1 // SPDX-License-Identifier: GPL-2.0-only
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/backing-dev.h>
16 #include <linux/vmacache.h>
17 #include <linux/shm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/syscalls.h>
22 #include <linux/capability.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/hugetlb.h>
29 #include <linux/shmem_fs.h>
30 #include <linux/profile.h>
31 #include <linux/export.h>
32 #include <linux/mount.h>
33 #include <linux/mempolicy.h>
34 #include <linux/rmap.h>
35 #include <linux/mmu_notifier.h>
36 #include <linux/mmdebug.h>
37 #include <linux/perf_event.h>
38 #include <linux/audit.h>
39 #include <linux/khugepaged.h>
40 #include <linux/uprobes.h>
41 #include <linux/rbtree_augmented.h>
42 #include <linux/notifier.h>
43 #include <linux/memory.h>
44 #include <linux/printk.h>
45 #include <linux/userfaultfd_k.h>
46 #include <linux/moduleparam.h>
47 #include <linux/pkeys.h>
48 #include <linux/oom.h>
49 #include <linux/sched/mm.h>
51 #include <linux/uaccess.h>
52 #include <asm/cacheflush.h>
54 #include <asm/mmu_context.h>
58 #ifndef arch_mmap_check
59 #define arch_mmap_check(addr, len, flags) (0)
62 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
64 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
65 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
67 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
69 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
70 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
73 static bool ignore_rlimit_data
;
74 core_param(ignore_rlimit_data
, ignore_rlimit_data
, bool, 0644);
76 static void unmap_region(struct mm_struct
*mm
,
77 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
78 unsigned long start
, unsigned long end
);
80 /* description of effects of mapping type and prot in current implementation.
81 * this is due to the limited x86 page protection hardware. The expected
82 * behavior is in parens:
85 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
90 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
94 pgprot_t protection_map
[16] __ro_after_init
= {
95 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
96 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
99 #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
100 static inline pgprot_t
arch_filter_pgprot(pgprot_t prot
)
106 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
108 pgprot_t ret
= __pgprot(pgprot_val(protection_map
[vm_flags
&
109 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
110 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
112 return arch_filter_pgprot(ret
);
114 EXPORT_SYMBOL(vm_get_page_prot
);
116 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
118 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
121 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
122 void vma_set_page_prot(struct vm_area_struct
*vma
)
124 unsigned long vm_flags
= vma
->vm_flags
;
125 pgprot_t vm_page_prot
;
127 vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
128 if (vma_wants_writenotify(vma
, vm_page_prot
)) {
129 vm_flags
&= ~VM_SHARED
;
130 vm_page_prot
= vm_pgprot_modify(vm_page_prot
, vm_flags
);
132 /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
133 WRITE_ONCE(vma
->vm_page_prot
, vm_page_prot
);
137 * Requires inode->i_mapping->i_mmap_rwsem
139 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
140 struct file
*file
, struct address_space
*mapping
)
142 if (vma
->vm_flags
& VM_DENYWRITE
)
143 atomic_inc(&file_inode(file
)->i_writecount
);
144 if (vma
->vm_flags
& VM_SHARED
)
145 mapping_unmap_writable(mapping
);
147 flush_dcache_mmap_lock(mapping
);
148 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
149 flush_dcache_mmap_unlock(mapping
);
153 * Unlink a file-based vm structure from its interval tree, to hide
154 * vma from rmap and vmtruncate before freeing its page tables.
156 void unlink_file_vma(struct vm_area_struct
*vma
)
158 struct file
*file
= vma
->vm_file
;
161 struct address_space
*mapping
= file
->f_mapping
;
162 i_mmap_lock_write(mapping
);
163 __remove_shared_vm_struct(vma
, file
, mapping
);
164 i_mmap_unlock_write(mapping
);
169 * Close a vm structure and free it, returning the next.
171 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
173 struct vm_area_struct
*next
= vma
->vm_next
;
176 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
177 vma
->vm_ops
->close(vma
);
180 mpol_put(vma_policy(vma
));
185 static int do_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
,
186 struct list_head
*uf
);
187 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
189 unsigned long retval
;
190 unsigned long newbrk
, oldbrk
, origbrk
;
191 struct mm_struct
*mm
= current
->mm
;
192 struct vm_area_struct
*next
;
193 unsigned long min_brk
;
195 bool downgraded
= false;
198 if (down_write_killable(&mm
->mmap_sem
))
203 #ifdef CONFIG_COMPAT_BRK
205 * CONFIG_COMPAT_BRK can still be overridden by setting
206 * randomize_va_space to 2, which will still cause mm->start_brk
207 * to be arbitrarily shifted
209 if (current
->brk_randomized
)
210 min_brk
= mm
->start_brk
;
212 min_brk
= mm
->end_data
;
214 min_brk
= mm
->start_brk
;
220 * Check against rlimit here. If this check is done later after the test
221 * of oldbrk with newbrk then it can escape the test and let the data
222 * segment grow beyond its set limit the in case where the limit is
223 * not page aligned -Ram Gupta
225 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
226 mm
->end_data
, mm
->start_data
))
229 newbrk
= PAGE_ALIGN(brk
);
230 oldbrk
= PAGE_ALIGN(mm
->brk
);
231 if (oldbrk
== newbrk
) {
237 * Always allow shrinking brk.
238 * __do_munmap() may downgrade mmap_sem to read.
240 if (brk
<= mm
->brk
) {
244 * mm->brk must to be protected by write mmap_sem so update it
245 * before downgrading mmap_sem. When __do_munmap() fails,
246 * mm->brk will be restored from origbrk.
249 ret
= __do_munmap(mm
, newbrk
, oldbrk
-newbrk
, &uf
, true);
253 } else if (ret
== 1) {
259 /* Check against existing mmap mappings. */
260 next
= find_vma(mm
, oldbrk
);
261 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
264 /* Ok, looks good - let it rip. */
265 if (do_brk_flags(oldbrk
, newbrk
-oldbrk
, 0, &uf
) < 0)
270 populate
= newbrk
> oldbrk
&& (mm
->def_flags
& VM_LOCKED
) != 0;
272 up_read(&mm
->mmap_sem
);
274 up_write(&mm
->mmap_sem
);
275 userfaultfd_unmap_complete(mm
, &uf
);
277 mm_populate(oldbrk
, newbrk
- oldbrk
);
282 up_write(&mm
->mmap_sem
);
286 static long vma_compute_subtree_gap(struct vm_area_struct
*vma
)
288 unsigned long max
, prev_end
, subtree_gap
;
291 * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
292 * allow two stack_guard_gaps between them here, and when choosing
293 * an unmapped area; whereas when expanding we only require one.
294 * That's a little inconsistent, but keeps the code here simpler.
296 max
= vm_start_gap(vma
);
298 prev_end
= vm_end_gap(vma
->vm_prev
);
304 if (vma
->vm_rb
.rb_left
) {
305 subtree_gap
= rb_entry(vma
->vm_rb
.rb_left
,
306 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
307 if (subtree_gap
> max
)
310 if (vma
->vm_rb
.rb_right
) {
311 subtree_gap
= rb_entry(vma
->vm_rb
.rb_right
,
312 struct vm_area_struct
, vm_rb
)->rb_subtree_gap
;
313 if (subtree_gap
> max
)
319 #ifdef CONFIG_DEBUG_VM_RB
320 static int browse_rb(struct mm_struct
*mm
)
322 struct rb_root
*root
= &mm
->mm_rb
;
323 int i
= 0, j
, bug
= 0;
324 struct rb_node
*nd
, *pn
= NULL
;
325 unsigned long prev
= 0, pend
= 0;
327 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
328 struct vm_area_struct
*vma
;
329 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
330 if (vma
->vm_start
< prev
) {
331 pr_emerg("vm_start %lx < prev %lx\n",
332 vma
->vm_start
, prev
);
335 if (vma
->vm_start
< pend
) {
336 pr_emerg("vm_start %lx < pend %lx\n",
337 vma
->vm_start
, pend
);
340 if (vma
->vm_start
> vma
->vm_end
) {
341 pr_emerg("vm_start %lx > vm_end %lx\n",
342 vma
->vm_start
, vma
->vm_end
);
345 spin_lock(&mm
->page_table_lock
);
346 if (vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
)) {
347 pr_emerg("free gap %lx, correct %lx\n",
349 vma_compute_subtree_gap(vma
));
352 spin_unlock(&mm
->page_table_lock
);
355 prev
= vma
->vm_start
;
359 for (nd
= pn
; nd
; nd
= rb_prev(nd
))
362 pr_emerg("backwards %d, forwards %d\n", j
, i
);
368 static void validate_mm_rb(struct rb_root
*root
, struct vm_area_struct
*ignore
)
372 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
373 struct vm_area_struct
*vma
;
374 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
375 VM_BUG_ON_VMA(vma
!= ignore
&&
376 vma
->rb_subtree_gap
!= vma_compute_subtree_gap(vma
),
381 static void validate_mm(struct mm_struct
*mm
)
385 unsigned long highest_address
= 0;
386 struct vm_area_struct
*vma
= mm
->mmap
;
389 struct anon_vma
*anon_vma
= vma
->anon_vma
;
390 struct anon_vma_chain
*avc
;
393 anon_vma_lock_read(anon_vma
);
394 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
395 anon_vma_interval_tree_verify(avc
);
396 anon_vma_unlock_read(anon_vma
);
399 highest_address
= vm_end_gap(vma
);
403 if (i
!= mm
->map_count
) {
404 pr_emerg("map_count %d vm_next %d\n", mm
->map_count
, i
);
407 if (highest_address
!= mm
->highest_vm_end
) {
408 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
409 mm
->highest_vm_end
, highest_address
);
413 if (i
!= mm
->map_count
) {
415 pr_emerg("map_count %d rb %d\n", mm
->map_count
, i
);
418 VM_BUG_ON_MM(bug
, mm
);
421 #define validate_mm_rb(root, ignore) do { } while (0)
422 #define validate_mm(mm) do { } while (0)
425 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks
, struct vm_area_struct
, vm_rb
,
426 unsigned long, rb_subtree_gap
, vma_compute_subtree_gap
)
429 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
430 * vma->vm_prev->vm_end values changed, without modifying the vma's position
433 static void vma_gap_update(struct vm_area_struct
*vma
)
436 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
437 * function that does exactly what we want.
439 vma_gap_callbacks_propagate(&vma
->vm_rb
, NULL
);
442 static inline void vma_rb_insert(struct vm_area_struct
*vma
,
443 struct rb_root
*root
)
445 /* All rb_subtree_gap values must be consistent prior to insertion */
446 validate_mm_rb(root
, NULL
);
448 rb_insert_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
451 static void __vma_rb_erase(struct vm_area_struct
*vma
, struct rb_root
*root
)
454 * Note rb_erase_augmented is a fairly large inline function,
455 * so make sure we instantiate it only once with our desired
456 * augmented rbtree callbacks.
458 rb_erase_augmented(&vma
->vm_rb
, root
, &vma_gap_callbacks
);
461 static __always_inline
void vma_rb_erase_ignore(struct vm_area_struct
*vma
,
462 struct rb_root
*root
,
463 struct vm_area_struct
*ignore
)
466 * All rb_subtree_gap values must be consistent prior to erase,
467 * with the possible exception of the "next" vma being erased if
468 * next->vm_start was reduced.
470 validate_mm_rb(root
, ignore
);
472 __vma_rb_erase(vma
, root
);
475 static __always_inline
void vma_rb_erase(struct vm_area_struct
*vma
,
476 struct rb_root
*root
)
479 * All rb_subtree_gap values must be consistent prior to erase,
480 * with the possible exception of the vma being erased.
482 validate_mm_rb(root
, vma
);
484 __vma_rb_erase(vma
, root
);
488 * vma has some anon_vma assigned, and is already inserted on that
489 * anon_vma's interval trees.
491 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
492 * vma must be removed from the anon_vma's interval trees using
493 * anon_vma_interval_tree_pre_update_vma().
495 * After the update, the vma will be reinserted using
496 * anon_vma_interval_tree_post_update_vma().
498 * The entire update must be protected by exclusive mmap_sem and by
499 * the root anon_vma's mutex.
502 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
504 struct anon_vma_chain
*avc
;
506 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
507 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
511 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
513 struct anon_vma_chain
*avc
;
515 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
516 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
519 static int find_vma_links(struct mm_struct
*mm
, unsigned long addr
,
520 unsigned long end
, struct vm_area_struct
**pprev
,
521 struct rb_node
***rb_link
, struct rb_node
**rb_parent
)
523 struct rb_node
**__rb_link
, *__rb_parent
, *rb_prev
;
525 __rb_link
= &mm
->mm_rb
.rb_node
;
526 rb_prev
= __rb_parent
= NULL
;
529 struct vm_area_struct
*vma_tmp
;
531 __rb_parent
= *__rb_link
;
532 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
534 if (vma_tmp
->vm_end
> addr
) {
535 /* Fail if an existing vma overlaps the area */
536 if (vma_tmp
->vm_start
< end
)
538 __rb_link
= &__rb_parent
->rb_left
;
540 rb_prev
= __rb_parent
;
541 __rb_link
= &__rb_parent
->rb_right
;
547 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
548 *rb_link
= __rb_link
;
549 *rb_parent
= __rb_parent
;
553 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
554 unsigned long addr
, unsigned long end
)
556 unsigned long nr_pages
= 0;
557 struct vm_area_struct
*vma
;
559 /* Find first overlaping mapping */
560 vma
= find_vma_intersection(mm
, addr
, end
);
564 nr_pages
= (min(end
, vma
->vm_end
) -
565 max(addr
, vma
->vm_start
)) >> PAGE_SHIFT
;
567 /* Iterate over the rest of the overlaps */
568 for (vma
= vma
->vm_next
; vma
; vma
= vma
->vm_next
) {
569 unsigned long overlap_len
;
571 if (vma
->vm_start
> end
)
574 overlap_len
= min(end
, vma
->vm_end
) - vma
->vm_start
;
575 nr_pages
+= overlap_len
>> PAGE_SHIFT
;
581 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
582 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
584 /* Update tracking information for the gap following the new vma. */
586 vma_gap_update(vma
->vm_next
);
588 mm
->highest_vm_end
= vm_end_gap(vma
);
591 * vma->vm_prev wasn't known when we followed the rbtree to find the
592 * correct insertion point for that vma. As a result, we could not
593 * update the vma vm_rb parents rb_subtree_gap values on the way down.
594 * So, we first insert the vma with a zero rb_subtree_gap value
595 * (to be consistent with what we did on the way down), and then
596 * immediately update the gap to the correct value. Finally we
597 * rebalance the rbtree after all augmented values have been set.
599 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
600 vma
->rb_subtree_gap
= 0;
602 vma_rb_insert(vma
, &mm
->mm_rb
);
605 static void __vma_link_file(struct vm_area_struct
*vma
)
611 struct address_space
*mapping
= file
->f_mapping
;
613 if (vma
->vm_flags
& VM_DENYWRITE
)
614 atomic_dec(&file_inode(file
)->i_writecount
);
615 if (vma
->vm_flags
& VM_SHARED
)
616 atomic_inc(&mapping
->i_mmap_writable
);
618 flush_dcache_mmap_lock(mapping
);
619 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
620 flush_dcache_mmap_unlock(mapping
);
625 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
626 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
627 struct rb_node
*rb_parent
)
629 __vma_link_list(mm
, vma
, prev
, rb_parent
);
630 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
633 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
634 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
635 struct rb_node
*rb_parent
)
637 struct address_space
*mapping
= NULL
;
640 mapping
= vma
->vm_file
->f_mapping
;
641 i_mmap_lock_write(mapping
);
644 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
645 __vma_link_file(vma
);
648 i_mmap_unlock_write(mapping
);
655 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
656 * mm's list and rbtree. It has already been inserted into the interval tree.
658 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
660 struct vm_area_struct
*prev
;
661 struct rb_node
**rb_link
, *rb_parent
;
663 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
664 &prev
, &rb_link
, &rb_parent
))
666 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
670 static __always_inline
void __vma_unlink_common(struct mm_struct
*mm
,
671 struct vm_area_struct
*vma
,
672 struct vm_area_struct
*prev
,
674 struct vm_area_struct
*ignore
)
676 struct vm_area_struct
*next
;
678 vma_rb_erase_ignore(vma
, &mm
->mm_rb
, ignore
);
681 prev
->vm_next
= next
;
685 prev
->vm_next
= next
;
690 next
->vm_prev
= prev
;
693 vmacache_invalidate(mm
);
696 static inline void __vma_unlink_prev(struct mm_struct
*mm
,
697 struct vm_area_struct
*vma
,
698 struct vm_area_struct
*prev
)
700 __vma_unlink_common(mm
, vma
, prev
, true, vma
);
704 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
705 * is already present in an i_mmap tree without adjusting the tree.
706 * The following helper function should be used when such adjustments
707 * are necessary. The "insert" vma (if any) is to be inserted
708 * before we drop the necessary locks.
710 int __vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
711 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
,
712 struct vm_area_struct
*expand
)
714 struct mm_struct
*mm
= vma
->vm_mm
;
715 struct vm_area_struct
*next
= vma
->vm_next
, *orig_vma
= vma
;
716 struct address_space
*mapping
= NULL
;
717 struct rb_root_cached
*root
= NULL
;
718 struct anon_vma
*anon_vma
= NULL
;
719 struct file
*file
= vma
->vm_file
;
720 bool start_changed
= false, end_changed
= false;
721 long adjust_next
= 0;
724 if (next
&& !insert
) {
725 struct vm_area_struct
*exporter
= NULL
, *importer
= NULL
;
727 if (end
>= next
->vm_end
) {
729 * vma expands, overlapping all the next, and
730 * perhaps the one after too (mprotect case 6).
731 * The only other cases that gets here are
732 * case 1, case 7 and case 8.
734 if (next
== expand
) {
736 * The only case where we don't expand "vma"
737 * and we expand "next" instead is case 8.
739 VM_WARN_ON(end
!= next
->vm_end
);
741 * remove_next == 3 means we're
742 * removing "vma" and that to do so we
743 * swapped "vma" and "next".
746 VM_WARN_ON(file
!= next
->vm_file
);
749 VM_WARN_ON(expand
!= vma
);
751 * case 1, 6, 7, remove_next == 2 is case 6,
752 * remove_next == 1 is case 1 or 7.
754 remove_next
= 1 + (end
> next
->vm_end
);
755 VM_WARN_ON(remove_next
== 2 &&
756 end
!= next
->vm_next
->vm_end
);
757 VM_WARN_ON(remove_next
== 1 &&
758 end
!= next
->vm_end
);
759 /* trim end to next, for case 6 first pass */
767 * If next doesn't have anon_vma, import from vma after
768 * next, if the vma overlaps with it.
770 if (remove_next
== 2 && !next
->anon_vma
)
771 exporter
= next
->vm_next
;
773 } else if (end
> next
->vm_start
) {
775 * vma expands, overlapping part of the next:
776 * mprotect case 5 shifting the boundary up.
778 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
781 VM_WARN_ON(expand
!= importer
);
782 } else if (end
< vma
->vm_end
) {
784 * vma shrinks, and !insert tells it's not
785 * split_vma inserting another: so it must be
786 * mprotect case 4 shifting the boundary down.
788 adjust_next
= -((vma
->vm_end
- end
) >> PAGE_SHIFT
);
791 VM_WARN_ON(expand
!= importer
);
795 * Easily overlooked: when mprotect shifts the boundary,
796 * make sure the expanding vma has anon_vma set if the
797 * shrinking vma had, to cover any anon pages imported.
799 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
802 importer
->anon_vma
= exporter
->anon_vma
;
803 error
= anon_vma_clone(importer
, exporter
);
809 vma_adjust_trans_huge(orig_vma
, start
, end
, adjust_next
);
812 mapping
= file
->f_mapping
;
813 root
= &mapping
->i_mmap
;
814 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
817 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
819 i_mmap_lock_write(mapping
);
822 * Put into interval tree now, so instantiated pages
823 * are visible to arm/parisc __flush_dcache_page
824 * throughout; but we cannot insert into address
825 * space until vma start or end is updated.
827 __vma_link_file(insert
);
831 anon_vma
= vma
->anon_vma
;
832 if (!anon_vma
&& adjust_next
)
833 anon_vma
= next
->anon_vma
;
835 VM_WARN_ON(adjust_next
&& next
->anon_vma
&&
836 anon_vma
!= next
->anon_vma
);
837 anon_vma_lock_write(anon_vma
);
838 anon_vma_interval_tree_pre_update_vma(vma
);
840 anon_vma_interval_tree_pre_update_vma(next
);
844 flush_dcache_mmap_lock(mapping
);
845 vma_interval_tree_remove(vma
, root
);
847 vma_interval_tree_remove(next
, root
);
850 if (start
!= vma
->vm_start
) {
851 vma
->vm_start
= start
;
852 start_changed
= true;
854 if (end
!= vma
->vm_end
) {
858 vma
->vm_pgoff
= pgoff
;
860 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
861 next
->vm_pgoff
+= adjust_next
;
866 vma_interval_tree_insert(next
, root
);
867 vma_interval_tree_insert(vma
, root
);
868 flush_dcache_mmap_unlock(mapping
);
873 * vma_merge has merged next into vma, and needs
874 * us to remove next before dropping the locks.
876 if (remove_next
!= 3)
877 __vma_unlink_prev(mm
, next
, vma
);
880 * vma is not before next if they've been
883 * pre-swap() next->vm_start was reduced so
884 * tell validate_mm_rb to ignore pre-swap()
885 * "next" (which is stored in post-swap()
888 __vma_unlink_common(mm
, next
, NULL
, false, vma
);
890 __remove_shared_vm_struct(next
, file
, mapping
);
893 * split_vma has split insert from vma, and needs
894 * us to insert it before dropping the locks
895 * (it may either follow vma or precede it).
897 __insert_vm_struct(mm
, insert
);
903 mm
->highest_vm_end
= vm_end_gap(vma
);
904 else if (!adjust_next
)
905 vma_gap_update(next
);
910 anon_vma_interval_tree_post_update_vma(vma
);
912 anon_vma_interval_tree_post_update_vma(next
);
913 anon_vma_unlock_write(anon_vma
);
916 i_mmap_unlock_write(mapping
);
927 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
931 anon_vma_merge(vma
, next
);
933 mpol_put(vma_policy(next
));
936 * In mprotect's case 6 (see comments on vma_merge),
937 * we must remove another next too. It would clutter
938 * up the code too much to do both in one go.
940 if (remove_next
!= 3) {
942 * If "next" was removed and vma->vm_end was
943 * expanded (up) over it, in turn
944 * "next->vm_prev->vm_end" changed and the
945 * "vma->vm_next" gap must be updated.
950 * For the scope of the comment "next" and
951 * "vma" considered pre-swap(): if "vma" was
952 * removed, next->vm_start was expanded (down)
953 * over it and the "next" gap must be updated.
954 * Because of the swap() the post-swap() "vma"
955 * actually points to pre-swap() "next"
956 * (post-swap() "next" as opposed is now a
961 if (remove_next
== 2) {
967 vma_gap_update(next
);
970 * If remove_next == 2 we obviously can't
973 * If remove_next == 3 we can't reach this
974 * path because pre-swap() next is always not
975 * NULL. pre-swap() "next" is not being
976 * removed and its next->vm_end is not altered
977 * (and furthermore "end" already matches
978 * next->vm_end in remove_next == 3).
980 * We reach this only in the remove_next == 1
981 * case if the "next" vma that was removed was
982 * the highest vma of the mm. However in such
983 * case next->vm_end == "end" and the extended
984 * "vma" has vma->vm_end == next->vm_end so
985 * mm->highest_vm_end doesn't need any update
986 * in remove_next == 1 case.
988 VM_WARN_ON(mm
->highest_vm_end
!= vm_end_gap(vma
));
1000 * If the vma has a ->close operation then the driver probably needs to release
1001 * per-vma resources, so we don't attempt to merge those.
1003 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
1004 struct file
*file
, unsigned long vm_flags
,
1005 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1008 * VM_SOFTDIRTY should not prevent from VMA merging, if we
1009 * match the flags but dirty bit -- the caller should mark
1010 * merged VMA as dirty. If dirty bit won't be excluded from
1011 * comparison, we increase pressure on the memory system forcing
1012 * the kernel to generate new VMAs when old one could be
1015 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
1017 if (vma
->vm_file
!= file
)
1019 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
1021 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
1026 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
1027 struct anon_vma
*anon_vma2
,
1028 struct vm_area_struct
*vma
)
1031 * The list_is_singular() test is to avoid merging VMA cloned from
1032 * parents. This can improve scalability caused by anon_vma lock.
1034 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
1035 list_is_singular(&vma
->anon_vma_chain
)))
1037 return anon_vma1
== anon_vma2
;
1041 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1042 * in front of (at a lower 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.
1047 * We don't check here for the merged mmap wrapping around the end of pagecache
1048 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1049 * wrap, nor mmaps which cover the final page at index -1UL.
1052 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1053 struct anon_vma
*anon_vma
, struct file
*file
,
1055 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1057 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1058 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1059 if (vma
->vm_pgoff
== vm_pgoff
)
1066 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1067 * beyond (at a higher virtual address and file offset than) the vma.
1069 * We cannot merge two vmas if they have differently assigned (non-NULL)
1070 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1073 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
1074 struct anon_vma
*anon_vma
, struct file
*file
,
1076 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1078 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
) &&
1079 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
1081 vm_pglen
= vma_pages(vma
);
1082 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
1089 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1090 * whether that can be merged with its predecessor or its successor.
1091 * Or both (it neatly fills a hole).
1093 * In most cases - when called for mmap, brk or mremap - [addr,end) is
1094 * certain not to be mapped by the time vma_merge is called; but when
1095 * called for mprotect, it is certain to be already mapped (either at
1096 * an offset within prev, or at the start of next), and the flags of
1097 * this area are about to be changed to vm_flags - and the no-change
1098 * case has already been eliminated.
1100 * The following mprotect cases have to be considered, where AAAA is
1101 * the area passed down from mprotect_fixup, never extending beyond one
1102 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1104 * AAAA AAAA AAAA AAAA
1105 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1106 * cannot merge might become might become might become
1107 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1108 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1109 * mremap move: PPPPXXXXXXXX 8
1111 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1112 * might become case 1 below case 2 below case 3 below
1114 * It is important for case 8 that the vma NNNN overlapping the
1115 * region AAAA is never going to extended over XXXX. Instead XXXX must
1116 * be extended in region AAAA and NNNN must be removed. This way in
1117 * all cases where vma_merge succeeds, the moment vma_adjust drops the
1118 * rmap_locks, the properties of the merged vma will be already
1119 * correct for the whole merged range. Some of those properties like
1120 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1121 * be correct for the whole merged range immediately after the
1122 * rmap_locks are released. Otherwise if XXXX would be removed and
1123 * NNNN would be extended over the XXXX range, remove_migration_ptes
1124 * or other rmap walkers (if working on addresses beyond the "end"
1125 * parameter) may establish ptes with the wrong permissions of NNNN
1126 * instead of the right permissions of XXXX.
1128 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
1129 struct vm_area_struct
*prev
, unsigned long addr
,
1130 unsigned long end
, unsigned long vm_flags
,
1131 struct anon_vma
*anon_vma
, struct file
*file
,
1132 pgoff_t pgoff
, struct mempolicy
*policy
,
1133 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
)
1135 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
1136 struct vm_area_struct
*area
, *next
;
1140 * We later require that vma->vm_flags == vm_flags,
1141 * so this tests vma->vm_flags & VM_SPECIAL, too.
1143 if (vm_flags
& VM_SPECIAL
)
1147 next
= prev
->vm_next
;
1151 if (area
&& area
->vm_end
== end
) /* cases 6, 7, 8 */
1152 next
= next
->vm_next
;
1154 /* verify some invariant that must be enforced by the caller */
1155 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
1156 VM_WARN_ON(area
&& end
> area
->vm_end
);
1157 VM_WARN_ON(addr
>= end
);
1160 * Can it merge with the predecessor?
1162 if (prev
&& prev
->vm_end
== addr
&&
1163 mpol_equal(vma_policy(prev
), policy
) &&
1164 can_vma_merge_after(prev
, vm_flags
,
1165 anon_vma
, file
, pgoff
,
1166 vm_userfaultfd_ctx
)) {
1168 * OK, it can. Can we now merge in the successor as well?
1170 if (next
&& end
== next
->vm_start
&&
1171 mpol_equal(policy
, vma_policy(next
)) &&
1172 can_vma_merge_before(next
, vm_flags
,
1175 vm_userfaultfd_ctx
) &&
1176 is_mergeable_anon_vma(prev
->anon_vma
,
1177 next
->anon_vma
, NULL
)) {
1179 err
= __vma_adjust(prev
, prev
->vm_start
,
1180 next
->vm_end
, prev
->vm_pgoff
, NULL
,
1182 } else /* cases 2, 5, 7 */
1183 err
= __vma_adjust(prev
, prev
->vm_start
,
1184 end
, prev
->vm_pgoff
, NULL
, prev
);
1187 khugepaged_enter_vma_merge(prev
, vm_flags
);
1192 * Can this new request be merged in front of next?
1194 if (next
&& end
== next
->vm_start
&&
1195 mpol_equal(policy
, vma_policy(next
)) &&
1196 can_vma_merge_before(next
, vm_flags
,
1197 anon_vma
, file
, pgoff
+pglen
,
1198 vm_userfaultfd_ctx
)) {
1199 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
1200 err
= __vma_adjust(prev
, prev
->vm_start
,
1201 addr
, prev
->vm_pgoff
, NULL
, next
);
1202 else { /* cases 3, 8 */
1203 err
= __vma_adjust(area
, addr
, next
->vm_end
,
1204 next
->vm_pgoff
- pglen
, NULL
, next
);
1206 * In case 3 area is already equal to next and
1207 * this is a noop, but in case 8 "area" has
1208 * been removed and next was expanded over it.
1214 khugepaged_enter_vma_merge(area
, vm_flags
);
1222 * Rough compatbility check to quickly see if it's even worth looking
1223 * at sharing an anon_vma.
1225 * They need to have the same vm_file, and the flags can only differ
1226 * in things that mprotect may change.
1228 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1229 * we can merge the two vma's. For example, we refuse to merge a vma if
1230 * there is a vm_ops->close() function, because that indicates that the
1231 * driver is doing some kind of reference counting. But that doesn't
1232 * really matter for the anon_vma sharing case.
1234 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1236 return a
->vm_end
== b
->vm_start
&&
1237 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1238 a
->vm_file
== b
->vm_file
&&
1239 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
|VM_SOFTDIRTY
)) &&
1240 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1244 * Do some basic sanity checking to see if we can re-use the anon_vma
1245 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1246 * the same as 'old', the other will be the new one that is trying
1247 * to share the anon_vma.
1249 * NOTE! This runs with mm_sem held for reading, so it is possible that
1250 * the anon_vma of 'old' is concurrently in the process of being set up
1251 * by another page fault trying to merge _that_. But that's ok: if it
1252 * is being set up, that automatically means that it will be a singleton
1253 * acceptable for merging, so we can do all of this optimistically. But
1254 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1256 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1257 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1258 * is to return an anon_vma that is "complex" due to having gone through
1261 * We also make sure that the two vma's are compatible (adjacent,
1262 * and with the same memory policies). That's all stable, even with just
1263 * a read lock on the mm_sem.
1265 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1267 if (anon_vma_compatible(a
, b
)) {
1268 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1270 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1277 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1278 * neighbouring vmas for a suitable anon_vma, before it goes off
1279 * to allocate a new anon_vma. It checks because a repetitive
1280 * sequence of mprotects and faults may otherwise lead to distinct
1281 * anon_vmas being allocated, preventing vma merge in subsequent
1284 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1286 struct anon_vma
*anon_vma
;
1287 struct vm_area_struct
*near
;
1289 near
= vma
->vm_next
;
1293 anon_vma
= reusable_anon_vma(near
, vma
, near
);
1297 near
= vma
->vm_prev
;
1301 anon_vma
= reusable_anon_vma(near
, near
, vma
);
1306 * There's no absolute need to look only at touching neighbours:
1307 * we could search further afield for "compatible" anon_vmas.
1308 * But it would probably just be a waste of time searching,
1309 * or lead to too many vmas hanging off the same anon_vma.
1310 * We're trying to allow mprotect remerging later on,
1311 * not trying to minimize memory used for anon_vmas.
1317 * If a hint addr is less than mmap_min_addr change hint to be as
1318 * low as possible but still greater than mmap_min_addr
1320 static inline unsigned long round_hint_to_min(unsigned long hint
)
1323 if (((void *)hint
!= NULL
) &&
1324 (hint
< mmap_min_addr
))
1325 return PAGE_ALIGN(mmap_min_addr
);
1329 static inline int mlock_future_check(struct mm_struct
*mm
,
1330 unsigned long flags
,
1333 unsigned long locked
, lock_limit
;
1335 /* mlock MCL_FUTURE? */
1336 if (flags
& VM_LOCKED
) {
1337 locked
= len
>> PAGE_SHIFT
;
1338 locked
+= mm
->locked_vm
;
1339 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1340 lock_limit
>>= PAGE_SHIFT
;
1341 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1347 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1349 if (S_ISREG(inode
->i_mode
))
1350 return MAX_LFS_FILESIZE
;
1352 if (S_ISBLK(inode
->i_mode
))
1353 return MAX_LFS_FILESIZE
;
1355 /* Special "we do even unsigned file positions" case */
1356 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1359 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1363 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1364 unsigned long pgoff
, unsigned long len
)
1366 u64 maxsize
= file_mmap_size_max(file
, inode
);
1368 if (maxsize
&& len
> maxsize
)
1371 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1377 * The caller must hold down_write(¤t->mm->mmap_sem).
1379 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1380 unsigned long len
, unsigned long prot
,
1381 unsigned long flags
, vm_flags_t vm_flags
,
1382 unsigned long pgoff
, unsigned long *populate
,
1383 struct list_head
*uf
)
1385 struct mm_struct
*mm
= current
->mm
;
1394 * Does the application expect PROT_READ to imply PROT_EXEC?
1396 * (the exception is when the underlying filesystem is noexec
1397 * mounted, in which case we dont add PROT_EXEC.)
1399 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1400 if (!(file
&& path_noexec(&file
->f_path
)))
1403 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1404 if (flags
& MAP_FIXED_NOREPLACE
)
1407 if (!(flags
& MAP_FIXED
))
1408 addr
= round_hint_to_min(addr
);
1410 /* Careful about overflows.. */
1411 len
= PAGE_ALIGN(len
);
1415 /* offset overflow? */
1416 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1419 /* Too many mappings? */
1420 if (mm
->map_count
> sysctl_max_map_count
)
1423 /* Obtain the address to map to. we verify (or select) it and ensure
1424 * that it represents a valid section of the address space.
1426 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1427 if (offset_in_page(addr
))
1430 if (flags
& MAP_FIXED_NOREPLACE
) {
1431 struct vm_area_struct
*vma
= find_vma(mm
, addr
);
1433 if (vma
&& vma
->vm_start
< addr
+ len
)
1437 if (prot
== PROT_EXEC
) {
1438 pkey
= execute_only_pkey(mm
);
1443 /* Do simple checking here so the lower-level routines won't have
1444 * to. we assume access permissions have been handled by the open
1445 * of the memory object, so we don't do any here.
1447 vm_flags
|= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1448 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1450 if (flags
& MAP_LOCKED
)
1451 if (!can_do_mlock())
1454 if (mlock_future_check(mm
, vm_flags
, len
))
1458 struct inode
*inode
= file_inode(file
);
1459 unsigned long flags_mask
;
1461 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1464 flags_mask
= LEGACY_MAP_MASK
| file
->f_op
->mmap_supported_flags
;
1466 switch (flags
& MAP_TYPE
) {
1469 * Force use of MAP_SHARED_VALIDATE with non-legacy
1470 * flags. E.g. MAP_SYNC is dangerous to use with
1471 * MAP_SHARED as you don't know which consistency model
1472 * you will get. We silently ignore unsupported flags
1473 * with MAP_SHARED to preserve backward compatibility.
1475 flags
&= LEGACY_MAP_MASK
;
1477 case MAP_SHARED_VALIDATE
:
1478 if (flags
& ~flags_mask
)
1480 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1484 * Make sure we don't allow writing to an append-only
1487 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1491 * Make sure there are no mandatory locks on the file.
1493 if (locks_verify_locked(file
))
1496 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1497 if (!(file
->f_mode
& FMODE_WRITE
))
1498 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1502 if (!(file
->f_mode
& FMODE_READ
))
1504 if (path_noexec(&file
->f_path
)) {
1505 if (vm_flags
& VM_EXEC
)
1507 vm_flags
&= ~VM_MAYEXEC
;
1510 if (!file
->f_op
->mmap
)
1512 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1520 switch (flags
& MAP_TYPE
) {
1522 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1528 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1532 * Set pgoff according to addr for anon_vma.
1534 pgoff
= addr
>> PAGE_SHIFT
;
1542 * Set 'VM_NORESERVE' if we should not account for the
1543 * memory use of this mapping.
1545 if (flags
& MAP_NORESERVE
) {
1546 /* We honor MAP_NORESERVE if allowed to overcommit */
1547 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1548 vm_flags
|= VM_NORESERVE
;
1550 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1551 if (file
&& is_file_hugepages(file
))
1552 vm_flags
|= VM_NORESERVE
;
1555 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1556 if (!IS_ERR_VALUE(addr
) &&
1557 ((vm_flags
& VM_LOCKED
) ||
1558 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1563 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1564 unsigned long prot
, unsigned long flags
,
1565 unsigned long fd
, unsigned long pgoff
)
1567 struct file
*file
= NULL
;
1568 unsigned long retval
;
1570 if (!(flags
& MAP_ANONYMOUS
)) {
1571 audit_mmap_fd(fd
, flags
);
1575 if (is_file_hugepages(file
))
1576 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1578 if (unlikely(flags
& MAP_HUGETLB
&& !is_file_hugepages(file
)))
1580 } else if (flags
& MAP_HUGETLB
) {
1581 struct user_struct
*user
= NULL
;
1584 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1588 len
= ALIGN(len
, huge_page_size(hs
));
1590 * VM_NORESERVE is used because the reservations will be
1591 * taken when vm_ops->mmap() is called
1592 * A dummy user value is used because we are not locking
1593 * memory so no accounting is necessary
1595 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1597 &user
, HUGETLB_ANONHUGE_INODE
,
1598 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1600 return PTR_ERR(file
);
1603 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1605 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1612 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1613 unsigned long, prot
, unsigned long, flags
,
1614 unsigned long, fd
, unsigned long, pgoff
)
1616 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1619 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1620 struct mmap_arg_struct
{
1624 unsigned long flags
;
1626 unsigned long offset
;
1629 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1631 struct mmap_arg_struct a
;
1633 if (copy_from_user(&a
, arg
, sizeof(a
)))
1635 if (offset_in_page(a
.offset
))
1638 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1639 a
.offset
>> PAGE_SHIFT
);
1641 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1644 * Some shared mappings will want the pages marked read-only
1645 * to track write events. If so, we'll downgrade vm_page_prot
1646 * to the private version (using protection_map[] without the
1649 int vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1651 vm_flags_t vm_flags
= vma
->vm_flags
;
1652 const struct vm_operations_struct
*vm_ops
= vma
->vm_ops
;
1654 /* If it was private or non-writable, the write bit is already clear */
1655 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1658 /* The backer wishes to know when pages are first written to? */
1659 if (vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
))
1662 /* The open routine did something to the protections that pgprot_modify
1663 * won't preserve? */
1664 if (pgprot_val(vm_page_prot
) !=
1665 pgprot_val(vm_pgprot_modify(vm_page_prot
, vm_flags
)))
1668 /* Do we need to track softdirty? */
1669 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY
) && !(vm_flags
& VM_SOFTDIRTY
))
1672 /* Specialty mapping? */
1673 if (vm_flags
& VM_PFNMAP
)
1676 /* Can the mapping track the dirty pages? */
1677 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1678 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1682 * We account for memory if it's a private writeable mapping,
1683 * not hugepages and VM_NORESERVE wasn't set.
1685 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1688 * hugetlb has its own accounting separate from the core VM
1689 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1691 if (file
&& is_file_hugepages(file
))
1694 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1697 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1698 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
1699 struct list_head
*uf
)
1701 struct mm_struct
*mm
= current
->mm
;
1702 struct vm_area_struct
*vma
, *prev
;
1704 struct rb_node
**rb_link
, *rb_parent
;
1705 unsigned long charged
= 0;
1707 /* Check against address space limit. */
1708 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
1709 unsigned long nr_pages
;
1712 * MAP_FIXED may remove pages of mappings that intersects with
1713 * requested mapping. Account for the pages it would unmap.
1715 nr_pages
= count_vma_pages_range(mm
, addr
, addr
+ len
);
1717 if (!may_expand_vm(mm
, vm_flags
,
1718 (len
>> PAGE_SHIFT
) - nr_pages
))
1722 /* Clear old maps */
1723 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
1725 if (do_munmap(mm
, addr
, len
, uf
))
1730 * Private writable mapping: check memory availability
1732 if (accountable_mapping(file
, vm_flags
)) {
1733 charged
= len
>> PAGE_SHIFT
;
1734 if (security_vm_enough_memory_mm(mm
, charged
))
1736 vm_flags
|= VM_ACCOUNT
;
1740 * Can we just expand an old mapping?
1742 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1743 NULL
, file
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
1748 * Determine the object being mapped and call the appropriate
1749 * specific mapper. the address has already been validated, but
1750 * not unmapped, but the maps are removed from the list.
1752 vma
= vm_area_alloc(mm
);
1758 vma
->vm_start
= addr
;
1759 vma
->vm_end
= addr
+ len
;
1760 vma
->vm_flags
= vm_flags
;
1761 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1762 vma
->vm_pgoff
= pgoff
;
1765 if (vm_flags
& VM_DENYWRITE
) {
1766 error
= deny_write_access(file
);
1770 if (vm_flags
& VM_SHARED
) {
1771 error
= mapping_map_writable(file
->f_mapping
);
1773 goto allow_write_and_free_vma
;
1776 /* ->mmap() can change vma->vm_file, but must guarantee that
1777 * vma_link() below can deny write-access if VM_DENYWRITE is set
1778 * and map writably if VM_SHARED is set. This usually means the
1779 * new file must not have been exposed to user-space, yet.
1781 vma
->vm_file
= get_file(file
);
1782 error
= call_mmap(file
, vma
);
1784 goto unmap_and_free_vma
;
1786 /* Can addr have changed??
1788 * Answer: Yes, several device drivers can do it in their
1789 * f_op->mmap method. -DaveM
1790 * Bug: If addr is changed, prev, rb_link, rb_parent should
1791 * be updated for vma_link()
1793 WARN_ON_ONCE(addr
!= vma
->vm_start
);
1795 addr
= vma
->vm_start
;
1796 vm_flags
= vma
->vm_flags
;
1797 } else if (vm_flags
& VM_SHARED
) {
1798 error
= shmem_zero_setup(vma
);
1802 vma_set_anonymous(vma
);
1805 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1806 /* Once vma denies write, undo our temporary denial count */
1808 if (vm_flags
& VM_SHARED
)
1809 mapping_unmap_writable(file
->f_mapping
);
1810 if (vm_flags
& VM_DENYWRITE
)
1811 allow_write_access(file
);
1813 file
= vma
->vm_file
;
1815 perf_event_mmap(vma
);
1817 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
1818 if (vm_flags
& VM_LOCKED
) {
1819 if ((vm_flags
& VM_SPECIAL
) || vma_is_dax(vma
) ||
1820 is_vm_hugetlb_page(vma
) ||
1821 vma
== get_gate_vma(current
->mm
))
1822 vma
->vm_flags
&= VM_LOCKED_CLEAR_MASK
;
1824 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1831 * New (or expanded) vma always get soft dirty status.
1832 * Otherwise user-space soft-dirty page tracker won't
1833 * be able to distinguish situation when vma area unmapped,
1834 * then new mapped in-place (which must be aimed as
1835 * a completely new data area).
1837 vma
->vm_flags
|= VM_SOFTDIRTY
;
1839 vma_set_page_prot(vma
);
1845 vma
->vm_file
= NULL
;
1847 /* Undo any partial mapping done by a device driver. */
1848 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1850 if (vm_flags
& VM_SHARED
)
1851 mapping_unmap_writable(file
->f_mapping
);
1852 allow_write_and_free_vma
:
1853 if (vm_flags
& VM_DENYWRITE
)
1854 allow_write_access(file
);
1859 vm_unacct_memory(charged
);
1863 unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1866 * We implement the search by looking for an rbtree node that
1867 * immediately follows a suitable gap. That is,
1868 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1869 * - gap_end = vma->vm_start >= info->low_limit + length;
1870 * - gap_end - gap_start >= length
1873 struct mm_struct
*mm
= current
->mm
;
1874 struct vm_area_struct
*vma
;
1875 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1877 /* Adjust search length to account for worst case alignment overhead */
1878 length
= info
->length
+ info
->align_mask
;
1879 if (length
< info
->length
)
1882 /* Adjust search limits by the desired length */
1883 if (info
->high_limit
< length
)
1885 high_limit
= info
->high_limit
- length
;
1887 if (info
->low_limit
> high_limit
)
1889 low_limit
= info
->low_limit
+ length
;
1891 /* Check if rbtree root looks promising */
1892 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1894 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1895 if (vma
->rb_subtree_gap
< length
)
1899 /* Visit left subtree if it looks promising */
1900 gap_end
= vm_start_gap(vma
);
1901 if (gap_end
>= low_limit
&& vma
->vm_rb
.rb_left
) {
1902 struct vm_area_struct
*left
=
1903 rb_entry(vma
->vm_rb
.rb_left
,
1904 struct vm_area_struct
, vm_rb
);
1905 if (left
->rb_subtree_gap
>= length
) {
1911 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
1913 /* Check if current node has a suitable gap */
1914 if (gap_start
> high_limit
)
1916 if (gap_end
>= low_limit
&&
1917 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
1920 /* Visit right subtree if it looks promising */
1921 if (vma
->vm_rb
.rb_right
) {
1922 struct vm_area_struct
*right
=
1923 rb_entry(vma
->vm_rb
.rb_right
,
1924 struct vm_area_struct
, vm_rb
);
1925 if (right
->rb_subtree_gap
>= length
) {
1931 /* Go back up the rbtree to find next candidate node */
1933 struct rb_node
*prev
= &vma
->vm_rb
;
1934 if (!rb_parent(prev
))
1936 vma
= rb_entry(rb_parent(prev
),
1937 struct vm_area_struct
, vm_rb
);
1938 if (prev
== vma
->vm_rb
.rb_left
) {
1939 gap_start
= vm_end_gap(vma
->vm_prev
);
1940 gap_end
= vm_start_gap(vma
);
1947 /* Check highest gap, which does not precede any rbtree node */
1948 gap_start
= mm
->highest_vm_end
;
1949 gap_end
= ULONG_MAX
; /* Only for VM_BUG_ON below */
1950 if (gap_start
> high_limit
)
1954 /* We found a suitable gap. Clip it with the original low_limit. */
1955 if (gap_start
< info
->low_limit
)
1956 gap_start
= info
->low_limit
;
1958 /* Adjust gap address to the desired alignment */
1959 gap_start
+= (info
->align_offset
- gap_start
) & info
->align_mask
;
1961 VM_BUG_ON(gap_start
+ info
->length
> info
->high_limit
);
1962 VM_BUG_ON(gap_start
+ info
->length
> gap_end
);
1966 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1968 struct mm_struct
*mm
= current
->mm
;
1969 struct vm_area_struct
*vma
;
1970 unsigned long length
, low_limit
, high_limit
, gap_start
, gap_end
;
1972 /* Adjust search length to account for worst case alignment overhead */
1973 length
= info
->length
+ info
->align_mask
;
1974 if (length
< info
->length
)
1978 * Adjust search limits by the desired length.
1979 * See implementation comment at top of unmapped_area().
1981 gap_end
= info
->high_limit
;
1982 if (gap_end
< length
)
1984 high_limit
= gap_end
- length
;
1986 if (info
->low_limit
> high_limit
)
1988 low_limit
= info
->low_limit
+ length
;
1990 /* Check highest gap, which does not precede any rbtree node */
1991 gap_start
= mm
->highest_vm_end
;
1992 if (gap_start
<= high_limit
)
1995 /* Check if rbtree root looks promising */
1996 if (RB_EMPTY_ROOT(&mm
->mm_rb
))
1998 vma
= rb_entry(mm
->mm_rb
.rb_node
, struct vm_area_struct
, vm_rb
);
1999 if (vma
->rb_subtree_gap
< length
)
2003 /* Visit right subtree if it looks promising */
2004 gap_start
= vma
->vm_prev
? vm_end_gap(vma
->vm_prev
) : 0;
2005 if (gap_start
<= high_limit
&& vma
->vm_rb
.rb_right
) {
2006 struct vm_area_struct
*right
=
2007 rb_entry(vma
->vm_rb
.rb_right
,
2008 struct vm_area_struct
, vm_rb
);
2009 if (right
->rb_subtree_gap
>= length
) {
2016 /* Check if current node has a suitable gap */
2017 gap_end
= vm_start_gap(vma
);
2018 if (gap_end
< low_limit
)
2020 if (gap_start
<= high_limit
&&
2021 gap_end
> gap_start
&& gap_end
- gap_start
>= length
)
2024 /* Visit left subtree if it looks promising */
2025 if (vma
->vm_rb
.rb_left
) {
2026 struct vm_area_struct
*left
=
2027 rb_entry(vma
->vm_rb
.rb_left
,
2028 struct vm_area_struct
, vm_rb
);
2029 if (left
->rb_subtree_gap
>= length
) {
2035 /* Go back up the rbtree to find next candidate node */
2037 struct rb_node
*prev
= &vma
->vm_rb
;
2038 if (!rb_parent(prev
))
2040 vma
= rb_entry(rb_parent(prev
),
2041 struct vm_area_struct
, vm_rb
);
2042 if (prev
== vma
->vm_rb
.rb_right
) {
2043 gap_start
= vma
->vm_prev
?
2044 vm_end_gap(vma
->vm_prev
) : 0;
2051 /* We found a suitable gap. Clip it with the original high_limit. */
2052 if (gap_end
> info
->high_limit
)
2053 gap_end
= info
->high_limit
;
2056 /* Compute highest gap address at the desired alignment */
2057 gap_end
-= info
->length
;
2058 gap_end
-= (gap_end
- info
->align_offset
) & info
->align_mask
;
2060 VM_BUG_ON(gap_end
< info
->low_limit
);
2061 VM_BUG_ON(gap_end
< gap_start
);
2066 #ifndef arch_get_mmap_end
2067 #define arch_get_mmap_end(addr) (TASK_SIZE)
2070 #ifndef arch_get_mmap_base
2071 #define arch_get_mmap_base(addr, base) (base)
2074 /* Get an address range which is currently unmapped.
2075 * For shmat() with addr=0.
2077 * Ugly calling convention alert:
2078 * Return value with the low bits set means error value,
2080 * if (ret & ~PAGE_MASK)
2083 * This function "knows" that -ENOMEM has the bits set.
2085 #ifndef HAVE_ARCH_UNMAPPED_AREA
2087 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
2088 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
2090 struct mm_struct
*mm
= current
->mm
;
2091 struct vm_area_struct
*vma
, *prev
;
2092 struct vm_unmapped_area_info info
;
2093 const unsigned long mmap_end
= arch_get_mmap_end(addr
);
2095 if (len
> mmap_end
- mmap_min_addr
)
2098 if (flags
& MAP_FIXED
)
2102 addr
= PAGE_ALIGN(addr
);
2103 vma
= find_vma_prev(mm
, addr
, &prev
);
2104 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
2105 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2106 (!prev
|| addr
>= vm_end_gap(prev
)))
2112 info
.low_limit
= mm
->mmap_base
;
2113 info
.high_limit
= mmap_end
;
2114 info
.align_mask
= 0;
2115 return vm_unmapped_area(&info
);
2120 * This mmap-allocator allocates new areas top-down from below the
2121 * stack's low limit (the base):
2123 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2125 arch_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
2126 unsigned long len
, unsigned long pgoff
,
2127 unsigned long flags
)
2129 struct vm_area_struct
*vma
, *prev
;
2130 struct mm_struct
*mm
= current
->mm
;
2131 struct vm_unmapped_area_info info
;
2132 const unsigned long mmap_end
= arch_get_mmap_end(addr
);
2134 /* requested length too big for entire address space */
2135 if (len
> mmap_end
- mmap_min_addr
)
2138 if (flags
& MAP_FIXED
)
2141 /* requesting a specific address */
2143 addr
= PAGE_ALIGN(addr
);
2144 vma
= find_vma_prev(mm
, addr
, &prev
);
2145 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
2146 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
2147 (!prev
|| addr
>= vm_end_gap(prev
)))
2151 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
2153 info
.low_limit
= max(PAGE_SIZE
, mmap_min_addr
);
2154 info
.high_limit
= arch_get_mmap_base(addr
, mm
->mmap_base
);
2155 info
.align_mask
= 0;
2156 addr
= vm_unmapped_area(&info
);
2159 * A failed mmap() very likely causes application failure,
2160 * so fall back to the bottom-up function here. This scenario
2161 * can happen with large stack limits and large mmap()
2164 if (offset_in_page(addr
)) {
2165 VM_BUG_ON(addr
!= -ENOMEM
);
2167 info
.low_limit
= TASK_UNMAPPED_BASE
;
2168 info
.high_limit
= mmap_end
;
2169 addr
= vm_unmapped_area(&info
);
2177 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
2178 unsigned long pgoff
, unsigned long flags
)
2180 unsigned long (*get_area
)(struct file
*, unsigned long,
2181 unsigned long, unsigned long, unsigned long);
2183 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
2187 /* Careful about overflows.. */
2188 if (len
> TASK_SIZE
)
2191 get_area
= current
->mm
->get_unmapped_area
;
2193 if (file
->f_op
->get_unmapped_area
)
2194 get_area
= file
->f_op
->get_unmapped_area
;
2195 } else if (flags
& MAP_SHARED
) {
2197 * mmap_region() will call shmem_zero_setup() to create a file,
2198 * so use shmem's get_unmapped_area in case it can be huge.
2199 * do_mmap_pgoff() will clear pgoff, so match alignment.
2202 get_area
= shmem_get_unmapped_area
;
2205 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
2206 if (IS_ERR_VALUE(addr
))
2209 if (addr
> TASK_SIZE
- len
)
2211 if (offset_in_page(addr
))
2214 error
= security_mmap_addr(addr
);
2215 return error
? error
: addr
;
2218 EXPORT_SYMBOL(get_unmapped_area
);
2220 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2221 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2223 struct rb_node
*rb_node
;
2224 struct vm_area_struct
*vma
;
2226 /* Check the cache first. */
2227 vma
= vmacache_find(mm
, addr
);
2231 rb_node
= mm
->mm_rb
.rb_node
;
2234 struct vm_area_struct
*tmp
;
2236 tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2238 if (tmp
->vm_end
> addr
) {
2240 if (tmp
->vm_start
<= addr
)
2242 rb_node
= rb_node
->rb_left
;
2244 rb_node
= rb_node
->rb_right
;
2248 vmacache_update(addr
, vma
);
2252 EXPORT_SYMBOL(find_vma
);
2255 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2257 struct vm_area_struct
*
2258 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2259 struct vm_area_struct
**pprev
)
2261 struct vm_area_struct
*vma
;
2263 vma
= find_vma(mm
, addr
);
2265 *pprev
= vma
->vm_prev
;
2267 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
2270 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
2271 rb_node
= rb_node
->rb_right
;
2278 * Verify that the stack growth is acceptable and
2279 * update accounting. This is shared with both the
2280 * grow-up and grow-down cases.
2282 static int acct_stack_growth(struct vm_area_struct
*vma
,
2283 unsigned long size
, unsigned long grow
)
2285 struct mm_struct
*mm
= vma
->vm_mm
;
2286 unsigned long new_start
;
2288 /* address space limit tests */
2289 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
2292 /* Stack limit test */
2293 if (size
> rlimit(RLIMIT_STACK
))
2296 /* mlock limit tests */
2297 if (vma
->vm_flags
& VM_LOCKED
) {
2298 unsigned long locked
;
2299 unsigned long limit
;
2300 locked
= mm
->locked_vm
+ grow
;
2301 limit
= rlimit(RLIMIT_MEMLOCK
);
2302 limit
>>= PAGE_SHIFT
;
2303 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
2307 /* Check to ensure the stack will not grow into a hugetlb-only region */
2308 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2310 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2314 * Overcommit.. This must be the final test, as it will
2315 * update security statistics.
2317 if (security_vm_enough_memory_mm(mm
, grow
))
2323 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2325 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2326 * vma is the last one with address > vma->vm_end. Have to extend vma.
2328 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2330 struct mm_struct
*mm
= vma
->vm_mm
;
2331 struct vm_area_struct
*next
;
2332 unsigned long gap_addr
;
2335 if (!(vma
->vm_flags
& VM_GROWSUP
))
2338 /* Guard against exceeding limits of the address space. */
2339 address
&= PAGE_MASK
;
2340 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2342 address
+= PAGE_SIZE
;
2344 /* Enforce stack_guard_gap */
2345 gap_addr
= address
+ stack_guard_gap
;
2347 /* Guard against overflow */
2348 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2349 gap_addr
= TASK_SIZE
;
2351 next
= vma
->vm_next
;
2352 if (next
&& next
->vm_start
< gap_addr
&&
2353 (next
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2354 if (!(next
->vm_flags
& VM_GROWSUP
))
2356 /* Check that both stack segments have the same anon_vma? */
2359 /* We must make sure the anon_vma is allocated. */
2360 if (unlikely(anon_vma_prepare(vma
)))
2364 * vma->vm_start/vm_end cannot change under us because the caller
2365 * is required to hold the mmap_sem in read mode. We need the
2366 * anon_vma lock to serialize against concurrent expand_stacks.
2368 anon_vma_lock_write(vma
->anon_vma
);
2370 /* Somebody else might have raced and expanded it already */
2371 if (address
> vma
->vm_end
) {
2372 unsigned long size
, grow
;
2374 size
= address
- vma
->vm_start
;
2375 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2378 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2379 error
= acct_stack_growth(vma
, size
, grow
);
2382 * vma_gap_update() doesn't support concurrent
2383 * updates, but we only hold a shared mmap_sem
2384 * lock here, so we need to protect against
2385 * concurrent vma expansions.
2386 * anon_vma_lock_write() doesn't help here, as
2387 * we don't guarantee that all growable vmas
2388 * in a mm share the same root anon vma.
2389 * So, we reuse mm->page_table_lock to guard
2390 * against concurrent vma expansions.
2392 spin_lock(&mm
->page_table_lock
);
2393 if (vma
->vm_flags
& VM_LOCKED
)
2394 mm
->locked_vm
+= grow
;
2395 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2396 anon_vma_interval_tree_pre_update_vma(vma
);
2397 vma
->vm_end
= address
;
2398 anon_vma_interval_tree_post_update_vma(vma
);
2400 vma_gap_update(vma
->vm_next
);
2402 mm
->highest_vm_end
= vm_end_gap(vma
);
2403 spin_unlock(&mm
->page_table_lock
);
2405 perf_event_mmap(vma
);
2409 anon_vma_unlock_write(vma
->anon_vma
);
2410 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2414 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2417 * vma is the first one with address < vma->vm_start. Have to extend vma.
2419 int expand_downwards(struct vm_area_struct
*vma
,
2420 unsigned long address
)
2422 struct mm_struct
*mm
= vma
->vm_mm
;
2423 struct vm_area_struct
*prev
;
2426 address
&= PAGE_MASK
;
2427 if (address
< mmap_min_addr
)
2430 /* Enforce stack_guard_gap */
2431 prev
= vma
->vm_prev
;
2432 /* Check that both stack segments have the same anon_vma? */
2433 if (prev
&& !(prev
->vm_flags
& VM_GROWSDOWN
) &&
2434 (prev
->vm_flags
& (VM_WRITE
|VM_READ
|VM_EXEC
))) {
2435 if (address
- prev
->vm_end
< stack_guard_gap
)
2439 /* We must make sure the anon_vma is allocated. */
2440 if (unlikely(anon_vma_prepare(vma
)))
2444 * vma->vm_start/vm_end cannot change under us because the caller
2445 * is required to hold the mmap_sem in read mode. We need the
2446 * anon_vma lock to serialize against concurrent expand_stacks.
2448 anon_vma_lock_write(vma
->anon_vma
);
2450 /* Somebody else might have raced and expanded it already */
2451 if (address
< vma
->vm_start
) {
2452 unsigned long size
, grow
;
2454 size
= vma
->vm_end
- address
;
2455 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2458 if (grow
<= vma
->vm_pgoff
) {
2459 error
= acct_stack_growth(vma
, size
, grow
);
2462 * vma_gap_update() doesn't support concurrent
2463 * updates, but we only hold a shared mmap_sem
2464 * lock here, so we need to protect against
2465 * concurrent vma expansions.
2466 * anon_vma_lock_write() doesn't help here, as
2467 * we don't guarantee that all growable vmas
2468 * in a mm share the same root anon vma.
2469 * So, we reuse mm->page_table_lock to guard
2470 * against concurrent vma expansions.
2472 spin_lock(&mm
->page_table_lock
);
2473 if (vma
->vm_flags
& VM_LOCKED
)
2474 mm
->locked_vm
+= grow
;
2475 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2476 anon_vma_interval_tree_pre_update_vma(vma
);
2477 vma
->vm_start
= address
;
2478 vma
->vm_pgoff
-= grow
;
2479 anon_vma_interval_tree_post_update_vma(vma
);
2480 vma_gap_update(vma
);
2481 spin_unlock(&mm
->page_table_lock
);
2483 perf_event_mmap(vma
);
2487 anon_vma_unlock_write(vma
->anon_vma
);
2488 khugepaged_enter_vma_merge(vma
, vma
->vm_flags
);
2493 /* enforced gap between the expanding stack and other mappings. */
2494 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2496 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2501 val
= simple_strtoul(p
, &endptr
, 10);
2503 stack_guard_gap
= val
<< PAGE_SHIFT
;
2507 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2509 #ifdef CONFIG_STACK_GROWSUP
2510 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2512 return expand_upwards(vma
, address
);
2515 struct vm_area_struct
*
2516 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2518 struct vm_area_struct
*vma
, *prev
;
2521 vma
= find_vma_prev(mm
, addr
, &prev
);
2522 if (vma
&& (vma
->vm_start
<= addr
))
2524 /* don't alter vm_end if the coredump is running */
2525 if (!prev
|| !mmget_still_valid(mm
) || expand_stack(prev
, addr
))
2527 if (prev
->vm_flags
& VM_LOCKED
)
2528 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2532 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
2534 return expand_downwards(vma
, address
);
2537 struct vm_area_struct
*
2538 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
2540 struct vm_area_struct
*vma
;
2541 unsigned long start
;
2544 vma
= find_vma(mm
, addr
);
2547 if (vma
->vm_start
<= addr
)
2549 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2551 /* don't alter vm_start if the coredump is running */
2552 if (!mmget_still_valid(mm
))
2554 start
= vma
->vm_start
;
2555 if (expand_stack(vma
, addr
))
2557 if (vma
->vm_flags
& VM_LOCKED
)
2558 populate_vma_page_range(vma
, addr
, start
, NULL
);
2563 EXPORT_SYMBOL_GPL(find_extend_vma
);
2566 * Ok - we have the memory areas we should free on the vma list,
2567 * so release them, and do the vma updates.
2569 * Called with the mm semaphore held.
2571 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
2573 unsigned long nr_accounted
= 0;
2575 /* Update high watermark before we lower total_vm */
2576 update_hiwater_vm(mm
);
2578 long nrpages
= vma_pages(vma
);
2580 if (vma
->vm_flags
& VM_ACCOUNT
)
2581 nr_accounted
+= nrpages
;
2582 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2583 vma
= remove_vma(vma
);
2585 vm_unacct_memory(nr_accounted
);
2590 * Get rid of page table information in the indicated region.
2592 * Called with the mm semaphore held.
2594 static void unmap_region(struct mm_struct
*mm
,
2595 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2596 unsigned long start
, unsigned long end
)
2598 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
2599 struct mmu_gather tlb
;
2602 tlb_gather_mmu(&tlb
, mm
, start
, end
);
2603 update_hiwater_rss(mm
);
2604 unmap_vmas(&tlb
, vma
, start
, end
);
2605 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2606 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
2607 tlb_finish_mmu(&tlb
, start
, end
);
2611 * Create a list of vma's touched by the unmap, removing them from the mm's
2612 * vma list as we go..
2615 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2616 struct vm_area_struct
*prev
, unsigned long end
)
2618 struct vm_area_struct
**insertion_point
;
2619 struct vm_area_struct
*tail_vma
= NULL
;
2621 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
2622 vma
->vm_prev
= NULL
;
2624 vma_rb_erase(vma
, &mm
->mm_rb
);
2628 } while (vma
&& vma
->vm_start
< end
);
2629 *insertion_point
= vma
;
2631 vma
->vm_prev
= prev
;
2632 vma_gap_update(vma
);
2634 mm
->highest_vm_end
= prev
? vm_end_gap(prev
) : 0;
2635 tail_vma
->vm_next
= NULL
;
2637 /* Kill the cache */
2638 vmacache_invalidate(mm
);
2642 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2643 * has already been checked or doesn't make sense to fail.
2645 int __split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2646 unsigned long addr
, int new_below
)
2648 struct vm_area_struct
*new;
2651 if (vma
->vm_ops
&& vma
->vm_ops
->split
) {
2652 err
= vma
->vm_ops
->split(vma
, addr
);
2657 new = vm_area_dup(vma
);
2664 new->vm_start
= addr
;
2665 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2668 err
= vma_dup_policy(vma
, new);
2672 err
= anon_vma_clone(new, vma
);
2679 if (new->vm_ops
&& new->vm_ops
->open
)
2680 new->vm_ops
->open(new);
2683 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2684 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2686 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2692 /* Clean everything up if vma_adjust failed. */
2693 if (new->vm_ops
&& new->vm_ops
->close
)
2694 new->vm_ops
->close(new);
2697 unlink_anon_vmas(new);
2699 mpol_put(vma_policy(new));
2706 * Split a vma into two pieces at address 'addr', a new vma is allocated
2707 * either for the first part or the tail.
2709 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2710 unsigned long addr
, int new_below
)
2712 if (mm
->map_count
>= sysctl_max_map_count
)
2715 return __split_vma(mm
, vma
, addr
, new_below
);
2718 /* Munmap is split into 2 main parts -- this part which finds
2719 * what needs doing, and the areas themselves, which do the
2720 * work. This now handles partial unmappings.
2721 * Jeremy Fitzhardinge <jeremy@goop.org>
2723 int __do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2724 struct list_head
*uf
, bool downgrade
)
2727 struct vm_area_struct
*vma
, *prev
, *last
;
2729 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2732 len
= PAGE_ALIGN(len
);
2738 * arch_unmap() might do unmaps itself. It must be called
2739 * and finish any rbtree manipulation before this code
2740 * runs and also starts to manipulate the rbtree.
2742 arch_unmap(mm
, start
, end
);
2744 /* Find the first overlapping VMA */
2745 vma
= find_vma(mm
, start
);
2748 prev
= vma
->vm_prev
;
2749 /* we have start < vma->vm_end */
2751 /* if it doesn't overlap, we have nothing.. */
2752 if (vma
->vm_start
>= end
)
2756 * If we need to split any vma, do it now to save pain later.
2758 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2759 * unmapped vm_area_struct will remain in use: so lower split_vma
2760 * places tmp vma above, and higher split_vma places tmp vma below.
2762 if (start
> vma
->vm_start
) {
2766 * Make sure that map_count on return from munmap() will
2767 * not exceed its limit; but let map_count go just above
2768 * its limit temporarily, to help free resources as expected.
2770 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2773 error
= __split_vma(mm
, vma
, start
, 0);
2779 /* Does it split the last one? */
2780 last
= find_vma(mm
, end
);
2781 if (last
&& end
> last
->vm_start
) {
2782 int error
= __split_vma(mm
, last
, end
, 1);
2786 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2790 * If userfaultfd_unmap_prep returns an error the vmas
2791 * will remain splitted, but userland will get a
2792 * highly unexpected error anyway. This is no
2793 * different than the case where the first of the two
2794 * __split_vma fails, but we don't undo the first
2795 * split, despite we could. This is unlikely enough
2796 * failure that it's not worth optimizing it for.
2798 int error
= userfaultfd_unmap_prep(vma
, start
, end
, uf
);
2804 * unlock any mlock()ed ranges before detaching vmas
2806 if (mm
->locked_vm
) {
2807 struct vm_area_struct
*tmp
= vma
;
2808 while (tmp
&& tmp
->vm_start
< end
) {
2809 if (tmp
->vm_flags
& VM_LOCKED
) {
2810 mm
->locked_vm
-= vma_pages(tmp
);
2811 munlock_vma_pages_all(tmp
);
2818 /* Detach vmas from rbtree */
2819 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2822 downgrade_write(&mm
->mmap_sem
);
2824 unmap_region(mm
, vma
, prev
, start
, end
);
2826 /* Fix up all other VM information */
2827 remove_vma_list(mm
, vma
);
2829 return downgrade
? 1 : 0;
2832 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2833 struct list_head
*uf
)
2835 return __do_munmap(mm
, start
, len
, uf
, false);
2838 static int __vm_munmap(unsigned long start
, size_t len
, bool downgrade
)
2841 struct mm_struct
*mm
= current
->mm
;
2844 if (down_write_killable(&mm
->mmap_sem
))
2847 ret
= __do_munmap(mm
, start
, len
, &uf
, downgrade
);
2849 * Returning 1 indicates mmap_sem is downgraded.
2850 * But 1 is not legal return value of vm_munmap() and munmap(), reset
2851 * it to 0 before return.
2854 up_read(&mm
->mmap_sem
);
2857 up_write(&mm
->mmap_sem
);
2859 userfaultfd_unmap_complete(mm
, &uf
);
2863 int vm_munmap(unsigned long start
, size_t len
)
2865 return __vm_munmap(start
, len
, false);
2867 EXPORT_SYMBOL(vm_munmap
);
2869 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2871 profile_munmap(addr
);
2872 return __vm_munmap(addr
, len
, true);
2877 * Emulation of deprecated remap_file_pages() syscall.
2879 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2880 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2883 struct mm_struct
*mm
= current
->mm
;
2884 struct vm_area_struct
*vma
;
2885 unsigned long populate
= 0;
2886 unsigned long ret
= -EINVAL
;
2887 struct file
*file
, *prfile
;
2889 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
2890 current
->comm
, current
->pid
);
2894 start
= start
& PAGE_MASK
;
2895 size
= size
& PAGE_MASK
;
2897 if (start
+ size
<= start
)
2900 /* Does pgoff wrap? */
2901 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2904 if (down_write_killable(&mm
->mmap_sem
))
2907 vma
= find_vma(mm
, start
);
2909 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2912 if (start
< vma
->vm_start
)
2915 if (start
+ size
> vma
->vm_end
) {
2916 struct vm_area_struct
*next
;
2918 for (next
= vma
->vm_next
; next
; next
= next
->vm_next
) {
2919 /* hole between vmas ? */
2920 if (next
->vm_start
!= next
->vm_prev
->vm_end
)
2923 if (next
->vm_file
!= vma
->vm_file
)
2926 if (next
->vm_flags
!= vma
->vm_flags
)
2929 if (start
+ size
<= next
->vm_end
)
2937 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
2938 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
2939 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
2941 flags
&= MAP_NONBLOCK
;
2942 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
2943 if (vma
->vm_flags
& VM_LOCKED
) {
2944 struct vm_area_struct
*tmp
;
2945 flags
|= MAP_LOCKED
;
2947 /* drop PG_Mlocked flag for over-mapped range */
2948 for (tmp
= vma
; tmp
->vm_start
>= start
+ size
;
2949 tmp
= tmp
->vm_next
) {
2951 * Split pmd and munlock page on the border
2954 vma_adjust_trans_huge(tmp
, start
, start
+ size
, 0);
2956 munlock_vma_pages_range(tmp
,
2957 max(tmp
->vm_start
, start
),
2958 min(tmp
->vm_end
, start
+ size
));
2963 file
= vma
->vm_file
;
2964 prfile
= vma
->vm_prfile
;
2965 ret
= do_mmap_pgoff(vma
->vm_file
, start
, size
,
2966 prot
, flags
, pgoff
, &populate
, NULL
);
2967 if (!IS_ERR_VALUE(ret
) && file
&& prfile
) {
2968 struct vm_area_struct
*new_vma
;
2970 new_vma
= find_vma(mm
, ret
);
2971 if (!new_vma
->vm_prfile
)
2972 new_vma
->vm_prfile
= prfile
;
2977 * two fput()s instead of vma_fput(vma),
2978 * coz vma may not be available anymore.
2984 up_write(&mm
->mmap_sem
);
2986 mm_populate(ret
, populate
);
2987 if (!IS_ERR_VALUE(ret
))
2993 * this is really a simplified "do_mmap". it only handles
2994 * anonymous maps. eventually we may be able to do some
2995 * brk-specific accounting here.
2997 static int do_brk_flags(unsigned long addr
, unsigned long len
, unsigned long flags
, struct list_head
*uf
)
2999 struct mm_struct
*mm
= current
->mm
;
3000 struct vm_area_struct
*vma
, *prev
;
3001 struct rb_node
**rb_link
, *rb_parent
;
3002 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
3005 /* Until we need other flags, refuse anything except VM_EXEC. */
3006 if ((flags
& (~VM_EXEC
)) != 0)
3008 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
3010 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
3011 if (offset_in_page(error
))
3014 error
= mlock_future_check(mm
, mm
->def_flags
, len
);
3019 * Clear old maps. this also does some error checking for us
3021 while (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
,
3023 if (do_munmap(mm
, addr
, len
, uf
))
3027 /* Check against address space limits *after* clearing old maps... */
3028 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
3031 if (mm
->map_count
> sysctl_max_map_count
)
3034 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
3037 /* Can we just expand an old private anonymous mapping? */
3038 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
3039 NULL
, NULL
, pgoff
, NULL
, NULL_VM_UFFD_CTX
);
3044 * create a vma struct for an anonymous mapping
3046 vma
= vm_area_alloc(mm
);
3048 vm_unacct_memory(len
>> PAGE_SHIFT
);
3052 vma_set_anonymous(vma
);
3053 vma
->vm_start
= addr
;
3054 vma
->vm_end
= addr
+ len
;
3055 vma
->vm_pgoff
= pgoff
;
3056 vma
->vm_flags
= flags
;
3057 vma
->vm_page_prot
= vm_get_page_prot(flags
);
3058 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3060 perf_event_mmap(vma
);
3061 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3062 mm
->data_vm
+= len
>> PAGE_SHIFT
;
3063 if (flags
& VM_LOCKED
)
3064 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
3065 vma
->vm_flags
|= VM_SOFTDIRTY
;
3069 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
3071 struct mm_struct
*mm
= current
->mm
;
3077 len
= PAGE_ALIGN(request
);
3083 if (down_write_killable(&mm
->mmap_sem
))
3086 ret
= do_brk_flags(addr
, len
, flags
, &uf
);
3087 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3088 up_write(&mm
->mmap_sem
);
3089 userfaultfd_unmap_complete(mm
, &uf
);
3090 if (populate
&& !ret
)
3091 mm_populate(addr
, len
);
3094 EXPORT_SYMBOL(vm_brk_flags
);
3096 int vm_brk(unsigned long addr
, unsigned long len
)
3098 return vm_brk_flags(addr
, len
, 0);
3100 EXPORT_SYMBOL(vm_brk
);
3102 /* Release all mmaps. */
3103 void exit_mmap(struct mm_struct
*mm
)
3105 struct mmu_gather tlb
;
3106 struct vm_area_struct
*vma
;
3107 unsigned long nr_accounted
= 0;
3109 /* mm's last user has gone, and its about to be pulled down */
3110 mmu_notifier_release(mm
);
3112 if (unlikely(mm_is_oom_victim(mm
))) {
3114 * Manually reap the mm to free as much memory as possible.
3115 * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3116 * this mm from further consideration. Taking mm->mmap_sem for
3117 * write after setting MMF_OOM_SKIP will guarantee that the oom
3118 * reaper will not run on this mm again after mmap_sem is
3121 * Nothing can be holding mm->mmap_sem here and the above call
3122 * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3123 * __oom_reap_task_mm() will not block.
3125 * This needs to be done before calling munlock_vma_pages_all(),
3126 * which clears VM_LOCKED, otherwise the oom reaper cannot
3129 (void)__oom_reap_task_mm(mm
);
3131 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3132 down_write(&mm
->mmap_sem
);
3133 up_write(&mm
->mmap_sem
);
3136 if (mm
->locked_vm
) {
3139 if (vma
->vm_flags
& VM_LOCKED
)
3140 munlock_vma_pages_all(vma
);
3148 if (!vma
) /* Can happen if dup_mmap() received an OOM */
3153 tlb_gather_mmu(&tlb
, mm
, 0, -1);
3154 /* update_hiwater_rss(mm) here? but nobody should be looking */
3155 /* Use -1 here to ensure all VMAs in the mm are unmapped */
3156 unmap_vmas(&tlb
, vma
, 0, -1);
3157 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, USER_PGTABLES_CEILING
);
3158 tlb_finish_mmu(&tlb
, 0, -1);
3161 * Walk the list again, actually closing and freeing it,
3162 * with preemption enabled, without holding any MM locks.
3165 if (vma
->vm_flags
& VM_ACCOUNT
)
3166 nr_accounted
+= vma_pages(vma
);
3167 vma
= remove_vma(vma
);
3169 vm_unacct_memory(nr_accounted
);
3172 /* Insert vm structure into process list sorted by address
3173 * and into the inode's i_mmap tree. If vm_file is non-NULL
3174 * then i_mmap_rwsem is taken here.
3176 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3178 struct vm_area_struct
*prev
;
3179 struct rb_node
**rb_link
, *rb_parent
;
3181 if (find_vma_links(mm
, vma
->vm_start
, vma
->vm_end
,
3182 &prev
, &rb_link
, &rb_parent
))
3184 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3185 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
3189 * The vm_pgoff of a purely anonymous vma should be irrelevant
3190 * until its first write fault, when page's anon_vma and index
3191 * are set. But now set the vm_pgoff it will almost certainly
3192 * end up with (unless mremap moves it elsewhere before that
3193 * first wfault), so /proc/pid/maps tells a consistent story.
3195 * By setting it to reflect the virtual start address of the
3196 * vma, merges and splits can happen in a seamless way, just
3197 * using the existing file pgoff checks and manipulations.
3198 * Similarly in do_mmap_pgoff and in do_brk.
3200 if (vma_is_anonymous(vma
)) {
3201 BUG_ON(vma
->anon_vma
);
3202 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3205 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
3210 * Copy the vma structure to a new location in the same mm,
3211 * prior to moving page table entries, to effect an mremap move.
3213 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3214 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3215 bool *need_rmap_locks
)
3217 struct vm_area_struct
*vma
= *vmap
;
3218 unsigned long vma_start
= vma
->vm_start
;
3219 struct mm_struct
*mm
= vma
->vm_mm
;
3220 struct vm_area_struct
*new_vma
, *prev
;
3221 struct rb_node
**rb_link
, *rb_parent
;
3222 bool faulted_in_anon_vma
= true;
3225 * If anonymous vma has not yet been faulted, update new pgoff
3226 * to match new location, to increase its chance of merging.
3228 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3229 pgoff
= addr
>> PAGE_SHIFT
;
3230 faulted_in_anon_vma
= false;
3233 if (find_vma_links(mm
, addr
, addr
+ len
, &prev
, &rb_link
, &rb_parent
))
3234 return NULL
; /* should never get here */
3235 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
3236 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
3237 vma
->vm_userfaultfd_ctx
);
3240 * Source vma may have been merged into new_vma
3242 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3243 vma_start
< new_vma
->vm_end
)) {
3245 * The only way we can get a vma_merge with
3246 * self during an mremap is if the vma hasn't
3247 * been faulted in yet and we were allowed to
3248 * reset the dst vma->vm_pgoff to the
3249 * destination address of the mremap to allow
3250 * the merge to happen. mremap must change the
3251 * vm_pgoff linearity between src and dst vmas
3252 * (in turn preventing a vma_merge) to be
3253 * safe. It is only safe to keep the vm_pgoff
3254 * linear if there are no pages mapped yet.
3256 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3257 *vmap
= vma
= new_vma
;
3259 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3261 new_vma
= vm_area_dup(vma
);
3264 new_vma
->vm_start
= addr
;
3265 new_vma
->vm_end
= addr
+ len
;
3266 new_vma
->vm_pgoff
= pgoff
;
3267 if (vma_dup_policy(vma
, new_vma
))
3269 if (anon_vma_clone(new_vma
, vma
))
3270 goto out_free_mempol
;
3271 if (new_vma
->vm_file
)
3272 vma_get_file(new_vma
);
3273 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3274 new_vma
->vm_ops
->open(new_vma
);
3275 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
3276 *need_rmap_locks
= false;
3281 mpol_put(vma_policy(new_vma
));
3283 vm_area_free(new_vma
);
3289 * Return true if the calling process may expand its vm space by the passed
3292 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3294 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3297 if (is_data_mapping(flags
) &&
3298 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3299 /* Workaround for Valgrind */
3300 if (rlimit(RLIMIT_DATA
) == 0 &&
3301 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3304 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3305 current
->comm
, current
->pid
,
3306 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3307 rlimit(RLIMIT_DATA
),
3308 ignore_rlimit_data
? "" : " or use boot option ignore_rlimit_data");
3310 if (!ignore_rlimit_data
)
3317 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3319 mm
->total_vm
+= npages
;
3321 if (is_exec_mapping(flags
))
3322 mm
->exec_vm
+= npages
;
3323 else if (is_stack_mapping(flags
))
3324 mm
->stack_vm
+= npages
;
3325 else if (is_data_mapping(flags
))
3326 mm
->data_vm
+= npages
;
3329 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
);
3332 * Having a close hook prevents vma merging regardless of flags.
3334 static void special_mapping_close(struct vm_area_struct
*vma
)
3338 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3340 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3343 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3345 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3347 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3351 return sm
->mremap(sm
, new_vma
);
3356 static const struct vm_operations_struct special_mapping_vmops
= {
3357 .close
= special_mapping_close
,
3358 .fault
= special_mapping_fault
,
3359 .mremap
= special_mapping_mremap
,
3360 .name
= special_mapping_name
,
3363 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3364 .close
= special_mapping_close
,
3365 .fault
= special_mapping_fault
,
3368 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
)
3370 struct vm_area_struct
*vma
= vmf
->vma
;
3372 struct page
**pages
;
3374 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3375 pages
= vma
->vm_private_data
;
3377 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3380 return sm
->fault(sm
, vmf
->vma
, vmf
);
3385 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3389 struct page
*page
= *pages
;
3395 return VM_FAULT_SIGBUS
;
3398 static struct vm_area_struct
*__install_special_mapping(
3399 struct mm_struct
*mm
,
3400 unsigned long addr
, unsigned long len
,
3401 unsigned long vm_flags
, void *priv
,
3402 const struct vm_operations_struct
*ops
)
3405 struct vm_area_struct
*vma
;
3407 vma
= vm_area_alloc(mm
);
3408 if (unlikely(vma
== NULL
))
3409 return ERR_PTR(-ENOMEM
);
3411 vma
->vm_start
= addr
;
3412 vma
->vm_end
= addr
+ len
;
3414 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
| VM_SOFTDIRTY
;
3415 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3418 vma
->vm_private_data
= priv
;
3420 ret
= insert_vm_struct(mm
, vma
);
3424 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3426 perf_event_mmap(vma
);
3432 return ERR_PTR(ret
);
3435 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3436 const struct vm_special_mapping
*sm
)
3438 return vma
->vm_private_data
== sm
&&
3439 (vma
->vm_ops
== &special_mapping_vmops
||
3440 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3444 * Called with mm->mmap_sem held for writing.
3445 * Insert a new vma covering the given region, with the given flags.
3446 * Its pages are supplied by the given array of struct page *.
3447 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3448 * The region past the last page supplied will always produce SIGBUS.
3449 * The array pointer and the pages it points to are assumed to stay alive
3450 * for as long as this mapping might exist.
3452 struct vm_area_struct
*_install_special_mapping(
3453 struct mm_struct
*mm
,
3454 unsigned long addr
, unsigned long len
,
3455 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3457 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3458 &special_mapping_vmops
);
3461 int install_special_mapping(struct mm_struct
*mm
,
3462 unsigned long addr
, unsigned long len
,
3463 unsigned long vm_flags
, struct page
**pages
)
3465 struct vm_area_struct
*vma
= __install_special_mapping(
3466 mm
, addr
, len
, vm_flags
, (void *)pages
,
3467 &legacy_special_mapping_vmops
);
3469 return PTR_ERR_OR_ZERO(vma
);
3472 static DEFINE_MUTEX(mm_all_locks_mutex
);
3474 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3476 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3478 * The LSB of head.next can't change from under us
3479 * because we hold the mm_all_locks_mutex.
3481 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_sem
);
3483 * We can safely modify head.next after taking the
3484 * anon_vma->root->rwsem. If some other vma in this mm shares
3485 * the same anon_vma we won't take it again.
3487 * No need of atomic instructions here, head.next
3488 * can't change from under us thanks to the
3489 * anon_vma->root->rwsem.
3491 if (__test_and_set_bit(0, (unsigned long *)
3492 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3497 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3499 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3501 * AS_MM_ALL_LOCKS can't change from under us because
3502 * we hold the mm_all_locks_mutex.
3504 * Operations on ->flags have to be atomic because
3505 * even if AS_MM_ALL_LOCKS is stable thanks to the
3506 * mm_all_locks_mutex, there may be other cpus
3507 * changing other bitflags in parallel to us.
3509 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3511 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_sem
);
3516 * This operation locks against the VM for all pte/vma/mm related
3517 * operations that could ever happen on a certain mm. This includes
3518 * vmtruncate, try_to_unmap, and all page faults.
3520 * The caller must take the mmap_sem in write mode before calling
3521 * mm_take_all_locks(). The caller isn't allowed to release the
3522 * mmap_sem until mm_drop_all_locks() returns.
3524 * mmap_sem in write mode is required in order to block all operations
3525 * that could modify pagetables and free pages without need of
3526 * altering the vma layout. It's also needed in write mode to avoid new
3527 * anon_vmas to be associated with existing vmas.
3529 * A single task can't take more than one mm_take_all_locks() in a row
3530 * or it would deadlock.
3532 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3533 * mapping->flags avoid to take the same lock twice, if more than one
3534 * vma in this mm is backed by the same anon_vma or address_space.
3536 * We take locks in following order, accordingly to comment at beginning
3538 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3540 * - all i_mmap_rwsem locks;
3541 * - all anon_vma->rwseml
3543 * We can take all locks within these types randomly because the VM code
3544 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3545 * mm_all_locks_mutex.
3547 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3548 * that may have to take thousand of locks.
3550 * mm_take_all_locks() can fail if it's interrupted by signals.
3552 int mm_take_all_locks(struct mm_struct
*mm
)
3554 struct vm_area_struct
*vma
;
3555 struct anon_vma_chain
*avc
;
3557 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3559 mutex_lock(&mm_all_locks_mutex
);
3561 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3562 if (signal_pending(current
))
3564 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3565 is_vm_hugetlb_page(vma
))
3566 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3569 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3570 if (signal_pending(current
))
3572 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3573 !is_vm_hugetlb_page(vma
))
3574 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3577 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3578 if (signal_pending(current
))
3581 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3582 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3588 mm_drop_all_locks(mm
);
3592 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3594 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3596 * The LSB of head.next can't change to 0 from under
3597 * us because we hold the mm_all_locks_mutex.
3599 * We must however clear the bitflag before unlocking
3600 * the vma so the users using the anon_vma->rb_root will
3601 * never see our bitflag.
3603 * No need of atomic instructions here, head.next
3604 * can't change from under us until we release the
3605 * anon_vma->root->rwsem.
3607 if (!__test_and_clear_bit(0, (unsigned long *)
3608 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3610 anon_vma_unlock_write(anon_vma
);
3614 static void vm_unlock_mapping(struct address_space
*mapping
)
3616 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3618 * AS_MM_ALL_LOCKS can't change to 0 from under us
3619 * because we hold the mm_all_locks_mutex.
3621 i_mmap_unlock_write(mapping
);
3622 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3629 * The mmap_sem cannot be released by the caller until
3630 * mm_drop_all_locks() returns.
3632 void mm_drop_all_locks(struct mm_struct
*mm
)
3634 struct vm_area_struct
*vma
;
3635 struct anon_vma_chain
*avc
;
3637 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
3638 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3640 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
3642 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3643 vm_unlock_anon_vma(avc
->anon_vma
);
3644 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3645 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3648 mutex_unlock(&mm_all_locks_mutex
);
3652 * initialise the percpu counter for VM
3654 void __init
mmap_init(void)
3658 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3663 * Initialise sysctl_user_reserve_kbytes.
3665 * This is intended to prevent a user from starting a single memory hogging
3666 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3669 * The default value is min(3% of free memory, 128MB)
3670 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3672 static int init_user_reserve(void)
3674 unsigned long free_kbytes
;
3676 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3678 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3681 subsys_initcall(init_user_reserve
);
3684 * Initialise sysctl_admin_reserve_kbytes.
3686 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3687 * to log in and kill a memory hogging process.
3689 * Systems with more than 256MB will reserve 8MB, enough to recover
3690 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3691 * only reserve 3% of free pages by default.
3693 static int init_admin_reserve(void)
3695 unsigned long free_kbytes
;
3697 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3699 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3702 subsys_initcall(init_admin_reserve
);
3705 * Reinititalise user and admin reserves if memory is added or removed.
3707 * The default user reserve max is 128MB, and the default max for the
3708 * admin reserve is 8MB. These are usually, but not always, enough to
3709 * enable recovery from a memory hogging process using login/sshd, a shell,
3710 * and tools like top. It may make sense to increase or even disable the
3711 * reserve depending on the existence of swap or variations in the recovery
3712 * tools. So, the admin may have changed them.
3714 * If memory is added and the reserves have been eliminated or increased above
3715 * the default max, then we'll trust the admin.
3717 * If memory is removed and there isn't enough free memory, then we
3718 * need to reset the reserves.
3720 * Otherwise keep the reserve set by the admin.
3722 static int reserve_mem_notifier(struct notifier_block
*nb
,
3723 unsigned long action
, void *data
)
3725 unsigned long tmp
, free_kbytes
;
3729 /* Default max is 128MB. Leave alone if modified by operator. */
3730 tmp
= sysctl_user_reserve_kbytes
;
3731 if (0 < tmp
&& tmp
< (1UL << 17))
3732 init_user_reserve();
3734 /* Default max is 8MB. Leave alone if modified by operator. */
3735 tmp
= sysctl_admin_reserve_kbytes
;
3736 if (0 < tmp
&& tmp
< (1UL << 13))
3737 init_admin_reserve();
3741 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
3743 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3744 init_user_reserve();
3745 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3746 sysctl_user_reserve_kbytes
);
3749 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3750 init_admin_reserve();
3751 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3752 sysctl_admin_reserve_kbytes
);
3761 static struct notifier_block reserve_mem_nb
= {
3762 .notifier_call
= reserve_mem_notifier
,
3765 static int __meminit
init_reserve_notifier(void)
3767 if (register_hotmemory_notifier(&reserve_mem_nb
))
3768 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3772 subsys_initcall(init_reserve_notifier
);