1 // SPDX-License-Identifier: GPL-2.0-only
5 * Replacement code for mm functions to support CPU's that don't
6 * have any form of memory management unit (thus no virtual memory).
8 * See Documentation/admin-guide/mm/nommu-mmap.rst
10 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
11 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
12 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
13 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
14 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/export.h>
21 #include <linux/sched/mm.h>
22 #include <linux/vmacache.h>
23 #include <linux/mman.h>
24 #include <linux/swap.h>
25 #include <linux/file.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/slab.h>
29 #include <linux/vmalloc.h>
30 #include <linux/blkdev.h>
31 #include <linux/backing-dev.h>
32 #include <linux/compiler.h>
33 #include <linux/mount.h>
34 #include <linux/personality.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/printk.h>
40 #include <linux/uaccess.h>
42 #include <asm/tlbflush.h>
43 #include <asm/mmu_context.h>
47 EXPORT_SYMBOL(high_memory
);
49 unsigned long max_mapnr
;
50 EXPORT_SYMBOL(max_mapnr
);
51 unsigned long highest_memmap_pfn
;
52 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
53 int heap_stack_gap
= 0;
55 atomic_long_t mmap_pages_allocated
;
57 EXPORT_SYMBOL(mem_map
);
59 /* list of mapped, potentially shareable regions */
60 static struct kmem_cache
*vm_region_jar
;
61 struct rb_root nommu_region_tree
= RB_ROOT
;
62 DECLARE_RWSEM(nommu_region_sem
);
64 const struct vm_operations_struct generic_file_vm_ops
= {
68 * Return the total memory allocated for this pointer, not
69 * just what the caller asked for.
71 * Doesn't have to be accurate, i.e. may have races.
73 unsigned int kobjsize(const void *objp
)
78 * If the object we have should not have ksize performed on it,
81 if (!objp
|| !virt_addr_valid(objp
))
84 page
= virt_to_head_page(objp
);
87 * If the allocator sets PageSlab, we know the pointer came from
94 * If it's not a compound page, see if we have a matching VMA
95 * region. This test is intentionally done in reverse order,
96 * so if there's no VMA, we still fall through and hand back
97 * PAGE_SIZE for 0-order pages.
99 if (!PageCompound(page
)) {
100 struct vm_area_struct
*vma
;
102 vma
= find_vma(current
->mm
, (unsigned long)objp
);
104 return vma
->vm_end
- vma
->vm_start
;
108 * The ksize() function is only guaranteed to work for pointers
109 * returned by kmalloc(). So handle arbitrary pointers here.
111 return page_size(page
);
115 * follow_pfn - look up PFN at a user virtual address
116 * @vma: memory mapping
117 * @address: user virtual address
118 * @pfn: location to store found PFN
120 * Only IO mappings and raw PFN mappings are allowed.
122 * Returns zero and the pfn at @pfn on success, -ve otherwise.
124 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
127 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
130 *pfn
= address
>> PAGE_SHIFT
;
133 EXPORT_SYMBOL(follow_pfn
);
135 LIST_HEAD(vmap_area_list
);
137 void vfree(const void *addr
)
141 EXPORT_SYMBOL(vfree
);
143 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
)
146 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
147 * returns only a logical address.
149 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
151 EXPORT_SYMBOL(__vmalloc
);
153 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
154 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
155 pgprot_t prot
, unsigned long vm_flags
, int node
,
158 return __vmalloc(size
, gfp_mask
);
161 void *__vmalloc_node(unsigned long size
, unsigned long align
, gfp_t gfp_mask
,
162 int node
, const void *caller
)
164 return __vmalloc(size
, gfp_mask
);
167 static void *__vmalloc_user_flags(unsigned long size
, gfp_t flags
)
171 ret
= __vmalloc(size
, flags
);
173 struct vm_area_struct
*vma
;
175 mmap_write_lock(current
->mm
);
176 vma
= find_vma(current
->mm
, (unsigned long)ret
);
178 vma
->vm_flags
|= VM_USERMAP
;
179 mmap_write_unlock(current
->mm
);
185 void *vmalloc_user(unsigned long size
)
187 return __vmalloc_user_flags(size
, GFP_KERNEL
| __GFP_ZERO
);
189 EXPORT_SYMBOL(vmalloc_user
);
191 struct page
*vmalloc_to_page(const void *addr
)
193 return virt_to_page(addr
);
195 EXPORT_SYMBOL(vmalloc_to_page
);
197 unsigned long vmalloc_to_pfn(const void *addr
)
199 return page_to_pfn(virt_to_page(addr
));
201 EXPORT_SYMBOL(vmalloc_to_pfn
);
203 long vread(char *buf
, char *addr
, unsigned long count
)
205 /* Don't allow overflow */
206 if ((unsigned long) buf
+ count
< count
)
207 count
= -(unsigned long) buf
;
209 memcpy(buf
, addr
, count
);
213 long vwrite(char *buf
, char *addr
, unsigned long count
)
215 /* Don't allow overflow */
216 if ((unsigned long) addr
+ count
< count
)
217 count
= -(unsigned long) addr
;
219 memcpy(addr
, buf
, count
);
224 * vmalloc - allocate virtually contiguous memory
226 * @size: allocation size
228 * Allocate enough pages to cover @size from the page level
229 * allocator and map them into contiguous kernel virtual space.
231 * For tight control over page level allocator and protection flags
232 * use __vmalloc() instead.
234 void *vmalloc(unsigned long size
)
236 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
);
238 EXPORT_SYMBOL(vmalloc
);
241 * vzalloc - allocate virtually contiguous memory with zero fill
243 * @size: allocation size
245 * Allocate enough pages to cover @size from the page level
246 * allocator and map them into contiguous kernel virtual space.
247 * The memory allocated is set to zero.
249 * For tight control over page level allocator and protection flags
250 * use __vmalloc() instead.
252 void *vzalloc(unsigned long size
)
254 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
256 EXPORT_SYMBOL(vzalloc
);
259 * vmalloc_node - allocate memory on a specific node
260 * @size: allocation size
263 * Allocate enough pages to cover @size from the page level
264 * allocator and map them into contiguous kernel virtual space.
266 * For tight control over page level allocator and protection flags
267 * use __vmalloc() instead.
269 void *vmalloc_node(unsigned long size
, int node
)
271 return vmalloc(size
);
273 EXPORT_SYMBOL(vmalloc_node
);
276 * vzalloc_node - allocate memory on a specific node with zero fill
277 * @size: allocation size
280 * Allocate enough pages to cover @size from the page level
281 * allocator and map them into contiguous kernel virtual space.
282 * The memory allocated is set to zero.
284 * For tight control over page level allocator and protection flags
285 * use __vmalloc() instead.
287 void *vzalloc_node(unsigned long size
, int node
)
289 return vzalloc(size
);
291 EXPORT_SYMBOL(vzalloc_node
);
294 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
295 * @size: allocation size
297 * Allocate enough 32bit PA addressable pages to cover @size from the
298 * page level allocator and map them into contiguous kernel virtual space.
300 void *vmalloc_32(unsigned long size
)
302 return __vmalloc(size
, GFP_KERNEL
);
304 EXPORT_SYMBOL(vmalloc_32
);
307 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
308 * @size: allocation size
310 * The resulting memory area is 32bit addressable and zeroed so it can be
311 * mapped to userspace without leaking data.
313 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
314 * remap_vmalloc_range() are permissible.
316 void *vmalloc_32_user(unsigned long size
)
319 * We'll have to sort out the ZONE_DMA bits for 64-bit,
320 * but for now this can simply use vmalloc_user() directly.
322 return vmalloc_user(size
);
324 EXPORT_SYMBOL(vmalloc_32_user
);
326 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
333 void vunmap(const void *addr
)
337 EXPORT_SYMBOL(vunmap
);
339 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
)
344 EXPORT_SYMBOL(vm_map_ram
);
346 void vm_unmap_ram(const void *mem
, unsigned int count
)
350 EXPORT_SYMBOL(vm_unmap_ram
);
352 void vm_unmap_aliases(void)
355 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
357 void free_vm_area(struct vm_struct
*area
)
361 EXPORT_SYMBOL_GPL(free_vm_area
);
363 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
368 EXPORT_SYMBOL(vm_insert_page
);
370 int vm_map_pages(struct vm_area_struct
*vma
, struct page
**pages
,
375 EXPORT_SYMBOL(vm_map_pages
);
377 int vm_map_pages_zero(struct vm_area_struct
*vma
, struct page
**pages
,
382 EXPORT_SYMBOL(vm_map_pages_zero
);
385 * sys_brk() for the most part doesn't need the global kernel
386 * lock, except when an application is doing something nasty
387 * like trying to un-brk an area that has already been mapped
388 * to a regular file. in this case, the unmapping will need
389 * to invoke file system routines that need the global lock.
391 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
393 struct mm_struct
*mm
= current
->mm
;
395 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
402 * Always allow shrinking brk
404 if (brk
<= mm
->brk
) {
410 * Ok, looks good - let it rip.
412 flush_icache_user_range(mm
->brk
, brk
);
413 return mm
->brk
= brk
;
417 * initialise the percpu counter for VM and region record slabs
419 void __init
mmap_init(void)
423 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
425 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
429 * validate the region tree
430 * - the caller must hold the region lock
432 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
433 static noinline
void validate_nommu_regions(void)
435 struct vm_region
*region
, *last
;
436 struct rb_node
*p
, *lastp
;
438 lastp
= rb_first(&nommu_region_tree
);
442 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
443 BUG_ON(last
->vm_end
<= last
->vm_start
);
444 BUG_ON(last
->vm_top
< last
->vm_end
);
446 while ((p
= rb_next(lastp
))) {
447 region
= rb_entry(p
, struct vm_region
, vm_rb
);
448 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
450 BUG_ON(region
->vm_end
<= region
->vm_start
);
451 BUG_ON(region
->vm_top
< region
->vm_end
);
452 BUG_ON(region
->vm_start
< last
->vm_top
);
458 static void validate_nommu_regions(void)
464 * add a region into the global tree
466 static void add_nommu_region(struct vm_region
*region
)
468 struct vm_region
*pregion
;
469 struct rb_node
**p
, *parent
;
471 validate_nommu_regions();
474 p
= &nommu_region_tree
.rb_node
;
477 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
478 if (region
->vm_start
< pregion
->vm_start
)
480 else if (region
->vm_start
> pregion
->vm_start
)
482 else if (pregion
== region
)
488 rb_link_node(®ion
->vm_rb
, parent
, p
);
489 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
491 validate_nommu_regions();
495 * delete a region from the global tree
497 static void delete_nommu_region(struct vm_region
*region
)
499 BUG_ON(!nommu_region_tree
.rb_node
);
501 validate_nommu_regions();
502 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
503 validate_nommu_regions();
507 * free a contiguous series of pages
509 static void free_page_series(unsigned long from
, unsigned long to
)
511 for (; from
< to
; from
+= PAGE_SIZE
) {
512 struct page
*page
= virt_to_page(from
);
514 atomic_long_dec(&mmap_pages_allocated
);
520 * release a reference to a region
521 * - the caller must hold the region semaphore for writing, which this releases
522 * - the region may not have been added to the tree yet, in which case vm_top
523 * will equal vm_start
525 static void __put_nommu_region(struct vm_region
*region
)
526 __releases(nommu_region_sem
)
528 BUG_ON(!nommu_region_tree
.rb_node
);
530 if (--region
->vm_usage
== 0) {
531 if (region
->vm_top
> region
->vm_start
)
532 delete_nommu_region(region
);
533 up_write(&nommu_region_sem
);
536 fput(region
->vm_file
);
538 /* IO memory and memory shared directly out of the pagecache
539 * from ramfs/tmpfs mustn't be released here */
540 if (region
->vm_flags
& VM_MAPPED_COPY
)
541 free_page_series(region
->vm_start
, region
->vm_top
);
542 kmem_cache_free(vm_region_jar
, region
);
544 up_write(&nommu_region_sem
);
549 * release a reference to a region
551 static void put_nommu_region(struct vm_region
*region
)
553 down_write(&nommu_region_sem
);
554 __put_nommu_region(region
);
558 * add a VMA into a process's mm_struct in the appropriate place in the list
559 * and tree and add to the address space's page tree also if not an anonymous
561 * - should be called with mm->mmap_lock held writelocked
563 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
565 struct vm_area_struct
*pvma
, *prev
;
566 struct address_space
*mapping
;
567 struct rb_node
**p
, *parent
, *rb_prev
;
569 BUG_ON(!vma
->vm_region
);
574 /* add the VMA to the mapping */
576 mapping
= vma
->vm_file
->f_mapping
;
578 i_mmap_lock_write(mapping
);
579 flush_dcache_mmap_lock(mapping
);
580 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
581 flush_dcache_mmap_unlock(mapping
);
582 i_mmap_unlock_write(mapping
);
585 /* add the VMA to the tree */
586 parent
= rb_prev
= NULL
;
587 p
= &mm
->mm_rb
.rb_node
;
590 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
592 /* sort by: start addr, end addr, VMA struct addr in that order
593 * (the latter is necessary as we may get identical VMAs) */
594 if (vma
->vm_start
< pvma
->vm_start
)
596 else if (vma
->vm_start
> pvma
->vm_start
) {
599 } else if (vma
->vm_end
< pvma
->vm_end
)
601 else if (vma
->vm_end
> pvma
->vm_end
) {
604 } else if (vma
< pvma
)
606 else if (vma
> pvma
) {
613 rb_link_node(&vma
->vm_rb
, parent
, p
);
614 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
616 /* add VMA to the VMA list also */
619 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
621 __vma_link_list(mm
, vma
, prev
);
625 * delete a VMA from its owning mm_struct and address space
627 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
630 struct address_space
*mapping
;
631 struct mm_struct
*mm
= vma
->vm_mm
;
632 struct task_struct
*curr
= current
;
635 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
636 /* if the vma is cached, invalidate the entire cache */
637 if (curr
->vmacache
.vmas
[i
] == vma
) {
638 vmacache_invalidate(mm
);
643 /* remove the VMA from the mapping */
645 mapping
= vma
->vm_file
->f_mapping
;
647 i_mmap_lock_write(mapping
);
648 flush_dcache_mmap_lock(mapping
);
649 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
650 flush_dcache_mmap_unlock(mapping
);
651 i_mmap_unlock_write(mapping
);
654 /* remove from the MM's tree and list */
655 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
657 __vma_unlink_list(mm
, vma
);
661 * destroy a VMA record
663 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
665 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
666 vma
->vm_ops
->close(vma
);
669 put_nommu_region(vma
->vm_region
);
674 * look up the first VMA in which addr resides, NULL if none
675 * - should be called with mm->mmap_lock at least held readlocked
677 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
679 struct vm_area_struct
*vma
;
681 /* check the cache first */
682 vma
= vmacache_find(mm
, addr
);
686 /* trawl the list (there may be multiple mappings in which addr
688 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
689 if (vma
->vm_start
> addr
)
691 if (vma
->vm_end
> addr
) {
692 vmacache_update(addr
, vma
);
699 EXPORT_SYMBOL(find_vma
);
703 * - we don't extend stack VMAs under NOMMU conditions
705 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
707 return find_vma(mm
, addr
);
711 * expand a stack to a given address
712 * - not supported under NOMMU conditions
714 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
720 * look up the first VMA exactly that exactly matches addr
721 * - should be called with mm->mmap_lock at least held readlocked
723 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
727 struct vm_area_struct
*vma
;
728 unsigned long end
= addr
+ len
;
730 /* check the cache first */
731 vma
= vmacache_find_exact(mm
, addr
, end
);
735 /* trawl the list (there may be multiple mappings in which addr
737 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
738 if (vma
->vm_start
< addr
)
740 if (vma
->vm_start
> addr
)
742 if (vma
->vm_end
== end
) {
743 vmacache_update(addr
, vma
);
752 * determine whether a mapping should be permitted and, if so, what sort of
753 * mapping we're capable of supporting
755 static int validate_mmap_request(struct file
*file
,
761 unsigned long *_capabilities
)
763 unsigned long capabilities
, rlen
;
766 /* do the simple checks first */
767 if (flags
& MAP_FIXED
)
770 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
771 (flags
& MAP_TYPE
) != MAP_SHARED
)
777 /* Careful about overflows.. */
778 rlen
= PAGE_ALIGN(len
);
779 if (!rlen
|| rlen
> TASK_SIZE
)
782 /* offset overflow? */
783 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
787 /* files must support mmap */
788 if (!file
->f_op
->mmap
)
791 /* work out if what we've got could possibly be shared
792 * - we support chardevs that provide their own "memory"
793 * - we support files/blockdevs that are memory backed
795 if (file
->f_op
->mmap_capabilities
) {
796 capabilities
= file
->f_op
->mmap_capabilities(file
);
798 /* no explicit capabilities set, so assume some
800 switch (file_inode(file
)->i_mode
& S_IFMT
) {
803 capabilities
= NOMMU_MAP_COPY
;
818 /* eliminate any capabilities that we can't support on this
820 if (!file
->f_op
->get_unmapped_area
)
821 capabilities
&= ~NOMMU_MAP_DIRECT
;
822 if (!(file
->f_mode
& FMODE_CAN_READ
))
823 capabilities
&= ~NOMMU_MAP_COPY
;
825 /* The file shall have been opened with read permission. */
826 if (!(file
->f_mode
& FMODE_READ
))
829 if (flags
& MAP_SHARED
) {
830 /* do checks for writing, appending and locking */
831 if ((prot
& PROT_WRITE
) &&
832 !(file
->f_mode
& FMODE_WRITE
))
835 if (IS_APPEND(file_inode(file
)) &&
836 (file
->f_mode
& FMODE_WRITE
))
839 if (locks_verify_locked(file
))
842 if (!(capabilities
& NOMMU_MAP_DIRECT
))
845 /* we mustn't privatise shared mappings */
846 capabilities
&= ~NOMMU_MAP_COPY
;
848 /* we're going to read the file into private memory we
850 if (!(capabilities
& NOMMU_MAP_COPY
))
853 /* we don't permit a private writable mapping to be
854 * shared with the backing device */
855 if (prot
& PROT_WRITE
)
856 capabilities
&= ~NOMMU_MAP_DIRECT
;
859 if (capabilities
& NOMMU_MAP_DIRECT
) {
860 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
861 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
862 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
864 capabilities
&= ~NOMMU_MAP_DIRECT
;
865 if (flags
& MAP_SHARED
) {
866 pr_warn("MAP_SHARED not completely supported on !MMU\n");
872 /* handle executable mappings and implied executable
874 if (path_noexec(&file
->f_path
)) {
875 if (prot
& PROT_EXEC
)
877 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
878 /* handle implication of PROT_EXEC by PROT_READ */
879 if (current
->personality
& READ_IMPLIES_EXEC
) {
880 if (capabilities
& NOMMU_MAP_EXEC
)
883 } else if ((prot
& PROT_READ
) &&
884 (prot
& PROT_EXEC
) &&
885 !(capabilities
& NOMMU_MAP_EXEC
)
887 /* backing file is not executable, try to copy */
888 capabilities
&= ~NOMMU_MAP_DIRECT
;
891 /* anonymous mappings are always memory backed and can be
894 capabilities
= NOMMU_MAP_COPY
;
896 /* handle PROT_EXEC implication by PROT_READ */
897 if ((prot
& PROT_READ
) &&
898 (current
->personality
& READ_IMPLIES_EXEC
))
902 /* allow the security API to have its say */
903 ret
= security_mmap_addr(addr
);
908 *_capabilities
= capabilities
;
913 * we've determined that we can make the mapping, now translate what we
914 * now know into VMA flags
916 static unsigned long determine_vm_flags(struct file
*file
,
919 unsigned long capabilities
)
921 unsigned long vm_flags
;
923 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
924 /* vm_flags |= mm->def_flags; */
926 if (!(capabilities
& NOMMU_MAP_DIRECT
)) {
927 /* attempt to share read-only copies of mapped file chunks */
928 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
929 if (file
&& !(prot
& PROT_WRITE
))
930 vm_flags
|= VM_MAYSHARE
;
932 /* overlay a shareable mapping on the backing device or inode
933 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
935 vm_flags
|= VM_MAYSHARE
| (capabilities
& NOMMU_VMFLAGS
);
936 if (flags
& MAP_SHARED
)
937 vm_flags
|= VM_SHARED
;
940 /* refuse to let anyone share private mappings with this process if
941 * it's being traced - otherwise breakpoints set in it may interfere
942 * with another untraced process
944 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
945 vm_flags
&= ~VM_MAYSHARE
;
951 * set up a shared mapping on a file (the driver or filesystem provides and
954 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
958 ret
= call_mmap(vma
->vm_file
, vma
);
960 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
966 /* getting -ENOSYS indicates that direct mmap isn't possible (as
967 * opposed to tried but failed) so we can only give a suitable error as
968 * it's not possible to make a private copy if MAP_SHARED was given */
973 * set up a private mapping or an anonymous shared mapping
975 static int do_mmap_private(struct vm_area_struct
*vma
,
976 struct vm_region
*region
,
978 unsigned long capabilities
)
980 unsigned long total
, point
;
984 /* invoke the file's mapping function so that it can keep track of
985 * shared mappings on devices or memory
986 * - VM_MAYSHARE will be set if it may attempt to share
988 if (capabilities
& NOMMU_MAP_DIRECT
) {
989 ret
= call_mmap(vma
->vm_file
, vma
);
991 /* shouldn't return success if we're not sharing */
992 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
993 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
999 /* getting an ENOSYS error indicates that direct mmap isn't
1000 * possible (as opposed to tried but failed) so we'll try to
1001 * make a private copy of the data and map that instead */
1005 /* allocate some memory to hold the mapping
1006 * - note that this may not return a page-aligned address if the object
1007 * we're allocating is smaller than a page
1009 order
= get_order(len
);
1011 point
= len
>> PAGE_SHIFT
;
1013 /* we don't want to allocate a power-of-2 sized page set */
1014 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
1017 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
1021 atomic_long_add(total
, &mmap_pages_allocated
);
1023 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1024 region
->vm_start
= (unsigned long) base
;
1025 region
->vm_end
= region
->vm_start
+ len
;
1026 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1028 vma
->vm_start
= region
->vm_start
;
1029 vma
->vm_end
= region
->vm_start
+ len
;
1032 /* read the contents of a file into the copy */
1035 fpos
= vma
->vm_pgoff
;
1036 fpos
<<= PAGE_SHIFT
;
1038 ret
= kernel_read(vma
->vm_file
, base
, len
, &fpos
);
1042 /* clear the last little bit */
1044 memset(base
+ ret
, 0, len
- ret
);
1047 vma_set_anonymous(vma
);
1053 free_page_series(region
->vm_start
, region
->vm_top
);
1054 region
->vm_start
= vma
->vm_start
= 0;
1055 region
->vm_end
= vma
->vm_end
= 0;
1060 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1061 len
, current
->pid
, current
->comm
);
1062 show_free_areas(0, NULL
);
1067 * handle mapping creation for uClinux
1069 unsigned long do_mmap(struct file
*file
,
1073 unsigned long flags
,
1074 unsigned long pgoff
,
1075 unsigned long *populate
,
1076 struct list_head
*uf
)
1078 struct vm_area_struct
*vma
;
1079 struct vm_region
*region
;
1081 vm_flags_t vm_flags
;
1082 unsigned long capabilities
, result
;
1087 /* decide whether we should attempt the mapping, and if so what sort of
1089 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1094 /* we ignore the address hint */
1096 len
= PAGE_ALIGN(len
);
1098 /* we've determined that we can make the mapping, now translate what we
1099 * now know into VMA flags */
1100 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1102 /* we're going to need to record the mapping */
1103 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1105 goto error_getting_region
;
1107 vma
= vm_area_alloc(current
->mm
);
1109 goto error_getting_vma
;
1111 region
->vm_usage
= 1;
1112 region
->vm_flags
= vm_flags
;
1113 region
->vm_pgoff
= pgoff
;
1115 vma
->vm_flags
= vm_flags
;
1116 vma
->vm_pgoff
= pgoff
;
1119 region
->vm_file
= get_file(file
);
1120 vma
->vm_file
= get_file(file
);
1123 down_write(&nommu_region_sem
);
1125 /* if we want to share, we need to check for regions created by other
1126 * mmap() calls that overlap with our proposed mapping
1127 * - we can only share with a superset match on most regular files
1128 * - shared mappings on character devices and memory backed files are
1129 * permitted to overlap inexactly as far as we are concerned for in
1130 * these cases, sharing is handled in the driver or filesystem rather
1133 if (vm_flags
& VM_MAYSHARE
) {
1134 struct vm_region
*pregion
;
1135 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1137 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1138 pgend
= pgoff
+ pglen
;
1140 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1141 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1143 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1146 /* search for overlapping mappings on the same file */
1147 if (file_inode(pregion
->vm_file
) !=
1151 if (pregion
->vm_pgoff
>= pgend
)
1154 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1155 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1156 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1157 if (pgoff
>= rpgend
)
1160 /* handle inexactly overlapping matches between
1162 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1163 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1164 /* new mapping is not a subset of the region */
1165 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1166 goto sharing_violation
;
1170 /* we've found a region we can share */
1171 pregion
->vm_usage
++;
1172 vma
->vm_region
= pregion
;
1173 start
= pregion
->vm_start
;
1174 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1175 vma
->vm_start
= start
;
1176 vma
->vm_end
= start
+ len
;
1178 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1179 vma
->vm_flags
|= VM_MAPPED_COPY
;
1181 ret
= do_mmap_shared_file(vma
);
1183 vma
->vm_region
= NULL
;
1186 pregion
->vm_usage
--;
1188 goto error_just_free
;
1191 fput(region
->vm_file
);
1192 kmem_cache_free(vm_region_jar
, region
);
1198 /* obtain the address at which to make a shared mapping
1199 * - this is the hook for quasi-memory character devices to
1200 * tell us the location of a shared mapping
1202 if (capabilities
& NOMMU_MAP_DIRECT
) {
1203 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1205 if (IS_ERR_VALUE(addr
)) {
1208 goto error_just_free
;
1210 /* the driver refused to tell us where to site
1211 * the mapping so we'll have to attempt to copy
1214 if (!(capabilities
& NOMMU_MAP_COPY
))
1215 goto error_just_free
;
1217 capabilities
&= ~NOMMU_MAP_DIRECT
;
1219 vma
->vm_start
= region
->vm_start
= addr
;
1220 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1225 vma
->vm_region
= region
;
1227 /* set up the mapping
1228 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1230 if (file
&& vma
->vm_flags
& VM_SHARED
)
1231 ret
= do_mmap_shared_file(vma
);
1233 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1235 goto error_just_free
;
1236 add_nommu_region(region
);
1238 /* clear anonymous mappings that don't ask for uninitialized data */
1239 if (!vma
->vm_file
&&
1240 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED
) ||
1241 !(flags
& MAP_UNINITIALIZED
)))
1242 memset((void *)region
->vm_start
, 0,
1243 region
->vm_end
- region
->vm_start
);
1245 /* okay... we have a mapping; now we have to register it */
1246 result
= vma
->vm_start
;
1248 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1251 add_vma_to_mm(current
->mm
, vma
);
1253 /* we flush the region from the icache only when the first executable
1254 * mapping of it is made */
1255 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1256 flush_icache_user_range(region
->vm_start
, region
->vm_end
);
1257 region
->vm_icache_flushed
= true;
1260 up_write(&nommu_region_sem
);
1265 up_write(&nommu_region_sem
);
1267 if (region
->vm_file
)
1268 fput(region
->vm_file
);
1269 kmem_cache_free(vm_region_jar
, region
);
1276 up_write(&nommu_region_sem
);
1277 pr_warn("Attempt to share mismatched mappings\n");
1282 kmem_cache_free(vm_region_jar
, region
);
1283 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1285 show_free_areas(0, NULL
);
1288 error_getting_region
:
1289 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1291 show_free_areas(0, NULL
);
1295 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1296 unsigned long prot
, unsigned long flags
,
1297 unsigned long fd
, unsigned long pgoff
)
1299 struct file
*file
= NULL
;
1300 unsigned long retval
= -EBADF
;
1302 audit_mmap_fd(fd
, flags
);
1303 if (!(flags
& MAP_ANONYMOUS
)) {
1309 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1311 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1319 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1320 unsigned long, prot
, unsigned long, flags
,
1321 unsigned long, fd
, unsigned long, pgoff
)
1323 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1326 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1327 struct mmap_arg_struct
{
1331 unsigned long flags
;
1333 unsigned long offset
;
1336 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1338 struct mmap_arg_struct a
;
1340 if (copy_from_user(&a
, arg
, sizeof(a
)))
1342 if (offset_in_page(a
.offset
))
1345 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1346 a
.offset
>> PAGE_SHIFT
);
1348 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1351 * split a vma into two pieces at address 'addr', a new vma is allocated either
1352 * for the first part or the tail.
1354 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1355 unsigned long addr
, int new_below
)
1357 struct vm_area_struct
*new;
1358 struct vm_region
*region
;
1359 unsigned long npages
;
1361 /* we're only permitted to split anonymous regions (these should have
1362 * only a single usage on the region) */
1366 if (mm
->map_count
>= sysctl_max_map_count
)
1369 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1373 new = vm_area_dup(vma
);
1375 kmem_cache_free(vm_region_jar
, region
);
1379 /* most fields are the same, copy all, and then fixup */
1380 *region
= *vma
->vm_region
;
1381 new->vm_region
= region
;
1383 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1386 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1388 region
->vm_start
= new->vm_start
= addr
;
1389 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1392 if (new->vm_ops
&& new->vm_ops
->open
)
1393 new->vm_ops
->open(new);
1395 delete_vma_from_mm(vma
);
1396 down_write(&nommu_region_sem
);
1397 delete_nommu_region(vma
->vm_region
);
1399 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1400 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1402 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1403 vma
->vm_region
->vm_top
= addr
;
1405 add_nommu_region(vma
->vm_region
);
1406 add_nommu_region(new->vm_region
);
1407 up_write(&nommu_region_sem
);
1408 add_vma_to_mm(mm
, vma
);
1409 add_vma_to_mm(mm
, new);
1414 * shrink a VMA by removing the specified chunk from either the beginning or
1417 static int shrink_vma(struct mm_struct
*mm
,
1418 struct vm_area_struct
*vma
,
1419 unsigned long from
, unsigned long to
)
1421 struct vm_region
*region
;
1423 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1425 delete_vma_from_mm(vma
);
1426 if (from
> vma
->vm_start
)
1430 add_vma_to_mm(mm
, vma
);
1432 /* cut the backing region down to size */
1433 region
= vma
->vm_region
;
1434 BUG_ON(region
->vm_usage
!= 1);
1436 down_write(&nommu_region_sem
);
1437 delete_nommu_region(region
);
1438 if (from
> region
->vm_start
) {
1439 to
= region
->vm_top
;
1440 region
->vm_top
= region
->vm_end
= from
;
1442 region
->vm_start
= to
;
1444 add_nommu_region(region
);
1445 up_write(&nommu_region_sem
);
1447 free_page_series(from
, to
);
1453 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1454 * VMA, though it need not cover the whole VMA
1456 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1458 struct vm_area_struct
*vma
;
1462 len
= PAGE_ALIGN(len
);
1468 /* find the first potentially overlapping VMA */
1469 vma
= find_vma(mm
, start
);
1473 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1474 current
->pid
, current
->comm
,
1475 start
, start
+ len
- 1);
1481 /* we're allowed to split an anonymous VMA but not a file-backed one */
1484 if (start
> vma
->vm_start
)
1486 if (end
== vma
->vm_end
)
1487 goto erase_whole_vma
;
1492 /* the chunk must be a subset of the VMA found */
1493 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1494 goto erase_whole_vma
;
1495 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1497 if (offset_in_page(start
))
1499 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1501 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1502 ret
= split_vma(mm
, vma
, start
, 1);
1506 return shrink_vma(mm
, vma
, start
, end
);
1510 delete_vma_from_mm(vma
);
1511 delete_vma(mm
, vma
);
1514 EXPORT_SYMBOL(do_munmap
);
1516 int vm_munmap(unsigned long addr
, size_t len
)
1518 struct mm_struct
*mm
= current
->mm
;
1521 mmap_write_lock(mm
);
1522 ret
= do_munmap(mm
, addr
, len
, NULL
);
1523 mmap_write_unlock(mm
);
1526 EXPORT_SYMBOL(vm_munmap
);
1528 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1530 return vm_munmap(addr
, len
);
1534 * release all the mappings made in a process's VM space
1536 void exit_mmap(struct mm_struct
*mm
)
1538 struct vm_area_struct
*vma
;
1545 while ((vma
= mm
->mmap
)) {
1546 mm
->mmap
= vma
->vm_next
;
1547 delete_vma_from_mm(vma
);
1548 delete_vma(mm
, vma
);
1553 int vm_brk(unsigned long addr
, unsigned long len
)
1559 * expand (or shrink) an existing mapping, potentially moving it at the same
1560 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1562 * under NOMMU conditions, we only permit changing a mapping's size, and only
1563 * as long as it stays within the region allocated by do_mmap_private() and the
1564 * block is not shareable
1566 * MREMAP_FIXED is not supported under NOMMU conditions
1568 static unsigned long do_mremap(unsigned long addr
,
1569 unsigned long old_len
, unsigned long new_len
,
1570 unsigned long flags
, unsigned long new_addr
)
1572 struct vm_area_struct
*vma
;
1574 /* insanity checks first */
1575 old_len
= PAGE_ALIGN(old_len
);
1576 new_len
= PAGE_ALIGN(new_len
);
1577 if (old_len
== 0 || new_len
== 0)
1578 return (unsigned long) -EINVAL
;
1580 if (offset_in_page(addr
))
1583 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1584 return (unsigned long) -EINVAL
;
1586 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1588 return (unsigned long) -EINVAL
;
1590 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1591 return (unsigned long) -EFAULT
;
1593 if (vma
->vm_flags
& VM_MAYSHARE
)
1594 return (unsigned long) -EPERM
;
1596 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1597 return (unsigned long) -ENOMEM
;
1599 /* all checks complete - do it */
1600 vma
->vm_end
= vma
->vm_start
+ new_len
;
1601 return vma
->vm_start
;
1604 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1605 unsigned long, new_len
, unsigned long, flags
,
1606 unsigned long, new_addr
)
1610 mmap_write_lock(current
->mm
);
1611 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1612 mmap_write_unlock(current
->mm
);
1616 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1617 unsigned int foll_flags
)
1622 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1623 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1625 if (addr
!= (pfn
<< PAGE_SHIFT
))
1628 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1631 EXPORT_SYMBOL(remap_pfn_range
);
1633 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1635 unsigned long pfn
= start
>> PAGE_SHIFT
;
1636 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1638 pfn
+= vma
->vm_pgoff
;
1639 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1641 EXPORT_SYMBOL(vm_iomap_memory
);
1643 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1644 unsigned long pgoff
)
1646 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1648 if (!(vma
->vm_flags
& VM_USERMAP
))
1651 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1652 vma
->vm_end
= vma
->vm_start
+ size
;
1656 EXPORT_SYMBOL(remap_vmalloc_range
);
1658 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1659 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1664 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
1669 EXPORT_SYMBOL(filemap_fault
);
1671 void filemap_map_pages(struct vm_fault
*vmf
,
1672 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1676 EXPORT_SYMBOL(filemap_map_pages
);
1678 int __access_remote_vm(struct mm_struct
*mm
, unsigned long addr
, void *buf
,
1679 int len
, unsigned int gup_flags
)
1681 struct vm_area_struct
*vma
;
1682 int write
= gup_flags
& FOLL_WRITE
;
1684 if (mmap_read_lock_killable(mm
))
1687 /* the access must start within one of the target process's mappings */
1688 vma
= find_vma(mm
, addr
);
1690 /* don't overrun this mapping */
1691 if (addr
+ len
>= vma
->vm_end
)
1692 len
= vma
->vm_end
- addr
;
1694 /* only read or write mappings where it is permitted */
1695 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1696 copy_to_user_page(vma
, NULL
, addr
,
1697 (void *) addr
, buf
, len
);
1698 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1699 copy_from_user_page(vma
, NULL
, addr
,
1700 buf
, (void *) addr
, len
);
1707 mmap_read_unlock(mm
);
1713 * access_remote_vm - access another process' address space
1714 * @mm: the mm_struct of the target address space
1715 * @addr: start address to access
1716 * @buf: source or destination buffer
1717 * @len: number of bytes to transfer
1718 * @gup_flags: flags modifying lookup behaviour
1720 * The caller must hold a reference on @mm.
1722 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1723 void *buf
, int len
, unsigned int gup_flags
)
1725 return __access_remote_vm(mm
, addr
, buf
, len
, gup_flags
);
1729 * Access another process' address space.
1730 * - source/target buffer must be kernel space
1732 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1733 unsigned int gup_flags
)
1735 struct mm_struct
*mm
;
1737 if (addr
+ len
< addr
)
1740 mm
= get_task_mm(tsk
);
1744 len
= __access_remote_vm(mm
, addr
, buf
, len
, gup_flags
);
1749 EXPORT_SYMBOL_GPL(access_process_vm
);
1752 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1753 * @inode: The inode to check
1754 * @size: The current filesize of the inode
1755 * @newsize: The proposed filesize of the inode
1757 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1758 * make sure that any outstanding VMAs aren't broken and then shrink the
1759 * vm_regions that extend beyond so that do_mmap() doesn't
1760 * automatically grant mappings that are too large.
1762 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1765 struct vm_area_struct
*vma
;
1766 struct vm_region
*region
;
1768 size_t r_size
, r_top
;
1770 low
= newsize
>> PAGE_SHIFT
;
1771 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1773 down_write(&nommu_region_sem
);
1774 i_mmap_lock_read(inode
->i_mapping
);
1776 /* search for VMAs that fall within the dead zone */
1777 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1778 /* found one - only interested if it's shared out of the page
1780 if (vma
->vm_flags
& VM_SHARED
) {
1781 i_mmap_unlock_read(inode
->i_mapping
);
1782 up_write(&nommu_region_sem
);
1783 return -ETXTBSY
; /* not quite true, but near enough */
1787 /* reduce any regions that overlap the dead zone - if in existence,
1788 * these will be pointed to by VMAs that don't overlap the dead zone
1790 * we don't check for any regions that start beyond the EOF as there
1793 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1794 if (!(vma
->vm_flags
& VM_SHARED
))
1797 region
= vma
->vm_region
;
1798 r_size
= region
->vm_top
- region
->vm_start
;
1799 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1801 if (r_top
> newsize
) {
1802 region
->vm_top
-= r_top
- newsize
;
1803 if (region
->vm_end
> region
->vm_top
)
1804 region
->vm_end
= region
->vm_top
;
1808 i_mmap_unlock_read(inode
->i_mapping
);
1809 up_write(&nommu_region_sem
);
1814 * Initialise sysctl_user_reserve_kbytes.
1816 * This is intended to prevent a user from starting a single memory hogging
1817 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1820 * The default value is min(3% of free memory, 128MB)
1821 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1823 static int __meminit
init_user_reserve(void)
1825 unsigned long free_kbytes
;
1827 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1829 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1832 subsys_initcall(init_user_reserve
);
1835 * Initialise sysctl_admin_reserve_kbytes.
1837 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1838 * to log in and kill a memory hogging process.
1840 * Systems with more than 256MB will reserve 8MB, enough to recover
1841 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1842 * only reserve 3% of free pages by default.
1844 static int __meminit
init_admin_reserve(void)
1846 unsigned long free_kbytes
;
1848 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1850 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1853 subsys_initcall(init_admin_reserve
);