4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
7 * See Documentation/nommu-mmap.txt
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/export.h>
20 #include <linux/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/printk.h>
38 #include <linux/uaccess.h>
40 #include <asm/tlbflush.h>
41 #include <asm/mmu_context.h>
45 EXPORT_SYMBOL(high_memory
);
47 unsigned long max_mapnr
;
48 EXPORT_SYMBOL(max_mapnr
);
49 unsigned long highest_memmap_pfn
;
50 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
51 int heap_stack_gap
= 0;
53 atomic_long_t mmap_pages_allocated
;
55 EXPORT_SYMBOL(mem_map
);
57 /* list of mapped, potentially shareable regions */
58 static struct kmem_cache
*vm_region_jar
;
59 struct rb_root nommu_region_tree
= RB_ROOT
;
60 DECLARE_RWSEM(nommu_region_sem
);
62 const struct vm_operations_struct generic_file_vm_ops
= {
66 * Return the total memory allocated for this pointer, not
67 * just what the caller asked for.
69 * Doesn't have to be accurate, i.e. may have races.
71 unsigned int kobjsize(const void *objp
)
76 * If the object we have should not have ksize performed on it,
79 if (!objp
|| !virt_addr_valid(objp
))
82 page
= virt_to_head_page(objp
);
85 * If the allocator sets PageSlab, we know the pointer came from
92 * If it's not a compound page, see if we have a matching VMA
93 * region. This test is intentionally done in reverse order,
94 * so if there's no VMA, we still fall through and hand back
95 * PAGE_SIZE for 0-order pages.
97 if (!PageCompound(page
)) {
98 struct vm_area_struct
*vma
;
100 vma
= find_vma(current
->mm
, (unsigned long)objp
);
102 return vma
->vm_end
- vma
->vm_start
;
106 * The ksize() function is only guaranteed to work for pointers
107 * returned by kmalloc(). So handle arbitrary pointers here.
109 return PAGE_SIZE
<< compound_order(page
);
112 static long __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
113 unsigned long start
, unsigned long nr_pages
,
114 unsigned int foll_flags
, struct page
**pages
,
115 struct vm_area_struct
**vmas
, int *nonblocking
)
117 struct vm_area_struct
*vma
;
118 unsigned long vm_flags
;
121 /* calculate required read or write permissions.
122 * If FOLL_FORCE is set, we only require the "MAY" flags.
124 vm_flags
= (foll_flags
& FOLL_WRITE
) ?
125 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
126 vm_flags
&= (foll_flags
& FOLL_FORCE
) ?
127 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
129 for (i
= 0; i
< nr_pages
; i
++) {
130 vma
= find_vma(mm
, start
);
132 goto finish_or_fault
;
134 /* protect what we can, including chardevs */
135 if ((vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
136 !(vm_flags
& vma
->vm_flags
))
137 goto finish_or_fault
;
140 pages
[i
] = virt_to_page(start
);
146 start
= (start
+ PAGE_SIZE
) & PAGE_MASK
;
152 return i
? : -EFAULT
;
156 * get a list of pages in an address range belonging to the specified process
157 * and indicate the VMA that covers each page
158 * - this is potentially dodgy as we may end incrementing the page count of a
159 * slab page or a secondary page from a compound page
160 * - don't permit access to VMAs that don't support it, such as I/O mappings
162 long get_user_pages(unsigned long start
, unsigned long nr_pages
,
163 unsigned int gup_flags
, struct page
**pages
,
164 struct vm_area_struct
**vmas
)
166 return __get_user_pages(current
, current
->mm
, start
, nr_pages
,
167 gup_flags
, pages
, vmas
, NULL
);
169 EXPORT_SYMBOL(get_user_pages
);
171 long get_user_pages_locked(unsigned long start
, unsigned long nr_pages
,
172 unsigned int gup_flags
, struct page
**pages
,
175 return get_user_pages(start
, nr_pages
, gup_flags
, pages
, NULL
);
177 EXPORT_SYMBOL(get_user_pages_locked
);
179 static long __get_user_pages_unlocked(struct task_struct
*tsk
,
180 struct mm_struct
*mm
, unsigned long start
,
181 unsigned long nr_pages
, struct page
**pages
,
182 unsigned int gup_flags
)
185 down_read(&mm
->mmap_sem
);
186 ret
= __get_user_pages(tsk
, mm
, start
, nr_pages
, gup_flags
, pages
,
188 up_read(&mm
->mmap_sem
);
192 long get_user_pages_unlocked(unsigned long start
, unsigned long nr_pages
,
193 struct page
**pages
, unsigned int gup_flags
)
195 return __get_user_pages_unlocked(current
, current
->mm
, start
, nr_pages
,
198 EXPORT_SYMBOL(get_user_pages_unlocked
);
201 * follow_pfn - look up PFN at a user virtual address
202 * @vma: memory mapping
203 * @address: user virtual address
204 * @pfn: location to store found PFN
206 * Only IO mappings and raw PFN mappings are allowed.
208 * Returns zero and the pfn at @pfn on success, -ve otherwise.
210 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
213 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
216 *pfn
= address
>> PAGE_SHIFT
;
219 EXPORT_SYMBOL(follow_pfn
);
221 LIST_HEAD(vmap_area_list
);
223 void vfree(const void *addr
)
227 EXPORT_SYMBOL(vfree
);
229 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
232 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
233 * returns only a logical address.
235 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
237 EXPORT_SYMBOL(__vmalloc
);
239 void *vmalloc_user(unsigned long size
)
243 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
246 struct vm_area_struct
*vma
;
248 down_write(¤t
->mm
->mmap_sem
);
249 vma
= find_vma(current
->mm
, (unsigned long)ret
);
251 vma
->vm_flags
|= VM_USERMAP
;
252 up_write(¤t
->mm
->mmap_sem
);
257 EXPORT_SYMBOL(vmalloc_user
);
259 struct page
*vmalloc_to_page(const void *addr
)
261 return virt_to_page(addr
);
263 EXPORT_SYMBOL(vmalloc_to_page
);
265 unsigned long vmalloc_to_pfn(const void *addr
)
267 return page_to_pfn(virt_to_page(addr
));
269 EXPORT_SYMBOL(vmalloc_to_pfn
);
271 long vread(char *buf
, char *addr
, unsigned long count
)
273 /* Don't allow overflow */
274 if ((unsigned long) buf
+ count
< count
)
275 count
= -(unsigned long) buf
;
277 memcpy(buf
, addr
, count
);
281 long vwrite(char *buf
, char *addr
, unsigned long count
)
283 /* Don't allow overflow */
284 if ((unsigned long) addr
+ count
< count
)
285 count
= -(unsigned long) addr
;
287 memcpy(addr
, buf
, count
);
292 * vmalloc - allocate virtually contiguous memory
294 * @size: allocation size
296 * Allocate enough pages to cover @size from the page level
297 * allocator and map them into contiguous kernel virtual space.
299 * For tight control over page level allocator and protection flags
300 * use __vmalloc() instead.
302 void *vmalloc(unsigned long size
)
304 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
306 EXPORT_SYMBOL(vmalloc
);
309 * vzalloc - allocate virtually contiguous memory with zero fill
311 * @size: allocation size
313 * Allocate enough pages to cover @size from the page level
314 * allocator and map them into contiguous kernel virtual space.
315 * The memory allocated is set to zero.
317 * For tight control over page level allocator and protection flags
318 * use __vmalloc() instead.
320 void *vzalloc(unsigned long size
)
322 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
325 EXPORT_SYMBOL(vzalloc
);
328 * vmalloc_node - allocate memory on a specific node
329 * @size: allocation size
332 * Allocate enough pages to cover @size from the page level
333 * allocator and map them into contiguous kernel virtual space.
335 * For tight control over page level allocator and protection flags
336 * use __vmalloc() instead.
338 void *vmalloc_node(unsigned long size
, int node
)
340 return vmalloc(size
);
342 EXPORT_SYMBOL(vmalloc_node
);
345 * vzalloc_node - allocate memory on a specific node with zero fill
346 * @size: allocation size
349 * Allocate enough pages to cover @size from the page level
350 * allocator and map them into contiguous kernel virtual space.
351 * The memory allocated is set to zero.
353 * For tight control over page level allocator and protection flags
354 * use __vmalloc() instead.
356 void *vzalloc_node(unsigned long size
, int node
)
358 return vzalloc(size
);
360 EXPORT_SYMBOL(vzalloc_node
);
362 #ifndef PAGE_KERNEL_EXEC
363 # define PAGE_KERNEL_EXEC PAGE_KERNEL
367 * vmalloc_exec - allocate virtually contiguous, executable memory
368 * @size: allocation size
370 * Kernel-internal function to allocate enough pages to cover @size
371 * the page level allocator and map them into contiguous and
372 * executable kernel virtual space.
374 * For tight control over page level allocator and protection flags
375 * use __vmalloc() instead.
378 void *vmalloc_exec(unsigned long size
)
380 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
384 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
385 * @size: allocation size
387 * Allocate enough 32bit PA addressable pages to cover @size from the
388 * page level allocator and map them into contiguous kernel virtual space.
390 void *vmalloc_32(unsigned long size
)
392 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
394 EXPORT_SYMBOL(vmalloc_32
);
397 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
398 * @size: allocation size
400 * The resulting memory area is 32bit addressable and zeroed so it can be
401 * mapped to userspace without leaking data.
403 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
404 * remap_vmalloc_range() are permissible.
406 void *vmalloc_32_user(unsigned long size
)
409 * We'll have to sort out the ZONE_DMA bits for 64-bit,
410 * but for now this can simply use vmalloc_user() directly.
412 return vmalloc_user(size
);
414 EXPORT_SYMBOL(vmalloc_32_user
);
416 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
423 void vunmap(const void *addr
)
427 EXPORT_SYMBOL(vunmap
);
429 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
434 EXPORT_SYMBOL(vm_map_ram
);
436 void vm_unmap_ram(const void *mem
, unsigned int count
)
440 EXPORT_SYMBOL(vm_unmap_ram
);
442 void vm_unmap_aliases(void)
445 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
448 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
451 void __weak
vmalloc_sync_all(void)
456 * alloc_vm_area - allocate a range of kernel address space
457 * @size: size of the area
459 * Returns: NULL on failure, vm_struct on success
461 * This function reserves a range of kernel address space, and
462 * allocates pagetables to map that range. No actual mappings
463 * are created. If the kernel address space is not shared
464 * between processes, it syncs the pagetable across all
467 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
472 EXPORT_SYMBOL_GPL(alloc_vm_area
);
474 void free_vm_area(struct vm_struct
*area
)
478 EXPORT_SYMBOL_GPL(free_vm_area
);
480 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
485 EXPORT_SYMBOL(vm_insert_page
);
488 * sys_brk() for the most part doesn't need the global kernel
489 * lock, except when an application is doing something nasty
490 * like trying to un-brk an area that has already been mapped
491 * to a regular file. in this case, the unmapping will need
492 * to invoke file system routines that need the global lock.
494 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
496 struct mm_struct
*mm
= current
->mm
;
498 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
505 * Always allow shrinking brk
507 if (brk
<= mm
->brk
) {
513 * Ok, looks good - let it rip.
515 flush_icache_range(mm
->brk
, brk
);
516 return mm
->brk
= brk
;
520 * initialise the percpu counter for VM and region record slabs
522 void __init
mmap_init(void)
526 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
528 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
532 * validate the region tree
533 * - the caller must hold the region lock
535 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
536 static noinline
void validate_nommu_regions(void)
538 struct vm_region
*region
, *last
;
539 struct rb_node
*p
, *lastp
;
541 lastp
= rb_first(&nommu_region_tree
);
545 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
546 BUG_ON(last
->vm_end
<= last
->vm_start
);
547 BUG_ON(last
->vm_top
< last
->vm_end
);
549 while ((p
= rb_next(lastp
))) {
550 region
= rb_entry(p
, struct vm_region
, vm_rb
);
551 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
553 BUG_ON(region
->vm_end
<= region
->vm_start
);
554 BUG_ON(region
->vm_top
< region
->vm_end
);
555 BUG_ON(region
->vm_start
< last
->vm_top
);
561 static void validate_nommu_regions(void)
567 * add a region into the global tree
569 static void add_nommu_region(struct vm_region
*region
)
571 struct vm_region
*pregion
;
572 struct rb_node
**p
, *parent
;
574 validate_nommu_regions();
577 p
= &nommu_region_tree
.rb_node
;
580 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
581 if (region
->vm_start
< pregion
->vm_start
)
583 else if (region
->vm_start
> pregion
->vm_start
)
585 else if (pregion
== region
)
591 rb_link_node(®ion
->vm_rb
, parent
, p
);
592 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
594 validate_nommu_regions();
598 * delete a region from the global tree
600 static void delete_nommu_region(struct vm_region
*region
)
602 BUG_ON(!nommu_region_tree
.rb_node
);
604 validate_nommu_regions();
605 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
606 validate_nommu_regions();
610 * free a contiguous series of pages
612 static void free_page_series(unsigned long from
, unsigned long to
)
614 for (; from
< to
; from
+= PAGE_SIZE
) {
615 struct page
*page
= virt_to_page(from
);
617 atomic_long_dec(&mmap_pages_allocated
);
623 * release a reference to a region
624 * - the caller must hold the region semaphore for writing, which this releases
625 * - the region may not have been added to the tree yet, in which case vm_top
626 * will equal vm_start
628 static void __put_nommu_region(struct vm_region
*region
)
629 __releases(nommu_region_sem
)
631 BUG_ON(!nommu_region_tree
.rb_node
);
633 if (--region
->vm_usage
== 0) {
634 if (region
->vm_top
> region
->vm_start
)
635 delete_nommu_region(region
);
636 up_write(&nommu_region_sem
);
639 fput(region
->vm_file
);
641 /* IO memory and memory shared directly out of the pagecache
642 * from ramfs/tmpfs mustn't be released here */
643 if (region
->vm_flags
& VM_MAPPED_COPY
)
644 free_page_series(region
->vm_start
, region
->vm_top
);
645 kmem_cache_free(vm_region_jar
, region
);
647 up_write(&nommu_region_sem
);
652 * release a reference to a region
654 static void put_nommu_region(struct vm_region
*region
)
656 down_write(&nommu_region_sem
);
657 __put_nommu_region(region
);
661 * update protection on a vma
663 static void protect_vma(struct vm_area_struct
*vma
, unsigned long flags
)
666 struct mm_struct
*mm
= vma
->vm_mm
;
667 long start
= vma
->vm_start
& PAGE_MASK
;
668 while (start
< vma
->vm_end
) {
669 protect_page(mm
, start
, flags
);
672 update_protections(mm
);
677 * add a VMA into a process's mm_struct in the appropriate place in the list
678 * and tree and add to the address space's page tree also if not an anonymous
680 * - should be called with mm->mmap_sem held writelocked
682 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
684 struct vm_area_struct
*pvma
, *prev
;
685 struct address_space
*mapping
;
686 struct rb_node
**p
, *parent
, *rb_prev
;
688 BUG_ON(!vma
->vm_region
);
693 protect_vma(vma
, vma
->vm_flags
);
695 /* add the VMA to the mapping */
697 mapping
= vma
->vm_file
->f_mapping
;
699 i_mmap_lock_write(mapping
);
700 flush_dcache_mmap_lock(mapping
);
701 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
702 flush_dcache_mmap_unlock(mapping
);
703 i_mmap_unlock_write(mapping
);
706 /* add the VMA to the tree */
707 parent
= rb_prev
= NULL
;
708 p
= &mm
->mm_rb
.rb_node
;
711 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
713 /* sort by: start addr, end addr, VMA struct addr in that order
714 * (the latter is necessary as we may get identical VMAs) */
715 if (vma
->vm_start
< pvma
->vm_start
)
717 else if (vma
->vm_start
> pvma
->vm_start
) {
720 } else if (vma
->vm_end
< pvma
->vm_end
)
722 else if (vma
->vm_end
> pvma
->vm_end
) {
725 } else if (vma
< pvma
)
727 else if (vma
> pvma
) {
734 rb_link_node(&vma
->vm_rb
, parent
, p
);
735 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
737 /* add VMA to the VMA list also */
740 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
742 __vma_link_list(mm
, vma
, prev
, parent
);
746 * delete a VMA from its owning mm_struct and address space
748 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
751 struct address_space
*mapping
;
752 struct mm_struct
*mm
= vma
->vm_mm
;
753 struct task_struct
*curr
= current
;
758 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
759 /* if the vma is cached, invalidate the entire cache */
760 if (curr
->vmacache
[i
] == vma
) {
761 vmacache_invalidate(mm
);
766 /* remove the VMA from the mapping */
768 mapping
= vma
->vm_file
->f_mapping
;
770 i_mmap_lock_write(mapping
);
771 flush_dcache_mmap_lock(mapping
);
772 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
773 flush_dcache_mmap_unlock(mapping
);
774 i_mmap_unlock_write(mapping
);
777 /* remove from the MM's tree and list */
778 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
781 vma
->vm_prev
->vm_next
= vma
->vm_next
;
783 mm
->mmap
= vma
->vm_next
;
786 vma
->vm_next
->vm_prev
= vma
->vm_prev
;
790 * destroy a VMA record
792 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
794 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
795 vma
->vm_ops
->close(vma
);
798 put_nommu_region(vma
->vm_region
);
799 kmem_cache_free(vm_area_cachep
, vma
);
803 * look up the first VMA in which addr resides, NULL if none
804 * - should be called with mm->mmap_sem at least held readlocked
806 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
808 struct vm_area_struct
*vma
;
810 /* check the cache first */
811 vma
= vmacache_find(mm
, addr
);
815 /* trawl the list (there may be multiple mappings in which addr
817 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
818 if (vma
->vm_start
> addr
)
820 if (vma
->vm_end
> addr
) {
821 vmacache_update(addr
, vma
);
828 EXPORT_SYMBOL(find_vma
);
832 * - we don't extend stack VMAs under NOMMU conditions
834 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
836 return find_vma(mm
, addr
);
840 * expand a stack to a given address
841 * - not supported under NOMMU conditions
843 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
849 * look up the first VMA exactly that exactly matches addr
850 * - should be called with mm->mmap_sem at least held readlocked
852 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
856 struct vm_area_struct
*vma
;
857 unsigned long end
= addr
+ len
;
859 /* check the cache first */
860 vma
= vmacache_find_exact(mm
, addr
, end
);
864 /* trawl the list (there may be multiple mappings in which addr
866 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
867 if (vma
->vm_start
< addr
)
869 if (vma
->vm_start
> addr
)
871 if (vma
->vm_end
== end
) {
872 vmacache_update(addr
, vma
);
881 * determine whether a mapping should be permitted and, if so, what sort of
882 * mapping we're capable of supporting
884 static int validate_mmap_request(struct file
*file
,
890 unsigned long *_capabilities
)
892 unsigned long capabilities
, rlen
;
895 /* do the simple checks first */
896 if (flags
& MAP_FIXED
)
899 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
900 (flags
& MAP_TYPE
) != MAP_SHARED
)
906 /* Careful about overflows.. */
907 rlen
= PAGE_ALIGN(len
);
908 if (!rlen
|| rlen
> TASK_SIZE
)
911 /* offset overflow? */
912 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
916 /* files must support mmap */
917 if (!file
->f_op
->mmap
)
920 /* work out if what we've got could possibly be shared
921 * - we support chardevs that provide their own "memory"
922 * - we support files/blockdevs that are memory backed
924 if (file
->f_op
->mmap_capabilities
) {
925 capabilities
= file
->f_op
->mmap_capabilities(file
);
927 /* no explicit capabilities set, so assume some
929 switch (file_inode(file
)->i_mode
& S_IFMT
) {
932 capabilities
= NOMMU_MAP_COPY
;
947 /* eliminate any capabilities that we can't support on this
949 if (!file
->f_op
->get_unmapped_area
)
950 capabilities
&= ~NOMMU_MAP_DIRECT
;
951 if (!(file
->f_mode
& FMODE_CAN_READ
))
952 capabilities
&= ~NOMMU_MAP_COPY
;
954 /* The file shall have been opened with read permission. */
955 if (!(file
->f_mode
& FMODE_READ
))
958 if (flags
& MAP_SHARED
) {
959 /* do checks for writing, appending and locking */
960 if ((prot
& PROT_WRITE
) &&
961 !(file
->f_mode
& FMODE_WRITE
))
964 if (IS_APPEND(file_inode(file
)) &&
965 (file
->f_mode
& FMODE_WRITE
))
968 if (locks_verify_locked(file
))
971 if (!(capabilities
& NOMMU_MAP_DIRECT
))
974 /* we mustn't privatise shared mappings */
975 capabilities
&= ~NOMMU_MAP_COPY
;
977 /* we're going to read the file into private memory we
979 if (!(capabilities
& NOMMU_MAP_COPY
))
982 /* we don't permit a private writable mapping to be
983 * shared with the backing device */
984 if (prot
& PROT_WRITE
)
985 capabilities
&= ~NOMMU_MAP_DIRECT
;
988 if (capabilities
& NOMMU_MAP_DIRECT
) {
989 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
990 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
991 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
993 capabilities
&= ~NOMMU_MAP_DIRECT
;
994 if (flags
& MAP_SHARED
) {
995 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1001 /* handle executable mappings and implied executable
1003 if (path_noexec(&file
->f_path
)) {
1004 if (prot
& PROT_EXEC
)
1006 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
1007 /* handle implication of PROT_EXEC by PROT_READ */
1008 if (current
->personality
& READ_IMPLIES_EXEC
) {
1009 if (capabilities
& NOMMU_MAP_EXEC
)
1012 } else if ((prot
& PROT_READ
) &&
1013 (prot
& PROT_EXEC
) &&
1014 !(capabilities
& NOMMU_MAP_EXEC
)
1016 /* backing file is not executable, try to copy */
1017 capabilities
&= ~NOMMU_MAP_DIRECT
;
1020 /* anonymous mappings are always memory backed and can be
1023 capabilities
= NOMMU_MAP_COPY
;
1025 /* handle PROT_EXEC implication by PROT_READ */
1026 if ((prot
& PROT_READ
) &&
1027 (current
->personality
& READ_IMPLIES_EXEC
))
1031 /* allow the security API to have its say */
1032 ret
= security_mmap_addr(addr
);
1037 *_capabilities
= capabilities
;
1042 * we've determined that we can make the mapping, now translate what we
1043 * now know into VMA flags
1045 static unsigned long determine_vm_flags(struct file
*file
,
1047 unsigned long flags
,
1048 unsigned long capabilities
)
1050 unsigned long vm_flags
;
1052 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
1053 /* vm_flags |= mm->def_flags; */
1055 if (!(capabilities
& NOMMU_MAP_DIRECT
)) {
1056 /* attempt to share read-only copies of mapped file chunks */
1057 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1058 if (file
&& !(prot
& PROT_WRITE
))
1059 vm_flags
|= VM_MAYSHARE
;
1061 /* overlay a shareable mapping on the backing device or inode
1062 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1064 vm_flags
|= VM_MAYSHARE
| (capabilities
& NOMMU_VMFLAGS
);
1065 if (flags
& MAP_SHARED
)
1066 vm_flags
|= VM_SHARED
;
1069 /* refuse to let anyone share private mappings with this process if
1070 * it's being traced - otherwise breakpoints set in it may interfere
1071 * with another untraced process
1073 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
1074 vm_flags
&= ~VM_MAYSHARE
;
1080 * set up a shared mapping on a file (the driver or filesystem provides and
1083 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1087 ret
= call_mmap(vma
->vm_file
, vma
);
1089 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1095 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1096 * opposed to tried but failed) so we can only give a suitable error as
1097 * it's not possible to make a private copy if MAP_SHARED was given */
1102 * set up a private mapping or an anonymous shared mapping
1104 static int do_mmap_private(struct vm_area_struct
*vma
,
1105 struct vm_region
*region
,
1107 unsigned long capabilities
)
1109 unsigned long total
, point
;
1113 /* invoke the file's mapping function so that it can keep track of
1114 * shared mappings on devices or memory
1115 * - VM_MAYSHARE will be set if it may attempt to share
1117 if (capabilities
& NOMMU_MAP_DIRECT
) {
1118 ret
= call_mmap(vma
->vm_file
, vma
);
1120 /* shouldn't return success if we're not sharing */
1121 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1122 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1128 /* getting an ENOSYS error indicates that direct mmap isn't
1129 * possible (as opposed to tried but failed) so we'll try to
1130 * make a private copy of the data and map that instead */
1134 /* allocate some memory to hold the mapping
1135 * - note that this may not return a page-aligned address if the object
1136 * we're allocating is smaller than a page
1138 order
= get_order(len
);
1140 point
= len
>> PAGE_SHIFT
;
1142 /* we don't want to allocate a power-of-2 sized page set */
1143 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
1146 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
1150 atomic_long_add(total
, &mmap_pages_allocated
);
1152 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1153 region
->vm_start
= (unsigned long) base
;
1154 region
->vm_end
= region
->vm_start
+ len
;
1155 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1157 vma
->vm_start
= region
->vm_start
;
1158 vma
->vm_end
= region
->vm_start
+ len
;
1161 /* read the contents of a file into the copy */
1162 mm_segment_t old_fs
;
1165 fpos
= vma
->vm_pgoff
;
1166 fpos
<<= PAGE_SHIFT
;
1170 ret
= __vfs_read(vma
->vm_file
, base
, len
, &fpos
);
1176 /* clear the last little bit */
1178 memset(base
+ ret
, 0, len
- ret
);
1185 free_page_series(region
->vm_start
, region
->vm_top
);
1186 region
->vm_start
= vma
->vm_start
= 0;
1187 region
->vm_end
= vma
->vm_end
= 0;
1192 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1193 len
, current
->pid
, current
->comm
);
1194 show_free_areas(0, NULL
);
1199 * handle mapping creation for uClinux
1201 unsigned long do_mmap(struct file
*file
,
1205 unsigned long flags
,
1206 vm_flags_t vm_flags
,
1207 unsigned long pgoff
,
1208 unsigned long *populate
,
1209 struct list_head
*uf
)
1211 struct vm_area_struct
*vma
;
1212 struct vm_region
*region
;
1214 unsigned long capabilities
, result
;
1219 /* decide whether we should attempt the mapping, and if so what sort of
1221 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1226 /* we ignore the address hint */
1228 len
= PAGE_ALIGN(len
);
1230 /* we've determined that we can make the mapping, now translate what we
1231 * now know into VMA flags */
1232 vm_flags
|= determine_vm_flags(file
, prot
, flags
, capabilities
);
1234 /* we're going to need to record the mapping */
1235 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1237 goto error_getting_region
;
1239 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1241 goto error_getting_vma
;
1243 region
->vm_usage
= 1;
1244 region
->vm_flags
= vm_flags
;
1245 region
->vm_pgoff
= pgoff
;
1247 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1248 vma
->vm_flags
= vm_flags
;
1249 vma
->vm_pgoff
= pgoff
;
1252 region
->vm_file
= get_file(file
);
1253 vma
->vm_file
= get_file(file
);
1256 down_write(&nommu_region_sem
);
1258 /* if we want to share, we need to check for regions created by other
1259 * mmap() calls that overlap with our proposed mapping
1260 * - we can only share with a superset match on most regular files
1261 * - shared mappings on character devices and memory backed files are
1262 * permitted to overlap inexactly as far as we are concerned for in
1263 * these cases, sharing is handled in the driver or filesystem rather
1266 if (vm_flags
& VM_MAYSHARE
) {
1267 struct vm_region
*pregion
;
1268 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1270 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1271 pgend
= pgoff
+ pglen
;
1273 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1274 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1276 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1279 /* search for overlapping mappings on the same file */
1280 if (file_inode(pregion
->vm_file
) !=
1284 if (pregion
->vm_pgoff
>= pgend
)
1287 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1288 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1289 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1290 if (pgoff
>= rpgend
)
1293 /* handle inexactly overlapping matches between
1295 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1296 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1297 /* new mapping is not a subset of the region */
1298 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1299 goto sharing_violation
;
1303 /* we've found a region we can share */
1304 pregion
->vm_usage
++;
1305 vma
->vm_region
= pregion
;
1306 start
= pregion
->vm_start
;
1307 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1308 vma
->vm_start
= start
;
1309 vma
->vm_end
= start
+ len
;
1311 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1312 vma
->vm_flags
|= VM_MAPPED_COPY
;
1314 ret
= do_mmap_shared_file(vma
);
1316 vma
->vm_region
= NULL
;
1319 pregion
->vm_usage
--;
1321 goto error_just_free
;
1324 fput(region
->vm_file
);
1325 kmem_cache_free(vm_region_jar
, region
);
1331 /* obtain the address at which to make a shared mapping
1332 * - this is the hook for quasi-memory character devices to
1333 * tell us the location of a shared mapping
1335 if (capabilities
& NOMMU_MAP_DIRECT
) {
1336 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1338 if (IS_ERR_VALUE(addr
)) {
1341 goto error_just_free
;
1343 /* the driver refused to tell us where to site
1344 * the mapping so we'll have to attempt to copy
1347 if (!(capabilities
& NOMMU_MAP_COPY
))
1348 goto error_just_free
;
1350 capabilities
&= ~NOMMU_MAP_DIRECT
;
1352 vma
->vm_start
= region
->vm_start
= addr
;
1353 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1358 vma
->vm_region
= region
;
1360 /* set up the mapping
1361 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1363 if (file
&& vma
->vm_flags
& VM_SHARED
)
1364 ret
= do_mmap_shared_file(vma
);
1366 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1368 goto error_just_free
;
1369 add_nommu_region(region
);
1371 /* clear anonymous mappings that don't ask for uninitialized data */
1372 if (!vma
->vm_file
&& !(flags
& MAP_UNINITIALIZED
))
1373 memset((void *)region
->vm_start
, 0,
1374 region
->vm_end
- region
->vm_start
);
1376 /* okay... we have a mapping; now we have to register it */
1377 result
= vma
->vm_start
;
1379 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1382 add_vma_to_mm(current
->mm
, vma
);
1384 /* we flush the region from the icache only when the first executable
1385 * mapping of it is made */
1386 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1387 flush_icache_range(region
->vm_start
, region
->vm_end
);
1388 region
->vm_icache_flushed
= true;
1391 up_write(&nommu_region_sem
);
1396 up_write(&nommu_region_sem
);
1398 if (region
->vm_file
)
1399 fput(region
->vm_file
);
1400 kmem_cache_free(vm_region_jar
, region
);
1403 kmem_cache_free(vm_area_cachep
, vma
);
1407 up_write(&nommu_region_sem
);
1408 pr_warn("Attempt to share mismatched mappings\n");
1413 kmem_cache_free(vm_region_jar
, region
);
1414 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1416 show_free_areas(0, NULL
);
1419 error_getting_region
:
1420 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1422 show_free_areas(0, NULL
);
1426 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1427 unsigned long, prot
, unsigned long, flags
,
1428 unsigned long, fd
, unsigned long, pgoff
)
1430 struct file
*file
= NULL
;
1431 unsigned long retval
= -EBADF
;
1433 audit_mmap_fd(fd
, flags
);
1434 if (!(flags
& MAP_ANONYMOUS
)) {
1440 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1442 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1450 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1451 struct mmap_arg_struct
{
1455 unsigned long flags
;
1457 unsigned long offset
;
1460 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1462 struct mmap_arg_struct a
;
1464 if (copy_from_user(&a
, arg
, sizeof(a
)))
1466 if (offset_in_page(a
.offset
))
1469 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1470 a
.offset
>> PAGE_SHIFT
);
1472 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1475 * split a vma into two pieces at address 'addr', a new vma is allocated either
1476 * for the first part or the tail.
1478 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1479 unsigned long addr
, int new_below
)
1481 struct vm_area_struct
*new;
1482 struct vm_region
*region
;
1483 unsigned long npages
;
1485 /* we're only permitted to split anonymous regions (these should have
1486 * only a single usage on the region) */
1490 if (mm
->map_count
>= sysctl_max_map_count
)
1493 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1497 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1499 kmem_cache_free(vm_region_jar
, region
);
1503 /* most fields are the same, copy all, and then fixup */
1505 *region
= *vma
->vm_region
;
1506 new->vm_region
= region
;
1508 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1511 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1513 region
->vm_start
= new->vm_start
= addr
;
1514 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1517 if (new->vm_ops
&& new->vm_ops
->open
)
1518 new->vm_ops
->open(new);
1520 delete_vma_from_mm(vma
);
1521 down_write(&nommu_region_sem
);
1522 delete_nommu_region(vma
->vm_region
);
1524 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1525 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1527 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1528 vma
->vm_region
->vm_top
= addr
;
1530 add_nommu_region(vma
->vm_region
);
1531 add_nommu_region(new->vm_region
);
1532 up_write(&nommu_region_sem
);
1533 add_vma_to_mm(mm
, vma
);
1534 add_vma_to_mm(mm
, new);
1539 * shrink a VMA by removing the specified chunk from either the beginning or
1542 static int shrink_vma(struct mm_struct
*mm
,
1543 struct vm_area_struct
*vma
,
1544 unsigned long from
, unsigned long to
)
1546 struct vm_region
*region
;
1548 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1550 delete_vma_from_mm(vma
);
1551 if (from
> vma
->vm_start
)
1555 add_vma_to_mm(mm
, vma
);
1557 /* cut the backing region down to size */
1558 region
= vma
->vm_region
;
1559 BUG_ON(region
->vm_usage
!= 1);
1561 down_write(&nommu_region_sem
);
1562 delete_nommu_region(region
);
1563 if (from
> region
->vm_start
) {
1564 to
= region
->vm_top
;
1565 region
->vm_top
= region
->vm_end
= from
;
1567 region
->vm_start
= to
;
1569 add_nommu_region(region
);
1570 up_write(&nommu_region_sem
);
1572 free_page_series(from
, to
);
1578 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1579 * VMA, though it need not cover the whole VMA
1581 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1583 struct vm_area_struct
*vma
;
1587 len
= PAGE_ALIGN(len
);
1593 /* find the first potentially overlapping VMA */
1594 vma
= find_vma(mm
, start
);
1598 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1599 current
->pid
, current
->comm
,
1600 start
, start
+ len
- 1);
1606 /* we're allowed to split an anonymous VMA but not a file-backed one */
1609 if (start
> vma
->vm_start
)
1611 if (end
== vma
->vm_end
)
1612 goto erase_whole_vma
;
1617 /* the chunk must be a subset of the VMA found */
1618 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1619 goto erase_whole_vma
;
1620 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1622 if (offset_in_page(start
))
1624 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1626 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1627 ret
= split_vma(mm
, vma
, start
, 1);
1631 return shrink_vma(mm
, vma
, start
, end
);
1635 delete_vma_from_mm(vma
);
1636 delete_vma(mm
, vma
);
1639 EXPORT_SYMBOL(do_munmap
);
1641 int vm_munmap(unsigned long addr
, size_t len
)
1643 struct mm_struct
*mm
= current
->mm
;
1646 down_write(&mm
->mmap_sem
);
1647 ret
= do_munmap(mm
, addr
, len
, NULL
);
1648 up_write(&mm
->mmap_sem
);
1651 EXPORT_SYMBOL(vm_munmap
);
1653 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1655 return vm_munmap(addr
, len
);
1659 * release all the mappings made in a process's VM space
1661 void exit_mmap(struct mm_struct
*mm
)
1663 struct vm_area_struct
*vma
;
1670 while ((vma
= mm
->mmap
)) {
1671 mm
->mmap
= vma
->vm_next
;
1672 delete_vma_from_mm(vma
);
1673 delete_vma(mm
, vma
);
1678 int vm_brk(unsigned long addr
, unsigned long len
)
1684 * expand (or shrink) an existing mapping, potentially moving it at the same
1685 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1687 * under NOMMU conditions, we only permit changing a mapping's size, and only
1688 * as long as it stays within the region allocated by do_mmap_private() and the
1689 * block is not shareable
1691 * MREMAP_FIXED is not supported under NOMMU conditions
1693 static unsigned long do_mremap(unsigned long addr
,
1694 unsigned long old_len
, unsigned long new_len
,
1695 unsigned long flags
, unsigned long new_addr
)
1697 struct vm_area_struct
*vma
;
1699 /* insanity checks first */
1700 old_len
= PAGE_ALIGN(old_len
);
1701 new_len
= PAGE_ALIGN(new_len
);
1702 if (old_len
== 0 || new_len
== 0)
1703 return (unsigned long) -EINVAL
;
1705 if (offset_in_page(addr
))
1708 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1709 return (unsigned long) -EINVAL
;
1711 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1713 return (unsigned long) -EINVAL
;
1715 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1716 return (unsigned long) -EFAULT
;
1718 if (vma
->vm_flags
& VM_MAYSHARE
)
1719 return (unsigned long) -EPERM
;
1721 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1722 return (unsigned long) -ENOMEM
;
1724 /* all checks complete - do it */
1725 vma
->vm_end
= vma
->vm_start
+ new_len
;
1726 return vma
->vm_start
;
1729 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1730 unsigned long, new_len
, unsigned long, flags
,
1731 unsigned long, new_addr
)
1735 down_write(¤t
->mm
->mmap_sem
);
1736 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1737 up_write(¤t
->mm
->mmap_sem
);
1741 struct page
*follow_page_mask(struct vm_area_struct
*vma
,
1742 unsigned long address
, unsigned int flags
,
1743 unsigned int *page_mask
)
1749 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1750 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1752 if (addr
!= (pfn
<< PAGE_SHIFT
))
1755 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1758 EXPORT_SYMBOL(remap_pfn_range
);
1760 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1762 unsigned long pfn
= start
>> PAGE_SHIFT
;
1763 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1765 pfn
+= vma
->vm_pgoff
;
1766 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1768 EXPORT_SYMBOL(vm_iomap_memory
);
1770 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1771 unsigned long pgoff
)
1773 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1775 if (!(vma
->vm_flags
& VM_USERMAP
))
1778 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1779 vma
->vm_end
= vma
->vm_start
+ size
;
1783 EXPORT_SYMBOL(remap_vmalloc_range
);
1785 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1786 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1791 void unmap_mapping_range(struct address_space
*mapping
,
1792 loff_t
const holebegin
, loff_t
const holelen
,
1796 EXPORT_SYMBOL(unmap_mapping_range
);
1798 int filemap_fault(struct vm_fault
*vmf
)
1803 EXPORT_SYMBOL(filemap_fault
);
1805 void filemap_map_pages(struct vm_fault
*vmf
,
1806 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1810 EXPORT_SYMBOL(filemap_map_pages
);
1812 int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
1813 unsigned long addr
, void *buf
, int len
, unsigned int gup_flags
)
1815 struct vm_area_struct
*vma
;
1816 int write
= gup_flags
& FOLL_WRITE
;
1818 down_read(&mm
->mmap_sem
);
1820 /* the access must start within one of the target process's mappings */
1821 vma
= find_vma(mm
, addr
);
1823 /* don't overrun this mapping */
1824 if (addr
+ len
>= vma
->vm_end
)
1825 len
= vma
->vm_end
- addr
;
1827 /* only read or write mappings where it is permitted */
1828 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1829 copy_to_user_page(vma
, NULL
, addr
,
1830 (void *) addr
, buf
, len
);
1831 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1832 copy_from_user_page(vma
, NULL
, addr
,
1833 buf
, (void *) addr
, len
);
1840 up_read(&mm
->mmap_sem
);
1846 * @access_remote_vm - access another process' address space
1847 * @mm: the mm_struct of the target address space
1848 * @addr: start address to access
1849 * @buf: source or destination buffer
1850 * @len: number of bytes to transfer
1851 * @gup_flags: flags modifying lookup behaviour
1853 * The caller must hold a reference on @mm.
1855 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1856 void *buf
, int len
, unsigned int gup_flags
)
1858 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, gup_flags
);
1862 * Access another process' address space.
1863 * - source/target buffer must be kernel space
1865 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1866 unsigned int gup_flags
)
1868 struct mm_struct
*mm
;
1870 if (addr
+ len
< addr
)
1873 mm
= get_task_mm(tsk
);
1877 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, gup_flags
);
1882 EXPORT_SYMBOL_GPL(access_process_vm
);
1885 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1886 * @inode: The inode to check
1887 * @size: The current filesize of the inode
1888 * @newsize: The proposed filesize of the inode
1890 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1891 * make sure that that any outstanding VMAs aren't broken and then shrink the
1892 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1893 * automatically grant mappings that are too large.
1895 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1898 struct vm_area_struct
*vma
;
1899 struct vm_region
*region
;
1901 size_t r_size
, r_top
;
1903 low
= newsize
>> PAGE_SHIFT
;
1904 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1906 down_write(&nommu_region_sem
);
1907 i_mmap_lock_read(inode
->i_mapping
);
1909 /* search for VMAs that fall within the dead zone */
1910 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1911 /* found one - only interested if it's shared out of the page
1913 if (vma
->vm_flags
& VM_SHARED
) {
1914 i_mmap_unlock_read(inode
->i_mapping
);
1915 up_write(&nommu_region_sem
);
1916 return -ETXTBSY
; /* not quite true, but near enough */
1920 /* reduce any regions that overlap the dead zone - if in existence,
1921 * these will be pointed to by VMAs that don't overlap the dead zone
1923 * we don't check for any regions that start beyond the EOF as there
1926 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1927 if (!(vma
->vm_flags
& VM_SHARED
))
1930 region
= vma
->vm_region
;
1931 r_size
= region
->vm_top
- region
->vm_start
;
1932 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1934 if (r_top
> newsize
) {
1935 region
->vm_top
-= r_top
- newsize
;
1936 if (region
->vm_end
> region
->vm_top
)
1937 region
->vm_end
= region
->vm_top
;
1941 i_mmap_unlock_read(inode
->i_mapping
);
1942 up_write(&nommu_region_sem
);
1947 * Initialise sysctl_user_reserve_kbytes.
1949 * This is intended to prevent a user from starting a single memory hogging
1950 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1953 * The default value is min(3% of free memory, 128MB)
1954 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1956 static int __meminit
init_user_reserve(void)
1958 unsigned long free_kbytes
;
1960 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1962 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1965 subsys_initcall(init_user_reserve
);
1968 * Initialise sysctl_admin_reserve_kbytes.
1970 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1971 * to log in and kill a memory hogging process.
1973 * Systems with more than 256MB will reserve 8MB, enough to recover
1974 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1975 * only reserve 3% of free pages by default.
1977 static int __meminit
init_admin_reserve(void)
1979 unsigned long free_kbytes
;
1981 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1983 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1986 subsys_initcall(init_admin_reserve
);