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 atomic_long_t vm_committed_space
= ATOMIC_LONG_INIT(0);
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
= 1; /* page trimming behaviour */
70 int heap_stack_gap
= 0;
72 atomic_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 len
, 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
< len
; 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 len
, 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
,
238 EXPORT_SYMBOL(get_user_pages
);
240 DEFINE_RWLOCK(vmlist_lock
);
241 struct vm_struct
*vmlist
;
243 void vfree(const void *addr
)
247 EXPORT_SYMBOL(vfree
);
249 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
252 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
253 * returns only a logical address.
255 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
257 EXPORT_SYMBOL(__vmalloc
);
259 void *vmalloc_user(unsigned long size
)
263 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
266 struct vm_area_struct
*vma
;
268 down_write(¤t
->mm
->mmap_sem
);
269 vma
= find_vma(current
->mm
, (unsigned long)ret
);
271 vma
->vm_flags
|= VM_USERMAP
;
272 up_write(¤t
->mm
->mmap_sem
);
277 EXPORT_SYMBOL(vmalloc_user
);
279 struct page
*vmalloc_to_page(const void *addr
)
281 return virt_to_page(addr
);
283 EXPORT_SYMBOL(vmalloc_to_page
);
285 unsigned long vmalloc_to_pfn(const void *addr
)
287 return page_to_pfn(virt_to_page(addr
));
289 EXPORT_SYMBOL(vmalloc_to_pfn
);
291 long vread(char *buf
, char *addr
, unsigned long count
)
293 memcpy(buf
, addr
, count
);
297 long vwrite(char *buf
, char *addr
, unsigned long count
)
299 /* Don't allow overflow */
300 if ((unsigned long) addr
+ count
< count
)
301 count
= -(unsigned long) addr
;
303 memcpy(addr
, buf
, count
);
308 * vmalloc - allocate virtually continguos memory
310 * @size: allocation size
312 * Allocate enough pages to cover @size from the page level
313 * allocator and map them into continguos kernel virtual space.
315 * For tight control over page level allocator and protection flags
316 * use __vmalloc() instead.
318 void *vmalloc(unsigned long size
)
320 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
322 EXPORT_SYMBOL(vmalloc
);
324 void *vmalloc_node(unsigned long size
, int node
)
326 return vmalloc(size
);
328 EXPORT_SYMBOL(vmalloc_node
);
330 #ifndef PAGE_KERNEL_EXEC
331 # define PAGE_KERNEL_EXEC PAGE_KERNEL
335 * vmalloc_exec - allocate virtually contiguous, executable memory
336 * @size: allocation size
338 * Kernel-internal function to allocate enough pages to cover @size
339 * the page level allocator and map them into contiguous and
340 * executable kernel virtual space.
342 * For tight control over page level allocator and protection flags
343 * use __vmalloc() instead.
346 void *vmalloc_exec(unsigned long size
)
348 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
352 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
353 * @size: allocation size
355 * Allocate enough 32bit PA addressable pages to cover @size from the
356 * page level allocator and map them into continguos kernel virtual space.
358 void *vmalloc_32(unsigned long size
)
360 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
362 EXPORT_SYMBOL(vmalloc_32
);
365 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
366 * @size: allocation size
368 * The resulting memory area is 32bit addressable and zeroed so it can be
369 * mapped to userspace without leaking data.
371 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
372 * remap_vmalloc_range() are permissible.
374 void *vmalloc_32_user(unsigned long size
)
377 * We'll have to sort out the ZONE_DMA bits for 64-bit,
378 * but for now this can simply use vmalloc_user() directly.
380 return vmalloc_user(size
);
382 EXPORT_SYMBOL(vmalloc_32_user
);
384 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
391 void vunmap(const void *addr
)
395 EXPORT_SYMBOL(vunmap
);
397 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
402 EXPORT_SYMBOL(vm_map_ram
);
404 void vm_unmap_ram(const void *mem
, unsigned int count
)
408 EXPORT_SYMBOL(vm_unmap_ram
);
410 void vm_unmap_aliases(void)
413 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
416 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
419 void __attribute__((weak
)) vmalloc_sync_all(void)
423 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
428 EXPORT_SYMBOL(vm_insert_page
);
431 * sys_brk() for the most part doesn't need the global kernel
432 * lock, except when an application is doing something nasty
433 * like trying to un-brk an area that has already been mapped
434 * to a regular file. in this case, the unmapping will need
435 * to invoke file system routines that need the global lock.
437 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
439 struct mm_struct
*mm
= current
->mm
;
441 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
448 * Always allow shrinking brk
450 if (brk
<= mm
->brk
) {
456 * Ok, looks good - let it rip.
458 return mm
->brk
= brk
;
462 * initialise the VMA and region record slabs
464 void __init
mmap_init(void)
466 vm_region_jar
= kmem_cache_create("vm_region_jar",
467 sizeof(struct vm_region
), 0,
469 vm_area_cachep
= kmem_cache_create("vm_area_struct",
470 sizeof(struct vm_area_struct
), 0,
475 * validate the region tree
476 * - the caller must hold the region lock
478 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
479 static noinline
void validate_nommu_regions(void)
481 struct vm_region
*region
, *last
;
482 struct rb_node
*p
, *lastp
;
484 lastp
= rb_first(&nommu_region_tree
);
488 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
489 if (unlikely(last
->vm_end
<= last
->vm_start
))
491 if (unlikely(last
->vm_top
< last
->vm_end
))
494 while ((p
= rb_next(lastp
))) {
495 region
= rb_entry(p
, struct vm_region
, vm_rb
);
496 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
498 if (unlikely(region
->vm_end
<= region
->vm_start
))
500 if (unlikely(region
->vm_top
< region
->vm_end
))
502 if (unlikely(region
->vm_start
< last
->vm_top
))
509 #define validate_nommu_regions() do {} while(0)
513 * add a region into the global tree
515 static void add_nommu_region(struct vm_region
*region
)
517 struct vm_region
*pregion
;
518 struct rb_node
**p
, *parent
;
520 validate_nommu_regions();
522 BUG_ON(region
->vm_start
& ~PAGE_MASK
);
525 p
= &nommu_region_tree
.rb_node
;
528 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
529 if (region
->vm_start
< pregion
->vm_start
)
531 else if (region
->vm_start
> pregion
->vm_start
)
533 else if (pregion
== region
)
539 rb_link_node(®ion
->vm_rb
, parent
, p
);
540 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
542 validate_nommu_regions();
546 * delete a region from the global tree
548 static void delete_nommu_region(struct vm_region
*region
)
550 BUG_ON(!nommu_region_tree
.rb_node
);
552 validate_nommu_regions();
553 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
554 validate_nommu_regions();
558 * free a contiguous series of pages
560 static void free_page_series(unsigned long from
, unsigned long to
)
562 for (; from
< to
; from
+= PAGE_SIZE
) {
563 struct page
*page
= virt_to_page(from
);
565 kdebug("- free %lx", from
);
566 atomic_dec(&mmap_pages_allocated
);
567 if (page_count(page
) != 1)
568 kdebug("free page %p [%d]", page
, page_count(page
));
574 * release a reference to a region
575 * - the caller must hold the region semaphore, which this releases
576 * - the region may not have been added to the tree yet, in which case vm_top
577 * will equal vm_start
579 static void __put_nommu_region(struct vm_region
*region
)
580 __releases(nommu_region_sem
)
582 kenter("%p{%d}", region
, atomic_read(®ion
->vm_usage
));
584 BUG_ON(!nommu_region_tree
.rb_node
);
586 if (atomic_dec_and_test(®ion
->vm_usage
)) {
587 if (region
->vm_top
> region
->vm_start
)
588 delete_nommu_region(region
);
589 up_write(&nommu_region_sem
);
592 fput(region
->vm_file
);
594 /* IO memory and memory shared directly out of the pagecache
595 * from ramfs/tmpfs mustn't be released here */
596 if (region
->vm_flags
& VM_MAPPED_COPY
) {
597 kdebug("free series");
598 free_page_series(region
->vm_start
, region
->vm_top
);
600 kmem_cache_free(vm_region_jar
, region
);
602 up_write(&nommu_region_sem
);
607 * release a reference to a region
609 static void put_nommu_region(struct vm_region
*region
)
611 down_write(&nommu_region_sem
);
612 __put_nommu_region(region
);
616 * add a VMA into a process's mm_struct in the appropriate place in the list
617 * and tree and add to the address space's page tree also if not an anonymous
619 * - should be called with mm->mmap_sem held writelocked
621 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
623 struct vm_area_struct
*pvma
, **pp
;
624 struct address_space
*mapping
;
625 struct rb_node
**p
, *parent
;
629 BUG_ON(!vma
->vm_region
);
634 /* add the VMA to the mapping */
636 mapping
= vma
->vm_file
->f_mapping
;
638 flush_dcache_mmap_lock(mapping
);
639 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
640 flush_dcache_mmap_unlock(mapping
);
643 /* add the VMA to the tree */
645 p
= &mm
->mm_rb
.rb_node
;
648 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
650 /* sort by: start addr, end addr, VMA struct addr in that order
651 * (the latter is necessary as we may get identical VMAs) */
652 if (vma
->vm_start
< pvma
->vm_start
)
654 else if (vma
->vm_start
> pvma
->vm_start
)
656 else if (vma
->vm_end
< pvma
->vm_end
)
658 else if (vma
->vm_end
> pvma
->vm_end
)
668 rb_link_node(&vma
->vm_rb
, parent
, p
);
669 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
671 /* add VMA to the VMA list also */
672 for (pp
= &mm
->mmap
; (pvma
= *pp
); pp
= &(*pp
)->vm_next
) {
673 if (pvma
->vm_start
> vma
->vm_start
)
675 if (pvma
->vm_start
< vma
->vm_start
)
677 if (pvma
->vm_end
< vma
->vm_end
)
686 * delete a VMA from its owning mm_struct and address space
688 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
690 struct vm_area_struct
**pp
;
691 struct address_space
*mapping
;
692 struct mm_struct
*mm
= vma
->vm_mm
;
697 if (mm
->mmap_cache
== vma
)
698 mm
->mmap_cache
= NULL
;
700 /* remove the VMA from the mapping */
702 mapping
= vma
->vm_file
->f_mapping
;
704 flush_dcache_mmap_lock(mapping
);
705 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
706 flush_dcache_mmap_unlock(mapping
);
709 /* remove from the MM's tree and list */
710 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
711 for (pp
= &mm
->mmap
; *pp
; pp
= &(*pp
)->vm_next
) {
722 * destroy a VMA record
724 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
727 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
728 vma
->vm_ops
->close(vma
);
731 if (vma
->vm_flags
& VM_EXECUTABLE
)
732 removed_exe_file_vma(mm
);
734 put_nommu_region(vma
->vm_region
);
735 kmem_cache_free(vm_area_cachep
, vma
);
739 * look up the first VMA in which addr resides, NULL if none
740 * - should be called with mm->mmap_sem at least held readlocked
742 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
744 struct vm_area_struct
*vma
;
745 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
747 /* check the cache first */
748 vma
= mm
->mmap_cache
;
749 if (vma
&& vma
->vm_start
<= addr
&& vma
->vm_end
> addr
)
752 /* trawl the tree (there may be multiple mappings in which addr
754 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
755 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
756 if (vma
->vm_start
> addr
)
758 if (vma
->vm_end
> addr
) {
759 mm
->mmap_cache
= vma
;
766 EXPORT_SYMBOL(find_vma
);
770 * - we don't extend stack VMAs under NOMMU conditions
772 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
774 return find_vma(mm
, addr
);
778 * expand a stack to a given address
779 * - not supported under NOMMU conditions
781 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
787 * look up the first VMA exactly that exactly matches addr
788 * - should be called with mm->mmap_sem at least held readlocked
790 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
794 struct vm_area_struct
*vma
;
795 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
796 unsigned long end
= addr
+ len
;
798 /* check the cache first */
799 vma
= mm
->mmap_cache
;
800 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_end
== end
)
803 /* trawl the tree (there may be multiple mappings in which addr
805 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
806 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
807 if (vma
->vm_start
< addr
)
809 if (vma
->vm_start
> addr
)
811 if (vma
->vm_end
== end
) {
812 mm
->mmap_cache
= vma
;
821 * determine whether a mapping should be permitted and, if so, what sort of
822 * mapping we're capable of supporting
824 static int validate_mmap_request(struct file
*file
,
830 unsigned long *_capabilities
)
832 unsigned long capabilities
, rlen
;
833 unsigned long reqprot
= prot
;
836 /* do the simple checks first */
837 if (flags
& MAP_FIXED
|| addr
) {
839 "%d: Can't do fixed-address/overlay mmap of RAM\n",
844 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
845 (flags
& MAP_TYPE
) != MAP_SHARED
)
851 /* Careful about overflows.. */
852 rlen
= PAGE_ALIGN(len
);
853 if (!rlen
|| rlen
> TASK_SIZE
)
856 /* offset overflow? */
857 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
861 /* validate file mapping requests */
862 struct address_space
*mapping
;
864 /* files must support mmap */
865 if (!file
->f_op
|| !file
->f_op
->mmap
)
868 /* work out if what we've got could possibly be shared
869 * - we support chardevs that provide their own "memory"
870 * - we support files/blockdevs that are memory backed
872 mapping
= file
->f_mapping
;
874 mapping
= file
->f_path
.dentry
->d_inode
->i_mapping
;
877 if (mapping
&& mapping
->backing_dev_info
)
878 capabilities
= mapping
->backing_dev_info
->capabilities
;
881 /* no explicit capabilities set, so assume some
883 switch (file
->f_path
.dentry
->d_inode
->i_mode
& S_IFMT
) {
886 capabilities
= BDI_CAP_MAP_COPY
;
901 /* eliminate any capabilities that we can't support on this
903 if (!file
->f_op
->get_unmapped_area
)
904 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
905 if (!file
->f_op
->read
)
906 capabilities
&= ~BDI_CAP_MAP_COPY
;
908 if (flags
& MAP_SHARED
) {
909 /* do checks for writing, appending and locking */
910 if ((prot
& PROT_WRITE
) &&
911 !(file
->f_mode
& FMODE_WRITE
))
914 if (IS_APPEND(file
->f_path
.dentry
->d_inode
) &&
915 (file
->f_mode
& FMODE_WRITE
))
918 if (locks_verify_locked(file
->f_path
.dentry
->d_inode
))
921 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
924 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
925 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
926 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
928 printk("MAP_SHARED not completely supported on !MMU\n");
932 /* we mustn't privatise shared mappings */
933 capabilities
&= ~BDI_CAP_MAP_COPY
;
936 /* we're going to read the file into private memory we
938 if (!(capabilities
& BDI_CAP_MAP_COPY
))
941 /* we don't permit a private writable mapping to be
942 * shared with the backing device */
943 if (prot
& PROT_WRITE
)
944 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
947 /* handle executable mappings and implied executable
949 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
950 if (prot
& PROT_EXEC
)
953 else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
954 /* handle implication of PROT_EXEC by PROT_READ */
955 if (current
->personality
& READ_IMPLIES_EXEC
) {
956 if (capabilities
& BDI_CAP_EXEC_MAP
)
960 else if ((prot
& PROT_READ
) &&
961 (prot
& PROT_EXEC
) &&
962 !(capabilities
& BDI_CAP_EXEC_MAP
)
964 /* backing file is not executable, try to copy */
965 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
969 /* anonymous mappings are always memory backed and can be
972 capabilities
= BDI_CAP_MAP_COPY
;
974 /* handle PROT_EXEC implication by PROT_READ */
975 if ((prot
& PROT_READ
) &&
976 (current
->personality
& READ_IMPLIES_EXEC
))
980 /* allow the security API to have its say */
981 ret
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
986 *_capabilities
= capabilities
;
991 * we've determined that we can make the mapping, now translate what we
992 * now know into VMA flags
994 static unsigned long determine_vm_flags(struct file
*file
,
997 unsigned long capabilities
)
999 unsigned long vm_flags
;
1001 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
1002 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1003 /* vm_flags |= mm->def_flags; */
1005 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
1006 /* attempt to share read-only copies of mapped file chunks */
1007 if (file
&& !(prot
& PROT_WRITE
))
1008 vm_flags
|= VM_MAYSHARE
;
1011 /* overlay a shareable mapping on the backing device or inode
1012 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1014 if (flags
& MAP_SHARED
)
1015 vm_flags
|= VM_MAYSHARE
| VM_SHARED
;
1016 else if ((((vm_flags
& capabilities
) ^ vm_flags
) & BDI_CAP_VMFLAGS
) == 0)
1017 vm_flags
|= VM_MAYSHARE
;
1020 /* refuse to let anyone share private mappings with this process if
1021 * it's being traced - otherwise breakpoints set in it may interfere
1022 * with another untraced process
1024 if ((flags
& MAP_PRIVATE
) && tracehook_expect_breakpoints(current
))
1025 vm_flags
&= ~VM_MAYSHARE
;
1031 * set up a shared mapping on a file (the driver or filesystem provides and
1034 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1038 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1040 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1046 /* getting an ENOSYS error indicates that direct mmap isn't
1047 * possible (as opposed to tried but failed) so we'll fall
1048 * through to making a private copy of the data and mapping
1054 * set up a private mapping or an anonymous shared mapping
1056 static int do_mmap_private(struct vm_area_struct
*vma
,
1057 struct vm_region
*region
,
1061 unsigned long total
, point
, n
, rlen
;
1065 /* invoke the file's mapping function so that it can keep track of
1066 * shared mappings on devices or memory
1067 * - VM_MAYSHARE will be set if it may attempt to share
1070 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1072 /* shouldn't return success if we're not sharing */
1073 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1074 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1080 /* getting an ENOSYS error indicates that direct mmap isn't
1081 * possible (as opposed to tried but failed) so we'll try to
1082 * make a private copy of the data and map that instead */
1085 rlen
= PAGE_ALIGN(len
);
1087 /* allocate some memory to hold the mapping
1088 * - note that this may not return a page-aligned address if the object
1089 * we're allocating is smaller than a page
1091 order
= get_order(rlen
);
1092 kdebug("alloc order %d for %lx", order
, len
);
1094 pages
= alloc_pages(GFP_KERNEL
, order
);
1099 atomic_add(total
, &mmap_pages_allocated
);
1101 point
= rlen
>> PAGE_SHIFT
;
1103 /* we allocated a power-of-2 sized page set, so we may want to trim off
1105 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
) {
1106 while (total
> point
) {
1107 order
= ilog2(total
- point
);
1109 kdebug("shave %lu/%lu @%lu", n
, total
- point
, total
);
1110 atomic_sub(n
, &mmap_pages_allocated
);
1112 set_page_refcounted(pages
+ total
);
1113 __free_pages(pages
+ total
, order
);
1117 for (point
= 1; point
< total
; point
++)
1118 set_page_refcounted(&pages
[point
]);
1120 base
= page_address(pages
);
1121 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1122 region
->vm_start
= (unsigned long) base
;
1123 region
->vm_end
= region
->vm_start
+ rlen
;
1124 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1126 vma
->vm_start
= region
->vm_start
;
1127 vma
->vm_end
= region
->vm_start
+ len
;
1130 /* read the contents of a file into the copy */
1131 mm_segment_t old_fs
;
1134 fpos
= vma
->vm_pgoff
;
1135 fpos
<<= PAGE_SHIFT
;
1139 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, rlen
, &fpos
);
1145 /* clear the last little bit */
1147 memset(base
+ ret
, 0, rlen
- ret
);
1150 /* if it's an anonymous mapping, then just clear it */
1151 memset(base
, 0, rlen
);
1157 free_page_series(region
->vm_start
, region
->vm_end
);
1158 region
->vm_start
= vma
->vm_start
= 0;
1159 region
->vm_end
= vma
->vm_end
= 0;
1164 printk("Allocation of length %lu from process %d failed\n",
1171 * handle mapping creation for uClinux
1173 unsigned long do_mmap_pgoff(struct file
*file
,
1177 unsigned long flags
,
1178 unsigned long pgoff
)
1180 struct vm_area_struct
*vma
;
1181 struct vm_region
*region
;
1183 unsigned long capabilities
, vm_flags
, result
;
1186 kenter(",%lx,%lx,%lx,%lx,%lx", addr
, len
, prot
, flags
, pgoff
);
1188 if (!(flags
& MAP_FIXED
))
1189 addr
= round_hint_to_min(addr
);
1191 /* decide whether we should attempt the mapping, and if so what sort of
1193 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1196 kleave(" = %d [val]", ret
);
1200 /* we've determined that we can make the mapping, now translate what we
1201 * now know into VMA flags */
1202 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1204 /* we're going to need to record the mapping */
1205 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1207 goto error_getting_region
;
1209 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1211 goto error_getting_vma
;
1213 atomic_set(®ion
->vm_usage
, 1);
1214 region
->vm_flags
= vm_flags
;
1215 region
->vm_pgoff
= pgoff
;
1217 INIT_LIST_HEAD(&vma
->anon_vma_node
);
1218 vma
->vm_flags
= vm_flags
;
1219 vma
->vm_pgoff
= pgoff
;
1222 region
->vm_file
= file
;
1224 vma
->vm_file
= file
;
1226 if (vm_flags
& VM_EXECUTABLE
) {
1227 added_exe_file_vma(current
->mm
);
1228 vma
->vm_mm
= current
->mm
;
1232 down_write(&nommu_region_sem
);
1234 /* if we want to share, we need to check for regions created by other
1235 * mmap() calls that overlap with our proposed mapping
1236 * - we can only share with a superset match on most regular files
1237 * - shared mappings on character devices and memory backed files are
1238 * permitted to overlap inexactly as far as we are concerned for in
1239 * these cases, sharing is handled in the driver or filesystem rather
1242 if (vm_flags
& VM_MAYSHARE
) {
1243 struct vm_region
*pregion
;
1244 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1246 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1247 pgend
= pgoff
+ pglen
;
1249 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1250 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1252 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1255 /* search for overlapping mappings on the same file */
1256 if (pregion
->vm_file
->f_path
.dentry
->d_inode
!=
1257 file
->f_path
.dentry
->d_inode
)
1260 if (pregion
->vm_pgoff
>= pgend
)
1263 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1264 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1265 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1266 if (pgoff
>= rpgend
)
1269 /* handle inexactly overlapping matches between
1271 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1272 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1273 /* new mapping is not a subset of the region */
1274 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1275 goto sharing_violation
;
1279 /* we've found a region we can share */
1280 atomic_inc(&pregion
->vm_usage
);
1281 vma
->vm_region
= pregion
;
1282 start
= pregion
->vm_start
;
1283 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1284 vma
->vm_start
= start
;
1285 vma
->vm_end
= start
+ len
;
1287 if (pregion
->vm_flags
& VM_MAPPED_COPY
) {
1288 kdebug("share copy");
1289 vma
->vm_flags
|= VM_MAPPED_COPY
;
1291 kdebug("share mmap");
1292 ret
= do_mmap_shared_file(vma
);
1294 vma
->vm_region
= NULL
;
1297 atomic_dec(&pregion
->vm_usage
);
1299 goto error_just_free
;
1302 fput(region
->vm_file
);
1303 kmem_cache_free(vm_region_jar
, region
);
1309 /* obtain the address at which to make a shared mapping
1310 * - this is the hook for quasi-memory character devices to
1311 * tell us the location of a shared mapping
1313 if (file
&& file
->f_op
->get_unmapped_area
) {
1314 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1316 if (IS_ERR((void *) addr
)) {
1318 if (ret
!= (unsigned long) -ENOSYS
)
1319 goto error_just_free
;
1321 /* the driver refused to tell us where to site
1322 * the mapping so we'll have to attempt to copy
1324 ret
= (unsigned long) -ENODEV
;
1325 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1326 goto error_just_free
;
1328 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1330 vma
->vm_start
= region
->vm_start
= addr
;
1331 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1336 vma
->vm_region
= region
;
1338 /* set up the mapping */
1339 if (file
&& vma
->vm_flags
& VM_SHARED
)
1340 ret
= do_mmap_shared_file(vma
);
1342 ret
= do_mmap_private(vma
, region
, len
);
1344 goto error_put_region
;
1346 add_nommu_region(region
);
1348 /* okay... we have a mapping; now we have to register it */
1349 result
= vma
->vm_start
;
1351 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1354 add_vma_to_mm(current
->mm
, vma
);
1356 up_write(&nommu_region_sem
);
1358 if (prot
& PROT_EXEC
)
1359 flush_icache_range(result
, result
+ len
);
1361 kleave(" = %lx", result
);
1365 __put_nommu_region(region
);
1369 if (vma
->vm_flags
& VM_EXECUTABLE
)
1370 removed_exe_file_vma(vma
->vm_mm
);
1372 kmem_cache_free(vm_area_cachep
, vma
);
1374 kleave(" = %d [pr]", ret
);
1378 up_write(&nommu_region_sem
);
1380 fput(region
->vm_file
);
1381 kmem_cache_free(vm_region_jar
, region
);
1383 if (vma
->vm_flags
& VM_EXECUTABLE
)
1384 removed_exe_file_vma(vma
->vm_mm
);
1385 kmem_cache_free(vm_area_cachep
, vma
);
1386 kleave(" = %d", ret
);
1390 up_write(&nommu_region_sem
);
1391 printk(KERN_WARNING
"Attempt to share mismatched mappings\n");
1396 kmem_cache_free(vm_region_jar
, region
);
1397 printk(KERN_WARNING
"Allocation of vma for %lu byte allocation"
1398 " from process %d failed\n",
1403 error_getting_region
:
1404 printk(KERN_WARNING
"Allocation of vm region for %lu byte allocation"
1405 " from process %d failed\n",
1410 EXPORT_SYMBOL(do_mmap_pgoff
);
1413 * split a vma into two pieces at address 'addr', a new vma is allocated either
1414 * for the first part or the tail.
1416 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1417 unsigned long addr
, int new_below
)
1419 struct vm_area_struct
*new;
1420 struct vm_region
*region
;
1421 unsigned long npages
;
1425 /* we're only permitted to split anonymous regions that have a single
1428 atomic_read(&vma
->vm_region
->vm_usage
) != 1)
1431 if (mm
->map_count
>= sysctl_max_map_count
)
1434 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1438 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1440 kmem_cache_free(vm_region_jar
, region
);
1444 /* most fields are the same, copy all, and then fixup */
1446 *region
= *vma
->vm_region
;
1447 new->vm_region
= region
;
1449 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1452 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1454 region
->vm_start
= new->vm_start
= addr
;
1455 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1458 if (new->vm_ops
&& new->vm_ops
->open
)
1459 new->vm_ops
->open(new);
1461 delete_vma_from_mm(vma
);
1462 down_write(&nommu_region_sem
);
1463 delete_nommu_region(vma
->vm_region
);
1465 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1466 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1468 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1469 vma
->vm_region
->vm_top
= addr
;
1471 add_nommu_region(vma
->vm_region
);
1472 add_nommu_region(new->vm_region
);
1473 up_write(&nommu_region_sem
);
1474 add_vma_to_mm(mm
, vma
);
1475 add_vma_to_mm(mm
, new);
1480 * shrink a VMA by removing the specified chunk from either the beginning or
1483 static int shrink_vma(struct mm_struct
*mm
,
1484 struct vm_area_struct
*vma
,
1485 unsigned long from
, unsigned long to
)
1487 struct vm_region
*region
;
1491 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1493 delete_vma_from_mm(vma
);
1494 if (from
> vma
->vm_start
)
1498 add_vma_to_mm(mm
, vma
);
1500 /* cut the backing region down to size */
1501 region
= vma
->vm_region
;
1502 BUG_ON(atomic_read(®ion
->vm_usage
) != 1);
1504 down_write(&nommu_region_sem
);
1505 delete_nommu_region(region
);
1506 if (from
> region
->vm_start
) {
1507 to
= region
->vm_top
;
1508 region
->vm_top
= region
->vm_end
= from
;
1510 region
->vm_start
= to
;
1512 add_nommu_region(region
);
1513 up_write(&nommu_region_sem
);
1515 free_page_series(from
, to
);
1521 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1522 * VMA, though it need not cover the whole VMA
1524 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1526 struct vm_area_struct
*vma
;
1528 unsigned long end
= start
+ len
;
1531 kenter(",%lx,%zx", start
, len
);
1536 /* find the first potentially overlapping VMA */
1537 vma
= find_vma(mm
, start
);
1540 "munmap of memory not mmapped by process %d (%s):"
1542 current
->pid
, current
->comm
, start
, start
+ len
- 1);
1546 /* we're allowed to split an anonymous VMA but not a file-backed one */
1549 if (start
> vma
->vm_start
) {
1550 kleave(" = -EINVAL [miss]");
1553 if (end
== vma
->vm_end
)
1554 goto erase_whole_vma
;
1555 rb
= rb_next(&vma
->vm_rb
);
1556 vma
= rb_entry(rb
, struct vm_area_struct
, vm_rb
);
1558 kleave(" = -EINVAL [split file]");
1561 /* the chunk must be a subset of the VMA found */
1562 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1563 goto erase_whole_vma
;
1564 if (start
< vma
->vm_start
|| end
> vma
->vm_end
) {
1565 kleave(" = -EINVAL [superset]");
1568 if (start
& ~PAGE_MASK
) {
1569 kleave(" = -EINVAL [unaligned start]");
1572 if (end
!= vma
->vm_end
&& end
& ~PAGE_MASK
) {
1573 kleave(" = -EINVAL [unaligned split]");
1576 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1577 ret
= split_vma(mm
, vma
, start
, 1);
1579 kleave(" = %d [split]", ret
);
1583 return shrink_vma(mm
, vma
, start
, end
);
1587 delete_vma_from_mm(vma
);
1588 delete_vma(mm
, vma
);
1592 EXPORT_SYMBOL(do_munmap
);
1594 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1597 struct mm_struct
*mm
= current
->mm
;
1599 down_write(&mm
->mmap_sem
);
1600 ret
= do_munmap(mm
, addr
, len
);
1601 up_write(&mm
->mmap_sem
);
1606 * release all the mappings made in a process's VM space
1608 void exit_mmap(struct mm_struct
*mm
)
1610 struct vm_area_struct
*vma
;
1619 while ((vma
= mm
->mmap
)) {
1620 mm
->mmap
= vma
->vm_next
;
1621 delete_vma_from_mm(vma
);
1622 delete_vma(mm
, vma
);
1628 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1634 * expand (or shrink) an existing mapping, potentially moving it at the same
1635 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1637 * under NOMMU conditions, we only permit changing a mapping's size, and only
1638 * as long as it stays within the region allocated by do_mmap_private() and the
1639 * block is not shareable
1641 * MREMAP_FIXED is not supported under NOMMU conditions
1643 unsigned long do_mremap(unsigned long addr
,
1644 unsigned long old_len
, unsigned long new_len
,
1645 unsigned long flags
, unsigned long new_addr
)
1647 struct vm_area_struct
*vma
;
1649 /* insanity checks first */
1650 if (old_len
== 0 || new_len
== 0)
1651 return (unsigned long) -EINVAL
;
1653 if (addr
& ~PAGE_MASK
)
1656 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1657 return (unsigned long) -EINVAL
;
1659 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1661 return (unsigned long) -EINVAL
;
1663 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1664 return (unsigned long) -EFAULT
;
1666 if (vma
->vm_flags
& VM_MAYSHARE
)
1667 return (unsigned long) -EPERM
;
1669 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1670 return (unsigned long) -ENOMEM
;
1672 /* all checks complete - do it */
1673 vma
->vm_end
= vma
->vm_start
+ new_len
;
1674 return vma
->vm_start
;
1676 EXPORT_SYMBOL(do_mremap
);
1678 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1679 unsigned long, new_len
, unsigned long, flags
,
1680 unsigned long, new_addr
)
1684 down_write(¤t
->mm
->mmap_sem
);
1685 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1686 up_write(¤t
->mm
->mmap_sem
);
1690 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1691 unsigned int foll_flags
)
1696 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long from
,
1697 unsigned long to
, unsigned long size
, pgprot_t prot
)
1699 vma
->vm_start
= vma
->vm_pgoff
<< PAGE_SHIFT
;
1702 EXPORT_SYMBOL(remap_pfn_range
);
1704 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1705 unsigned long pgoff
)
1707 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1709 if (!(vma
->vm_flags
& VM_USERMAP
))
1712 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1713 vma
->vm_end
= vma
->vm_start
+ size
;
1717 EXPORT_SYMBOL(remap_vmalloc_range
);
1719 void swap_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1723 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1724 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1729 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1733 void unmap_mapping_range(struct address_space
*mapping
,
1734 loff_t
const holebegin
, loff_t
const holelen
,
1738 EXPORT_SYMBOL(unmap_mapping_range
);
1741 * ask for an unmapped area at which to create a mapping on a file
1743 unsigned long get_unmapped_area(struct file
*file
, unsigned long addr
,
1744 unsigned long len
, unsigned long pgoff
,
1745 unsigned long flags
)
1747 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long,
1748 unsigned long, unsigned long);
1750 get_area
= current
->mm
->get_unmapped_area
;
1751 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1752 get_area
= file
->f_op
->get_unmapped_area
;
1757 return get_area(file
, addr
, len
, pgoff
, flags
);
1759 EXPORT_SYMBOL(get_unmapped_area
);
1762 * Check that a process has enough memory to allocate a new virtual
1763 * mapping. 0 means there is enough memory for the allocation to
1764 * succeed and -ENOMEM implies there is not.
1766 * We currently support three overcommit policies, which are set via the
1767 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1769 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1770 * Additional code 2002 Jul 20 by Robert Love.
1772 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1774 * Note this is a helper function intended to be used by LSMs which
1775 * wish to use this logic.
1777 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1779 unsigned long free
, allowed
;
1781 vm_acct_memory(pages
);
1784 * Sometimes we want to use more memory than we have
1786 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1789 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1792 free
= global_page_state(NR_FILE_PAGES
);
1793 free
+= nr_swap_pages
;
1796 * Any slabs which are created with the
1797 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1798 * which are reclaimable, under pressure. The dentry
1799 * cache and most inode caches should fall into this
1801 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1804 * Leave the last 3% for root
1813 * nr_free_pages() is very expensive on large systems,
1814 * only call if we're about to fail.
1816 n
= nr_free_pages();
1819 * Leave reserved pages. The pages are not for anonymous pages.
1821 if (n
<= totalreserve_pages
)
1824 n
-= totalreserve_pages
;
1827 * Leave the last 3% for root
1839 allowed
= totalram_pages
* sysctl_overcommit_ratio
/ 100;
1841 * Leave the last 3% for root
1844 allowed
-= allowed
/ 32;
1845 allowed
+= total_swap_pages
;
1847 /* Don't let a single process grow too big:
1848 leave 3% of the size of this process for other processes */
1850 allowed
-= mm
->total_vm
/ 32;
1853 * cast `allowed' as a signed long because vm_committed_space
1854 * sometimes has a negative value
1856 if (atomic_long_read(&vm_committed_space
) < (long)allowed
)
1859 vm_unacct_memory(pages
);
1864 int in_gate_area_no_task(unsigned long addr
)
1869 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1874 EXPORT_SYMBOL(filemap_fault
);
1877 * Access another process' address space.
1878 * - source/target buffer must be kernel space
1880 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
1882 struct vm_area_struct
*vma
;
1883 struct mm_struct
*mm
;
1885 if (addr
+ len
< addr
)
1888 mm
= get_task_mm(tsk
);
1892 down_read(&mm
->mmap_sem
);
1894 /* the access must start within one of the target process's mappings */
1895 vma
= find_vma(mm
, addr
);
1897 /* don't overrun this mapping */
1898 if (addr
+ len
>= vma
->vm_end
)
1899 len
= vma
->vm_end
- addr
;
1901 /* only read or write mappings where it is permitted */
1902 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1903 len
-= copy_to_user((void *) addr
, buf
, len
);
1904 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1905 len
-= copy_from_user(buf
, (void *) addr
, len
);
1912 up_read(&mm
->mmap_sem
);