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/mm_inline.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/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
44 #include <linux/userfaultfd_k.h>
45 #include <linux/moduleparam.h>
46 #include <linux/pkeys.h>
47 #include <linux/oom.h>
48 #include <linux/sched/mm.h>
49 #include <linux/ksm.h>
51 #include <linux/uaccess.h>
52 #include <asm/cacheflush.h>
54 #include <asm/mmu_context.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/mmap.h>
61 #ifndef arch_mmap_check
62 #define arch_mmap_check(addr, len, flags) (0)
65 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
66 const int mmap_rnd_bits_min
= CONFIG_ARCH_MMAP_RND_BITS_MIN
;
67 int mmap_rnd_bits_max __ro_after_init
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
68 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
70 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
71 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
72 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
73 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
76 static bool ignore_rlimit_data
;
77 core_param(ignore_rlimit_data
, ignore_rlimit_data
, bool, 0644);
79 static void unmap_region(struct mm_struct
*mm
, struct ma_state
*mas
,
80 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
81 struct vm_area_struct
*next
, unsigned long start
,
82 unsigned long end
, unsigned long tree_end
, bool mm_wr_locked
);
84 static pgprot_t
vm_pgprot_modify(pgprot_t oldprot
, unsigned long vm_flags
)
86 return pgprot_modify(oldprot
, vm_get_page_prot(vm_flags
));
89 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
90 void vma_set_page_prot(struct vm_area_struct
*vma
)
92 unsigned long vm_flags
= vma
->vm_flags
;
93 pgprot_t vm_page_prot
;
95 vm_page_prot
= vm_pgprot_modify(vma
->vm_page_prot
, vm_flags
);
96 if (vma_wants_writenotify(vma
, vm_page_prot
)) {
97 vm_flags
&= ~VM_SHARED
;
98 vm_page_prot
= vm_pgprot_modify(vm_page_prot
, vm_flags
);
100 /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
101 WRITE_ONCE(vma
->vm_page_prot
, vm_page_prot
);
105 * Requires inode->i_mapping->i_mmap_rwsem
107 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
108 struct address_space
*mapping
)
110 if (vma_is_shared_maywrite(vma
))
111 mapping_unmap_writable(mapping
);
113 flush_dcache_mmap_lock(mapping
);
114 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
115 flush_dcache_mmap_unlock(mapping
);
119 * Unlink a file-based vm structure from its interval tree, to hide
120 * vma from rmap and vmtruncate before freeing its page tables.
122 void unlink_file_vma(struct vm_area_struct
*vma
)
124 struct file
*file
= vma
->vm_file
;
127 struct address_space
*mapping
= file
->f_mapping
;
128 i_mmap_lock_write(mapping
);
129 __remove_shared_vm_struct(vma
, mapping
);
130 i_mmap_unlock_write(mapping
);
134 void unlink_file_vma_batch_init(struct unlink_vma_file_batch
*vb
)
139 static void unlink_file_vma_batch_process(struct unlink_vma_file_batch
*vb
)
141 struct address_space
*mapping
;
144 mapping
= vb
->vmas
[0]->vm_file
->f_mapping
;
145 i_mmap_lock_write(mapping
);
146 for (i
= 0; i
< vb
->count
; i
++) {
147 VM_WARN_ON_ONCE(vb
->vmas
[i
]->vm_file
->f_mapping
!= mapping
);
148 __remove_shared_vm_struct(vb
->vmas
[i
], mapping
);
150 i_mmap_unlock_write(mapping
);
152 unlink_file_vma_batch_init(vb
);
155 void unlink_file_vma_batch_add(struct unlink_vma_file_batch
*vb
,
156 struct vm_area_struct
*vma
)
158 if (vma
->vm_file
== NULL
)
161 if ((vb
->count
> 0 && vb
->vmas
[0]->vm_file
!= vma
->vm_file
) ||
162 vb
->count
== ARRAY_SIZE(vb
->vmas
))
163 unlink_file_vma_batch_process(vb
);
165 vb
->vmas
[vb
->count
] = vma
;
169 void unlink_file_vma_batch_final(struct unlink_vma_file_batch
*vb
)
172 unlink_file_vma_batch_process(vb
);
176 * Close a vm structure and free it.
178 static void remove_vma(struct vm_area_struct
*vma
, bool unreachable
)
181 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
182 vma
->vm_ops
->close(vma
);
185 mpol_put(vma_policy(vma
));
192 static inline struct vm_area_struct
*vma_prev_limit(struct vma_iterator
*vmi
,
195 return mas_prev(&vmi
->mas
, min
);
199 * check_brk_limits() - Use platform specific check of range & verify mlock
201 * @addr: The address to check
202 * @len: The size of increase.
204 * Return: 0 on success.
206 static int check_brk_limits(unsigned long addr
, unsigned long len
)
208 unsigned long mapped_addr
;
210 mapped_addr
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
211 if (IS_ERR_VALUE(mapped_addr
))
214 return mlock_future_ok(current
->mm
, current
->mm
->def_flags
, len
)
217 static int do_brk_flags(struct vma_iterator
*vmi
, struct vm_area_struct
*brkvma
,
218 unsigned long addr
, unsigned long request
, unsigned long flags
);
219 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
221 unsigned long newbrk
, oldbrk
, origbrk
;
222 struct mm_struct
*mm
= current
->mm
;
223 struct vm_area_struct
*brkvma
, *next
= NULL
;
224 unsigned long min_brk
;
225 bool populate
= false;
227 struct vma_iterator vmi
;
229 if (mmap_write_lock_killable(mm
))
234 #ifdef CONFIG_COMPAT_BRK
236 * CONFIG_COMPAT_BRK can still be overridden by setting
237 * randomize_va_space to 2, which will still cause mm->start_brk
238 * to be arbitrarily shifted
240 if (current
->brk_randomized
)
241 min_brk
= mm
->start_brk
;
243 min_brk
= mm
->end_data
;
245 min_brk
= mm
->start_brk
;
251 * Check against rlimit here. If this check is done later after the test
252 * of oldbrk with newbrk then it can escape the test and let the data
253 * segment grow beyond its set limit the in case where the limit is
254 * not page aligned -Ram Gupta
256 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
257 mm
->end_data
, mm
->start_data
))
260 newbrk
= PAGE_ALIGN(brk
);
261 oldbrk
= PAGE_ALIGN(mm
->brk
);
262 if (oldbrk
== newbrk
) {
267 /* Always allow shrinking brk. */
268 if (brk
<= mm
->brk
) {
269 /* Search one past newbrk */
270 vma_iter_init(&vmi
, mm
, newbrk
);
271 brkvma
= vma_find(&vmi
, oldbrk
);
272 if (!brkvma
|| brkvma
->vm_start
>= oldbrk
)
273 goto out
; /* mapping intersects with an existing non-brk vma. */
275 * mm->brk must be protected by write mmap_lock.
276 * do_vma_munmap() will drop the lock on success, so update it
277 * before calling do_vma_munmap().
280 if (do_vma_munmap(&vmi
, brkvma
, newbrk
, oldbrk
, &uf
, true))
283 goto success_unlocked
;
286 if (check_brk_limits(oldbrk
, newbrk
- oldbrk
))
290 * Only check if the next VMA is within the stack_guard_gap of the
293 vma_iter_init(&vmi
, mm
, oldbrk
);
294 next
= vma_find(&vmi
, newbrk
+ PAGE_SIZE
+ stack_guard_gap
);
295 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
298 brkvma
= vma_prev_limit(&vmi
, mm
->start_brk
);
299 /* Ok, looks good - let it rip. */
300 if (do_brk_flags(&vmi
, brkvma
, oldbrk
, newbrk
- oldbrk
, 0) < 0)
304 if (mm
->def_flags
& VM_LOCKED
)
308 mmap_write_unlock(mm
);
310 userfaultfd_unmap_complete(mm
, &uf
);
312 mm_populate(oldbrk
, newbrk
- oldbrk
);
317 mmap_write_unlock(mm
);
321 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
322 static void validate_mm(struct mm_struct
*mm
)
326 struct vm_area_struct
*vma
;
327 VMA_ITERATOR(vmi
, mm
, 0);
329 mt_validate(&mm
->mm_mt
);
330 for_each_vma(vmi
, vma
) {
331 #ifdef CONFIG_DEBUG_VM_RB
332 struct anon_vma
*anon_vma
= vma
->anon_vma
;
333 struct anon_vma_chain
*avc
;
335 unsigned long vmi_start
, vmi_end
;
338 vmi_start
= vma_iter_addr(&vmi
);
339 vmi_end
= vma_iter_end(&vmi
);
340 if (VM_WARN_ON_ONCE_MM(vma
->vm_end
!= vmi_end
, mm
))
343 if (VM_WARN_ON_ONCE_MM(vma
->vm_start
!= vmi_start
, mm
))
347 pr_emerg("issue in %s\n", current
->comm
);
350 pr_emerg("tree range: %px start %lx end %lx\n", vma
,
351 vmi_start
, vmi_end
- 1);
352 vma_iter_dump_tree(&vmi
);
355 #ifdef CONFIG_DEBUG_VM_RB
357 anon_vma_lock_read(anon_vma
);
358 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
359 anon_vma_interval_tree_verify(avc
);
360 anon_vma_unlock_read(anon_vma
);
365 if (i
!= mm
->map_count
) {
366 pr_emerg("map_count %d vma iterator %d\n", mm
->map_count
, i
);
369 VM_BUG_ON_MM(bug
, mm
);
372 #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */
373 #define validate_mm(mm) do { } while (0)
374 #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
377 * vma has some anon_vma assigned, and is already inserted on that
378 * anon_vma's interval trees.
380 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
381 * vma must be removed from the anon_vma's interval trees using
382 * anon_vma_interval_tree_pre_update_vma().
384 * After the update, the vma will be reinserted using
385 * anon_vma_interval_tree_post_update_vma().
387 * The entire update must be protected by exclusive mmap_lock and by
388 * the root anon_vma's mutex.
391 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
393 struct anon_vma_chain
*avc
;
395 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
396 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
400 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
402 struct anon_vma_chain
*avc
;
404 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
405 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
408 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
409 unsigned long addr
, unsigned long end
)
411 VMA_ITERATOR(vmi
, mm
, addr
);
412 struct vm_area_struct
*vma
;
413 unsigned long nr_pages
= 0;
415 for_each_vma_range(vmi
, vma
, end
) {
416 unsigned long vm_start
= max(addr
, vma
->vm_start
);
417 unsigned long vm_end
= min(end
, vma
->vm_end
);
419 nr_pages
+= PHYS_PFN(vm_end
- vm_start
);
425 static void __vma_link_file(struct vm_area_struct
*vma
,
426 struct address_space
*mapping
)
428 if (vma_is_shared_maywrite(vma
))
429 mapping_allow_writable(mapping
);
431 flush_dcache_mmap_lock(mapping
);
432 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
433 flush_dcache_mmap_unlock(mapping
);
436 static void vma_link_file(struct vm_area_struct
*vma
)
438 struct file
*file
= vma
->vm_file
;
439 struct address_space
*mapping
;
442 mapping
= file
->f_mapping
;
443 i_mmap_lock_write(mapping
);
444 __vma_link_file(vma
, mapping
);
445 i_mmap_unlock_write(mapping
);
449 static int vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
451 VMA_ITERATOR(vmi
, mm
, 0);
453 vma_iter_config(&vmi
, vma
->vm_start
, vma
->vm_end
);
454 if (vma_iter_prealloc(&vmi
, vma
))
457 vma_start_write(vma
);
458 vma_iter_store(&vmi
, vma
);
466 * init_multi_vma_prep() - Initializer for struct vma_prepare
467 * @vp: The vma_prepare struct
468 * @vma: The vma that will be altered once locked
469 * @next: The next vma if it is to be adjusted
470 * @remove: The first vma to be removed
471 * @remove2: The second vma to be removed
473 static inline void init_multi_vma_prep(struct vma_prepare
*vp
,
474 struct vm_area_struct
*vma
, struct vm_area_struct
*next
,
475 struct vm_area_struct
*remove
, struct vm_area_struct
*remove2
)
477 memset(vp
, 0, sizeof(struct vma_prepare
));
479 vp
->anon_vma
= vma
->anon_vma
;
481 vp
->remove2
= remove2
;
483 if (!vp
->anon_vma
&& next
)
484 vp
->anon_vma
= next
->anon_vma
;
486 vp
->file
= vma
->vm_file
;
488 vp
->mapping
= vma
->vm_file
->f_mapping
;
493 * init_vma_prep() - Initializer wrapper for vma_prepare struct
494 * @vp: The vma_prepare struct
495 * @vma: The vma that will be altered once locked
497 static inline void init_vma_prep(struct vma_prepare
*vp
,
498 struct vm_area_struct
*vma
)
500 init_multi_vma_prep(vp
, vma
, NULL
, NULL
, NULL
);
505 * vma_prepare() - Helper function for handling locking VMAs prior to altering
506 * @vp: The initialized vma_prepare struct
508 static inline void vma_prepare(struct vma_prepare
*vp
)
511 uprobe_munmap(vp
->vma
, vp
->vma
->vm_start
, vp
->vma
->vm_end
);
514 uprobe_munmap(vp
->adj_next
, vp
->adj_next
->vm_start
,
515 vp
->adj_next
->vm_end
);
517 i_mmap_lock_write(vp
->mapping
);
518 if (vp
->insert
&& vp
->insert
->vm_file
) {
520 * Put into interval tree now, so instantiated pages
521 * are visible to arm/parisc __flush_dcache_page
522 * throughout; but we cannot insert into address
523 * space until vma start or end is updated.
525 __vma_link_file(vp
->insert
,
526 vp
->insert
->vm_file
->f_mapping
);
531 anon_vma_lock_write(vp
->anon_vma
);
532 anon_vma_interval_tree_pre_update_vma(vp
->vma
);
534 anon_vma_interval_tree_pre_update_vma(vp
->adj_next
);
538 flush_dcache_mmap_lock(vp
->mapping
);
539 vma_interval_tree_remove(vp
->vma
, &vp
->mapping
->i_mmap
);
541 vma_interval_tree_remove(vp
->adj_next
,
542 &vp
->mapping
->i_mmap
);
548 * vma_complete- Helper function for handling the unlocking after altering VMAs,
549 * or for inserting a VMA.
551 * @vp: The vma_prepare struct
552 * @vmi: The vma iterator
555 static inline void vma_complete(struct vma_prepare
*vp
,
556 struct vma_iterator
*vmi
, struct mm_struct
*mm
)
560 vma_interval_tree_insert(vp
->adj_next
,
561 &vp
->mapping
->i_mmap
);
562 vma_interval_tree_insert(vp
->vma
, &vp
->mapping
->i_mmap
);
563 flush_dcache_mmap_unlock(vp
->mapping
);
566 if (vp
->remove
&& vp
->file
) {
567 __remove_shared_vm_struct(vp
->remove
, vp
->mapping
);
569 __remove_shared_vm_struct(vp
->remove2
, vp
->mapping
);
570 } else if (vp
->insert
) {
572 * split_vma has split insert from vma, and needs
573 * us to insert it before dropping the locks
574 * (it may either follow vma or precede it).
576 vma_iter_store(vmi
, vp
->insert
);
581 anon_vma_interval_tree_post_update_vma(vp
->vma
);
583 anon_vma_interval_tree_post_update_vma(vp
->adj_next
);
584 anon_vma_unlock_write(vp
->anon_vma
);
588 i_mmap_unlock_write(vp
->mapping
);
589 uprobe_mmap(vp
->vma
);
592 uprobe_mmap(vp
->adj_next
);
597 vma_mark_detached(vp
->remove
, true);
599 uprobe_munmap(vp
->remove
, vp
->remove
->vm_start
,
603 if (vp
->remove
->anon_vma
)
604 anon_vma_merge(vp
->vma
, vp
->remove
);
606 mpol_put(vma_policy(vp
->remove
));
608 WARN_ON_ONCE(vp
->vma
->vm_end
< vp
->remove
->vm_end
);
609 vm_area_free(vp
->remove
);
612 * In mprotect's case 6 (see comments on vma_merge),
613 * we are removing both mid and next vmas
616 vp
->remove
= vp
->remove2
;
621 if (vp
->insert
&& vp
->file
)
622 uprobe_mmap(vp
->insert
);
627 * dup_anon_vma() - Helper function to duplicate anon_vma
628 * @dst: The destination VMA
629 * @src: The source VMA
630 * @dup: Pointer to the destination VMA when successful.
632 * Returns: 0 on success.
634 static inline int dup_anon_vma(struct vm_area_struct
*dst
,
635 struct vm_area_struct
*src
, struct vm_area_struct
**dup
)
638 * Easily overlooked: when mprotect shifts the boundary, make sure the
639 * expanding vma has anon_vma set if the shrinking vma had, to cover any
640 * anon pages imported.
642 if (src
->anon_vma
&& !dst
->anon_vma
) {
645 vma_assert_write_locked(dst
);
646 dst
->anon_vma
= src
->anon_vma
;
647 ret
= anon_vma_clone(dst
, src
);
658 * vma_expand - Expand an existing VMA
660 * @vmi: The vma iterator
661 * @vma: The vma to expand
662 * @start: The start of the vma
663 * @end: The exclusive end of the vma
664 * @pgoff: The page offset of vma
665 * @next: The current of next vma.
667 * Expand @vma to @start and @end. Can expand off the start and end. Will
668 * expand over @next if it's different from @vma and @end == @next->vm_end.
669 * Checking if the @vma can expand and merge with @next needs to be handled by
672 * Returns: 0 on success
674 int vma_expand(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
675 unsigned long start
, unsigned long end
, pgoff_t pgoff
,
676 struct vm_area_struct
*next
)
678 struct vm_area_struct
*anon_dup
= NULL
;
679 bool remove_next
= false;
680 struct vma_prepare vp
;
682 vma_start_write(vma
);
683 if (next
&& (vma
!= next
) && (end
== next
->vm_end
)) {
687 vma_start_write(next
);
688 ret
= dup_anon_vma(vma
, next
, &anon_dup
);
693 init_multi_vma_prep(&vp
, vma
, NULL
, remove_next
? next
: NULL
, NULL
);
694 /* Not merging but overwriting any part of next is not handled. */
695 VM_WARN_ON(next
&& !vp
.remove
&&
696 next
!= vma
&& end
> next
->vm_start
);
697 /* Only handles expanding */
698 VM_WARN_ON(vma
->vm_start
< start
|| vma
->vm_end
> end
);
700 /* Note: vma iterator must be pointing to 'start' */
701 vma_iter_config(vmi
, start
, end
);
702 if (vma_iter_prealloc(vmi
, vma
))
706 vma_adjust_trans_huge(vma
, start
, end
, 0);
707 vma_set_range(vma
, start
, end
, pgoff
);
708 vma_iter_store(vmi
, vma
);
710 vma_complete(&vp
, vmi
, vma
->vm_mm
);
715 unlink_anon_vmas(anon_dup
);
720 * vma_shrink() - Reduce an existing VMAs memory area
721 * @vmi: The vma iterator
722 * @vma: The VMA to modify
723 * @start: The new start
726 * Returns: 0 on success, -ENOMEM otherwise
728 int vma_shrink(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
729 unsigned long start
, unsigned long end
, pgoff_t pgoff
)
731 struct vma_prepare vp
;
733 WARN_ON((vma
->vm_start
!= start
) && (vma
->vm_end
!= end
));
735 if (vma
->vm_start
< start
)
736 vma_iter_config(vmi
, vma
->vm_start
, start
);
738 vma_iter_config(vmi
, end
, vma
->vm_end
);
740 if (vma_iter_prealloc(vmi
, NULL
))
743 vma_start_write(vma
);
745 init_vma_prep(&vp
, vma
);
747 vma_adjust_trans_huge(vma
, start
, end
, 0);
750 vma_set_range(vma
, start
, end
, pgoff
);
751 vma_complete(&vp
, vmi
, vma
->vm_mm
);
756 * If the vma has a ->close operation then the driver probably needs to release
757 * per-vma resources, so we don't attempt to merge those if the caller indicates
758 * the current vma may be removed as part of the merge.
760 static inline bool is_mergeable_vma(struct vm_area_struct
*vma
,
761 struct file
*file
, unsigned long vm_flags
,
762 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
763 struct anon_vma_name
*anon_name
, bool may_remove_vma
)
766 * VM_SOFTDIRTY should not prevent from VMA merging, if we
767 * match the flags but dirty bit -- the caller should mark
768 * merged VMA as dirty. If dirty bit won't be excluded from
769 * comparison, we increase pressure on the memory system forcing
770 * the kernel to generate new VMAs when old one could be
773 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
775 if (vma
->vm_file
!= file
)
777 if (may_remove_vma
&& vma
->vm_ops
&& vma
->vm_ops
->close
)
779 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
781 if (!anon_vma_name_eq(anon_vma_name(vma
), anon_name
))
786 static inline bool is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
787 struct anon_vma
*anon_vma2
, struct vm_area_struct
*vma
)
790 * The list_is_singular() test is to avoid merging VMA cloned from
791 * parents. This can improve scalability caused by anon_vma lock.
793 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
794 list_is_singular(&vma
->anon_vma_chain
)))
796 return anon_vma1
== anon_vma2
;
800 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
801 * in front of (at a lower virtual address and file offset than) the vma.
803 * We cannot merge two vmas if they have differently assigned (non-NULL)
804 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
806 * We don't check here for the merged mmap wrapping around the end of pagecache
807 * indices (16TB on ia32) because do_mmap() does not permit mmap's which
808 * wrap, nor mmaps which cover the final page at index -1UL.
810 * We assume the vma may be removed as part of the merge.
813 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
814 struct anon_vma
*anon_vma
, struct file
*file
,
815 pgoff_t vm_pgoff
, struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
816 struct anon_vma_name
*anon_name
)
818 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
, true) &&
819 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
820 if (vma
->vm_pgoff
== vm_pgoff
)
827 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
828 * beyond (at a higher virtual address and file offset than) the vma.
830 * We cannot merge two vmas if they have differently assigned (non-NULL)
831 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
833 * We assume that vma is not removed as part of the merge.
836 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
837 struct anon_vma
*anon_vma
, struct file
*file
,
838 pgoff_t vm_pgoff
, struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
839 struct anon_vma_name
*anon_name
)
841 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
, false) &&
842 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
844 vm_pglen
= vma_pages(vma
);
845 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
852 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
853 * figure out whether that can be merged with its predecessor or its
854 * successor. Or both (it neatly fills a hole).
856 * In most cases - when called for mmap, brk or mremap - [addr,end) is
857 * certain not to be mapped by the time vma_merge is called; but when
858 * called for mprotect, it is certain to be already mapped (either at
859 * an offset within prev, or at the start of next), and the flags of
860 * this area are about to be changed to vm_flags - and the no-change
861 * case has already been eliminated.
863 * The following mprotect cases have to be considered, where **** is
864 * the area passed down from mprotect_fixup, never extending beyond one
865 * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts
866 * at the same address as **** and is of the same or larger span, and
867 * NNNN the next vma after ****:
870 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC
871 * cannot merge might become might become
872 * PPNNNNNNNNNN PPPPPPPPPPCC
873 * mmap, brk or case 4 below case 5 below
876 * PPPP NNNN PPPPCCCCNNNN
877 * might become might become
878 * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or
879 * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or
880 * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8
882 * It is important for case 8 that the vma CCCC overlapping the
883 * region **** is never going to extended over NNNN. Instead NNNN must
884 * be extended in region **** and CCCC must be removed. This way in
885 * all cases where vma_merge succeeds, the moment vma_merge drops the
886 * rmap_locks, the properties of the merged vma will be already
887 * correct for the whole merged range. Some of those properties like
888 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
889 * be correct for the whole merged range immediately after the
890 * rmap_locks are released. Otherwise if NNNN would be removed and
891 * CCCC would be extended over the NNNN range, remove_migration_ptes
892 * or other rmap walkers (if working on addresses beyond the "end"
893 * parameter) may establish ptes with the wrong permissions of CCCC
894 * instead of the right permissions of NNNN.
897 * PPPP is represented by *prev
898 * CCCC is represented by *curr or not represented at all (NULL)
899 * NNNN is represented by *next or not represented at all (NULL)
900 * **** is not represented - it will be merged and the vma containing the
901 * area is returned, or the function will return NULL
903 static struct vm_area_struct
904 *vma_merge(struct vma_iterator
*vmi
, struct vm_area_struct
*prev
,
905 struct vm_area_struct
*src
, unsigned long addr
, unsigned long end
,
906 unsigned long vm_flags
, pgoff_t pgoff
, struct mempolicy
*policy
,
907 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
908 struct anon_vma_name
*anon_name
)
910 struct mm_struct
*mm
= src
->vm_mm
;
911 struct anon_vma
*anon_vma
= src
->anon_vma
;
912 struct file
*file
= src
->vm_file
;
913 struct vm_area_struct
*curr
, *next
, *res
;
914 struct vm_area_struct
*vma
, *adjust
, *remove
, *remove2
;
915 struct vm_area_struct
*anon_dup
= NULL
;
916 struct vma_prepare vp
;
919 bool merge_prev
= false;
920 bool merge_next
= false;
921 bool vma_expanded
= false;
922 unsigned long vma_start
= addr
;
923 unsigned long vma_end
= end
;
924 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
928 * We later require that vma->vm_flags == vm_flags,
929 * so this tests vma->vm_flags & VM_SPECIAL, too.
931 if (vm_flags
& VM_SPECIAL
)
934 /* Does the input range span an existing VMA? (cases 5 - 8) */
935 curr
= find_vma_intersection(mm
, prev
? prev
->vm_end
: 0, end
);
937 if (!curr
|| /* cases 1 - 4 */
938 end
== curr
->vm_end
) /* cases 6 - 8, adjacent VMA */
939 next
= vma_lookup(mm
, end
);
941 next
= NULL
; /* case 5 */
944 vma_start
= prev
->vm_start
;
945 vma_pgoff
= prev
->vm_pgoff
;
947 /* Can we merge the predecessor? */
948 if (addr
== prev
->vm_end
&& mpol_equal(vma_policy(prev
), policy
)
949 && can_vma_merge_after(prev
, vm_flags
, anon_vma
, file
,
950 pgoff
, vm_userfaultfd_ctx
, anon_name
)) {
956 /* Can we merge the successor? */
957 if (next
&& mpol_equal(policy
, vma_policy(next
)) &&
958 can_vma_merge_before(next
, vm_flags
, anon_vma
, file
, pgoff
+pglen
,
959 vm_userfaultfd_ctx
, anon_name
)) {
963 /* Verify some invariant that must be enforced by the caller. */
964 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
965 VM_WARN_ON(curr
&& (addr
!= curr
->vm_start
|| end
> curr
->vm_end
));
966 VM_WARN_ON(addr
>= end
);
968 if (!merge_prev
&& !merge_next
)
969 return NULL
; /* Not mergeable. */
972 vma_start_write(prev
);
975 remove
= remove2
= adjust
= NULL
;
977 /* Can we merge both the predecessor and the successor? */
978 if (merge_prev
&& merge_next
&&
979 is_mergeable_anon_vma(prev
->anon_vma
, next
->anon_vma
, NULL
)) {
980 vma_start_write(next
);
981 remove
= next
; /* case 1 */
982 vma_end
= next
->vm_end
;
983 err
= dup_anon_vma(prev
, next
, &anon_dup
);
984 if (curr
) { /* case 6 */
985 vma_start_write(curr
);
989 * Note that the dup_anon_vma below cannot overwrite err
990 * since the first caller would do nothing unless next
994 err
= dup_anon_vma(prev
, curr
, &anon_dup
);
996 } else if (merge_prev
) { /* case 2 */
998 vma_start_write(curr
);
999 if (end
== curr
->vm_end
) { /* case 7 */
1001 * can_vma_merge_after() assumed we would not be
1002 * removing prev vma, so it skipped the check
1003 * for vm_ops->close, but we are removing curr
1005 if (curr
->vm_ops
&& curr
->vm_ops
->close
)
1008 } else { /* case 5 */
1010 adj_start
= (end
- curr
->vm_start
);
1013 err
= dup_anon_vma(prev
, curr
, &anon_dup
);
1015 } else { /* merge_next */
1016 vma_start_write(next
);
1018 if (prev
&& addr
< prev
->vm_end
) { /* case 4 */
1019 vma_start_write(prev
);
1022 adj_start
= -(prev
->vm_end
- addr
);
1023 err
= dup_anon_vma(next
, prev
, &anon_dup
);
1026 * Note that cases 3 and 8 are the ONLY ones where prev
1027 * is permitted to be (but is not necessarily) NULL.
1029 vma
= next
; /* case 3 */
1031 vma_end
= next
->vm_end
;
1032 vma_pgoff
= next
->vm_pgoff
- pglen
;
1033 if (curr
) { /* case 8 */
1034 vma_pgoff
= curr
->vm_pgoff
;
1035 vma_start_write(curr
);
1037 err
= dup_anon_vma(next
, curr
, &anon_dup
);
1042 /* Error in anon_vma clone. */
1046 if (vma_start
< vma
->vm_start
|| vma_end
> vma
->vm_end
)
1047 vma_expanded
= true;
1050 vma_iter_config(vmi
, vma_start
, vma_end
);
1052 vma_iter_config(vmi
, adjust
->vm_start
+ adj_start
,
1056 if (vma_iter_prealloc(vmi
, vma
))
1059 init_multi_vma_prep(&vp
, vma
, adjust
, remove
, remove2
);
1060 VM_WARN_ON(vp
.anon_vma
&& adjust
&& adjust
->anon_vma
&&
1061 vp
.anon_vma
!= adjust
->anon_vma
);
1064 vma_adjust_trans_huge(vma
, vma_start
, vma_end
, adj_start
);
1065 vma_set_range(vma
, vma_start
, vma_end
, vma_pgoff
);
1068 vma_iter_store(vmi
, vma
);
1071 adjust
->vm_start
+= adj_start
;
1072 adjust
->vm_pgoff
+= adj_start
>> PAGE_SHIFT
;
1073 if (adj_start
< 0) {
1074 WARN_ON(vma_expanded
);
1075 vma_iter_store(vmi
, next
);
1079 vma_complete(&vp
, vmi
, mm
);
1080 khugepaged_enter_vma(res
, vm_flags
);
1085 unlink_anon_vmas(anon_dup
);
1088 vma_iter_set(vmi
, addr
);
1094 * Rough compatibility check to quickly see if it's even worth looking
1095 * at sharing an anon_vma.
1097 * They need to have the same vm_file, and the flags can only differ
1098 * in things that mprotect may change.
1100 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1101 * we can merge the two vma's. For example, we refuse to merge a vma if
1102 * there is a vm_ops->close() function, because that indicates that the
1103 * driver is doing some kind of reference counting. But that doesn't
1104 * really matter for the anon_vma sharing case.
1106 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1108 return a
->vm_end
== b
->vm_start
&&
1109 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1110 a
->vm_file
== b
->vm_file
&&
1111 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_ACCESS_FLAGS
| VM_SOFTDIRTY
)) &&
1112 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1116 * Do some basic sanity checking to see if we can re-use the anon_vma
1117 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1118 * the same as 'old', the other will be the new one that is trying
1119 * to share the anon_vma.
1121 * NOTE! This runs with mmap_lock held for reading, so it is possible that
1122 * the anon_vma of 'old' is concurrently in the process of being set up
1123 * by another page fault trying to merge _that_. But that's ok: if it
1124 * is being set up, that automatically means that it will be a singleton
1125 * acceptable for merging, so we can do all of this optimistically. But
1126 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1128 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1129 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1130 * is to return an anon_vma that is "complex" due to having gone through
1133 * We also make sure that the two vma's are compatible (adjacent,
1134 * and with the same memory policies). That's all stable, even with just
1135 * a read lock on the mmap_lock.
1137 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1139 if (anon_vma_compatible(a
, b
)) {
1140 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1142 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1149 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1150 * neighbouring vmas for a suitable anon_vma, before it goes off
1151 * to allocate a new anon_vma. It checks because a repetitive
1152 * sequence of mprotects and faults may otherwise lead to distinct
1153 * anon_vmas being allocated, preventing vma merge in subsequent
1156 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1158 struct anon_vma
*anon_vma
= NULL
;
1159 struct vm_area_struct
*prev
, *next
;
1160 VMA_ITERATOR(vmi
, vma
->vm_mm
, vma
->vm_end
);
1162 /* Try next first. */
1163 next
= vma_iter_load(&vmi
);
1165 anon_vma
= reusable_anon_vma(next
, vma
, next
);
1170 prev
= vma_prev(&vmi
);
1171 VM_BUG_ON_VMA(prev
!= vma
, vma
);
1172 prev
= vma_prev(&vmi
);
1173 /* Try prev next. */
1175 anon_vma
= reusable_anon_vma(prev
, prev
, vma
);
1178 * We might reach here with anon_vma == NULL if we can't find
1179 * any reusable anon_vma.
1180 * There's no absolute need to look only at touching neighbours:
1181 * we could search further afield for "compatible" anon_vmas.
1182 * But it would probably just be a waste of time searching,
1183 * or lead to too many vmas hanging off the same anon_vma.
1184 * We're trying to allow mprotect remerging later on,
1185 * not trying to minimize memory used for anon_vmas.
1191 * If a hint addr is less than mmap_min_addr change hint to be as
1192 * low as possible but still greater than mmap_min_addr
1194 static inline unsigned long round_hint_to_min(unsigned long hint
)
1197 if (((void *)hint
!= NULL
) &&
1198 (hint
< mmap_min_addr
))
1199 return PAGE_ALIGN(mmap_min_addr
);
1203 bool mlock_future_ok(struct mm_struct
*mm
, unsigned long flags
,
1204 unsigned long bytes
)
1206 unsigned long locked_pages
, limit_pages
;
1208 if (!(flags
& VM_LOCKED
) || capable(CAP_IPC_LOCK
))
1211 locked_pages
= bytes
>> PAGE_SHIFT
;
1212 locked_pages
+= mm
->locked_vm
;
1214 limit_pages
= rlimit(RLIMIT_MEMLOCK
);
1215 limit_pages
>>= PAGE_SHIFT
;
1217 return locked_pages
<= limit_pages
;
1220 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1222 if (S_ISREG(inode
->i_mode
))
1223 return MAX_LFS_FILESIZE
;
1225 if (S_ISBLK(inode
->i_mode
))
1226 return MAX_LFS_FILESIZE
;
1228 if (S_ISSOCK(inode
->i_mode
))
1229 return MAX_LFS_FILESIZE
;
1231 /* Special "we do even unsigned file positions" case */
1232 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1235 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1239 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1240 unsigned long pgoff
, unsigned long len
)
1242 u64 maxsize
= file_mmap_size_max(file
, inode
);
1244 if (maxsize
&& len
> maxsize
)
1247 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1253 * The caller must write-lock current->mm->mmap_lock.
1255 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1256 unsigned long len
, unsigned long prot
,
1257 unsigned long flags
, vm_flags_t vm_flags
,
1258 unsigned long pgoff
, unsigned long *populate
,
1259 struct list_head
*uf
)
1261 struct mm_struct
*mm
= current
->mm
;
1270 * Does the application expect PROT_READ to imply PROT_EXEC?
1272 * (the exception is when the underlying filesystem is noexec
1273 * mounted, in which case we don't add PROT_EXEC.)
1275 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1276 if (!(file
&& path_noexec(&file
->f_path
)))
1279 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1280 if (flags
& MAP_FIXED_NOREPLACE
)
1283 if (!(flags
& MAP_FIXED
))
1284 addr
= round_hint_to_min(addr
);
1286 /* Careful about overflows.. */
1287 len
= PAGE_ALIGN(len
);
1291 /* offset overflow? */
1292 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1295 /* Too many mappings? */
1296 if (mm
->map_count
> sysctl_max_map_count
)
1300 * addr is returned from get_unmapped_area,
1301 * There are two cases:
1302 * 1> MAP_FIXED == false
1303 * unallocated memory, no need to check sealing.
1304 * 1> MAP_FIXED == true
1305 * sealing is checked inside mmap_region when
1306 * do_vmi_munmap is called.
1309 if (prot
== PROT_EXEC
) {
1310 pkey
= execute_only_pkey(mm
);
1315 /* Do simple checking here so the lower-level routines won't have
1316 * to. we assume access permissions have been handled by the open
1317 * of the memory object, so we don't do any here.
1319 vm_flags
|= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1320 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1322 /* Obtain the address to map to. we verify (or select) it and ensure
1323 * that it represents a valid section of the address space.
1325 addr
= __get_unmapped_area(file
, addr
, len
, pgoff
, flags
, vm_flags
);
1326 if (IS_ERR_VALUE(addr
))
1329 if (flags
& MAP_FIXED_NOREPLACE
) {
1330 if (find_vma_intersection(mm
, addr
, addr
+ len
))
1334 if (flags
& MAP_LOCKED
)
1335 if (!can_do_mlock())
1338 if (!mlock_future_ok(mm
, vm_flags
, len
))
1342 struct inode
*inode
= file_inode(file
);
1343 unsigned long flags_mask
;
1345 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1348 flags_mask
= LEGACY_MAP_MASK
;
1349 if (file
->f_op
->fop_flags
& FOP_MMAP_SYNC
)
1350 flags_mask
|= MAP_SYNC
;
1352 switch (flags
& MAP_TYPE
) {
1355 * Force use of MAP_SHARED_VALIDATE with non-legacy
1356 * flags. E.g. MAP_SYNC is dangerous to use with
1357 * MAP_SHARED as you don't know which consistency model
1358 * you will get. We silently ignore unsupported flags
1359 * with MAP_SHARED to preserve backward compatibility.
1361 flags
&= LEGACY_MAP_MASK
;
1363 case MAP_SHARED_VALIDATE
:
1364 if (flags
& ~flags_mask
)
1366 if (prot
& PROT_WRITE
) {
1367 if (!(file
->f_mode
& FMODE_WRITE
))
1369 if (IS_SWAPFILE(file
->f_mapping
->host
))
1374 * Make sure we don't allow writing to an append-only
1377 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1380 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1381 if (!(file
->f_mode
& FMODE_WRITE
))
1382 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1385 if (!(file
->f_mode
& FMODE_READ
))
1387 if (path_noexec(&file
->f_path
)) {
1388 if (vm_flags
& VM_EXEC
)
1390 vm_flags
&= ~VM_MAYEXEC
;
1393 if (!file
->f_op
->mmap
)
1395 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1403 switch (flags
& MAP_TYPE
) {
1405 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1411 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1414 if (VM_DROPPABLE
== VM_NONE
)
1417 * A locked or stack area makes no sense to be droppable.
1419 * Also, since droppable pages can just go away at any time
1420 * it makes no sense to copy them on fork or dump them.
1422 * And don't attempt to combine with hugetlb for now.
1424 if (flags
& (MAP_LOCKED
| MAP_HUGETLB
))
1426 if (vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
))
1429 vm_flags
|= VM_DROPPABLE
;
1432 * If the pages can be dropped, then it doesn't make
1433 * sense to reserve them.
1435 vm_flags
|= VM_NORESERVE
;
1438 * Likewise, they're volatile enough that they
1439 * shouldn't survive forks or coredumps.
1441 vm_flags
|= VM_WIPEONFORK
| VM_DONTDUMP
;
1445 * Set pgoff according to addr for anon_vma.
1447 pgoff
= addr
>> PAGE_SHIFT
;
1455 * Set 'VM_NORESERVE' if we should not account for the
1456 * memory use of this mapping.
1458 if (flags
& MAP_NORESERVE
) {
1459 /* We honor MAP_NORESERVE if allowed to overcommit */
1460 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1461 vm_flags
|= VM_NORESERVE
;
1463 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1464 if (file
&& is_file_hugepages(file
))
1465 vm_flags
|= VM_NORESERVE
;
1468 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1469 if (!IS_ERR_VALUE(addr
) &&
1470 ((vm_flags
& VM_LOCKED
) ||
1471 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1476 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1477 unsigned long prot
, unsigned long flags
,
1478 unsigned long fd
, unsigned long pgoff
)
1480 struct file
*file
= NULL
;
1481 unsigned long retval
;
1483 if (!(flags
& MAP_ANONYMOUS
)) {
1484 audit_mmap_fd(fd
, flags
);
1488 if (is_file_hugepages(file
)) {
1489 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1490 } else if (unlikely(flags
& MAP_HUGETLB
)) {
1494 } else if (flags
& MAP_HUGETLB
) {
1497 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1501 len
= ALIGN(len
, huge_page_size(hs
));
1503 * VM_NORESERVE is used because the reservations will be
1504 * taken when vm_ops->mmap() is called
1506 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1508 HUGETLB_ANONHUGE_INODE
,
1509 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1511 return PTR_ERR(file
);
1514 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1521 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1522 unsigned long, prot
, unsigned long, flags
,
1523 unsigned long, fd
, unsigned long, pgoff
)
1525 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1528 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1529 struct mmap_arg_struct
{
1533 unsigned long flags
;
1535 unsigned long offset
;
1538 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1540 struct mmap_arg_struct a
;
1542 if (copy_from_user(&a
, arg
, sizeof(a
)))
1544 if (offset_in_page(a
.offset
))
1547 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1548 a
.offset
>> PAGE_SHIFT
);
1550 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1552 static bool vm_ops_needs_writenotify(const struct vm_operations_struct
*vm_ops
)
1554 return vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
);
1557 static bool vma_is_shared_writable(struct vm_area_struct
*vma
)
1559 return (vma
->vm_flags
& (VM_WRITE
| VM_SHARED
)) ==
1560 (VM_WRITE
| VM_SHARED
);
1563 static bool vma_fs_can_writeback(struct vm_area_struct
*vma
)
1565 /* No managed pages to writeback. */
1566 if (vma
->vm_flags
& VM_PFNMAP
)
1569 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1570 mapping_can_writeback(vma
->vm_file
->f_mapping
);
1574 * Does this VMA require the underlying folios to have their dirty state
1577 bool vma_needs_dirty_tracking(struct vm_area_struct
*vma
)
1579 /* Only shared, writable VMAs require dirty tracking. */
1580 if (!vma_is_shared_writable(vma
))
1583 /* Does the filesystem need to be notified? */
1584 if (vm_ops_needs_writenotify(vma
->vm_ops
))
1588 * Even if the filesystem doesn't indicate a need for writenotify, if it
1589 * can writeback, dirty tracking is still required.
1591 return vma_fs_can_writeback(vma
);
1595 * Some shared mappings will want the pages marked read-only
1596 * to track write events. If so, we'll downgrade vm_page_prot
1597 * to the private version (using protection_map[] without the
1600 bool vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1602 /* If it was private or non-writable, the write bit is already clear */
1603 if (!vma_is_shared_writable(vma
))
1606 /* The backer wishes to know when pages are first written to? */
1607 if (vm_ops_needs_writenotify(vma
->vm_ops
))
1610 /* The open routine did something to the protections that pgprot_modify
1611 * won't preserve? */
1612 if (pgprot_val(vm_page_prot
) !=
1613 pgprot_val(vm_pgprot_modify(vm_page_prot
, vma
->vm_flags
)))
1617 * Do we need to track softdirty? hugetlb does not support softdirty
1620 if (vma_soft_dirty_enabled(vma
) && !is_vm_hugetlb_page(vma
))
1623 /* Do we need write faults for uffd-wp tracking? */
1624 if (userfaultfd_wp(vma
))
1627 /* Can the mapping track the dirty pages? */
1628 return vma_fs_can_writeback(vma
);
1632 * We account for memory if it's a private writeable mapping,
1633 * not hugepages and VM_NORESERVE wasn't set.
1635 static inline bool accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1638 * hugetlb has its own accounting separate from the core VM
1639 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1641 if (file
&& is_file_hugepages(file
))
1644 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1648 * unmapped_area() - Find an area between the low_limit and the high_limit with
1649 * the correct alignment and offset, all from @info. Note: current->mm is used
1652 * @info: The unmapped area information including the range [low_limit -
1653 * high_limit), the alignment offset and mask.
1655 * Return: A memory address or -ENOMEM.
1657 static unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1659 unsigned long length
, gap
;
1660 unsigned long low_limit
, high_limit
;
1661 struct vm_area_struct
*tmp
;
1662 VMA_ITERATOR(vmi
, current
->mm
, 0);
1664 /* Adjust search length to account for worst case alignment overhead */
1665 length
= info
->length
+ info
->align_mask
+ info
->start_gap
;
1666 if (length
< info
->length
)
1669 low_limit
= info
->low_limit
;
1670 if (low_limit
< mmap_min_addr
)
1671 low_limit
= mmap_min_addr
;
1672 high_limit
= info
->high_limit
;
1674 if (vma_iter_area_lowest(&vmi
, low_limit
, high_limit
, length
))
1678 * Adjust for the gap first so it doesn't interfere with the
1679 * later alignment. The first step is the minimum needed to
1680 * fulill the start gap, the next steps is the minimum to align
1681 * that. It is the minimum needed to fulill both.
1683 gap
= vma_iter_addr(&vmi
) + info
->start_gap
;
1684 gap
+= (info
->align_offset
- gap
) & info
->align_mask
;
1685 tmp
= vma_next(&vmi
);
1686 if (tmp
&& (tmp
->vm_flags
& VM_STARTGAP_FLAGS
)) { /* Avoid prev check if possible */
1687 if (vm_start_gap(tmp
) < gap
+ length
- 1) {
1688 low_limit
= tmp
->vm_end
;
1689 vma_iter_reset(&vmi
);
1693 tmp
= vma_prev(&vmi
);
1694 if (tmp
&& vm_end_gap(tmp
) > gap
) {
1695 low_limit
= vm_end_gap(tmp
);
1696 vma_iter_reset(&vmi
);
1705 * unmapped_area_topdown() - Find an area between the low_limit and the
1706 * high_limit with the correct alignment and offset at the highest available
1707 * address, all from @info. Note: current->mm is used for the search.
1709 * @info: The unmapped area information including the range [low_limit -
1710 * high_limit), the alignment offset and mask.
1712 * Return: A memory address or -ENOMEM.
1714 static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1716 unsigned long length
, gap
, gap_end
;
1717 unsigned long low_limit
, high_limit
;
1718 struct vm_area_struct
*tmp
;
1719 VMA_ITERATOR(vmi
, current
->mm
, 0);
1721 /* Adjust search length to account for worst case alignment overhead */
1722 length
= info
->length
+ info
->align_mask
+ info
->start_gap
;
1723 if (length
< info
->length
)
1726 low_limit
= info
->low_limit
;
1727 if (low_limit
< mmap_min_addr
)
1728 low_limit
= mmap_min_addr
;
1729 high_limit
= info
->high_limit
;
1731 if (vma_iter_area_highest(&vmi
, low_limit
, high_limit
, length
))
1734 gap
= vma_iter_end(&vmi
) - info
->length
;
1735 gap
-= (gap
- info
->align_offset
) & info
->align_mask
;
1736 gap_end
= vma_iter_end(&vmi
);
1737 tmp
= vma_next(&vmi
);
1738 if (tmp
&& (tmp
->vm_flags
& VM_STARTGAP_FLAGS
)) { /* Avoid prev check if possible */
1739 if (vm_start_gap(tmp
) < gap_end
) {
1740 high_limit
= vm_start_gap(tmp
);
1741 vma_iter_reset(&vmi
);
1745 tmp
= vma_prev(&vmi
);
1746 if (tmp
&& vm_end_gap(tmp
) > gap
) {
1747 high_limit
= tmp
->vm_start
;
1748 vma_iter_reset(&vmi
);
1757 * Search for an unmapped address range.
1759 * We are looking for a range that:
1760 * - does not intersect with any VMA;
1761 * - is contained within the [low_limit, high_limit) interval;
1762 * - is at least the desired size.
1763 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
1765 unsigned long vm_unmapped_area(struct vm_unmapped_area_info
*info
)
1769 if (info
->flags
& VM_UNMAPPED_AREA_TOPDOWN
)
1770 addr
= unmapped_area_topdown(info
);
1772 addr
= unmapped_area(info
);
1774 trace_vm_unmapped_area(addr
, info
);
1778 /* Get an address range which is currently unmapped.
1779 * For shmat() with addr=0.
1781 * Ugly calling convention alert:
1782 * Return value with the low bits set means error value,
1784 * if (ret & ~PAGE_MASK)
1787 * This function "knows" that -ENOMEM has the bits set.
1790 generic_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1791 unsigned long len
, unsigned long pgoff
,
1792 unsigned long flags
)
1794 struct mm_struct
*mm
= current
->mm
;
1795 struct vm_area_struct
*vma
, *prev
;
1796 struct vm_unmapped_area_info info
= {};
1797 const unsigned long mmap_end
= arch_get_mmap_end(addr
, len
, flags
);
1799 if (len
> mmap_end
- mmap_min_addr
)
1802 if (flags
& MAP_FIXED
)
1806 addr
= PAGE_ALIGN(addr
);
1807 vma
= find_vma_prev(mm
, addr
, &prev
);
1808 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
1809 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1810 (!prev
|| addr
>= vm_end_gap(prev
)))
1815 info
.low_limit
= mm
->mmap_base
;
1816 info
.high_limit
= mmap_end
;
1817 return vm_unmapped_area(&info
);
1820 #ifndef HAVE_ARCH_UNMAPPED_AREA
1822 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1823 unsigned long len
, unsigned long pgoff
,
1824 unsigned long flags
)
1826 return generic_get_unmapped_area(filp
, addr
, len
, pgoff
, flags
);
1831 * This mmap-allocator allocates new areas top-down from below the
1832 * stack's low limit (the base):
1835 generic_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
1836 unsigned long len
, unsigned long pgoff
,
1837 unsigned long flags
)
1839 struct vm_area_struct
*vma
, *prev
;
1840 struct mm_struct
*mm
= current
->mm
;
1841 struct vm_unmapped_area_info info
= {};
1842 const unsigned long mmap_end
= arch_get_mmap_end(addr
, len
, flags
);
1844 /* requested length too big for entire address space */
1845 if (len
> mmap_end
- mmap_min_addr
)
1848 if (flags
& MAP_FIXED
)
1851 /* requesting a specific address */
1853 addr
= PAGE_ALIGN(addr
);
1854 vma
= find_vma_prev(mm
, addr
, &prev
);
1855 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
1856 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1857 (!prev
|| addr
>= vm_end_gap(prev
)))
1861 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1863 info
.low_limit
= PAGE_SIZE
;
1864 info
.high_limit
= arch_get_mmap_base(addr
, mm
->mmap_base
);
1865 addr
= vm_unmapped_area(&info
);
1868 * A failed mmap() very likely causes application failure,
1869 * so fall back to the bottom-up function here. This scenario
1870 * can happen with large stack limits and large mmap()
1873 if (offset_in_page(addr
)) {
1874 VM_BUG_ON(addr
!= -ENOMEM
);
1876 info
.low_limit
= TASK_UNMAPPED_BASE
;
1877 info
.high_limit
= mmap_end
;
1878 addr
= vm_unmapped_area(&info
);
1884 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1886 arch_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
1887 unsigned long len
, unsigned long pgoff
,
1888 unsigned long flags
)
1890 return generic_get_unmapped_area_topdown(filp
, addr
, len
, pgoff
, flags
);
1894 #ifndef HAVE_ARCH_UNMAPPED_AREA_VMFLAGS
1896 arch_get_unmapped_area_vmflags(struct file
*filp
, unsigned long addr
, unsigned long len
,
1897 unsigned long pgoff
, unsigned long flags
, vm_flags_t vm_flags
)
1899 return arch_get_unmapped_area(filp
, addr
, len
, pgoff
, flags
);
1903 arch_get_unmapped_area_topdown_vmflags(struct file
*filp
, unsigned long addr
,
1904 unsigned long len
, unsigned long pgoff
,
1905 unsigned long flags
, vm_flags_t vm_flags
)
1907 return arch_get_unmapped_area_topdown(filp
, addr
, len
, pgoff
, flags
);
1911 unsigned long mm_get_unmapped_area_vmflags(struct mm_struct
*mm
, struct file
*filp
,
1912 unsigned long addr
, unsigned long len
,
1913 unsigned long pgoff
, unsigned long flags
,
1914 vm_flags_t vm_flags
)
1916 if (test_bit(MMF_TOPDOWN
, &mm
->flags
))
1917 return arch_get_unmapped_area_topdown_vmflags(filp
, addr
, len
, pgoff
,
1919 return arch_get_unmapped_area_vmflags(filp
, addr
, len
, pgoff
, flags
, vm_flags
);
1923 __get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1924 unsigned long pgoff
, unsigned long flags
, vm_flags_t vm_flags
)
1926 unsigned long (*get_area
)(struct file
*, unsigned long,
1927 unsigned long, unsigned long, unsigned long)
1930 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1934 /* Careful about overflows.. */
1935 if (len
> TASK_SIZE
)
1939 if (file
->f_op
->get_unmapped_area
)
1940 get_area
= file
->f_op
->get_unmapped_area
;
1941 } else if (flags
& MAP_SHARED
) {
1943 * mmap_region() will call shmem_zero_setup() to create a file,
1944 * so use shmem's get_unmapped_area in case it can be huge.
1946 get_area
= shmem_get_unmapped_area
;
1949 /* Always treat pgoff as zero for anonymous memory. */
1954 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1955 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
1956 /* Ensures that larger anonymous mappings are THP aligned. */
1957 addr
= thp_get_unmapped_area_vmflags(file
, addr
, len
,
1958 pgoff
, flags
, vm_flags
);
1960 addr
= mm_get_unmapped_area_vmflags(current
->mm
, file
, addr
, len
,
1961 pgoff
, flags
, vm_flags
);
1963 if (IS_ERR_VALUE(addr
))
1966 if (addr
> TASK_SIZE
- len
)
1968 if (offset_in_page(addr
))
1971 error
= security_mmap_addr(addr
);
1972 return error
? error
: addr
;
1976 mm_get_unmapped_area(struct mm_struct
*mm
, struct file
*file
,
1977 unsigned long addr
, unsigned long len
,
1978 unsigned long pgoff
, unsigned long flags
)
1980 if (test_bit(MMF_TOPDOWN
, &mm
->flags
))
1981 return arch_get_unmapped_area_topdown(file
, addr
, len
, pgoff
, flags
);
1982 return arch_get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1984 EXPORT_SYMBOL(mm_get_unmapped_area
);
1987 * find_vma_intersection() - Look up the first VMA which intersects the interval
1988 * @mm: The process address space.
1989 * @start_addr: The inclusive start user address.
1990 * @end_addr: The exclusive end user address.
1992 * Returns: The first VMA within the provided range, %NULL otherwise. Assumes
1993 * start_addr < end_addr.
1995 struct vm_area_struct
*find_vma_intersection(struct mm_struct
*mm
,
1996 unsigned long start_addr
,
1997 unsigned long end_addr
)
1999 unsigned long index
= start_addr
;
2001 mmap_assert_locked(mm
);
2002 return mt_find(&mm
->mm_mt
, &index
, end_addr
- 1);
2004 EXPORT_SYMBOL(find_vma_intersection
);
2007 * find_vma() - Find the VMA for a given address, or the next VMA.
2008 * @mm: The mm_struct to check
2009 * @addr: The address
2011 * Returns: The VMA associated with addr, or the next VMA.
2012 * May return %NULL in the case of no VMA at addr or above.
2014 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
2016 unsigned long index
= addr
;
2018 mmap_assert_locked(mm
);
2019 return mt_find(&mm
->mm_mt
, &index
, ULONG_MAX
);
2021 EXPORT_SYMBOL(find_vma
);
2024 * find_vma_prev() - Find the VMA for a given address, or the next vma and
2025 * set %pprev to the previous VMA, if any.
2026 * @mm: The mm_struct to check
2027 * @addr: The address
2028 * @pprev: The pointer to set to the previous VMA
2030 * Note that RCU lock is missing here since the external mmap_lock() is used
2033 * Returns: The VMA associated with @addr, or the next vma.
2034 * May return %NULL in the case of no vma at addr or above.
2036 struct vm_area_struct
*
2037 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
2038 struct vm_area_struct
**pprev
)
2040 struct vm_area_struct
*vma
;
2041 VMA_ITERATOR(vmi
, mm
, addr
);
2043 vma
= vma_iter_load(&vmi
);
2044 *pprev
= vma_prev(&vmi
);
2046 vma
= vma_next(&vmi
);
2051 * Verify that the stack growth is acceptable and
2052 * update accounting. This is shared with both the
2053 * grow-up and grow-down cases.
2055 static int acct_stack_growth(struct vm_area_struct
*vma
,
2056 unsigned long size
, unsigned long grow
)
2058 struct mm_struct
*mm
= vma
->vm_mm
;
2059 unsigned long new_start
;
2061 /* address space limit tests */
2062 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
2065 /* Stack limit test */
2066 if (size
> rlimit(RLIMIT_STACK
))
2069 /* mlock limit tests */
2070 if (!mlock_future_ok(mm
, vma
->vm_flags
, grow
<< PAGE_SHIFT
))
2073 /* Check to ensure the stack will not grow into a hugetlb-only region */
2074 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
2076 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
2080 * Overcommit.. This must be the final test, as it will
2081 * update security statistics.
2083 if (security_vm_enough_memory_mm(mm
, grow
))
2089 #if defined(CONFIG_STACK_GROWSUP)
2091 * PA-RISC uses this for its stack.
2092 * vma is the last one with address > vma->vm_end. Have to extend vma.
2094 static int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
2096 struct mm_struct
*mm
= vma
->vm_mm
;
2097 struct vm_area_struct
*next
;
2098 unsigned long gap_addr
;
2100 VMA_ITERATOR(vmi
, mm
, vma
->vm_start
);
2102 if (!(vma
->vm_flags
& VM_GROWSUP
))
2105 /* Guard against exceeding limits of the address space. */
2106 address
&= PAGE_MASK
;
2107 if (address
>= (TASK_SIZE
& PAGE_MASK
))
2109 address
+= PAGE_SIZE
;
2111 /* Enforce stack_guard_gap */
2112 gap_addr
= address
+ stack_guard_gap
;
2114 /* Guard against overflow */
2115 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
2116 gap_addr
= TASK_SIZE
;
2118 next
= find_vma_intersection(mm
, vma
->vm_end
, gap_addr
);
2119 if (next
&& vma_is_accessible(next
)) {
2120 if (!(next
->vm_flags
& VM_GROWSUP
))
2122 /* Check that both stack segments have the same anon_vma? */
2126 vma_iter_prev_range_limit(&vmi
, address
);
2128 vma_iter_config(&vmi
, vma
->vm_start
, address
);
2129 if (vma_iter_prealloc(&vmi
, vma
))
2132 /* We must make sure the anon_vma is allocated. */
2133 if (unlikely(anon_vma_prepare(vma
))) {
2134 vma_iter_free(&vmi
);
2138 /* Lock the VMA before expanding to prevent concurrent page faults */
2139 vma_start_write(vma
);
2141 * vma->vm_start/vm_end cannot change under us because the caller
2142 * is required to hold the mmap_lock in read mode. We need the
2143 * anon_vma lock to serialize against concurrent expand_stacks.
2145 anon_vma_lock_write(vma
->anon_vma
);
2147 /* Somebody else might have raced and expanded it already */
2148 if (address
> vma
->vm_end
) {
2149 unsigned long size
, grow
;
2151 size
= address
- vma
->vm_start
;
2152 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
2155 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
2156 error
= acct_stack_growth(vma
, size
, grow
);
2159 * We only hold a shared mmap_lock lock here, so
2160 * we need to protect against concurrent vma
2161 * expansions. anon_vma_lock_write() doesn't
2162 * help here, as we don't guarantee that all
2163 * growable vmas in a mm share the same root
2164 * anon vma. So, we reuse mm->page_table_lock
2165 * to guard against concurrent vma expansions.
2167 spin_lock(&mm
->page_table_lock
);
2168 if (vma
->vm_flags
& VM_LOCKED
)
2169 mm
->locked_vm
+= grow
;
2170 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2171 anon_vma_interval_tree_pre_update_vma(vma
);
2172 vma
->vm_end
= address
;
2173 /* Overwrite old entry in mtree. */
2174 vma_iter_store(&vmi
, vma
);
2175 anon_vma_interval_tree_post_update_vma(vma
);
2176 spin_unlock(&mm
->page_table_lock
);
2178 perf_event_mmap(vma
);
2182 anon_vma_unlock_write(vma
->anon_vma
);
2183 vma_iter_free(&vmi
);
2187 #endif /* CONFIG_STACK_GROWSUP */
2190 * vma is the first one with address < vma->vm_start. Have to extend vma.
2191 * mmap_lock held for writing.
2193 int expand_downwards(struct vm_area_struct
*vma
, unsigned long address
)
2195 struct mm_struct
*mm
= vma
->vm_mm
;
2196 struct vm_area_struct
*prev
;
2198 VMA_ITERATOR(vmi
, mm
, vma
->vm_start
);
2200 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2203 address
&= PAGE_MASK
;
2204 if (address
< mmap_min_addr
|| address
< FIRST_USER_ADDRESS
)
2207 /* Enforce stack_guard_gap */
2208 prev
= vma_prev(&vmi
);
2209 /* Check that both stack segments have the same anon_vma? */
2211 if (!(prev
->vm_flags
& VM_GROWSDOWN
) &&
2212 vma_is_accessible(prev
) &&
2213 (address
- prev
->vm_end
< stack_guard_gap
))
2218 vma_iter_next_range_limit(&vmi
, vma
->vm_start
);
2220 vma_iter_config(&vmi
, address
, vma
->vm_end
);
2221 if (vma_iter_prealloc(&vmi
, vma
))
2224 /* We must make sure the anon_vma is allocated. */
2225 if (unlikely(anon_vma_prepare(vma
))) {
2226 vma_iter_free(&vmi
);
2230 /* Lock the VMA before expanding to prevent concurrent page faults */
2231 vma_start_write(vma
);
2233 * vma->vm_start/vm_end cannot change under us because the caller
2234 * is required to hold the mmap_lock in read mode. We need the
2235 * anon_vma lock to serialize against concurrent expand_stacks.
2237 anon_vma_lock_write(vma
->anon_vma
);
2239 /* Somebody else might have raced and expanded it already */
2240 if (address
< vma
->vm_start
) {
2241 unsigned long size
, grow
;
2243 size
= vma
->vm_end
- address
;
2244 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2247 if (grow
<= vma
->vm_pgoff
) {
2248 error
= acct_stack_growth(vma
, size
, grow
);
2251 * We only hold a shared mmap_lock lock here, so
2252 * we need to protect against concurrent vma
2253 * expansions. anon_vma_lock_write() doesn't
2254 * help here, as we don't guarantee that all
2255 * growable vmas in a mm share the same root
2256 * anon vma. So, we reuse mm->page_table_lock
2257 * to guard against concurrent vma expansions.
2259 spin_lock(&mm
->page_table_lock
);
2260 if (vma
->vm_flags
& VM_LOCKED
)
2261 mm
->locked_vm
+= grow
;
2262 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2263 anon_vma_interval_tree_pre_update_vma(vma
);
2264 vma
->vm_start
= address
;
2265 vma
->vm_pgoff
-= grow
;
2266 /* Overwrite old entry in mtree. */
2267 vma_iter_store(&vmi
, vma
);
2268 anon_vma_interval_tree_post_update_vma(vma
);
2269 spin_unlock(&mm
->page_table_lock
);
2271 perf_event_mmap(vma
);
2275 anon_vma_unlock_write(vma
->anon_vma
);
2276 vma_iter_free(&vmi
);
2281 /* enforced gap between the expanding stack and other mappings. */
2282 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2284 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2289 val
= simple_strtoul(p
, &endptr
, 10);
2291 stack_guard_gap
= val
<< PAGE_SHIFT
;
2295 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2297 #ifdef CONFIG_STACK_GROWSUP
2298 int expand_stack_locked(struct vm_area_struct
*vma
, unsigned long address
)
2300 return expand_upwards(vma
, address
);
2303 struct vm_area_struct
*find_extend_vma_locked(struct mm_struct
*mm
, unsigned long addr
)
2305 struct vm_area_struct
*vma
, *prev
;
2308 vma
= find_vma_prev(mm
, addr
, &prev
);
2309 if (vma
&& (vma
->vm_start
<= addr
))
2313 if (expand_stack_locked(prev
, addr
))
2315 if (prev
->vm_flags
& VM_LOCKED
)
2316 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2320 int expand_stack_locked(struct vm_area_struct
*vma
, unsigned long address
)
2322 return expand_downwards(vma
, address
);
2325 struct vm_area_struct
*find_extend_vma_locked(struct mm_struct
*mm
, unsigned long addr
)
2327 struct vm_area_struct
*vma
;
2328 unsigned long start
;
2331 vma
= find_vma(mm
, addr
);
2334 if (vma
->vm_start
<= addr
)
2336 start
= vma
->vm_start
;
2337 if (expand_stack_locked(vma
, addr
))
2339 if (vma
->vm_flags
& VM_LOCKED
)
2340 populate_vma_page_range(vma
, addr
, start
, NULL
);
2345 #if defined(CONFIG_STACK_GROWSUP)
2347 #define vma_expand_up(vma,addr) expand_upwards(vma, addr)
2348 #define vma_expand_down(vma, addr) (-EFAULT)
2352 #define vma_expand_up(vma,addr) (-EFAULT)
2353 #define vma_expand_down(vma, addr) expand_downwards(vma, addr)
2358 * expand_stack(): legacy interface for page faulting. Don't use unless
2361 * This is called with the mm locked for reading, drops the lock, takes
2362 * the lock for writing, tries to look up a vma again, expands it if
2363 * necessary, and downgrades the lock to reading again.
2365 * If no vma is found or it can't be expanded, it returns NULL and has
2368 struct vm_area_struct
*expand_stack(struct mm_struct
*mm
, unsigned long addr
)
2370 struct vm_area_struct
*vma
, *prev
;
2372 mmap_read_unlock(mm
);
2373 if (mmap_write_lock_killable(mm
))
2376 vma
= find_vma_prev(mm
, addr
, &prev
);
2377 if (vma
&& vma
->vm_start
<= addr
)
2380 if (prev
&& !vma_expand_up(prev
, addr
)) {
2385 if (vma
&& !vma_expand_down(vma
, addr
))
2388 mmap_write_unlock(mm
);
2392 mmap_write_downgrade(mm
);
2397 * Ok - we have the memory areas we should free on a maple tree so release them,
2398 * and do the vma updates.
2400 * Called with the mm semaphore held.
2402 static inline void remove_mt(struct mm_struct
*mm
, struct ma_state
*mas
)
2404 unsigned long nr_accounted
= 0;
2405 struct vm_area_struct
*vma
;
2407 /* Update high watermark before we lower total_vm */
2408 update_hiwater_vm(mm
);
2409 mas_for_each(mas
, vma
, ULONG_MAX
) {
2410 long nrpages
= vma_pages(vma
);
2412 if (vma
->vm_flags
& VM_ACCOUNT
)
2413 nr_accounted
+= nrpages
;
2414 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2415 remove_vma(vma
, false);
2417 vm_unacct_memory(nr_accounted
);
2421 * Get rid of page table information in the indicated region.
2423 * Called with the mm semaphore held.
2425 static void unmap_region(struct mm_struct
*mm
, struct ma_state
*mas
,
2426 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2427 struct vm_area_struct
*next
, unsigned long start
,
2428 unsigned long end
, unsigned long tree_end
, bool mm_wr_locked
)
2430 struct mmu_gather tlb
;
2431 unsigned long mt_start
= mas
->index
;
2434 tlb_gather_mmu(&tlb
, mm
);
2435 update_hiwater_rss(mm
);
2436 unmap_vmas(&tlb
, mas
, vma
, start
, end
, tree_end
, mm_wr_locked
);
2437 mas_set(mas
, mt_start
);
2438 free_pgtables(&tlb
, mas
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2439 next
? next
->vm_start
: USER_PGTABLES_CEILING
,
2441 tlb_finish_mmu(&tlb
);
2445 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2446 * has already been checked or doesn't make sense to fail.
2447 * VMA Iterator will point to the end VMA.
2449 static int __split_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2450 unsigned long addr
, int new_below
)
2452 struct vma_prepare vp
;
2453 struct vm_area_struct
*new;
2456 WARN_ON(vma
->vm_start
>= addr
);
2457 WARN_ON(vma
->vm_end
<= addr
);
2459 if (vma
->vm_ops
&& vma
->vm_ops
->may_split
) {
2460 err
= vma
->vm_ops
->may_split(vma
, addr
);
2465 new = vm_area_dup(vma
);
2472 new->vm_start
= addr
;
2473 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2477 vma_iter_config(vmi
, new->vm_start
, new->vm_end
);
2478 if (vma_iter_prealloc(vmi
, new))
2481 err
= vma_dup_policy(vma
, new);
2485 err
= anon_vma_clone(new, vma
);
2490 get_file(new->vm_file
);
2492 if (new->vm_ops
&& new->vm_ops
->open
)
2493 new->vm_ops
->open(new);
2495 vma_start_write(vma
);
2496 vma_start_write(new);
2498 init_vma_prep(&vp
, vma
);
2501 vma_adjust_trans_huge(vma
, vma
->vm_start
, addr
, 0);
2504 vma
->vm_start
= addr
;
2505 vma
->vm_pgoff
+= (addr
- new->vm_start
) >> PAGE_SHIFT
;
2510 /* vma_complete stores the new vma */
2511 vma_complete(&vp
, vmi
, vma
->vm_mm
);
2519 mpol_put(vma_policy(new));
2528 * Split a vma into two pieces at address 'addr', a new vma is allocated
2529 * either for the first part or the tail.
2531 static int split_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2532 unsigned long addr
, int new_below
)
2534 if (vma
->vm_mm
->map_count
>= sysctl_max_map_count
)
2537 return __split_vma(vmi
, vma
, addr
, new_below
);
2541 * We are about to modify one or multiple of a VMA's flags, policy, userfaultfd
2542 * context and anonymous VMA name within the range [start, end).
2544 * As a result, we might be able to merge the newly modified VMA range with an
2545 * adjacent VMA with identical properties.
2547 * If no merge is possible and the range does not span the entirety of the VMA,
2548 * we then need to split the VMA to accommodate the change.
2550 * The function returns either the merged VMA, the original VMA if a split was
2551 * required instead, or an error if the split failed.
2553 struct vm_area_struct
*vma_modify(struct vma_iterator
*vmi
,
2554 struct vm_area_struct
*prev
,
2555 struct vm_area_struct
*vma
,
2556 unsigned long start
, unsigned long end
,
2557 unsigned long vm_flags
,
2558 struct mempolicy
*policy
,
2559 struct vm_userfaultfd_ctx uffd_ctx
,
2560 struct anon_vma_name
*anon_name
)
2562 pgoff_t pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
2563 struct vm_area_struct
*merged
;
2565 merged
= vma_merge(vmi
, prev
, vma
, start
, end
, vm_flags
,
2566 pgoff
, policy
, uffd_ctx
, anon_name
);
2570 if (vma
->vm_start
< start
) {
2571 int err
= split_vma(vmi
, vma
, start
, 1);
2574 return ERR_PTR(err
);
2577 if (vma
->vm_end
> end
) {
2578 int err
= split_vma(vmi
, vma
, end
, 0);
2581 return ERR_PTR(err
);
2588 * Attempt to merge a newly mapped VMA with those adjacent to it. The caller
2589 * must ensure that [start, end) does not overlap any existing VMA.
2591 static struct vm_area_struct
2592 *vma_merge_new_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*prev
,
2593 struct vm_area_struct
*vma
, unsigned long start
,
2594 unsigned long end
, pgoff_t pgoff
)
2596 return vma_merge(vmi
, prev
, vma
, start
, end
, vma
->vm_flags
, pgoff
,
2597 vma_policy(vma
), vma
->vm_userfaultfd_ctx
, anon_vma_name(vma
));
2601 * Expand vma by delta bytes, potentially merging with an immediately adjacent
2602 * VMA with identical properties.
2604 struct vm_area_struct
*vma_merge_extend(struct vma_iterator
*vmi
,
2605 struct vm_area_struct
*vma
,
2606 unsigned long delta
)
2608 pgoff_t pgoff
= vma
->vm_pgoff
+ vma_pages(vma
);
2610 /* vma is specified as prev, so case 1 or 2 will apply. */
2611 return vma_merge(vmi
, vma
, vma
, vma
->vm_end
, vma
->vm_end
+ delta
,
2612 vma
->vm_flags
, pgoff
, vma_policy(vma
),
2613 vma
->vm_userfaultfd_ctx
, anon_vma_name(vma
));
2617 * do_vmi_align_munmap() - munmap the aligned region from @start to @end.
2618 * @vmi: The vma iterator
2619 * @vma: The starting vm_area_struct
2620 * @mm: The mm_struct
2621 * @start: The aligned start address to munmap.
2622 * @end: The aligned end address to munmap.
2623 * @uf: The userfaultfd list_head
2624 * @unlock: Set to true to drop the mmap_lock. unlocking only happens on
2627 * Return: 0 on success and drops the lock if so directed, error and leaves the
2628 * lock held otherwise.
2631 do_vmi_align_munmap(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2632 struct mm_struct
*mm
, unsigned long start
,
2633 unsigned long end
, struct list_head
*uf
, bool unlock
)
2635 struct vm_area_struct
*prev
, *next
= NULL
;
2636 struct maple_tree mt_detach
;
2638 int error
= -ENOMEM
;
2639 unsigned long locked_vm
= 0;
2640 MA_STATE(mas_detach
, &mt_detach
, 0, 0);
2641 mt_init_flags(&mt_detach
, vmi
->mas
.tree
->ma_flags
& MT_FLAGS_LOCK_MASK
);
2642 mt_on_stack(mt_detach
);
2645 * If we need to split any vma, do it now to save pain later.
2647 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2648 * unmapped vm_area_struct will remain in use: so lower split_vma
2649 * places tmp vma above, and higher split_vma places tmp vma below.
2652 /* Does it split the first one? */
2653 if (start
> vma
->vm_start
) {
2656 * Make sure that map_count on return from munmap() will
2657 * not exceed its limit; but let map_count go just above
2658 * its limit temporarily, to help free resources as expected.
2660 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2661 goto map_count_exceeded
;
2663 error
= __split_vma(vmi
, vma
, start
, 1);
2665 goto start_split_failed
;
2669 * Detach a range of VMAs from the mm. Using next as a temp variable as
2670 * it is always overwritten.
2674 /* Does it split the end? */
2675 if (next
->vm_end
> end
) {
2676 error
= __split_vma(vmi
, next
, end
, 0);
2678 goto end_split_failed
;
2680 vma_start_write(next
);
2681 mas_set(&mas_detach
, count
);
2682 error
= mas_store_gfp(&mas_detach
, next
, GFP_KERNEL
);
2684 goto munmap_gather_failed
;
2685 vma_mark_detached(next
, true);
2686 if (next
->vm_flags
& VM_LOCKED
)
2687 locked_vm
+= vma_pages(next
);
2692 * If userfaultfd_unmap_prep returns an error the vmas
2693 * will remain split, but userland will get a
2694 * highly unexpected error anyway. This is no
2695 * different than the case where the first of the two
2696 * __split_vma fails, but we don't undo the first
2697 * split, despite we could. This is unlikely enough
2698 * failure that it's not worth optimizing it for.
2700 error
= userfaultfd_unmap_prep(next
, start
, end
, uf
);
2703 goto userfaultfd_error
;
2705 #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
2706 BUG_ON(next
->vm_start
< start
);
2707 BUG_ON(next
->vm_start
> end
);
2709 } for_each_vma_range(*vmi
, next
, end
);
2711 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
2712 /* Make sure no VMAs are about to be lost. */
2714 MA_STATE(test
, &mt_detach
, 0, 0);
2715 struct vm_area_struct
*vma_mas
, *vma_test
;
2718 vma_iter_set(vmi
, start
);
2720 vma_test
= mas_find(&test
, count
- 1);
2721 for_each_vma_range(*vmi
, vma_mas
, end
) {
2722 BUG_ON(vma_mas
!= vma_test
);
2724 vma_test
= mas_next(&test
, count
- 1);
2727 BUG_ON(count
!= test_count
);
2731 while (vma_iter_addr(vmi
) > start
)
2732 vma_iter_prev_range(vmi
);
2734 error
= vma_iter_clear_gfp(vmi
, start
, end
, GFP_KERNEL
);
2736 goto clear_tree_failed
;
2738 /* Point of no return */
2739 mm
->locked_vm
-= locked_vm
;
2740 mm
->map_count
-= count
;
2742 mmap_write_downgrade(mm
);
2744 prev
= vma_iter_prev_range(vmi
);
2745 next
= vma_next(vmi
);
2747 vma_iter_prev_range(vmi
);
2750 * We can free page tables without write-locking mmap_lock because VMAs
2751 * were isolated before we downgraded mmap_lock.
2753 mas_set(&mas_detach
, 1);
2754 unmap_region(mm
, &mas_detach
, vma
, prev
, next
, start
, end
, count
,
2756 /* Statistics and freeing VMAs */
2757 mas_set(&mas_detach
, 0);
2758 remove_mt(mm
, &mas_detach
);
2761 mmap_read_unlock(mm
);
2763 __mt_destroy(&mt_detach
);
2768 munmap_gather_failed
:
2770 mas_set(&mas_detach
, 0);
2771 mas_for_each(&mas_detach
, next
, end
)
2772 vma_mark_detached(next
, false);
2774 __mt_destroy(&mt_detach
);
2782 * do_vmi_munmap() - munmap a given range.
2783 * @vmi: The vma iterator
2784 * @mm: The mm_struct
2785 * @start: The start address to munmap
2786 * @len: The length of the range to munmap
2787 * @uf: The userfaultfd list_head
2788 * @unlock: set to true if the user wants to drop the mmap_lock on success
2790 * This function takes a @mas that is either pointing to the previous VMA or set
2791 * to MA_START and sets it up to remove the mapping(s). The @len will be
2792 * aligned and any arch_unmap work will be preformed.
2794 * Return: 0 on success and drops the lock if so directed, error and leaves the
2795 * lock held otherwise.
2797 int do_vmi_munmap(struct vma_iterator
*vmi
, struct mm_struct
*mm
,
2798 unsigned long start
, size_t len
, struct list_head
*uf
,
2802 struct vm_area_struct
*vma
;
2804 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2807 end
= start
+ PAGE_ALIGN(len
);
2812 * Check if memory is sealed before arch_unmap.
2813 * Prevent unmapping a sealed VMA.
2814 * can_modify_mm assumes we have acquired the lock on MM.
2816 if (unlikely(!can_modify_mm(mm
, start
, end
)))
2819 /* arch_unmap() might do unmaps itself. */
2820 arch_unmap(mm
, start
, end
);
2822 /* Find the first overlapping VMA */
2823 vma
= vma_find(vmi
, end
);
2826 mmap_write_unlock(mm
);
2830 return do_vmi_align_munmap(vmi
, vma
, mm
, start
, end
, uf
, unlock
);
2833 /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
2834 * @mm: The mm_struct
2835 * @start: The start address to munmap
2836 * @len: The length to be munmapped.
2837 * @uf: The userfaultfd list_head
2839 * Return: 0 on success, error otherwise.
2841 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2842 struct list_head
*uf
)
2844 VMA_ITERATOR(vmi
, mm
, start
);
2846 return do_vmi_munmap(&vmi
, mm
, start
, len
, uf
, false);
2849 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
2850 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
2851 struct list_head
*uf
)
2853 struct mm_struct
*mm
= current
->mm
;
2854 struct vm_area_struct
*vma
= NULL
;
2855 struct vm_area_struct
*next
, *prev
, *merge
;
2856 pgoff_t pglen
= len
>> PAGE_SHIFT
;
2857 unsigned long charged
= 0;
2858 unsigned long end
= addr
+ len
;
2859 unsigned long merge_start
= addr
, merge_end
= end
;
2860 bool writable_file_mapping
= false;
2863 VMA_ITERATOR(vmi
, mm
, addr
);
2865 /* Check against address space limit. */
2866 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
2867 unsigned long nr_pages
;
2870 * MAP_FIXED may remove pages of mappings that intersects with
2871 * requested mapping. Account for the pages it would unmap.
2873 nr_pages
= count_vma_pages_range(mm
, addr
, end
);
2875 if (!may_expand_vm(mm
, vm_flags
,
2876 (len
>> PAGE_SHIFT
) - nr_pages
))
2880 /* Unmap any existing mapping in the area */
2881 error
= do_vmi_munmap(&vmi
, mm
, addr
, len
, uf
, false);
2882 if (error
== -EPERM
)
2888 * Private writable mapping: check memory availability
2890 if (accountable_mapping(file
, vm_flags
)) {
2891 charged
= len
>> PAGE_SHIFT
;
2892 if (security_vm_enough_memory_mm(mm
, charged
))
2894 vm_flags
|= VM_ACCOUNT
;
2897 next
= vma_next(&vmi
);
2898 prev
= vma_prev(&vmi
);
2899 if (vm_flags
& VM_SPECIAL
) {
2901 vma_iter_next_range(&vmi
);
2905 /* Attempt to expand an old mapping */
2907 if (next
&& next
->vm_start
== end
&& !vma_policy(next
) &&
2908 can_vma_merge_before(next
, vm_flags
, NULL
, file
, pgoff
+pglen
,
2909 NULL_VM_UFFD_CTX
, NULL
)) {
2910 merge_end
= next
->vm_end
;
2912 vm_pgoff
= next
->vm_pgoff
- pglen
;
2916 if (prev
&& prev
->vm_end
== addr
&& !vma_policy(prev
) &&
2917 (vma
? can_vma_merge_after(prev
, vm_flags
, vma
->anon_vma
, file
,
2918 pgoff
, vma
->vm_userfaultfd_ctx
, NULL
) :
2919 can_vma_merge_after(prev
, vm_flags
, NULL
, file
, pgoff
,
2920 NULL_VM_UFFD_CTX
, NULL
))) {
2921 merge_start
= prev
->vm_start
;
2923 vm_pgoff
= prev
->vm_pgoff
;
2925 vma_iter_next_range(&vmi
);
2928 /* Actually expand, if possible */
2930 !vma_expand(&vmi
, vma
, merge_start
, merge_end
, vm_pgoff
, next
)) {
2931 khugepaged_enter_vma(vma
, vm_flags
);
2936 vma_iter_set(&vmi
, addr
);
2940 * Determine the object being mapped and call the appropriate
2941 * specific mapper. the address has already been validated, but
2942 * not unmapped, but the maps are removed from the list.
2944 vma
= vm_area_alloc(mm
);
2950 vma_iter_config(&vmi
, addr
, end
);
2951 vma_set_range(vma
, addr
, end
, pgoff
);
2952 vm_flags_init(vma
, vm_flags
);
2953 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
2956 vma
->vm_file
= get_file(file
);
2957 error
= call_mmap(file
, vma
);
2959 goto unmap_and_free_vma
;
2961 if (vma_is_shared_maywrite(vma
)) {
2962 error
= mapping_map_writable(file
->f_mapping
);
2964 goto close_and_free_vma
;
2966 writable_file_mapping
= true;
2970 * Expansion is handled above, merging is handled below.
2971 * Drivers should not alter the address of the VMA.
2974 if (WARN_ON((addr
!= vma
->vm_start
)))
2975 goto close_and_free_vma
;
2977 vma_iter_config(&vmi
, addr
, end
);
2979 * If vm_flags changed after call_mmap(), we should try merge
2980 * vma again as we may succeed this time.
2982 if (unlikely(vm_flags
!= vma
->vm_flags
&& prev
)) {
2983 merge
= vma_merge_new_vma(&vmi
, prev
, vma
,
2984 vma
->vm_start
, vma
->vm_end
,
2988 * ->mmap() can change vma->vm_file and fput
2989 * the original file. So fput the vma->vm_file
2990 * here or we would add an extra fput for file
2991 * and cause general protection fault
2997 /* Update vm_flags to pick up the change. */
2998 vm_flags
= vma
->vm_flags
;
2999 goto unmap_writable
;
3003 vm_flags
= vma
->vm_flags
;
3004 } else if (vm_flags
& VM_SHARED
) {
3005 error
= shmem_zero_setup(vma
);
3009 vma_set_anonymous(vma
);
3012 if (map_deny_write_exec(vma
, vma
->vm_flags
)) {
3014 goto close_and_free_vma
;
3017 /* Allow architectures to sanity-check the vm_flags */
3019 if (!arch_validate_flags(vma
->vm_flags
))
3020 goto close_and_free_vma
;
3023 if (vma_iter_prealloc(&vmi
, vma
))
3024 goto close_and_free_vma
;
3026 /* Lock the VMA since it is modified after insertion into VMA tree */
3027 vma_start_write(vma
);
3028 vma_iter_store(&vmi
, vma
);
3033 * vma_merge() calls khugepaged_enter_vma() either, the below
3034 * call covers the non-merge case.
3036 khugepaged_enter_vma(vma
, vma
->vm_flags
);
3038 /* Once vma denies write, undo our temporary denial count */
3040 if (writable_file_mapping
)
3041 mapping_unmap_writable(file
->f_mapping
);
3042 file
= vma
->vm_file
;
3045 perf_event_mmap(vma
);
3047 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
3048 if (vm_flags
& VM_LOCKED
) {
3049 if ((vm_flags
& VM_SPECIAL
) || vma_is_dax(vma
) ||
3050 is_vm_hugetlb_page(vma
) ||
3051 vma
== get_gate_vma(current
->mm
))
3052 vm_flags_clear(vma
, VM_LOCKED_MASK
);
3054 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
3061 * New (or expanded) vma always get soft dirty status.
3062 * Otherwise user-space soft-dirty page tracker won't
3063 * be able to distinguish situation when vma area unmapped,
3064 * then new mapped in-place (which must be aimed as
3065 * a completely new data area).
3067 vm_flags_set(vma
, VM_SOFTDIRTY
);
3069 vma_set_page_prot(vma
);
3075 if (file
&& vma
->vm_ops
&& vma
->vm_ops
->close
)
3076 vma
->vm_ops
->close(vma
);
3078 if (file
|| vma
->vm_file
) {
3081 vma
->vm_file
= NULL
;
3083 vma_iter_set(&vmi
, vma
->vm_end
);
3084 /* Undo any partial mapping done by a device driver. */
3085 unmap_region(mm
, &vmi
.mas
, vma
, prev
, next
, vma
->vm_start
,
3086 vma
->vm_end
, vma
->vm_end
, true);
3088 if (writable_file_mapping
)
3089 mapping_unmap_writable(file
->f_mapping
);
3094 vm_unacct_memory(charged
);
3099 static int __vm_munmap(unsigned long start
, size_t len
, bool unlock
)
3102 struct mm_struct
*mm
= current
->mm
;
3104 VMA_ITERATOR(vmi
, mm
, start
);
3106 if (mmap_write_lock_killable(mm
))
3109 ret
= do_vmi_munmap(&vmi
, mm
, start
, len
, &uf
, unlock
);
3111 mmap_write_unlock(mm
);
3113 userfaultfd_unmap_complete(mm
, &uf
);
3117 int vm_munmap(unsigned long start
, size_t len
)
3119 return __vm_munmap(start
, len
, false);
3121 EXPORT_SYMBOL(vm_munmap
);
3123 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
3125 addr
= untagged_addr(addr
);
3126 return __vm_munmap(addr
, len
, true);
3131 * Emulation of deprecated remap_file_pages() syscall.
3133 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
3134 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
3137 struct mm_struct
*mm
= current
->mm
;
3138 struct vm_area_struct
*vma
;
3139 unsigned long populate
= 0;
3140 unsigned long ret
= -EINVAL
;
3143 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
3144 current
->comm
, current
->pid
);
3148 start
= start
& PAGE_MASK
;
3149 size
= size
& PAGE_MASK
;
3151 if (start
+ size
<= start
)
3154 /* Does pgoff wrap? */
3155 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
3158 if (mmap_write_lock_killable(mm
))
3161 vma
= vma_lookup(mm
, start
);
3163 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
3166 if (start
+ size
> vma
->vm_end
) {
3167 VMA_ITERATOR(vmi
, mm
, vma
->vm_end
);
3168 struct vm_area_struct
*next
, *prev
= vma
;
3170 for_each_vma_range(vmi
, next
, start
+ size
) {
3171 /* hole between vmas ? */
3172 if (next
->vm_start
!= prev
->vm_end
)
3175 if (next
->vm_file
!= vma
->vm_file
)
3178 if (next
->vm_flags
!= vma
->vm_flags
)
3181 if (start
+ size
<= next
->vm_end
)
3191 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
3192 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
3193 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
3195 flags
&= MAP_NONBLOCK
;
3196 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
3197 if (vma
->vm_flags
& VM_LOCKED
)
3198 flags
|= MAP_LOCKED
;
3200 file
= get_file(vma
->vm_file
);
3201 ret
= do_mmap(vma
->vm_file
, start
, size
,
3202 prot
, flags
, 0, pgoff
, &populate
, NULL
);
3205 mmap_write_unlock(mm
);
3207 mm_populate(ret
, populate
);
3208 if (!IS_ERR_VALUE(ret
))
3214 * do_vma_munmap() - Unmap a full or partial vma.
3215 * @vmi: The vma iterator pointing at the vma
3216 * @vma: The first vma to be munmapped
3217 * @start: the start of the address to unmap
3218 * @end: The end of the address to unmap
3219 * @uf: The userfaultfd list_head
3220 * @unlock: Drop the lock on success
3222 * unmaps a VMA mapping when the vma iterator is already in position.
3223 * Does not handle alignment.
3225 * Return: 0 on success drops the lock of so directed, error on failure and will
3226 * still hold the lock.
3228 int do_vma_munmap(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
3229 unsigned long start
, unsigned long end
, struct list_head
*uf
,
3232 struct mm_struct
*mm
= vma
->vm_mm
;
3235 * Check if memory is sealed before arch_unmap.
3236 * Prevent unmapping a sealed VMA.
3237 * can_modify_mm assumes we have acquired the lock on MM.
3239 if (unlikely(!can_modify_mm(mm
, start
, end
)))
3242 arch_unmap(mm
, start
, end
);
3243 return do_vmi_align_munmap(vmi
, vma
, mm
, start
, end
, uf
, unlock
);
3247 * do_brk_flags() - Increase the brk vma if the flags match.
3248 * @vmi: The vma iterator
3249 * @addr: The start address
3250 * @len: The length of the increase
3252 * @flags: The VMA Flags
3254 * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags
3255 * do not match then create a new anonymous VMA. Eventually we may be able to
3256 * do some brk-specific accounting here.
3258 static int do_brk_flags(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
3259 unsigned long addr
, unsigned long len
, unsigned long flags
)
3261 struct mm_struct
*mm
= current
->mm
;
3262 struct vma_prepare vp
;
3265 * Check against address space limits by the changed size
3266 * Note: This happens *after* clearing old mappings in some code paths.
3268 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
3269 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
3272 if (mm
->map_count
> sysctl_max_map_count
)
3275 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
3279 * Expand the existing vma if possible; Note that singular lists do not
3280 * occur after forking, so the expand will only happen on new VMAs.
3282 if (vma
&& vma
->vm_end
== addr
&& !vma_policy(vma
) &&
3283 can_vma_merge_after(vma
, flags
, NULL
, NULL
,
3284 addr
>> PAGE_SHIFT
, NULL_VM_UFFD_CTX
, NULL
)) {
3285 vma_iter_config(vmi
, vma
->vm_start
, addr
+ len
);
3286 if (vma_iter_prealloc(vmi
, vma
))
3289 vma_start_write(vma
);
3291 init_vma_prep(&vp
, vma
);
3293 vma_adjust_trans_huge(vma
, vma
->vm_start
, addr
+ len
, 0);
3294 vma
->vm_end
= addr
+ len
;
3295 vm_flags_set(vma
, VM_SOFTDIRTY
);
3296 vma_iter_store(vmi
, vma
);
3298 vma_complete(&vp
, vmi
, mm
);
3299 khugepaged_enter_vma(vma
, flags
);
3304 vma_iter_next_range(vmi
);
3305 /* create a vma struct for an anonymous mapping */
3306 vma
= vm_area_alloc(mm
);
3310 vma_set_anonymous(vma
);
3311 vma_set_range(vma
, addr
, addr
+ len
, addr
>> PAGE_SHIFT
);
3312 vm_flags_init(vma
, flags
);
3313 vma
->vm_page_prot
= vm_get_page_prot(flags
);
3314 vma_start_write(vma
);
3315 if (vma_iter_store_gfp(vmi
, vma
, GFP_KERNEL
))
3316 goto mas_store_fail
;
3322 perf_event_mmap(vma
);
3323 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3324 mm
->data_vm
+= len
>> PAGE_SHIFT
;
3325 if (flags
& VM_LOCKED
)
3326 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
3327 vm_flags_set(vma
, VM_SOFTDIRTY
);
3333 vm_unacct_memory(len
>> PAGE_SHIFT
);
3337 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
3339 struct mm_struct
*mm
= current
->mm
;
3340 struct vm_area_struct
*vma
= NULL
;
3345 VMA_ITERATOR(vmi
, mm
, addr
);
3347 len
= PAGE_ALIGN(request
);
3353 /* Until we need other flags, refuse anything except VM_EXEC. */
3354 if ((flags
& (~VM_EXEC
)) != 0)
3357 if (mmap_write_lock_killable(mm
))
3360 ret
= check_brk_limits(addr
, len
);
3364 ret
= do_vmi_munmap(&vmi
, mm
, addr
, len
, &uf
, 0);
3368 vma
= vma_prev(&vmi
);
3369 ret
= do_brk_flags(&vmi
, vma
, addr
, len
, flags
);
3370 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3371 mmap_write_unlock(mm
);
3372 userfaultfd_unmap_complete(mm
, &uf
);
3373 if (populate
&& !ret
)
3374 mm_populate(addr
, len
);
3379 mmap_write_unlock(mm
);
3382 EXPORT_SYMBOL(vm_brk_flags
);
3384 /* Release all mmaps. */
3385 void exit_mmap(struct mm_struct
*mm
)
3387 struct mmu_gather tlb
;
3388 struct vm_area_struct
*vma
;
3389 unsigned long nr_accounted
= 0;
3390 VMA_ITERATOR(vmi
, mm
, 0);
3393 /* mm's last user has gone, and its about to be pulled down */
3394 mmu_notifier_release(mm
);
3399 vma
= vma_next(&vmi
);
3400 if (!vma
|| unlikely(xa_is_zero(vma
))) {
3401 /* Can happen if dup_mmap() received an OOM */
3402 mmap_read_unlock(mm
);
3403 mmap_write_lock(mm
);
3409 tlb_gather_mmu_fullmm(&tlb
, mm
);
3410 /* update_hiwater_rss(mm) here? but nobody should be looking */
3411 /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
3412 unmap_vmas(&tlb
, &vmi
.mas
, vma
, 0, ULONG_MAX
, ULONG_MAX
, false);
3413 mmap_read_unlock(mm
);
3416 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
3417 * because the memory has been already freed.
3419 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3420 mmap_write_lock(mm
);
3421 mt_clear_in_rcu(&mm
->mm_mt
);
3422 vma_iter_set(&vmi
, vma
->vm_end
);
3423 free_pgtables(&tlb
, &vmi
.mas
, vma
, FIRST_USER_ADDRESS
,
3424 USER_PGTABLES_CEILING
, true);
3425 tlb_finish_mmu(&tlb
);
3428 * Walk the list again, actually closing and freeing it, with preemption
3429 * enabled, without holding any MM locks besides the unreachable
3432 vma_iter_set(&vmi
, vma
->vm_end
);
3434 if (vma
->vm_flags
& VM_ACCOUNT
)
3435 nr_accounted
+= vma_pages(vma
);
3436 remove_vma(vma
, true);
3439 vma
= vma_next(&vmi
);
3440 } while (vma
&& likely(!xa_is_zero(vma
)));
3442 BUG_ON(count
!= mm
->map_count
);
3444 trace_exit_mmap(mm
);
3446 __mt_destroy(&mm
->mm_mt
);
3447 mmap_write_unlock(mm
);
3448 vm_unacct_memory(nr_accounted
);
3451 /* Insert vm structure into process list sorted by address
3452 * and into the inode's i_mmap tree. If vm_file is non-NULL
3453 * then i_mmap_rwsem is taken here.
3455 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3457 unsigned long charged
= vma_pages(vma
);
3460 if (find_vma_intersection(mm
, vma
->vm_start
, vma
->vm_end
))
3463 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3464 security_vm_enough_memory_mm(mm
, charged
))
3468 * The vm_pgoff of a purely anonymous vma should be irrelevant
3469 * until its first write fault, when page's anon_vma and index
3470 * are set. But now set the vm_pgoff it will almost certainly
3471 * end up with (unless mremap moves it elsewhere before that
3472 * first wfault), so /proc/pid/maps tells a consistent story.
3474 * By setting it to reflect the virtual start address of the
3475 * vma, merges and splits can happen in a seamless way, just
3476 * using the existing file pgoff checks and manipulations.
3477 * Similarly in do_mmap and in do_brk_flags.
3479 if (vma_is_anonymous(vma
)) {
3480 BUG_ON(vma
->anon_vma
);
3481 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3484 if (vma_link(mm
, vma
)) {
3485 if (vma
->vm_flags
& VM_ACCOUNT
)
3486 vm_unacct_memory(charged
);
3494 * Copy the vma structure to a new location in the same mm,
3495 * prior to moving page table entries, to effect an mremap move.
3497 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3498 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3499 bool *need_rmap_locks
)
3501 struct vm_area_struct
*vma
= *vmap
;
3502 unsigned long vma_start
= vma
->vm_start
;
3503 struct mm_struct
*mm
= vma
->vm_mm
;
3504 struct vm_area_struct
*new_vma
, *prev
;
3505 bool faulted_in_anon_vma
= true;
3506 VMA_ITERATOR(vmi
, mm
, addr
);
3509 * If anonymous vma has not yet been faulted, update new pgoff
3510 * to match new location, to increase its chance of merging.
3512 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3513 pgoff
= addr
>> PAGE_SHIFT
;
3514 faulted_in_anon_vma
= false;
3517 new_vma
= find_vma_prev(mm
, addr
, &prev
);
3518 if (new_vma
&& new_vma
->vm_start
< addr
+ len
)
3519 return NULL
; /* should never get here */
3521 new_vma
= vma_merge_new_vma(&vmi
, prev
, vma
, addr
, addr
+ len
, pgoff
);
3524 * Source vma may have been merged into new_vma
3526 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3527 vma_start
< new_vma
->vm_end
)) {
3529 * The only way we can get a vma_merge with
3530 * self during an mremap is if the vma hasn't
3531 * been faulted in yet and we were allowed to
3532 * reset the dst vma->vm_pgoff to the
3533 * destination address of the mremap to allow
3534 * the merge to happen. mremap must change the
3535 * vm_pgoff linearity between src and dst vmas
3536 * (in turn preventing a vma_merge) to be
3537 * safe. It is only safe to keep the vm_pgoff
3538 * linear if there are no pages mapped yet.
3540 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3541 *vmap
= vma
= new_vma
;
3543 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3545 new_vma
= vm_area_dup(vma
);
3548 vma_set_range(new_vma
, addr
, addr
+ len
, pgoff
);
3549 if (vma_dup_policy(vma
, new_vma
))
3551 if (anon_vma_clone(new_vma
, vma
))
3552 goto out_free_mempol
;
3553 if (new_vma
->vm_file
)
3554 get_file(new_vma
->vm_file
);
3555 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3556 new_vma
->vm_ops
->open(new_vma
);
3557 if (vma_link(mm
, new_vma
))
3559 *need_rmap_locks
= false;
3564 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->close
)
3565 new_vma
->vm_ops
->close(new_vma
);
3567 if (new_vma
->vm_file
)
3568 fput(new_vma
->vm_file
);
3570 unlink_anon_vmas(new_vma
);
3572 mpol_put(vma_policy(new_vma
));
3574 vm_area_free(new_vma
);
3580 * Return true if the calling process may expand its vm space by the passed
3583 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3585 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3588 if (is_data_mapping(flags
) &&
3589 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3590 /* Workaround for Valgrind */
3591 if (rlimit(RLIMIT_DATA
) == 0 &&
3592 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3595 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3596 current
->comm
, current
->pid
,
3597 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3598 rlimit(RLIMIT_DATA
),
3599 ignore_rlimit_data
? "" : " or use boot option ignore_rlimit_data");
3601 if (!ignore_rlimit_data
)
3608 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3610 WRITE_ONCE(mm
->total_vm
, READ_ONCE(mm
->total_vm
)+npages
);
3612 if (is_exec_mapping(flags
))
3613 mm
->exec_vm
+= npages
;
3614 else if (is_stack_mapping(flags
))
3615 mm
->stack_vm
+= npages
;
3616 else if (is_data_mapping(flags
))
3617 mm
->data_vm
+= npages
;
3620 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
);
3623 * Having a close hook prevents vma merging regardless of flags.
3625 static void special_mapping_close(struct vm_area_struct
*vma
)
3629 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3631 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3634 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3636 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3638 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3642 return sm
->mremap(sm
, new_vma
);
3647 static int special_mapping_split(struct vm_area_struct
*vma
, unsigned long addr
)
3650 * Forbid splitting special mappings - kernel has expectations over
3651 * the number of pages in mapping. Together with VM_DONTEXPAND
3652 * the size of vma should stay the same over the special mapping's
3658 static const struct vm_operations_struct special_mapping_vmops
= {
3659 .close
= special_mapping_close
,
3660 .fault
= special_mapping_fault
,
3661 .mremap
= special_mapping_mremap
,
3662 .name
= special_mapping_name
,
3663 /* vDSO code relies that VVAR can't be accessed remotely */
3665 .may_split
= special_mapping_split
,
3668 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3669 .close
= special_mapping_close
,
3670 .fault
= special_mapping_fault
,
3673 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
)
3675 struct vm_area_struct
*vma
= vmf
->vma
;
3677 struct page
**pages
;
3679 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3680 pages
= vma
->vm_private_data
;
3682 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3685 return sm
->fault(sm
, vmf
->vma
, vmf
);
3690 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3694 struct page
*page
= *pages
;
3700 return VM_FAULT_SIGBUS
;
3703 static struct vm_area_struct
*__install_special_mapping(
3704 struct mm_struct
*mm
,
3705 unsigned long addr
, unsigned long len
,
3706 unsigned long vm_flags
, void *priv
,
3707 const struct vm_operations_struct
*ops
)
3710 struct vm_area_struct
*vma
;
3712 vma
= vm_area_alloc(mm
);
3713 if (unlikely(vma
== NULL
))
3714 return ERR_PTR(-ENOMEM
);
3716 vma_set_range(vma
, addr
, addr
+ len
, 0);
3717 vm_flags_init(vma
, (vm_flags
| mm
->def_flags
|
3718 VM_DONTEXPAND
| VM_SOFTDIRTY
) & ~VM_LOCKED_MASK
);
3719 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3722 vma
->vm_private_data
= priv
;
3724 ret
= insert_vm_struct(mm
, vma
);
3728 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3730 perf_event_mmap(vma
);
3736 return ERR_PTR(ret
);
3739 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3740 const struct vm_special_mapping
*sm
)
3742 return vma
->vm_private_data
== sm
&&
3743 (vma
->vm_ops
== &special_mapping_vmops
||
3744 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3748 * Called with mm->mmap_lock held for writing.
3749 * Insert a new vma covering the given region, with the given flags.
3750 * Its pages are supplied by the given array of struct page *.
3751 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3752 * The region past the last page supplied will always produce SIGBUS.
3753 * The array pointer and the pages it points to are assumed to stay alive
3754 * for as long as this mapping might exist.
3756 struct vm_area_struct
*_install_special_mapping(
3757 struct mm_struct
*mm
,
3758 unsigned long addr
, unsigned long len
,
3759 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3761 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3762 &special_mapping_vmops
);
3765 int install_special_mapping(struct mm_struct
*mm
,
3766 unsigned long addr
, unsigned long len
,
3767 unsigned long vm_flags
, struct page
**pages
)
3769 struct vm_area_struct
*vma
= __install_special_mapping(
3770 mm
, addr
, len
, vm_flags
, (void *)pages
,
3771 &legacy_special_mapping_vmops
);
3773 return PTR_ERR_OR_ZERO(vma
);
3776 static DEFINE_MUTEX(mm_all_locks_mutex
);
3778 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3780 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3782 * The LSB of head.next can't change from under us
3783 * because we hold the mm_all_locks_mutex.
3785 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_lock
);
3787 * We can safely modify head.next after taking the
3788 * anon_vma->root->rwsem. If some other vma in this mm shares
3789 * the same anon_vma we won't take it again.
3791 * No need of atomic instructions here, head.next
3792 * can't change from under us thanks to the
3793 * anon_vma->root->rwsem.
3795 if (__test_and_set_bit(0, (unsigned long *)
3796 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3801 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3803 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3805 * AS_MM_ALL_LOCKS can't change from under us because
3806 * we hold the mm_all_locks_mutex.
3808 * Operations on ->flags have to be atomic because
3809 * even if AS_MM_ALL_LOCKS is stable thanks to the
3810 * mm_all_locks_mutex, there may be other cpus
3811 * changing other bitflags in parallel to us.
3813 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3815 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_lock
);
3820 * This operation locks against the VM for all pte/vma/mm related
3821 * operations that could ever happen on a certain mm. This includes
3822 * vmtruncate, try_to_unmap, and all page faults.
3824 * The caller must take the mmap_lock in write mode before calling
3825 * mm_take_all_locks(). The caller isn't allowed to release the
3826 * mmap_lock until mm_drop_all_locks() returns.
3828 * mmap_lock in write mode is required in order to block all operations
3829 * that could modify pagetables and free pages without need of
3830 * altering the vma layout. It's also needed in write mode to avoid new
3831 * anon_vmas to be associated with existing vmas.
3833 * A single task can't take more than one mm_take_all_locks() in a row
3834 * or it would deadlock.
3836 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3837 * mapping->flags avoid to take the same lock twice, if more than one
3838 * vma in this mm is backed by the same anon_vma or address_space.
3840 * We take locks in following order, accordingly to comment at beginning
3842 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3844 * - all vmas marked locked
3845 * - all i_mmap_rwsem locks;
3846 * - all anon_vma->rwseml
3848 * We can take all locks within these types randomly because the VM code
3849 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3850 * mm_all_locks_mutex.
3852 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3853 * that may have to take thousand of locks.
3855 * mm_take_all_locks() can fail if it's interrupted by signals.
3857 int mm_take_all_locks(struct mm_struct
*mm
)
3859 struct vm_area_struct
*vma
;
3860 struct anon_vma_chain
*avc
;
3861 VMA_ITERATOR(vmi
, mm
, 0);
3863 mmap_assert_write_locked(mm
);
3865 mutex_lock(&mm_all_locks_mutex
);
3868 * vma_start_write() does not have a complement in mm_drop_all_locks()
3869 * because vma_start_write() is always asymmetrical; it marks a VMA as
3870 * being written to until mmap_write_unlock() or mmap_write_downgrade()
3873 for_each_vma(vmi
, vma
) {
3874 if (signal_pending(current
))
3876 vma_start_write(vma
);
3879 vma_iter_init(&vmi
, mm
, 0);
3880 for_each_vma(vmi
, vma
) {
3881 if (signal_pending(current
))
3883 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3884 is_vm_hugetlb_page(vma
))
3885 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3888 vma_iter_init(&vmi
, mm
, 0);
3889 for_each_vma(vmi
, vma
) {
3890 if (signal_pending(current
))
3892 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3893 !is_vm_hugetlb_page(vma
))
3894 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3897 vma_iter_init(&vmi
, mm
, 0);
3898 for_each_vma(vmi
, vma
) {
3899 if (signal_pending(current
))
3902 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3903 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3909 mm_drop_all_locks(mm
);
3913 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3915 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3917 * The LSB of head.next can't change to 0 from under
3918 * us because we hold the mm_all_locks_mutex.
3920 * We must however clear the bitflag before unlocking
3921 * the vma so the users using the anon_vma->rb_root will
3922 * never see our bitflag.
3924 * No need of atomic instructions here, head.next
3925 * can't change from under us until we release the
3926 * anon_vma->root->rwsem.
3928 if (!__test_and_clear_bit(0, (unsigned long *)
3929 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3931 anon_vma_unlock_write(anon_vma
);
3935 static void vm_unlock_mapping(struct address_space
*mapping
)
3937 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3939 * AS_MM_ALL_LOCKS can't change to 0 from under us
3940 * because we hold the mm_all_locks_mutex.
3942 i_mmap_unlock_write(mapping
);
3943 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3950 * The mmap_lock cannot be released by the caller until
3951 * mm_drop_all_locks() returns.
3953 void mm_drop_all_locks(struct mm_struct
*mm
)
3955 struct vm_area_struct
*vma
;
3956 struct anon_vma_chain
*avc
;
3957 VMA_ITERATOR(vmi
, mm
, 0);
3959 mmap_assert_write_locked(mm
);
3960 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3962 for_each_vma(vmi
, vma
) {
3964 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3965 vm_unlock_anon_vma(avc
->anon_vma
);
3966 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3967 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3970 mutex_unlock(&mm_all_locks_mutex
);
3974 * initialise the percpu counter for VM
3976 void __init
mmap_init(void)
3980 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3985 * Initialise sysctl_user_reserve_kbytes.
3987 * This is intended to prevent a user from starting a single memory hogging
3988 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3991 * The default value is min(3% of free memory, 128MB)
3992 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3994 static int init_user_reserve(void)
3996 unsigned long free_kbytes
;
3998 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
4000 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, SZ_128K
);
4003 subsys_initcall(init_user_reserve
);
4006 * Initialise sysctl_admin_reserve_kbytes.
4008 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
4009 * to log in and kill a memory hogging process.
4011 * Systems with more than 256MB will reserve 8MB, enough to recover
4012 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
4013 * only reserve 3% of free pages by default.
4015 static int init_admin_reserve(void)
4017 unsigned long free_kbytes
;
4019 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
4021 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, SZ_8K
);
4024 subsys_initcall(init_admin_reserve
);
4027 * Reinititalise user and admin reserves if memory is added or removed.
4029 * The default user reserve max is 128MB, and the default max for the
4030 * admin reserve is 8MB. These are usually, but not always, enough to
4031 * enable recovery from a memory hogging process using login/sshd, a shell,
4032 * and tools like top. It may make sense to increase or even disable the
4033 * reserve depending on the existence of swap or variations in the recovery
4034 * tools. So, the admin may have changed them.
4036 * If memory is added and the reserves have been eliminated or increased above
4037 * the default max, then we'll trust the admin.
4039 * If memory is removed and there isn't enough free memory, then we
4040 * need to reset the reserves.
4042 * Otherwise keep the reserve set by the admin.
4044 static int reserve_mem_notifier(struct notifier_block
*nb
,
4045 unsigned long action
, void *data
)
4047 unsigned long tmp
, free_kbytes
;
4051 /* Default max is 128MB. Leave alone if modified by operator. */
4052 tmp
= sysctl_user_reserve_kbytes
;
4053 if (tmp
> 0 && tmp
< SZ_128K
)
4054 init_user_reserve();
4056 /* Default max is 8MB. Leave alone if modified by operator. */
4057 tmp
= sysctl_admin_reserve_kbytes
;
4058 if (tmp
> 0 && tmp
< SZ_8K
)
4059 init_admin_reserve();
4063 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
4065 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
4066 init_user_reserve();
4067 pr_info("vm.user_reserve_kbytes reset to %lu\n",
4068 sysctl_user_reserve_kbytes
);
4071 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
4072 init_admin_reserve();
4073 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
4074 sysctl_admin_reserve_kbytes
);
4083 static int __meminit
init_reserve_notifier(void)
4085 if (hotplug_memory_notifier(reserve_mem_notifier
, DEFAULT_CALLBACK_PRI
))
4086 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
4090 subsys_initcall(init_reserve_notifier
);