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 /* amount of vm to protect from userspace access */
73 unsigned long mmap_min_addr
= CONFIG_DEFAULT_MMAP_MIN_ADDR
;
75 atomic_long_t mmap_pages_allocated
;
77 EXPORT_SYMBOL(mem_map
);
78 EXPORT_SYMBOL(num_physpages
);
80 /* list of mapped, potentially shareable regions */
81 static struct kmem_cache
*vm_region_jar
;
82 struct rb_root nommu_region_tree
= RB_ROOT
;
83 DECLARE_RWSEM(nommu_region_sem
);
85 struct vm_operations_struct generic_file_vm_ops
= {
89 * Handle all mappings that got truncated by a "truncate()"
92 * NOTE! We have to be ready to update the memory sharing
93 * between the file and the memory map for a potential last
94 * incomplete page. Ugly, but necessary.
96 int vmtruncate(struct inode
*inode
, loff_t offset
)
98 struct address_space
*mapping
= inode
->i_mapping
;
101 if (inode
->i_size
< offset
)
103 i_size_write(inode
, offset
);
105 truncate_inode_pages(mapping
, offset
);
109 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
110 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
112 if (offset
> inode
->i_sb
->s_maxbytes
)
114 i_size_write(inode
, offset
);
117 if (inode
->i_op
->truncate
)
118 inode
->i_op
->truncate(inode
);
121 send_sig(SIGXFSZ
, current
, 0);
126 EXPORT_SYMBOL(vmtruncate
);
129 * Return the total memory allocated for this pointer, not
130 * just what the caller asked for.
132 * Doesn't have to be accurate, i.e. may have races.
134 unsigned int kobjsize(const void *objp
)
139 * If the object we have should not have ksize performed on it,
142 if (!objp
|| !virt_addr_valid(objp
))
145 page
= virt_to_head_page(objp
);
148 * If the allocator sets PageSlab, we know the pointer came from
155 * If it's not a compound page, see if we have a matching VMA
156 * region. This test is intentionally done in reverse order,
157 * so if there's no VMA, we still fall through and hand back
158 * PAGE_SIZE for 0-order pages.
160 if (!PageCompound(page
)) {
161 struct vm_area_struct
*vma
;
163 vma
= find_vma(current
->mm
, (unsigned long)objp
);
165 return vma
->vm_end
- vma
->vm_start
;
169 * The ksize() function is only guaranteed to work for pointers
170 * returned by kmalloc(). So handle arbitrary pointers here.
172 return PAGE_SIZE
<< compound_order(page
);
175 int __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
176 unsigned long start
, int nr_pages
, int flags
,
177 struct page
**pages
, struct vm_area_struct
**vmas
)
179 struct vm_area_struct
*vma
;
180 unsigned long vm_flags
;
182 int write
= !!(flags
& GUP_FLAGS_WRITE
);
183 int force
= !!(flags
& GUP_FLAGS_FORCE
);
184 int ignore
= !!(flags
& GUP_FLAGS_IGNORE_VMA_PERMISSIONS
);
186 /* calculate required read or write permissions.
187 * - if 'force' is set, we only require the "MAY" flags.
189 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
190 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
192 for (i
= 0; i
< nr_pages
; i
++) {
193 vma
= find_vma(mm
, start
);
195 goto finish_or_fault
;
197 /* protect what we can, including chardevs */
198 if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
) ||
199 (!ignore
&& !(vm_flags
& vma
->vm_flags
)))
200 goto finish_or_fault
;
203 pages
[i
] = virt_to_page(start
);
205 page_cache_get(pages
[i
]);
215 return i
? : -EFAULT
;
220 * get a list of pages in an address range belonging to the specified process
221 * and indicate the VMA that covers each page
222 * - this is potentially dodgy as we may end incrementing the page count of a
223 * slab page or a secondary page from a compound page
224 * - don't permit access to VMAs that don't support it, such as I/O mappings
226 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
227 unsigned long start
, int nr_pages
, int write
, int force
,
228 struct page
**pages
, struct vm_area_struct
**vmas
)
233 flags
|= GUP_FLAGS_WRITE
;
235 flags
|= GUP_FLAGS_FORCE
;
237 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
);
239 EXPORT_SYMBOL(get_user_pages
);
241 DEFINE_RWLOCK(vmlist_lock
);
242 struct vm_struct
*vmlist
;
244 void vfree(const void *addr
)
248 EXPORT_SYMBOL(vfree
);
250 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
253 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
254 * returns only a logical address.
256 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
258 EXPORT_SYMBOL(__vmalloc
);
260 void *vmalloc_user(unsigned long size
)
264 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
267 struct vm_area_struct
*vma
;
269 down_write(¤t
->mm
->mmap_sem
);
270 vma
= find_vma(current
->mm
, (unsigned long)ret
);
272 vma
->vm_flags
|= VM_USERMAP
;
273 up_write(¤t
->mm
->mmap_sem
);
278 EXPORT_SYMBOL(vmalloc_user
);
280 struct page
*vmalloc_to_page(const void *addr
)
282 return virt_to_page(addr
);
284 EXPORT_SYMBOL(vmalloc_to_page
);
286 unsigned long vmalloc_to_pfn(const void *addr
)
288 return page_to_pfn(virt_to_page(addr
));
290 EXPORT_SYMBOL(vmalloc_to_pfn
);
292 long vread(char *buf
, char *addr
, unsigned long count
)
294 memcpy(buf
, addr
, count
);
298 long vwrite(char *buf
, char *addr
, unsigned long count
)
300 /* Don't allow overflow */
301 if ((unsigned long) addr
+ count
< count
)
302 count
= -(unsigned long) addr
;
304 memcpy(addr
, buf
, count
);
309 * vmalloc - allocate virtually continguos memory
311 * @size: allocation size
313 * Allocate enough pages to cover @size from the page level
314 * allocator and map them into continguos kernel virtual space.
316 * For tight control over page level allocator and protection flags
317 * use __vmalloc() instead.
319 void *vmalloc(unsigned long size
)
321 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
323 EXPORT_SYMBOL(vmalloc
);
325 void *vmalloc_node(unsigned long size
, int node
)
327 return vmalloc(size
);
329 EXPORT_SYMBOL(vmalloc_node
);
331 #ifndef PAGE_KERNEL_EXEC
332 # define PAGE_KERNEL_EXEC PAGE_KERNEL
336 * vmalloc_exec - allocate virtually contiguous, executable memory
337 * @size: allocation size
339 * Kernel-internal function to allocate enough pages to cover @size
340 * the page level allocator and map them into contiguous and
341 * executable kernel virtual space.
343 * For tight control over page level allocator and protection flags
344 * use __vmalloc() instead.
347 void *vmalloc_exec(unsigned long size
)
349 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
353 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
354 * @size: allocation size
356 * Allocate enough 32bit PA addressable pages to cover @size from the
357 * page level allocator and map them into continguos kernel virtual space.
359 void *vmalloc_32(unsigned long size
)
361 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
363 EXPORT_SYMBOL(vmalloc_32
);
366 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
367 * @size: allocation size
369 * The resulting memory area is 32bit addressable and zeroed so it can be
370 * mapped to userspace without leaking data.
372 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
373 * remap_vmalloc_range() are permissible.
375 void *vmalloc_32_user(unsigned long size
)
378 * We'll have to sort out the ZONE_DMA bits for 64-bit,
379 * but for now this can simply use vmalloc_user() directly.
381 return vmalloc_user(size
);
383 EXPORT_SYMBOL(vmalloc_32_user
);
385 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
392 void vunmap(const void *addr
)
396 EXPORT_SYMBOL(vunmap
);
398 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
403 EXPORT_SYMBOL(vm_map_ram
);
405 void vm_unmap_ram(const void *mem
, unsigned int count
)
409 EXPORT_SYMBOL(vm_unmap_ram
);
411 void vm_unmap_aliases(void)
414 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
417 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
420 void __attribute__((weak
)) vmalloc_sync_all(void)
424 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
429 EXPORT_SYMBOL(vm_insert_page
);
432 * sys_brk() for the most part doesn't need the global kernel
433 * lock, except when an application is doing something nasty
434 * like trying to un-brk an area that has already been mapped
435 * to a regular file. in this case, the unmapping will need
436 * to invoke file system routines that need the global lock.
438 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
440 struct mm_struct
*mm
= current
->mm
;
442 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
449 * Always allow shrinking brk
451 if (brk
<= mm
->brk
) {
457 * Ok, looks good - let it rip.
459 return mm
->brk
= brk
;
463 * initialise the VMA and region record slabs
465 void __init
mmap_init(void)
469 ret
= percpu_counter_init(&vm_committed_as
, 0);
471 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
);
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 BUG_ON(unlikely(last
->vm_end
<= last
->vm_start
));
490 BUG_ON(unlikely(last
->vm_top
< last
->vm_end
));
492 while ((p
= rb_next(lastp
))) {
493 region
= rb_entry(p
, struct vm_region
, vm_rb
);
494 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
496 BUG_ON(unlikely(region
->vm_end
<= region
->vm_start
));
497 BUG_ON(unlikely(region
->vm_top
< region
->vm_end
));
498 BUG_ON(unlikely(region
->vm_start
< last
->vm_top
));
504 static void validate_nommu_regions(void)
510 * add a region into the global tree
512 static void add_nommu_region(struct vm_region
*region
)
514 struct vm_region
*pregion
;
515 struct rb_node
**p
, *parent
;
517 validate_nommu_regions();
520 p
= &nommu_region_tree
.rb_node
;
523 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
524 if (region
->vm_start
< pregion
->vm_start
)
526 else if (region
->vm_start
> pregion
->vm_start
)
528 else if (pregion
== region
)
534 rb_link_node(®ion
->vm_rb
, parent
, p
);
535 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
537 validate_nommu_regions();
541 * delete a region from the global tree
543 static void delete_nommu_region(struct vm_region
*region
)
545 BUG_ON(!nommu_region_tree
.rb_node
);
547 validate_nommu_regions();
548 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
549 validate_nommu_regions();
553 * free a contiguous series of pages
555 static void free_page_series(unsigned long from
, unsigned long to
)
557 for (; from
< to
; from
+= PAGE_SIZE
) {
558 struct page
*page
= virt_to_page(from
);
560 kdebug("- free %lx", from
);
561 atomic_long_dec(&mmap_pages_allocated
);
562 if (page_count(page
) != 1)
563 kdebug("free page %p: refcount not one: %d",
564 page
, page_count(page
));
570 * release a reference to a region
571 * - the caller must hold the region semaphore for writing, which this releases
572 * - the region may not have been added to the tree yet, in which case vm_top
573 * will equal vm_start
575 static void __put_nommu_region(struct vm_region
*region
)
576 __releases(nommu_region_sem
)
578 kenter("%p{%d}", region
, atomic_read(®ion
->vm_usage
));
580 BUG_ON(!nommu_region_tree
.rb_node
);
582 if (atomic_dec_and_test(®ion
->vm_usage
)) {
583 if (region
->vm_top
> region
->vm_start
)
584 delete_nommu_region(region
);
585 up_write(&nommu_region_sem
);
588 fput(region
->vm_file
);
590 /* IO memory and memory shared directly out of the pagecache
591 * from ramfs/tmpfs mustn't be released here */
592 if (region
->vm_flags
& VM_MAPPED_COPY
) {
593 kdebug("free series");
594 free_page_series(region
->vm_start
, region
->vm_top
);
596 kmem_cache_free(vm_region_jar
, region
);
598 up_write(&nommu_region_sem
);
603 * release a reference to a region
605 static void put_nommu_region(struct vm_region
*region
)
607 down_write(&nommu_region_sem
);
608 __put_nommu_region(region
);
612 * add a VMA into a process's mm_struct in the appropriate place in the list
613 * and tree and add to the address space's page tree also if not an anonymous
615 * - should be called with mm->mmap_sem held writelocked
617 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
619 struct vm_area_struct
*pvma
, **pp
;
620 struct address_space
*mapping
;
621 struct rb_node
**p
, *parent
;
625 BUG_ON(!vma
->vm_region
);
630 /* add the VMA to the mapping */
632 mapping
= vma
->vm_file
->f_mapping
;
634 flush_dcache_mmap_lock(mapping
);
635 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
636 flush_dcache_mmap_unlock(mapping
);
639 /* add the VMA to the tree */
641 p
= &mm
->mm_rb
.rb_node
;
644 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
646 /* sort by: start addr, end addr, VMA struct addr in that order
647 * (the latter is necessary as we may get identical VMAs) */
648 if (vma
->vm_start
< pvma
->vm_start
)
650 else if (vma
->vm_start
> pvma
->vm_start
)
652 else if (vma
->vm_end
< pvma
->vm_end
)
654 else if (vma
->vm_end
> pvma
->vm_end
)
664 rb_link_node(&vma
->vm_rb
, parent
, p
);
665 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
667 /* add VMA to the VMA list also */
668 for (pp
= &mm
->mmap
; (pvma
= *pp
); pp
= &(*pp
)->vm_next
) {
669 if (pvma
->vm_start
> vma
->vm_start
)
671 if (pvma
->vm_start
< vma
->vm_start
)
673 if (pvma
->vm_end
< vma
->vm_end
)
682 * delete a VMA from its owning mm_struct and address space
684 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
686 struct vm_area_struct
**pp
;
687 struct address_space
*mapping
;
688 struct mm_struct
*mm
= vma
->vm_mm
;
693 if (mm
->mmap_cache
== vma
)
694 mm
->mmap_cache
= NULL
;
696 /* remove the VMA from the mapping */
698 mapping
= vma
->vm_file
->f_mapping
;
700 flush_dcache_mmap_lock(mapping
);
701 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
702 flush_dcache_mmap_unlock(mapping
);
705 /* remove from the MM's tree and list */
706 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
707 for (pp
= &mm
->mmap
; *pp
; pp
= &(*pp
)->vm_next
) {
718 * destroy a VMA record
720 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
723 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
724 vma
->vm_ops
->close(vma
);
727 if (vma
->vm_flags
& VM_EXECUTABLE
)
728 removed_exe_file_vma(mm
);
730 put_nommu_region(vma
->vm_region
);
731 kmem_cache_free(vm_area_cachep
, vma
);
735 * look up the first VMA in which addr resides, NULL if none
736 * - should be called with mm->mmap_sem at least held readlocked
738 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
740 struct vm_area_struct
*vma
;
741 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
743 /* check the cache first */
744 vma
= mm
->mmap_cache
;
745 if (vma
&& vma
->vm_start
<= addr
&& vma
->vm_end
> addr
)
748 /* trawl the tree (there may be multiple mappings in which addr
750 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
751 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
752 if (vma
->vm_start
> addr
)
754 if (vma
->vm_end
> addr
) {
755 mm
->mmap_cache
= vma
;
762 EXPORT_SYMBOL(find_vma
);
766 * - we don't extend stack VMAs under NOMMU conditions
768 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
770 return find_vma(mm
, addr
);
774 * expand a stack to a given address
775 * - not supported under NOMMU conditions
777 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
783 * look up the first VMA exactly that exactly matches addr
784 * - should be called with mm->mmap_sem at least held readlocked
786 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
790 struct vm_area_struct
*vma
;
791 struct rb_node
*n
= mm
->mm_rb
.rb_node
;
792 unsigned long end
= addr
+ len
;
794 /* check the cache first */
795 vma
= mm
->mmap_cache
;
796 if (vma
&& vma
->vm_start
== addr
&& vma
->vm_end
== end
)
799 /* trawl the tree (there may be multiple mappings in which addr
801 for (n
= rb_first(&mm
->mm_rb
); n
; n
= rb_next(n
)) {
802 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
803 if (vma
->vm_start
< addr
)
805 if (vma
->vm_start
> addr
)
807 if (vma
->vm_end
== end
) {
808 mm
->mmap_cache
= vma
;
817 * determine whether a mapping should be permitted and, if so, what sort of
818 * mapping we're capable of supporting
820 static int validate_mmap_request(struct file
*file
,
826 unsigned long *_capabilities
)
828 unsigned long capabilities
, rlen
;
829 unsigned long reqprot
= prot
;
832 /* do the simple checks first */
833 if (flags
& MAP_FIXED
|| addr
) {
835 "%d: Can't do fixed-address/overlay mmap of RAM\n",
840 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
841 (flags
& MAP_TYPE
) != MAP_SHARED
)
847 /* Careful about overflows.. */
848 rlen
= PAGE_ALIGN(len
);
849 if (!rlen
|| rlen
> TASK_SIZE
)
852 /* offset overflow? */
853 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
857 /* validate file mapping requests */
858 struct address_space
*mapping
;
860 /* files must support mmap */
861 if (!file
->f_op
|| !file
->f_op
->mmap
)
864 /* work out if what we've got could possibly be shared
865 * - we support chardevs that provide their own "memory"
866 * - we support files/blockdevs that are memory backed
868 mapping
= file
->f_mapping
;
870 mapping
= file
->f_path
.dentry
->d_inode
->i_mapping
;
873 if (mapping
&& mapping
->backing_dev_info
)
874 capabilities
= mapping
->backing_dev_info
->capabilities
;
877 /* no explicit capabilities set, so assume some
879 switch (file
->f_path
.dentry
->d_inode
->i_mode
& S_IFMT
) {
882 capabilities
= BDI_CAP_MAP_COPY
;
897 /* eliminate any capabilities that we can't support on this
899 if (!file
->f_op
->get_unmapped_area
)
900 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
901 if (!file
->f_op
->read
)
902 capabilities
&= ~BDI_CAP_MAP_COPY
;
904 if (flags
& MAP_SHARED
) {
905 /* do checks for writing, appending and locking */
906 if ((prot
& PROT_WRITE
) &&
907 !(file
->f_mode
& FMODE_WRITE
))
910 if (IS_APPEND(file
->f_path
.dentry
->d_inode
) &&
911 (file
->f_mode
& FMODE_WRITE
))
914 if (locks_verify_locked(file
->f_path
.dentry
->d_inode
))
917 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
920 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
921 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
922 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
924 printk("MAP_SHARED not completely supported on !MMU\n");
928 /* we mustn't privatise shared mappings */
929 capabilities
&= ~BDI_CAP_MAP_COPY
;
932 /* we're going to read the file into private memory we
934 if (!(capabilities
& BDI_CAP_MAP_COPY
))
937 /* we don't permit a private writable mapping to be
938 * shared with the backing device */
939 if (prot
& PROT_WRITE
)
940 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
943 /* handle executable mappings and implied executable
945 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
946 if (prot
& PROT_EXEC
)
949 else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
950 /* handle implication of PROT_EXEC by PROT_READ */
951 if (current
->personality
& READ_IMPLIES_EXEC
) {
952 if (capabilities
& BDI_CAP_EXEC_MAP
)
956 else if ((prot
& PROT_READ
) &&
957 (prot
& PROT_EXEC
) &&
958 !(capabilities
& BDI_CAP_EXEC_MAP
)
960 /* backing file is not executable, try to copy */
961 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
965 /* anonymous mappings are always memory backed and can be
968 capabilities
= BDI_CAP_MAP_COPY
;
970 /* handle PROT_EXEC implication by PROT_READ */
971 if ((prot
& PROT_READ
) &&
972 (current
->personality
& READ_IMPLIES_EXEC
))
976 /* allow the security API to have its say */
977 ret
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
982 *_capabilities
= capabilities
;
987 * we've determined that we can make the mapping, now translate what we
988 * now know into VMA flags
990 static unsigned long determine_vm_flags(struct file
*file
,
993 unsigned long capabilities
)
995 unsigned long vm_flags
;
997 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
998 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
999 /* vm_flags |= mm->def_flags; */
1001 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
1002 /* attempt to share read-only copies of mapped file chunks */
1003 if (file
&& !(prot
& PROT_WRITE
))
1004 vm_flags
|= VM_MAYSHARE
;
1007 /* overlay a shareable mapping on the backing device or inode
1008 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1010 if (flags
& MAP_SHARED
)
1011 vm_flags
|= VM_MAYSHARE
| VM_SHARED
;
1012 else if ((((vm_flags
& capabilities
) ^ vm_flags
) & BDI_CAP_VMFLAGS
) == 0)
1013 vm_flags
|= VM_MAYSHARE
;
1016 /* refuse to let anyone share private mappings with this process if
1017 * it's being traced - otherwise breakpoints set in it may interfere
1018 * with another untraced process
1020 if ((flags
& MAP_PRIVATE
) && tracehook_expect_breakpoints(current
))
1021 vm_flags
&= ~VM_MAYSHARE
;
1027 * set up a shared mapping on a file (the driver or filesystem provides and
1030 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1034 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1036 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1042 /* getting an ENOSYS error indicates that direct mmap isn't
1043 * possible (as opposed to tried but failed) so we'll fall
1044 * through to making a private copy of the data and mapping
1050 * set up a private mapping or an anonymous shared mapping
1052 static int do_mmap_private(struct vm_area_struct
*vma
,
1053 struct vm_region
*region
,
1057 unsigned long total
, point
, n
, rlen
;
1061 /* invoke the file's mapping function so that it can keep track of
1062 * shared mappings on devices or memory
1063 * - VM_MAYSHARE will be set if it may attempt to share
1066 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1068 /* shouldn't return success if we're not sharing */
1069 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1070 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1076 /* getting an ENOSYS error indicates that direct mmap isn't
1077 * possible (as opposed to tried but failed) so we'll try to
1078 * make a private copy of the data and map that instead */
1081 rlen
= PAGE_ALIGN(len
);
1083 /* allocate some memory to hold the mapping
1084 * - note that this may not return a page-aligned address if the object
1085 * we're allocating is smaller than a page
1087 order
= get_order(rlen
);
1088 kdebug("alloc order %d for %lx", order
, len
);
1090 pages
= alloc_pages(GFP_KERNEL
, order
);
1095 atomic_long_add(total
, &mmap_pages_allocated
);
1097 point
= rlen
>> PAGE_SHIFT
;
1099 /* we allocated a power-of-2 sized page set, so we may want to trim off
1101 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
) {
1102 while (total
> point
) {
1103 order
= ilog2(total
- point
);
1105 kdebug("shave %lu/%lu @%lu", n
, total
- point
, total
);
1106 atomic_long_sub(n
, &mmap_pages_allocated
);
1108 set_page_refcounted(pages
+ total
);
1109 __free_pages(pages
+ total
, order
);
1113 for (point
= 1; point
< total
; point
++)
1114 set_page_refcounted(&pages
[point
]);
1116 base
= page_address(pages
);
1117 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1118 region
->vm_start
= (unsigned long) base
;
1119 region
->vm_end
= region
->vm_start
+ rlen
;
1120 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1122 vma
->vm_start
= region
->vm_start
;
1123 vma
->vm_end
= region
->vm_start
+ len
;
1126 /* read the contents of a file into the copy */
1127 mm_segment_t old_fs
;
1130 fpos
= vma
->vm_pgoff
;
1131 fpos
<<= PAGE_SHIFT
;
1135 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, rlen
, &fpos
);
1141 /* clear the last little bit */
1143 memset(base
+ ret
, 0, rlen
- ret
);
1146 /* if it's an anonymous mapping, then just clear it */
1147 memset(base
, 0, rlen
);
1153 free_page_series(region
->vm_start
, region
->vm_end
);
1154 region
->vm_start
= vma
->vm_start
= 0;
1155 region
->vm_end
= vma
->vm_end
= 0;
1160 printk("Allocation of length %lu from process %d (%s) failed\n",
1161 len
, current
->pid
, current
->comm
);
1167 * handle mapping creation for uClinux
1169 unsigned long do_mmap_pgoff(struct file
*file
,
1173 unsigned long flags
,
1174 unsigned long pgoff
)
1176 struct vm_area_struct
*vma
;
1177 struct vm_region
*region
;
1179 unsigned long capabilities
, vm_flags
, result
;
1182 kenter(",%lx,%lx,%lx,%lx,%lx", addr
, len
, prot
, flags
, pgoff
);
1184 if (!(flags
& MAP_FIXED
))
1185 addr
= round_hint_to_min(addr
);
1187 /* decide whether we should attempt the mapping, and if so what sort of
1189 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1192 kleave(" = %d [val]", ret
);
1196 /* we've determined that we can make the mapping, now translate what we
1197 * now know into VMA flags */
1198 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1200 /* we're going to need to record the mapping */
1201 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1203 goto error_getting_region
;
1205 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1207 goto error_getting_vma
;
1209 atomic_set(®ion
->vm_usage
, 1);
1210 region
->vm_flags
= vm_flags
;
1211 region
->vm_pgoff
= pgoff
;
1213 INIT_LIST_HEAD(&vma
->anon_vma_node
);
1214 vma
->vm_flags
= vm_flags
;
1215 vma
->vm_pgoff
= pgoff
;
1218 region
->vm_file
= file
;
1220 vma
->vm_file
= file
;
1222 if (vm_flags
& VM_EXECUTABLE
) {
1223 added_exe_file_vma(current
->mm
);
1224 vma
->vm_mm
= current
->mm
;
1228 down_write(&nommu_region_sem
);
1230 /* if we want to share, we need to check for regions created by other
1231 * mmap() calls that overlap with our proposed mapping
1232 * - we can only share with a superset match on most regular files
1233 * - shared mappings on character devices and memory backed files are
1234 * permitted to overlap inexactly as far as we are concerned for in
1235 * these cases, sharing is handled in the driver or filesystem rather
1238 if (vm_flags
& VM_MAYSHARE
) {
1239 struct vm_region
*pregion
;
1240 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1242 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1243 pgend
= pgoff
+ pglen
;
1245 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1246 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1248 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1251 /* search for overlapping mappings on the same file */
1252 if (pregion
->vm_file
->f_path
.dentry
->d_inode
!=
1253 file
->f_path
.dentry
->d_inode
)
1256 if (pregion
->vm_pgoff
>= pgend
)
1259 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1260 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1261 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1262 if (pgoff
>= rpgend
)
1265 /* handle inexactly overlapping matches between
1267 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1268 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1269 /* new mapping is not a subset of the region */
1270 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1271 goto sharing_violation
;
1275 /* we've found a region we can share */
1276 atomic_inc(&pregion
->vm_usage
);
1277 vma
->vm_region
= pregion
;
1278 start
= pregion
->vm_start
;
1279 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1280 vma
->vm_start
= start
;
1281 vma
->vm_end
= start
+ len
;
1283 if (pregion
->vm_flags
& VM_MAPPED_COPY
) {
1284 kdebug("share copy");
1285 vma
->vm_flags
|= VM_MAPPED_COPY
;
1287 kdebug("share mmap");
1288 ret
= do_mmap_shared_file(vma
);
1290 vma
->vm_region
= NULL
;
1293 atomic_dec(&pregion
->vm_usage
);
1295 goto error_just_free
;
1298 fput(region
->vm_file
);
1299 kmem_cache_free(vm_region_jar
, region
);
1305 /* obtain the address at which to make a shared mapping
1306 * - this is the hook for quasi-memory character devices to
1307 * tell us the location of a shared mapping
1309 if (file
&& file
->f_op
->get_unmapped_area
) {
1310 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1312 if (IS_ERR((void *) addr
)) {
1314 if (ret
!= (unsigned long) -ENOSYS
)
1315 goto error_just_free
;
1317 /* the driver refused to tell us where to site
1318 * the mapping so we'll have to attempt to copy
1320 ret
= (unsigned long) -ENODEV
;
1321 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1322 goto error_just_free
;
1324 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1326 vma
->vm_start
= region
->vm_start
= addr
;
1327 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1332 vma
->vm_region
= region
;
1334 /* set up the mapping */
1335 if (file
&& vma
->vm_flags
& VM_SHARED
)
1336 ret
= do_mmap_shared_file(vma
);
1338 ret
= do_mmap_private(vma
, region
, len
);
1340 goto error_put_region
;
1342 add_nommu_region(region
);
1344 /* okay... we have a mapping; now we have to register it */
1345 result
= vma
->vm_start
;
1347 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1350 add_vma_to_mm(current
->mm
, vma
);
1352 up_write(&nommu_region_sem
);
1354 if (prot
& PROT_EXEC
)
1355 flush_icache_range(result
, result
+ len
);
1357 kleave(" = %lx", result
);
1361 __put_nommu_region(region
);
1365 if (vma
->vm_flags
& VM_EXECUTABLE
)
1366 removed_exe_file_vma(vma
->vm_mm
);
1368 kmem_cache_free(vm_area_cachep
, vma
);
1370 kleave(" = %d [pr]", ret
);
1374 up_write(&nommu_region_sem
);
1376 fput(region
->vm_file
);
1377 kmem_cache_free(vm_region_jar
, region
);
1379 if (vma
->vm_flags
& VM_EXECUTABLE
)
1380 removed_exe_file_vma(vma
->vm_mm
);
1381 kmem_cache_free(vm_area_cachep
, vma
);
1382 kleave(" = %d", ret
);
1386 up_write(&nommu_region_sem
);
1387 printk(KERN_WARNING
"Attempt to share mismatched mappings\n");
1392 kmem_cache_free(vm_region_jar
, region
);
1393 printk(KERN_WARNING
"Allocation of vma for %lu byte allocation"
1394 " from process %d failed\n",
1399 error_getting_region
:
1400 printk(KERN_WARNING
"Allocation of vm region for %lu byte allocation"
1401 " from process %d failed\n",
1406 EXPORT_SYMBOL(do_mmap_pgoff
);
1409 * split a vma into two pieces at address 'addr', a new vma is allocated either
1410 * for the first part or the tail.
1412 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1413 unsigned long addr
, int new_below
)
1415 struct vm_area_struct
*new;
1416 struct vm_region
*region
;
1417 unsigned long npages
;
1421 /* we're only permitted to split anonymous regions that have a single
1424 atomic_read(&vma
->vm_region
->vm_usage
) != 1)
1427 if (mm
->map_count
>= sysctl_max_map_count
)
1430 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1434 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1436 kmem_cache_free(vm_region_jar
, region
);
1440 /* most fields are the same, copy all, and then fixup */
1442 *region
= *vma
->vm_region
;
1443 new->vm_region
= region
;
1445 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1448 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1450 region
->vm_start
= new->vm_start
= addr
;
1451 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1454 if (new->vm_ops
&& new->vm_ops
->open
)
1455 new->vm_ops
->open(new);
1457 delete_vma_from_mm(vma
);
1458 down_write(&nommu_region_sem
);
1459 delete_nommu_region(vma
->vm_region
);
1461 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1462 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1464 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1465 vma
->vm_region
->vm_top
= addr
;
1467 add_nommu_region(vma
->vm_region
);
1468 add_nommu_region(new->vm_region
);
1469 up_write(&nommu_region_sem
);
1470 add_vma_to_mm(mm
, vma
);
1471 add_vma_to_mm(mm
, new);
1476 * shrink a VMA by removing the specified chunk from either the beginning or
1479 static int shrink_vma(struct mm_struct
*mm
,
1480 struct vm_area_struct
*vma
,
1481 unsigned long from
, unsigned long to
)
1483 struct vm_region
*region
;
1487 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1489 delete_vma_from_mm(vma
);
1490 if (from
> vma
->vm_start
)
1494 add_vma_to_mm(mm
, vma
);
1496 /* cut the backing region down to size */
1497 region
= vma
->vm_region
;
1498 BUG_ON(atomic_read(®ion
->vm_usage
) != 1);
1500 down_write(&nommu_region_sem
);
1501 delete_nommu_region(region
);
1502 if (from
> region
->vm_start
) {
1503 to
= region
->vm_top
;
1504 region
->vm_top
= region
->vm_end
= from
;
1506 region
->vm_start
= to
;
1508 add_nommu_region(region
);
1509 up_write(&nommu_region_sem
);
1511 free_page_series(from
, to
);
1517 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1518 * VMA, though it need not cover the whole VMA
1520 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1522 struct vm_area_struct
*vma
;
1524 unsigned long end
= start
+ len
;
1527 kenter(",%lx,%zx", start
, len
);
1532 /* find the first potentially overlapping VMA */
1533 vma
= find_vma(mm
, start
);
1535 static int limit
= 0;
1538 "munmap of memory not mmapped by process %d"
1539 " (%s): 0x%lx-0x%lx\n",
1540 current
->pid
, current
->comm
,
1541 start
, start
+ len
- 1);
1547 /* we're allowed to split an anonymous VMA but not a file-backed one */
1550 if (start
> vma
->vm_start
) {
1551 kleave(" = -EINVAL [miss]");
1554 if (end
== vma
->vm_end
)
1555 goto erase_whole_vma
;
1556 rb
= rb_next(&vma
->vm_rb
);
1557 vma
= rb_entry(rb
, struct vm_area_struct
, vm_rb
);
1559 kleave(" = -EINVAL [split file]");
1562 /* the chunk must be a subset of the VMA found */
1563 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1564 goto erase_whole_vma
;
1565 if (start
< vma
->vm_start
|| end
> vma
->vm_end
) {
1566 kleave(" = -EINVAL [superset]");
1569 if (start
& ~PAGE_MASK
) {
1570 kleave(" = -EINVAL [unaligned start]");
1573 if (end
!= vma
->vm_end
&& end
& ~PAGE_MASK
) {
1574 kleave(" = -EINVAL [unaligned split]");
1577 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1578 ret
= split_vma(mm
, vma
, start
, 1);
1580 kleave(" = %d [split]", ret
);
1584 return shrink_vma(mm
, vma
, start
, end
);
1588 delete_vma_from_mm(vma
);
1589 delete_vma(mm
, vma
);
1593 EXPORT_SYMBOL(do_munmap
);
1595 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1598 struct mm_struct
*mm
= current
->mm
;
1600 down_write(&mm
->mmap_sem
);
1601 ret
= do_munmap(mm
, addr
, len
);
1602 up_write(&mm
->mmap_sem
);
1607 * release all the mappings made in a process's VM space
1609 void exit_mmap(struct mm_struct
*mm
)
1611 struct vm_area_struct
*vma
;
1620 while ((vma
= mm
->mmap
)) {
1621 mm
->mmap
= vma
->vm_next
;
1622 delete_vma_from_mm(vma
);
1623 delete_vma(mm
, vma
);
1629 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1635 * expand (or shrink) an existing mapping, potentially moving it at the same
1636 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1638 * under NOMMU conditions, we only permit changing a mapping's size, and only
1639 * as long as it stays within the region allocated by do_mmap_private() and the
1640 * block is not shareable
1642 * MREMAP_FIXED is not supported under NOMMU conditions
1644 unsigned long do_mremap(unsigned long addr
,
1645 unsigned long old_len
, unsigned long new_len
,
1646 unsigned long flags
, unsigned long new_addr
)
1648 struct vm_area_struct
*vma
;
1650 /* insanity checks first */
1651 if (old_len
== 0 || new_len
== 0)
1652 return (unsigned long) -EINVAL
;
1654 if (addr
& ~PAGE_MASK
)
1657 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1658 return (unsigned long) -EINVAL
;
1660 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1662 return (unsigned long) -EINVAL
;
1664 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1665 return (unsigned long) -EFAULT
;
1667 if (vma
->vm_flags
& VM_MAYSHARE
)
1668 return (unsigned long) -EPERM
;
1670 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1671 return (unsigned long) -ENOMEM
;
1673 /* all checks complete - do it */
1674 vma
->vm_end
= vma
->vm_start
+ new_len
;
1675 return vma
->vm_start
;
1677 EXPORT_SYMBOL(do_mremap
);
1679 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1680 unsigned long, new_len
, unsigned long, flags
,
1681 unsigned long, new_addr
)
1685 down_write(¤t
->mm
->mmap_sem
);
1686 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1687 up_write(¤t
->mm
->mmap_sem
);
1691 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1692 unsigned int foll_flags
)
1697 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long from
,
1698 unsigned long to
, unsigned long size
, pgprot_t prot
)
1700 vma
->vm_start
= vma
->vm_pgoff
<< PAGE_SHIFT
;
1703 EXPORT_SYMBOL(remap_pfn_range
);
1705 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1706 unsigned long pgoff
)
1708 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1710 if (!(vma
->vm_flags
& VM_USERMAP
))
1713 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1714 vma
->vm_end
= vma
->vm_start
+ size
;
1718 EXPORT_SYMBOL(remap_vmalloc_range
);
1720 void swap_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1724 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1725 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1730 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1734 void unmap_mapping_range(struct address_space
*mapping
,
1735 loff_t
const holebegin
, loff_t
const holelen
,
1739 EXPORT_SYMBOL(unmap_mapping_range
);
1742 * ask for an unmapped area at which to create a mapping on a file
1744 unsigned long get_unmapped_area(struct file
*file
, unsigned long addr
,
1745 unsigned long len
, unsigned long pgoff
,
1746 unsigned long flags
)
1748 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long,
1749 unsigned long, unsigned long);
1751 get_area
= current
->mm
->get_unmapped_area
;
1752 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1753 get_area
= file
->f_op
->get_unmapped_area
;
1758 return get_area(file
, addr
, len
, pgoff
, flags
);
1760 EXPORT_SYMBOL(get_unmapped_area
);
1763 * Check that a process has enough memory to allocate a new virtual
1764 * mapping. 0 means there is enough memory for the allocation to
1765 * succeed and -ENOMEM implies there is not.
1767 * We currently support three overcommit policies, which are set via the
1768 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1770 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1771 * Additional code 2002 Jul 20 by Robert Love.
1773 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1775 * Note this is a helper function intended to be used by LSMs which
1776 * wish to use this logic.
1778 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1780 unsigned long free
, allowed
;
1782 vm_acct_memory(pages
);
1785 * Sometimes we want to use more memory than we have
1787 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1790 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1793 free
= global_page_state(NR_FILE_PAGES
);
1794 free
+= nr_swap_pages
;
1797 * Any slabs which are created with the
1798 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1799 * which are reclaimable, under pressure. The dentry
1800 * cache and most inode caches should fall into this
1802 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1805 * Leave the last 3% for root
1814 * nr_free_pages() is very expensive on large systems,
1815 * only call if we're about to fail.
1817 n
= nr_free_pages();
1820 * Leave reserved pages. The pages are not for anonymous pages.
1822 if (n
<= totalreserve_pages
)
1825 n
-= totalreserve_pages
;
1828 * Leave the last 3% for root
1840 allowed
= totalram_pages
* sysctl_overcommit_ratio
/ 100;
1842 * Leave the last 3% for root
1845 allowed
-= allowed
/ 32;
1846 allowed
+= total_swap_pages
;
1848 /* Don't let a single process grow too big:
1849 leave 3% of the size of this process for other processes */
1851 allowed
-= mm
->total_vm
/ 32;
1853 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
1857 vm_unacct_memory(pages
);
1862 int in_gate_area_no_task(unsigned long addr
)
1867 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1872 EXPORT_SYMBOL(filemap_fault
);
1875 * Access another process' address space.
1876 * - source/target buffer must be kernel space
1878 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
1880 struct vm_area_struct
*vma
;
1881 struct mm_struct
*mm
;
1883 if (addr
+ len
< addr
)
1886 mm
= get_task_mm(tsk
);
1890 down_read(&mm
->mmap_sem
);
1892 /* the access must start within one of the target process's mappings */
1893 vma
= find_vma(mm
, addr
);
1895 /* don't overrun this mapping */
1896 if (addr
+ len
>= vma
->vm_end
)
1897 len
= vma
->vm_end
- addr
;
1899 /* only read or write mappings where it is permitted */
1900 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1901 len
-= copy_to_user((void *) addr
, buf
, len
);
1902 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1903 len
-= copy_from_user(buf
, (void *) addr
, len
);
1910 up_read(&mm
->mmap_sem
);