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/nommu-mmap.txt
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_exec - allocate virtually contiguous, executable memory
295 * @size: allocation size
297 * Kernel-internal function to allocate enough pages to cover @size
298 * the page level allocator and map them into contiguous and
299 * executable kernel virtual space.
301 * For tight control over page level allocator and protection flags
302 * use __vmalloc() instead.
305 void *vmalloc_exec(unsigned long size
)
307 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
);
311 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
312 * @size: allocation size
314 * Allocate enough 32bit PA addressable pages to cover @size from the
315 * page level allocator and map them into contiguous kernel virtual space.
317 void *vmalloc_32(unsigned long size
)
319 return __vmalloc(size
, GFP_KERNEL
);
321 EXPORT_SYMBOL(vmalloc_32
);
324 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
325 * @size: allocation size
327 * The resulting memory area is 32bit addressable and zeroed so it can be
328 * mapped to userspace without leaking data.
330 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
331 * remap_vmalloc_range() are permissible.
333 void *vmalloc_32_user(unsigned long size
)
336 * We'll have to sort out the ZONE_DMA bits for 64-bit,
337 * but for now this can simply use vmalloc_user() directly.
339 return vmalloc_user(size
);
341 EXPORT_SYMBOL(vmalloc_32_user
);
343 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
350 void vunmap(const void *addr
)
354 EXPORT_SYMBOL(vunmap
);
356 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
)
361 EXPORT_SYMBOL(vm_map_ram
);
363 void vm_unmap_ram(const void *mem
, unsigned int count
)
367 EXPORT_SYMBOL(vm_unmap_ram
);
369 void vm_unmap_aliases(void)
372 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
374 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
379 EXPORT_SYMBOL_GPL(alloc_vm_area
);
381 void free_vm_area(struct vm_struct
*area
)
385 EXPORT_SYMBOL_GPL(free_vm_area
);
387 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
392 EXPORT_SYMBOL(vm_insert_page
);
394 int vm_map_pages(struct vm_area_struct
*vma
, struct page
**pages
,
399 EXPORT_SYMBOL(vm_map_pages
);
401 int vm_map_pages_zero(struct vm_area_struct
*vma
, struct page
**pages
,
406 EXPORT_SYMBOL(vm_map_pages_zero
);
409 * sys_brk() for the most part doesn't need the global kernel
410 * lock, except when an application is doing something nasty
411 * like trying to un-brk an area that has already been mapped
412 * to a regular file. in this case, the unmapping will need
413 * to invoke file system routines that need the global lock.
415 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
417 struct mm_struct
*mm
= current
->mm
;
419 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
426 * Always allow shrinking brk
428 if (brk
<= mm
->brk
) {
434 * Ok, looks good - let it rip.
436 flush_icache_user_range(mm
->brk
, brk
);
437 return mm
->brk
= brk
;
441 * initialise the percpu counter for VM and region record slabs
443 void __init
mmap_init(void)
447 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
449 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
453 * validate the region tree
454 * - the caller must hold the region lock
456 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
457 static noinline
void validate_nommu_regions(void)
459 struct vm_region
*region
, *last
;
460 struct rb_node
*p
, *lastp
;
462 lastp
= rb_first(&nommu_region_tree
);
466 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
467 BUG_ON(last
->vm_end
<= last
->vm_start
);
468 BUG_ON(last
->vm_top
< last
->vm_end
);
470 while ((p
= rb_next(lastp
))) {
471 region
= rb_entry(p
, struct vm_region
, vm_rb
);
472 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
474 BUG_ON(region
->vm_end
<= region
->vm_start
);
475 BUG_ON(region
->vm_top
< region
->vm_end
);
476 BUG_ON(region
->vm_start
< last
->vm_top
);
482 static void validate_nommu_regions(void)
488 * add a region into the global tree
490 static void add_nommu_region(struct vm_region
*region
)
492 struct vm_region
*pregion
;
493 struct rb_node
**p
, *parent
;
495 validate_nommu_regions();
498 p
= &nommu_region_tree
.rb_node
;
501 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
502 if (region
->vm_start
< pregion
->vm_start
)
504 else if (region
->vm_start
> pregion
->vm_start
)
506 else if (pregion
== region
)
512 rb_link_node(®ion
->vm_rb
, parent
, p
);
513 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
515 validate_nommu_regions();
519 * delete a region from the global tree
521 static void delete_nommu_region(struct vm_region
*region
)
523 BUG_ON(!nommu_region_tree
.rb_node
);
525 validate_nommu_regions();
526 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
527 validate_nommu_regions();
531 * free a contiguous series of pages
533 static void free_page_series(unsigned long from
, unsigned long to
)
535 for (; from
< to
; from
+= PAGE_SIZE
) {
536 struct page
*page
= virt_to_page(from
);
538 atomic_long_dec(&mmap_pages_allocated
);
544 * release a reference to a region
545 * - the caller must hold the region semaphore for writing, which this releases
546 * - the region may not have been added to the tree yet, in which case vm_top
547 * will equal vm_start
549 static void __put_nommu_region(struct vm_region
*region
)
550 __releases(nommu_region_sem
)
552 BUG_ON(!nommu_region_tree
.rb_node
);
554 if (--region
->vm_usage
== 0) {
555 if (region
->vm_top
> region
->vm_start
)
556 delete_nommu_region(region
);
557 up_write(&nommu_region_sem
);
560 fput(region
->vm_file
);
562 /* IO memory and memory shared directly out of the pagecache
563 * from ramfs/tmpfs mustn't be released here */
564 if (region
->vm_flags
& VM_MAPPED_COPY
)
565 free_page_series(region
->vm_start
, region
->vm_top
);
566 kmem_cache_free(vm_region_jar
, region
);
568 up_write(&nommu_region_sem
);
573 * release a reference to a region
575 static void put_nommu_region(struct vm_region
*region
)
577 down_write(&nommu_region_sem
);
578 __put_nommu_region(region
);
582 * add a VMA into a process's mm_struct in the appropriate place in the list
583 * and tree and add to the address space's page tree also if not an anonymous
585 * - should be called with mm->mmap_lock held writelocked
587 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
589 struct vm_area_struct
*pvma
, *prev
;
590 struct address_space
*mapping
;
591 struct rb_node
**p
, *parent
, *rb_prev
;
593 BUG_ON(!vma
->vm_region
);
598 /* add the VMA to the mapping */
600 mapping
= vma
->vm_file
->f_mapping
;
602 i_mmap_lock_write(mapping
);
603 flush_dcache_mmap_lock(mapping
);
604 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
605 flush_dcache_mmap_unlock(mapping
);
606 i_mmap_unlock_write(mapping
);
609 /* add the VMA to the tree */
610 parent
= rb_prev
= NULL
;
611 p
= &mm
->mm_rb
.rb_node
;
614 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
616 /* sort by: start addr, end addr, VMA struct addr in that order
617 * (the latter is necessary as we may get identical VMAs) */
618 if (vma
->vm_start
< pvma
->vm_start
)
620 else if (vma
->vm_start
> pvma
->vm_start
) {
623 } else if (vma
->vm_end
< pvma
->vm_end
)
625 else if (vma
->vm_end
> pvma
->vm_end
) {
628 } else if (vma
< pvma
)
630 else if (vma
> pvma
) {
637 rb_link_node(&vma
->vm_rb
, parent
, p
);
638 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
640 /* add VMA to the VMA list also */
643 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
645 __vma_link_list(mm
, vma
, prev
);
649 * delete a VMA from its owning mm_struct and address space
651 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
654 struct address_space
*mapping
;
655 struct mm_struct
*mm
= vma
->vm_mm
;
656 struct task_struct
*curr
= current
;
659 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
660 /* if the vma is cached, invalidate the entire cache */
661 if (curr
->vmacache
.vmas
[i
] == vma
) {
662 vmacache_invalidate(mm
);
667 /* remove the VMA from the mapping */
669 mapping
= vma
->vm_file
->f_mapping
;
671 i_mmap_lock_write(mapping
);
672 flush_dcache_mmap_lock(mapping
);
673 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
674 flush_dcache_mmap_unlock(mapping
);
675 i_mmap_unlock_write(mapping
);
678 /* remove from the MM's tree and list */
679 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
681 __vma_unlink_list(mm
, vma
);
685 * destroy a VMA record
687 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
689 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
690 vma
->vm_ops
->close(vma
);
693 put_nommu_region(vma
->vm_region
);
698 * look up the first VMA in which addr resides, NULL if none
699 * - should be called with mm->mmap_lock at least held readlocked
701 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
703 struct vm_area_struct
*vma
;
705 /* check the cache first */
706 vma
= vmacache_find(mm
, addr
);
710 /* trawl the list (there may be multiple mappings in which addr
712 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
713 if (vma
->vm_start
> addr
)
715 if (vma
->vm_end
> addr
) {
716 vmacache_update(addr
, vma
);
723 EXPORT_SYMBOL(find_vma
);
727 * - we don't extend stack VMAs under NOMMU conditions
729 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
731 return find_vma(mm
, addr
);
735 * expand a stack to a given address
736 * - not supported under NOMMU conditions
738 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
744 * look up the first VMA exactly that exactly matches addr
745 * - should be called with mm->mmap_lock at least held readlocked
747 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
751 struct vm_area_struct
*vma
;
752 unsigned long end
= addr
+ len
;
754 /* check the cache first */
755 vma
= vmacache_find_exact(mm
, addr
, end
);
759 /* trawl the list (there may be multiple mappings in which addr
761 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
762 if (vma
->vm_start
< addr
)
764 if (vma
->vm_start
> addr
)
766 if (vma
->vm_end
== end
) {
767 vmacache_update(addr
, vma
);
776 * determine whether a mapping should be permitted and, if so, what sort of
777 * mapping we're capable of supporting
779 static int validate_mmap_request(struct file
*file
,
785 unsigned long *_capabilities
)
787 unsigned long capabilities
, rlen
;
790 /* do the simple checks first */
791 if (flags
& MAP_FIXED
)
794 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
795 (flags
& MAP_TYPE
) != MAP_SHARED
)
801 /* Careful about overflows.. */
802 rlen
= PAGE_ALIGN(len
);
803 if (!rlen
|| rlen
> TASK_SIZE
)
806 /* offset overflow? */
807 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
811 /* files must support mmap */
812 if (!file
->f_op
->mmap
)
815 /* work out if what we've got could possibly be shared
816 * - we support chardevs that provide their own "memory"
817 * - we support files/blockdevs that are memory backed
819 if (file
->f_op
->mmap_capabilities
) {
820 capabilities
= file
->f_op
->mmap_capabilities(file
);
822 /* no explicit capabilities set, so assume some
824 switch (file_inode(file
)->i_mode
& S_IFMT
) {
827 capabilities
= NOMMU_MAP_COPY
;
842 /* eliminate any capabilities that we can't support on this
844 if (!file
->f_op
->get_unmapped_area
)
845 capabilities
&= ~NOMMU_MAP_DIRECT
;
846 if (!(file
->f_mode
& FMODE_CAN_READ
))
847 capabilities
&= ~NOMMU_MAP_COPY
;
849 /* The file shall have been opened with read permission. */
850 if (!(file
->f_mode
& FMODE_READ
))
853 if (flags
& MAP_SHARED
) {
854 /* do checks for writing, appending and locking */
855 if ((prot
& PROT_WRITE
) &&
856 !(file
->f_mode
& FMODE_WRITE
))
859 if (IS_APPEND(file_inode(file
)) &&
860 (file
->f_mode
& FMODE_WRITE
))
863 if (locks_verify_locked(file
))
866 if (!(capabilities
& NOMMU_MAP_DIRECT
))
869 /* we mustn't privatise shared mappings */
870 capabilities
&= ~NOMMU_MAP_COPY
;
872 /* we're going to read the file into private memory we
874 if (!(capabilities
& NOMMU_MAP_COPY
))
877 /* we don't permit a private writable mapping to be
878 * shared with the backing device */
879 if (prot
& PROT_WRITE
)
880 capabilities
&= ~NOMMU_MAP_DIRECT
;
883 if (capabilities
& NOMMU_MAP_DIRECT
) {
884 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
885 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
886 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
888 capabilities
&= ~NOMMU_MAP_DIRECT
;
889 if (flags
& MAP_SHARED
) {
890 pr_warn("MAP_SHARED not completely supported on !MMU\n");
896 /* handle executable mappings and implied executable
898 if (path_noexec(&file
->f_path
)) {
899 if (prot
& PROT_EXEC
)
901 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
902 /* handle implication of PROT_EXEC by PROT_READ */
903 if (current
->personality
& READ_IMPLIES_EXEC
) {
904 if (capabilities
& NOMMU_MAP_EXEC
)
907 } else if ((prot
& PROT_READ
) &&
908 (prot
& PROT_EXEC
) &&
909 !(capabilities
& NOMMU_MAP_EXEC
)
911 /* backing file is not executable, try to copy */
912 capabilities
&= ~NOMMU_MAP_DIRECT
;
915 /* anonymous mappings are always memory backed and can be
918 capabilities
= NOMMU_MAP_COPY
;
920 /* handle PROT_EXEC implication by PROT_READ */
921 if ((prot
& PROT_READ
) &&
922 (current
->personality
& READ_IMPLIES_EXEC
))
926 /* allow the security API to have its say */
927 ret
= security_mmap_addr(addr
);
932 *_capabilities
= capabilities
;
937 * we've determined that we can make the mapping, now translate what we
938 * now know into VMA flags
940 static unsigned long determine_vm_flags(struct file
*file
,
943 unsigned long capabilities
)
945 unsigned long vm_flags
;
947 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
948 /* vm_flags |= mm->def_flags; */
950 if (!(capabilities
& NOMMU_MAP_DIRECT
)) {
951 /* attempt to share read-only copies of mapped file chunks */
952 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
953 if (file
&& !(prot
& PROT_WRITE
))
954 vm_flags
|= VM_MAYSHARE
;
956 /* overlay a shareable mapping on the backing device or inode
957 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
959 vm_flags
|= VM_MAYSHARE
| (capabilities
& NOMMU_VMFLAGS
);
960 if (flags
& MAP_SHARED
)
961 vm_flags
|= VM_SHARED
;
964 /* refuse to let anyone share private mappings with this process if
965 * it's being traced - otherwise breakpoints set in it may interfere
966 * with another untraced process
968 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
969 vm_flags
&= ~VM_MAYSHARE
;
975 * set up a shared mapping on a file (the driver or filesystem provides and
978 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
982 ret
= call_mmap(vma
->vm_file
, vma
);
984 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
990 /* getting -ENOSYS indicates that direct mmap isn't possible (as
991 * opposed to tried but failed) so we can only give a suitable error as
992 * it's not possible to make a private copy if MAP_SHARED was given */
997 * set up a private mapping or an anonymous shared mapping
999 static int do_mmap_private(struct vm_area_struct
*vma
,
1000 struct vm_region
*region
,
1002 unsigned long capabilities
)
1004 unsigned long total
, point
;
1008 /* invoke the file's mapping function so that it can keep track of
1009 * shared mappings on devices or memory
1010 * - VM_MAYSHARE will be set if it may attempt to share
1012 if (capabilities
& NOMMU_MAP_DIRECT
) {
1013 ret
= call_mmap(vma
->vm_file
, vma
);
1015 /* shouldn't return success if we're not sharing */
1016 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1017 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1023 /* getting an ENOSYS error indicates that direct mmap isn't
1024 * possible (as opposed to tried but failed) so we'll try to
1025 * make a private copy of the data and map that instead */
1029 /* allocate some memory to hold the mapping
1030 * - note that this may not return a page-aligned address if the object
1031 * we're allocating is smaller than a page
1033 order
= get_order(len
);
1035 point
= len
>> PAGE_SHIFT
;
1037 /* we don't want to allocate a power-of-2 sized page set */
1038 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
1041 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
1045 atomic_long_add(total
, &mmap_pages_allocated
);
1047 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1048 region
->vm_start
= (unsigned long) base
;
1049 region
->vm_end
= region
->vm_start
+ len
;
1050 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1052 vma
->vm_start
= region
->vm_start
;
1053 vma
->vm_end
= region
->vm_start
+ len
;
1056 /* read the contents of a file into the copy */
1059 fpos
= vma
->vm_pgoff
;
1060 fpos
<<= PAGE_SHIFT
;
1062 ret
= kernel_read(vma
->vm_file
, base
, len
, &fpos
);
1066 /* clear the last little bit */
1068 memset(base
+ ret
, 0, len
- ret
);
1071 vma_set_anonymous(vma
);
1077 free_page_series(region
->vm_start
, region
->vm_top
);
1078 region
->vm_start
= vma
->vm_start
= 0;
1079 region
->vm_end
= vma
->vm_end
= 0;
1084 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1085 len
, current
->pid
, current
->comm
);
1086 show_free_areas(0, NULL
);
1091 * handle mapping creation for uClinux
1093 unsigned long do_mmap(struct file
*file
,
1097 unsigned long flags
,
1098 vm_flags_t vm_flags
,
1099 unsigned long pgoff
,
1100 unsigned long *populate
,
1101 struct list_head
*uf
)
1103 struct vm_area_struct
*vma
;
1104 struct vm_region
*region
;
1106 unsigned long capabilities
, result
;
1111 /* decide whether we should attempt the mapping, and if so what sort of
1113 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1118 /* we ignore the address hint */
1120 len
= PAGE_ALIGN(len
);
1122 /* we've determined that we can make the mapping, now translate what we
1123 * now know into VMA flags */
1124 vm_flags
|= determine_vm_flags(file
, prot
, flags
, capabilities
);
1126 /* we're going to need to record the mapping */
1127 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1129 goto error_getting_region
;
1131 vma
= vm_area_alloc(current
->mm
);
1133 goto error_getting_vma
;
1135 region
->vm_usage
= 1;
1136 region
->vm_flags
= vm_flags
;
1137 region
->vm_pgoff
= pgoff
;
1139 vma
->vm_flags
= vm_flags
;
1140 vma
->vm_pgoff
= pgoff
;
1143 region
->vm_file
= get_file(file
);
1144 vma
->vm_file
= get_file(file
);
1147 down_write(&nommu_region_sem
);
1149 /* if we want to share, we need to check for regions created by other
1150 * mmap() calls that overlap with our proposed mapping
1151 * - we can only share with a superset match on most regular files
1152 * - shared mappings on character devices and memory backed files are
1153 * permitted to overlap inexactly as far as we are concerned for in
1154 * these cases, sharing is handled in the driver or filesystem rather
1157 if (vm_flags
& VM_MAYSHARE
) {
1158 struct vm_region
*pregion
;
1159 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1161 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1162 pgend
= pgoff
+ pglen
;
1164 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1165 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1167 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1170 /* search for overlapping mappings on the same file */
1171 if (file_inode(pregion
->vm_file
) !=
1175 if (pregion
->vm_pgoff
>= pgend
)
1178 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1179 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1180 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1181 if (pgoff
>= rpgend
)
1184 /* handle inexactly overlapping matches between
1186 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1187 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1188 /* new mapping is not a subset of the region */
1189 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1190 goto sharing_violation
;
1194 /* we've found a region we can share */
1195 pregion
->vm_usage
++;
1196 vma
->vm_region
= pregion
;
1197 start
= pregion
->vm_start
;
1198 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1199 vma
->vm_start
= start
;
1200 vma
->vm_end
= start
+ len
;
1202 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1203 vma
->vm_flags
|= VM_MAPPED_COPY
;
1205 ret
= do_mmap_shared_file(vma
);
1207 vma
->vm_region
= NULL
;
1210 pregion
->vm_usage
--;
1212 goto error_just_free
;
1215 fput(region
->vm_file
);
1216 kmem_cache_free(vm_region_jar
, region
);
1222 /* obtain the address at which to make a shared mapping
1223 * - this is the hook for quasi-memory character devices to
1224 * tell us the location of a shared mapping
1226 if (capabilities
& NOMMU_MAP_DIRECT
) {
1227 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1229 if (IS_ERR_VALUE(addr
)) {
1232 goto error_just_free
;
1234 /* the driver refused to tell us where to site
1235 * the mapping so we'll have to attempt to copy
1238 if (!(capabilities
& NOMMU_MAP_COPY
))
1239 goto error_just_free
;
1241 capabilities
&= ~NOMMU_MAP_DIRECT
;
1243 vma
->vm_start
= region
->vm_start
= addr
;
1244 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1249 vma
->vm_region
= region
;
1251 /* set up the mapping
1252 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1254 if (file
&& vma
->vm_flags
& VM_SHARED
)
1255 ret
= do_mmap_shared_file(vma
);
1257 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1259 goto error_just_free
;
1260 add_nommu_region(region
);
1262 /* clear anonymous mappings that don't ask for uninitialized data */
1263 if (!vma
->vm_file
&&
1264 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED
) ||
1265 !(flags
& MAP_UNINITIALIZED
)))
1266 memset((void *)region
->vm_start
, 0,
1267 region
->vm_end
- region
->vm_start
);
1269 /* okay... we have a mapping; now we have to register it */
1270 result
= vma
->vm_start
;
1272 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1275 add_vma_to_mm(current
->mm
, vma
);
1277 /* we flush the region from the icache only when the first executable
1278 * mapping of it is made */
1279 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1280 flush_icache_user_range(region
->vm_start
, region
->vm_end
);
1281 region
->vm_icache_flushed
= true;
1284 up_write(&nommu_region_sem
);
1289 up_write(&nommu_region_sem
);
1291 if (region
->vm_file
)
1292 fput(region
->vm_file
);
1293 kmem_cache_free(vm_region_jar
, region
);
1300 up_write(&nommu_region_sem
);
1301 pr_warn("Attempt to share mismatched mappings\n");
1306 kmem_cache_free(vm_region_jar
, region
);
1307 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1309 show_free_areas(0, NULL
);
1312 error_getting_region
:
1313 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1315 show_free_areas(0, NULL
);
1319 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1320 unsigned long prot
, unsigned long flags
,
1321 unsigned long fd
, unsigned long pgoff
)
1323 struct file
*file
= NULL
;
1324 unsigned long retval
= -EBADF
;
1326 audit_mmap_fd(fd
, flags
);
1327 if (!(flags
& MAP_ANONYMOUS
)) {
1333 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1335 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1343 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1344 unsigned long, prot
, unsigned long, flags
,
1345 unsigned long, fd
, unsigned long, pgoff
)
1347 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1350 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1351 struct mmap_arg_struct
{
1355 unsigned long flags
;
1357 unsigned long offset
;
1360 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1362 struct mmap_arg_struct a
;
1364 if (copy_from_user(&a
, arg
, sizeof(a
)))
1366 if (offset_in_page(a
.offset
))
1369 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1370 a
.offset
>> PAGE_SHIFT
);
1372 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1375 * split a vma into two pieces at address 'addr', a new vma is allocated either
1376 * for the first part or the tail.
1378 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1379 unsigned long addr
, int new_below
)
1381 struct vm_area_struct
*new;
1382 struct vm_region
*region
;
1383 unsigned long npages
;
1385 /* we're only permitted to split anonymous regions (these should have
1386 * only a single usage on the region) */
1390 if (mm
->map_count
>= sysctl_max_map_count
)
1393 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1397 new = vm_area_dup(vma
);
1399 kmem_cache_free(vm_region_jar
, region
);
1403 /* most fields are the same, copy all, and then fixup */
1404 *region
= *vma
->vm_region
;
1405 new->vm_region
= region
;
1407 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1410 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1412 region
->vm_start
= new->vm_start
= addr
;
1413 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1416 if (new->vm_ops
&& new->vm_ops
->open
)
1417 new->vm_ops
->open(new);
1419 delete_vma_from_mm(vma
);
1420 down_write(&nommu_region_sem
);
1421 delete_nommu_region(vma
->vm_region
);
1423 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1424 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1426 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1427 vma
->vm_region
->vm_top
= addr
;
1429 add_nommu_region(vma
->vm_region
);
1430 add_nommu_region(new->vm_region
);
1431 up_write(&nommu_region_sem
);
1432 add_vma_to_mm(mm
, vma
);
1433 add_vma_to_mm(mm
, new);
1438 * shrink a VMA by removing the specified chunk from either the beginning or
1441 static int shrink_vma(struct mm_struct
*mm
,
1442 struct vm_area_struct
*vma
,
1443 unsigned long from
, unsigned long to
)
1445 struct vm_region
*region
;
1447 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1449 delete_vma_from_mm(vma
);
1450 if (from
> vma
->vm_start
)
1454 add_vma_to_mm(mm
, vma
);
1456 /* cut the backing region down to size */
1457 region
= vma
->vm_region
;
1458 BUG_ON(region
->vm_usage
!= 1);
1460 down_write(&nommu_region_sem
);
1461 delete_nommu_region(region
);
1462 if (from
> region
->vm_start
) {
1463 to
= region
->vm_top
;
1464 region
->vm_top
= region
->vm_end
= from
;
1466 region
->vm_start
= to
;
1468 add_nommu_region(region
);
1469 up_write(&nommu_region_sem
);
1471 free_page_series(from
, to
);
1477 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1478 * VMA, though it need not cover the whole VMA
1480 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1482 struct vm_area_struct
*vma
;
1486 len
= PAGE_ALIGN(len
);
1492 /* find the first potentially overlapping VMA */
1493 vma
= find_vma(mm
, start
);
1497 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1498 current
->pid
, current
->comm
,
1499 start
, start
+ len
- 1);
1505 /* we're allowed to split an anonymous VMA but not a file-backed one */
1508 if (start
> vma
->vm_start
)
1510 if (end
== vma
->vm_end
)
1511 goto erase_whole_vma
;
1516 /* the chunk must be a subset of the VMA found */
1517 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1518 goto erase_whole_vma
;
1519 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1521 if (offset_in_page(start
))
1523 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1525 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1526 ret
= split_vma(mm
, vma
, start
, 1);
1530 return shrink_vma(mm
, vma
, start
, end
);
1534 delete_vma_from_mm(vma
);
1535 delete_vma(mm
, vma
);
1538 EXPORT_SYMBOL(do_munmap
);
1540 int vm_munmap(unsigned long addr
, size_t len
)
1542 struct mm_struct
*mm
= current
->mm
;
1545 mmap_write_lock(mm
);
1546 ret
= do_munmap(mm
, addr
, len
, NULL
);
1547 mmap_write_unlock(mm
);
1550 EXPORT_SYMBOL(vm_munmap
);
1552 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1554 return vm_munmap(addr
, len
);
1558 * release all the mappings made in a process's VM space
1560 void exit_mmap(struct mm_struct
*mm
)
1562 struct vm_area_struct
*vma
;
1569 while ((vma
= mm
->mmap
)) {
1570 mm
->mmap
= vma
->vm_next
;
1571 delete_vma_from_mm(vma
);
1572 delete_vma(mm
, vma
);
1577 int vm_brk(unsigned long addr
, unsigned long len
)
1583 * expand (or shrink) an existing mapping, potentially moving it at the same
1584 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1586 * under NOMMU conditions, we only permit changing a mapping's size, and only
1587 * as long as it stays within the region allocated by do_mmap_private() and the
1588 * block is not shareable
1590 * MREMAP_FIXED is not supported under NOMMU conditions
1592 static unsigned long do_mremap(unsigned long addr
,
1593 unsigned long old_len
, unsigned long new_len
,
1594 unsigned long flags
, unsigned long new_addr
)
1596 struct vm_area_struct
*vma
;
1598 /* insanity checks first */
1599 old_len
= PAGE_ALIGN(old_len
);
1600 new_len
= PAGE_ALIGN(new_len
);
1601 if (old_len
== 0 || new_len
== 0)
1602 return (unsigned long) -EINVAL
;
1604 if (offset_in_page(addr
))
1607 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1608 return (unsigned long) -EINVAL
;
1610 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1612 return (unsigned long) -EINVAL
;
1614 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1615 return (unsigned long) -EFAULT
;
1617 if (vma
->vm_flags
& VM_MAYSHARE
)
1618 return (unsigned long) -EPERM
;
1620 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1621 return (unsigned long) -ENOMEM
;
1623 /* all checks complete - do it */
1624 vma
->vm_end
= vma
->vm_start
+ new_len
;
1625 return vma
->vm_start
;
1628 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1629 unsigned long, new_len
, unsigned long, flags
,
1630 unsigned long, new_addr
)
1634 mmap_write_lock(current
->mm
);
1635 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1636 mmap_write_unlock(current
->mm
);
1640 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1641 unsigned int foll_flags
)
1646 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1647 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1649 if (addr
!= (pfn
<< PAGE_SHIFT
))
1652 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1655 EXPORT_SYMBOL(remap_pfn_range
);
1657 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1659 unsigned long pfn
= start
>> PAGE_SHIFT
;
1660 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1662 pfn
+= vma
->vm_pgoff
;
1663 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1665 EXPORT_SYMBOL(vm_iomap_memory
);
1667 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1668 unsigned long pgoff
)
1670 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1672 if (!(vma
->vm_flags
& VM_USERMAP
))
1675 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1676 vma
->vm_end
= vma
->vm_start
+ size
;
1680 EXPORT_SYMBOL(remap_vmalloc_range
);
1682 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1683 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1688 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
1693 EXPORT_SYMBOL(filemap_fault
);
1695 void filemap_map_pages(struct vm_fault
*vmf
,
1696 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1700 EXPORT_SYMBOL(filemap_map_pages
);
1702 int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
1703 unsigned long addr
, void *buf
, int len
, unsigned int gup_flags
)
1705 struct vm_area_struct
*vma
;
1706 int write
= gup_flags
& FOLL_WRITE
;
1708 if (mmap_read_lock_killable(mm
))
1711 /* the access must start within one of the target process's mappings */
1712 vma
= find_vma(mm
, addr
);
1714 /* don't overrun this mapping */
1715 if (addr
+ len
>= vma
->vm_end
)
1716 len
= vma
->vm_end
- addr
;
1718 /* only read or write mappings where it is permitted */
1719 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1720 copy_to_user_page(vma
, NULL
, addr
,
1721 (void *) addr
, buf
, len
);
1722 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1723 copy_from_user_page(vma
, NULL
, addr
,
1724 buf
, (void *) addr
, len
);
1731 mmap_read_unlock(mm
);
1737 * access_remote_vm - access another process' address space
1738 * @mm: the mm_struct of the target address space
1739 * @addr: start address to access
1740 * @buf: source or destination buffer
1741 * @len: number of bytes to transfer
1742 * @gup_flags: flags modifying lookup behaviour
1744 * The caller must hold a reference on @mm.
1746 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1747 void *buf
, int len
, unsigned int gup_flags
)
1749 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, gup_flags
);
1753 * Access another process' address space.
1754 * - source/target buffer must be kernel space
1756 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1757 unsigned int gup_flags
)
1759 struct mm_struct
*mm
;
1761 if (addr
+ len
< addr
)
1764 mm
= get_task_mm(tsk
);
1768 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, gup_flags
);
1773 EXPORT_SYMBOL_GPL(access_process_vm
);
1776 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1777 * @inode: The inode to check
1778 * @size: The current filesize of the inode
1779 * @newsize: The proposed filesize of the inode
1781 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1782 * make sure that that any outstanding VMAs aren't broken and then shrink the
1783 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1784 * automatically grant mappings that are too large.
1786 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1789 struct vm_area_struct
*vma
;
1790 struct vm_region
*region
;
1792 size_t r_size
, r_top
;
1794 low
= newsize
>> PAGE_SHIFT
;
1795 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1797 down_write(&nommu_region_sem
);
1798 i_mmap_lock_read(inode
->i_mapping
);
1800 /* search for VMAs that fall within the dead zone */
1801 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1802 /* found one - only interested if it's shared out of the page
1804 if (vma
->vm_flags
& VM_SHARED
) {
1805 i_mmap_unlock_read(inode
->i_mapping
);
1806 up_write(&nommu_region_sem
);
1807 return -ETXTBSY
; /* not quite true, but near enough */
1811 /* reduce any regions that overlap the dead zone - if in existence,
1812 * these will be pointed to by VMAs that don't overlap the dead zone
1814 * we don't check for any regions that start beyond the EOF as there
1817 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1818 if (!(vma
->vm_flags
& VM_SHARED
))
1821 region
= vma
->vm_region
;
1822 r_size
= region
->vm_top
- region
->vm_start
;
1823 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1825 if (r_top
> newsize
) {
1826 region
->vm_top
-= r_top
- newsize
;
1827 if (region
->vm_end
> region
->vm_top
)
1828 region
->vm_end
= region
->vm_top
;
1832 i_mmap_unlock_read(inode
->i_mapping
);
1833 up_write(&nommu_region_sem
);
1838 * Initialise sysctl_user_reserve_kbytes.
1840 * This is intended to prevent a user from starting a single memory hogging
1841 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1844 * The default value is min(3% of free memory, 128MB)
1845 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1847 static int __meminit
init_user_reserve(void)
1849 unsigned long free_kbytes
;
1851 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1853 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1856 subsys_initcall(init_user_reserve
);
1859 * Initialise sysctl_admin_reserve_kbytes.
1861 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1862 * to log in and kill a memory hogging process.
1864 * Systems with more than 256MB will reserve 8MB, enough to recover
1865 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1866 * only reserve 3% of free pages by default.
1868 static int __meminit
init_admin_reserve(void)
1870 unsigned long free_kbytes
;
1872 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1874 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1877 subsys_initcall(init_admin_reserve
);