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 const int mmap_rnd_bits_max
= CONFIG_ARCH_MMAP_RND_BITS_MAX
;
68 int mmap_rnd_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_BITS
;
70 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
71 const int mmap_rnd_compat_bits_min
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN
;
72 const int mmap_rnd_compat_bits_max
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX
;
73 int mmap_rnd_compat_bits __read_mostly
= CONFIG_ARCH_MMAP_RND_COMPAT_BITS
;
76 static bool ignore_rlimit_data
;
77 core_param(ignore_rlimit_data
, ignore_rlimit_data
, bool, 0644);
79 static void unmap_region(struct mm_struct
*mm
, 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 file
*file
, struct address_space
*mapping
)
110 if (vma
->vm_flags
& VM_SHARED
)
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
, file
, mapping
);
130 i_mmap_unlock_write(mapping
);
135 * Close a vm structure and free it.
137 static void remove_vma(struct vm_area_struct
*vma
, bool unreachable
)
140 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
141 vma
->vm_ops
->close(vma
);
144 mpol_put(vma_policy(vma
));
151 static inline struct vm_area_struct
*vma_prev_limit(struct vma_iterator
*vmi
,
154 return mas_prev(&vmi
->mas
, min
);
158 * check_brk_limits() - Use platform specific check of range & verify mlock
160 * @addr: The address to check
161 * @len: The size of increase.
163 * Return: 0 on success.
165 static int check_brk_limits(unsigned long addr
, unsigned long len
)
167 unsigned long mapped_addr
;
169 mapped_addr
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
170 if (IS_ERR_VALUE(mapped_addr
))
173 return mlock_future_ok(current
->mm
, current
->mm
->def_flags
, len
)
176 static int do_brk_flags(struct vma_iterator
*vmi
, struct vm_area_struct
*brkvma
,
177 unsigned long addr
, unsigned long request
, unsigned long flags
);
178 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
180 unsigned long newbrk
, oldbrk
, origbrk
;
181 struct mm_struct
*mm
= current
->mm
;
182 struct vm_area_struct
*brkvma
, *next
= NULL
;
183 unsigned long min_brk
;
184 bool populate
= false;
186 struct vma_iterator vmi
;
188 if (mmap_write_lock_killable(mm
))
193 #ifdef CONFIG_COMPAT_BRK
195 * CONFIG_COMPAT_BRK can still be overridden by setting
196 * randomize_va_space to 2, which will still cause mm->start_brk
197 * to be arbitrarily shifted
199 if (current
->brk_randomized
)
200 min_brk
= mm
->start_brk
;
202 min_brk
= mm
->end_data
;
204 min_brk
= mm
->start_brk
;
210 * Check against rlimit here. If this check is done later after the test
211 * of oldbrk with newbrk then it can escape the test and let the data
212 * segment grow beyond its set limit the in case where the limit is
213 * not page aligned -Ram Gupta
215 if (check_data_rlimit(rlimit(RLIMIT_DATA
), brk
, mm
->start_brk
,
216 mm
->end_data
, mm
->start_data
))
219 newbrk
= PAGE_ALIGN(brk
);
220 oldbrk
= PAGE_ALIGN(mm
->brk
);
221 if (oldbrk
== newbrk
) {
226 /* Always allow shrinking brk. */
227 if (brk
<= mm
->brk
) {
228 /* Search one past newbrk */
229 vma_iter_init(&vmi
, mm
, newbrk
);
230 brkvma
= vma_find(&vmi
, oldbrk
);
231 if (!brkvma
|| brkvma
->vm_start
>= oldbrk
)
232 goto out
; /* mapping intersects with an existing non-brk vma. */
234 * mm->brk must be protected by write mmap_lock.
235 * do_vma_munmap() will drop the lock on success, so update it
236 * before calling do_vma_munmap().
239 if (do_vma_munmap(&vmi
, brkvma
, newbrk
, oldbrk
, &uf
, true))
242 goto success_unlocked
;
245 if (check_brk_limits(oldbrk
, newbrk
- oldbrk
))
249 * Only check if the next VMA is within the stack_guard_gap of the
252 vma_iter_init(&vmi
, mm
, oldbrk
);
253 next
= vma_find(&vmi
, newbrk
+ PAGE_SIZE
+ stack_guard_gap
);
254 if (next
&& newbrk
+ PAGE_SIZE
> vm_start_gap(next
))
257 brkvma
= vma_prev_limit(&vmi
, mm
->start_brk
);
258 /* Ok, looks good - let it rip. */
259 if (do_brk_flags(&vmi
, brkvma
, oldbrk
, newbrk
- oldbrk
, 0) < 0)
263 if (mm
->def_flags
& VM_LOCKED
)
267 mmap_write_unlock(mm
);
269 userfaultfd_unmap_complete(mm
, &uf
);
271 mm_populate(oldbrk
, newbrk
- oldbrk
);
276 mmap_write_unlock(mm
);
280 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
281 static void validate_mm(struct mm_struct
*mm
)
285 struct vm_area_struct
*vma
;
286 VMA_ITERATOR(vmi
, mm
, 0);
288 mt_validate(&mm
->mm_mt
);
289 for_each_vma(vmi
, vma
) {
290 #ifdef CONFIG_DEBUG_VM_RB
291 struct anon_vma
*anon_vma
= vma
->anon_vma
;
292 struct anon_vma_chain
*avc
;
294 unsigned long vmi_start
, vmi_end
;
297 vmi_start
= vma_iter_addr(&vmi
);
298 vmi_end
= vma_iter_end(&vmi
);
299 if (VM_WARN_ON_ONCE_MM(vma
->vm_end
!= vmi_end
, mm
))
302 if (VM_WARN_ON_ONCE_MM(vma
->vm_start
!= vmi_start
, mm
))
306 pr_emerg("issue in %s\n", current
->comm
);
309 pr_emerg("tree range: %px start %lx end %lx\n", vma
,
310 vmi_start
, vmi_end
- 1);
311 vma_iter_dump_tree(&vmi
);
314 #ifdef CONFIG_DEBUG_VM_RB
316 anon_vma_lock_read(anon_vma
);
317 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
318 anon_vma_interval_tree_verify(avc
);
319 anon_vma_unlock_read(anon_vma
);
324 if (i
!= mm
->map_count
) {
325 pr_emerg("map_count %d vma iterator %d\n", mm
->map_count
, i
);
328 VM_BUG_ON_MM(bug
, mm
);
331 #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */
332 #define validate_mm(mm) do { } while (0)
333 #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
336 * vma has some anon_vma assigned, and is already inserted on that
337 * anon_vma's interval trees.
339 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
340 * vma must be removed from the anon_vma's interval trees using
341 * anon_vma_interval_tree_pre_update_vma().
343 * After the update, the vma will be reinserted using
344 * anon_vma_interval_tree_post_update_vma().
346 * The entire update must be protected by exclusive mmap_lock and by
347 * the root anon_vma's mutex.
350 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct
*vma
)
352 struct anon_vma_chain
*avc
;
354 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
355 anon_vma_interval_tree_remove(avc
, &avc
->anon_vma
->rb_root
);
359 anon_vma_interval_tree_post_update_vma(struct vm_area_struct
*vma
)
361 struct anon_vma_chain
*avc
;
363 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
364 anon_vma_interval_tree_insert(avc
, &avc
->anon_vma
->rb_root
);
367 static unsigned long count_vma_pages_range(struct mm_struct
*mm
,
368 unsigned long addr
, unsigned long end
)
370 VMA_ITERATOR(vmi
, mm
, addr
);
371 struct vm_area_struct
*vma
;
372 unsigned long nr_pages
= 0;
374 for_each_vma_range(vmi
, vma
, end
) {
375 unsigned long vm_start
= max(addr
, vma
->vm_start
);
376 unsigned long vm_end
= min(end
, vma
->vm_end
);
378 nr_pages
+= PHYS_PFN(vm_end
- vm_start
);
384 static void __vma_link_file(struct vm_area_struct
*vma
,
385 struct address_space
*mapping
)
387 if (vma
->vm_flags
& VM_SHARED
)
388 mapping_allow_writable(mapping
);
390 flush_dcache_mmap_lock(mapping
);
391 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
392 flush_dcache_mmap_unlock(mapping
);
395 static int vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
397 VMA_ITERATOR(vmi
, mm
, 0);
398 struct address_space
*mapping
= NULL
;
400 vma_iter_config(&vmi
, vma
->vm_start
, vma
->vm_end
);
401 if (vma_iter_prealloc(&vmi
, vma
))
404 vma_start_write(vma
);
406 vma_iter_store(&vmi
, vma
);
409 mapping
= vma
->vm_file
->f_mapping
;
410 i_mmap_lock_write(mapping
);
411 __vma_link_file(vma
, mapping
);
412 i_mmap_unlock_write(mapping
);
421 * init_multi_vma_prep() - Initializer for struct vma_prepare
422 * @vp: The vma_prepare struct
423 * @vma: The vma that will be altered once locked
424 * @next: The next vma if it is to be adjusted
425 * @remove: The first vma to be removed
426 * @remove2: The second vma to be removed
428 static inline void init_multi_vma_prep(struct vma_prepare
*vp
,
429 struct vm_area_struct
*vma
, struct vm_area_struct
*next
,
430 struct vm_area_struct
*remove
, struct vm_area_struct
*remove2
)
432 memset(vp
, 0, sizeof(struct vma_prepare
));
434 vp
->anon_vma
= vma
->anon_vma
;
436 vp
->remove2
= remove2
;
438 if (!vp
->anon_vma
&& next
)
439 vp
->anon_vma
= next
->anon_vma
;
441 vp
->file
= vma
->vm_file
;
443 vp
->mapping
= vma
->vm_file
->f_mapping
;
448 * init_vma_prep() - Initializer wrapper for vma_prepare struct
449 * @vp: The vma_prepare struct
450 * @vma: The vma that will be altered once locked
452 static inline void init_vma_prep(struct vma_prepare
*vp
,
453 struct vm_area_struct
*vma
)
455 init_multi_vma_prep(vp
, vma
, NULL
, NULL
, NULL
);
460 * vma_prepare() - Helper function for handling locking VMAs prior to altering
461 * @vp: The initialized vma_prepare struct
463 static inline void vma_prepare(struct vma_prepare
*vp
)
466 uprobe_munmap(vp
->vma
, vp
->vma
->vm_start
, vp
->vma
->vm_end
);
469 uprobe_munmap(vp
->adj_next
, vp
->adj_next
->vm_start
,
470 vp
->adj_next
->vm_end
);
472 i_mmap_lock_write(vp
->mapping
);
473 if (vp
->insert
&& vp
->insert
->vm_file
) {
475 * Put into interval tree now, so instantiated pages
476 * are visible to arm/parisc __flush_dcache_page
477 * throughout; but we cannot insert into address
478 * space until vma start or end is updated.
480 __vma_link_file(vp
->insert
,
481 vp
->insert
->vm_file
->f_mapping
);
486 anon_vma_lock_write(vp
->anon_vma
);
487 anon_vma_interval_tree_pre_update_vma(vp
->vma
);
489 anon_vma_interval_tree_pre_update_vma(vp
->adj_next
);
493 flush_dcache_mmap_lock(vp
->mapping
);
494 vma_interval_tree_remove(vp
->vma
, &vp
->mapping
->i_mmap
);
496 vma_interval_tree_remove(vp
->adj_next
,
497 &vp
->mapping
->i_mmap
);
503 * vma_complete- Helper function for handling the unlocking after altering VMAs,
504 * or for inserting a VMA.
506 * @vp: The vma_prepare struct
507 * @vmi: The vma iterator
510 static inline void vma_complete(struct vma_prepare
*vp
,
511 struct vma_iterator
*vmi
, struct mm_struct
*mm
)
515 vma_interval_tree_insert(vp
->adj_next
,
516 &vp
->mapping
->i_mmap
);
517 vma_interval_tree_insert(vp
->vma
, &vp
->mapping
->i_mmap
);
518 flush_dcache_mmap_unlock(vp
->mapping
);
521 if (vp
->remove
&& vp
->file
) {
522 __remove_shared_vm_struct(vp
->remove
, vp
->file
, vp
->mapping
);
524 __remove_shared_vm_struct(vp
->remove2
, vp
->file
,
526 } else if (vp
->insert
) {
528 * split_vma has split insert from vma, and needs
529 * us to insert it before dropping the locks
530 * (it may either follow vma or precede it).
532 vma_iter_store(vmi
, vp
->insert
);
537 anon_vma_interval_tree_post_update_vma(vp
->vma
);
539 anon_vma_interval_tree_post_update_vma(vp
->adj_next
);
540 anon_vma_unlock_write(vp
->anon_vma
);
544 i_mmap_unlock_write(vp
->mapping
);
545 uprobe_mmap(vp
->vma
);
548 uprobe_mmap(vp
->adj_next
);
553 vma_mark_detached(vp
->remove
, true);
555 uprobe_munmap(vp
->remove
, vp
->remove
->vm_start
,
559 if (vp
->remove
->anon_vma
)
560 anon_vma_merge(vp
->vma
, vp
->remove
);
562 mpol_put(vma_policy(vp
->remove
));
564 WARN_ON_ONCE(vp
->vma
->vm_end
< vp
->remove
->vm_end
);
565 vm_area_free(vp
->remove
);
568 * In mprotect's case 6 (see comments on vma_merge),
569 * we are removing both mid and next vmas
572 vp
->remove
= vp
->remove2
;
577 if (vp
->insert
&& vp
->file
)
578 uprobe_mmap(vp
->insert
);
583 * dup_anon_vma() - Helper function to duplicate anon_vma
584 * @dst: The destination VMA
585 * @src: The source VMA
587 * Returns: 0 on success.
589 static inline int dup_anon_vma(struct vm_area_struct
*dst
,
590 struct vm_area_struct
*src
)
593 * Easily overlooked: when mprotect shifts the boundary, make sure the
594 * expanding vma has anon_vma set if the shrinking vma had, to cover any
595 * anon pages imported.
597 if (src
->anon_vma
&& !dst
->anon_vma
) {
598 vma_assert_write_locked(dst
);
599 dst
->anon_vma
= src
->anon_vma
;
600 return anon_vma_clone(dst
, src
);
607 * vma_expand - Expand an existing VMA
609 * @vmi: The vma iterator
610 * @vma: The vma to expand
611 * @start: The start of the vma
612 * @end: The exclusive end of the vma
613 * @pgoff: The page offset of vma
614 * @next: The current of next vma.
616 * Expand @vma to @start and @end. Can expand off the start and end. Will
617 * expand over @next if it's different from @vma and @end == @next->vm_end.
618 * Checking if the @vma can expand and merge with @next needs to be handled by
621 * Returns: 0 on success
623 int vma_expand(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
624 unsigned long start
, unsigned long end
, pgoff_t pgoff
,
625 struct vm_area_struct
*next
)
627 bool remove_next
= false;
628 struct vma_prepare vp
;
630 vma_start_write(vma
);
631 if (next
&& (vma
!= next
) && (end
== next
->vm_end
)) {
635 vma_start_write(next
);
636 ret
= dup_anon_vma(vma
, next
);
641 init_multi_vma_prep(&vp
, vma
, NULL
, remove_next
? next
: NULL
, NULL
);
642 /* Not merging but overwriting any part of next is not handled. */
643 VM_WARN_ON(next
&& !vp
.remove
&&
644 next
!= vma
&& end
> next
->vm_start
);
645 /* Only handles expanding */
646 VM_WARN_ON(vma
->vm_start
< start
|| vma
->vm_end
> end
);
648 /* Note: vma iterator must be pointing to 'start' */
649 vma_iter_config(vmi
, start
, end
);
650 if (vma_iter_prealloc(vmi
, vma
))
654 vma_adjust_trans_huge(vma
, start
, end
, 0);
655 vma
->vm_start
= start
;
657 vma
->vm_pgoff
= pgoff
;
658 vma_iter_store(vmi
, vma
);
660 vma_complete(&vp
, vmi
, vma
->vm_mm
);
668 * vma_shrink() - Reduce an existing VMAs memory area
669 * @vmi: The vma iterator
670 * @vma: The VMA to modify
671 * @start: The new start
674 * Returns: 0 on success, -ENOMEM otherwise
676 int vma_shrink(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
677 unsigned long start
, unsigned long end
, pgoff_t pgoff
)
679 struct vma_prepare vp
;
681 WARN_ON((vma
->vm_start
!= start
) && (vma
->vm_end
!= end
));
683 if (vma
->vm_start
< start
)
684 vma_iter_config(vmi
, vma
->vm_start
, start
);
686 vma_iter_config(vmi
, end
, vma
->vm_end
);
688 if (vma_iter_prealloc(vmi
, NULL
))
691 vma_start_write(vma
);
693 init_vma_prep(&vp
, vma
);
695 vma_adjust_trans_huge(vma
, start
, end
, 0);
698 vma
->vm_start
= start
;
700 vma
->vm_pgoff
= pgoff
;
701 vma_complete(&vp
, vmi
, vma
->vm_mm
);
706 * If the vma has a ->close operation then the driver probably needs to release
707 * per-vma resources, so we don't attempt to merge those if the caller indicates
708 * the current vma may be removed as part of the merge.
710 static inline bool is_mergeable_vma(struct vm_area_struct
*vma
,
711 struct file
*file
, unsigned long vm_flags
,
712 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
713 struct anon_vma_name
*anon_name
, bool may_remove_vma
)
716 * VM_SOFTDIRTY should not prevent from VMA merging, if we
717 * match the flags but dirty bit -- the caller should mark
718 * merged VMA as dirty. If dirty bit won't be excluded from
719 * comparison, we increase pressure on the memory system forcing
720 * the kernel to generate new VMAs when old one could be
723 if ((vma
->vm_flags
^ vm_flags
) & ~VM_SOFTDIRTY
)
725 if (vma
->vm_file
!= file
)
727 if (may_remove_vma
&& vma
->vm_ops
&& vma
->vm_ops
->close
)
729 if (!is_mergeable_vm_userfaultfd_ctx(vma
, vm_userfaultfd_ctx
))
731 if (!anon_vma_name_eq(anon_vma_name(vma
), anon_name
))
736 static inline bool is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
737 struct anon_vma
*anon_vma2
, struct vm_area_struct
*vma
)
740 * The list_is_singular() test is to avoid merging VMA cloned from
741 * parents. This can improve scalability caused by anon_vma lock.
743 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
744 list_is_singular(&vma
->anon_vma_chain
)))
746 return anon_vma1
== anon_vma2
;
750 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
751 * in front of (at a lower virtual address and file offset than) the vma.
753 * We cannot merge two vmas if they have differently assigned (non-NULL)
754 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
756 * We don't check here for the merged mmap wrapping around the end of pagecache
757 * indices (16TB on ia32) because do_mmap() does not permit mmap's which
758 * wrap, nor mmaps which cover the final page at index -1UL.
760 * We assume the vma may be removed as part of the merge.
763 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
764 struct anon_vma
*anon_vma
, struct file
*file
,
765 pgoff_t vm_pgoff
, struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
766 struct anon_vma_name
*anon_name
)
768 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
, true) &&
769 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
770 if (vma
->vm_pgoff
== vm_pgoff
)
777 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
778 * beyond (at a higher virtual address and file offset than) the vma.
780 * We cannot merge two vmas if they have differently assigned (non-NULL)
781 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
783 * We assume that vma is not removed as part of the merge.
786 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
787 struct anon_vma
*anon_vma
, struct file
*file
,
788 pgoff_t vm_pgoff
, struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
789 struct anon_vma_name
*anon_name
)
791 if (is_mergeable_vma(vma
, file
, vm_flags
, vm_userfaultfd_ctx
, anon_name
, false) &&
792 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
794 vm_pglen
= vma_pages(vma
);
795 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
802 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
803 * figure out whether that can be merged with its predecessor or its
804 * successor. Or both (it neatly fills a hole).
806 * In most cases - when called for mmap, brk or mremap - [addr,end) is
807 * certain not to be mapped by the time vma_merge is called; but when
808 * called for mprotect, it is certain to be already mapped (either at
809 * an offset within prev, or at the start of next), and the flags of
810 * this area are about to be changed to vm_flags - and the no-change
811 * case has already been eliminated.
813 * The following mprotect cases have to be considered, where **** is
814 * the area passed down from mprotect_fixup, never extending beyond one
815 * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts
816 * at the same address as **** and is of the same or larger span, and
817 * NNNN the next vma after ****:
820 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC
821 * cannot merge might become might become
822 * PPNNNNNNNNNN PPPPPPPPPPCC
823 * mmap, brk or case 4 below case 5 below
826 * PPPP NNNN PPPPCCCCNNNN
827 * might become might become
828 * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or
829 * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or
830 * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8
832 * It is important for case 8 that the vma CCCC overlapping the
833 * region **** is never going to extended over NNNN. Instead NNNN must
834 * be extended in region **** and CCCC must be removed. This way in
835 * all cases where vma_merge succeeds, the moment vma_merge drops the
836 * rmap_locks, the properties of the merged vma will be already
837 * correct for the whole merged range. Some of those properties like
838 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
839 * be correct for the whole merged range immediately after the
840 * rmap_locks are released. Otherwise if NNNN would be removed and
841 * CCCC would be extended over the NNNN range, remove_migration_ptes
842 * or other rmap walkers (if working on addresses beyond the "end"
843 * parameter) may establish ptes with the wrong permissions of CCCC
844 * instead of the right permissions of NNNN.
847 * PPPP is represented by *prev
848 * CCCC is represented by *curr or not represented at all (NULL)
849 * NNNN is represented by *next or not represented at all (NULL)
850 * **** is not represented - it will be merged and the vma containing the
851 * area is returned, or the function will return NULL
853 struct vm_area_struct
*vma_merge(struct vma_iterator
*vmi
, struct mm_struct
*mm
,
854 struct vm_area_struct
*prev
, unsigned long addr
,
855 unsigned long end
, unsigned long vm_flags
,
856 struct anon_vma
*anon_vma
, struct file
*file
,
857 pgoff_t pgoff
, struct mempolicy
*policy
,
858 struct vm_userfaultfd_ctx vm_userfaultfd_ctx
,
859 struct anon_vma_name
*anon_name
)
861 struct vm_area_struct
*curr
, *next
, *res
;
862 struct vm_area_struct
*vma
, *adjust
, *remove
, *remove2
;
863 struct vma_prepare vp
;
866 bool merge_prev
= false;
867 bool merge_next
= false;
868 bool vma_expanded
= false;
869 unsigned long vma_start
= addr
;
870 unsigned long vma_end
= end
;
871 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
875 * We later require that vma->vm_flags == vm_flags,
876 * so this tests vma->vm_flags & VM_SPECIAL, too.
878 if (vm_flags
& VM_SPECIAL
)
881 /* Does the input range span an existing VMA? (cases 5 - 8) */
882 curr
= find_vma_intersection(mm
, prev
? prev
->vm_end
: 0, end
);
884 if (!curr
|| /* cases 1 - 4 */
885 end
== curr
->vm_end
) /* cases 6 - 8, adjacent VMA */
886 next
= vma_lookup(mm
, end
);
888 next
= NULL
; /* case 5 */
891 vma_start
= prev
->vm_start
;
892 vma_pgoff
= prev
->vm_pgoff
;
894 /* Can we merge the predecessor? */
895 if (addr
== prev
->vm_end
&& mpol_equal(vma_policy(prev
), policy
)
896 && can_vma_merge_after(prev
, vm_flags
, anon_vma
, file
,
897 pgoff
, vm_userfaultfd_ctx
, anon_name
)) {
903 /* Can we merge the successor? */
904 if (next
&& mpol_equal(policy
, vma_policy(next
)) &&
905 can_vma_merge_before(next
, vm_flags
, anon_vma
, file
, pgoff
+pglen
,
906 vm_userfaultfd_ctx
, anon_name
)) {
910 /* Verify some invariant that must be enforced by the caller. */
911 VM_WARN_ON(prev
&& addr
<= prev
->vm_start
);
912 VM_WARN_ON(curr
&& (addr
!= curr
->vm_start
|| end
> curr
->vm_end
));
913 VM_WARN_ON(addr
>= end
);
915 if (!merge_prev
&& !merge_next
)
916 return NULL
; /* Not mergeable. */
919 vma_start_write(prev
);
922 remove
= remove2
= adjust
= NULL
;
924 /* Can we merge both the predecessor and the successor? */
925 if (merge_prev
&& merge_next
&&
926 is_mergeable_anon_vma(prev
->anon_vma
, next
->anon_vma
, NULL
)) {
927 vma_start_write(next
);
928 remove
= next
; /* case 1 */
929 vma_end
= next
->vm_end
;
930 err
= dup_anon_vma(prev
, next
);
931 if (curr
) { /* case 6 */
932 vma_start_write(curr
);
936 err
= dup_anon_vma(prev
, curr
);
938 } else if (merge_prev
) { /* case 2 */
940 vma_start_write(curr
);
941 err
= dup_anon_vma(prev
, curr
);
942 if (end
== curr
->vm_end
) { /* case 7 */
944 } else { /* case 5 */
946 adj_start
= (end
- curr
->vm_start
);
949 } else { /* merge_next */
950 vma_start_write(next
);
952 if (prev
&& addr
< prev
->vm_end
) { /* case 4 */
953 vma_start_write(prev
);
956 adj_start
= -(prev
->vm_end
- addr
);
957 err
= dup_anon_vma(next
, prev
);
960 * Note that cases 3 and 8 are the ONLY ones where prev
961 * is permitted to be (but is not necessarily) NULL.
963 vma
= next
; /* case 3 */
965 vma_end
= next
->vm_end
;
966 vma_pgoff
= next
->vm_pgoff
- pglen
;
967 if (curr
) { /* case 8 */
968 vma_pgoff
= curr
->vm_pgoff
;
969 vma_start_write(curr
);
971 err
= dup_anon_vma(next
, curr
);
976 /* Error in anon_vma clone. */
980 if (vma_start
< vma
->vm_start
|| vma_end
> vma
->vm_end
)
984 vma_iter_config(vmi
, vma_start
, vma_end
);
986 vma_iter_config(vmi
, adjust
->vm_start
+ adj_start
,
990 if (vma_iter_prealloc(vmi
, vma
))
993 init_multi_vma_prep(&vp
, vma
, adjust
, remove
, remove2
);
994 VM_WARN_ON(vp
.anon_vma
&& adjust
&& adjust
->anon_vma
&&
995 vp
.anon_vma
!= adjust
->anon_vma
);
998 vma_adjust_trans_huge(vma
, vma_start
, vma_end
, adj_start
);
1000 vma
->vm_start
= vma_start
;
1001 vma
->vm_end
= vma_end
;
1002 vma
->vm_pgoff
= vma_pgoff
;
1005 vma_iter_store(vmi
, vma
);
1008 adjust
->vm_start
+= adj_start
;
1009 adjust
->vm_pgoff
+= adj_start
>> PAGE_SHIFT
;
1010 if (adj_start
< 0) {
1011 WARN_ON(vma_expanded
);
1012 vma_iter_store(vmi
, next
);
1016 vma_complete(&vp
, vmi
, mm
);
1017 khugepaged_enter_vma(res
, vm_flags
);
1022 * Rough compatibility check to quickly see if it's even worth looking
1023 * at sharing an anon_vma.
1025 * They need to have the same vm_file, and the flags can only differ
1026 * in things that mprotect may change.
1028 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1029 * we can merge the two vma's. For example, we refuse to merge a vma if
1030 * there is a vm_ops->close() function, because that indicates that the
1031 * driver is doing some kind of reference counting. But that doesn't
1032 * really matter for the anon_vma sharing case.
1034 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1036 return a
->vm_end
== b
->vm_start
&&
1037 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
1038 a
->vm_file
== b
->vm_file
&&
1039 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_ACCESS_FLAGS
| VM_SOFTDIRTY
)) &&
1040 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
1044 * Do some basic sanity checking to see if we can re-use the anon_vma
1045 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1046 * the same as 'old', the other will be the new one that is trying
1047 * to share the anon_vma.
1049 * NOTE! This runs with mmap_lock held for reading, so it is possible that
1050 * the anon_vma of 'old' is concurrently in the process of being set up
1051 * by another page fault trying to merge _that_. But that's ok: if it
1052 * is being set up, that automatically means that it will be a singleton
1053 * acceptable for merging, so we can do all of this optimistically. But
1054 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1056 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1057 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1058 * is to return an anon_vma that is "complex" due to having gone through
1061 * We also make sure that the two vma's are compatible (adjacent,
1062 * and with the same memory policies). That's all stable, even with just
1063 * a read lock on the mmap_lock.
1065 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
1067 if (anon_vma_compatible(a
, b
)) {
1068 struct anon_vma
*anon_vma
= READ_ONCE(old
->anon_vma
);
1070 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
1077 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1078 * neighbouring vmas for a suitable anon_vma, before it goes off
1079 * to allocate a new anon_vma. It checks because a repetitive
1080 * sequence of mprotects and faults may otherwise lead to distinct
1081 * anon_vmas being allocated, preventing vma merge in subsequent
1084 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
1086 MA_STATE(mas
, &vma
->vm_mm
->mm_mt
, vma
->vm_end
, vma
->vm_end
);
1087 struct anon_vma
*anon_vma
= NULL
;
1088 struct vm_area_struct
*prev
, *next
;
1090 /* Try next first. */
1091 next
= mas_walk(&mas
);
1093 anon_vma
= reusable_anon_vma(next
, vma
, next
);
1098 prev
= mas_prev(&mas
, 0);
1099 VM_BUG_ON_VMA(prev
!= vma
, vma
);
1100 prev
= mas_prev(&mas
, 0);
1101 /* Try prev next. */
1103 anon_vma
= reusable_anon_vma(prev
, prev
, vma
);
1106 * We might reach here with anon_vma == NULL if we can't find
1107 * any reusable anon_vma.
1108 * There's no absolute need to look only at touching neighbours:
1109 * we could search further afield for "compatible" anon_vmas.
1110 * But it would probably just be a waste of time searching,
1111 * or lead to too many vmas hanging off the same anon_vma.
1112 * We're trying to allow mprotect remerging later on,
1113 * not trying to minimize memory used for anon_vmas.
1119 * If a hint addr is less than mmap_min_addr change hint to be as
1120 * low as possible but still greater than mmap_min_addr
1122 static inline unsigned long round_hint_to_min(unsigned long hint
)
1125 if (((void *)hint
!= NULL
) &&
1126 (hint
< mmap_min_addr
))
1127 return PAGE_ALIGN(mmap_min_addr
);
1131 bool mlock_future_ok(struct mm_struct
*mm
, unsigned long flags
,
1132 unsigned long bytes
)
1134 unsigned long locked_pages
, limit_pages
;
1136 if (!(flags
& VM_LOCKED
) || capable(CAP_IPC_LOCK
))
1139 locked_pages
= bytes
>> PAGE_SHIFT
;
1140 locked_pages
+= mm
->locked_vm
;
1142 limit_pages
= rlimit(RLIMIT_MEMLOCK
);
1143 limit_pages
>>= PAGE_SHIFT
;
1145 return locked_pages
<= limit_pages
;
1148 static inline u64
file_mmap_size_max(struct file
*file
, struct inode
*inode
)
1150 if (S_ISREG(inode
->i_mode
))
1151 return MAX_LFS_FILESIZE
;
1153 if (S_ISBLK(inode
->i_mode
))
1154 return MAX_LFS_FILESIZE
;
1156 if (S_ISSOCK(inode
->i_mode
))
1157 return MAX_LFS_FILESIZE
;
1159 /* Special "we do even unsigned file positions" case */
1160 if (file
->f_mode
& FMODE_UNSIGNED_OFFSET
)
1163 /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1167 static inline bool file_mmap_ok(struct file
*file
, struct inode
*inode
,
1168 unsigned long pgoff
, unsigned long len
)
1170 u64 maxsize
= file_mmap_size_max(file
, inode
);
1172 if (maxsize
&& len
> maxsize
)
1175 if (pgoff
> maxsize
>> PAGE_SHIFT
)
1181 * The caller must write-lock current->mm->mmap_lock.
1183 unsigned long do_mmap(struct file
*file
, unsigned long addr
,
1184 unsigned long len
, unsigned long prot
,
1185 unsigned long flags
, vm_flags_t vm_flags
,
1186 unsigned long pgoff
, unsigned long *populate
,
1187 struct list_head
*uf
)
1189 struct mm_struct
*mm
= current
->mm
;
1198 * Does the application expect PROT_READ to imply PROT_EXEC?
1200 * (the exception is when the underlying filesystem is noexec
1201 * mounted, in which case we dont add PROT_EXEC.)
1203 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1204 if (!(file
&& path_noexec(&file
->f_path
)))
1207 /* force arch specific MAP_FIXED handling in get_unmapped_area */
1208 if (flags
& MAP_FIXED_NOREPLACE
)
1211 if (!(flags
& MAP_FIXED
))
1212 addr
= round_hint_to_min(addr
);
1214 /* Careful about overflows.. */
1215 len
= PAGE_ALIGN(len
);
1219 /* offset overflow? */
1220 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1223 /* Too many mappings? */
1224 if (mm
->map_count
> sysctl_max_map_count
)
1227 /* Obtain the address to map to. we verify (or select) it and ensure
1228 * that it represents a valid section of the address space.
1230 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1231 if (IS_ERR_VALUE(addr
))
1234 if (flags
& MAP_FIXED_NOREPLACE
) {
1235 if (find_vma_intersection(mm
, addr
, addr
+ len
))
1239 if (prot
== PROT_EXEC
) {
1240 pkey
= execute_only_pkey(mm
);
1245 /* Do simple checking here so the lower-level routines won't have
1246 * to. we assume access permissions have been handled by the open
1247 * of the memory object, so we don't do any here.
1249 vm_flags
|= calc_vm_prot_bits(prot
, pkey
) | calc_vm_flag_bits(flags
) |
1250 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1252 if (flags
& MAP_LOCKED
)
1253 if (!can_do_mlock())
1256 if (!mlock_future_ok(mm
, vm_flags
, len
))
1260 struct inode
*inode
= file_inode(file
);
1261 unsigned long flags_mask
;
1263 if (!file_mmap_ok(file
, inode
, pgoff
, len
))
1266 flags_mask
= LEGACY_MAP_MASK
| file
->f_op
->mmap_supported_flags
;
1268 switch (flags
& MAP_TYPE
) {
1271 * Force use of MAP_SHARED_VALIDATE with non-legacy
1272 * flags. E.g. MAP_SYNC is dangerous to use with
1273 * MAP_SHARED as you don't know which consistency model
1274 * you will get. We silently ignore unsupported flags
1275 * with MAP_SHARED to preserve backward compatibility.
1277 flags
&= LEGACY_MAP_MASK
;
1279 case MAP_SHARED_VALIDATE
:
1280 if (flags
& ~flags_mask
)
1282 if (prot
& PROT_WRITE
) {
1283 if (!(file
->f_mode
& FMODE_WRITE
))
1285 if (IS_SWAPFILE(file
->f_mapping
->host
))
1290 * Make sure we don't allow writing to an append-only
1293 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1296 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1297 if (!(file
->f_mode
& FMODE_WRITE
))
1298 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1301 if (!(file
->f_mode
& FMODE_READ
))
1303 if (path_noexec(&file
->f_path
)) {
1304 if (vm_flags
& VM_EXEC
)
1306 vm_flags
&= ~VM_MAYEXEC
;
1309 if (!file
->f_op
->mmap
)
1311 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1319 switch (flags
& MAP_TYPE
) {
1321 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1327 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1331 * Set pgoff according to addr for anon_vma.
1333 pgoff
= addr
>> PAGE_SHIFT
;
1341 * Set 'VM_NORESERVE' if we should not account for the
1342 * memory use of this mapping.
1344 if (flags
& MAP_NORESERVE
) {
1345 /* We honor MAP_NORESERVE if allowed to overcommit */
1346 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1347 vm_flags
|= VM_NORESERVE
;
1349 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1350 if (file
&& is_file_hugepages(file
))
1351 vm_flags
|= VM_NORESERVE
;
1354 addr
= mmap_region(file
, addr
, len
, vm_flags
, pgoff
, uf
);
1355 if (!IS_ERR_VALUE(addr
) &&
1356 ((vm_flags
& VM_LOCKED
) ||
1357 (flags
& (MAP_POPULATE
| MAP_NONBLOCK
)) == MAP_POPULATE
))
1362 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1363 unsigned long prot
, unsigned long flags
,
1364 unsigned long fd
, unsigned long pgoff
)
1366 struct file
*file
= NULL
;
1367 unsigned long retval
;
1369 if (!(flags
& MAP_ANONYMOUS
)) {
1370 audit_mmap_fd(fd
, flags
);
1374 if (is_file_hugepages(file
)) {
1375 len
= ALIGN(len
, huge_page_size(hstate_file(file
)));
1376 } else if (unlikely(flags
& MAP_HUGETLB
)) {
1380 } else if (flags
& MAP_HUGETLB
) {
1383 hs
= hstate_sizelog((flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1387 len
= ALIGN(len
, huge_page_size(hs
));
1389 * VM_NORESERVE is used because the reservations will be
1390 * taken when vm_ops->mmap() is called
1392 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
,
1394 HUGETLB_ANONHUGE_INODE
,
1395 (flags
>> MAP_HUGE_SHIFT
) & MAP_HUGE_MASK
);
1397 return PTR_ERR(file
);
1400 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1407 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1408 unsigned long, prot
, unsigned long, flags
,
1409 unsigned long, fd
, unsigned long, pgoff
)
1411 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1414 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1415 struct mmap_arg_struct
{
1419 unsigned long flags
;
1421 unsigned long offset
;
1424 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1426 struct mmap_arg_struct a
;
1428 if (copy_from_user(&a
, arg
, sizeof(a
)))
1430 if (offset_in_page(a
.offset
))
1433 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1434 a
.offset
>> PAGE_SHIFT
);
1436 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1438 static bool vm_ops_needs_writenotify(const struct vm_operations_struct
*vm_ops
)
1440 return vm_ops
&& (vm_ops
->page_mkwrite
|| vm_ops
->pfn_mkwrite
);
1443 static bool vma_is_shared_writable(struct vm_area_struct
*vma
)
1445 return (vma
->vm_flags
& (VM_WRITE
| VM_SHARED
)) ==
1446 (VM_WRITE
| VM_SHARED
);
1449 static bool vma_fs_can_writeback(struct vm_area_struct
*vma
)
1451 /* No managed pages to writeback. */
1452 if (vma
->vm_flags
& VM_PFNMAP
)
1455 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1456 mapping_can_writeback(vma
->vm_file
->f_mapping
);
1460 * Does this VMA require the underlying folios to have their dirty state
1463 bool vma_needs_dirty_tracking(struct vm_area_struct
*vma
)
1465 /* Only shared, writable VMAs require dirty tracking. */
1466 if (!vma_is_shared_writable(vma
))
1469 /* Does the filesystem need to be notified? */
1470 if (vm_ops_needs_writenotify(vma
->vm_ops
))
1474 * Even if the filesystem doesn't indicate a need for writenotify, if it
1475 * can writeback, dirty tracking is still required.
1477 return vma_fs_can_writeback(vma
);
1481 * Some shared mappings will want the pages marked read-only
1482 * to track write events. If so, we'll downgrade vm_page_prot
1483 * to the private version (using protection_map[] without the
1486 int vma_wants_writenotify(struct vm_area_struct
*vma
, pgprot_t vm_page_prot
)
1488 /* If it was private or non-writable, the write bit is already clear */
1489 if (!vma_is_shared_writable(vma
))
1492 /* The backer wishes to know when pages are first written to? */
1493 if (vm_ops_needs_writenotify(vma
->vm_ops
))
1496 /* The open routine did something to the protections that pgprot_modify
1497 * won't preserve? */
1498 if (pgprot_val(vm_page_prot
) !=
1499 pgprot_val(vm_pgprot_modify(vm_page_prot
, vma
->vm_flags
)))
1503 * Do we need to track softdirty? hugetlb does not support softdirty
1506 if (vma_soft_dirty_enabled(vma
) && !is_vm_hugetlb_page(vma
))
1509 /* Do we need write faults for uffd-wp tracking? */
1510 if (userfaultfd_wp(vma
))
1513 /* Can the mapping track the dirty pages? */
1514 return vma_fs_can_writeback(vma
);
1518 * We account for memory if it's a private writeable mapping,
1519 * not hugepages and VM_NORESERVE wasn't set.
1521 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1524 * hugetlb has its own accounting separate from the core VM
1525 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1527 if (file
&& is_file_hugepages(file
))
1530 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1534 * unmapped_area() - Find an area between the low_limit and the high_limit with
1535 * the correct alignment and offset, all from @info. Note: current->mm is used
1538 * @info: The unmapped area information including the range [low_limit -
1539 * high_limit), the alignment offset and mask.
1541 * Return: A memory address or -ENOMEM.
1543 static unsigned long unmapped_area(struct vm_unmapped_area_info
*info
)
1545 unsigned long length
, gap
;
1546 unsigned long low_limit
, high_limit
;
1547 struct vm_area_struct
*tmp
;
1549 MA_STATE(mas
, ¤t
->mm
->mm_mt
, 0, 0);
1551 /* Adjust search length to account for worst case alignment overhead */
1552 length
= info
->length
+ info
->align_mask
;
1553 if (length
< info
->length
)
1556 low_limit
= info
->low_limit
;
1557 if (low_limit
< mmap_min_addr
)
1558 low_limit
= mmap_min_addr
;
1559 high_limit
= info
->high_limit
;
1561 if (mas_empty_area(&mas
, low_limit
, high_limit
- 1, length
))
1565 gap
+= (info
->align_offset
- gap
) & info
->align_mask
;
1566 tmp
= mas_next(&mas
, ULONG_MAX
);
1567 if (tmp
&& (tmp
->vm_flags
& VM_STARTGAP_FLAGS
)) { /* Avoid prev check if possible */
1568 if (vm_start_gap(tmp
) < gap
+ length
- 1) {
1569 low_limit
= tmp
->vm_end
;
1574 tmp
= mas_prev(&mas
, 0);
1575 if (tmp
&& vm_end_gap(tmp
) > gap
) {
1576 low_limit
= vm_end_gap(tmp
);
1586 * unmapped_area_topdown() - Find an area between the low_limit and the
1587 * high_limit with the correct alignment and offset at the highest available
1588 * address, all from @info. Note: current->mm is used for the search.
1590 * @info: The unmapped area information including the range [low_limit -
1591 * high_limit), the alignment offset and mask.
1593 * Return: A memory address or -ENOMEM.
1595 static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info
*info
)
1597 unsigned long length
, gap
, gap_end
;
1598 unsigned long low_limit
, high_limit
;
1599 struct vm_area_struct
*tmp
;
1601 MA_STATE(mas
, ¤t
->mm
->mm_mt
, 0, 0);
1602 /* Adjust search length to account for worst case alignment overhead */
1603 length
= info
->length
+ info
->align_mask
;
1604 if (length
< info
->length
)
1607 low_limit
= info
->low_limit
;
1608 if (low_limit
< mmap_min_addr
)
1609 low_limit
= mmap_min_addr
;
1610 high_limit
= info
->high_limit
;
1612 if (mas_empty_area_rev(&mas
, low_limit
, high_limit
- 1, length
))
1615 gap
= mas
.last
+ 1 - info
->length
;
1616 gap
-= (gap
- info
->align_offset
) & info
->align_mask
;
1618 tmp
= mas_next(&mas
, ULONG_MAX
);
1619 if (tmp
&& (tmp
->vm_flags
& VM_STARTGAP_FLAGS
)) { /* Avoid prev check if possible */
1620 if (vm_start_gap(tmp
) <= gap_end
) {
1621 high_limit
= vm_start_gap(tmp
);
1626 tmp
= mas_prev(&mas
, 0);
1627 if (tmp
&& vm_end_gap(tmp
) > gap
) {
1628 high_limit
= tmp
->vm_start
;
1638 * Search for an unmapped address range.
1640 * We are looking for a range that:
1641 * - does not intersect with any VMA;
1642 * - is contained within the [low_limit, high_limit) interval;
1643 * - is at least the desired size.
1644 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
1646 unsigned long vm_unmapped_area(struct vm_unmapped_area_info
*info
)
1650 if (info
->flags
& VM_UNMAPPED_AREA_TOPDOWN
)
1651 addr
= unmapped_area_topdown(info
);
1653 addr
= unmapped_area(info
);
1655 trace_vm_unmapped_area(addr
, info
);
1659 /* Get an address range which is currently unmapped.
1660 * For shmat() with addr=0.
1662 * Ugly calling convention alert:
1663 * Return value with the low bits set means error value,
1665 * if (ret & ~PAGE_MASK)
1668 * This function "knows" that -ENOMEM has the bits set.
1671 generic_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1672 unsigned long len
, unsigned long pgoff
,
1673 unsigned long flags
)
1675 struct mm_struct
*mm
= current
->mm
;
1676 struct vm_area_struct
*vma
, *prev
;
1677 struct vm_unmapped_area_info info
;
1678 const unsigned long mmap_end
= arch_get_mmap_end(addr
, len
, flags
);
1680 if (len
> mmap_end
- mmap_min_addr
)
1683 if (flags
& MAP_FIXED
)
1687 addr
= PAGE_ALIGN(addr
);
1688 vma
= find_vma_prev(mm
, addr
, &prev
);
1689 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
1690 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1691 (!prev
|| addr
>= vm_end_gap(prev
)))
1697 info
.low_limit
= mm
->mmap_base
;
1698 info
.high_limit
= mmap_end
;
1699 info
.align_mask
= 0;
1700 info
.align_offset
= 0;
1701 return vm_unmapped_area(&info
);
1704 #ifndef HAVE_ARCH_UNMAPPED_AREA
1706 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1707 unsigned long len
, unsigned long pgoff
,
1708 unsigned long flags
)
1710 return generic_get_unmapped_area(filp
, addr
, len
, pgoff
, flags
);
1715 * This mmap-allocator allocates new areas top-down from below the
1716 * stack's low limit (the base):
1719 generic_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
1720 unsigned long len
, unsigned long pgoff
,
1721 unsigned long flags
)
1723 struct vm_area_struct
*vma
, *prev
;
1724 struct mm_struct
*mm
= current
->mm
;
1725 struct vm_unmapped_area_info info
;
1726 const unsigned long mmap_end
= arch_get_mmap_end(addr
, len
, flags
);
1728 /* requested length too big for entire address space */
1729 if (len
> mmap_end
- mmap_min_addr
)
1732 if (flags
& MAP_FIXED
)
1735 /* requesting a specific address */
1737 addr
= PAGE_ALIGN(addr
);
1738 vma
= find_vma_prev(mm
, addr
, &prev
);
1739 if (mmap_end
- len
>= addr
&& addr
>= mmap_min_addr
&&
1740 (!vma
|| addr
+ len
<= vm_start_gap(vma
)) &&
1741 (!prev
|| addr
>= vm_end_gap(prev
)))
1745 info
.flags
= VM_UNMAPPED_AREA_TOPDOWN
;
1747 info
.low_limit
= PAGE_SIZE
;
1748 info
.high_limit
= arch_get_mmap_base(addr
, mm
->mmap_base
);
1749 info
.align_mask
= 0;
1750 info
.align_offset
= 0;
1751 addr
= vm_unmapped_area(&info
);
1754 * A failed mmap() very likely causes application failure,
1755 * so fall back to the bottom-up function here. This scenario
1756 * can happen with large stack limits and large mmap()
1759 if (offset_in_page(addr
)) {
1760 VM_BUG_ON(addr
!= -ENOMEM
);
1762 info
.low_limit
= TASK_UNMAPPED_BASE
;
1763 info
.high_limit
= mmap_end
;
1764 addr
= vm_unmapped_area(&info
);
1770 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1772 arch_get_unmapped_area_topdown(struct file
*filp
, unsigned long addr
,
1773 unsigned long len
, unsigned long pgoff
,
1774 unsigned long flags
)
1776 return generic_get_unmapped_area_topdown(filp
, addr
, len
, pgoff
, flags
);
1781 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1782 unsigned long pgoff
, unsigned long flags
)
1784 unsigned long (*get_area
)(struct file
*, unsigned long,
1785 unsigned long, unsigned long, unsigned long);
1787 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1791 /* Careful about overflows.. */
1792 if (len
> TASK_SIZE
)
1795 get_area
= current
->mm
->get_unmapped_area
;
1797 if (file
->f_op
->get_unmapped_area
)
1798 get_area
= file
->f_op
->get_unmapped_area
;
1799 } else if (flags
& MAP_SHARED
) {
1801 * mmap_region() will call shmem_zero_setup() to create a file,
1802 * so use shmem's get_unmapped_area in case it can be huge.
1803 * do_mmap() will clear pgoff, so match alignment.
1806 get_area
= shmem_get_unmapped_area
;
1809 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1810 if (IS_ERR_VALUE(addr
))
1813 if (addr
> TASK_SIZE
- len
)
1815 if (offset_in_page(addr
))
1818 error
= security_mmap_addr(addr
);
1819 return error
? error
: addr
;
1822 EXPORT_SYMBOL(get_unmapped_area
);
1825 * find_vma_intersection() - Look up the first VMA which intersects the interval
1826 * @mm: The process address space.
1827 * @start_addr: The inclusive start user address.
1828 * @end_addr: The exclusive end user address.
1830 * Returns: The first VMA within the provided range, %NULL otherwise. Assumes
1831 * start_addr < end_addr.
1833 struct vm_area_struct
*find_vma_intersection(struct mm_struct
*mm
,
1834 unsigned long start_addr
,
1835 unsigned long end_addr
)
1837 unsigned long index
= start_addr
;
1839 mmap_assert_locked(mm
);
1840 return mt_find(&mm
->mm_mt
, &index
, end_addr
- 1);
1842 EXPORT_SYMBOL(find_vma_intersection
);
1845 * find_vma() - Find the VMA for a given address, or the next VMA.
1846 * @mm: The mm_struct to check
1847 * @addr: The address
1849 * Returns: The VMA associated with addr, or the next VMA.
1850 * May return %NULL in the case of no VMA at addr or above.
1852 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1854 unsigned long index
= addr
;
1856 mmap_assert_locked(mm
);
1857 return mt_find(&mm
->mm_mt
, &index
, ULONG_MAX
);
1859 EXPORT_SYMBOL(find_vma
);
1862 * find_vma_prev() - Find the VMA for a given address, or the next vma and
1863 * set %pprev to the previous VMA, if any.
1864 * @mm: The mm_struct to check
1865 * @addr: The address
1866 * @pprev: The pointer to set to the previous VMA
1868 * Note that RCU lock is missing here since the external mmap_lock() is used
1871 * Returns: The VMA associated with @addr, or the next vma.
1872 * May return %NULL in the case of no vma at addr or above.
1874 struct vm_area_struct
*
1875 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1876 struct vm_area_struct
**pprev
)
1878 struct vm_area_struct
*vma
;
1879 MA_STATE(mas
, &mm
->mm_mt
, addr
, addr
);
1881 vma
= mas_walk(&mas
);
1882 *pprev
= mas_prev(&mas
, 0);
1884 vma
= mas_next(&mas
, ULONG_MAX
);
1889 * Verify that the stack growth is acceptable and
1890 * update accounting. This is shared with both the
1891 * grow-up and grow-down cases.
1893 static int acct_stack_growth(struct vm_area_struct
*vma
,
1894 unsigned long size
, unsigned long grow
)
1896 struct mm_struct
*mm
= vma
->vm_mm
;
1897 unsigned long new_start
;
1899 /* address space limit tests */
1900 if (!may_expand_vm(mm
, vma
->vm_flags
, grow
))
1903 /* Stack limit test */
1904 if (size
> rlimit(RLIMIT_STACK
))
1907 /* mlock limit tests */
1908 if (!mlock_future_ok(mm
, vma
->vm_flags
, grow
<< PAGE_SHIFT
))
1911 /* Check to ensure the stack will not grow into a hugetlb-only region */
1912 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1914 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1918 * Overcommit.. This must be the final test, as it will
1919 * update security statistics.
1921 if (security_vm_enough_memory_mm(mm
, grow
))
1927 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1929 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1930 * vma is the last one with address > vma->vm_end. Have to extend vma.
1932 static int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1934 struct mm_struct
*mm
= vma
->vm_mm
;
1935 struct vm_area_struct
*next
;
1936 unsigned long gap_addr
;
1938 MA_STATE(mas
, &mm
->mm_mt
, vma
->vm_start
, address
);
1940 if (!(vma
->vm_flags
& VM_GROWSUP
))
1943 /* Guard against exceeding limits of the address space. */
1944 address
&= PAGE_MASK
;
1945 if (address
>= (TASK_SIZE
& PAGE_MASK
))
1947 address
+= PAGE_SIZE
;
1949 /* Enforce stack_guard_gap */
1950 gap_addr
= address
+ stack_guard_gap
;
1952 /* Guard against overflow */
1953 if (gap_addr
< address
|| gap_addr
> TASK_SIZE
)
1954 gap_addr
= TASK_SIZE
;
1956 next
= find_vma_intersection(mm
, vma
->vm_end
, gap_addr
);
1957 if (next
&& vma_is_accessible(next
)) {
1958 if (!(next
->vm_flags
& VM_GROWSUP
))
1960 /* Check that both stack segments have the same anon_vma? */
1964 mas_prev_range(&mas
, address
);
1966 __mas_set_range(&mas
, vma
->vm_start
, address
- 1);
1967 if (mas_preallocate(&mas
, vma
, GFP_KERNEL
))
1970 /* We must make sure the anon_vma is allocated. */
1971 if (unlikely(anon_vma_prepare(vma
))) {
1976 /* Lock the VMA before expanding to prevent concurrent page faults */
1977 vma_start_write(vma
);
1979 * vma->vm_start/vm_end cannot change under us because the caller
1980 * is required to hold the mmap_lock in read mode. We need the
1981 * anon_vma lock to serialize against concurrent expand_stacks.
1983 anon_vma_lock_write(vma
->anon_vma
);
1985 /* Somebody else might have raced and expanded it already */
1986 if (address
> vma
->vm_end
) {
1987 unsigned long size
, grow
;
1989 size
= address
- vma
->vm_start
;
1990 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1993 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1994 error
= acct_stack_growth(vma
, size
, grow
);
1997 * We only hold a shared mmap_lock lock here, so
1998 * we need to protect against concurrent vma
1999 * expansions. anon_vma_lock_write() doesn't
2000 * help here, as we don't guarantee that all
2001 * growable vmas in a mm share the same root
2002 * anon vma. So, we reuse mm->page_table_lock
2003 * to guard against concurrent vma expansions.
2005 spin_lock(&mm
->page_table_lock
);
2006 if (vma
->vm_flags
& VM_LOCKED
)
2007 mm
->locked_vm
+= grow
;
2008 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2009 anon_vma_interval_tree_pre_update_vma(vma
);
2010 vma
->vm_end
= address
;
2011 /* Overwrite old entry in mtree. */
2012 mas_store_prealloc(&mas
, vma
);
2013 anon_vma_interval_tree_post_update_vma(vma
);
2014 spin_unlock(&mm
->page_table_lock
);
2016 perf_event_mmap(vma
);
2020 anon_vma_unlock_write(vma
->anon_vma
);
2021 khugepaged_enter_vma(vma
, vma
->vm_flags
);
2026 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2029 * vma is the first one with address < vma->vm_start. Have to extend vma.
2030 * mmap_lock held for writing.
2032 int expand_downwards(struct vm_area_struct
*vma
, unsigned long address
)
2034 struct mm_struct
*mm
= vma
->vm_mm
;
2035 MA_STATE(mas
, &mm
->mm_mt
, vma
->vm_start
, vma
->vm_start
);
2036 struct vm_area_struct
*prev
;
2039 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
2042 address
&= PAGE_MASK
;
2043 if (address
< mmap_min_addr
|| address
< FIRST_USER_ADDRESS
)
2046 /* Enforce stack_guard_gap */
2047 prev
= mas_prev(&mas
, 0);
2048 /* Check that both stack segments have the same anon_vma? */
2050 if (!(prev
->vm_flags
& VM_GROWSDOWN
) &&
2051 vma_is_accessible(prev
) &&
2052 (address
- prev
->vm_end
< stack_guard_gap
))
2057 mas_next_range(&mas
, vma
->vm_start
);
2059 __mas_set_range(&mas
, address
, vma
->vm_end
- 1);
2060 if (mas_preallocate(&mas
, vma
, GFP_KERNEL
))
2063 /* We must make sure the anon_vma is allocated. */
2064 if (unlikely(anon_vma_prepare(vma
))) {
2069 /* Lock the VMA before expanding to prevent concurrent page faults */
2070 vma_start_write(vma
);
2072 * vma->vm_start/vm_end cannot change under us because the caller
2073 * is required to hold the mmap_lock in read mode. We need the
2074 * anon_vma lock to serialize against concurrent expand_stacks.
2076 anon_vma_lock_write(vma
->anon_vma
);
2078 /* Somebody else might have raced and expanded it already */
2079 if (address
< vma
->vm_start
) {
2080 unsigned long size
, grow
;
2082 size
= vma
->vm_end
- address
;
2083 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
2086 if (grow
<= vma
->vm_pgoff
) {
2087 error
= acct_stack_growth(vma
, size
, grow
);
2090 * We only hold a shared mmap_lock lock here, so
2091 * we need to protect against concurrent vma
2092 * expansions. anon_vma_lock_write() doesn't
2093 * help here, as we don't guarantee that all
2094 * growable vmas in a mm share the same root
2095 * anon vma. So, we reuse mm->page_table_lock
2096 * to guard against concurrent vma expansions.
2098 spin_lock(&mm
->page_table_lock
);
2099 if (vma
->vm_flags
& VM_LOCKED
)
2100 mm
->locked_vm
+= grow
;
2101 vm_stat_account(mm
, vma
->vm_flags
, grow
);
2102 anon_vma_interval_tree_pre_update_vma(vma
);
2103 vma
->vm_start
= address
;
2104 vma
->vm_pgoff
-= grow
;
2105 /* Overwrite old entry in mtree. */
2106 mas_store_prealloc(&mas
, vma
);
2107 anon_vma_interval_tree_post_update_vma(vma
);
2108 spin_unlock(&mm
->page_table_lock
);
2110 perf_event_mmap(vma
);
2114 anon_vma_unlock_write(vma
->anon_vma
);
2115 khugepaged_enter_vma(vma
, vma
->vm_flags
);
2121 /* enforced gap between the expanding stack and other mappings. */
2122 unsigned long stack_guard_gap
= 256UL<<PAGE_SHIFT
;
2124 static int __init
cmdline_parse_stack_guard_gap(char *p
)
2129 val
= simple_strtoul(p
, &endptr
, 10);
2131 stack_guard_gap
= val
<< PAGE_SHIFT
;
2135 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap
);
2137 #ifdef CONFIG_STACK_GROWSUP
2138 int expand_stack_locked(struct vm_area_struct
*vma
, unsigned long address
)
2140 return expand_upwards(vma
, address
);
2143 struct vm_area_struct
*find_extend_vma_locked(struct mm_struct
*mm
, unsigned long addr
)
2145 struct vm_area_struct
*vma
, *prev
;
2148 vma
= find_vma_prev(mm
, addr
, &prev
);
2149 if (vma
&& (vma
->vm_start
<= addr
))
2153 if (expand_stack_locked(prev
, addr
))
2155 if (prev
->vm_flags
& VM_LOCKED
)
2156 populate_vma_page_range(prev
, addr
, prev
->vm_end
, NULL
);
2160 int expand_stack_locked(struct vm_area_struct
*vma
, unsigned long address
)
2162 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
)))
2164 return expand_downwards(vma
, address
);
2167 struct vm_area_struct
*find_extend_vma_locked(struct mm_struct
*mm
, unsigned long addr
)
2169 struct vm_area_struct
*vma
;
2170 unsigned long start
;
2173 vma
= find_vma(mm
, addr
);
2176 if (vma
->vm_start
<= addr
)
2178 start
= vma
->vm_start
;
2179 if (expand_stack_locked(vma
, addr
))
2181 if (vma
->vm_flags
& VM_LOCKED
)
2182 populate_vma_page_range(vma
, addr
, start
, NULL
);
2188 * IA64 has some horrid mapping rules: it can expand both up and down,
2189 * but with various special rules.
2191 * We'll get rid of this architecture eventually, so the ugliness is
2195 static inline bool vma_expand_ok(struct vm_area_struct
*vma
, unsigned long addr
)
2197 return REGION_NUMBER(addr
) == REGION_NUMBER(vma
->vm_start
) &&
2198 REGION_OFFSET(addr
) < RGN_MAP_LIMIT
;
2202 * IA64 stacks grow down, but there's a special register backing store
2203 * that can grow up. Only sequentially, though, so the new address must
2206 static inline int vma_expand_up(struct vm_area_struct
*vma
, unsigned long addr
)
2208 if (!vma_expand_ok(vma
, addr
))
2210 if (vma
->vm_end
!= (addr
& PAGE_MASK
))
2212 return expand_upwards(vma
, addr
);
2215 static inline bool vma_expand_down(struct vm_area_struct
*vma
, unsigned long addr
)
2217 if (!vma_expand_ok(vma
, addr
))
2219 return expand_downwards(vma
, addr
);
2222 #elif defined(CONFIG_STACK_GROWSUP)
2224 #define vma_expand_up(vma,addr) expand_upwards(vma, addr)
2225 #define vma_expand_down(vma, addr) (-EFAULT)
2229 #define vma_expand_up(vma,addr) (-EFAULT)
2230 #define vma_expand_down(vma, addr) expand_downwards(vma, addr)
2235 * expand_stack(): legacy interface for page faulting. Don't use unless
2238 * This is called with the mm locked for reading, drops the lock, takes
2239 * the lock for writing, tries to look up a vma again, expands it if
2240 * necessary, and downgrades the lock to reading again.
2242 * If no vma is found or it can't be expanded, it returns NULL and has
2245 struct vm_area_struct
*expand_stack(struct mm_struct
*mm
, unsigned long addr
)
2247 struct vm_area_struct
*vma
, *prev
;
2249 mmap_read_unlock(mm
);
2250 if (mmap_write_lock_killable(mm
))
2253 vma
= find_vma_prev(mm
, addr
, &prev
);
2254 if (vma
&& vma
->vm_start
<= addr
)
2257 if (prev
&& !vma_expand_up(prev
, addr
)) {
2262 if (vma
&& !vma_expand_down(vma
, addr
))
2265 mmap_write_unlock(mm
);
2269 mmap_write_downgrade(mm
);
2274 * Ok - we have the memory areas we should free on a maple tree so release them,
2275 * and do the vma updates.
2277 * Called with the mm semaphore held.
2279 static inline void remove_mt(struct mm_struct
*mm
, struct ma_state
*mas
)
2281 unsigned long nr_accounted
= 0;
2282 struct vm_area_struct
*vma
;
2284 /* Update high watermark before we lower total_vm */
2285 update_hiwater_vm(mm
);
2286 mas_for_each(mas
, vma
, ULONG_MAX
) {
2287 long nrpages
= vma_pages(vma
);
2289 if (vma
->vm_flags
& VM_ACCOUNT
)
2290 nr_accounted
+= nrpages
;
2291 vm_stat_account(mm
, vma
->vm_flags
, -nrpages
);
2292 remove_vma(vma
, false);
2294 vm_unacct_memory(nr_accounted
);
2298 * Get rid of page table information in the indicated region.
2300 * Called with the mm semaphore held.
2302 static void unmap_region(struct mm_struct
*mm
, struct ma_state
*mas
,
2303 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
2304 struct vm_area_struct
*next
, unsigned long start
,
2305 unsigned long end
, unsigned long tree_end
, bool mm_wr_locked
)
2307 struct mmu_gather tlb
;
2308 unsigned long mt_start
= mas
->index
;
2311 tlb_gather_mmu(&tlb
, mm
);
2312 update_hiwater_rss(mm
);
2313 unmap_vmas(&tlb
, mas
, vma
, start
, end
, tree_end
, mm_wr_locked
);
2314 mas_set(mas
, mt_start
);
2315 free_pgtables(&tlb
, mas
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
2316 next
? next
->vm_start
: USER_PGTABLES_CEILING
,
2318 tlb_finish_mmu(&tlb
);
2322 * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2323 * has already been checked or doesn't make sense to fail.
2324 * VMA Iterator will point to the end VMA.
2326 int __split_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2327 unsigned long addr
, int new_below
)
2329 struct vma_prepare vp
;
2330 struct vm_area_struct
*new;
2333 WARN_ON(vma
->vm_start
>= addr
);
2334 WARN_ON(vma
->vm_end
<= addr
);
2336 if (vma
->vm_ops
&& vma
->vm_ops
->may_split
) {
2337 err
= vma
->vm_ops
->may_split(vma
, addr
);
2342 new = vm_area_dup(vma
);
2349 new->vm_start
= addr
;
2350 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2354 vma_iter_config(vmi
, new->vm_start
, new->vm_end
);
2355 if (vma_iter_prealloc(vmi
, new))
2358 err
= vma_dup_policy(vma
, new);
2362 err
= anon_vma_clone(new, vma
);
2367 get_file(new->vm_file
);
2369 if (new->vm_ops
&& new->vm_ops
->open
)
2370 new->vm_ops
->open(new);
2372 vma_start_write(vma
);
2373 vma_start_write(new);
2375 init_vma_prep(&vp
, vma
);
2378 vma_adjust_trans_huge(vma
, vma
->vm_start
, addr
, 0);
2381 vma
->vm_start
= addr
;
2382 vma
->vm_pgoff
+= (addr
- new->vm_start
) >> PAGE_SHIFT
;
2387 /* vma_complete stores the new vma */
2388 vma_complete(&vp
, vmi
, vma
->vm_mm
);
2396 mpol_put(vma_policy(new));
2405 * Split a vma into two pieces at address 'addr', a new vma is allocated
2406 * either for the first part or the tail.
2408 int split_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2409 unsigned long addr
, int new_below
)
2411 if (vma
->vm_mm
->map_count
>= sysctl_max_map_count
)
2414 return __split_vma(vmi
, vma
, addr
, new_below
);
2418 * do_vmi_align_munmap() - munmap the aligned region from @start to @end.
2419 * @vmi: The vma iterator
2420 * @vma: The starting vm_area_struct
2421 * @mm: The mm_struct
2422 * @start: The aligned start address to munmap.
2423 * @end: The aligned end address to munmap.
2424 * @uf: The userfaultfd list_head
2425 * @unlock: Set to true to drop the mmap_lock. unlocking only happens on
2428 * Return: 0 on success and drops the lock if so directed, error and leaves the
2429 * lock held otherwise.
2432 do_vmi_align_munmap(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
2433 struct mm_struct
*mm
, unsigned long start
,
2434 unsigned long end
, struct list_head
*uf
, bool unlock
)
2436 struct vm_area_struct
*prev
, *next
= NULL
;
2437 struct maple_tree mt_detach
;
2439 int error
= -ENOMEM
;
2440 unsigned long locked_vm
= 0;
2441 MA_STATE(mas_detach
, &mt_detach
, 0, 0);
2442 mt_init_flags(&mt_detach
, vmi
->mas
.tree
->ma_flags
& MT_FLAGS_LOCK_MASK
);
2443 mt_on_stack(mt_detach
);
2446 * If we need to split any vma, do it now to save pain later.
2448 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2449 * unmapped vm_area_struct will remain in use: so lower split_vma
2450 * places tmp vma above, and higher split_vma places tmp vma below.
2453 /* Does it split the first one? */
2454 if (start
> vma
->vm_start
) {
2457 * Make sure that map_count on return from munmap() will
2458 * not exceed its limit; but let map_count go just above
2459 * its limit temporarily, to help free resources as expected.
2461 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2462 goto map_count_exceeded
;
2464 error
= __split_vma(vmi
, vma
, start
, 1);
2466 goto start_split_failed
;
2470 * Detach a range of VMAs from the mm. Using next as a temp variable as
2471 * it is always overwritten.
2475 /* Does it split the end? */
2476 if (next
->vm_end
> end
) {
2477 error
= __split_vma(vmi
, next
, end
, 0);
2479 goto end_split_failed
;
2481 vma_start_write(next
);
2482 mas_set(&mas_detach
, count
);
2483 error
= mas_store_gfp(&mas_detach
, next
, GFP_KERNEL
);
2485 goto munmap_gather_failed
;
2486 vma_mark_detached(next
, true);
2487 if (next
->vm_flags
& VM_LOCKED
)
2488 locked_vm
+= vma_pages(next
);
2493 * If userfaultfd_unmap_prep returns an error the vmas
2494 * will remain split, but userland will get a
2495 * highly unexpected error anyway. This is no
2496 * different than the case where the first of the two
2497 * __split_vma fails, but we don't undo the first
2498 * split, despite we could. This is unlikely enough
2499 * failure that it's not worth optimizing it for.
2501 error
= userfaultfd_unmap_prep(next
, start
, end
, uf
);
2504 goto userfaultfd_error
;
2506 #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
2507 BUG_ON(next
->vm_start
< start
);
2508 BUG_ON(next
->vm_start
> end
);
2510 } for_each_vma_range(*vmi
, next
, end
);
2512 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
2513 /* Make sure no VMAs are about to be lost. */
2515 MA_STATE(test
, &mt_detach
, 0, 0);
2516 struct vm_area_struct
*vma_mas
, *vma_test
;
2519 vma_iter_set(vmi
, start
);
2521 vma_test
= mas_find(&test
, count
- 1);
2522 for_each_vma_range(*vmi
, vma_mas
, end
) {
2523 BUG_ON(vma_mas
!= vma_test
);
2525 vma_test
= mas_next(&test
, count
- 1);
2528 BUG_ON(count
!= test_count
);
2532 while (vma_iter_addr(vmi
) > start
)
2533 vma_iter_prev_range(vmi
);
2535 error
= vma_iter_clear_gfp(vmi
, start
, end
, GFP_KERNEL
);
2537 goto clear_tree_failed
;
2539 /* Point of no return */
2540 mm
->locked_vm
-= locked_vm
;
2541 mm
->map_count
-= count
;
2543 mmap_write_downgrade(mm
);
2545 prev
= vma_iter_prev_range(vmi
);
2546 next
= vma_next(vmi
);
2548 vma_iter_prev_range(vmi
);
2551 * We can free page tables without write-locking mmap_lock because VMAs
2552 * were isolated before we downgraded mmap_lock.
2554 mas_set(&mas_detach
, 1);
2555 unmap_region(mm
, &mas_detach
, vma
, prev
, next
, start
, end
, count
,
2557 /* Statistics and freeing VMAs */
2558 mas_set(&mas_detach
, 0);
2559 remove_mt(mm
, &mas_detach
);
2562 mmap_read_unlock(mm
);
2564 __mt_destroy(&mt_detach
);
2569 munmap_gather_failed
:
2571 mas_set(&mas_detach
, 0);
2572 mas_for_each(&mas_detach
, next
, end
)
2573 vma_mark_detached(next
, false);
2575 __mt_destroy(&mt_detach
);
2583 * do_vmi_munmap() - munmap a given range.
2584 * @vmi: The vma iterator
2585 * @mm: The mm_struct
2586 * @start: The start address to munmap
2587 * @len: The length of the range to munmap
2588 * @uf: The userfaultfd list_head
2589 * @unlock: set to true if the user wants to drop the mmap_lock on success
2591 * This function takes a @mas that is either pointing to the previous VMA or set
2592 * to MA_START and sets it up to remove the mapping(s). The @len will be
2593 * aligned and any arch_unmap work will be preformed.
2595 * Return: 0 on success and drops the lock if so directed, error and leaves the
2596 * lock held otherwise.
2598 int do_vmi_munmap(struct vma_iterator
*vmi
, struct mm_struct
*mm
,
2599 unsigned long start
, size_t len
, struct list_head
*uf
,
2603 struct vm_area_struct
*vma
;
2605 if ((offset_in_page(start
)) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2608 end
= start
+ PAGE_ALIGN(len
);
2612 /* arch_unmap() might do unmaps itself. */
2613 arch_unmap(mm
, start
, end
);
2615 /* Find the first overlapping VMA */
2616 vma
= vma_find(vmi
, end
);
2619 mmap_write_unlock(mm
);
2623 return do_vmi_align_munmap(vmi
, vma
, mm
, start
, end
, uf
, unlock
);
2626 /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
2627 * @mm: The mm_struct
2628 * @start: The start address to munmap
2629 * @len: The length to be munmapped.
2630 * @uf: The userfaultfd list_head
2632 * Return: 0 on success, error otherwise.
2634 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
,
2635 struct list_head
*uf
)
2637 VMA_ITERATOR(vmi
, mm
, start
);
2639 return do_vmi_munmap(&vmi
, mm
, start
, len
, uf
, false);
2642 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
2643 unsigned long len
, vm_flags_t vm_flags
, unsigned long pgoff
,
2644 struct list_head
*uf
)
2646 struct mm_struct
*mm
= current
->mm
;
2647 struct vm_area_struct
*vma
= NULL
;
2648 struct vm_area_struct
*next
, *prev
, *merge
;
2649 pgoff_t pglen
= len
>> PAGE_SHIFT
;
2650 unsigned long charged
= 0;
2651 unsigned long end
= addr
+ len
;
2652 unsigned long merge_start
= addr
, merge_end
= end
;
2655 VMA_ITERATOR(vmi
, mm
, addr
);
2657 /* Check against address space limit. */
2658 if (!may_expand_vm(mm
, vm_flags
, len
>> PAGE_SHIFT
)) {
2659 unsigned long nr_pages
;
2662 * MAP_FIXED may remove pages of mappings that intersects with
2663 * requested mapping. Account for the pages it would unmap.
2665 nr_pages
= count_vma_pages_range(mm
, addr
, end
);
2667 if (!may_expand_vm(mm
, vm_flags
,
2668 (len
>> PAGE_SHIFT
) - nr_pages
))
2672 /* Unmap any existing mapping in the area */
2673 if (do_vmi_munmap(&vmi
, mm
, addr
, len
, uf
, false))
2677 * Private writable mapping: check memory availability
2679 if (accountable_mapping(file
, vm_flags
)) {
2680 charged
= len
>> PAGE_SHIFT
;
2681 if (security_vm_enough_memory_mm(mm
, charged
))
2683 vm_flags
|= VM_ACCOUNT
;
2686 next
= vma_next(&vmi
);
2687 prev
= vma_prev(&vmi
);
2688 if (vm_flags
& VM_SPECIAL
) {
2690 vma_iter_next_range(&vmi
);
2694 /* Attempt to expand an old mapping */
2696 if (next
&& next
->vm_start
== end
&& !vma_policy(next
) &&
2697 can_vma_merge_before(next
, vm_flags
, NULL
, file
, pgoff
+pglen
,
2698 NULL_VM_UFFD_CTX
, NULL
)) {
2699 merge_end
= next
->vm_end
;
2701 vm_pgoff
= next
->vm_pgoff
- pglen
;
2705 if (prev
&& prev
->vm_end
== addr
&& !vma_policy(prev
) &&
2706 (vma
? can_vma_merge_after(prev
, vm_flags
, vma
->anon_vma
, file
,
2707 pgoff
, vma
->vm_userfaultfd_ctx
, NULL
) :
2708 can_vma_merge_after(prev
, vm_flags
, NULL
, file
, pgoff
,
2709 NULL_VM_UFFD_CTX
, NULL
))) {
2710 merge_start
= prev
->vm_start
;
2712 vm_pgoff
= prev
->vm_pgoff
;
2714 vma_iter_next_range(&vmi
);
2717 /* Actually expand, if possible */
2719 !vma_expand(&vmi
, vma
, merge_start
, merge_end
, vm_pgoff
, next
)) {
2720 khugepaged_enter_vma(vma
, vm_flags
);
2725 vma_iter_set(&vmi
, addr
);
2729 * Determine the object being mapped and call the appropriate
2730 * specific mapper. the address has already been validated, but
2731 * not unmapped, but the maps are removed from the list.
2733 vma
= vm_area_alloc(mm
);
2739 vma_iter_config(&vmi
, addr
, end
);
2740 vma
->vm_start
= addr
;
2742 vm_flags_init(vma
, vm_flags
);
2743 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
2744 vma
->vm_pgoff
= pgoff
;
2747 if (vm_flags
& VM_SHARED
) {
2748 error
= mapping_map_writable(file
->f_mapping
);
2753 vma
->vm_file
= get_file(file
);
2754 error
= call_mmap(file
, vma
);
2756 goto unmap_and_free_vma
;
2759 * Expansion is handled above, merging is handled below.
2760 * Drivers should not alter the address of the VMA.
2763 if (WARN_ON((addr
!= vma
->vm_start
)))
2764 goto close_and_free_vma
;
2766 vma_iter_config(&vmi
, addr
, end
);
2768 * If vm_flags changed after call_mmap(), we should try merge
2769 * vma again as we may succeed this time.
2771 if (unlikely(vm_flags
!= vma
->vm_flags
&& prev
)) {
2772 merge
= vma_merge(&vmi
, mm
, prev
, vma
->vm_start
,
2773 vma
->vm_end
, vma
->vm_flags
, NULL
,
2774 vma
->vm_file
, vma
->vm_pgoff
, NULL
,
2775 NULL_VM_UFFD_CTX
, NULL
);
2778 * ->mmap() can change vma->vm_file and fput
2779 * the original file. So fput the vma->vm_file
2780 * here or we would add an extra fput for file
2781 * and cause general protection fault
2787 /* Update vm_flags to pick up the change. */
2788 vm_flags
= vma
->vm_flags
;
2789 goto unmap_writable
;
2793 vm_flags
= vma
->vm_flags
;
2794 } else if (vm_flags
& VM_SHARED
) {
2795 error
= shmem_zero_setup(vma
);
2799 vma_set_anonymous(vma
);
2802 if (map_deny_write_exec(vma
, vma
->vm_flags
)) {
2804 goto close_and_free_vma
;
2807 /* Allow architectures to sanity-check the vm_flags */
2809 if (!arch_validate_flags(vma
->vm_flags
))
2810 goto close_and_free_vma
;
2813 if (vma_iter_prealloc(&vmi
, vma
))
2814 goto close_and_free_vma
;
2816 /* Lock the VMA since it is modified after insertion into VMA tree */
2817 vma_start_write(vma
);
2818 vma_iter_store(&vmi
, vma
);
2821 i_mmap_lock_write(vma
->vm_file
->f_mapping
);
2822 if (vma
->vm_flags
& VM_SHARED
)
2823 mapping_allow_writable(vma
->vm_file
->f_mapping
);
2825 flush_dcache_mmap_lock(vma
->vm_file
->f_mapping
);
2826 vma_interval_tree_insert(vma
, &vma
->vm_file
->f_mapping
->i_mmap
);
2827 flush_dcache_mmap_unlock(vma
->vm_file
->f_mapping
);
2828 i_mmap_unlock_write(vma
->vm_file
->f_mapping
);
2832 * vma_merge() calls khugepaged_enter_vma() either, the below
2833 * call covers the non-merge case.
2835 khugepaged_enter_vma(vma
, vma
->vm_flags
);
2837 /* Once vma denies write, undo our temporary denial count */
2839 if (file
&& vm_flags
& VM_SHARED
)
2840 mapping_unmap_writable(file
->f_mapping
);
2841 file
= vma
->vm_file
;
2844 perf_event_mmap(vma
);
2846 vm_stat_account(mm
, vm_flags
, len
>> PAGE_SHIFT
);
2847 if (vm_flags
& VM_LOCKED
) {
2848 if ((vm_flags
& VM_SPECIAL
) || vma_is_dax(vma
) ||
2849 is_vm_hugetlb_page(vma
) ||
2850 vma
== get_gate_vma(current
->mm
))
2851 vm_flags_clear(vma
, VM_LOCKED_MASK
);
2853 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2860 * New (or expanded) vma always get soft dirty status.
2861 * Otherwise user-space soft-dirty page tracker won't
2862 * be able to distinguish situation when vma area unmapped,
2863 * then new mapped in-place (which must be aimed as
2864 * a completely new data area).
2866 vm_flags_set(vma
, VM_SOFTDIRTY
);
2868 vma_set_page_prot(vma
);
2874 if (file
&& vma
->vm_ops
&& vma
->vm_ops
->close
)
2875 vma
->vm_ops
->close(vma
);
2877 if (file
|| vma
->vm_file
) {
2880 vma
->vm_file
= NULL
;
2882 vma_iter_set(&vmi
, vma
->vm_end
);
2883 /* Undo any partial mapping done by a device driver. */
2884 unmap_region(mm
, &vmi
.mas
, vma
, prev
, next
, vma
->vm_start
,
2885 vma
->vm_end
, vma
->vm_end
, true);
2887 if (file
&& (vm_flags
& VM_SHARED
))
2888 mapping_unmap_writable(file
->f_mapping
);
2893 vm_unacct_memory(charged
);
2898 static int __vm_munmap(unsigned long start
, size_t len
, bool unlock
)
2901 struct mm_struct
*mm
= current
->mm
;
2903 VMA_ITERATOR(vmi
, mm
, start
);
2905 if (mmap_write_lock_killable(mm
))
2908 ret
= do_vmi_munmap(&vmi
, mm
, start
, len
, &uf
, unlock
);
2910 mmap_write_unlock(mm
);
2912 userfaultfd_unmap_complete(mm
, &uf
);
2916 int vm_munmap(unsigned long start
, size_t len
)
2918 return __vm_munmap(start
, len
, false);
2920 EXPORT_SYMBOL(vm_munmap
);
2922 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2924 addr
= untagged_addr(addr
);
2925 return __vm_munmap(addr
, len
, true);
2930 * Emulation of deprecated remap_file_pages() syscall.
2932 SYSCALL_DEFINE5(remap_file_pages
, unsigned long, start
, unsigned long, size
,
2933 unsigned long, prot
, unsigned long, pgoff
, unsigned long, flags
)
2936 struct mm_struct
*mm
= current
->mm
;
2937 struct vm_area_struct
*vma
;
2938 unsigned long populate
= 0;
2939 unsigned long ret
= -EINVAL
;
2942 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
2943 current
->comm
, current
->pid
);
2947 start
= start
& PAGE_MASK
;
2948 size
= size
& PAGE_MASK
;
2950 if (start
+ size
<= start
)
2953 /* Does pgoff wrap? */
2954 if (pgoff
+ (size
>> PAGE_SHIFT
) < pgoff
)
2957 if (mmap_write_lock_killable(mm
))
2960 vma
= vma_lookup(mm
, start
);
2962 if (!vma
|| !(vma
->vm_flags
& VM_SHARED
))
2965 if (start
+ size
> vma
->vm_end
) {
2966 VMA_ITERATOR(vmi
, mm
, vma
->vm_end
);
2967 struct vm_area_struct
*next
, *prev
= vma
;
2969 for_each_vma_range(vmi
, next
, start
+ size
) {
2970 /* hole between vmas ? */
2971 if (next
->vm_start
!= prev
->vm_end
)
2974 if (next
->vm_file
!= vma
->vm_file
)
2977 if (next
->vm_flags
!= vma
->vm_flags
)
2980 if (start
+ size
<= next
->vm_end
)
2990 prot
|= vma
->vm_flags
& VM_READ
? PROT_READ
: 0;
2991 prot
|= vma
->vm_flags
& VM_WRITE
? PROT_WRITE
: 0;
2992 prot
|= vma
->vm_flags
& VM_EXEC
? PROT_EXEC
: 0;
2994 flags
&= MAP_NONBLOCK
;
2995 flags
|= MAP_SHARED
| MAP_FIXED
| MAP_POPULATE
;
2996 if (vma
->vm_flags
& VM_LOCKED
)
2997 flags
|= MAP_LOCKED
;
2999 file
= get_file(vma
->vm_file
);
3000 ret
= do_mmap(vma
->vm_file
, start
, size
,
3001 prot
, flags
, 0, pgoff
, &populate
, NULL
);
3004 mmap_write_unlock(mm
);
3006 mm_populate(ret
, populate
);
3007 if (!IS_ERR_VALUE(ret
))
3013 * do_vma_munmap() - Unmap a full or partial vma.
3014 * @vmi: The vma iterator pointing at the vma
3015 * @vma: The first vma to be munmapped
3016 * @start: the start of the address to unmap
3017 * @end: The end of the address to unmap
3018 * @uf: The userfaultfd list_head
3019 * @unlock: Drop the lock on success
3021 * unmaps a VMA mapping when the vma iterator is already in position.
3022 * Does not handle alignment.
3024 * Return: 0 on success drops the lock of so directed, error on failure and will
3025 * still hold the lock.
3027 int do_vma_munmap(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
3028 unsigned long start
, unsigned long end
, struct list_head
*uf
,
3031 struct mm_struct
*mm
= vma
->vm_mm
;
3033 arch_unmap(mm
, start
, end
);
3034 return do_vmi_align_munmap(vmi
, vma
, mm
, start
, end
, uf
, unlock
);
3038 * do_brk_flags() - Increase the brk vma if the flags match.
3039 * @vmi: The vma iterator
3040 * @addr: The start address
3041 * @len: The length of the increase
3043 * @flags: The VMA Flags
3045 * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags
3046 * do not match then create a new anonymous VMA. Eventually we may be able to
3047 * do some brk-specific accounting here.
3049 static int do_brk_flags(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
3050 unsigned long addr
, unsigned long len
, unsigned long flags
)
3052 struct mm_struct
*mm
= current
->mm
;
3053 struct vma_prepare vp
;
3056 * Check against address space limits by the changed size
3057 * Note: This happens *after* clearing old mappings in some code paths.
3059 flags
|= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
3060 if (!may_expand_vm(mm
, flags
, len
>> PAGE_SHIFT
))
3063 if (mm
->map_count
> sysctl_max_map_count
)
3066 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
3070 * Expand the existing vma if possible; Note that singular lists do not
3071 * occur after forking, so the expand will only happen on new VMAs.
3073 if (vma
&& vma
->vm_end
== addr
&& !vma_policy(vma
) &&
3074 can_vma_merge_after(vma
, flags
, NULL
, NULL
,
3075 addr
>> PAGE_SHIFT
, NULL_VM_UFFD_CTX
, NULL
)) {
3076 vma_iter_config(vmi
, vma
->vm_start
, addr
+ len
);
3077 if (vma_iter_prealloc(vmi
, vma
))
3080 vma_start_write(vma
);
3082 init_vma_prep(&vp
, vma
);
3084 vma_adjust_trans_huge(vma
, vma
->vm_start
, addr
+ len
, 0);
3085 vma
->vm_end
= addr
+ len
;
3086 vm_flags_set(vma
, VM_SOFTDIRTY
);
3087 vma_iter_store(vmi
, vma
);
3089 vma_complete(&vp
, vmi
, mm
);
3090 khugepaged_enter_vma(vma
, flags
);
3095 vma_iter_next_range(vmi
);
3096 /* create a vma struct for an anonymous mapping */
3097 vma
= vm_area_alloc(mm
);
3101 vma_set_anonymous(vma
);
3102 vma
->vm_start
= addr
;
3103 vma
->vm_end
= addr
+ len
;
3104 vma
->vm_pgoff
= addr
>> PAGE_SHIFT
;
3105 vm_flags_init(vma
, flags
);
3106 vma
->vm_page_prot
= vm_get_page_prot(flags
);
3107 vma_start_write(vma
);
3108 if (vma_iter_store_gfp(vmi
, vma
, GFP_KERNEL
))
3109 goto mas_store_fail
;
3115 perf_event_mmap(vma
);
3116 mm
->total_vm
+= len
>> PAGE_SHIFT
;
3117 mm
->data_vm
+= len
>> PAGE_SHIFT
;
3118 if (flags
& VM_LOCKED
)
3119 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
3120 vm_flags_set(vma
, VM_SOFTDIRTY
);
3126 vm_unacct_memory(len
>> PAGE_SHIFT
);
3130 int vm_brk_flags(unsigned long addr
, unsigned long request
, unsigned long flags
)
3132 struct mm_struct
*mm
= current
->mm
;
3133 struct vm_area_struct
*vma
= NULL
;
3138 VMA_ITERATOR(vmi
, mm
, addr
);
3140 len
= PAGE_ALIGN(request
);
3146 if (mmap_write_lock_killable(mm
))
3149 /* Until we need other flags, refuse anything except VM_EXEC. */
3150 if ((flags
& (~VM_EXEC
)) != 0)
3153 ret
= check_brk_limits(addr
, len
);
3157 ret
= do_vmi_munmap(&vmi
, mm
, addr
, len
, &uf
, 0);
3161 vma
= vma_prev(&vmi
);
3162 ret
= do_brk_flags(&vmi
, vma
, addr
, len
, flags
);
3163 populate
= ((mm
->def_flags
& VM_LOCKED
) != 0);
3164 mmap_write_unlock(mm
);
3165 userfaultfd_unmap_complete(mm
, &uf
);
3166 if (populate
&& !ret
)
3167 mm_populate(addr
, len
);
3172 mmap_write_unlock(mm
);
3175 EXPORT_SYMBOL(vm_brk_flags
);
3177 int vm_brk(unsigned long addr
, unsigned long len
)
3179 return vm_brk_flags(addr
, len
, 0);
3181 EXPORT_SYMBOL(vm_brk
);
3183 /* Release all mmaps. */
3184 void exit_mmap(struct mm_struct
*mm
)
3186 struct mmu_gather tlb
;
3187 struct vm_area_struct
*vma
;
3188 unsigned long nr_accounted
= 0;
3189 MA_STATE(mas
, &mm
->mm_mt
, 0, 0);
3192 /* mm's last user has gone, and its about to be pulled down */
3193 mmu_notifier_release(mm
);
3198 vma
= mas_find(&mas
, ULONG_MAX
);
3200 /* Can happen if dup_mmap() received an OOM */
3201 mmap_read_unlock(mm
);
3207 tlb_gather_mmu_fullmm(&tlb
, mm
);
3208 /* update_hiwater_rss(mm) here? but nobody should be looking */
3209 /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
3210 unmap_vmas(&tlb
, &mas
, vma
, 0, ULONG_MAX
, ULONG_MAX
, false);
3211 mmap_read_unlock(mm
);
3214 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
3215 * because the memory has been already freed.
3217 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
3218 mmap_write_lock(mm
);
3219 mt_clear_in_rcu(&mm
->mm_mt
);
3220 mas_set(&mas
, vma
->vm_end
);
3221 free_pgtables(&tlb
, &mas
, vma
, FIRST_USER_ADDRESS
,
3222 USER_PGTABLES_CEILING
, true);
3223 tlb_finish_mmu(&tlb
);
3226 * Walk the list again, actually closing and freeing it, with preemption
3227 * enabled, without holding any MM locks besides the unreachable
3230 mas_set(&mas
, vma
->vm_end
);
3232 if (vma
->vm_flags
& VM_ACCOUNT
)
3233 nr_accounted
+= vma_pages(vma
);
3234 remove_vma(vma
, true);
3237 } while ((vma
= mas_find(&mas
, ULONG_MAX
)) != NULL
);
3239 BUG_ON(count
!= mm
->map_count
);
3241 trace_exit_mmap(mm
);
3242 __mt_destroy(&mm
->mm_mt
);
3243 mmap_write_unlock(mm
);
3244 vm_unacct_memory(nr_accounted
);
3247 /* Insert vm structure into process list sorted by address
3248 * and into the inode's i_mmap tree. If vm_file is non-NULL
3249 * then i_mmap_rwsem is taken here.
3251 int insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
3253 unsigned long charged
= vma_pages(vma
);
3256 if (find_vma_intersection(mm
, vma
->vm_start
, vma
->vm_end
))
3259 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
3260 security_vm_enough_memory_mm(mm
, charged
))
3264 * The vm_pgoff of a purely anonymous vma should be irrelevant
3265 * until its first write fault, when page's anon_vma and index
3266 * are set. But now set the vm_pgoff it will almost certainly
3267 * end up with (unless mremap moves it elsewhere before that
3268 * first wfault), so /proc/pid/maps tells a consistent story.
3270 * By setting it to reflect the virtual start address of the
3271 * vma, merges and splits can happen in a seamless way, just
3272 * using the existing file pgoff checks and manipulations.
3273 * Similarly in do_mmap and in do_brk_flags.
3275 if (vma_is_anonymous(vma
)) {
3276 BUG_ON(vma
->anon_vma
);
3277 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
3280 if (vma_link(mm
, vma
)) {
3281 vm_unacct_memory(charged
);
3289 * Copy the vma structure to a new location in the same mm,
3290 * prior to moving page table entries, to effect an mremap move.
3292 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
3293 unsigned long addr
, unsigned long len
, pgoff_t pgoff
,
3294 bool *need_rmap_locks
)
3296 struct vm_area_struct
*vma
= *vmap
;
3297 unsigned long vma_start
= vma
->vm_start
;
3298 struct mm_struct
*mm
= vma
->vm_mm
;
3299 struct vm_area_struct
*new_vma
, *prev
;
3300 bool faulted_in_anon_vma
= true;
3301 VMA_ITERATOR(vmi
, mm
, addr
);
3304 * If anonymous vma has not yet been faulted, update new pgoff
3305 * to match new location, to increase its chance of merging.
3307 if (unlikely(vma_is_anonymous(vma
) && !vma
->anon_vma
)) {
3308 pgoff
= addr
>> PAGE_SHIFT
;
3309 faulted_in_anon_vma
= false;
3312 new_vma
= find_vma_prev(mm
, addr
, &prev
);
3313 if (new_vma
&& new_vma
->vm_start
< addr
+ len
)
3314 return NULL
; /* should never get here */
3316 new_vma
= vma_merge(&vmi
, mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
3317 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
),
3318 vma
->vm_userfaultfd_ctx
, anon_vma_name(vma
));
3321 * Source vma may have been merged into new_vma
3323 if (unlikely(vma_start
>= new_vma
->vm_start
&&
3324 vma_start
< new_vma
->vm_end
)) {
3326 * The only way we can get a vma_merge with
3327 * self during an mremap is if the vma hasn't
3328 * been faulted in yet and we were allowed to
3329 * reset the dst vma->vm_pgoff to the
3330 * destination address of the mremap to allow
3331 * the merge to happen. mremap must change the
3332 * vm_pgoff linearity between src and dst vmas
3333 * (in turn preventing a vma_merge) to be
3334 * safe. It is only safe to keep the vm_pgoff
3335 * linear if there are no pages mapped yet.
3337 VM_BUG_ON_VMA(faulted_in_anon_vma
, new_vma
);
3338 *vmap
= vma
= new_vma
;
3340 *need_rmap_locks
= (new_vma
->vm_pgoff
<= vma
->vm_pgoff
);
3342 new_vma
= vm_area_dup(vma
);
3345 new_vma
->vm_start
= addr
;
3346 new_vma
->vm_end
= addr
+ len
;
3347 new_vma
->vm_pgoff
= pgoff
;
3348 if (vma_dup_policy(vma
, new_vma
))
3350 if (anon_vma_clone(new_vma
, vma
))
3351 goto out_free_mempol
;
3352 if (new_vma
->vm_file
)
3353 get_file(new_vma
->vm_file
);
3354 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
3355 new_vma
->vm_ops
->open(new_vma
);
3356 if (vma_link(mm
, new_vma
))
3358 *need_rmap_locks
= false;
3363 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->close
)
3364 new_vma
->vm_ops
->close(new_vma
);
3366 if (new_vma
->vm_file
)
3367 fput(new_vma
->vm_file
);
3369 unlink_anon_vmas(new_vma
);
3371 mpol_put(vma_policy(new_vma
));
3373 vm_area_free(new_vma
);
3379 * Return true if the calling process may expand its vm space by the passed
3382 bool may_expand_vm(struct mm_struct
*mm
, vm_flags_t flags
, unsigned long npages
)
3384 if (mm
->total_vm
+ npages
> rlimit(RLIMIT_AS
) >> PAGE_SHIFT
)
3387 if (is_data_mapping(flags
) &&
3388 mm
->data_vm
+ npages
> rlimit(RLIMIT_DATA
) >> PAGE_SHIFT
) {
3389 /* Workaround for Valgrind */
3390 if (rlimit(RLIMIT_DATA
) == 0 &&
3391 mm
->data_vm
+ npages
<= rlimit_max(RLIMIT_DATA
) >> PAGE_SHIFT
)
3394 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3395 current
->comm
, current
->pid
,
3396 (mm
->data_vm
+ npages
) << PAGE_SHIFT
,
3397 rlimit(RLIMIT_DATA
),
3398 ignore_rlimit_data
? "" : " or use boot option ignore_rlimit_data");
3400 if (!ignore_rlimit_data
)
3407 void vm_stat_account(struct mm_struct
*mm
, vm_flags_t flags
, long npages
)
3409 WRITE_ONCE(mm
->total_vm
, READ_ONCE(mm
->total_vm
)+npages
);
3411 if (is_exec_mapping(flags
))
3412 mm
->exec_vm
+= npages
;
3413 else if (is_stack_mapping(flags
))
3414 mm
->stack_vm
+= npages
;
3415 else if (is_data_mapping(flags
))
3416 mm
->data_vm
+= npages
;
3419 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
);
3422 * Having a close hook prevents vma merging regardless of flags.
3424 static void special_mapping_close(struct vm_area_struct
*vma
)
3428 static const char *special_mapping_name(struct vm_area_struct
*vma
)
3430 return ((struct vm_special_mapping
*)vma
->vm_private_data
)->name
;
3433 static int special_mapping_mremap(struct vm_area_struct
*new_vma
)
3435 struct vm_special_mapping
*sm
= new_vma
->vm_private_data
;
3437 if (WARN_ON_ONCE(current
->mm
!= new_vma
->vm_mm
))
3441 return sm
->mremap(sm
, new_vma
);
3446 static int special_mapping_split(struct vm_area_struct
*vma
, unsigned long addr
)
3449 * Forbid splitting special mappings - kernel has expectations over
3450 * the number of pages in mapping. Together with VM_DONTEXPAND
3451 * the size of vma should stay the same over the special mapping's
3457 static const struct vm_operations_struct special_mapping_vmops
= {
3458 .close
= special_mapping_close
,
3459 .fault
= special_mapping_fault
,
3460 .mremap
= special_mapping_mremap
,
3461 .name
= special_mapping_name
,
3462 /* vDSO code relies that VVAR can't be accessed remotely */
3464 .may_split
= special_mapping_split
,
3467 static const struct vm_operations_struct legacy_special_mapping_vmops
= {
3468 .close
= special_mapping_close
,
3469 .fault
= special_mapping_fault
,
3472 static vm_fault_t
special_mapping_fault(struct vm_fault
*vmf
)
3474 struct vm_area_struct
*vma
= vmf
->vma
;
3476 struct page
**pages
;
3478 if (vma
->vm_ops
== &legacy_special_mapping_vmops
) {
3479 pages
= vma
->vm_private_data
;
3481 struct vm_special_mapping
*sm
= vma
->vm_private_data
;
3484 return sm
->fault(sm
, vmf
->vma
, vmf
);
3489 for (pgoff
= vmf
->pgoff
; pgoff
&& *pages
; ++pages
)
3493 struct page
*page
= *pages
;
3499 return VM_FAULT_SIGBUS
;
3502 static struct vm_area_struct
*__install_special_mapping(
3503 struct mm_struct
*mm
,
3504 unsigned long addr
, unsigned long len
,
3505 unsigned long vm_flags
, void *priv
,
3506 const struct vm_operations_struct
*ops
)
3509 struct vm_area_struct
*vma
;
3511 vma
= vm_area_alloc(mm
);
3512 if (unlikely(vma
== NULL
))
3513 return ERR_PTR(-ENOMEM
);
3515 vma
->vm_start
= addr
;
3516 vma
->vm_end
= addr
+ len
;
3518 vm_flags_init(vma
, (vm_flags
| mm
->def_flags
|
3519 VM_DONTEXPAND
| VM_SOFTDIRTY
) & ~VM_LOCKED_MASK
);
3520 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
3523 vma
->vm_private_data
= priv
;
3525 ret
= insert_vm_struct(mm
, vma
);
3529 vm_stat_account(mm
, vma
->vm_flags
, len
>> PAGE_SHIFT
);
3531 perf_event_mmap(vma
);
3537 return ERR_PTR(ret
);
3540 bool vma_is_special_mapping(const struct vm_area_struct
*vma
,
3541 const struct vm_special_mapping
*sm
)
3543 return vma
->vm_private_data
== sm
&&
3544 (vma
->vm_ops
== &special_mapping_vmops
||
3545 vma
->vm_ops
== &legacy_special_mapping_vmops
);
3549 * Called with mm->mmap_lock held for writing.
3550 * Insert a new vma covering the given region, with the given flags.
3551 * Its pages are supplied by the given array of struct page *.
3552 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3553 * The region past the last page supplied will always produce SIGBUS.
3554 * The array pointer and the pages it points to are assumed to stay alive
3555 * for as long as this mapping might exist.
3557 struct vm_area_struct
*_install_special_mapping(
3558 struct mm_struct
*mm
,
3559 unsigned long addr
, unsigned long len
,
3560 unsigned long vm_flags
, const struct vm_special_mapping
*spec
)
3562 return __install_special_mapping(mm
, addr
, len
, vm_flags
, (void *)spec
,
3563 &special_mapping_vmops
);
3566 int install_special_mapping(struct mm_struct
*mm
,
3567 unsigned long addr
, unsigned long len
,
3568 unsigned long vm_flags
, struct page
**pages
)
3570 struct vm_area_struct
*vma
= __install_special_mapping(
3571 mm
, addr
, len
, vm_flags
, (void *)pages
,
3572 &legacy_special_mapping_vmops
);
3574 return PTR_ERR_OR_ZERO(vma
);
3577 static DEFINE_MUTEX(mm_all_locks_mutex
);
3579 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
3581 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3583 * The LSB of head.next can't change from under us
3584 * because we hold the mm_all_locks_mutex.
3586 down_write_nest_lock(&anon_vma
->root
->rwsem
, &mm
->mmap_lock
);
3588 * We can safely modify head.next after taking the
3589 * anon_vma->root->rwsem. If some other vma in this mm shares
3590 * the same anon_vma we won't take it again.
3592 * No need of atomic instructions here, head.next
3593 * can't change from under us thanks to the
3594 * anon_vma->root->rwsem.
3596 if (__test_and_set_bit(0, (unsigned long *)
3597 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3602 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
3604 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3606 * AS_MM_ALL_LOCKS can't change from under us because
3607 * we hold the mm_all_locks_mutex.
3609 * Operations on ->flags have to be atomic because
3610 * even if AS_MM_ALL_LOCKS is stable thanks to the
3611 * mm_all_locks_mutex, there may be other cpus
3612 * changing other bitflags in parallel to us.
3614 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
3616 down_write_nest_lock(&mapping
->i_mmap_rwsem
, &mm
->mmap_lock
);
3621 * This operation locks against the VM for all pte/vma/mm related
3622 * operations that could ever happen on a certain mm. This includes
3623 * vmtruncate, try_to_unmap, and all page faults.
3625 * The caller must take the mmap_lock in write mode before calling
3626 * mm_take_all_locks(). The caller isn't allowed to release the
3627 * mmap_lock until mm_drop_all_locks() returns.
3629 * mmap_lock in write mode is required in order to block all operations
3630 * that could modify pagetables and free pages without need of
3631 * altering the vma layout. It's also needed in write mode to avoid new
3632 * anon_vmas to be associated with existing vmas.
3634 * A single task can't take more than one mm_take_all_locks() in a row
3635 * or it would deadlock.
3637 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3638 * mapping->flags avoid to take the same lock twice, if more than one
3639 * vma in this mm is backed by the same anon_vma or address_space.
3641 * We take locks in following order, accordingly to comment at beginning
3643 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3645 * - all vmas marked locked
3646 * - all i_mmap_rwsem locks;
3647 * - all anon_vma->rwseml
3649 * We can take all locks within these types randomly because the VM code
3650 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3651 * mm_all_locks_mutex.
3653 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3654 * that may have to take thousand of locks.
3656 * mm_take_all_locks() can fail if it's interrupted by signals.
3658 int mm_take_all_locks(struct mm_struct
*mm
)
3660 struct vm_area_struct
*vma
;
3661 struct anon_vma_chain
*avc
;
3662 MA_STATE(mas
, &mm
->mm_mt
, 0, 0);
3664 mmap_assert_write_locked(mm
);
3666 mutex_lock(&mm_all_locks_mutex
);
3669 * vma_start_write() does not have a complement in mm_drop_all_locks()
3670 * because vma_start_write() is always asymmetrical; it marks a VMA as
3671 * being written to until mmap_write_unlock() or mmap_write_downgrade()
3674 mas_for_each(&mas
, vma
, ULONG_MAX
) {
3675 if (signal_pending(current
))
3677 vma_start_write(vma
);
3681 mas_for_each(&mas
, vma
, ULONG_MAX
) {
3682 if (signal_pending(current
))
3684 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3685 is_vm_hugetlb_page(vma
))
3686 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3690 mas_for_each(&mas
, vma
, ULONG_MAX
) {
3691 if (signal_pending(current
))
3693 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
3694 !is_vm_hugetlb_page(vma
))
3695 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
3699 mas_for_each(&mas
, vma
, ULONG_MAX
) {
3700 if (signal_pending(current
))
3703 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3704 vm_lock_anon_vma(mm
, avc
->anon_vma
);
3710 mm_drop_all_locks(mm
);
3714 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
3716 if (test_bit(0, (unsigned long *) &anon_vma
->root
->rb_root
.rb_root
.rb_node
)) {
3718 * The LSB of head.next can't change to 0 from under
3719 * us because we hold the mm_all_locks_mutex.
3721 * We must however clear the bitflag before unlocking
3722 * the vma so the users using the anon_vma->rb_root will
3723 * never see our bitflag.
3725 * No need of atomic instructions here, head.next
3726 * can't change from under us until we release the
3727 * anon_vma->root->rwsem.
3729 if (!__test_and_clear_bit(0, (unsigned long *)
3730 &anon_vma
->root
->rb_root
.rb_root
.rb_node
))
3732 anon_vma_unlock_write(anon_vma
);
3736 static void vm_unlock_mapping(struct address_space
*mapping
)
3738 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
3740 * AS_MM_ALL_LOCKS can't change to 0 from under us
3741 * because we hold the mm_all_locks_mutex.
3743 i_mmap_unlock_write(mapping
);
3744 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
3751 * The mmap_lock cannot be released by the caller until
3752 * mm_drop_all_locks() returns.
3754 void mm_drop_all_locks(struct mm_struct
*mm
)
3756 struct vm_area_struct
*vma
;
3757 struct anon_vma_chain
*avc
;
3758 MA_STATE(mas
, &mm
->mm_mt
, 0, 0);
3760 mmap_assert_write_locked(mm
);
3761 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
3763 mas_for_each(&mas
, vma
, ULONG_MAX
) {
3765 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
3766 vm_unlock_anon_vma(avc
->anon_vma
);
3767 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
3768 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
3771 mutex_unlock(&mm_all_locks_mutex
);
3775 * initialise the percpu counter for VM
3777 void __init
mmap_init(void)
3781 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
3786 * Initialise sysctl_user_reserve_kbytes.
3788 * This is intended to prevent a user from starting a single memory hogging
3789 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3792 * The default value is min(3% of free memory, 128MB)
3793 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3795 static int init_user_reserve(void)
3797 unsigned long free_kbytes
;
3799 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
3801 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
3804 subsys_initcall(init_user_reserve
);
3807 * Initialise sysctl_admin_reserve_kbytes.
3809 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3810 * to log in and kill a memory hogging process.
3812 * Systems with more than 256MB will reserve 8MB, enough to recover
3813 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3814 * only reserve 3% of free pages by default.
3816 static int init_admin_reserve(void)
3818 unsigned long free_kbytes
;
3820 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
3822 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
3825 subsys_initcall(init_admin_reserve
);
3828 * Reinititalise user and admin reserves if memory is added or removed.
3830 * The default user reserve max is 128MB, and the default max for the
3831 * admin reserve is 8MB. These are usually, but not always, enough to
3832 * enable recovery from a memory hogging process using login/sshd, a shell,
3833 * and tools like top. It may make sense to increase or even disable the
3834 * reserve depending on the existence of swap or variations in the recovery
3835 * tools. So, the admin may have changed them.
3837 * If memory is added and the reserves have been eliminated or increased above
3838 * the default max, then we'll trust the admin.
3840 * If memory is removed and there isn't enough free memory, then we
3841 * need to reset the reserves.
3843 * Otherwise keep the reserve set by the admin.
3845 static int reserve_mem_notifier(struct notifier_block
*nb
,
3846 unsigned long action
, void *data
)
3848 unsigned long tmp
, free_kbytes
;
3852 /* Default max is 128MB. Leave alone if modified by operator. */
3853 tmp
= sysctl_user_reserve_kbytes
;
3854 if (0 < tmp
&& tmp
< (1UL << 17))
3855 init_user_reserve();
3857 /* Default max is 8MB. Leave alone if modified by operator. */
3858 tmp
= sysctl_admin_reserve_kbytes
;
3859 if (0 < tmp
&& tmp
< (1UL << 13))
3860 init_admin_reserve();
3864 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
3866 if (sysctl_user_reserve_kbytes
> free_kbytes
) {
3867 init_user_reserve();
3868 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3869 sysctl_user_reserve_kbytes
);
3872 if (sysctl_admin_reserve_kbytes
> free_kbytes
) {
3873 init_admin_reserve();
3874 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3875 sysctl_admin_reserve_kbytes
);
3884 static int __meminit
init_reserve_notifier(void)
3886 if (hotplug_memory_notifier(reserve_mem_notifier
, DEFAULT_CALLBACK_PRI
))
3887 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3891 subsys_initcall(init_reserve_notifier
);