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) 2004-2005 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 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>
39 unsigned long max_mapnr
;
40 unsigned long num_physpages
;
41 unsigned long askedalloc
, realalloc
;
42 atomic_long_t vm_committed_space
= ATOMIC_LONG_INIT(0);
43 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
44 int sysctl_overcommit_ratio
= 50; /* default is 50% */
45 int sysctl_max_map_count
= DEFAULT_MAX_MAP_COUNT
;
46 int heap_stack_gap
= 0;
48 EXPORT_SYMBOL(mem_map
);
49 EXPORT_SYMBOL(num_physpages
);
51 /* list of shareable VMAs */
52 struct rb_root nommu_vma_tree
= RB_ROOT
;
53 DECLARE_RWSEM(nommu_vma_sem
);
55 struct vm_operations_struct generic_file_vm_ops
= {
59 * Handle all mappings that got truncated by a "truncate()"
62 * NOTE! We have to be ready to update the memory sharing
63 * between the file and the memory map for a potential last
64 * incomplete page. Ugly, but necessary.
66 int vmtruncate(struct inode
*inode
, loff_t offset
)
68 struct address_space
*mapping
= inode
->i_mapping
;
71 if (inode
->i_size
< offset
)
73 i_size_write(inode
, offset
);
75 truncate_inode_pages(mapping
, offset
);
79 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
80 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
82 if (offset
> inode
->i_sb
->s_maxbytes
)
84 i_size_write(inode
, offset
);
87 if (inode
->i_op
&& inode
->i_op
->truncate
)
88 inode
->i_op
->truncate(inode
);
91 send_sig(SIGXFSZ
, current
, 0);
96 EXPORT_SYMBOL(vmtruncate
);
99 * Return the total memory allocated for this pointer, not
100 * just what the caller asked for.
102 * Doesn't have to be accurate, i.e. may have races.
104 unsigned int kobjsize(const void *objp
)
109 * If the object we have should not have ksize performed on it,
112 if (!objp
|| !virt_addr_valid(objp
))
115 page
= virt_to_head_page(objp
);
118 * If the allocator sets PageSlab, we know the pointer came from
125 * The ksize() function is only guaranteed to work for pointers
126 * returned by kmalloc(). So handle arbitrary pointers here.
128 return PAGE_SIZE
<< compound_order(page
);
132 * get a list of pages in an address range belonging to the specified process
133 * and indicate the VMA that covers each page
134 * - this is potentially dodgy as we may end incrementing the page count of a
135 * slab page or a secondary page from a compound page
136 * - don't permit access to VMAs that don't support it, such as I/O mappings
138 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
139 unsigned long start
, int len
, int write
, int force
,
140 struct page
**pages
, struct vm_area_struct
**vmas
)
142 struct vm_area_struct
*vma
;
143 unsigned long vm_flags
;
146 /* calculate required read or write permissions.
147 * - if 'force' is set, we only require the "MAY" flags.
149 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
150 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
152 for (i
= 0; i
< len
; i
++) {
153 vma
= find_vma(mm
, start
);
155 goto finish_or_fault
;
157 /* protect what we can, including chardevs */
158 if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
) ||
159 !(vm_flags
& vma
->vm_flags
))
160 goto finish_or_fault
;
163 pages
[i
] = virt_to_page(start
);
165 page_cache_get(pages
[i
]);
175 return i
? : -EFAULT
;
177 EXPORT_SYMBOL(get_user_pages
);
179 DEFINE_RWLOCK(vmlist_lock
);
180 struct vm_struct
*vmlist
;
182 void vfree(const void *addr
)
186 EXPORT_SYMBOL(vfree
);
188 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
191 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
192 * returns only a logical address.
194 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
196 EXPORT_SYMBOL(__vmalloc
);
198 void *vmalloc_user(unsigned long size
)
202 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
205 struct vm_area_struct
*vma
;
207 down_write(¤t
->mm
->mmap_sem
);
208 vma
= find_vma(current
->mm
, (unsigned long)ret
);
210 vma
->vm_flags
|= VM_USERMAP
;
211 up_write(¤t
->mm
->mmap_sem
);
216 EXPORT_SYMBOL(vmalloc_user
);
218 struct page
*vmalloc_to_page(const void *addr
)
220 return virt_to_page(addr
);
222 EXPORT_SYMBOL(vmalloc_to_page
);
224 unsigned long vmalloc_to_pfn(const void *addr
)
226 return page_to_pfn(virt_to_page(addr
));
228 EXPORT_SYMBOL(vmalloc_to_pfn
);
230 long vread(char *buf
, char *addr
, unsigned long count
)
232 memcpy(buf
, addr
, count
);
236 long vwrite(char *buf
, char *addr
, unsigned long count
)
238 /* Don't allow overflow */
239 if ((unsigned long) addr
+ count
< count
)
240 count
= -(unsigned long) addr
;
242 memcpy(addr
, buf
, count
);
247 * vmalloc - allocate virtually continguos memory
249 * @size: allocation size
251 * Allocate enough pages to cover @size from the page level
252 * allocator and map them into continguos kernel virtual space.
254 * For tight control over page level allocator and protection flags
255 * use __vmalloc() instead.
257 void *vmalloc(unsigned long size
)
259 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
261 EXPORT_SYMBOL(vmalloc
);
263 void *vmalloc_node(unsigned long size
, int node
)
265 return vmalloc(size
);
267 EXPORT_SYMBOL(vmalloc_node
);
270 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
271 * @size: allocation size
273 * Allocate enough 32bit PA addressable pages to cover @size from the
274 * page level allocator and map them into continguos kernel virtual space.
276 void *vmalloc_32(unsigned long size
)
278 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
280 EXPORT_SYMBOL(vmalloc_32
);
283 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
284 * @size: allocation size
286 * The resulting memory area is 32bit addressable and zeroed so it can be
287 * mapped to userspace without leaking data.
289 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
290 * remap_vmalloc_range() are permissible.
292 void *vmalloc_32_user(unsigned long size
)
295 * We'll have to sort out the ZONE_DMA bits for 64-bit,
296 * but for now this can simply use vmalloc_user() directly.
298 return vmalloc_user(size
);
300 EXPORT_SYMBOL(vmalloc_32_user
);
302 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
309 void vunmap(const void *addr
)
313 EXPORT_SYMBOL(vunmap
);
316 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
319 void __attribute__((weak
)) vmalloc_sync_all(void)
323 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
328 EXPORT_SYMBOL(vm_insert_page
);
331 * sys_brk() for the most part doesn't need the global kernel
332 * lock, except when an application is doing something nasty
333 * like trying to un-brk an area that has already been mapped
334 * to a regular file. in this case, the unmapping will need
335 * to invoke file system routines that need the global lock.
337 asmlinkage
unsigned long sys_brk(unsigned long brk
)
339 struct mm_struct
*mm
= current
->mm
;
341 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
348 * Always allow shrinking brk
350 if (brk
<= mm
->brk
) {
356 * Ok, looks good - let it rip.
358 return mm
->brk
= brk
;
362 static void show_process_blocks(void)
364 struct vm_list_struct
*vml
;
366 printk("Process blocks %d:", current
->pid
);
368 for (vml
= ¤t
->mm
->context
.vmlist
; vml
; vml
= vml
->next
) {
369 printk(" %p: %p", vml
, vml
->vma
);
371 printk(" (%d @%lx #%d)",
372 kobjsize((void *) vml
->vma
->vm_start
),
374 atomic_read(&vml
->vma
->vm_usage
));
375 printk(vml
->next
? " ->" : ".\n");
381 * add a VMA into a process's mm_struct in the appropriate place in the list
382 * - should be called with mm->mmap_sem held writelocked
384 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_list_struct
*vml
)
386 struct vm_list_struct
**ppv
;
388 for (ppv
= ¤t
->mm
->context
.vmlist
; *ppv
; ppv
= &(*ppv
)->next
)
389 if ((*ppv
)->vma
->vm_start
> vml
->vma
->vm_start
)
397 * look up the first VMA in which addr resides, NULL if none
398 * - should be called with mm->mmap_sem at least held readlocked
400 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
402 struct vm_list_struct
*loop
, *vml
;
404 /* search the vm_start ordered list */
406 for (loop
= mm
->context
.vmlist
; loop
; loop
= loop
->next
) {
407 if (loop
->vma
->vm_start
> addr
)
412 if (vml
&& vml
->vma
->vm_end
> addr
)
417 EXPORT_SYMBOL(find_vma
);
421 * - we don't extend stack VMAs under NOMMU conditions
423 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
425 return find_vma(mm
, addr
);
428 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
434 * look up the first VMA exactly that exactly matches addr
435 * - should be called with mm->mmap_sem at least held readlocked
437 static inline struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
440 struct vm_list_struct
*vml
;
442 /* search the vm_start ordered list */
443 for (vml
= mm
->context
.vmlist
; vml
; vml
= vml
->next
) {
444 if (vml
->vma
->vm_start
== addr
)
446 if (vml
->vma
->vm_start
> addr
)
454 * find a VMA in the global tree
456 static inline struct vm_area_struct
*find_nommu_vma(unsigned long start
)
458 struct vm_area_struct
*vma
;
459 struct rb_node
*n
= nommu_vma_tree
.rb_node
;
462 vma
= rb_entry(n
, struct vm_area_struct
, vm_rb
);
464 if (start
< vma
->vm_start
)
466 else if (start
> vma
->vm_start
)
476 * add a VMA in the global tree
478 static void add_nommu_vma(struct vm_area_struct
*vma
)
480 struct vm_area_struct
*pvma
;
481 struct address_space
*mapping
;
482 struct rb_node
**p
= &nommu_vma_tree
.rb_node
;
483 struct rb_node
*parent
= NULL
;
485 /* add the VMA to the mapping */
487 mapping
= vma
->vm_file
->f_mapping
;
489 flush_dcache_mmap_lock(mapping
);
490 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
491 flush_dcache_mmap_unlock(mapping
);
494 /* add the VMA to the master list */
497 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
499 if (vma
->vm_start
< pvma
->vm_start
) {
502 else if (vma
->vm_start
> pvma
->vm_start
) {
506 /* mappings are at the same address - this can only
507 * happen for shared-mem chardevs and shared file
508 * mappings backed by ramfs/tmpfs */
509 BUG_ON(!(pvma
->vm_flags
& VM_SHARED
));
520 rb_link_node(&vma
->vm_rb
, parent
, p
);
521 rb_insert_color(&vma
->vm_rb
, &nommu_vma_tree
);
525 * delete a VMA from the global list
527 static void delete_nommu_vma(struct vm_area_struct
*vma
)
529 struct address_space
*mapping
;
531 /* remove the VMA from the mapping */
533 mapping
= vma
->vm_file
->f_mapping
;
535 flush_dcache_mmap_lock(mapping
);
536 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
537 flush_dcache_mmap_unlock(mapping
);
540 /* remove from the master list */
541 rb_erase(&vma
->vm_rb
, &nommu_vma_tree
);
545 * determine whether a mapping should be permitted and, if so, what sort of
546 * mapping we're capable of supporting
548 static int validate_mmap_request(struct file
*file
,
554 unsigned long *_capabilities
)
556 unsigned long capabilities
;
557 unsigned long reqprot
= prot
;
560 /* do the simple checks first */
561 if (flags
& MAP_FIXED
|| addr
) {
563 "%d: Can't do fixed-address/overlay mmap of RAM\n",
568 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
569 (flags
& MAP_TYPE
) != MAP_SHARED
)
575 /* Careful about overflows.. */
576 len
= PAGE_ALIGN(len
);
577 if (!len
|| len
> TASK_SIZE
)
580 /* offset overflow? */
581 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
585 /* validate file mapping requests */
586 struct address_space
*mapping
;
588 /* files must support mmap */
589 if (!file
->f_op
|| !file
->f_op
->mmap
)
592 /* work out if what we've got could possibly be shared
593 * - we support chardevs that provide their own "memory"
594 * - we support files/blockdevs that are memory backed
596 mapping
= file
->f_mapping
;
598 mapping
= file
->f_path
.dentry
->d_inode
->i_mapping
;
601 if (mapping
&& mapping
->backing_dev_info
)
602 capabilities
= mapping
->backing_dev_info
->capabilities
;
605 /* no explicit capabilities set, so assume some
607 switch (file
->f_path
.dentry
->d_inode
->i_mode
& S_IFMT
) {
610 capabilities
= BDI_CAP_MAP_COPY
;
625 /* eliminate any capabilities that we can't support on this
627 if (!file
->f_op
->get_unmapped_area
)
628 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
629 if (!file
->f_op
->read
)
630 capabilities
&= ~BDI_CAP_MAP_COPY
;
632 if (flags
& MAP_SHARED
) {
633 /* do checks for writing, appending and locking */
634 if ((prot
& PROT_WRITE
) &&
635 !(file
->f_mode
& FMODE_WRITE
))
638 if (IS_APPEND(file
->f_path
.dentry
->d_inode
) &&
639 (file
->f_mode
& FMODE_WRITE
))
642 if (locks_verify_locked(file
->f_path
.dentry
->d_inode
))
645 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
648 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
649 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
650 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
652 printk("MAP_SHARED not completely supported on !MMU\n");
656 /* we mustn't privatise shared mappings */
657 capabilities
&= ~BDI_CAP_MAP_COPY
;
660 /* we're going to read the file into private memory we
662 if (!(capabilities
& BDI_CAP_MAP_COPY
))
665 /* we don't permit a private writable mapping to be
666 * shared with the backing device */
667 if (prot
& PROT_WRITE
)
668 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
671 /* handle executable mappings and implied executable
673 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
674 if (prot
& PROT_EXEC
)
677 else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
678 /* handle implication of PROT_EXEC by PROT_READ */
679 if (current
->personality
& READ_IMPLIES_EXEC
) {
680 if (capabilities
& BDI_CAP_EXEC_MAP
)
684 else if ((prot
& PROT_READ
) &&
685 (prot
& PROT_EXEC
) &&
686 !(capabilities
& BDI_CAP_EXEC_MAP
)
688 /* backing file is not executable, try to copy */
689 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
693 /* anonymous mappings are always memory backed and can be
696 capabilities
= BDI_CAP_MAP_COPY
;
698 /* handle PROT_EXEC implication by PROT_READ */
699 if ((prot
& PROT_READ
) &&
700 (current
->personality
& READ_IMPLIES_EXEC
))
704 /* allow the security API to have its say */
705 ret
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
710 *_capabilities
= capabilities
;
715 * we've determined that we can make the mapping, now translate what we
716 * now know into VMA flags
718 static unsigned long determine_vm_flags(struct file
*file
,
721 unsigned long capabilities
)
723 unsigned long vm_flags
;
725 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
726 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
727 /* vm_flags |= mm->def_flags; */
729 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
730 /* attempt to share read-only copies of mapped file chunks */
731 if (file
&& !(prot
& PROT_WRITE
))
732 vm_flags
|= VM_MAYSHARE
;
735 /* overlay a shareable mapping on the backing device or inode
736 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
738 if (flags
& MAP_SHARED
)
739 vm_flags
|= VM_MAYSHARE
| VM_SHARED
;
740 else if ((((vm_flags
& capabilities
) ^ vm_flags
) & BDI_CAP_VMFLAGS
) == 0)
741 vm_flags
|= VM_MAYSHARE
;
744 /* refuse to let anyone share private mappings with this process if
745 * it's being traced - otherwise breakpoints set in it may interfere
746 * with another untraced process
748 if ((flags
& MAP_PRIVATE
) && tracehook_expect_breakpoints(current
))
749 vm_flags
&= ~VM_MAYSHARE
;
755 * set up a shared mapping on a file
757 static int do_mmap_shared_file(struct vm_area_struct
*vma
, unsigned long len
)
761 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
765 /* getting an ENOSYS error indicates that direct mmap isn't
766 * possible (as opposed to tried but failed) so we'll fall
767 * through to making a private copy of the data and mapping
773 * set up a private mapping or an anonymous shared mapping
775 static int do_mmap_private(struct vm_area_struct
*vma
, unsigned long len
)
780 /* invoke the file's mapping function so that it can keep track of
781 * shared mappings on devices or memory
782 * - VM_MAYSHARE will be set if it may attempt to share
785 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
786 if (ret
!= -ENOSYS
) {
787 /* shouldn't return success if we're not sharing */
788 BUG_ON(ret
== 0 && !(vma
->vm_flags
& VM_MAYSHARE
));
789 return ret
; /* success or a real error */
792 /* getting an ENOSYS error indicates that direct mmap isn't
793 * possible (as opposed to tried but failed) so we'll try to
794 * make a private copy of the data and map that instead */
797 /* allocate some memory to hold the mapping
798 * - note that this may not return a page-aligned address if the object
799 * we're allocating is smaller than a page
801 base
= kmalloc(len
, GFP_KERNEL
|__GFP_COMP
);
805 vma
->vm_start
= (unsigned long) base
;
806 vma
->vm_end
= vma
->vm_start
+ len
;
807 vma
->vm_flags
|= VM_MAPPED_COPY
;
810 if (len
+ WARN_ON_SLACK
<= kobjsize(result
))
811 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n",
812 len
, current
->pid
, kobjsize(result
) - len
);
816 /* read the contents of a file into the copy */
820 fpos
= vma
->vm_pgoff
;
825 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, len
, &fpos
);
831 /* clear the last little bit */
833 memset(base
+ ret
, 0, len
- ret
);
836 /* if it's an anonymous mapping, then just clear it */
837 memset(base
, 0, len
);
848 printk("Allocation of length %lu from process %d failed\n",
855 * handle mapping creation for uClinux
857 unsigned long do_mmap_pgoff(struct file
*file
,
864 struct vm_list_struct
*vml
= NULL
;
865 struct vm_area_struct
*vma
= NULL
;
867 unsigned long capabilities
, vm_flags
;
871 if (!(flags
& MAP_FIXED
))
872 addr
= round_hint_to_min(addr
);
874 /* decide whether we should attempt the mapping, and if so what sort of
876 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
881 /* we've determined that we can make the mapping, now translate what we
882 * now know into VMA flags */
883 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
885 /* we're going to need to record the mapping if it works */
886 vml
= kzalloc(sizeof(struct vm_list_struct
), GFP_KERNEL
);
888 goto error_getting_vml
;
890 down_write(&nommu_vma_sem
);
892 /* if we want to share, we need to check for VMAs created by other
893 * mmap() calls that overlap with our proposed mapping
894 * - we can only share with an exact match on most regular files
895 * - shared mappings on character devices and memory backed files are
896 * permitted to overlap inexactly as far as we are concerned for in
897 * these cases, sharing is handled in the driver or filesystem rather
900 if (vm_flags
& VM_MAYSHARE
) {
901 unsigned long pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
902 unsigned long vmpglen
;
904 /* suppress VMA sharing for shared regions */
905 if (vm_flags
& VM_SHARED
&&
906 capabilities
& BDI_CAP_MAP_DIRECT
)
907 goto dont_share_VMAs
;
909 for (rb
= rb_first(&nommu_vma_tree
); rb
; rb
= rb_next(rb
)) {
910 vma
= rb_entry(rb
, struct vm_area_struct
, vm_rb
);
912 if (!(vma
->vm_flags
& VM_MAYSHARE
))
915 /* search for overlapping mappings on the same file */
916 if (vma
->vm_file
->f_path
.dentry
->d_inode
!= file
->f_path
.dentry
->d_inode
)
919 if (vma
->vm_pgoff
>= pgoff
+ pglen
)
922 vmpglen
= vma
->vm_end
- vma
->vm_start
+ PAGE_SIZE
- 1;
923 vmpglen
>>= PAGE_SHIFT
;
924 if (pgoff
>= vma
->vm_pgoff
+ vmpglen
)
927 /* handle inexactly overlapping matches between mappings */
928 if (vma
->vm_pgoff
!= pgoff
|| vmpglen
!= pglen
) {
929 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
930 goto sharing_violation
;
934 /* we've found a VMA we can share */
935 atomic_inc(&vma
->vm_usage
);
938 result
= (void *) vma
->vm_start
;
945 /* obtain the address at which to make a shared mapping
946 * - this is the hook for quasi-memory character devices to
947 * tell us the location of a shared mapping
949 if (file
&& file
->f_op
->get_unmapped_area
) {
950 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
952 if (IS_ERR((void *) addr
)) {
954 if (ret
!= (unsigned long) -ENOSYS
)
957 /* the driver refused to tell us where to site
958 * the mapping so we'll have to attempt to copy
960 ret
= (unsigned long) -ENODEV
;
961 if (!(capabilities
& BDI_CAP_MAP_COPY
))
964 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
969 /* we're going to need a VMA struct as well */
970 vma
= kzalloc(sizeof(struct vm_area_struct
), GFP_KERNEL
);
972 goto error_getting_vma
;
974 INIT_LIST_HEAD(&vma
->anon_vma_node
);
975 atomic_set(&vma
->vm_usage
, 1);
978 if (vm_flags
& VM_EXECUTABLE
) {
979 added_exe_file_vma(current
->mm
);
980 vma
->vm_mm
= current
->mm
;
984 vma
->vm_flags
= vm_flags
;
985 vma
->vm_start
= addr
;
986 vma
->vm_end
= addr
+ len
;
987 vma
->vm_pgoff
= pgoff
;
991 /* set up the mapping */
992 if (file
&& vma
->vm_flags
& VM_SHARED
)
993 ret
= do_mmap_shared_file(vma
, len
);
995 ret
= do_mmap_private(vma
, len
);
999 /* okay... we have a mapping; now we have to register it */
1000 result
= (void *) vma
->vm_start
;
1002 if (vma
->vm_flags
& VM_MAPPED_COPY
) {
1003 realalloc
+= kobjsize(result
);
1007 realalloc
+= kobjsize(vma
);
1008 askedalloc
+= sizeof(*vma
);
1010 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1015 realalloc
+= kobjsize(vml
);
1016 askedalloc
+= sizeof(*vml
);
1018 add_vma_to_mm(current
->mm
, vml
);
1020 up_write(&nommu_vma_sem
);
1022 if (prot
& PROT_EXEC
)
1023 flush_icache_range((unsigned long) result
,
1024 (unsigned long) result
+ len
);
1027 printk("do_mmap:\n");
1028 show_process_blocks();
1031 return (unsigned long) result
;
1034 up_write(&nommu_vma_sem
);
1039 if (vma
->vm_flags
& VM_EXECUTABLE
)
1040 removed_exe_file_vma(vma
->vm_mm
);
1047 up_write(&nommu_vma_sem
);
1048 printk("Attempt to share mismatched mappings\n");
1053 up_write(&nommu_vma_sem
);
1055 printk("Allocation of vma for %lu byte allocation from process %d failed\n",
1061 printk("Allocation of vml for %lu byte allocation from process %d failed\n",
1066 EXPORT_SYMBOL(do_mmap_pgoff
);
1069 * handle mapping disposal for uClinux
1071 static void put_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1074 down_write(&nommu_vma_sem
);
1076 if (atomic_dec_and_test(&vma
->vm_usage
)) {
1077 delete_nommu_vma(vma
);
1079 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
1080 vma
->vm_ops
->close(vma
);
1082 /* IO memory and memory shared directly out of the pagecache from
1083 * ramfs/tmpfs mustn't be released here */
1084 if (vma
->vm_flags
& VM_MAPPED_COPY
) {
1085 realalloc
-= kobjsize((void *) vma
->vm_start
);
1086 askedalloc
-= vma
->vm_end
- vma
->vm_start
;
1087 kfree((void *) vma
->vm_start
);
1090 realalloc
-= kobjsize(vma
);
1091 askedalloc
-= sizeof(*vma
);
1095 if (vma
->vm_flags
& VM_EXECUTABLE
)
1096 removed_exe_file_vma(mm
);
1101 up_write(&nommu_vma_sem
);
1107 * - under NOMMU conditions the parameters must match exactly to the mapping to
1110 int do_munmap(struct mm_struct
*mm
, unsigned long addr
, size_t len
)
1112 struct vm_list_struct
*vml
, **parent
;
1113 unsigned long end
= addr
+ len
;
1116 printk("do_munmap:\n");
1119 for (parent
= &mm
->context
.vmlist
; *parent
; parent
= &(*parent
)->next
) {
1120 if ((*parent
)->vma
->vm_start
> addr
)
1122 if ((*parent
)->vma
->vm_start
== addr
&&
1123 ((len
== 0) || ((*parent
)->vma
->vm_end
== end
)))
1127 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
1128 current
->pid
, current
->comm
, (void *) addr
);
1134 put_vma(mm
, vml
->vma
);
1136 *parent
= vml
->next
;
1137 realalloc
-= kobjsize(vml
);
1138 askedalloc
-= sizeof(*vml
);
1141 update_hiwater_vm(mm
);
1142 mm
->total_vm
-= len
>> PAGE_SHIFT
;
1145 show_process_blocks();
1150 EXPORT_SYMBOL(do_munmap
);
1152 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
1155 struct mm_struct
*mm
= current
->mm
;
1157 down_write(&mm
->mmap_sem
);
1158 ret
= do_munmap(mm
, addr
, len
);
1159 up_write(&mm
->mmap_sem
);
1164 * Release all mappings
1166 void exit_mmap(struct mm_struct
* mm
)
1168 struct vm_list_struct
*tmp
;
1172 printk("Exit_mmap:\n");
1177 while ((tmp
= mm
->context
.vmlist
)) {
1178 mm
->context
.vmlist
= tmp
->next
;
1179 put_vma(mm
, tmp
->vma
);
1181 realalloc
-= kobjsize(tmp
);
1182 askedalloc
-= sizeof(*tmp
);
1187 show_process_blocks();
1192 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1198 * expand (or shrink) an existing mapping, potentially moving it at the same
1199 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1201 * under NOMMU conditions, we only permit changing a mapping's size, and only
1202 * as long as it stays within the hole allocated by the kmalloc() call in
1203 * do_mmap_pgoff() and the block is not shareable
1205 * MREMAP_FIXED is not supported under NOMMU conditions
1207 unsigned long do_mremap(unsigned long addr
,
1208 unsigned long old_len
, unsigned long new_len
,
1209 unsigned long flags
, unsigned long new_addr
)
1211 struct vm_area_struct
*vma
;
1213 /* insanity checks first */
1215 return (unsigned long) -EINVAL
;
1217 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1218 return (unsigned long) -EINVAL
;
1220 vma
= find_vma_exact(current
->mm
, addr
);
1222 return (unsigned long) -EINVAL
;
1224 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1225 return (unsigned long) -EFAULT
;
1227 if (vma
->vm_flags
& VM_MAYSHARE
)
1228 return (unsigned long) -EPERM
;
1230 if (new_len
> kobjsize((void *) addr
))
1231 return (unsigned long) -ENOMEM
;
1233 /* all checks complete - do it */
1234 vma
->vm_end
= vma
->vm_start
+ new_len
;
1236 askedalloc
-= old_len
;
1237 askedalloc
+= new_len
;
1239 return vma
->vm_start
;
1241 EXPORT_SYMBOL(do_mremap
);
1243 asmlinkage
unsigned long sys_mremap(unsigned long addr
,
1244 unsigned long old_len
, unsigned long new_len
,
1245 unsigned long flags
, unsigned long new_addr
)
1249 down_write(¤t
->mm
->mmap_sem
);
1250 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1251 up_write(¤t
->mm
->mmap_sem
);
1255 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1256 unsigned int foll_flags
)
1261 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long from
,
1262 unsigned long to
, unsigned long size
, pgprot_t prot
)
1264 vma
->vm_start
= vma
->vm_pgoff
<< PAGE_SHIFT
;
1267 EXPORT_SYMBOL(remap_pfn_range
);
1269 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1270 unsigned long pgoff
)
1272 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1274 if (!(vma
->vm_flags
& VM_USERMAP
))
1277 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1278 vma
->vm_end
= vma
->vm_start
+ size
;
1282 EXPORT_SYMBOL(remap_vmalloc_range
);
1284 void swap_unplug_io_fn(struct backing_dev_info
*bdi
, struct page
*page
)
1288 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1289 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1294 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1298 void unmap_mapping_range(struct address_space
*mapping
,
1299 loff_t
const holebegin
, loff_t
const holelen
,
1303 EXPORT_SYMBOL(unmap_mapping_range
);
1306 * ask for an unmapped area at which to create a mapping on a file
1308 unsigned long get_unmapped_area(struct file
*file
, unsigned long addr
,
1309 unsigned long len
, unsigned long pgoff
,
1310 unsigned long flags
)
1312 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long,
1313 unsigned long, unsigned long);
1315 get_area
= current
->mm
->get_unmapped_area
;
1316 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1317 get_area
= file
->f_op
->get_unmapped_area
;
1322 return get_area(file
, addr
, len
, pgoff
, flags
);
1324 EXPORT_SYMBOL(get_unmapped_area
);
1327 * Check that a process has enough memory to allocate a new virtual
1328 * mapping. 0 means there is enough memory for the allocation to
1329 * succeed and -ENOMEM implies there is not.
1331 * We currently support three overcommit policies, which are set via the
1332 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1334 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1335 * Additional code 2002 Jul 20 by Robert Love.
1337 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1339 * Note this is a helper function intended to be used by LSMs which
1340 * wish to use this logic.
1342 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1344 unsigned long free
, allowed
;
1346 vm_acct_memory(pages
);
1349 * Sometimes we want to use more memory than we have
1351 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1354 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1357 free
= global_page_state(NR_FILE_PAGES
);
1358 free
+= nr_swap_pages
;
1361 * Any slabs which are created with the
1362 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1363 * which are reclaimable, under pressure. The dentry
1364 * cache and most inode caches should fall into this
1366 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1369 * Leave the last 3% for root
1378 * nr_free_pages() is very expensive on large systems,
1379 * only call if we're about to fail.
1381 n
= nr_free_pages();
1384 * Leave reserved pages. The pages are not for anonymous pages.
1386 if (n
<= totalreserve_pages
)
1389 n
-= totalreserve_pages
;
1392 * Leave the last 3% for root
1404 allowed
= totalram_pages
* sysctl_overcommit_ratio
/ 100;
1406 * Leave the last 3% for root
1409 allowed
-= allowed
/ 32;
1410 allowed
+= total_swap_pages
;
1412 /* Don't let a single process grow too big:
1413 leave 3% of the size of this process for other processes */
1414 allowed
-= current
->mm
->total_vm
/ 32;
1417 * cast `allowed' as a signed long because vm_committed_space
1418 * sometimes has a negative value
1420 if (atomic_long_read(&vm_committed_space
) < (long)allowed
)
1423 vm_unacct_memory(pages
);
1428 int in_gate_area_no_task(unsigned long addr
)
1433 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1438 EXPORT_SYMBOL(filemap_fault
);
1441 * Access another process' address space.
1442 * - source/target buffer must be kernel space
1444 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
1446 struct vm_area_struct
*vma
;
1447 struct mm_struct
*mm
;
1449 if (addr
+ len
< addr
)
1452 mm
= get_task_mm(tsk
);
1456 down_read(&mm
->mmap_sem
);
1458 /* the access must start within one of the target process's mappings */
1459 vma
= find_vma(mm
, addr
);
1461 /* don't overrun this mapping */
1462 if (addr
+ len
>= vma
->vm_end
)
1463 len
= vma
->vm_end
- addr
;
1465 /* only read or write mappings where it is permitted */
1466 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1467 len
-= copy_to_user((void *) addr
, buf
, len
);
1468 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1469 len
-= copy_from_user(buf
, (void *) addr
, len
);
1476 up_read(&mm
->mmap_sem
);