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 pgprot_t protection_map
[16] __ro_after_init
= {
94 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
95 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
98 #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
99 static inline pgprot_t
arch_filter_pgprot(pgprot_t prot
)
105 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
107 pgprot_t ret
= __pgprot(pgprot_val(protection_map
[vm_flags
&
108 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
109 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
111 return arch_filter_pgprot(ret
);
113 EXPORT_SYMBOL(vm_get_page_prot
);
115 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
117 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
120 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
121 void vma_set_page_prot(struct vm_area_struct
*vma
)
123 unsigned long vm_flags
= vma
->vm_flags
;
124 pgprot_t vm_page_prot
;
126 vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
127 if (vma_wants_writenotify(vma
, vm_page_prot
)) {
128 vm_flags
&= ~VM_SHARED
;
129 vm_page_prot
= vm_pgprot_modify(vm_page_prot
, vm_flags
);
131 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
132 WRITE_ONCE(vma
->vm_page_prot
, vm_page_prot
);
136 * Requires inode->i_mapping->i_mmap_rwsem
138 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
139 struct file
*file
, struct address_space
*mapping
)
141 if (vma
->vm_flags
& VM_DENYWRITE
)
142 atomic_inc(&file_inode(file
)->i_writecount
);
143 if (vma
->vm_flags
& VM_SHARED
)
144 mapping_unmap_writable(mapping
);
146 flush_dcache_mmap_lock(mapping
);
147 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
148 flush_dcache_mmap_unlock(mapping
);
152 * Unlink a file-based vm structure from its interval tree, to hide
153 * vma from rmap and vmtruncate before freeing its page tables.
155 void unlink_file_vma(struct vm_area_struct
*vma
)
157 struct file
*file
= vma
->vm_file
;
160 struct address_space
*mapping
= file
->f_mapping
;
161 i_mmap_lock_write(mapping
);
162 __remove_shared_vm_struct(vma
, file
, mapping
);
163 i_mmap_unlock_write(mapping
);
168 * Close a vm structure and free it, returning the next.
170 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
172 struct vm_area_struct
*next
= vma
->vm_next
;
175 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
176 vma
->vm_ops
->close(vma
);
179 mpol_put(vma_policy(vma
));
184 static int do_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
,
185 struct list_head
*uf
);
186 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
188 unsigned long retval
;
189 unsigned long newbrk
, oldbrk
;
190 struct mm_struct
*mm
= current
->mm
;
191 struct vm_area_struct
*next
;
192 unsigned long min_brk
;
196 if (down_write_killable(&mm
->mmap_sem
))
199 #ifdef CONFIG_COMPAT_BRK
201 * CONFIG_COMPAT_BRK can still be overridden by setting
202 * randomize_va_space to 2, which will still cause mm->start_brk
203 * to be arbitrarily shifted
205 if (current
->brk_randomized
)
206 min_brk
= mm
->start_brk
;
208 min_brk
= mm
->end_data
;
210 min_brk
= mm
->start_brk
;
216 * Check against rlimit here. If this check is done later after the test
217 * of oldbrk with newbrk then it can escape the test and let the data
218 * segment grow beyond its set limit the in case where the limit is
219 * not page aligned -Ram Gupta
221 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
222 mm
->end_data
, mm
->start_data
))
225 newbrk
= PAGE_ALIGN(brk
);
226 oldbrk
= PAGE_ALIGN(mm
->brk
);
227 if (oldbrk
== newbrk
)
230 /* Always allow shrinking brk. */
231 if (brk
<= mm
->brk
) {
232 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
, &uf
))
237 /* Check against existing mmap mappings. */
238 next
= find_vma(mm
, oldbrk
);
239 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
242 /* Ok, looks good - let it rip. */
243 if (do_brk_flags(oldbrk
, newbrk
-oldbrk
, 0, &uf
) < 0)
248 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
249 up_write(&mm
->mmap_sem
);
250 userfaultfd_unmap_complete(mm
, &uf
);
252 mm_populate(oldbrk
, newbrk
- oldbrk
);
257 up_write(&mm
->mmap_sem
);
261 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
263 unsigned long max
, prev_end
, subtree_gap
;
266 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
267 * allow two stack_guard_gaps between them here, and when choosing
268 * an unmapped area; whereas when expanding we only require one.
269 * That's a little inconsistent, but keeps the code here simpler.
271 max
= vm_start_gap(vma
);
273 prev_end
= vm_end_gap(vma
->vm_prev
);
279 if (vma
->vm_rb
.rb_left
) {
280 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
281 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
282 if (subtree_gap
> max
)
285 if (vma
->vm_rb
.rb_right
) {
286 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
287 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
288 if (subtree_gap
> max
)
294 #ifdef CONFIG_DEBUG_VM_RB
295 static int browse_rb(struct mm_struct
*mm
)
297 struct rb_root
*root
= &mm
->mm_rb
;
298 int i
= 0, j
, bug
= 0;
299 struct rb_node
*nd
, *pn
= NULL
;
300 unsigned long prev
= 0, pend
= 0;
302 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
303 struct vm_area_struct
*vma
;
304 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
305 if (vma
->vm_start
< prev
) {
306 pr_emerg("vm_start %lx < prev %lx\n",
307 vma
->vm_start
, prev
);
310 if (vma
->vm_start
< pend
) {
311 pr_emerg("vm_start %lx < pend %lx\n",
312 vma
->vm_start
, pend
);
315 if (vma
->vm_start
> vma
->vm_end
) {
316 pr_emerg("vm_start %lx > vm_end %lx\n",
317 vma
->vm_start
, vma
->vm_end
);
320 spin_lock(&mm
->page_table_lock
);
321 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
322 pr_emerg("free gap %lx, correct %lx\n",
324 vma_compute_subtree_gap(vma
));
327 spin_unlock(&mm
->page_table_lock
);
330 prev
= vma
->vm_start
;
334 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
337 pr_emerg("backwards %d, forwards %d\n", j
, i
);
343 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
347 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
348 struct vm_area_struct
*vma
;
349 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
350 VM_BUG_ON_VMA(vma
!= ignore
&&
351 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
356 static void validate_mm(struct mm_struct
*mm
)
360 unsigned long highest_address
= 0;
361 struct vm_area_struct
*vma
= mm
->mmap
;
364 struct anon_vma
*anon_vma
= vma
->anon_vma
;
365 struct anon_vma_chain
*avc
;
368 anon_vma_lock_read(anon_vma
);
369 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
370 anon_vma_interval_tree_verify(avc
);
371 anon_vma_unlock_read(anon_vma
);
374 highest_address
= vm_end_gap(vma
);
378 if (i
!= mm
->map_count
) {
379 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
382 if (highest_address
!= mm
->highest_vm_end
) {
383 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
384 mm
->highest_vm_end
, highest_address
);
388 if (i
!= mm
->map_count
) {
390 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
393 VM_BUG_ON_MM(bug
, mm
);
396 #define validate_mm_rb(root, ignore) do { } while (0)
397 #define validate_mm(mm) do { } while (0)
400 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
401 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
404 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
405 * vma->vm_prev->vm_end values changed, without modifying the vma's position
408 static void vma_gap_update(struct vm_area_struct
*vma
)
411 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
412 * function that does exacltly what we want.
414 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
417 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
418 struct rb_root
*root
)
420 /* All rb_subtree_gap values must be consistent prior to insertion */
421 validate_mm_rb(root
, NULL
);
423 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
426 static void __vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
429 * Note rb_erase_augmented is a fairly large inline function,
430 * so make sure we instantiate it only once with our desired
431 * augmented rbtree callbacks.
433 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
436 static __always_inline
void vma_rb_erase_ignore(struct vm_area_struct
*vma
,
437 struct rb_root
*root
,
438 struct vm_area_struct
*ignore
)
441 * All rb_subtree_gap values must be consistent prior to erase,
442 * with the possible exception of the "next" vma being erased if
443 * next->vm_start was reduced.
445 validate_mm_rb(root
, ignore
);
447 __vma_rb_erase(vma
, root
);
450 static __always_inline
void vma_rb_erase(struct vm_area_struct
*vma
,
451 struct rb_root
*root
)
454 * All rb_subtree_gap values must be consistent prior to erase,
455 * with the possible exception of the vma being erased.
457 validate_mm_rb(root
, vma
);
459 __vma_rb_erase(vma
, root
);
463 * vma has some anon_vma assigned, and is already inserted on that
464 * anon_vma's interval trees.
466 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
467 * vma must be removed from the anon_vma's interval trees using
468 * anon_vma_interval_tree_pre_update_vma().
470 * After the update, the vma will be reinserted using
471 * anon_vma_interval_tree_post_update_vma().
473 * The entire update must be protected by exclusive mmap_sem and by
474 * the root anon_vma's mutex.
477 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
479 struct anon_vma_chain
*avc
;
481 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
482 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
486 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
488 struct anon_vma_chain
*avc
;
490 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
491 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
494 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
495 unsigned long end
, struct vm_area_struct
**pprev
,
496 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
498 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
500 __rb_link
= &mm
->mm_rb
.rb_node
;
501 rb_prev
= __rb_parent
= NULL
;
504 struct vm_area_struct
*vma_tmp
;
506 __rb_parent
= *__rb_link
;
507 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
509 if (vma_tmp
->vm_end
> addr
) {
510 /* Fail if an existing vma overlaps the area */
511 if (vma_tmp
->vm_start
< end
)
513 __rb_link
= &__rb_parent
->rb_left
;
515 rb_prev
= __rb_parent
;
516 __rb_link
= &__rb_parent
->rb_right
;
522 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
523 *rb_link
= __rb_link
;
524 *rb_parent
= __rb_parent
;
528 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
529 unsigned long addr
, unsigned long end
)
531 unsigned long nr_pages
= 0;
532 struct vm_area_struct
*vma
;
534 /* Find first overlaping mapping */
535 vma
= find_vma_intersection(mm
, addr
, end
);
539 nr_pages
= (min(end
, vma
->vm_end
) -
540 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
542 /* Iterate over the rest of the overlaps */
543 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
544 unsigned long overlap_len
;
546 if (vma
->vm_start
> end
)
549 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
550 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
556 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
557 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
559 /* Update tracking information for the gap following the new vma. */
561 vma_gap_update(vma
->vm_next
);
563 mm
->highest_vm_end
= vm_end_gap(vma
);
566 * vma->vm_prev wasn't known when we followed the rbtree to find the
567 * correct insertion point for that vma. As a result, we could not
568 * update the vma vm_rb parents rb_subtree_gap values on the way down.
569 * So, we first insert the vma with a zero rb_subtree_gap value
570 * (to be consistent with what we did on the way down), and then
571 * immediately update the gap to the correct value. Finally we
572 * rebalance the rbtree after all augmented values have been set.
574 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
575 vma
->rb_subtree_gap
= 0;
577 vma_rb_insert(vma
, &mm
->mm_rb
);
580 static void __vma_link_file(struct vm_area_struct
*vma
)
586 struct address_space
*mapping
= file
->f_mapping
;
588 if (vma
->vm_flags
& VM_DENYWRITE
)
589 atomic_dec(&file_inode(file
)->i_writecount
);
590 if (vma
->vm_flags
& VM_SHARED
)
591 atomic_inc(&mapping
->i_mmap_writable
);
593 flush_dcache_mmap_lock(mapping
);
594 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
595 flush_dcache_mmap_unlock(mapping
);
600 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
601 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
602 struct rb_node
*rb_parent
)
604 __vma_link_list(mm
, vma
, prev
, rb_parent
);
605 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
608 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
609 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
610 struct rb_node
*rb_parent
)
612 struct address_space
*mapping
= NULL
;
615 mapping
= vma
->vm_file
->f_mapping
;
616 i_mmap_lock_write(mapping
);
619 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
620 __vma_link_file(vma
);
623 i_mmap_unlock_write(mapping
);
630 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
631 * mm's list and rbtree. It has already been inserted into the interval tree.
633 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
635 struct vm_area_struct
*prev
;
636 struct rb_node
**rb_link
, *rb_parent
;
638 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
639 &prev
, &rb_link
, &rb_parent
))
641 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
645 static __always_inline
void __vma_unlink_common(struct mm_struct
*mm
,
646 struct vm_area_struct
*vma
,
647 struct vm_area_struct
*prev
,
649 struct vm_area_struct
*ignore
)
651 struct vm_area_struct
*next
;
653 vma_rb_erase_ignore(vma
, &mm
->mm_rb
, ignore
);
656 prev
->vm_next
= next
;
660 prev
->vm_next
= next
;
665 next
->vm_prev
= prev
;
668 vmacache_invalidate(mm
);
671 static inline void __vma_unlink_prev(struct mm_struct
*mm
,
672 struct vm_area_struct
*vma
,
673 struct vm_area_struct
*prev
)
675 __vma_unlink_common(mm
, vma
, prev
, true, vma
);
679 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
680 * is already present in an i_mmap tree without adjusting the tree.
681 * The following helper function should be used when such adjustments
682 * are necessary. The "insert" vma (if any) is to be inserted
683 * before we drop the necessary locks.
685 int __vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
686 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
,
687 struct vm_area_struct
*expand
)
689 struct mm_struct
*mm
= vma
->vm_mm
;
690 struct vm_area_struct
*next
= vma
->vm_next
, *orig_vma
= vma
;
691 struct address_space
*mapping
= NULL
;
692 struct rb_root_cached
*root
= NULL
;
693 struct anon_vma
*anon_vma
= NULL
;
694 struct file
*file
= vma
->vm_file
;
695 bool start_changed
= false, end_changed
= false;
696 long adjust_next
= 0;
699 if (next
&& !insert
) {
700 struct vm_area_struct
*exporter
= NULL
, *importer
= NULL
;
702 if (end
>= next
->vm_end
) {
704 * vma expands, overlapping all the next, and
705 * perhaps the one after too (mprotect case 6).
706 * The only other cases that gets here are
707 * case 1, case 7 and case 8.
709 if (next
== expand
) {
711 * The only case where we don't expand "vma"
712 * and we expand "next" instead is case 8.
714 VM_WARN_ON(end
!= next
->vm_end
);
716 * remove_next == 3 means we're
717 * removing "vma" and that to do so we
718 * swapped "vma" and "next".
721 VM_WARN_ON(file
!= next
->vm_file
);
724 VM_WARN_ON(expand
!= vma
);
726 * case 1, 6, 7, remove_next == 2 is case 6,
727 * remove_next == 1 is case 1 or 7.
729 remove_next
= 1 + (end
> next
->vm_end
);
730 VM_WARN_ON(remove_next
== 2 &&
731 end
!= next
->vm_next
->vm_end
);
732 VM_WARN_ON(remove_next
== 1 &&
733 end
!= next
->vm_end
);
734 /* trim end to next, for case 6 first pass */
742 * If next doesn't have anon_vma, import from vma after
743 * next, if the vma overlaps with it.
745 if (remove_next
== 2 && !next
->anon_vma
)
746 exporter
= next
->vm_next
;
748 } else if (end
> next
->vm_start
) {
750 * vma expands, overlapping part of the next:
751 * mprotect case 5 shifting the boundary up.
753 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
756 VM_WARN_ON(expand
!= importer
);
757 } else if (end
< vma
->vm_end
) {
759 * vma shrinks, and !insert tells it's not
760 * split_vma inserting another: so it must be
761 * mprotect case 4 shifting the boundary down.
763 adjust_next
= -((vma
->vm_end
- end
) >> PAGE_SHIFT
);
766 VM_WARN_ON(expand
!= importer
);
770 * Easily overlooked: when mprotect shifts the boundary,
771 * make sure the expanding vma has anon_vma set if the
772 * shrinking vma had, to cover any anon pages imported.
774 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
777 importer
->anon_vma
= exporter
->anon_vma
;
778 error
= anon_vma_clone(importer
, exporter
);
784 vma_adjust_trans_huge(orig_vma
, start
, end
, adjust_next
);
787 mapping
= file
->f_mapping
;
788 root
= &mapping
->i_mmap
;
789 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
792 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
794 i_mmap_lock_write(mapping
);
797 * Put into interval tree now, so instantiated pages
798 * are visible to arm/parisc __flush_dcache_page
799 * throughout; but we cannot insert into address
800 * space until vma start or end is updated.
802 __vma_link_file(insert
);
806 anon_vma
= vma
->anon_vma
;
807 if (!anon_vma
&& adjust_next
)
808 anon_vma
= next
->anon_vma
;
810 VM_WARN_ON(adjust_next
&& next
->anon_vma
&&
811 anon_vma
!= next
->anon_vma
);
812 anon_vma_lock_write(anon_vma
);
813 anon_vma_interval_tree_pre_update_vma(vma
);
815 anon_vma_interval_tree_pre_update_vma(next
);
819 flush_dcache_mmap_lock(mapping
);
820 vma_interval_tree_remove(vma
, root
);
822 vma_interval_tree_remove(next
, root
);
825 if (start
!= vma
->vm_start
) {
826 vma
->vm_start
= start
;
827 start_changed
= true;
829 if (end
!= vma
->vm_end
) {
833 vma
->vm_pgoff
= pgoff
;
835 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
836 next
->vm_pgoff
+= adjust_next
;
841 vma_interval_tree_insert(next
, root
);
842 vma_interval_tree_insert(vma
, root
);
843 flush_dcache_mmap_unlock(mapping
);
848 * vma_merge has merged next into vma, and needs
849 * us to remove next before dropping the locks.
851 if (remove_next
!= 3)
852 __vma_unlink_prev(mm
, next
, vma
);
855 * vma is not before next if they've been
858 * pre-swap() next->vm_start was reduced so
859 * tell validate_mm_rb to ignore pre-swap()
860 * "next" (which is stored in post-swap()
863 __vma_unlink_common(mm
, next
, NULL
, false, vma
);
865 __remove_shared_vm_struct(next
, file
, mapping
);
868 * split_vma has split insert from vma, and needs
869 * us to insert it before dropping the locks
870 * (it may either follow vma or precede it).
872 __insert_vm_struct(mm
, insert
);
878 mm
->highest_vm_end
= vm_end_gap(vma
);
879 else if (!adjust_next
)
880 vma_gap_update(next
);
885 anon_vma_interval_tree_post_update_vma(vma
);
887 anon_vma_interval_tree_post_update_vma(next
);
888 anon_vma_unlock_write(anon_vma
);
891 i_mmap_unlock_write(mapping
);
902 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
906 anon_vma_merge(vma
, next
);
908 mpol_put(vma_policy(next
));
911 * In mprotect's case 6 (see comments on vma_merge),
912 * we must remove another next too. It would clutter
913 * up the code too much to do both in one go.
915 if (remove_next
!= 3) {
917 * If "next" was removed and vma->vm_end was
918 * expanded (up) over it, in turn
919 * "next->vm_prev->vm_end" changed and the
920 * "vma->vm_next" gap must be updated.
925 * For the scope of the comment "next" and
926 * "vma" considered pre-swap(): if "vma" was
927 * removed, next->vm_start was expanded (down)
928 * over it and the "next" gap must be updated.
929 * Because of the swap() the post-swap() "vma"
930 * actually points to pre-swap() "next"
931 * (post-swap() "next" as opposed is now a
936 if (remove_next
== 2) {
942 vma_gap_update(next
);
945 * If remove_next == 2 we obviously can't
948 * If remove_next == 3 we can't reach this
949 * path because pre-swap() next is always not
950 * NULL. pre-swap() "next" is not being
951 * removed and its next->vm_end is not altered
952 * (and furthermore "end" already matches
953 * next->vm_end in remove_next == 3).
955 * We reach this only in the remove_next == 1
956 * case if the "next" vma that was removed was
957 * the highest vma of the mm. However in such
958 * case next->vm_end == "end" and the extended
959 * "vma" has vma->vm_end == next->vm_end so
960 * mm->highest_vm_end doesn't need any update
961 * in remove_next == 1 case.
963 VM_WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
975 * If the vma has a ->close operation then the driver probably needs to release
976 * per-vma resources, so we don't attempt to merge those.
978 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
979 struct file
*file
, unsigned long vm_flags
,
980 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
983 * VM_SOFTDIRTY should not prevent from VMA merging, if we
984 * match the flags but dirty bit -- the caller should mark
985 * merged VMA as dirty. If dirty bit won't be excluded from
986 * comparison, we increase pressue on the memory system forcing
987 * the kernel to generate new VMAs when old one could be
990 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
992 if (vma
->vm_file
!= file
)
994 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
996 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
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
)
1032 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1033 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1034 if (vma
->vm_pgoff
== vm_pgoff
)
1041 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1042 * beyond (at a higher virtual address and file offset than) the vma.
1044 * We cannot merge two vmas if they have differently assigned (non-NULL)
1045 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1048 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1049 struct anon_vma
*anon_vma
, struct file
*file
,
1051 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1053 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1054 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1056 vm_pglen
= vma_pages(vma
);
1057 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1064 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1065 * whether that can be merged with its predecessor or its successor.
1066 * Or both (it neatly fills a hole).
1068 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1069 * certain not to be mapped by the time vma_merge is called; but when
1070 * called for mprotect, it is certain to be already mapped (either at
1071 * an offset within prev, or at the start of next), and the flags of
1072 * this area are about to be changed to vm_flags - and the no-change
1073 * case has already been eliminated.
1075 * The following mprotect cases have to be considered, where AAAA is
1076 * the area passed down from mprotect_fixup, never extending beyond one
1077 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1079 * AAAA AAAA AAAA AAAA
1080 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1081 * cannot merge might become might become might become
1082 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1083 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1084 * mremap move: PPPPXXXXXXXX 8
1086 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1087 * might become case 1 below case 2 below case 3 below
1089 * It is important for case 8 that the the vma NNNN overlapping the
1090 * region AAAA is never going to extended over XXXX. Instead XXXX must
1091 * be extended in region AAAA and NNNN must be removed. This way in
1092 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1093 * rmap_locks, the properties of the merged vma will be already
1094 * correct for the whole merged range. Some of those properties like
1095 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1096 * be correct for the whole merged range immediately after the
1097 * rmap_locks are released. Otherwise if XXXX would be removed and
1098 * NNNN would be extended over the XXXX range, remove_migration_ptes
1099 * or other rmap walkers (if working on addresses beyond the "end"
1100 * parameter) may establish ptes with the wrong permissions of NNNN
1101 * instead of the right permissions of XXXX.
1103 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1104 struct vm_area_struct
*prev
, unsigned long addr
,
1105 unsigned long end
, unsigned long vm_flags
,
1106 struct anon_vma
*anon_vma
, struct file
*file
,
1107 pgoff_t pgoff
, struct mempolicy
*policy
,
1108 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1110 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1111 struct vm_area_struct
*area
, *next
;
1115 * We later require that vma->vm_flags == vm_flags,
1116 * so this tests vma->vm_flags & VM_SPECIAL, too.
1118 if (vm_flags
& VM_SPECIAL
)
1122 next
= prev
->vm_next
;
1126 if (area
&& area
->vm_end
== end
) /* cases 6, 7, 8 */
1127 next
= next
->vm_next
;
1129 /* verify some invariant that must be enforced by the caller */
1130 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
1131 VM_WARN_ON(area
&& end
> area
->vm_end
);
1132 VM_WARN_ON(addr
>= end
);
1135 * Can it merge with the predecessor?
1137 if (prev
&& prev
->vm_end
== addr
&&
1138 mpol_equal(vma_policy(prev
), policy
) &&
1139 can_vma_merge_after(prev
, vm_flags
,
1140 anon_vma
, file
, pgoff
,
1141 vm_userfaultfd_ctx
)) {
1143 * OK, it can. Can we now merge in the successor as well?
1145 if (next
&& end
== next
->vm_start
&&
1146 mpol_equal(policy
, vma_policy(next
)) &&
1147 can_vma_merge_before(next
, vm_flags
,
1150 vm_userfaultfd_ctx
) &&
1151 is_mergeable_anon_vma(prev
->anon_vma
,
1152 next
->anon_vma
, NULL
)) {
1154 err
= __vma_adjust(prev
, prev
->vm_start
,
1155 next
->vm_end
, prev
->vm_pgoff
, NULL
,
1157 } else /* cases 2, 5, 7 */
1158 err
= __vma_adjust(prev
, prev
->vm_start
,
1159 end
, prev
->vm_pgoff
, NULL
, prev
);
1162 khugepaged_enter_vma_merge(prev
, vm_flags
);
1167 * Can this new request be merged in front of next?
1169 if (next
&& end
== next
->vm_start
&&
1170 mpol_equal(policy
, vma_policy(next
)) &&
1171 can_vma_merge_before(next
, vm_flags
,
1172 anon_vma
, file
, pgoff
+pglen
,
1173 vm_userfaultfd_ctx
)) {
1174 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1175 err
= __vma_adjust(prev
, prev
->vm_start
,
1176 addr
, prev
->vm_pgoff
, NULL
, next
);
1177 else { /* cases 3, 8 */
1178 err
= __vma_adjust(area
, addr
, next
->vm_end
,
1179 next
->vm_pgoff
- pglen
, NULL
, next
);
1181 * In case 3 area is already equal to next and
1182 * this is a noop, but in case 8 "area" has
1183 * been removed and next was expanded over it.
1189 khugepaged_enter_vma_merge(area
, vm_flags
);
1197 * Rough compatbility check to quickly see if it's even worth looking
1198 * at sharing an anon_vma.
1200 * They need to have the same vm_file, and the flags can only differ
1201 * in things that mprotect may change.
1203 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1204 * we can merge the two vma's. For example, we refuse to merge a vma if
1205 * there is a vm_ops->close() function, because that indicates that the
1206 * driver is doing some kind of reference counting. But that doesn't
1207 * really matter for the anon_vma sharing case.
1209 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1211 return a
->vm_end
== b
->vm_start
&&
1212 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1213 a
->vm_file
== b
->vm_file
&&
1214 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
|VM_SOFTDIRTY
)) &&
1215 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1219 * Do some basic sanity checking to see if we can re-use the anon_vma
1220 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1221 * the same as 'old', the other will be the new one that is trying
1222 * to share the anon_vma.
1224 * NOTE! This runs with mm_sem held for reading, so it is possible that
1225 * the anon_vma of 'old' is concurrently in the process of being set up
1226 * by another page fault trying to merge _that_. But that's ok: if it
1227 * is being set up, that automatically means that it will be a singleton
1228 * acceptable for merging, so we can do all of this optimistically. But
1229 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1231 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1232 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1233 * is to return an anon_vma that is "complex" due to having gone through
1236 * We also make sure that the two vma's are compatible (adjacent,
1237 * and with the same memory policies). That's all stable, even with just
1238 * a read lock on the mm_sem.
1240 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1242 if (anon_vma_compatible(a
, b
)) {
1243 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1245 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1252 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1253 * neighbouring vmas for a suitable anon_vma, before it goes off
1254 * to allocate a new anon_vma. It checks because a repetitive
1255 * sequence of mprotects and faults may otherwise lead to distinct
1256 * anon_vmas being allocated, preventing vma merge in subsequent
1259 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1261 struct anon_vma
*anon_vma
;
1262 struct vm_area_struct
*near
;
1264 near
= vma
->vm_next
;
1268 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1272 near
= vma
->vm_prev
;
1276 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1281 * There's no absolute need to look only at touching neighbours:
1282 * we could search further afield for "compatible" anon_vmas.
1283 * But it would probably just be a waste of time searching,
1284 * or lead to too many vmas hanging off the same anon_vma.
1285 * We're trying to allow mprotect remerging later on,
1286 * not trying to minimize memory used for anon_vmas.
1292 * If a hint addr is less than mmap_min_addr change hint to be as
1293 * low as possible but still greater than mmap_min_addr
1295 static inline unsigned long round_hint_to_min(unsigned long hint
)
1298 if (((void *)hint
!= NULL
) &&
1299 (hint
< mmap_min_addr
))
1300 return PAGE_ALIGN(mmap_min_addr
);
1304 static inline int mlock_future_check(struct mm_struct
*mm
,
1305 unsigned long flags
,
1308 unsigned long locked
, lock_limit
;
1310 /* mlock MCL_FUTURE? */
1311 if (flags
& VM_LOCKED
) {
1312 locked
= len
>> PAGE_SHIFT
;
1313 locked
+= mm
->locked_vm
;
1314 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1315 lock_limit
>>= PAGE_SHIFT
;
1316 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1322 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1324 if (S_ISREG(inode
->i_mode
))
1325 return MAX_LFS_FILESIZE
;
1327 if (S_ISBLK(inode
->i_mode
))
1328 return MAX_LFS_FILESIZE
;
1330 /* Special "we do even unsigned file positions" case */
1331 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1334 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1338 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1339 unsigned long pgoff
, unsigned long len
)
1341 u64 maxsize
= file_mmap_size_max(file
, inode
);
1343 if (maxsize
&& len
> maxsize
)
1346 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1352 * The caller must hold down_write(¤t->mm->mmap_sem).
1354 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1355 unsigned long len
, unsigned long prot
,
1356 unsigned long flags
, vm_flags_t vm_flags
,
1357 unsigned long pgoff
, unsigned long *populate
,
1358 struct list_head
*uf
)
1360 struct mm_struct
*mm
= current
->mm
;
1369 * Does the application expect PROT_READ to imply PROT_EXEC?
1371 * (the exception is when the underlying filesystem is noexec
1372 * mounted, in which case we dont add PROT_EXEC.)
1374 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1375 if (!(file
&& path_noexec(&file
->f_path
)))
1378 if (!(flags
& MAP_FIXED
))
1379 addr
= round_hint_to_min(addr
);
1381 /* Careful about overflows.. */
1382 len
= PAGE_ALIGN(len
);
1386 /* offset overflow? */
1387 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1390 /* Too many mappings? */
1391 if (mm
->map_count
> sysctl_max_map_count
)
1394 /* Obtain the address to map to. we verify (or select) it and ensure
1395 * that it represents a valid section of the address space.
1397 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1398 if (offset_in_page(addr
))
1401 if (prot
== PROT_EXEC
) {
1402 pkey
= execute_only_pkey(mm
);
1407 /* Do simple checking here so the lower-level routines won't have
1408 * to. we assume access permissions have been handled by the open
1409 * of the memory object, so we don't do any here.
1411 vm_flags
|= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1412 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1414 if (flags
& MAP_LOCKED
)
1415 if (!can_do_mlock())
1418 if (mlock_future_check(mm
, vm_flags
, len
))
1422 struct inode
*inode
= file_inode(file
);
1423 unsigned long flags_mask
;
1425 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1428 flags_mask
= LEGACY_MAP_MASK
| file
->f_op
->mmap_supported_flags
;
1430 switch (flags
& MAP_TYPE
) {
1433 * Force use of MAP_SHARED_VALIDATE with non-legacy
1434 * flags. E.g. MAP_SYNC is dangerous to use with
1435 * MAP_SHARED as you don't know which consistency model
1436 * you will get. We silently ignore unsupported flags
1437 * with MAP_SHARED to preserve backward compatibility.
1439 flags
&= LEGACY_MAP_MASK
;
1441 case MAP_SHARED_VALIDATE
:
1442 if (flags
& ~flags_mask
)
1444 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1448 * Make sure we don't allow writing to an append-only
1451 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1455 * Make sure there are no mandatory locks on the file.
1457 if (locks_verify_locked(file
))
1460 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1461 if (!(file
->f_mode
& FMODE_WRITE
))
1462 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1466 if (!(file
->f_mode
& FMODE_READ
))
1468 if (path_noexec(&file
->f_path
)) {
1469 if (vm_flags
& VM_EXEC
)
1471 vm_flags
&= ~VM_MAYEXEC
;
1474 if (!file
->f_op
->mmap
)
1476 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1484 switch (flags
& MAP_TYPE
) {
1486 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1492 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1496 * Set pgoff according to addr for anon_vma.
1498 pgoff
= addr
>> PAGE_SHIFT
;
1506 * Set 'VM_NORESERVE' if we should not account for the
1507 * memory use of this mapping.
1509 if (flags
& MAP_NORESERVE
) {
1510 /* We honor MAP_NORESERVE if allowed to overcommit */
1511 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1512 vm_flags
|= VM_NORESERVE
;
1514 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1515 if (file
&& is_file_hugepages(file
))
1516 vm_flags
|= VM_NORESERVE
;
1519 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1520 if (!IS_ERR_VALUE(addr
) &&
1521 ((vm_flags
& VM_LOCKED
) ||
1522 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1527 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1528 unsigned long, prot
, unsigned long, flags
,
1529 unsigned long, fd
, unsigned long, pgoff
)
1531 struct file
*file
= NULL
;
1532 unsigned long retval
;
1534 if (!(flags
& MAP_ANONYMOUS
)) {
1535 audit_mmap_fd(fd
, flags
);
1539 if (is_file_hugepages(file
))
1540 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1542 if (unlikely(flags
& MAP_HUGETLB
&& !is_file_hugepages(file
)))
1544 } else if (flags
& MAP_HUGETLB
) {
1545 struct user_struct
*user
= NULL
;
1548 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1552 len
= ALIGN(len
, huge_page_size(hs
));
1554 * VM_NORESERVE is used because the reservations will be
1555 * taken when vm_ops->mmap() is called
1556 * A dummy user value is used because we are not locking
1557 * memory so no accounting is necessary
1559 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1561 &user
, HUGETLB_ANONHUGE_INODE
,
1562 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1564 return PTR_ERR(file
);
1567 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1569 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1576 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1577 struct mmap_arg_struct
{
1581 unsigned long flags
;
1583 unsigned long offset
;
1586 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1588 struct mmap_arg_struct a
;
1590 if (copy_from_user(&a
, arg
, sizeof(a
)))
1592 if (offset_in_page(a
.offset
))
1595 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1596 a
.offset
>> PAGE_SHIFT
);
1598 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1601 * Some shared mappigns will want the pages marked read-only
1602 * to track write events. If so, we'll downgrade vm_page_prot
1603 * to the private version (using protection_map[] without the
1606 int vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1608 vm_flags_t vm_flags
= vma
->vm_flags
;
1609 const struct vm_operations_struct
*vm_ops
= vma
->vm_ops
;
1611 /* If it was private or non-writable, the write bit is already clear */
1612 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1615 /* The backer wishes to know when pages are first written to? */
1616 if (vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
))
1619 /* The open routine did something to the protections that pgprot_modify
1620 * won't preserve? */
1621 if (pgprot_val(vm_page_prot
) !=
1622 pgprot_val(vm_pgprot_modify(vm_page_prot
, vm_flags
)))
1625 /* Do we need to track softdirty? */
1626 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1629 /* Specialty mapping? */
1630 if (vm_flags
& VM_PFNMAP
)
1633 /* Can the mapping track the dirty pages? */
1634 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1635 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1639 * We account for memory if it's a private writeable mapping,
1640 * not hugepages and VM_NORESERVE wasn't set.
1642 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1645 * hugetlb has its own accounting separate from the core VM
1646 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1648 if (file
&& is_file_hugepages(file
))
1651 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1654 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1655 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
1656 struct list_head
*uf
)
1658 struct mm_struct
*mm
= current
->mm
;
1659 struct vm_area_struct
*vma
, *prev
;
1661 struct rb_node
**rb_link
, *rb_parent
;
1662 unsigned long charged
= 0;
1664 /* Check against address space limit. */
1665 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
1666 unsigned long nr_pages
;
1669 * MAP_FIXED may remove pages of mappings that intersects with
1670 * requested mapping. Account for the pages it would unmap.
1672 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1674 if (!may_expand_vm(mm
, vm_flags
,
1675 (len
>> PAGE_SHIFT
) - nr_pages
))
1679 /* Clear old maps */
1680 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
1682 if (do_munmap(mm
, addr
, len
, uf
))
1687 * Private writable mapping: check memory availability
1689 if (accountable_mapping(file
, vm_flags
)) {
1690 charged
= len
>> PAGE_SHIFT
;
1691 if (security_vm_enough_memory_mm(mm
, charged
))
1693 vm_flags
|= VM_ACCOUNT
;
1697 * Can we just expand an old mapping?
1699 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1700 NULL
, file
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
1705 * Determine the object being mapped and call the appropriate
1706 * specific mapper. the address has already been validated, but
1707 * not unmapped, but the maps are removed from the list.
1709 vma
= vm_area_alloc(mm
);
1715 vma
->vm_start
= addr
;
1716 vma
->vm_end
= addr
+ len
;
1717 vma
->vm_flags
= vm_flags
;
1718 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1719 vma
->vm_pgoff
= pgoff
;
1722 if (vm_flags
& VM_DENYWRITE
) {
1723 error
= deny_write_access(file
);
1727 if (vm_flags
& VM_SHARED
) {
1728 error
= mapping_map_writable(file
->f_mapping
);
1730 goto allow_write_and_free_vma
;
1733 /* ->mmap() can change vma->vm_file, but must guarantee that
1734 * vma_link() below can deny write-access if VM_DENYWRITE is set
1735 * and map writably if VM_SHARED is set. This usually means the
1736 * new file must not have been exposed to user-space, yet.
1738 vma
->vm_file
= get_file(file
);
1739 error
= call_mmap(file
, vma
);
1741 goto unmap_and_free_vma
;
1743 /* Can addr have changed??
1745 * Answer: Yes, several device drivers can do it in their
1746 * f_op->mmap method. -DaveM
1747 * Bug: If addr is changed, prev, rb_link, rb_parent should
1748 * be updated for vma_link()
1750 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1752 addr
= vma
->vm_start
;
1753 vm_flags
= vma
->vm_flags
;
1754 } else if (vm_flags
& VM_SHARED
) {
1755 error
= shmem_zero_setup(vma
);
1760 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1761 /* Once vma denies write, undo our temporary denial count */
1763 if (vm_flags
& VM_SHARED
)
1764 mapping_unmap_writable(file
->f_mapping
);
1765 if (vm_flags
& VM_DENYWRITE
)
1766 allow_write_access(file
);
1768 file
= vma
->vm_file
;
1770 perf_event_mmap(vma
);
1772 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
1773 if (vm_flags
& VM_LOCKED
) {
1774 if (!((vm_flags
& VM_SPECIAL
) || is_vm_hugetlb_page(vma
) ||
1775 vma
== get_gate_vma(current
->mm
)))
1776 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1778 vma
->vm_flags
&= VM_LOCKED_CLEAR_MASK
;
1785 * New (or expanded) vma always get soft dirty status.
1786 * Otherwise user-space soft-dirty page tracker won't
1787 * be able to distinguish situation when vma area unmapped,
1788 * then new mapped in-place (which must be aimed as
1789 * a completely new data area).
1791 vma
->vm_flags
|= VM_SOFTDIRTY
;
1793 vma_set_page_prot(vma
);
1799 vma
->vm_file
= NULL
;
1801 /* Undo any partial mapping done by a device driver. */
1802 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1804 if (vm_flags
& VM_SHARED
)
1805 mapping_unmap_writable(file
->f_mapping
);
1806 allow_write_and_free_vma
:
1807 if (vm_flags
& VM_DENYWRITE
)
1808 allow_write_access(file
);
1813 vm_unacct_memory(charged
);
1817 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1820 * We implement the search by looking for an rbtree node that
1821 * immediately follows a suitable gap. That is,
1822 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1823 * - gap_end = vma->vm_start >= info->low_limit + length;
1824 * - gap_end - gap_start >= length
1827 struct mm_struct
*mm
= current
->mm
;
1828 struct vm_area_struct
*vma
;
1829 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1831 /* Adjust search length to account for worst case alignment overhead */
1832 length
= info
->length
+ info
->align_mask
;
1833 if (length
< info
->length
)
1836 /* Adjust search limits by the desired length */
1837 if (info
->high_limit
< length
)
1839 high_limit
= info
->high_limit
- length
;
1841 if (info
->low_limit
> high_limit
)
1843 low_limit
= info
->low_limit
+ length
;
1845 /* Check if rbtree root looks promising */
1846 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1848 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1849 if (vma
->rb_subtree_gap
< length
)
1853 /* Visit left subtree if it looks promising */
1854 gap_end
= vm_start_gap(vma
);
1855 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1856 struct vm_area_struct
*left
=
1857 rb_entry(vma
->vm_rb
.rb_left
,
1858 struct vm_area_struct
, vm_rb
);
1859 if (left
->rb_subtree_gap
>= length
) {
1865 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1867 /* Check if current node has a suitable gap */
1868 if (gap_start
> high_limit
)
1870 if (gap_end
>= low_limit
&&
1871 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1874 /* Visit right subtree if it looks promising */
1875 if (vma
->vm_rb
.rb_right
) {
1876 struct vm_area_struct
*right
=
1877 rb_entry(vma
->vm_rb
.rb_right
,
1878 struct vm_area_struct
, vm_rb
);
1879 if (right
->rb_subtree_gap
>= length
) {
1885 /* Go back up the rbtree to find next candidate node */
1887 struct rb_node
*prev
= &vma
->vm_rb
;
1888 if (!rb_parent(prev
))
1890 vma
= rb_entry(rb_parent(prev
),
1891 struct vm_area_struct
, vm_rb
);
1892 if (prev
== vma
->vm_rb
.rb_left
) {
1893 gap_start
= vm_end_gap(vma
->vm_prev
);
1894 gap_end
= vm_start_gap(vma
);
1901 /* Check highest gap, which does not precede any rbtree node */
1902 gap_start
= mm
->highest_vm_end
;
1903 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1904 if (gap_start
> high_limit
)
1908 /* We found a suitable gap. Clip it with the original low_limit. */
1909 if (gap_start
< info
->low_limit
)
1910 gap_start
= info
->low_limit
;
1912 /* Adjust gap address to the desired alignment */
1913 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1915 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1916 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1920 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1922 struct mm_struct
*mm
= current
->mm
;
1923 struct vm_area_struct
*vma
;
1924 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1926 /* Adjust search length to account for worst case alignment overhead */
1927 length
= info
->length
+ info
->align_mask
;
1928 if (length
< info
->length
)
1932 * Adjust search limits by the desired length.
1933 * See implementation comment at top of unmapped_area().
1935 gap_end
= info
->high_limit
;
1936 if (gap_end
< length
)
1938 high_limit
= gap_end
- length
;
1940 if (info
->low_limit
> high_limit
)
1942 low_limit
= info
->low_limit
+ length
;
1944 /* Check highest gap, which does not precede any rbtree node */
1945 gap_start
= mm
->highest_vm_end
;
1946 if (gap_start
<= high_limit
)
1949 /* Check if rbtree root looks promising */
1950 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1952 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1953 if (vma
->rb_subtree_gap
< length
)
1957 /* Visit right subtree if it looks promising */
1958 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1959 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
1960 struct vm_area_struct
*right
=
1961 rb_entry(vma
->vm_rb
.rb_right
,
1962 struct vm_area_struct
, vm_rb
);
1963 if (right
->rb_subtree_gap
>= length
) {
1970 /* Check if current node has a suitable gap */
1971 gap_end
= vm_start_gap(vma
);
1972 if (gap_end
< low_limit
)
1974 if (gap_start
<= high_limit
&&
1975 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1978 /* Visit left subtree if it looks promising */
1979 if (vma
->vm_rb
.rb_left
) {
1980 struct vm_area_struct
*left
=
1981 rb_entry(vma
->vm_rb
.rb_left
,
1982 struct vm_area_struct
, vm_rb
);
1983 if (left
->rb_subtree_gap
>= length
) {
1989 /* Go back up the rbtree to find next candidate node */
1991 struct rb_node
*prev
= &vma
->vm_rb
;
1992 if (!rb_parent(prev
))
1994 vma
= rb_entry(rb_parent(prev
),
1995 struct vm_area_struct
, vm_rb
);
1996 if (prev
== vma
->vm_rb
.rb_right
) {
1997 gap_start
= vma
->vm_prev
?
1998 vm_end_gap(vma
->vm_prev
) : 0;
2005 /* We found a suitable gap. Clip it with the original high_limit. */
2006 if (gap_end
> info
->high_limit
)
2007 gap_end
= info
->high_limit
;
2010 /* Compute highest gap address at the desired alignment */
2011 gap_end
-= info
->length
;
2012 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
2014 VM_BUG_ON(gap_end
< info
->low_limit
);
2015 VM_BUG_ON(gap_end
< gap_start
);
2019 /* Get an address range which is currently unmapped.
2020 * For shmat() with addr=0.
2022 * Ugly calling convention alert:
2023 * Return value with the low bits set means error value,
2025 * if (ret & ~PAGE_MASK)
2028 * This function "knows" that -ENOMEM has the bits set.
2030 #ifndef HAVE_ARCH_UNMAPPED_AREA
2032 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
2033 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
2035 struct mm_struct
*mm
= current
->mm
;
2036 struct vm_area_struct
*vma
, *prev
;
2037 struct vm_unmapped_area_info info
;
2039 if (len
> TASK_SIZE
- mmap_min_addr
)
2042 if (flags
& MAP_FIXED
)
2046 addr
= PAGE_ALIGN(addr
);
2047 vma
= find_vma_prev(mm
, addr
, &prev
);
2048 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
2049 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2050 (!prev
|| addr
>= vm_end_gap(prev
)))
2056 info
.low_limit
= mm
->mmap_base
;
2057 info
.high_limit
= TASK_SIZE
;
2058 info
.align_mask
= 0;
2059 return vm_unmapped_area(&info
);
2064 * This mmap-allocator allocates new areas top-down from below the
2065 * stack's low limit (the base):
2067 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2069 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
2070 const unsigned long len
, const unsigned long pgoff
,
2071 const unsigned long flags
)
2073 struct vm_area_struct
*vma
, *prev
;
2074 struct mm_struct
*mm
= current
->mm
;
2075 unsigned long addr
= addr0
;
2076 struct vm_unmapped_area_info info
;
2078 /* requested length too big for entire address space */
2079 if (len
> TASK_SIZE
- mmap_min_addr
)
2082 if (flags
& MAP_FIXED
)
2085 /* requesting a specific address */
2087 addr
= PAGE_ALIGN(addr
);
2088 vma
= find_vma_prev(mm
, addr
, &prev
);
2089 if (TASK_SIZE
- len
>= addr
&& addr
>= mmap_min_addr
&&
2090 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2091 (!prev
|| addr
>= vm_end_gap(prev
)))
2095 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
2097 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
2098 info
.high_limit
= mm
->mmap_base
;
2099 info
.align_mask
= 0;
2100 addr
= vm_unmapped_area(&info
);
2103 * A failed mmap() very likely causes application failure,
2104 * so fall back to the bottom-up function here. This scenario
2105 * can happen with large stack limits and large mmap()
2108 if (offset_in_page(addr
)) {
2109 VM_BUG_ON(addr
!= -ENOMEM
);
2111 info
.low_limit
= TASK_UNMAPPED_BASE
;
2112 info
.high_limit
= TASK_SIZE
;
2113 addr
= vm_unmapped_area(&info
);
2121 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2122 unsigned long pgoff
, unsigned long flags
)
2124 unsigned long (*get_area
)(struct file
*, unsigned long,
2125 unsigned long, unsigned long, unsigned long);
2127 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2131 /* Careful about overflows.. */
2132 if (len
> TASK_SIZE
)
2135 get_area
= current
->mm
->get_unmapped_area
;
2137 if (file
->f_op
->get_unmapped_area
)
2138 get_area
= file
->f_op
->get_unmapped_area
;
2139 } else if (flags
& MAP_SHARED
) {
2141 * mmap_region() will call shmem_zero_setup() to create a file,
2142 * so use shmem's get_unmapped_area in case it can be huge.
2143 * do_mmap_pgoff() will clear pgoff, so match alignment.
2146 get_area
= shmem_get_unmapped_area
;
2149 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2150 if (IS_ERR_VALUE(addr
))
2153 if (addr
> TASK_SIZE
- len
)
2155 if (offset_in_page(addr
))
2158 error
= security_mmap_addr(addr
);
2159 return error
? error
: addr
;
2162 EXPORT_SYMBOL(get_unmapped_area
);
2164 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2165 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2167 struct rb_node
*rb_node
;
2168 struct vm_area_struct
*vma
;
2170 /* Check the cache first. */
2171 vma
= vmacache_find(mm
, addr
);
2175 rb_node
= mm
->mm_rb
.rb_node
;
2178 struct vm_area_struct
*tmp
;
2180 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2182 if (tmp
->vm_end
> addr
) {
2184 if (tmp
->vm_start
<= addr
)
2186 rb_node
= rb_node
->rb_left
;
2188 rb_node
= rb_node
->rb_right
;
2192 vmacache_update(addr
, vma
);
2196 EXPORT_SYMBOL(find_vma
);
2199 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2201 struct vm_area_struct
*
2202 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2203 struct vm_area_struct
**pprev
)
2205 struct vm_area_struct
*vma
;
2207 vma
= find_vma(mm
, addr
);
2209 *pprev
= vma
->vm_prev
;
2211 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2214 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2215 rb_node
= rb_node
->rb_right
;
2222 * Verify that the stack growth is acceptable and
2223 * update accounting. This is shared with both the
2224 * grow-up and grow-down cases.
2226 static int acct_stack_growth(struct vm_area_struct
*vma
,
2227 unsigned long size
, unsigned long grow
)
2229 struct mm_struct
*mm
= vma
->vm_mm
;
2230 unsigned long new_start
;
2232 /* address space limit tests */
2233 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
2236 /* Stack limit test */
2237 if (size
> rlimit(RLIMIT_STACK
))
2240 /* mlock limit tests */
2241 if (vma
->vm_flags
& VM_LOCKED
) {
2242 unsigned long locked
;
2243 unsigned long limit
;
2244 locked
= mm
->locked_vm
+ grow
;
2245 limit
= rlimit(RLIMIT_MEMLOCK
);
2246 limit
>>= PAGE_SHIFT
;
2247 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2251 /* Check to ensure the stack will not grow into a hugetlb-only region */
2252 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2254 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2258 * Overcommit.. This must be the final test, as it will
2259 * update security statistics.
2261 if (security_vm_enough_memory_mm(mm
, grow
))
2267 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2269 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2270 * vma is the last one with address > vma->vm_end. Have to extend vma.
2272 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2274 struct mm_struct
*mm
= vma
->vm_mm
;
2275 struct vm_area_struct
*next
;
2276 unsigned long gap_addr
;
2279 if (!(vma
->vm_flags
& VM_GROWSUP
))
2282 /* Guard against exceeding limits of the address space. */
2283 address
&= PAGE_MASK
;
2284 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2286 address
+= PAGE_SIZE
;
2288 /* Enforce stack_guard_gap */
2289 gap_addr
= address
+ stack_guard_gap
;
2291 /* Guard against overflow */
2292 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2293 gap_addr
= TASK_SIZE
;
2295 next
= vma
->vm_next
;
2296 if (next
&& next
->vm_start
< gap_addr
&&
2297 (next
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2298 if (!(next
->vm_flags
& VM_GROWSUP
))
2300 /* Check that both stack segments have the same anon_vma? */
2303 /* We must make sure the anon_vma is allocated. */
2304 if (unlikely(anon_vma_prepare(vma
)))
2308 * vma->vm_start/vm_end cannot change under us because the caller
2309 * is required to hold the mmap_sem in read mode. We need the
2310 * anon_vma lock to serialize against concurrent expand_stacks.
2312 anon_vma_lock_write(vma
->anon_vma
);
2314 /* Somebody else might have raced and expanded it already */
2315 if (address
> vma
->vm_end
) {
2316 unsigned long size
, grow
;
2318 size
= address
- vma
->vm_start
;
2319 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2322 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2323 error
= acct_stack_growth(vma
, size
, grow
);
2326 * vma_gap_update() doesn't support concurrent
2327 * updates, but we only hold a shared mmap_sem
2328 * lock here, so we need to protect against
2329 * concurrent vma expansions.
2330 * anon_vma_lock_write() doesn't help here, as
2331 * we don't guarantee that all growable vmas
2332 * in a mm share the same root anon vma.
2333 * So, we reuse mm->page_table_lock to guard
2334 * against concurrent vma expansions.
2336 spin_lock(&mm
->page_table_lock
);
2337 if (vma
->vm_flags
& VM_LOCKED
)
2338 mm
->locked_vm
+= grow
;
2339 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2340 anon_vma_interval_tree_pre_update_vma(vma
);
2341 vma
->vm_end
= address
;
2342 anon_vma_interval_tree_post_update_vma(vma
);
2344 vma_gap_update(vma
->vm_next
);
2346 mm
->highest_vm_end
= vm_end_gap(vma
);
2347 spin_unlock(&mm
->page_table_lock
);
2349 perf_event_mmap(vma
);
2353 anon_vma_unlock_write(vma
->anon_vma
);
2354 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2358 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2361 * vma is the first one with address < vma->vm_start. Have to extend vma.
2363 int expand_downwards(struct vm_area_struct
*vma
,
2364 unsigned long address
)
2366 struct mm_struct
*mm
= vma
->vm_mm
;
2367 struct vm_area_struct
*prev
;
2370 address
&= PAGE_MASK
;
2371 if (address
< mmap_min_addr
)
2374 /* Enforce stack_guard_gap */
2375 prev
= vma
->vm_prev
;
2376 /* Check that both stack segments have the same anon_vma? */
2377 if (prev
&& !(prev
->vm_flags
& VM_GROWSDOWN
) &&
2378 (prev
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2379 if (address
- prev
->vm_end
< stack_guard_gap
)
2383 /* We must make sure the anon_vma is allocated. */
2384 if (unlikely(anon_vma_prepare(vma
)))
2388 * vma->vm_start/vm_end cannot change under us because the caller
2389 * is required to hold the mmap_sem in read mode. We need the
2390 * anon_vma lock to serialize against concurrent expand_stacks.
2392 anon_vma_lock_write(vma
->anon_vma
);
2394 /* Somebody else might have raced and expanded it already */
2395 if (address
< vma
->vm_start
) {
2396 unsigned long size
, grow
;
2398 size
= vma
->vm_end
- address
;
2399 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2402 if (grow
<= vma
->vm_pgoff
) {
2403 error
= acct_stack_growth(vma
, size
, grow
);
2406 * vma_gap_update() doesn't support concurrent
2407 * updates, but we only hold a shared mmap_sem
2408 * lock here, so we need to protect against
2409 * concurrent vma expansions.
2410 * anon_vma_lock_write() doesn't help here, as
2411 * we don't guarantee that all growable vmas
2412 * in a mm share the same root anon vma.
2413 * So, we reuse mm->page_table_lock to guard
2414 * against concurrent vma expansions.
2416 spin_lock(&mm
->page_table_lock
);
2417 if (vma
->vm_flags
& VM_LOCKED
)
2418 mm
->locked_vm
+= grow
;
2419 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2420 anon_vma_interval_tree_pre_update_vma(vma
);
2421 vma
->vm_start
= address
;
2422 vma
->vm_pgoff
-= grow
;
2423 anon_vma_interval_tree_post_update_vma(vma
);
2424 vma_gap_update(vma
);
2425 spin_unlock(&mm
->page_table_lock
);
2427 perf_event_mmap(vma
);
2431 anon_vma_unlock_write(vma
->anon_vma
);
2432 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2437 /* enforced gap between the expanding stack and other mappings. */
2438 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2440 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2445 val
= simple_strtoul(p
, &endptr
, 10);
2447 stack_guard_gap
= val
<< PAGE_SHIFT
;
2451 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2453 #ifdef CONFIG_STACK_GROWSUP
2454 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2456 return expand_upwards(vma
, address
);
2459 struct vm_area_struct
*
2460 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2462 struct vm_area_struct
*vma
, *prev
;
2465 vma
= find_vma_prev(mm
, addr
, &prev
);
2466 if (vma
&& (vma
->vm_start
<= addr
))
2468 /* don't alter vm_end if the coredump is running */
2469 if (!prev
|| !mmget_still_valid(mm
) || expand_stack(prev
, addr
))
2471 if (prev
->vm_flags
& VM_LOCKED
)
2472 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2476 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2478 return expand_downwards(vma
, address
);
2481 struct vm_area_struct
*
2482 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2484 struct vm_area_struct
*vma
;
2485 unsigned long start
;
2488 vma
= find_vma(mm
, addr
);
2491 if (vma
->vm_start
<= addr
)
2493 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2495 /* don't alter vm_start if the coredump is running */
2496 if (!mmget_still_valid(mm
))
2498 start
= vma
->vm_start
;
2499 if (expand_stack(vma
, addr
))
2501 if (vma
->vm_flags
& VM_LOCKED
)
2502 populate_vma_page_range(vma
, addr
, start
, NULL
);
2507 EXPORT_SYMBOL_GPL(find_extend_vma
);
2510 * Ok - we have the memory areas we should free on the vma list,
2511 * so release them, and do the vma updates.
2513 * Called with the mm semaphore held.
2515 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2517 unsigned long nr_accounted
= 0;
2519 /* Update high watermark before we lower total_vm */
2520 update_hiwater_vm(mm
);
2522 long nrpages
= vma_pages(vma
);
2524 if (vma
->vm_flags
& VM_ACCOUNT
)
2525 nr_accounted
+= nrpages
;
2526 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2527 vma
= remove_vma(vma
);
2529 vm_unacct_memory(nr_accounted
);
2534 * Get rid of page table information in the indicated region.
2536 * Called with the mm semaphore held.
2538 static void unmap_region(struct mm_struct
*mm
,
2539 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2540 unsigned long start
, unsigned long end
)
2542 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2543 struct mmu_gather tlb
;
2546 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2547 update_hiwater_rss(mm
);
2548 unmap_vmas(&tlb
, vma
, start
, end
);
2549 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2550 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2551 tlb_finish_mmu(&tlb
, start
, end
);
2555 * Create a list of vma's touched by the unmap, removing them from the mm's
2556 * vma list as we go..
2559 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2560 struct vm_area_struct
*prev
, unsigned long end
)
2562 struct vm_area_struct
**insertion_point
;
2563 struct vm_area_struct
*tail_vma
= NULL
;
2565 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2566 vma
->vm_prev
= NULL
;
2568 vma_rb_erase(vma
, &mm
->mm_rb
);
2572 } while (vma
&& vma
->vm_start
< end
);
2573 *insertion_point
= vma
;
2575 vma
->vm_prev
= prev
;
2576 vma_gap_update(vma
);
2578 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2579 tail_vma
->vm_next
= NULL
;
2581 /* Kill the cache */
2582 vmacache_invalidate(mm
);
2586 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2587 * has already been checked or doesn't make sense to fail.
2589 int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2590 unsigned long addr
, int new_below
)
2592 struct vm_area_struct
*new;
2595 if (vma
->vm_ops
&& vma
->vm_ops
->split
) {
2596 err
= vma
->vm_ops
->split(vma
, addr
);
2601 new = vm_area_dup(vma
);
2608 new->vm_start
= addr
;
2609 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2612 err
= vma_dup_policy(vma
, new);
2616 err
= anon_vma_clone(new, vma
);
2623 if (new->vm_ops
&& new->vm_ops
->open
)
2624 new->vm_ops
->open(new);
2627 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2628 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2630 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2636 /* Clean everything up if vma_adjust failed. */
2637 if (new->vm_ops
&& new->vm_ops
->close
)
2638 new->vm_ops
->close(new);
2641 unlink_anon_vmas(new);
2643 mpol_put(vma_policy(new));
2650 * Split a vma into two pieces at address 'addr', a new vma is allocated
2651 * either for the first part or the tail.
2653 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2654 unsigned long addr
, int new_below
)
2656 if (mm
->map_count
>= sysctl_max_map_count
)
2659 return __split_vma(mm
, vma
, addr
, new_below
);
2662 /* Munmap is split into 2 main parts -- this part which finds
2663 * what needs doing, and the areas themselves, which do the
2664 * work. This now handles partial unmappings.
2665 * Jeremy Fitzhardinge <jeremy@goop.org>
2667 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2668 struct list_head
*uf
)
2671 struct vm_area_struct
*vma
, *prev
, *last
;
2673 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2676 len
= PAGE_ALIGN(len
);
2680 /* Find the first overlapping VMA */
2681 vma
= find_vma(mm
, start
);
2684 prev
= vma
->vm_prev
;
2685 /* we have start < vma->vm_end */
2687 /* if it doesn't overlap, we have nothing.. */
2689 if (vma
->vm_start
>= end
)
2693 * If we need to split any vma, do it now to save pain later.
2695 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2696 * unmapped vm_area_struct will remain in use: so lower split_vma
2697 * places tmp vma above, and higher split_vma places tmp vma below.
2699 if (start
> vma
->vm_start
) {
2703 * Make sure that map_count on return from munmap() will
2704 * not exceed its limit; but let map_count go just above
2705 * its limit temporarily, to help free resources as expected.
2707 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2710 error
= __split_vma(mm
, vma
, start
, 0);
2716 /* Does it split the last one? */
2717 last
= find_vma(mm
, end
);
2718 if (last
&& end
> last
->vm_start
) {
2719 int error
= __split_vma(mm
, last
, end
, 1);
2723 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2727 * If userfaultfd_unmap_prep returns an error the vmas
2728 * will remain splitted, but userland will get a
2729 * highly unexpected error anyway. This is no
2730 * different than the case where the first of the two
2731 * __split_vma fails, but we don't undo the first
2732 * split, despite we could. This is unlikely enough
2733 * failure that it's not worth optimizing it for.
2735 int error
= userfaultfd_unmap_prep(vma
, start
, end
, uf
);
2741 * unlock any mlock()ed ranges before detaching vmas
2743 if (mm
->locked_vm
) {
2744 struct vm_area_struct
*tmp
= vma
;
2745 while (tmp
&& tmp
->vm_start
< end
) {
2746 if (tmp
->vm_flags
& VM_LOCKED
) {
2747 mm
->locked_vm
-= vma_pages(tmp
);
2748 munlock_vma_pages_all(tmp
);
2755 * Remove the vma's, and unmap the actual pages
2757 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2758 unmap_region(mm
, vma
, prev
, start
, end
);
2760 arch_unmap(mm
, vma
, start
, end
);
2762 /* Fix up all other VM information */
2763 remove_vma_list(mm
, vma
);
2768 int vm_munmap(unsigned long start
, size_t len
)
2771 struct mm_struct
*mm
= current
->mm
;
2774 if (down_write_killable(&mm
->mmap_sem
))
2777 ret
= do_munmap(mm
, start
, len
, &uf
);
2778 up_write(&mm
->mmap_sem
);
2779 userfaultfd_unmap_complete(mm
, &uf
);
2782 EXPORT_SYMBOL(vm_munmap
);
2784 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2786 profile_munmap(addr
);
2787 return vm_munmap(addr
, len
);
2792 * Emulation of deprecated remap_file_pages() syscall.
2794 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2795 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2798 struct mm_struct
*mm
= current
->mm
;
2799 struct vm_area_struct
*vma
;
2800 unsigned long populate
= 0;
2801 unsigned long ret
= -EINVAL
;
2802 struct file
*file
, *prfile
;
2804 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2805 current
->comm
, current
->pid
);
2809 start
= start
& PAGE_MASK
;
2810 size
= size
& PAGE_MASK
;
2812 if (start
+ size
<= start
)
2815 /* Does pgoff wrap? */
2816 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2819 if (down_write_killable(&mm
->mmap_sem
))
2822 vma
= find_vma(mm
, start
);
2824 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2827 if (start
< vma
->vm_start
)
2830 if (start
+ size
> vma
->vm_end
) {
2831 struct vm_area_struct
*next
;
2833 for (next
= vma
->vm_next
; next
; next
= next
->vm_next
) {
2834 /* hole between vmas ? */
2835 if (next
->vm_start
!= next
->vm_prev
->vm_end
)
2838 if (next
->vm_file
!= vma
->vm_file
)
2841 if (next
->vm_flags
!= vma
->vm_flags
)
2844 if (start
+ size
<= next
->vm_end
)
2852 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
2853 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
2854 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
2856 flags
&= MAP_NONBLOCK
;
2857 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
2858 if (vma
->vm_flags
& VM_LOCKED
) {
2859 struct vm_area_struct
*tmp
;
2860 flags
|= MAP_LOCKED
;
2862 /* drop PG_Mlocked flag for over-mapped range */
2863 for (tmp
= vma
; tmp
->vm_start
>= start
+ size
;
2864 tmp
= tmp
->vm_next
) {
2866 * Split pmd and munlock page on the border
2869 vma_adjust_trans_huge(tmp
, start
, start
+ size
, 0);
2871 munlock_vma_pages_range(tmp
,
2872 max(tmp
->vm_start
, start
),
2873 min(tmp
->vm_end
, start
+ size
));
2878 file
= vma
->vm_file
;
2879 prfile
= vma
->vm_prfile
;
2880 ret
= do_mmap_pgoff(vma
->vm_file
, start
, size
,
2881 prot
, flags
, pgoff
, &populate
, NULL
);
2882 if (!IS_ERR_VALUE(ret
) && file
&& prfile
) {
2883 struct vm_area_struct
*new_vma
;
2885 new_vma
= find_vma(mm
, ret
);
2886 if (!new_vma
->vm_prfile
)
2887 new_vma
->vm_prfile
= prfile
;
2892 * two fput()s instead of vma_fput(vma),
2893 * coz vma may not be available anymore.
2899 up_write(&mm
->mmap_sem
);
2901 mm_populate(ret
, populate
);
2902 if (!IS_ERR_VALUE(ret
))
2907 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2909 #ifdef CONFIG_DEBUG_VM
2910 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2912 up_read(&mm
->mmap_sem
);
2918 * this is really a simplified "do_mmap". it only handles
2919 * anonymous maps. eventually we may be able to do some
2920 * brk-specific accounting here.
2922 static int do_brk_flags(unsigned long addr
, unsigned long len
, unsigned long flags
, struct list_head
*uf
)
2924 struct mm_struct
*mm
= current
->mm
;
2925 struct vm_area_struct
*vma
, *prev
;
2926 struct rb_node
**rb_link
, *rb_parent
;
2927 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2930 /* Until we need other flags, refuse anything except VM_EXEC. */
2931 if ((flags
& (~VM_EXEC
)) != 0)
2933 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2935 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2936 if (offset_in_page(error
))
2939 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
2944 * mm->mmap_sem is required to protect against another thread
2945 * changing the mappings in case we sleep.
2947 verify_mm_writelocked(mm
);
2950 * Clear old maps. this also does some error checking for us
2952 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
2954 if (do_munmap(mm
, addr
, len
, uf
))
2958 /* Check against address space limits *after* clearing old maps... */
2959 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
2962 if (mm
->map_count
> sysctl_max_map_count
)
2965 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2968 /* Can we just expand an old private anonymous mapping? */
2969 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2970 NULL
, NULL
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
2975 * create a vma struct for an anonymous mapping
2977 vma
= vm_area_alloc(mm
);
2979 vm_unacct_memory(len
>> PAGE_SHIFT
);
2983 vma
->vm_start
= addr
;
2984 vma
->vm_end
= addr
+ len
;
2985 vma
->vm_pgoff
= pgoff
;
2986 vma
->vm_flags
= flags
;
2987 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2988 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2990 perf_event_mmap(vma
);
2991 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2992 mm
->data_vm
+= len
>> PAGE_SHIFT
;
2993 if (flags
& VM_LOCKED
)
2994 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2995 vma
->vm_flags
|= VM_SOFTDIRTY
;
2999 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
3001 struct mm_struct
*mm
= current
->mm
;
3007 len
= PAGE_ALIGN(request
);
3013 if (down_write_killable(&mm
->mmap_sem
))
3016 ret
= do_brk_flags(addr
, len
, flags
, &uf
);
3017 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3018 up_write(&mm
->mmap_sem
);
3019 userfaultfd_unmap_complete(mm
, &uf
);
3020 if (populate
&& !ret
)
3021 mm_populate(addr
, len
);
3024 EXPORT_SYMBOL(vm_brk_flags
);
3026 int vm_brk(unsigned long addr
, unsigned long len
)
3028 return vm_brk_flags(addr
, len
, 0);
3030 EXPORT_SYMBOL(vm_brk
);
3032 /* Release all mmaps. */
3033 void exit_mmap(struct mm_struct
*mm
)
3035 struct mmu_gather tlb
;
3036 struct vm_area_struct
*vma
;
3037 unsigned long nr_accounted
= 0;
3039 /* mm's last user has gone, and its about to be pulled down */
3040 mmu_notifier_release(mm
);
3042 if (unlikely(mm_is_oom_victim(mm
))) {
3044 * Manually reap the mm to free as much memory as possible.
3045 * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3046 * this mm from further consideration. Taking mm->mmap_sem for
3047 * write after setting MMF_OOM_SKIP will guarantee that the oom
3048 * reaper will not run on this mm again after mmap_sem is
3051 * Nothing can be holding mm->mmap_sem here and the above call
3052 * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3053 * __oom_reap_task_mm() will not block.
3055 * This needs to be done before calling munlock_vma_pages_all(),
3056 * which clears VM_LOCKED, otherwise the oom reaper cannot
3059 mutex_lock(&oom_lock
);
3060 __oom_reap_task_mm(mm
);
3061 mutex_unlock(&oom_lock
);
3063 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3064 down_write(&mm
->mmap_sem
);
3065 up_write(&mm
->mmap_sem
);
3068 if (mm
->locked_vm
) {
3071 if (vma
->vm_flags
& VM_LOCKED
)
3072 munlock_vma_pages_all(vma
);
3080 if (!vma
) /* Can happen if dup_mmap() received an OOM */
3085 tlb_gather_mmu(&tlb
, mm
, 0, -1);
3086 /* update_hiwater_rss(mm) here? but nobody should be looking */
3087 /* Use -1 here to ensure all VMAs in the mm are unmapped */
3088 unmap_vmas(&tlb
, vma
, 0, -1);
3089 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
3090 tlb_finish_mmu(&tlb
, 0, -1);
3093 * Walk the list again, actually closing and freeing it,
3094 * with preemption enabled, without holding any MM locks.
3097 if (vma
->vm_flags
& VM_ACCOUNT
)
3098 nr_accounted
+= vma_pages(vma
);
3099 vma
= remove_vma(vma
);
3101 vm_unacct_memory(nr_accounted
);
3104 /* Insert vm structure into process list sorted by address
3105 * and into the inode's i_mmap tree. If vm_file is non-NULL
3106 * then i_mmap_rwsem is taken here.
3108 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3110 struct vm_area_struct
*prev
;
3111 struct rb_node
**rb_link
, *rb_parent
;
3113 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
3114 &prev
, &rb_link
, &rb_parent
))
3116 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3117 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
3121 * The vm_pgoff of a purely anonymous vma should be irrelevant
3122 * until its first write fault, when page's anon_vma and index
3123 * are set. But now set the vm_pgoff it will almost certainly
3124 * end up with (unless mremap moves it elsewhere before that
3125 * first wfault), so /proc/pid/maps tells a consistent story.
3127 * By setting it to reflect the virtual start address of the
3128 * vma, merges and splits can happen in a seamless way, just
3129 * using the existing file pgoff checks and manipulations.
3130 * Similarly in do_mmap_pgoff and in do_brk.
3132 if (vma_is_anonymous(vma
)) {
3133 BUG_ON(vma
->anon_vma
);
3134 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3137 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3142 * Copy the vma structure to a new location in the same mm,
3143 * prior to moving page table entries, to effect an mremap move.
3145 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3146 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3147 bool *need_rmap_locks
)
3149 struct vm_area_struct
*vma
= *vmap
;
3150 unsigned long vma_start
= vma
->vm_start
;
3151 struct mm_struct
*mm
= vma
->vm_mm
;
3152 struct vm_area_struct
*new_vma
, *prev
;
3153 struct rb_node
**rb_link
, *rb_parent
;
3154 bool faulted_in_anon_vma
= true;
3157 * If anonymous vma has not yet been faulted, update new pgoff
3158 * to match new location, to increase its chance of merging.
3160 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3161 pgoff
= addr
>> PAGE_SHIFT
;
3162 faulted_in_anon_vma
= false;
3165 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
3166 return NULL
; /* should never get here */
3167 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
3168 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
3169 vma
->vm_userfaultfd_ctx
);
3172 * Source vma may have been merged into new_vma
3174 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3175 vma_start
< new_vma
->vm_end
)) {
3177 * The only way we can get a vma_merge with
3178 * self during an mremap is if the vma hasn't
3179 * been faulted in yet and we were allowed to
3180 * reset the dst vma->vm_pgoff to the
3181 * destination address of the mremap to allow
3182 * the merge to happen. mremap must change the
3183 * vm_pgoff linearity between src and dst vmas
3184 * (in turn preventing a vma_merge) to be
3185 * safe. It is only safe to keep the vm_pgoff
3186 * linear if there are no pages mapped yet.
3188 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3189 *vmap
= vma
= new_vma
;
3191 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3193 new_vma
= vm_area_dup(vma
);
3196 new_vma
->vm_start
= addr
;
3197 new_vma
->vm_end
= addr
+ len
;
3198 new_vma
->vm_pgoff
= pgoff
;
3199 if (vma_dup_policy(vma
, new_vma
))
3201 if (anon_vma_clone(new_vma
, vma
))
3202 goto out_free_mempol
;
3203 if (new_vma
->vm_file
)
3204 vma_get_file(new_vma
);
3205 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3206 new_vma
->vm_ops
->open(new_vma
);
3207 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
3208 *need_rmap_locks
= false;
3213 mpol_put(vma_policy(new_vma
));
3215 vm_area_free(new_vma
);
3221 * Return true if the calling process may expand its vm space by the passed
3224 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3226 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3229 if (is_data_mapping(flags
) &&
3230 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3231 /* Workaround for Valgrind */
3232 if (rlimit(RLIMIT_DATA
) == 0 &&
3233 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3235 if (!ignore_rlimit_data
) {
3236 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n",
3237 current
->comm
, current
->pid
,
3238 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3239 rlimit(RLIMIT_DATA
));
3247 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3249 mm
->total_vm
+= npages
;
3251 if (is_exec_mapping(flags
))
3252 mm
->exec_vm
+= npages
;
3253 else if (is_stack_mapping(flags
))
3254 mm
->stack_vm
+= npages
;
3255 else if (is_data_mapping(flags
))
3256 mm
->data_vm
+= npages
;
3259 static int special_mapping_fault(struct vm_fault
*vmf
);
3262 * Having a close hook prevents vma merging regardless of flags.
3264 static void special_mapping_close(struct vm_area_struct
*vma
)
3268 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3270 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3273 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3275 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3277 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3281 return sm
->mremap(sm
, new_vma
);
3286 static const struct vm_operations_struct special_mapping_vmops
= {
3287 .close
= special_mapping_close
,
3288 .fault
= special_mapping_fault
,
3289 .mremap
= special_mapping_mremap
,
3290 .name
= special_mapping_name
,
3293 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3294 .close
= special_mapping_close
,
3295 .fault
= special_mapping_fault
,
3298 static int special_mapping_fault(struct vm_fault
*vmf
)
3300 struct vm_area_struct
*vma
= vmf
->vma
;
3302 struct page
**pages
;
3304 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3305 pages
= vma
->vm_private_data
;
3307 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3310 return sm
->fault(sm
, vmf
->vma
, vmf
);
3315 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3319 struct page
*page
= *pages
;
3325 return VM_FAULT_SIGBUS
;
3328 static struct vm_area_struct
*__install_special_mapping(
3329 struct mm_struct
*mm
,
3330 unsigned long addr
, unsigned long len
,
3331 unsigned long vm_flags
, void *priv
,
3332 const struct vm_operations_struct
*ops
)
3335 struct vm_area_struct
*vma
;
3337 vma
= vm_area_alloc(mm
);
3338 if (unlikely(vma
== NULL
))
3339 return ERR_PTR(-ENOMEM
);
3341 vma
->vm_start
= addr
;
3342 vma
->vm_end
= addr
+ len
;
3344 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
3345 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3348 vma
->vm_private_data
= priv
;
3350 ret
= insert_vm_struct(mm
, vma
);
3354 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3356 perf_event_mmap(vma
);
3362 return ERR_PTR(ret
);
3365 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3366 const struct vm_special_mapping
*sm
)
3368 return vma
->vm_private_data
== sm
&&
3369 (vma
->vm_ops
== &special_mapping_vmops
||
3370 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3374 * Called with mm->mmap_sem held for writing.
3375 * Insert a new vma covering the given region, with the given flags.
3376 * Its pages are supplied by the given array of struct page *.
3377 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3378 * The region past the last page supplied will always produce SIGBUS.
3379 * The array pointer and the pages it points to are assumed to stay alive
3380 * for as long as this mapping might exist.
3382 struct vm_area_struct
*_install_special_mapping(
3383 struct mm_struct
*mm
,
3384 unsigned long addr
, unsigned long len
,
3385 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3387 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3388 &special_mapping_vmops
);
3391 int install_special_mapping(struct mm_struct
*mm
,
3392 unsigned long addr
, unsigned long len
,
3393 unsigned long vm_flags
, struct page
**pages
)
3395 struct vm_area_struct
*vma
= __install_special_mapping(
3396 mm
, addr
, len
, vm_flags
, (void *)pages
,
3397 &legacy_special_mapping_vmops
);
3399 return PTR_ERR_OR_ZERO(vma
);
3402 static DEFINE_MUTEX(mm_all_locks_mutex
);
3404 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3406 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3408 * The LSB of head.next can't change from under us
3409 * because we hold the mm_all_locks_mutex.
3411 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3413 * We can safely modify head.next after taking the
3414 * anon_vma->root->rwsem. If some other vma in this mm shares
3415 * the same anon_vma we won't take it again.
3417 * No need of atomic instructions here, head.next
3418 * can't change from under us thanks to the
3419 * anon_vma->root->rwsem.
3421 if (__test_and_set_bit(0, (unsigned long *)
3422 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3427 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3429 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3431 * AS_MM_ALL_LOCKS can't change from under us because
3432 * we hold the mm_all_locks_mutex.
3434 * Operations on ->flags have to be atomic because
3435 * even if AS_MM_ALL_LOCKS is stable thanks to the
3436 * mm_all_locks_mutex, there may be other cpus
3437 * changing other bitflags in parallel to us.
3439 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3441 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_sem
);
3446 * This operation locks against the VM for all pte/vma/mm related
3447 * operations that could ever happen on a certain mm. This includes
3448 * vmtruncate, try_to_unmap, and all page faults.
3450 * The caller must take the mmap_sem in write mode before calling
3451 * mm_take_all_locks(). The caller isn't allowed to release the
3452 * mmap_sem until mm_drop_all_locks() returns.
3454 * mmap_sem in write mode is required in order to block all operations
3455 * that could modify pagetables and free pages without need of
3456 * altering the vma layout. It's also needed in write mode to avoid new
3457 * anon_vmas to be associated with existing vmas.
3459 * A single task can't take more than one mm_take_all_locks() in a row
3460 * or it would deadlock.
3462 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3463 * mapping->flags avoid to take the same lock twice, if more than one
3464 * vma in this mm is backed by the same anon_vma or address_space.
3466 * We take locks in following order, accordingly to comment at beginning
3468 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3470 * - all i_mmap_rwsem locks;
3471 * - all anon_vma->rwseml
3473 * We can take all locks within these types randomly because the VM code
3474 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3475 * mm_all_locks_mutex.
3477 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3478 * that may have to take thousand of locks.
3480 * mm_take_all_locks() can fail if it's interrupted by signals.
3482 int mm_take_all_locks(struct mm_struct
*mm
)
3484 struct vm_area_struct
*vma
;
3485 struct anon_vma_chain
*avc
;
3487 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3489 mutex_lock(&mm_all_locks_mutex
);
3491 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3492 if (signal_pending(current
))
3494 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3495 is_vm_hugetlb_page(vma
))
3496 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3499 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3500 if (signal_pending(current
))
3502 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3503 !is_vm_hugetlb_page(vma
))
3504 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3507 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3508 if (signal_pending(current
))
3511 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3512 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3518 mm_drop_all_locks(mm
);
3522 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3524 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3526 * The LSB of head.next can't change to 0 from under
3527 * us because we hold the mm_all_locks_mutex.
3529 * We must however clear the bitflag before unlocking
3530 * the vma so the users using the anon_vma->rb_root will
3531 * never see our bitflag.
3533 * No need of atomic instructions here, head.next
3534 * can't change from under us until we release the
3535 * anon_vma->root->rwsem.
3537 if (!__test_and_clear_bit(0, (unsigned long *)
3538 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3540 anon_vma_unlock_write(anon_vma
);
3544 static void vm_unlock_mapping(struct address_space
*mapping
)
3546 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3548 * AS_MM_ALL_LOCKS can't change to 0 from under us
3549 * because we hold the mm_all_locks_mutex.
3551 i_mmap_unlock_write(mapping
);
3552 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3559 * The mmap_sem cannot be released by the caller until
3560 * mm_drop_all_locks() returns.
3562 void mm_drop_all_locks(struct mm_struct
*mm
)
3564 struct vm_area_struct
*vma
;
3565 struct anon_vma_chain
*avc
;
3567 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3568 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3570 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3572 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3573 vm_unlock_anon_vma(avc
->anon_vma
);
3574 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3575 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3578 mutex_unlock(&mm_all_locks_mutex
);
3582 * initialise the percpu counter for VM
3584 void __init
mmap_init(void)
3588 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3593 * Initialise sysctl_user_reserve_kbytes.
3595 * This is intended to prevent a user from starting a single memory hogging
3596 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3599 * The default value is min(3% of free memory, 128MB)
3600 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3602 static int init_user_reserve(void)
3604 unsigned long free_kbytes
;
3606 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3608 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3611 subsys_initcall(init_user_reserve
);
3614 * Initialise sysctl_admin_reserve_kbytes.
3616 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3617 * to log in and kill a memory hogging process.
3619 * Systems with more than 256MB will reserve 8MB, enough to recover
3620 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3621 * only reserve 3% of free pages by default.
3623 static int init_admin_reserve(void)
3625 unsigned long free_kbytes
;
3627 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3629 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3632 subsys_initcall(init_admin_reserve
);
3635 * Reinititalise user and admin reserves if memory is added or removed.
3637 * The default user reserve max is 128MB, and the default max for the
3638 * admin reserve is 8MB. These are usually, but not always, enough to
3639 * enable recovery from a memory hogging process using login/sshd, a shell,
3640 * and tools like top. It may make sense to increase or even disable the
3641 * reserve depending on the existence of swap or variations in the recovery
3642 * tools. So, the admin may have changed them.
3644 * If memory is added and the reserves have been eliminated or increased above
3645 * the default max, then we'll trust the admin.
3647 * If memory is removed and there isn't enough free memory, then we
3648 * need to reset the reserves.
3650 * Otherwise keep the reserve set by the admin.
3652 static int reserve_mem_notifier(struct notifier_block
*nb
,
3653 unsigned long action
, void *data
)
3655 unsigned long tmp
, free_kbytes
;
3659 /* Default max is 128MB. Leave alone if modified by operator. */
3660 tmp
= sysctl_user_reserve_kbytes
;
3661 if (0 < tmp
&& tmp
< (1UL << 17))
3662 init_user_reserve();
3664 /* Default max is 8MB. Leave alone if modified by operator. */
3665 tmp
= sysctl_admin_reserve_kbytes
;
3666 if (0 < tmp
&& tmp
< (1UL << 13))
3667 init_admin_reserve();
3671 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3673 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3674 init_user_reserve();
3675 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3676 sysctl_user_reserve_kbytes
);
3679 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3680 init_admin_reserve();
3681 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3682 sysctl_admin_reserve_kbytes
);
3691 static struct notifier_block reserve_mem_nb
= {
3692 .notifier_call
= reserve_mem_notifier
,
3695 static int __meminit
init_reserve_notifier(void)
3697 if (register_hotmemory_notifier(&reserve_mem_nb
))
3698 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3702 subsys_initcall(init_reserve_notifier
);