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) 2007-2009 Paul Mundt <lethal@linux-sh.org>
16 #include <linux/module.h>
18 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/file.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/tracehook.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mount.h>
29 #include <linux/personality.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
33 #include <asm/uaccess.h>
35 #include <asm/tlbflush.h>
38 static inline __attribute__((format(printf
, 1, 2)))
39 void no_printk(const char *fmt
, ...)
44 #define kenter(FMT, ...) \
45 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
46 #define kleave(FMT, ...) \
47 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
48 #define kdebug(FMT, ...) \
49 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
51 #define kenter(FMT, ...) \
52 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
53 #define kleave(FMT, ...) \
54 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
55 #define kdebug(FMT, ...) \
56 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
63 unsigned long max_mapnr
;
64 unsigned long num_physpages
;
65 struct percpu_counter vm_committed_as
;
66 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
67 int sysctl_overcommit_ratio
= 50; /* default is 50% */
68 int sysctl_max_map_count
= DEFAULT_MAX_MAP_COUNT
;
69 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
70 int heap_stack_gap
= 0;
72 atomic_long_t mmap_pages_allocated
;
74 EXPORT_SYMBOL(mem_map
);
75 EXPORT_SYMBOL(num_physpages
);
77 /* list of mapped, potentially shareable regions */
78 static struct kmem_cache
*vm_region_jar
;
79 struct rb_root nommu_region_tree
= RB_ROOT
;
80 DECLARE_RWSEM(nommu_region_sem
);
82 struct vm_operations_struct generic_file_vm_ops
= {
86 * Handle all mappings that got truncated by a "truncate()"
89 * NOTE! We have to be ready to update the memory sharing
90 * between the file and the memory map for a potential last
91 * incomplete page. Ugly, but necessary.
93 int vmtruncate(struct inode
*inode
, loff_t offset
)
95 struct address_space
*mapping
= inode
->i_mapping
;
98 if (inode
->i_size
< offset
)
100 i_size_write(inode
, offset
);
102 truncate_inode_pages(mapping
, offset
);
106 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
107 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
109 if (offset
> inode
->i_sb
->s_maxbytes
)
111 i_size_write(inode
, offset
);
114 if (inode
->i_op
->truncate
)
115 inode
->i_op
->truncate(inode
);
118 send_sig(SIGXFSZ
, current
, 0);
123 EXPORT_SYMBOL(vmtruncate
);
126 * Return the total memory allocated for this pointer, not
127 * just what the caller asked for.
129 * Doesn't have to be accurate, i.e. may have races.
131 unsigned int kobjsize(const void *objp
)
136 * If the object we have should not have ksize performed on it,
139 if (!objp
|| !virt_addr_valid(objp
))
142 page
= virt_to_head_page(objp
);
145 * If the allocator sets PageSlab, we know the pointer came from
152 * If it's not a compound page, see if we have a matching VMA
153 * region. This test is intentionally done in reverse order,
154 * so if there's no VMA, we still fall through and hand back
155 * PAGE_SIZE for 0-order pages.
157 if (!PageCompound(page
)) {
158 struct vm_area_struct
*vma
;
160 vma
= find_vma(current
->mm
, (unsigned long)objp
);
162 return vma
->vm_end
- vma
->vm_start
;
166 * The ksize() function is only guaranteed to work for pointers
167 * returned by kmalloc(). So handle arbitrary pointers here.
169 return PAGE_SIZE
<< compound_order(page
);
172 int __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
173 unsigned long start
, int nr_pages
, int flags
,
174 struct page
**pages
, struct vm_area_struct
**vmas
)
176 struct vm_area_struct
*vma
;
177 unsigned long vm_flags
;
179 int write
= !!(flags
& GUP_FLAGS_WRITE
);
180 int force
= !!(flags
& GUP_FLAGS_FORCE
);
181 int ignore
= !!(flags
& GUP_FLAGS_IGNORE_VMA_PERMISSIONS
);
183 /* calculate required read or write permissions.
184 * - if 'force' is set, we only require the "MAY" flags.
186 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
187 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
189 for (i
= 0; i
< nr_pages
; i
++) {
190 vma
= find_vma(mm
, start
);
192 goto finish_or_fault
;
194 /* protect what we can, including chardevs */
195 if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
) ||
196 (!ignore
&& !(vm_flags
& vma
->vm_flags
)))
197 goto finish_or_fault
;
200 pages
[i
] = virt_to_page(start
);
202 page_cache_get(pages
[i
]);
212 return i
? : -EFAULT
;
217 * get a list of pages in an address range belonging to the specified process
218 * and indicate the VMA that covers each page
219 * - this is potentially dodgy as we may end incrementing the page count of a
220 * slab page or a secondary page from a compound page
221 * - don't permit access to VMAs that don't support it, such as I/O mappings
223 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
224 unsigned long start
, int nr_pages
, int write
, int force
,
225 struct page
**pages
, struct vm_area_struct
**vmas
)
230 flags
|= GUP_FLAGS_WRITE
;
232 flags
|= GUP_FLAGS_FORCE
;
234 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
);
236 EXPORT_SYMBOL(get_user_pages
);
239 * follow_pfn - look up PFN at a user virtual address
240 * @vma: memory mapping
241 * @address: user virtual address
242 * @pfn: location to store found PFN
244 * Only IO mappings and raw PFN mappings are allowed.
246 * Returns zero and the pfn at @pfn on success, -ve otherwise.
248 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
251 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
254 *pfn
= address
>> PAGE_SHIFT
;
257 EXPORT_SYMBOL(follow_pfn
);
259 DEFINE_RWLOCK(vmlist_lock
);
260 struct vm_struct
*vmlist
;
262 void vfree(const void *addr
)
266 EXPORT_SYMBOL(vfree
);
268 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
271 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
272 * returns only a logical address.
274 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
276 EXPORT_SYMBOL(__vmalloc
);
278 void *vmalloc_user(unsigned long size
)
282 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
285 struct vm_area_struct
*vma
;
287 down_write(¤t
->mm
->mmap_sem
);
288 vma
= find_vma(current
->mm
, (unsigned long)ret
);
290 vma
->vm_flags
|= VM_USERMAP
;
291 up_write(¤t
->mm
->mmap_sem
);
296 EXPORT_SYMBOL(vmalloc_user
);
298 struct page
*vmalloc_to_page(const void *addr
)
300 return virt_to_page(addr
);
302 EXPORT_SYMBOL(vmalloc_to_page
);
304 unsigned long vmalloc_to_pfn(const void *addr
)
306 return page_to_pfn(virt_to_page(addr
));
308 EXPORT_SYMBOL(vmalloc_to_pfn
);
310 long vread(char *buf
, char *addr
, unsigned long count
)
312 memcpy(buf
, addr
, count
);
316 long vwrite(char *buf
, char *addr
, unsigned long count
)
318 /* Don't allow overflow */
319 if ((unsigned long) addr
+ count
< count
)
320 count
= -(unsigned long) addr
;
322 memcpy(addr
, buf
, count
);
327 * vmalloc - allocate virtually continguos memory
329 * @size: allocation size
331 * Allocate enough pages to cover @size from the page level
332 * allocator and map them into continguos kernel virtual space.
334 * For tight control over page level allocator and protection flags
335 * use __vmalloc() instead.
337 void *vmalloc(unsigned long size
)
339 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
341 EXPORT_SYMBOL(vmalloc
);
343 void *vmalloc_node(unsigned long size
, int node
)
345 return vmalloc(size
);
347 EXPORT_SYMBOL(vmalloc_node
);
349 #ifndef PAGE_KERNEL_EXEC
350 # define PAGE_KERNEL_EXEC PAGE_KERNEL
354 * vmalloc_exec - allocate virtually contiguous, executable memory
355 * @size: allocation size
357 * Kernel-internal function to allocate enough pages to cover @size
358 * the page level allocator and map them into contiguous and
359 * executable kernel virtual space.
361 * For tight control over page level allocator and protection flags
362 * use __vmalloc() instead.
365 void *vmalloc_exec(unsigned long size
)
367 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
371 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
372 * @size: allocation size
374 * Allocate enough 32bit PA addressable pages to cover @size from the
375 * page level allocator and map them into continguos kernel virtual space.
377 void *vmalloc_32(unsigned long size
)
379 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
381 EXPORT_SYMBOL(vmalloc_32
);
384 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
385 * @size: allocation size
387 * The resulting memory area is 32bit addressable and zeroed so it can be
388 * mapped to userspace without leaking data.
390 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
391 * remap_vmalloc_range() are permissible.
393 void *vmalloc_32_user(unsigned long size
)
396 * We'll have to sort out the ZONE_DMA bits for 64-bit,
397 * but for now this can simply use vmalloc_user() directly.
399 return vmalloc_user(size
);
401 EXPORT_SYMBOL(vmalloc_32_user
);
403 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
410 void vunmap(const void *addr
)
414 EXPORT_SYMBOL(vunmap
);
416 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
421 EXPORT_SYMBOL(vm_map_ram
);
423 void vm_unmap_ram(const void *mem
, unsigned int count
)
427 EXPORT_SYMBOL(vm_unmap_ram
);
429 void vm_unmap_aliases(void)
432 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
435 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
438 void __attribute__((weak
)) vmalloc_sync_all(void)
442 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
447 EXPORT_SYMBOL(vm_insert_page
);
450 * sys_brk() for the most part doesn't need the global kernel
451 * lock, except when an application is doing something nasty
452 * like trying to un-brk an area that has already been mapped
453 * to a regular file. in this case, the unmapping will need
454 * to invoke file system routines that need the global lock.
456 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
458 struct mm_struct
*mm
= current
->mm
;
460 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
467 * Always allow shrinking brk
469 if (brk
<= mm
->brk
) {
475 * Ok, looks good - let it rip.
477 return mm
->brk
= brk
;
481 * initialise the VMA and region record slabs
483 void __init
mmap_init(void)
487 ret
= percpu_counter_init(&vm_committed_as
, 0);
489 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
);
493 * validate the region tree
494 * - the caller must hold the region lock
496 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
497 static noinline
void validate_nommu_regions(void)
499 struct vm_region
*region
, *last
;
500 struct rb_node
*p
, *lastp
;
502 lastp
= rb_first(&nommu_region_tree
);
506 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
507 BUG_ON(unlikely(last
->vm_end
<= last
->vm_start
));
508 BUG_ON(unlikely(last
->vm_top
< last
->vm_end
));
510 while ((p
= rb_next(lastp
))) {
511 region
= rb_entry(p
, struct vm_region
, vm_rb
);
512 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
514 BUG_ON(unlikely(region
->vm_end
<= region
->vm_start
));
515 BUG_ON(unlikely(region
->vm_top
< region
->vm_end
));
516 BUG_ON(unlikely(region
->vm_start
< last
->vm_top
));
522 static void validate_nommu_regions(void)
528 * add a region into the global tree
530 static void add_nommu_region(struct vm_region
*region
)
532 struct vm_region
*pregion
;
533 struct rb_node
**p
, *parent
;
535 validate_nommu_regions();
538 p
= &nommu_region_tree
.rb_node
;
541 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
542 if (region
->vm_start
< pregion
->vm_start
)
544 else if (region
->vm_start
> pregion
->vm_start
)
546 else if (pregion
== region
)
552 rb_link_node(®ion
->vm_rb
, parent
, p
);
553 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
555 validate_nommu_regions();
559 * delete a region from the global tree
561 static void delete_nommu_region(struct vm_region
*region
)
563 BUG_ON(!nommu_region_tree
.rb_node
);
565 validate_nommu_regions();
566 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
567 validate_nommu_regions();
571 * free a contiguous series of pages
573 static void free_page_series(unsigned long from
, unsigned long to
)
575 for (; from
< to
; from
+= PAGE_SIZE
) {
576 struct page
*page
= virt_to_page(from
);
578 kdebug("- free %lx", from
);
579 atomic_long_dec(&mmap_pages_allocated
);
580 if (page_count(page
) != 1)
581 kdebug("free page %p: refcount not one: %d",
582 page
, page_count(page
));
588 * release a reference to a region
589 * - the caller must hold the region semaphore for writing, which this releases
590 * - the region may not have been added to the tree yet, in which case vm_top
591 * will equal vm_start
593 static void __put_nommu_region(struct vm_region
*region
)
594 __releases(nommu_region_sem
)
596 kenter("%p{%d}", region
, atomic_read(®ion
->vm_usage
));
598 BUG_ON(!nommu_region_tree
.rb_node
);
600 if (atomic_dec_and_test(®ion
->vm_usage
)) {
601 if (region
->vm_top
> region
->vm_start
)
602 delete_nommu_region(region
);
603 up_write(&nommu_region_sem
);
606 fput(region
->vm_file
);
608 /* IO memory and memory shared directly out of the pagecache
609 * from ramfs/tmpfs mustn't be released here */
610 if (region
->vm_flags
& VM_MAPPED_COPY
) {
611 kdebug("free series");
612 free_page_series(region
->vm_start
, region
->vm_top
);
614 kmem_cache_free(vm_region_jar
, region
);
616 up_write(&nommu_region_sem
);
621 * release a reference to a region
623 static void put_nommu_region(struct vm_region
*region
)
625 down_write(&nommu_region_sem
);
626 __put_nommu_region(region
);
630 * add a VMA into a process's mm_struct in the appropriate place in the list
631 * and tree and add to the address space's page tree also if not an anonymous
633 * - should be called with mm->mmap_sem held writelocked
635 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
637 struct vm_area_struct
*pvma
, **pp
;
638 struct address_space
*mapping
;
639 struct rb_node
**p
, *parent
;
643 BUG_ON(!vma
->vm_region
);
648 /* add the VMA to the mapping */
650 mapping
= vma
->vm_file
->f_mapping
;
652 flush_dcache_mmap_lock(mapping
);
653 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
654 flush_dcache_mmap_unlock(mapping
);
657 /* add the VMA to the tree */
659 p
= &mm
->mm_rb
.rb_node
;
662 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
664 /* sort by: start addr, end addr, VMA struct addr in that order
665 * (the latter is necessary as we may get identical VMAs) */
666 if (vma
->vm_start
< pvma
->vm_start
)
668 else if (vma
->vm_start
> pvma
->vm_start
)
670 else if (vma
->vm_end
< pvma
->vm_end
)
672 else if (vma
->vm_end
> pvma
->vm_end
)
682 rb_link_node(&vma
->vm_rb
, parent
, p
);
683 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
685 /* add VMA to the VMA list also */
686 for (pp
= &mm
->mmap
; (pvma
= *pp
); pp
= &(*pp
)->vm_next
) {
687 if (pvma
->vm_start
> vma
->vm_start
)
689 if (pvma
->vm_start
< vma
->vm_start
)
691 if (pvma
->vm_end
< vma
->vm_end
)
700 * delete a VMA from its owning mm_struct and address space
702 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
704 struct vm_area_struct
**pp
;
705 struct address_space
*mapping
;
706 struct mm_struct
*mm
= vma
->vm_mm
;
711 if (mm
->mmap_cache
== vma
)
712 mm
->mmap_cache
= NULL
;
714 /* remove the VMA from the mapping */
716 mapping
= vma
->vm_file
->f_mapping
;
718 flush_dcache_mmap_lock(mapping
);
719 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
720 flush_dcache_mmap_unlock(mapping
);
723 /* remove from the MM's tree and list */
724 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
725 for (pp
= &mm
->mmap
; *pp
; pp
= &(*pp
)->vm_next
) {
736 * destroy a VMA record
738 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
741 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
742 vma
->vm_ops
->close(vma
);
745 if (vma
->vm_flags
& VM_EXECUTABLE
)
746 removed_exe_file_vma(mm
);
748 put_nommu_region(vma
->vm_region
);
749 kmem_cache_free(vm_area_cachep
, vma
);
753 * look up the first VMA in which addr resides, NULL if none
754 * - should be called with mm->mmap_sem at least held readlocked
756 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
758 struct vm_area_struct
*vma
;
759 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
761 /* check the cache first */
762 vma
= mm
->mmap_cache
;
763 if (vma
&& vma
->vm_start
<= addr
&& vma
->vm_end
> addr
)
766 /* trawl the tree (there may be multiple mappings in which addr
768 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
769 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
770 if (vma
->vm_start
> addr
)
772 if (vma
->vm_end
> addr
) {
773 mm
->mmap_cache
= vma
;
780 EXPORT_SYMBOL(find_vma
);
784 * - we don't extend stack VMAs under NOMMU conditions
786 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
788 return find_vma(mm
, addr
);
792 * expand a stack to a given address
793 * - not supported under NOMMU conditions
795 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
801 * look up the first VMA exactly that exactly matches addr
802 * - should be called with mm->mmap_sem at least held readlocked
804 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
808 struct vm_area_struct
*vma
;
809 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
810 unsigned long end
= addr
+ len
;
812 /* check the cache first */
813 vma
= mm
->mmap_cache
;
814 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_end
== end
)
817 /* trawl the tree (there may be multiple mappings in which addr
819 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
820 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
821 if (vma
->vm_start
< addr
)
823 if (vma
->vm_start
> addr
)
825 if (vma
->vm_end
== end
) {
826 mm
->mmap_cache
= vma
;
835 * determine whether a mapping should be permitted and, if so, what sort of
836 * mapping we're capable of supporting
838 static int validate_mmap_request(struct file
*file
,
844 unsigned long *_capabilities
)
846 unsigned long capabilities
, rlen
;
847 unsigned long reqprot
= prot
;
850 /* do the simple checks first */
851 if (flags
& MAP_FIXED
|| addr
) {
853 "%d: Can't do fixed-address/overlay mmap of RAM\n",
858 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
859 (flags
& MAP_TYPE
) != MAP_SHARED
)
865 /* Careful about overflows.. */
866 rlen
= PAGE_ALIGN(len
);
867 if (!rlen
|| rlen
> TASK_SIZE
)
870 /* offset overflow? */
871 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
875 /* validate file mapping requests */
876 struct address_space
*mapping
;
878 /* files must support mmap */
879 if (!file
->f_op
|| !file
->f_op
->mmap
)
882 /* work out if what we've got could possibly be shared
883 * - we support chardevs that provide their own "memory"
884 * - we support files/blockdevs that are memory backed
886 mapping
= file
->f_mapping
;
888 mapping
= file
->f_path
.dentry
->d_inode
->i_mapping
;
891 if (mapping
&& mapping
->backing_dev_info
)
892 capabilities
= mapping
->backing_dev_info
->capabilities
;
895 /* no explicit capabilities set, so assume some
897 switch (file
->f_path
.dentry
->d_inode
->i_mode
& S_IFMT
) {
900 capabilities
= BDI_CAP_MAP_COPY
;
915 /* eliminate any capabilities that we can't support on this
917 if (!file
->f_op
->get_unmapped_area
)
918 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
919 if (!file
->f_op
->read
)
920 capabilities
&= ~BDI_CAP_MAP_COPY
;
922 if (flags
& MAP_SHARED
) {
923 /* do checks for writing, appending and locking */
924 if ((prot
& PROT_WRITE
) &&
925 !(file
->f_mode
& FMODE_WRITE
))
928 if (IS_APPEND(file
->f_path
.dentry
->d_inode
) &&
929 (file
->f_mode
& FMODE_WRITE
))
932 if (locks_verify_locked(file
->f_path
.dentry
->d_inode
))
935 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
938 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
939 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
940 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
942 printk("MAP_SHARED not completely supported on !MMU\n");
946 /* we mustn't privatise shared mappings */
947 capabilities
&= ~BDI_CAP_MAP_COPY
;
950 /* we're going to read the file into private memory we
952 if (!(capabilities
& BDI_CAP_MAP_COPY
))
955 /* we don't permit a private writable mapping to be
956 * shared with the backing device */
957 if (prot
& PROT_WRITE
)
958 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
961 /* handle executable mappings and implied executable
963 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
964 if (prot
& PROT_EXEC
)
967 else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
968 /* handle implication of PROT_EXEC by PROT_READ */
969 if (current
->personality
& READ_IMPLIES_EXEC
) {
970 if (capabilities
& BDI_CAP_EXEC_MAP
)
974 else if ((prot
& PROT_READ
) &&
975 (prot
& PROT_EXEC
) &&
976 !(capabilities
& BDI_CAP_EXEC_MAP
)
978 /* backing file is not executable, try to copy */
979 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
983 /* anonymous mappings are always memory backed and can be
986 capabilities
= BDI_CAP_MAP_COPY
;
988 /* handle PROT_EXEC implication by PROT_READ */
989 if ((prot
& PROT_READ
) &&
990 (current
->personality
& READ_IMPLIES_EXEC
))
994 /* allow the security API to have its say */
995 ret
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1000 *_capabilities
= capabilities
;
1005 * we've determined that we can make the mapping, now translate what we
1006 * now know into VMA flags
1008 static unsigned long determine_vm_flags(struct file
*file
,
1010 unsigned long flags
,
1011 unsigned long capabilities
)
1013 unsigned long vm_flags
;
1015 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
1016 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1017 /* vm_flags |= mm->def_flags; */
1019 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
1020 /* attempt to share read-only copies of mapped file chunks */
1021 if (file
&& !(prot
& PROT_WRITE
))
1022 vm_flags
|= VM_MAYSHARE
;
1025 /* overlay a shareable mapping on the backing device or inode
1026 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1028 if (flags
& MAP_SHARED
)
1029 vm_flags
|= VM_MAYSHARE
| VM_SHARED
;
1030 else if ((((vm_flags
& capabilities
) ^ vm_flags
) & BDI_CAP_VMFLAGS
) == 0)
1031 vm_flags
|= VM_MAYSHARE
;
1034 /* refuse to let anyone share private mappings with this process if
1035 * it's being traced - otherwise breakpoints set in it may interfere
1036 * with another untraced process
1038 if ((flags
& MAP_PRIVATE
) && tracehook_expect_breakpoints(current
))
1039 vm_flags
&= ~VM_MAYSHARE
;
1045 * set up a shared mapping on a file (the driver or filesystem provides and
1048 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1052 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1054 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1060 /* getting an ENOSYS error indicates that direct mmap isn't
1061 * possible (as opposed to tried but failed) so we'll fall
1062 * through to making a private copy of the data and mapping
1068 * set up a private mapping or an anonymous shared mapping
1070 static int do_mmap_private(struct vm_area_struct
*vma
,
1071 struct vm_region
*region
,
1075 unsigned long total
, point
, n
, rlen
;
1079 /* invoke the file's mapping function so that it can keep track of
1080 * shared mappings on devices or memory
1081 * - VM_MAYSHARE will be set if it may attempt to share
1084 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1086 /* shouldn't return success if we're not sharing */
1087 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1088 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1094 /* getting an ENOSYS error indicates that direct mmap isn't
1095 * possible (as opposed to tried but failed) so we'll try to
1096 * make a private copy of the data and map that instead */
1099 rlen
= PAGE_ALIGN(len
);
1101 /* allocate some memory to hold the mapping
1102 * - note that this may not return a page-aligned address if the object
1103 * we're allocating is smaller than a page
1105 order
= get_order(rlen
);
1106 kdebug("alloc order %d for %lx", order
, len
);
1108 pages
= alloc_pages(GFP_KERNEL
, order
);
1113 atomic_long_add(total
, &mmap_pages_allocated
);
1115 point
= rlen
>> PAGE_SHIFT
;
1117 /* we allocated a power-of-2 sized page set, so we may want to trim off
1119 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
) {
1120 while (total
> point
) {
1121 order
= ilog2(total
- point
);
1123 kdebug("shave %lu/%lu @%lu", n
, total
- point
, total
);
1124 atomic_long_sub(n
, &mmap_pages_allocated
);
1126 set_page_refcounted(pages
+ total
);
1127 __free_pages(pages
+ total
, order
);
1131 for (point
= 1; point
< total
; point
++)
1132 set_page_refcounted(&pages
[point
]);
1134 base
= page_address(pages
);
1135 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1136 region
->vm_start
= (unsigned long) base
;
1137 region
->vm_end
= region
->vm_start
+ rlen
;
1138 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1140 vma
->vm_start
= region
->vm_start
;
1141 vma
->vm_end
= region
->vm_start
+ len
;
1144 /* read the contents of a file into the copy */
1145 mm_segment_t old_fs
;
1148 fpos
= vma
->vm_pgoff
;
1149 fpos
<<= PAGE_SHIFT
;
1153 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, rlen
, &fpos
);
1159 /* clear the last little bit */
1161 memset(base
+ ret
, 0, rlen
- ret
);
1164 /* if it's an anonymous mapping, then just clear it */
1165 memset(base
, 0, rlen
);
1171 free_page_series(region
->vm_start
, region
->vm_end
);
1172 region
->vm_start
= vma
->vm_start
= 0;
1173 region
->vm_end
= vma
->vm_end
= 0;
1178 printk("Allocation of length %lu from process %d (%s) failed\n",
1179 len
, current
->pid
, current
->comm
);
1185 * handle mapping creation for uClinux
1187 unsigned long do_mmap_pgoff(struct file
*file
,
1191 unsigned long flags
,
1192 unsigned long pgoff
)
1194 struct vm_area_struct
*vma
;
1195 struct vm_region
*region
;
1197 unsigned long capabilities
, vm_flags
, result
;
1200 kenter(",%lx,%lx,%lx,%lx,%lx", addr
, len
, prot
, flags
, pgoff
);
1202 if (!(flags
& MAP_FIXED
))
1203 addr
= round_hint_to_min(addr
);
1205 /* decide whether we should attempt the mapping, and if so what sort of
1207 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1210 kleave(" = %d [val]", ret
);
1214 /* we've determined that we can make the mapping, now translate what we
1215 * now know into VMA flags */
1216 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1218 /* we're going to need to record the mapping */
1219 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1221 goto error_getting_region
;
1223 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1225 goto error_getting_vma
;
1227 atomic_set(®ion
->vm_usage
, 1);
1228 region
->vm_flags
= vm_flags
;
1229 region
->vm_pgoff
= pgoff
;
1231 INIT_LIST_HEAD(&vma
->anon_vma_node
);
1232 vma
->vm_flags
= vm_flags
;
1233 vma
->vm_pgoff
= pgoff
;
1236 region
->vm_file
= file
;
1238 vma
->vm_file
= file
;
1240 if (vm_flags
& VM_EXECUTABLE
) {
1241 added_exe_file_vma(current
->mm
);
1242 vma
->vm_mm
= current
->mm
;
1246 down_write(&nommu_region_sem
);
1248 /* if we want to share, we need to check for regions created by other
1249 * mmap() calls that overlap with our proposed mapping
1250 * - we can only share with a superset match on most regular files
1251 * - shared mappings on character devices and memory backed files are
1252 * permitted to overlap inexactly as far as we are concerned for in
1253 * these cases, sharing is handled in the driver or filesystem rather
1256 if (vm_flags
& VM_MAYSHARE
) {
1257 struct vm_region
*pregion
;
1258 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1260 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1261 pgend
= pgoff
+ pglen
;
1263 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1264 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1266 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1269 /* search for overlapping mappings on the same file */
1270 if (pregion
->vm_file
->f_path
.dentry
->d_inode
!=
1271 file
->f_path
.dentry
->d_inode
)
1274 if (pregion
->vm_pgoff
>= pgend
)
1277 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1278 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1279 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1280 if (pgoff
>= rpgend
)
1283 /* handle inexactly overlapping matches between
1285 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1286 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1287 /* new mapping is not a subset of the region */
1288 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1289 goto sharing_violation
;
1293 /* we've found a region we can share */
1294 atomic_inc(&pregion
->vm_usage
);
1295 vma
->vm_region
= pregion
;
1296 start
= pregion
->vm_start
;
1297 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1298 vma
->vm_start
= start
;
1299 vma
->vm_end
= start
+ len
;
1301 if (pregion
->vm_flags
& VM_MAPPED_COPY
) {
1302 kdebug("share copy");
1303 vma
->vm_flags
|= VM_MAPPED_COPY
;
1305 kdebug("share mmap");
1306 ret
= do_mmap_shared_file(vma
);
1308 vma
->vm_region
= NULL
;
1311 atomic_dec(&pregion
->vm_usage
);
1313 goto error_just_free
;
1316 fput(region
->vm_file
);
1317 kmem_cache_free(vm_region_jar
, region
);
1323 /* obtain the address at which to make a shared mapping
1324 * - this is the hook for quasi-memory character devices to
1325 * tell us the location of a shared mapping
1327 if (file
&& file
->f_op
->get_unmapped_area
) {
1328 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1330 if (IS_ERR((void *) addr
)) {
1332 if (ret
!= (unsigned long) -ENOSYS
)
1333 goto error_just_free
;
1335 /* the driver refused to tell us where to site
1336 * the mapping so we'll have to attempt to copy
1338 ret
= (unsigned long) -ENODEV
;
1339 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1340 goto error_just_free
;
1342 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1344 vma
->vm_start
= region
->vm_start
= addr
;
1345 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1350 vma
->vm_region
= region
;
1352 /* set up the mapping */
1353 if (file
&& vma
->vm_flags
& VM_SHARED
)
1354 ret
= do_mmap_shared_file(vma
);
1356 ret
= do_mmap_private(vma
, region
, len
);
1358 goto error_put_region
;
1360 add_nommu_region(region
);
1362 /* okay... we have a mapping; now we have to register it */
1363 result
= vma
->vm_start
;
1365 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1368 add_vma_to_mm(current
->mm
, vma
);
1370 up_write(&nommu_region_sem
);
1372 if (prot
& PROT_EXEC
)
1373 flush_icache_range(result
, result
+ len
);
1375 kleave(" = %lx", result
);
1379 __put_nommu_region(region
);
1383 if (vma
->vm_flags
& VM_EXECUTABLE
)
1384 removed_exe_file_vma(vma
->vm_mm
);
1386 kmem_cache_free(vm_area_cachep
, vma
);
1388 kleave(" = %d [pr]", ret
);
1392 up_write(&nommu_region_sem
);
1394 fput(region
->vm_file
);
1395 kmem_cache_free(vm_region_jar
, region
);
1397 if (vma
->vm_flags
& VM_EXECUTABLE
)
1398 removed_exe_file_vma(vma
->vm_mm
);
1399 kmem_cache_free(vm_area_cachep
, vma
);
1400 kleave(" = %d", ret
);
1404 up_write(&nommu_region_sem
);
1405 printk(KERN_WARNING
"Attempt to share mismatched mappings\n");
1410 kmem_cache_free(vm_region_jar
, region
);
1411 printk(KERN_WARNING
"Allocation of vma for %lu byte allocation"
1412 " from process %d failed\n",
1417 error_getting_region
:
1418 printk(KERN_WARNING
"Allocation of vm region for %lu byte allocation"
1419 " from process %d failed\n",
1424 EXPORT_SYMBOL(do_mmap_pgoff
);
1427 * split a vma into two pieces at address 'addr', a new vma is allocated either
1428 * for the first part or the tail.
1430 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1431 unsigned long addr
, int new_below
)
1433 struct vm_area_struct
*new;
1434 struct vm_region
*region
;
1435 unsigned long npages
;
1439 /* we're only permitted to split anonymous regions that have a single
1442 atomic_read(&vma
->vm_region
->vm_usage
) != 1)
1445 if (mm
->map_count
>= sysctl_max_map_count
)
1448 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1452 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1454 kmem_cache_free(vm_region_jar
, region
);
1458 /* most fields are the same, copy all, and then fixup */
1460 *region
= *vma
->vm_region
;
1461 new->vm_region
= region
;
1463 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1466 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1468 region
->vm_start
= new->vm_start
= addr
;
1469 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1472 if (new->vm_ops
&& new->vm_ops
->open
)
1473 new->vm_ops
->open(new);
1475 delete_vma_from_mm(vma
);
1476 down_write(&nommu_region_sem
);
1477 delete_nommu_region(vma
->vm_region
);
1479 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1480 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1482 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1483 vma
->vm_region
->vm_top
= addr
;
1485 add_nommu_region(vma
->vm_region
);
1486 add_nommu_region(new->vm_region
);
1487 up_write(&nommu_region_sem
);
1488 add_vma_to_mm(mm
, vma
);
1489 add_vma_to_mm(mm
, new);
1494 * shrink a VMA by removing the specified chunk from either the beginning or
1497 static int shrink_vma(struct mm_struct
*mm
,
1498 struct vm_area_struct
*vma
,
1499 unsigned long from
, unsigned long to
)
1501 struct vm_region
*region
;
1505 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1507 delete_vma_from_mm(vma
);
1508 if (from
> vma
->vm_start
)
1512 add_vma_to_mm(mm
, vma
);
1514 /* cut the backing region down to size */
1515 region
= vma
->vm_region
;
1516 BUG_ON(atomic_read(®ion
->vm_usage
) != 1);
1518 down_write(&nommu_region_sem
);
1519 delete_nommu_region(region
);
1520 if (from
> region
->vm_start
) {
1521 to
= region
->vm_top
;
1522 region
->vm_top
= region
->vm_end
= from
;
1524 region
->vm_start
= to
;
1526 add_nommu_region(region
);
1527 up_write(&nommu_region_sem
);
1529 free_page_series(from
, to
);
1535 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1536 * VMA, though it need not cover the whole VMA
1538 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1540 struct vm_area_struct
*vma
;
1542 unsigned long end
= start
+ len
;
1545 kenter(",%lx,%zx", start
, len
);
1550 /* find the first potentially overlapping VMA */
1551 vma
= find_vma(mm
, start
);
1553 static int limit
= 0;
1556 "munmap of memory not mmapped by process %d"
1557 " (%s): 0x%lx-0x%lx\n",
1558 current
->pid
, current
->comm
,
1559 start
, start
+ len
- 1);
1565 /* we're allowed to split an anonymous VMA but not a file-backed one */
1568 if (start
> vma
->vm_start
) {
1569 kleave(" = -EINVAL [miss]");
1572 if (end
== vma
->vm_end
)
1573 goto erase_whole_vma
;
1574 rb
= rb_next(&vma
->vm_rb
);
1575 vma
= rb_entry(rb
, struct vm_area_struct
, vm_rb
);
1577 kleave(" = -EINVAL [split file]");
1580 /* the chunk must be a subset of the VMA found */
1581 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1582 goto erase_whole_vma
;
1583 if (start
< vma
->vm_start
|| end
> vma
->vm_end
) {
1584 kleave(" = -EINVAL [superset]");
1587 if (start
& ~PAGE_MASK
) {
1588 kleave(" = -EINVAL [unaligned start]");
1591 if (end
!= vma
->vm_end
&& end
& ~PAGE_MASK
) {
1592 kleave(" = -EINVAL [unaligned split]");
1595 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1596 ret
= split_vma(mm
, vma
, start
, 1);
1598 kleave(" = %d [split]", ret
);
1602 return shrink_vma(mm
, vma
, start
, end
);
1606 delete_vma_from_mm(vma
);
1607 delete_vma(mm
, vma
);
1611 EXPORT_SYMBOL(do_munmap
);
1613 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1616 struct mm_struct
*mm
= current
->mm
;
1618 down_write(&mm
->mmap_sem
);
1619 ret
= do_munmap(mm
, addr
, len
);
1620 up_write(&mm
->mmap_sem
);
1625 * release all the mappings made in a process's VM space
1627 void exit_mmap(struct mm_struct
*mm
)
1629 struct vm_area_struct
*vma
;
1638 while ((vma
= mm
->mmap
)) {
1639 mm
->mmap
= vma
->vm_next
;
1640 delete_vma_from_mm(vma
);
1641 delete_vma(mm
, vma
);
1647 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1653 * expand (or shrink) an existing mapping, potentially moving it at the same
1654 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1656 * under NOMMU conditions, we only permit changing a mapping's size, and only
1657 * as long as it stays within the region allocated by do_mmap_private() and the
1658 * block is not shareable
1660 * MREMAP_FIXED is not supported under NOMMU conditions
1662 unsigned long do_mremap(unsigned long addr
,
1663 unsigned long old_len
, unsigned long new_len
,
1664 unsigned long flags
, unsigned long new_addr
)
1666 struct vm_area_struct
*vma
;
1668 /* insanity checks first */
1669 if (old_len
== 0 || new_len
== 0)
1670 return (unsigned long) -EINVAL
;
1672 if (addr
& ~PAGE_MASK
)
1675 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1676 return (unsigned long) -EINVAL
;
1678 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1680 return (unsigned long) -EINVAL
;
1682 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1683 return (unsigned long) -EFAULT
;
1685 if (vma
->vm_flags
& VM_MAYSHARE
)
1686 return (unsigned long) -EPERM
;
1688 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1689 return (unsigned long) -ENOMEM
;
1691 /* all checks complete - do it */
1692 vma
->vm_end
= vma
->vm_start
+ new_len
;
1693 return vma
->vm_start
;
1695 EXPORT_SYMBOL(do_mremap
);
1697 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1698 unsigned long, new_len
, unsigned long, flags
,
1699 unsigned long, new_addr
)
1703 down_write(¤t
->mm
->mmap_sem
);
1704 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1705 up_write(¤t
->mm
->mmap_sem
);
1709 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1710 unsigned int foll_flags
)
1715 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long from
,
1716 unsigned long to
, unsigned long size
, pgprot_t prot
)
1718 vma
->vm_start
= vma
->vm_pgoff
<< PAGE_SHIFT
;
1721 EXPORT_SYMBOL(remap_pfn_range
);
1723 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1724 unsigned long pgoff
)
1726 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1728 if (!(vma
->vm_flags
& VM_USERMAP
))
1731 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1732 vma
->vm_end
= vma
->vm_start
+ size
;
1736 EXPORT_SYMBOL(remap_vmalloc_range
);
1738 void swap_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1742 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1743 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1748 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1752 void unmap_mapping_range(struct address_space
*mapping
,
1753 loff_t
const holebegin
, loff_t
const holelen
,
1757 EXPORT_SYMBOL(unmap_mapping_range
);
1760 * ask for an unmapped area at which to create a mapping on a file
1762 unsigned long get_unmapped_area(struct file
*file
, unsigned long addr
,
1763 unsigned long len
, unsigned long pgoff
,
1764 unsigned long flags
)
1766 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long,
1767 unsigned long, unsigned long);
1769 get_area
= current
->mm
->get_unmapped_area
;
1770 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1771 get_area
= file
->f_op
->get_unmapped_area
;
1776 return get_area(file
, addr
, len
, pgoff
, flags
);
1778 EXPORT_SYMBOL(get_unmapped_area
);
1781 * Check that a process has enough memory to allocate a new virtual
1782 * mapping. 0 means there is enough memory for the allocation to
1783 * succeed and -ENOMEM implies there is not.
1785 * We currently support three overcommit policies, which are set via the
1786 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1788 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1789 * Additional code 2002 Jul 20 by Robert Love.
1791 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1793 * Note this is a helper function intended to be used by LSMs which
1794 * wish to use this logic.
1796 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1798 unsigned long free
, allowed
;
1800 vm_acct_memory(pages
);
1803 * Sometimes we want to use more memory than we have
1805 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1808 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1811 free
= global_page_state(NR_FILE_PAGES
);
1812 free
+= nr_swap_pages
;
1815 * Any slabs which are created with the
1816 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1817 * which are reclaimable, under pressure. The dentry
1818 * cache and most inode caches should fall into this
1820 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1823 * Leave the last 3% for root
1832 * nr_free_pages() is very expensive on large systems,
1833 * only call if we're about to fail.
1835 n
= nr_free_pages();
1838 * Leave reserved pages. The pages are not for anonymous pages.
1840 if (n
<= totalreserve_pages
)
1843 n
-= totalreserve_pages
;
1846 * Leave the last 3% for root
1858 allowed
= totalram_pages
* sysctl_overcommit_ratio
/ 100;
1860 * Leave the last 3% for root
1863 allowed
-= allowed
/ 32;
1864 allowed
+= total_swap_pages
;
1866 /* Don't let a single process grow too big:
1867 leave 3% of the size of this process for other processes */
1869 allowed
-= mm
->total_vm
/ 32;
1871 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
1875 vm_unacct_memory(pages
);
1880 int in_gate_area_no_task(unsigned long addr
)
1885 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1890 EXPORT_SYMBOL(filemap_fault
);
1893 * Access another process' address space.
1894 * - source/target buffer must be kernel space
1896 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
1898 struct vm_area_struct
*vma
;
1899 struct mm_struct
*mm
;
1901 if (addr
+ len
< addr
)
1904 mm
= get_task_mm(tsk
);
1908 down_read(&mm
->mmap_sem
);
1910 /* the access must start within one of the target process's mappings */
1911 vma
= find_vma(mm
, addr
);
1913 /* don't overrun this mapping */
1914 if (addr
+ len
>= vma
->vm_end
)
1915 len
= vma
->vm_end
- addr
;
1917 /* only read or write mappings where it is permitted */
1918 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1919 len
-= copy_to_user((void *) addr
, buf
, len
);
1920 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1921 len
-= copy_from_user(buf
, (void *) addr
, len
);
1928 up_read(&mm
->mmap_sem
);