area_struct
*vma
, *prev
;
1652 struct rb_node
**rb_link
, *rb_parent
;
1653 unsigned long charged
= 0;
1655 /* Check against address space limit. */
1656 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
1657 unsigned long nr_pages
;
1660 * MAP_FIXED may remove pages of mappings that intersects with
1661 * requested mapping. Account for the pages it would unmap.
1663 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1665 if (!may_expand_vm(mm
, vm_flags
,
1666 (len
>> PAGE_SHIFT
) - nr_pages
))
1670 /* Clear old maps */
1671 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
1673 if (do_munmap(mm
, addr
, len
, uf
))
1678 * Private writable mapping: check memory availability
1680 if (accountable_mapping(file
, vm_flags
)) {
1681 charged
= len
>> PAGE_SHIFT
;
1682 if (security_vm_enough_memory_mm(mm
, charged
))
1684 vm_flags
|= VM_ACCOUNT
;
1688 * Can we just expand an old mapping?
1690 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1691 NULL
, file
, pgoff
, NULL
, NULL_VM_UFFD_CTX
, NULL
);
1696 * Determine the object being mapped and call the appropriate
1697 * specific mapper. the address has already been validated, but
1698 * not unmapped, but the maps are removed from the list.
1700 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1707 vma
->vm_start
= addr
;
1708 vma
->vm_end
= addr
+ len
;
1709 vma
->vm_flags
= vm_flags
;
1710 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1711 vma
->vm_pgoff
= pgoff
;
1712 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1715 if (vm_flags
& VM_DENYWRITE
) {
1716 error
= deny_write_access(file
);
1720 if (vm_flags
& VM_SHARED
) {
1721 error
= mapping_map_writable(file
->f_mapping
);
1723 goto allow_write_and_free_vma
;
1726 /* ->mmap() can change vma->vm_file, but must guarantee that
1727 * vma_link() below can deny write-access if VM_DENYWRITE is set
1728 * and map writably if VM_SHARED is set. This usually means the
1729 * new file must not have been exposed to user-space, yet.
1731 vma
->vm_file
= get_file(file
);
1732 error
= call_mmap(file
, vma
);
1734 goto unmap_and_free_vma
;
1736 /* Can addr have changed??
1738 * Answer: Yes, several device drivers can do it in their
1739 * f_op->mmap method. -DaveM
1740 * Bug: If addr is changed, prev, rb_link, rb_parent should
1741 * be updated for vma_link()
1743 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1745 addr
= vma
->vm_start
;
1746 vm_flags
= vma
->vm_flags
;
1747 } else if (vm_flags
& VM_SHARED
) {
1748 error
= shmem_zero_setup(vma
);
1753 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1754 /* Once vma denies write, undo our temporary denial count */
1756 if (vm_flags
& VM_SHARED
)
1757 mapping_unmap_writable(file
->f_mapping
);
1758 if (vm_flags
& VM_DENYWRITE
)
1759 allow_write_access(file
);
1761 file
= vma
->vm_file
;
1763 perf_event_mmap(vma
);
1765 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
1766 if (vm_flags
& VM_LOCKED
) {
1767 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1768 vma
== get_gate_vma(current
->mm
)))
1769 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1771 vma
->vm_flags
&= VM_LOCKED_CLEAR_MASK
;
1778 * New (or expanded) vma always get soft dirty status.
1779 * Otherwise user-space soft-dirty page tracker won't
1780 * be able to distinguish situation when vma area unmapped,
1781 * then new mapped in-place (which must be aimed as
1782 * a completely new data area).
1784 vma
->vm_flags
|= VM_SOFTDIRTY
;
1786 vma_set_page_prot(vma
);
1791 vma
->vm_file
= NULL
;
1794 /* Undo any partial mapping done by a device driver. */
1795 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1797 if (vm_flags
& VM_SHARED
)
1798 mapping_unmap_writable(file
->f_mapping
);
1799 allow_write_and_free_vma
:
1800 if (vm_flags
& VM_DENYWRITE
)
1801 allow_write_access(file
);
1803 kmem_cache_free(vm_area_cachep
, vma
);
1806 vm_unacct_memory(charged
);
1810 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1813 * We implement the search by looking for an rbtree node that
1814 * immediately follows a suitable gap. That is,
1815 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1816 * - gap_end = vma->vm_start >= info->low_limit + length;
1817 * - gap_end - gap_start >= length
1820 struct mm_struct
*mm
= current
->mm
;
1821 struct vm_area_struct
*vma
;
1822 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1824 /* Adjust search length to account for worst case alignment overhead */
1825 length
= info
->length
+ info
->align_mask
;
1826 if (length
< info
->length
)
1829 /* Adjust search limits by the desired length */
1830 if (info
->high_limit
< length
)
1832 high_limit
= info
->high_limit
- length
;
1834 if (info
->low_limit
> high_limit
)
1836 low_limit
= info
->low_limit
+ length
;
1838 /* Check if rbtree root looks promising */
1839 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1841 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1842 if (vma
->rb_subtree_gap
< length
)
1846 /* Visit left subtree if it looks promising */
1847 gap_end
= vm_start_gap(vma
);
1848 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1849 struct vm_area_struct
*left
=
1850 rb_entry(vma
->vm_rb
.rb_left
,
1851 struct vm_area_struct
, vm_rb
);
1852 if (left
->rb_subtree_gap
>= length
) {
1858 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1860 /* Check if current node has a suitable gap */
1861 if (gap_start
> high_limit
)
1863 if (gap_end
>= low_limit
&&
1864 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1867 /* Visit right subtree if it looks promising */
1868 if (vma
->vm_rb
.rb_right
) {
1869 struct vm_area_struct
*right
=
1870 rb_entry(vma
->vm_rb
.rb_right
,
1871 struct vm_area_struct
, vm_rb
);
1872 if (right
->rb_subtree_gap
>= length
) {
1878 /* Go back up the rbtree to find next candidate node */
1880 struct rb_node
*prev
= &vma
->vm_rb
;
1881 if (!rb_parent(prev
))
1883 vma
= rb_entry(rb_parent(prev
),
1884 struct vm_area_struct
, vm_rb
);
1885 if (prev
== vma
->vm_rb
.rb_left
) {
1886 gap_start
= vm_end_gap(vma
->vm_prev
);
1887 gap_end
= vm_start_gap(vma
);
1894 /* Check highest gap, which does not precede any rbtree node */
1895 gap_start
= mm
->highest_vm_end
;
1896 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1897 if (gap_start
> high_limit
)
1901 /* We found a suitable gap. Clip it with the original low_limit. */
1902 if (gap_start
< info
->low_limit
)
1903 gap_start
= info
->low_limit
;
1905 /* Adjust gap address to the desired alignment */
1906 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1908 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1909 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1913 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1915 struct mm_struct
*mm
= current
->mm
;
1916 struct vm_area_struct
*vma
;
1917 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1919 /* Adjust search length to account for worst case alignment overhead */
1920 length
= info
->length
+ info
->align_mask
;
1921 if (length
< info
->length
)
1925 * Adjust search limits by the desired length.
1926 * See implementation comment at top of unmapped_area().
1928 gap_end
= info
->high_limit
;
1929 if (gap_end
< length
)
1931 high_limit
= gap_end
- length
;
1933 if (info
->low_limit
> high_limit
)
1935 low_limit
= info
->low_limit
+ length
;
1937 /* Check highest gap, which does not precede any rbtree node */
1938 gap_start
= mm
->highest_vm_end
;
1939 if (gap_start
<= high_limit
)
1942 /* Check if rbtree root looks promising */
1943 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1945 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1946 if (vma
->rb_subtree_gap
< length
)
1950 /* Visit right subtree if it looks promising */
1951 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1952 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1953 struct vm_area_struct
*right
=
1954 rb_entry(vma
->vm_rb
.rb_right
,
1955 struct vm_area_struct
, vm_rb
);
1956 if (right
->rb_subtree_gap
>= length
) {
1963 /* Check if current node has a suitable gap */
1964 gap_end
= vm_start_gap(vma
);
1965 if (gap_end
< low_limit
)
1967 if (gap_start
<= high_limit
&&
1968 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1971 /* Visit left subtree if it looks promising */
1972 if (vma
->vm_rb
.rb_left
) {
1973 struct vm_area_struct
*left
=
1974 rb_entry(vma
->vm_rb
.rb_left
,
1975 struct vm_area_struct
, vm_rb
);
1976 if (left
->rb_subtree_gap
>= length
) {
1982 /* Go back up the rbtree to find next candidate node */
1984 struct rb_node
*prev
= &vma
->vm_rb
;
1985 if (!rb_parent(prev
))
1987 vma
= rb_entry(rb_parent(prev
),
1988 struct vm_area_struct
, vm_rb
);
1989 if (prev
== vma
->vm_rb
.rb_right
) {
1990 gap_start
= vma
->vm_prev
?
1991 vm_end_gap(vma
->vm_prev
) : 0;
1998 /* We found a suitable gap. Clip it with the original high_limit. */
1999 if (gap_end
> info
->high_limit
)
2000 gap_end
= info
->high_limit
;
2003 /* Compute highest gap address at the desired alignment */
2004 gap_end
-= info
->length
;
2005 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
2007 VM_BUG_ON(gap_end
< info
->low_limit
);
2008 VM_BUG_ON(gap_end
< gap_start
);
2012 /* Get an address range which is currently unmapped.
2013 * For shmat() with addr=0.
2015 * Ugly calling convention alert:
2016 * Return value with the low bits set means error value,
2018 * if (ret & ~PAGE_MASK)
2021 * This function "knows" that -ENOMEM has the bits set.
2023 #ifndef HAVE_ARCH_UNMAPPED_AREA
2025 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
2026 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
2028 struct mm_struct
*mm
= current
->mm
;
2029 struct vm_area_struct
*vma
, *prev
;
2030 struct vm_unmapped_area_info info
;
2032 if (len
> TASK_SIZE
- mmap_min_addr
)
2035 if (flags
& MAP_FIXED
)
2039 addr
= PAGE_ALIGN(addr
);
2040 vma
= find_vma_prev(mm
, addr
, &prev
);
2041 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
2042 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2043 (!prev
|| addr
>= vm_end_gap(prev
)))
2049 info
.low_limit
= mm
->mmap_base
;
2050 info
.high_limit
= TASK_SIZE
;
2051 info
.align_mask
= 0;
2052 return vm_unmapped_area(&info
);
2057 * This mmap-allocator allocates new areas top-down from below the
2058 * stack's low limit (the base):
2060 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2062 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
2063 const unsigned long len
, const unsigned long pgoff
,
2064 const unsigned long flags
)
2066 struct vm_area_struct
*vma
, *prev
;
2067 struct mm_struct
*mm
= current
->mm
;
2068 unsigned long addr
= addr0
;
2069 struct vm_unmapped_area_info info
;
2071 /* requested length too big for entire address space */
2072 if (len
> TASK_SIZE
- mmap_min_addr
)
2075 if (flags
& MAP_FIXED
)
2078 /* requesting a specific address */
2080 addr
= PAGE_ALIGN(addr
);
2081 vma
= find_vma_prev(mm
, addr
, &prev
);
2082 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
2083 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2084 (!prev
|| addr
>= vm_end_gap(prev
)))
2088 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
2090 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
2091 info
.high_limit
= mm
->mmap_base
;
2092 info
.align_mask
= 0;
2093 addr
= vm_unmapped_area(&info
);
2096 * A failed mmap() very likely causes application failure,
2097 * so fall back to the bottom-up function here. This scenario
2098 * can happen with large stack limits and large mmap()
2101 if (offset_in_page(addr
)) {
2102 VM_BUG_ON(addr
!= -ENOMEM
);
2104 info
.low_limit
= TASK_UNMAPPED_BASE
;
2105 info
.high_limit
= TASK_SIZE
;
2106 addr
= vm_unmapped_area(&info
);
2114 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2115 unsigned long pgoff
, unsigned long flags
)
2117 unsigned long (*get_area
)(struct file
*, unsigned long,
2118 unsigned long, unsigned long, unsigned long);
2120 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2124 /* Careful about overflows.. */
2125 if (len
> TASK_SIZE
)
2128 get_area
= current
->mm
->get_unmapped_area
;
2130 if (file
->f_op
->get_unmapped_area
)
2131 get_area
= file
->f_op
->get_unmapped_area
;
2132 } else if (flags
& MAP_SHARED
) {
2134 * mmap_region() will call shmem_zero_setup() to create a file,
2135 * so use shmem's get_unmapped_area in case it can be huge.
2136 * do_mmap_pgoff() will clear pgoff, so match alignment.
2139 get_area
= shmem_get_unmapped_area
;
2142 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2143 if (IS_ERR_VALUE(addr
))
2146 if (addr
> TASK_SIZE
- len
)
2148 if (offset_in_page(addr
))
2151 error
= security_mmap_addr(addr
);
2152 return error
? error
: addr
;
2155 EXPORT_SYMBOL(get_unmapped_area
);
2157 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2158 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2160 struct rb_node
*rb_node
;
2161 struct vm_area_struct
*vma
;
2163 /* Check the cache first. */
2164 vma
= vmacache_find(mm
, addr
);
2168 rb_node
= mm
->mm_rb
.rb_node
;
2171 struct vm_area_struct
*tmp
;
2173 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2175 if (tmp
->vm_end
> addr
) {
2177 if (tmp
->vm_start
<= addr
)
2179 rb_node
= rb_node
->rb_left
;
2181 rb_node
= rb_node
->rb_right
;
2185 vmacache_update(addr
, vma
);
2189 EXPORT_SYMBOL(find_vma
);
2192 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2194 struct vm_area_struct
*
2195 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2196 struct vm_area_struct
**pprev
)
2198 struct vm_area_struct
*vma
;
2200 vma
= find_vma(mm
, addr
);
2202 *pprev
= vma
->vm_prev
;
2204 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2207 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2208 rb_node
= rb_node
->rb_right
;
2215 * Verify that the stack growth is acceptable and
2216 * update accounting. This is shared with both the
2217 * grow-up and grow-down cases.
2219 static int acct_stack_growth(struct vm_area_struct
*vma
,
2220 unsigned long size
, unsigned long grow
)
2222 struct mm_struct
*mm
= vma
->vm_mm
;
2223 unsigned long new_start
;
2225 /* address space limit tests */
2226 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
2229 /* Stack limit test */
2230 if (size
> rlimit(RLIMIT_STACK
))
2233 /* mlock limit tests */
2234 if (vma
->vm_flags
& VM_LOCKED
) {
2235 unsigned long locked
;
2236 unsigned long limit
;
2237 locked
= mm
->locked_vm
+ grow
;
2238 limit
= rlimit(RLIMIT_MEMLOCK
);
2239 limit
>>= PAGE_SHIFT
;
2240 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2244 /* Check to ensure the stack will not grow into a hugetlb-only region */
2245 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2247 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2251 * Overcommit.. This must be the final test, as it will
2252 * update security statistics.
2254 if (security_vm_enough_memory_mm(mm
, grow
))
2260 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2262 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2263 * vma is the last one with address > vma->vm_end. Have to extend vma.
2265 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2267 struct mm_struct
*mm
= vma
->vm_mm
;
2268 struct vm_area_struct
*next
;
2269 unsigned long gap_addr
;
2272 if (!(vma
->vm_flags
& VM_GROWSUP
))
2275 /* Guard against exceeding limits of the address space. */
2276 address
&= PAGE_MASK
;
2277 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2279 address
+= PAGE_SIZE
;
2281 /* Enforce stack_guard_gap */
2282 gap_addr
= address
+ stack_guard_gap
;
2284 /* Guard against overflow */
2285 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2286 gap_addr
= TASK_SIZE
;
2288 next
= vma
->vm_next
;
2289 if (next
&& next
->vm_start
< gap_addr
&&
2290 (next
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2291 if (!(next
->vm_flags
& VM_GROWSUP
))
2293 /* Check that both stack segments have the same anon_vma? */
2296 /* We must make sure the anon_vma is allocated. */
2297 if (unlikely(anon_vma_prepare(vma
)))
2301 * vma->vm_start/vm_end cannot change under us because the caller
2302 * is required to hold the mmap_sem in read mode. We need the
2303 * anon_vma lock to serialize against concurrent expand_stacks.
2305 anon_vma_lock_write(vma
->anon_vma
);
2307 /* Somebody else might have raced and expanded it already */
2308 if (address
> vma
->vm_end
) {
2309 unsigned long size
, grow
;
2311 size
= address
- vma
->vm_start
;
2312 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2315 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2316 error
= acct_stack_growth(vma
, size
, grow
);
2319 * vma_gap_update() doesn't support concurrent
2320 * updates, but we only hold a shared mmap_sem
2321 * lock here, so we need to protect against
2322 * concurrent vma expansions.
2323 * anon_vma_lock_write() doesn't help here, as
2324 * we don't guarantee that all growable vmas
2325 * in a mm share the same root anon vma.
2326 * So, we reuse mm->page_table_lock to guard
2327 * against concurrent vma expansions.
2329 spin_lock(&mm
->page_table_lock
);
2330 if (vma
->vm_flags
& VM_LOCKED
)
2331 mm
->locked_vm
+= grow
;
2332 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2333 anon_vma_interval_tree_pre_update_vma(vma
);
2334 vma
->vm_end
= address
;
2335 anon_vma_interval_tree_post_update_vma(vma
);
2337 vma_gap_update(vma
->vm_next
);
2339 mm
->highest_vm_end
= vm_end_gap(vma
);
2340 spin_unlock(&mm
->page_table_lock
);
2342 perf_event_mmap(vma
);
2346 anon_vma_unlock_write(vma
->anon_vma
);
2347 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2351 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2354 * vma is the first one with address < vma->vm_start. Have to extend vma.
2356 int expand_downwards(struct vm_area_struct
*vma
,
2357 unsigned long address
)
2359 struct mm_struct
*mm
= vma
->vm_mm
;
2360 struct vm_area_struct
*prev
;
2363 address
&= PAGE_MASK
;
2364 if (address
< mmap_min_addr
)
2367 /* Enforce stack_guard_gap */
2368 prev
= vma
->vm_prev
;
2369 /* Check that both stack segments have the same anon_vma? */
2370 if (prev
&& !(prev
->vm_flags
& VM_GROWSDOWN
) &&
2371 (prev
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2372 if (address
- prev
->vm_end
< stack_guard_gap
)
2376 /* We must make sure the anon_vma is allocated. */
2377 if (unlikely(anon_vma_prepare(vma
)))
2381 * vma->vm_start/vm_end cannot change under us because the caller
2382 * is required to hold the mmap_sem in read mode. We need the
2383 * anon_vma lock to serialize against concurrent expand_stacks.
2385 anon_vma_lock_write(vma
->anon_vma
);
2387 /* Somebody else might have raced and expanded it already */
2388 if (address
< vma
->vm_start
) {
2389 unsigned long size
, grow
;
2391 size
= vma
->vm_end
- address
;
2392 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2395 if (grow
<= vma
->vm_pgoff
) {
2396 error
= acct_stack_growth(vma
, size
, grow
);
2399 * vma_gap_update() doesn't support concurrent
2400 * updates, but we only hold a shared mmap_sem
2401 * lock here, so we need to protect against
2402 * concurrent vma expansions.
2403 * anon_vma_lock_write() doesn't help here, as
2404 * we don't guarantee that all growable vmas
2405 * in a mm share the same root anon vma.
2406 * So, we reuse mm->page_table_lock to guard
2407 * against concurrent vma expansions.
2409 spin_lock(&mm
->page_table_lock
);
2410 if (vma
->vm_flags
& VM_LOCKED
)
2411 mm
->locked_vm
+= grow
;
2412 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2413 anon_vma_interval_tree_pre_update_vma(vma
);
2414 vma
->vm_start
= address
;
2415 vma
->vm_pgoff
-= grow
;
2416 anon_vma_interval_tree_post_update_vma(vma
);
2417 vma_gap_update(vma
);
2418 spin_unlock(&mm
->page_table_lock
);
2420 perf_event_mmap(vma
);
2424 anon_vma_unlock_write(vma
->anon_vma
);
2425 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2430 /* enforced gap between the expanding stack and other mappings. */
2431 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2433 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2438 val
= simple_strtoul(p
, &endptr
, 10);
2440 stack_guard_gap
= val
<< PAGE_SHIFT
;
2444 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2446 #ifdef CONFIG_STACK_GROWSUP
2447 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2449 return expand_upwards(vma
, address
);
2452 struct vm_area_struct
*
2453 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2455 struct vm_area_struct
*vma
, *prev
;
2458 vma
= find_vma_prev(mm
, addr
, &prev
);
2459 if (vma
&& (vma
->vm_start
<= addr
))
2461 /* don't alter vm_end if the coredump is running */
2462 if (!prev
|| !mmget_still_valid(mm
) || expand_stack(prev
, addr
))
2464 if (prev
->vm_flags
& VM_LOCKED
)
2465 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2469 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2471 return expand_downwards(vma
, address
);
2474 struct vm_area_struct
*
2475 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2477 struct vm_area_struct
*vma
;
2478 unsigned long start
;
2481 vma
= find_vma(mm
, addr
);
2484 if (vma
->vm_start
<= addr
)
2486 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2488 /* don't alter vm_start if the coredump is running */
2489 if (!mmget_still_valid(mm
))
2491 start
= vma
->vm_start
;
2492 if (expand_stack(vma
, addr
))
2494 if (vma
->vm_flags
& VM_LOCKED
)
2495 populate_vma_page_range(vma
, addr
, start
, NULL
);
2500 EXPORT_SYMBOL_GPL(find_extend_vma
);
2503 * Ok - we have the memory areas we should free on the vma list,
2504 * so release them, and do the vma updates.
2506 * Called with the mm semaphore held.
2508 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2510 unsigned long nr_accounted
= 0;
2512 /* Update high watermark before we lower total_vm */
2513 update_hiwater_vm(mm
);
2515 long nrpages
= vma_pages(vma
);
2517 if (vma
->vm_flags
& VM_ACCOUNT
)
2518 nr_accounted
+= nrpages
;
2519 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2520 vma
= remove_vma(vma
);
2522 vm_unacct_memory(nr_accounted
);
2527 * Get rid of page table information in the indicated region.
2529 * Called with the mm semaphore held.
2531 static void unmap_region(struct mm_struct
*mm
,
2532 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2533 unsigned long start
, unsigned long end
)
2535 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2536 struct mmu_gather tlb
;
2539 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2540 update_hiwater_rss(mm
);
2541 unmap_vmas(&tlb
, vma
, start
, end
);
2542 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2543 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2544 tlb_finish_mmu(&tlb
, start
, end
);
2548 * Create a list of vma's touched by the unmap, removing them from the mm's
2549 * vma list as we go..
2552 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2553 struct vm_area_struct
*prev
, unsigned long end
)
2555 struct vm_area_struct
**insertion_point
;
2556 struct vm_area_struct
*tail_vma
= NULL
;
2558 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2559 vma
->vm_prev
= NULL
;
2561 vma_rb_erase(vma
, &mm
->mm_rb
);
2565 } while (vma
&& vma
->vm_start
< end
);
2566 *insertion_point
= vma
;
2568 vma
->vm_prev
= prev
;
2569 vma_gap_update(vma
);
2571 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2572 tail_vma
->vm_next
= NULL
;
2574 /* Kill the cache */
2575 vmacache_invalidate(mm
);
2579 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2580 * has already been checked or doesn't make sense to fail.
2582 int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2583 unsigned long addr
, int new_below
)
2585 struct vm_area_struct
*new;
2588 if (vma
->vm_ops
&& vma
->vm_ops
->split
) {
2589 err
= vma
->vm_ops
->split(vma
, addr
);
2594 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2598 /* most fields are the same, copy all, and then fixup */
2601 INIT_LIST_HEAD(&new->anon_vma_chain
);
2606 new->vm_start
= addr
;
2607 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2610 err
= vma_dup_policy(vma
, new);
2614 err
= anon_vma_clone(new, vma
);
2619 get_file(new->vm_file
);
2621 if (new->vm_ops
&& new->vm_ops
->open
)
2622 new->vm_ops
->open(new);
2625 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2626 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2628 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2634 /* Clean everything up if vma_adjust failed. */
2635 if (new->vm_ops
&& new->vm_ops
->close
)
2636 new->vm_ops
->close(new);
2639 unlink_anon_vmas(new);
2641 mpol_put(vma_policy(new));
2643 kmem_cache_free(vm_area_cachep
, new);
2648 * Split a vma into two pieces at address 'addr', a new vma is allocated
2649 * either for the first part or the tail.
2651 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2652 unsigned long addr
, int new_below
)
2654 if (mm
->map_count
>= sysctl_max_map_count
)
2657 return __split_vma(mm
, vma
, addr
, new_below
);
2660 /* Munmap is split into 2 main parts -- this part which finds
2661 * what needs doing, and the areas themselves, which do the
2662 * work. This now handles partial unmappings.
2663 * Jeremy Fitzhardinge <jeremy@goop.org>
2665 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2666 struct list_head
*uf
)
2669 struct vm_area_struct
*vma
, *prev
, *last
;
2671 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2674 len
= PAGE_ALIGN(len
);
2678 /* Find the first overlapping VMA */
2679 vma
= find_vma(mm
, start
);
2682 prev
= vma
->vm_prev
;
2683 /* we have start < vma->vm_end */
2685 /* if it doesn't overlap, we have nothing.. */
2687 if (vma
->vm_start
>= end
)
2691 * If we need to split any vma, do it now to save pain later.
2693 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2694 * unmapped vm_area_struct will remain in use: so lower split_vma
2695 * places tmp vma above, and higher split_vma places tmp vma below.
2697 if (start
> vma
->vm_start
) {
2701 * Make sure that map_count on return from munmap() will
2702 * not exceed its limit; but let map_count go just above
2703 * its limit temporarily, to help free resources as expected.
2705 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2708 error
= __split_vma(mm
, vma
, start
, 0);
2714 /* Does it split the last one? */
2715 last
= find_vma(mm
, end
);
2716 if (last
&& end
> last
->vm_start
) {
2717 int error
= __split_vma(mm
, last
, end
, 1);
2721 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2725 * If userfaultfd_unmap_prep returns an error the vmas
2726 * will remain splitted, but userland will get a
2727 * highly unexpected error anyway. This is no
2728 * different than the case where the first of the two
2729 * __split_vma fails, but we don't undo the first
2730 * split, despite we could. This is unlikely enough
2731 * failure that it's not worth optimizing it for.
2733 int error
= userfaultfd_unmap_prep(vma
, start
, end
, uf
);
2739 * unlock any mlock()ed ranges before detaching vmas
2741 if (mm
->locked_vm
) {
2742 struct vm_area_struct
*tmp
= vma
;
2743 while (tmp
&& tmp
->vm_start
< end
) {
2744 if (tmp
->vm_flags
& VM_LOCKED
) {
2745 mm
->locked_vm
-= vma_pages(tmp
);
2746 munlock_vma_pages_all(tmp
);
2753 * Remove the vma's, and unmap the actual pages
2755 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2756 unmap_region(mm
, vma
, prev
, start
, end
);
2758 arch_unmap(mm
, vma
, start
, end
);
2760 /* Fix up all other VM information */
2761 remove_vma_list(mm
, vma
);
2765 EXPORT_SYMBOL(do_munmap
);
2767 int vm_munmap(unsigned long start
, size_t len
)
2770 struct mm_struct
*mm
= current
->mm
;
2773 if (down_write_killable(&mm
->mmap_sem
))
2776 ret
= do_munmap(mm
, start
, len
, &uf
);
2777 up_write(&mm
->mmap_sem
);
2778 userfaultfd_unmap_complete(mm
, &uf
);
2781 EXPORT_SYMBOL(vm_munmap
);
2783 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2785 profile_munmap(addr
);
2786 return vm_munmap(addr
, len
);
2791 * Emulation of deprecated remap_file_pages() syscall.
2793 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2794 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2797 struct mm_struct
*mm
= current
->mm
;
2798 struct vm_area_struct
*vma
;
2799 unsigned long populate
= 0;
2800 unsigned long ret
= -EINVAL
;
2803 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2804 current
->comm
, current
->pid
);
2808 start
= start
& PAGE_MASK
;
2809 size
= size
& PAGE_MASK
;
2811 if (start
+ size
<= start
)
2814 /* Does pgoff wrap? */
2815 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2818 if (down_write_killable(&mm
->mmap_sem
))
2821 vma
= find_vma(mm
, start
);
2823 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2826 if (start
< vma
->vm_start
)
2829 if (start
+ size
> vma
->vm_end
) {
2830 struct vm_area_struct
*next
;
2832 for (next
= vma
->vm_next
; next
; next
= next
->vm_next
) {
2833 /* hole between vmas ? */
2834 if (next
->vm_start
!= next
->vm_prev
->vm_end
)
2837 if (next
->vm_file
!= vma
->vm_file
)
2840 if (next
->vm_flags
!= vma
->vm_flags
)
2843 if (start
+ size
<= next
->vm_end
)
2851 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
2852 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
2853 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
2855 flags
&= MAP_NONBLOCK
;
2856 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
2857 if (vma
->vm_flags
& VM_LOCKED
) {
2858 struct vm_area_struct
*tmp
;
2859 flags
|= MAP_LOCKED
;
2861 /* drop PG_Mlocked flag for over-mapped range */
2862 for (tmp
= vma
; tmp
->vm_start
>= start
+ size
;
2863 tmp
= tmp
->vm_next
) {
2865 * Split pmd and munlock page on the border
2868 vma_adjust_trans_huge(tmp
, start
, start
+ size
, 0);
2870 munlock_vma_pages_range(tmp
,
2871 max(tmp
->vm_start
, start
),
2872 min(tmp
->vm_end
, start
+ size
));
2876 file
= get_file(vma
->vm_file
);
2877 ret
= do_mmap_pgoff(vma
->vm_file
, start
, size
,
2878 prot
, flags
, pgoff
, &populate
, NULL
);
2881 up_write(&mm
->mmap_sem
);
2883 mm_populate(ret
, populate
);
2884 if (!IS_ERR_VALUE(ret
))
2889 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2891 #ifdef CONFIG_DEBUG_VM
2892 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2894 up_read(&mm
->mmap_sem
);
2900 * this is really a simplified "do_mmap". it only handles
2901 * anonymous maps. eventually we may be able to do some
2902 * brk-specific accounting here.
2904 static int do_brk_flags(unsigned long addr
, unsigned long len
, unsigned long flags
, struct list_head
*uf
)
2906 struct mm_struct
*mm
= current
->mm
;
2907 struct vm_area_struct
*vma
, *prev
;
2908 struct rb_node
**rb_link
, *rb_parent
;
2909 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2912 /* Until we need other flags, refuse anything except VM_EXEC. */
2913 if ((flags
& (~VM_EXEC
)) != 0)
2915 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2917 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2918 if (offset_in_page(error
))
2921 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
2926 * mm->mmap_sem is required to protect against another thread
2927 * changing the mappings in case we sleep.
2929 verify_mm_writelocked(mm
);
2932 * Clear old maps. this also does some error checking for us
2934 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
2936 if (do_munmap(mm
, addr
, len
, uf
))
2940 /* Check against address space limits *after* clearing old maps... */
2941 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
2944 if (mm
->map_count
> sysctl_max_map_count
)
2947 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2950 /* Can we just expand an old private anonymous mapping? */
2951 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2952 NULL
, NULL
, pgoff
, NULL
, NULL_VM_UFFD_CTX
, NULL
);
2957 * create a vma struct for an anonymous mapping
2959 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2961 vm_unacct_memory(len
>> PAGE_SHIFT
);
2965 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2967 vma
->vm_start
= addr
;
2968 vma
->vm_end
= addr
+ len
;
2969 vma
->vm_pgoff
= pgoff
;
2970 vma
->vm_flags
= flags
;
2971 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2972 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2974 perf_event_mmap(vma
);
2975 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2976 mm
->data_vm
+= len
>> PAGE_SHIFT
;
2977 if (flags
& VM_LOCKED
)
2978 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2979 vma
->vm_flags
|= VM_SOFTDIRTY
;
2983 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
2985 struct mm_struct
*mm
= current
->mm
;
2991 len
= PAGE_ALIGN(request
);
2997 if (down_write_killable(&mm
->mmap_sem
))
3000 ret
= do_brk_flags(addr
, len
, flags
, &uf
);
3001 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3002 up_write(&mm
->mmap_sem
);
3003 userfaultfd_unmap_complete(mm
, &uf
);
3004 if (populate
&& !ret
)
3005 mm_populate(addr
, len
);
3008 EXPORT_SYMBOL(vm_brk_flags
);
3010 int vm_brk(unsigned long addr
, unsigned long len
)
3012 return vm_brk_flags(addr
, len
, 0);
3014 EXPORT_SYMBOL(vm_brk
);
3016 /* Release all mmaps. */
3017 void exit_mmap(struct mm_struct
*mm
)
3019 struct mmu_gather tlb
;
3020 struct vm_area_struct
*vma
;
3021 unsigned long nr_accounted
= 0;
3023 /* mm's last user has gone, and its about to be pulled down */
3024 mmu_notifier_release(mm
);
3026 if (unlikely(mm_is_oom_victim(mm
))) {
3028 * Manually reap the mm to free as much memory as possible.
3029 * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3030 * this mm from further consideration. Taking mm->mmap_sem for
3031 * write after setting MMF_OOM_SKIP will guarantee that the oom
3032 * reaper will not run on this mm again after mmap_sem is
3035 * Nothing can be holding mm->mmap_sem here and the above call
3036 * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3037 * __oom_reap_task_mm() will not block.
3039 * This needs to be done before calling munlock_vma_pages_all(),
3040 * which clears VM_LOCKED, otherwise the oom reaper cannot
3043 mutex_lock(&oom_lock
);
3044 __oom_reap_task_mm(mm
);
3045 mutex_unlock(&oom_lock
);
3047 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3048 down_write(&mm
->mmap_sem
);
3049 up_write(&mm
->mmap_sem
);
3052 if (mm
->locked_vm
) {
3055 if (vma
->vm_flags
& VM_LOCKED
)
3056 munlock_vma_pages_all(vma
);
3064 if (!vma
) /* Can happen if dup_mmap() received an OOM */
3069 tlb_gather_mmu(&tlb
, mm
, 0, -1);
3070 /* update_hiwater_rss(mm) here? but nobody should be looking */
3071 /* Use -1 here to ensure all VMAs in the mm are unmapped */
3072 unmap_vmas(&tlb
, vma
, 0, -1);
3073 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
3074 tlb_finish_mmu(&tlb
, 0, -1);
3077 * Walk the list again, actually closing and freeing it,
3078 * with preemption enabled, without holding any MM locks.
3081 if (vma
->vm_flags
& VM_ACCOUNT
)
3082 nr_accounted
+= vma_pages(vma
);
3083 vma
= remove_vma(vma
);
3085 vm_unacct_memory(nr_accounted
);
3088 /* Insert vm structure into process list sorted by address
3089 * and into the inode's i_mmap tree. If vm_file is non-NULL
3090 * then i_mmap_rwsem is taken here.
3092 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3094 struct vm_area_struct
*prev
;
3095 struct rb_node
**rb_link
, *rb_parent
;
3097 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
3098 &prev
, &rb_link
, &rb_parent
))
3100 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3101 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
3105 * The vm_pgoff of a purely anonymous vma should be irrelevant
3106 * until its first write fault, when page's anon_vma and index
3107 * are set. But now set the vm_pgoff it will almost certainly
3108 * end up with (unless mremap moves it elsewhere before that
3109 * first wfault), so /proc/pid/maps tells a consistent story.
3111 * By setting it to reflect the virtual start address of the
3112 * vma, merges and splits can happen in a seamless way, just
3113 * using the existing file pgoff checks and manipulations.
3114 * Similarly in do_mmap_pgoff and in do_brk.
3116 if (vma_is_anonymous(vma
)) {
3117 BUG_ON(vma
->anon_vma
);
3118 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3121 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3126 * Copy the vma structure to a new location in the same mm,
3127 * prior to moving page table entries, to effect an mremap move.
3129 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3130 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3131 bool *need_rmap_locks
)
3133 struct vm_area_struct
*vma
= *vmap
;
3134 unsigned long vma_start
= vma
->vm_start
;
3135 struct mm_struct
*mm
= vma
->vm_mm
;
3136 struct vm_area_struct
*new_vma
, *prev
;
3137 struct rb_node
**rb_link
, *rb_parent
;
3138 bool faulted_in_anon_vma
= true;
3141 * If anonymous vma has not yet been faulted, update new pgoff
3142 * to match new location, to increase its chance of merging.
3144 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3145 pgoff
= addr
>> PAGE_SHIFT
;
3146 faulted_in_anon_vma
= false;
3149 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
3150 return NULL
; /* should never get here */
3151 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
3152 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
3153 vma
->vm_userfaultfd_ctx
, vma_get_anon_name(vma
));
3156 * Source vma may have been merged into new_vma
3158 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3159 vma_start
< new_vma
->vm_end
)) {
3161 * The only way we can get a vma_merge with
3162 * self during an mremap is if the vma hasn't
3163 * been faulted in yet and we were allowed to
3164 * reset the dst vma->vm_pgoff to the
3165 * destination address of the mremap to allow
3166 * the merge to happen. mremap must change the
3167 * vm_pgoff linearity between src and dst vmas
3168 * (in turn preventing a vma_merge) to be
3169 * safe. It is only safe to keep the vm_pgoff
3170 * linear if there are no pages mapped yet.
3172 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3173 *vmap
= vma
= new_vma
;
3175 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3177 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
3181 new_vma
->vm_start
= addr
;
3182 new_vma
->vm_end
= addr
+ len
;
3183 new_vma
->vm_pgoff
= pgoff
;
3184 if (vma_dup_policy(vma
, new_vma
))
3186 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
3187 if (anon_vma_clone(new_vma
, vma
))
3188 goto out_free_mempol
;
3189 if (new_vma
->vm_file
)
3190 get_file(new_vma
->vm_file
);
3191 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3192 new_vma
->vm_ops
->open(new_vma
);
3193 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
3194 *need_rmap_locks
= false;
3199 mpol_put(vma_policy(new_vma
));
3201 kmem_cache_free(vm_area_cachep
, new_vma
);
3207 * Return true if the calling process may expand its vm space by the passed
3210 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3212 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3215 if (is_data_mapping(flags
) &&
3216 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3217 /* Workaround for Valgrind */
3218 if (rlimit(RLIMIT_DATA
) == 0 &&
3219 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3221 if (!ignore_rlimit_data
) {
3222 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n",
3223 current
->comm
, current
->pid
,
3224 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3225 rlimit(RLIMIT_DATA
));
3233 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3235 mm
->total_vm
+= npages
;
3237 if (is_exec_mapping(flags
))
3238 mm
->exec_vm
+= npages
;
3239 else if (is_stack_mapping(flags
))
3240 mm
->stack_vm
+= npages
;
3241 else if (is_data_mapping(flags
))
3242 mm
->data_vm
+= npages
;
3245 static int special_mapping_fault(struct vm_fault
*vmf
);
3248 * Having a close hook prevents vma merging regardless of flags.
3250 static void special_mapping_close(struct vm_area_struct
*vma
)
3254 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3256 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3259 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3261 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3263 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3267 return sm
->mremap(sm
, new_vma
);
3272 static const struct vm_operations_struct special_mapping_vmops
= {
3273 .close
= special_mapping_close
,
3274 .fault
= special_mapping_fault
,
3275 .mremap
= special_mapping_mremap
,
3276 .name
= special_mapping_name
,
3279 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3280 .close
= special_mapping_close
,
3281 .fault
= special_mapping_fault
,
3284 static int special_mapping_fault(struct vm_fault
*vmf
)
3286 struct vm_area_struct
*vma
= vmf
->vma
;
3288 struct page
**pages
;
3290 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3291 pages
= vma
->vm_private_data
;
3293 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3296 return sm
->fault(sm
, vmf
->vma
, vmf
);
3301 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3305 struct page
*page
= *pages
;
3311 return VM_FAULT_SIGBUS
;
3314 static struct vm_area_struct
*__install_special_mapping(
3315 struct mm_struct
*mm
,
3316 unsigned long addr
, unsigned long len
,
3317 unsigned long vm_flags
, void *priv
,
3318 const struct vm_operations_struct
*ops
)
3321 struct vm_area_struct
*vma
;
3323 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
3324 if (unlikely(vma
== NULL
))
3325 return ERR_PTR(-ENOMEM
);
3327 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
3329 vma
->vm_start
= addr
;
3330 vma
->vm_end
= addr
+ len
;
3332 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
3333 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3336 vma
->vm_private_data
= priv
;
3338 ret
= insert_vm_struct(mm
, vma
);
3342 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3344 perf_event_mmap(vma
);
3349 kmem_cache_free(vm_area_cachep
, vma
);
3350 return ERR_PTR(ret
);
3353 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3354 const struct vm_special_mapping
*sm
)
3356 return vma
->vm_private_data
== sm
&&
3357 (vma
->vm_ops
== &special_mapping_vmops
||
3358 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3362 * Called with mm->mmap_sem held for writing.
3363 * Insert a new vma covering the given region, with the given flags.
3364 * Its pages are supplied by the given array of struct page *.
3365 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3366 * The region past the last page supplied will always produce SIGBUS.
3367 * The array pointer and the pages it points to are assumed to stay alive
3368 * for as long as this mapping might exist.
3370 struct vm_area_struct
*_install_special_mapping(
3371 struct mm_struct
*mm
,
3372 unsigned long addr
, unsigned long len
,
3373 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3375 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3376 &special_mapping_vmops
);
3379 int install_special_mapping(struct mm_struct
*mm
,
3380 unsigned long addr
, unsigned long len
,
3381 unsigned long vm_flags
, struct page
**pages
)
3383 struct vm_area_struct
*vma
= __install_special_mapping(
3384 mm
, addr
, len
, vm_flags
, (void *)pages
,
3385 &legacy_special_mapping_vmops
);
3387 return PTR_ERR_OR_ZERO(vma
);
3390 static DEFINE_MUTEX(mm_all_locks_mutex
);
3392 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3394 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3396 * The LSB of head.next can't change from under us
3397 * because we hold the mm_all_locks_mutex.
3399 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3401 * We can safely modify head.next after taking the
3402 * anon_vma->root->rwsem. If some other vma in this mm shares
3403 * the same anon_vma we won't take it again.
3405 * No need of atomic instructions here, head.next
3406 * can't change from under us thanks to the
3407 * anon_vma->root->rwsem.
3409 if (__test_and_set_bit(0, (unsigned long *)
3410 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3415 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3417 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3419 * AS_MM_ALL_LOCKS can't change from under us because
3420 * we hold the mm_all_locks_mutex.
3422 * Operations on ->flags have to be atomic because
3423 * even if AS_MM_ALL_LOCKS is stable thanks to the
3424 * mm_all_locks_mutex, there may be other cpus
3425 * changing other bitflags in parallel to us.
3427 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3429 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_sem
);
3434 * This operation locks against the VM for all pte/vma/mm related
3435 * operations that could ever happen on a certain mm. This includes
3436 * vmtruncate, try_to_unmap, and all page faults.
3438 * The caller must take the mmap_sem in write mode before calling
3439 * mm_take_all_locks(). The caller isn't allowed to release the
3440 * mmap_sem until mm_drop_all_locks() returns.
3442 * mmap_sem in write mode is required in order to block all operations
3443 * that could modify pagetables and free pages without need of
3444 * altering the vma layout. It's also needed in write mode to avoid new
3445 * anon_vmas to be associated with existing vmas.
3447 * A single task can't take more than one mm_take_all_locks() in a row
3448 * or it would deadlock.
3450 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3451 * mapping->flags avoid to take the same lock twice, if more than one
3452 * vma in this mm is backed by the same anon_vma or address_space.
3454 * We take locks in following order, accordingly to comment at beginning
3456 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3458 * - all i_mmap_rwsem locks;
3459 * - all anon_vma->rwseml
3461 * We can take all locks within these types randomly because the VM code
3462 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3463 * mm_all_locks_mutex.
3465 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3466 * that may have to take thousand of locks.
3468 * mm_take_all_locks() can fail if it's interrupted by signals.
3470 int mm_take_all_locks(struct mm_struct
*mm
)
3472 struct vm_area_struct
*vma
;
3473 struct anon_vma_chain
*avc
;
3475 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3477 mutex_lock(&mm_all_locks_mutex
);
3479 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3480 if (signal_pending(current
))
3482 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3483 is_vm_hugetlb_page(vma
))
3484 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3487 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3488 if (signal_pending(current
))
3490 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3491 !is_vm_hugetlb_page(vma
))
3492 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3495 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3496 if (signal_pending(current
))
3499 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3500 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3506 mm_drop_all_locks(mm
);
3510 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3512 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3514 * The LSB of head.next can't change to 0 from under
3515 * us because we hold the mm_all_locks_mutex.
3517 * We must however clear the bitflag before unlocking
3518 * the vma so the users using the anon_vma->rb_root will
3519 * never see our bitflag.
3521 * No need of atomic instructions here, head.next
3522 * can't change from under us until we release the
3523 * anon_vma->root->rwsem.
3525 if (!__test_and_clear_bit(0, (unsigned long *)
3526 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3528 anon_vma_unlock_write(anon_vma
);
3532 static void vm_unlock_mapping(struct address_space
*mapping
)
3534 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3536 * AS_MM_ALL_LOCKS can't change to 0 from under us
3537 * because we hold the mm_all_locks_mutex.
3539 i_mmap_unlock_write(mapping
);
3540 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3547 * The mmap_sem cannot be released by the caller until
3548 * mm_drop_all_locks() returns.
3550 void mm_drop_all_locks(struct mm_struct
*mm
)
3552 struct vm_area_struct
*vma
;
3553 struct anon_vma_chain
*avc
;
3555 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3556 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3558 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3560 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3561 vm_unlock_anon_vma(avc
->anon_vma
);
3562 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3563 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3566 mutex_unlock(&mm_all_locks_mutex
);
3570 * initialise the percpu counter for VM
3572 void __init
mmap_init(void)
3576 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3581 * Initialise sysctl_user_reserve_kbytes.
3583 * This is intended to prevent a user from starting a single memory hogging
3584 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3587 * The default value is min(3% of free memory, 128MB)
3588 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3590 static int init_user_reserve(void)
3592 unsigned long free_kbytes
;
3594 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3596 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3599 subsys_initcall(init_user_reserve
);
3602 * Initialise sysctl_admin_reserve_kbytes.
3604 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3605 * to log in and kill a memory hogging process.
3607 * Systems with more than 256MB will reserve 8MB, enough to recover
3608 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3609 * only reserve 3% of free pages by default.
3611 static int init_admin_reserve(void)
3613 unsigned long free_kbytes
;
3615 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3617 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3620 subsys_initcall(init_admin_reserve
);
3623 * Reinititalise user and admin reserves if memory is added or removed.
3625 * The default user reserve max is 128MB, and the default max for the
3626 * admin reserve is 8MB. These are usually, but not always, enough to
3627 * enable recovery from a memory hogging process using login/sshd, a shell,
3628 * and tools like top. It may make sense to increase or even disable the
3629 * reserve depending on the existence of swap or variations in the recovery
3630 * tools. So, the admin may have changed them.
3632 * If memory is added and the reserves have been eliminated or increased above
3633 * the default max, then we'll trust the admin.
3635 * If memory is removed and there isn't enough free memory, then we
3636 * need to reset the reserves.
3638 * Otherwise keep the reserve set by the admin.
3640 static int reserve_mem_notifier(struct notifier_block
*nb
,
3641 unsigned long action
, void *data
)
3643 unsigned long tmp
, free_kbytes
;
3647 /* Default max is 128MB. Leave alone if modified by operator. */
3648 tmp
= sysctl_user_reserve_kbytes
;
3649 if (0 < tmp
&& tmp
< (1UL << 17))
3650 init_user_reserve();
3652 /* Default max is 8MB. Leave alone if modified by operator. */
3653 tmp
= sysctl_admin_reserve_kbytes
;
3654 if (0 < tmp
&& tmp
< (1UL << 13))
3655 init_admin_reserve();
3659 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3661 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3662 init_user_reserve();
3663 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3664 sysctl_user_reserve_kbytes
);
3667 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3668 init_admin_reserve();
3669 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3670 sysctl_admin_reserve_kbytes
);
3679 static struct notifier_block reserve_mem_nb
= {
3680 .notifier_call
= reserve_mem_notifier
,
3683 static int __meminit
init_reserve_notifier(void)
3685 if (register_hotmemory_notifier(&reserve_mem_nb
))
3686 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3690 subsys_initcall(init_reserve_notifier
);