4 * Replacement code for mm functions to support CPU's that don't
5 * have any form of memory management unit (thus no virtual memory).
7 * See Documentation/nommu-mmap.txt
9 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
10 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
11 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
12 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
13 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
16 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
18 #include <linux/export.h>
20 #include <linux/sched/mm.h>
21 #include <linux/vmacache.h>
22 #include <linux/mman.h>
23 #include <linux/swap.h>
24 #include <linux/file.h>
25 #include <linux/highmem.h>
26 #include <linux/pagemap.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/compiler.h>
32 #include <linux/mount.h>
33 #include <linux/personality.h>
34 #include <linux/security.h>
35 #include <linux/syscalls.h>
36 #include <linux/audit.h>
37 #include <linux/printk.h>
39 #include <linux/uaccess.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
46 EXPORT_SYMBOL(high_memory
);
48 unsigned long max_mapnr
;
49 EXPORT_SYMBOL(max_mapnr
);
50 unsigned long highest_memmap_pfn
;
51 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
52 int heap_stack_gap
= 0;
54 atomic_long_t mmap_pages_allocated
;
56 EXPORT_SYMBOL(mem_map
);
58 /* list of mapped, potentially shareable regions */
59 static struct kmem_cache
*vm_region_jar
;
60 struct rb_root nommu_region_tree
= RB_ROOT
;
61 DECLARE_RWSEM(nommu_region_sem
);
63 const struct vm_operations_struct generic_file_vm_ops
= {
67 * Return the total memory allocated for this pointer, not
68 * just what the caller asked for.
70 * Doesn't have to be accurate, i.e. may have races.
72 unsigned int kobjsize(const void *objp
)
77 * If the object we have should not have ksize performed on it,
80 if (!objp
|| !virt_addr_valid(objp
))
83 page
= virt_to_head_page(objp
);
86 * If the allocator sets PageSlab, we know the pointer came from
93 * If it's not a compound page, see if we have a matching VMA
94 * region. This test is intentionally done in reverse order,
95 * so if there's no VMA, we still fall through and hand back
96 * PAGE_SIZE for 0-order pages.
98 if (!PageCompound(page
)) {
99 struct vm_area_struct
*vma
;
101 vma
= find_vma(current
->mm
, (unsigned long)objp
);
103 return vma
->vm_end
- vma
->vm_start
;
107 * The ksize() function is only guaranteed to work for pointers
108 * returned by kmalloc(). So handle arbitrary pointers here.
110 return PAGE_SIZE
<< compound_order(page
);
113 static long __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
114 unsigned long start
, unsigned long nr_pages
,
115 unsigned int foll_flags
, struct page
**pages
,
116 struct vm_area_struct
**vmas
, int *nonblocking
)
118 struct vm_area_struct
*vma
;
119 unsigned long vm_flags
;
122 /* calculate required read or write permissions.
123 * If FOLL_FORCE is set, we only require the "MAY" flags.
125 vm_flags
= (foll_flags
& FOLL_WRITE
) ?
126 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
127 vm_flags
&= (foll_flags
& FOLL_FORCE
) ?
128 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
130 for (i
= 0; i
< nr_pages
; i
++) {
131 vma
= find_vma(mm
, start
);
133 goto finish_or_fault
;
135 /* protect what we can, including chardevs */
136 if ((vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
137 !(vm_flags
& vma
->vm_flags
))
138 goto finish_or_fault
;
141 pages
[i
] = virt_to_page(start
);
147 start
= (start
+ PAGE_SIZE
) & PAGE_MASK
;
153 return i
? : -EFAULT
;
157 * get a list of pages in an address range belonging to the specified process
158 * and indicate the VMA that covers each page
159 * - this is potentially dodgy as we may end incrementing the page count of a
160 * slab page or a secondary page from a compound page
161 * - don't permit access to VMAs that don't support it, such as I/O mappings
163 long get_user_pages(unsigned long start
, unsigned long nr_pages
,
164 unsigned int gup_flags
, struct page
**pages
,
165 struct vm_area_struct
**vmas
)
167 return __get_user_pages(current
, current
->mm
, start
, nr_pages
,
168 gup_flags
, pages
, vmas
, NULL
);
170 EXPORT_SYMBOL(get_user_pages
);
172 long get_user_pages_locked(unsigned long start
, unsigned long nr_pages
,
173 unsigned int gup_flags
, struct page
**pages
,
176 return get_user_pages(start
, nr_pages
, gup_flags
, pages
, NULL
);
178 EXPORT_SYMBOL(get_user_pages_locked
);
180 static long __get_user_pages_unlocked(struct task_struct
*tsk
,
181 struct mm_struct
*mm
, unsigned long start
,
182 unsigned long nr_pages
, struct page
**pages
,
183 unsigned int gup_flags
)
186 down_read(&mm
->mmap_sem
);
187 ret
= __get_user_pages(tsk
, mm
, start
, nr_pages
, gup_flags
, pages
,
189 up_read(&mm
->mmap_sem
);
193 long get_user_pages_unlocked(unsigned long start
, unsigned long nr_pages
,
194 struct page
**pages
, unsigned int gup_flags
)
196 return __get_user_pages_unlocked(current
, current
->mm
, start
, nr_pages
,
199 EXPORT_SYMBOL(get_user_pages_unlocked
);
202 * follow_pfn - look up PFN at a user virtual address
203 * @vma: memory mapping
204 * @address: user virtual address
205 * @pfn: location to store found PFN
207 * Only IO mappings and raw PFN mappings are allowed.
209 * Returns zero and the pfn at @pfn on success, -ve otherwise.
211 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
214 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
217 *pfn
= address
>> PAGE_SHIFT
;
220 EXPORT_SYMBOL(follow_pfn
);
222 LIST_HEAD(vmap_area_list
);
224 void vfree(const void *addr
)
228 EXPORT_SYMBOL(vfree
);
230 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
233 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
234 * returns only a logical address.
236 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
238 EXPORT_SYMBOL(__vmalloc
);
240 void *vmalloc_user(unsigned long size
)
244 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
247 struct vm_area_struct
*vma
;
249 down_write(¤t
->mm
->mmap_sem
);
250 vma
= find_vma(current
->mm
, (unsigned long)ret
);
252 vma
->vm_flags
|= VM_USERMAP
;
253 up_write(¤t
->mm
->mmap_sem
);
258 EXPORT_SYMBOL(vmalloc_user
);
260 struct page
*vmalloc_to_page(const void *addr
)
262 return virt_to_page(addr
);
264 EXPORT_SYMBOL(vmalloc_to_page
);
266 unsigned long vmalloc_to_pfn(const void *addr
)
268 return page_to_pfn(virt_to_page(addr
));
270 EXPORT_SYMBOL(vmalloc_to_pfn
);
272 long vread(char *buf
, char *addr
, unsigned long count
)
274 /* Don't allow overflow */
275 if ((unsigned long) buf
+ count
< count
)
276 count
= -(unsigned long) buf
;
278 memcpy(buf
, addr
, count
);
282 long vwrite(char *buf
, char *addr
, unsigned long count
)
284 /* Don't allow overflow */
285 if ((unsigned long) addr
+ count
< count
)
286 count
= -(unsigned long) addr
;
288 memcpy(addr
, buf
, count
);
293 * vmalloc - allocate virtually contiguous memory
295 * @size: allocation size
297 * Allocate enough pages to cover @size from the page level
298 * allocator and map them into contiguous kernel virtual space.
300 * For tight control over page level allocator and protection flags
301 * use __vmalloc() instead.
303 void *vmalloc(unsigned long size
)
305 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
307 EXPORT_SYMBOL(vmalloc
);
310 * vzalloc - allocate virtually contiguous memory with zero fill
312 * @size: allocation size
314 * Allocate enough pages to cover @size from the page level
315 * allocator and map them into contiguous kernel virtual space.
316 * The memory allocated is set to zero.
318 * For tight control over page level allocator and protection flags
319 * use __vmalloc() instead.
321 void *vzalloc(unsigned long size
)
323 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
326 EXPORT_SYMBOL(vzalloc
);
329 * vmalloc_node - allocate memory on a specific node
330 * @size: allocation size
333 * Allocate enough pages to cover @size from the page level
334 * allocator and map them into contiguous kernel virtual space.
336 * For tight control over page level allocator and protection flags
337 * use __vmalloc() instead.
339 void *vmalloc_node(unsigned long size
, int node
)
341 return vmalloc(size
);
343 EXPORT_SYMBOL(vmalloc_node
);
346 * vzalloc_node - allocate memory on a specific node with zero fill
347 * @size: allocation size
350 * Allocate enough pages to cover @size from the page level
351 * allocator and map them into contiguous kernel virtual space.
352 * The memory allocated is set to zero.
354 * For tight control over page level allocator and protection flags
355 * use __vmalloc() instead.
357 void *vzalloc_node(unsigned long size
, int node
)
359 return vzalloc(size
);
361 EXPORT_SYMBOL(vzalloc_node
);
363 #ifndef PAGE_KERNEL_EXEC
364 # define PAGE_KERNEL_EXEC PAGE_KERNEL
368 * vmalloc_exec - allocate virtually contiguous, executable memory
369 * @size: allocation size
371 * Kernel-internal function to allocate enough pages to cover @size
372 * the page level allocator and map them into contiguous and
373 * executable kernel virtual space.
375 * For tight control over page level allocator and protection flags
376 * use __vmalloc() instead.
379 void *vmalloc_exec(unsigned long size
)
381 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
385 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
386 * @size: allocation size
388 * Allocate enough 32bit PA addressable pages to cover @size from the
389 * page level allocator and map them into contiguous kernel virtual space.
391 void *vmalloc_32(unsigned long size
)
393 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
395 EXPORT_SYMBOL(vmalloc_32
);
398 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
399 * @size: allocation size
401 * The resulting memory area is 32bit addressable and zeroed so it can be
402 * mapped to userspace without leaking data.
404 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
405 * remap_vmalloc_range() are permissible.
407 void *vmalloc_32_user(unsigned long size
)
410 * We'll have to sort out the ZONE_DMA bits for 64-bit,
411 * but for now this can simply use vmalloc_user() directly.
413 return vmalloc_user(size
);
415 EXPORT_SYMBOL(vmalloc_32_user
);
417 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
424 void vunmap(const void *addr
)
428 EXPORT_SYMBOL(vunmap
);
430 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
435 EXPORT_SYMBOL(vm_map_ram
);
437 void vm_unmap_ram(const void *mem
, unsigned int count
)
441 EXPORT_SYMBOL(vm_unmap_ram
);
443 void vm_unmap_aliases(void)
446 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
449 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
452 void __weak
vmalloc_sync_all(void)
457 * alloc_vm_area - allocate a range of kernel address space
458 * @size: size of the area
460 * Returns: NULL on failure, vm_struct on success
462 * This function reserves a range of kernel address space, and
463 * allocates pagetables to map that range. No actual mappings
464 * are created. If the kernel address space is not shared
465 * between processes, it syncs the pagetable across all
468 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
473 EXPORT_SYMBOL_GPL(alloc_vm_area
);
475 void free_vm_area(struct vm_struct
*area
)
479 EXPORT_SYMBOL_GPL(free_vm_area
);
481 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
486 EXPORT_SYMBOL(vm_insert_page
);
489 * sys_brk() for the most part doesn't need the global kernel
490 * lock, except when an application is doing something nasty
491 * like trying to un-brk an area that has already been mapped
492 * to a regular file. in this case, the unmapping will need
493 * to invoke file system routines that need the global lock.
495 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
497 struct mm_struct
*mm
= current
->mm
;
499 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
506 * Always allow shrinking brk
508 if (brk
<= mm
->brk
) {
514 * Ok, looks good - let it rip.
516 flush_icache_range(mm
->brk
, brk
);
517 return mm
->brk
= brk
;
521 * initialise the percpu counter for VM and region record slabs
523 void __init
mmap_init(void)
527 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
529 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
533 * validate the region tree
534 * - the caller must hold the region lock
536 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
537 static noinline
void validate_nommu_regions(void)
539 struct vm_region
*region
, *last
;
540 struct rb_node
*p
, *lastp
;
542 lastp
= rb_first(&nommu_region_tree
);
546 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
547 BUG_ON(last
->vm_end
<= last
->vm_start
);
548 BUG_ON(last
->vm_top
< last
->vm_end
);
550 while ((p
= rb_next(lastp
))) {
551 region
= rb_entry(p
, struct vm_region
, vm_rb
);
552 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
554 BUG_ON(region
->vm_end
<= region
->vm_start
);
555 BUG_ON(region
->vm_top
< region
->vm_end
);
556 BUG_ON(region
->vm_start
< last
->vm_top
);
562 static void validate_nommu_regions(void)
568 * add a region into the global tree
570 static void add_nommu_region(struct vm_region
*region
)
572 struct vm_region
*pregion
;
573 struct rb_node
**p
, *parent
;
575 validate_nommu_regions();
578 p
= &nommu_region_tree
.rb_node
;
581 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
582 if (region
->vm_start
< pregion
->vm_start
)
584 else if (region
->vm_start
> pregion
->vm_start
)
586 else if (pregion
== region
)
592 rb_link_node(®ion
->vm_rb
, parent
, p
);
593 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
595 validate_nommu_regions();
599 * delete a region from the global tree
601 static void delete_nommu_region(struct vm_region
*region
)
603 BUG_ON(!nommu_region_tree
.rb_node
);
605 validate_nommu_regions();
606 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
607 validate_nommu_regions();
611 * free a contiguous series of pages
613 static void free_page_series(unsigned long from
, unsigned long to
)
615 for (; from
< to
; from
+= PAGE_SIZE
) {
616 struct page
*page
= virt_to_page(from
);
618 atomic_long_dec(&mmap_pages_allocated
);
624 * release a reference to a region
625 * - the caller must hold the region semaphore for writing, which this releases
626 * - the region may not have been added to the tree yet, in which case vm_top
627 * will equal vm_start
629 static void __put_nommu_region(struct vm_region
*region
)
630 __releases(nommu_region_sem
)
632 BUG_ON(!nommu_region_tree
.rb_node
);
634 if (--region
->vm_usage
== 0) {
635 if (region
->vm_top
> region
->vm_start
)
636 delete_nommu_region(region
);
637 up_write(&nommu_region_sem
);
640 fput(region
->vm_file
);
642 /* IO memory and memory shared directly out of the pagecache
643 * from ramfs/tmpfs mustn't be released here */
644 if (region
->vm_flags
& VM_MAPPED_COPY
)
645 free_page_series(region
->vm_start
, region
->vm_top
);
646 kmem_cache_free(vm_region_jar
, region
);
648 up_write(&nommu_region_sem
);
653 * release a reference to a region
655 static void put_nommu_region(struct vm_region
*region
)
657 down_write(&nommu_region_sem
);
658 __put_nommu_region(region
);
662 * update protection on a vma
664 static void protect_vma(struct vm_area_struct
*vma
, unsigned long flags
)
667 struct mm_struct
*mm
= vma
->vm_mm
;
668 long start
= vma
->vm_start
& PAGE_MASK
;
669 while (start
< vma
->vm_end
) {
670 protect_page(mm
, start
, flags
);
673 update_protections(mm
);
678 * add a VMA into a process's mm_struct in the appropriate place in the list
679 * and tree and add to the address space's page tree also if not an anonymous
681 * - should be called with mm->mmap_sem held writelocked
683 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
685 struct vm_area_struct
*pvma
, *prev
;
686 struct address_space
*mapping
;
687 struct rb_node
**p
, *parent
, *rb_prev
;
689 BUG_ON(!vma
->vm_region
);
694 protect_vma(vma
, vma
->vm_flags
);
696 /* add the VMA to the mapping */
698 mapping
= vma
->vm_file
->f_mapping
;
700 i_mmap_lock_write(mapping
);
701 flush_dcache_mmap_lock(mapping
);
702 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
703 flush_dcache_mmap_unlock(mapping
);
704 i_mmap_unlock_write(mapping
);
707 /* add the VMA to the tree */
708 parent
= rb_prev
= NULL
;
709 p
= &mm
->mm_rb
.rb_node
;
712 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
714 /* sort by: start addr, end addr, VMA struct addr in that order
715 * (the latter is necessary as we may get identical VMAs) */
716 if (vma
->vm_start
< pvma
->vm_start
)
718 else if (vma
->vm_start
> pvma
->vm_start
) {
721 } else if (vma
->vm_end
< pvma
->vm_end
)
723 else if (vma
->vm_end
> pvma
->vm_end
) {
726 } else if (vma
< pvma
)
728 else if (vma
> pvma
) {
735 rb_link_node(&vma
->vm_rb
, parent
, p
);
736 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
738 /* add VMA to the VMA list also */
741 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
743 __vma_link_list(mm
, vma
, prev
, parent
);
747 * delete a VMA from its owning mm_struct and address space
749 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
752 struct address_space
*mapping
;
753 struct mm_struct
*mm
= vma
->vm_mm
;
754 struct task_struct
*curr
= current
;
759 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
760 /* if the vma is cached, invalidate the entire cache */
761 if (curr
->vmacache
.vmas
[i
] == vma
) {
762 vmacache_invalidate(mm
);
767 /* remove the VMA from the mapping */
769 mapping
= vma
->vm_file
->f_mapping
;
771 i_mmap_lock_write(mapping
);
772 flush_dcache_mmap_lock(mapping
);
773 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
774 flush_dcache_mmap_unlock(mapping
);
775 i_mmap_unlock_write(mapping
);
778 /* remove from the MM's tree and list */
779 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
782 vma
->vm_prev
->vm_next
= vma
->vm_next
;
784 mm
->mmap
= vma
->vm_next
;
787 vma
->vm_next
->vm_prev
= vma
->vm_prev
;
791 * destroy a VMA record
793 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
795 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
796 vma
->vm_ops
->close(vma
);
799 put_nommu_region(vma
->vm_region
);
800 kmem_cache_free(vm_area_cachep
, vma
);
804 * look up the first VMA in which addr resides, NULL if none
805 * - should be called with mm->mmap_sem at least held readlocked
807 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
809 struct vm_area_struct
*vma
;
811 /* check the cache first */
812 vma
= vmacache_find(mm
, addr
);
816 /* trawl the list (there may be multiple mappings in which addr
818 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
819 if (vma
->vm_start
> addr
)
821 if (vma
->vm_end
> addr
) {
822 vmacache_update(addr
, vma
);
829 EXPORT_SYMBOL(find_vma
);
833 * - we don't extend stack VMAs under NOMMU conditions
835 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
837 return find_vma(mm
, addr
);
841 * expand a stack to a given address
842 * - not supported under NOMMU conditions
844 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
850 * look up the first VMA exactly that exactly matches addr
851 * - should be called with mm->mmap_sem at least held readlocked
853 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
857 struct vm_area_struct
*vma
;
858 unsigned long end
= addr
+ len
;
860 /* check the cache first */
861 vma
= vmacache_find_exact(mm
, addr
, end
);
865 /* trawl the list (there may be multiple mappings in which addr
867 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
868 if (vma
->vm_start
< addr
)
870 if (vma
->vm_start
> addr
)
872 if (vma
->vm_end
== end
) {
873 vmacache_update(addr
, vma
);
882 * determine whether a mapping should be permitted and, if so, what sort of
883 * mapping we're capable of supporting
885 static int validate_mmap_request(struct file
*file
,
891 unsigned long *_capabilities
)
893 unsigned long capabilities
, rlen
;
896 /* do the simple checks first */
897 if (flags
& MAP_FIXED
)
900 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
901 (flags
& MAP_TYPE
) != MAP_SHARED
)
907 /* Careful about overflows.. */
908 rlen
= PAGE_ALIGN(len
);
909 if (!rlen
|| rlen
> TASK_SIZE
)
912 /* offset overflow? */
913 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
917 /* files must support mmap */
918 if (!file
->f_op
->mmap
)
921 /* work out if what we've got could possibly be shared
922 * - we support chardevs that provide their own "memory"
923 * - we support files/blockdevs that are memory backed
925 if (file
->f_op
->mmap_capabilities
) {
926 capabilities
= file
->f_op
->mmap_capabilities(file
);
928 /* no explicit capabilities set, so assume some
930 switch (file_inode(file
)->i_mode
& S_IFMT
) {
933 capabilities
= NOMMU_MAP_COPY
;
948 /* eliminate any capabilities that we can't support on this
950 if (!file
->f_op
->get_unmapped_area
)
951 capabilities
&= ~NOMMU_MAP_DIRECT
;
952 if (!(file
->f_mode
& FMODE_CAN_READ
))
953 capabilities
&= ~NOMMU_MAP_COPY
;
955 /* The file shall have been opened with read permission. */
956 if (!(file
->f_mode
& FMODE_READ
))
959 if (flags
& MAP_SHARED
) {
960 /* do checks for writing, appending and locking */
961 if ((prot
& PROT_WRITE
) &&
962 !(file
->f_mode
& FMODE_WRITE
))
965 if (IS_APPEND(file_inode(file
)) &&
966 (file
->f_mode
& FMODE_WRITE
))
969 if (locks_verify_locked(file
))
972 if (!(capabilities
& NOMMU_MAP_DIRECT
))
975 /* we mustn't privatise shared mappings */
976 capabilities
&= ~NOMMU_MAP_COPY
;
978 /* we're going to read the file into private memory we
980 if (!(capabilities
& NOMMU_MAP_COPY
))
983 /* we don't permit a private writable mapping to be
984 * shared with the backing device */
985 if (prot
& PROT_WRITE
)
986 capabilities
&= ~NOMMU_MAP_DIRECT
;
989 if (capabilities
& NOMMU_MAP_DIRECT
) {
990 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
991 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
992 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
994 capabilities
&= ~NOMMU_MAP_DIRECT
;
995 if (flags
& MAP_SHARED
) {
996 pr_warn("MAP_SHARED not completely supported on !MMU\n");
1002 /* handle executable mappings and implied executable
1004 if (path_noexec(&file
->f_path
)) {
1005 if (prot
& PROT_EXEC
)
1007 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
1008 /* handle implication of PROT_EXEC by PROT_READ */
1009 if (current
->personality
& READ_IMPLIES_EXEC
) {
1010 if (capabilities
& NOMMU_MAP_EXEC
)
1013 } else if ((prot
& PROT_READ
) &&
1014 (prot
& PROT_EXEC
) &&
1015 !(capabilities
& NOMMU_MAP_EXEC
)
1017 /* backing file is not executable, try to copy */
1018 capabilities
&= ~NOMMU_MAP_DIRECT
;
1021 /* anonymous mappings are always memory backed and can be
1024 capabilities
= NOMMU_MAP_COPY
;
1026 /* handle PROT_EXEC implication by PROT_READ */
1027 if ((prot
& PROT_READ
) &&
1028 (current
->personality
& READ_IMPLIES_EXEC
))
1032 /* allow the security API to have its say */
1033 ret
= security_mmap_addr(addr
);
1038 *_capabilities
= capabilities
;
1043 * we've determined that we can make the mapping, now translate what we
1044 * now know into VMA flags
1046 static unsigned long determine_vm_flags(struct file
*file
,
1048 unsigned long flags
,
1049 unsigned long capabilities
)
1051 unsigned long vm_flags
;
1053 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
1054 /* vm_flags |= mm->def_flags; */
1056 if (!(capabilities
& NOMMU_MAP_DIRECT
)) {
1057 /* attempt to share read-only copies of mapped file chunks */
1058 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1059 if (file
&& !(prot
& PROT_WRITE
))
1060 vm_flags
|= VM_MAYSHARE
;
1062 /* overlay a shareable mapping on the backing device or inode
1063 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1065 vm_flags
|= VM_MAYSHARE
| (capabilities
& NOMMU_VMFLAGS
);
1066 if (flags
& MAP_SHARED
)
1067 vm_flags
|= VM_SHARED
;
1070 /* refuse to let anyone share private mappings with this process if
1071 * it's being traced - otherwise breakpoints set in it may interfere
1072 * with another untraced process
1074 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
1075 vm_flags
&= ~VM_MAYSHARE
;
1081 * set up a shared mapping on a file (the driver or filesystem provides and
1084 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1088 ret
= call_mmap(vma
->vm_file
, vma
);
1090 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1096 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1097 * opposed to tried but failed) so we can only give a suitable error as
1098 * it's not possible to make a private copy if MAP_SHARED was given */
1103 * set up a private mapping or an anonymous shared mapping
1105 static int do_mmap_private(struct vm_area_struct
*vma
,
1106 struct vm_region
*region
,
1108 unsigned long capabilities
)
1110 unsigned long total
, point
;
1114 /* invoke the file's mapping function so that it can keep track of
1115 * shared mappings on devices or memory
1116 * - VM_MAYSHARE will be set if it may attempt to share
1118 if (capabilities
& NOMMU_MAP_DIRECT
) {
1119 ret
= call_mmap(vma
->vm_file
, vma
);
1121 /* shouldn't return success if we're not sharing */
1122 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1123 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1129 /* getting an ENOSYS error indicates that direct mmap isn't
1130 * possible (as opposed to tried but failed) so we'll try to
1131 * make a private copy of the data and map that instead */
1135 /* allocate some memory to hold the mapping
1136 * - note that this may not return a page-aligned address if the object
1137 * we're allocating is smaller than a page
1139 order
= get_order(len
);
1141 point
= len
>> PAGE_SHIFT
;
1143 /* we don't want to allocate a power-of-2 sized page set */
1144 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
1147 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
1151 atomic_long_add(total
, &mmap_pages_allocated
);
1153 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1154 region
->vm_start
= (unsigned long) base
;
1155 region
->vm_end
= region
->vm_start
+ len
;
1156 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1158 vma
->vm_start
= region
->vm_start
;
1159 vma
->vm_end
= region
->vm_start
+ len
;
1162 /* read the contents of a file into the copy */
1163 mm_segment_t old_fs
;
1166 fpos
= vma
->vm_pgoff
;
1167 fpos
<<= PAGE_SHIFT
;
1171 ret
= __vfs_read(vma
->vm_file
, base
, len
, &fpos
);
1177 /* clear the last little bit */
1179 memset(base
+ ret
, 0, len
- ret
);
1186 free_page_series(region
->vm_start
, region
->vm_top
);
1187 region
->vm_start
= vma
->vm_start
= 0;
1188 region
->vm_end
= vma
->vm_end
= 0;
1193 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1194 len
, current
->pid
, current
->comm
);
1195 show_free_areas(0, NULL
);
1200 * handle mapping creation for uClinux
1202 unsigned long do_mmap(struct file
*file
,
1206 unsigned long flags
,
1207 vm_flags_t vm_flags
,
1208 unsigned long pgoff
,
1209 unsigned long *populate
,
1210 struct list_head
*uf
)
1212 struct vm_area_struct
*vma
;
1213 struct vm_region
*region
;
1215 unsigned long capabilities
, result
;
1220 /* decide whether we should attempt the mapping, and if so what sort of
1222 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1227 /* we ignore the address hint */
1229 len
= PAGE_ALIGN(len
);
1231 /* we've determined that we can make the mapping, now translate what we
1232 * now know into VMA flags */
1233 vm_flags
|= determine_vm_flags(file
, prot
, flags
, capabilities
);
1235 /* we're going to need to record the mapping */
1236 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1238 goto error_getting_region
;
1240 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1242 goto error_getting_vma
;
1244 region
->vm_usage
= 1;
1245 region
->vm_flags
= vm_flags
;
1246 region
->vm_pgoff
= pgoff
;
1248 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1249 vma
->vm_flags
= vm_flags
;
1250 vma
->vm_pgoff
= pgoff
;
1253 region
->vm_file
= get_file(file
);
1254 vma
->vm_file
= get_file(file
);
1257 down_write(&nommu_region_sem
);
1259 /* if we want to share, we need to check for regions created by other
1260 * mmap() calls that overlap with our proposed mapping
1261 * - we can only share with a superset match on most regular files
1262 * - shared mappings on character devices and memory backed files are
1263 * permitted to overlap inexactly as far as we are concerned for in
1264 * these cases, sharing is handled in the driver or filesystem rather
1267 if (vm_flags
& VM_MAYSHARE
) {
1268 struct vm_region
*pregion
;
1269 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1271 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1272 pgend
= pgoff
+ pglen
;
1274 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1275 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1277 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1280 /* search for overlapping mappings on the same file */
1281 if (file_inode(pregion
->vm_file
) !=
1285 if (pregion
->vm_pgoff
>= pgend
)
1288 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1289 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1290 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1291 if (pgoff
>= rpgend
)
1294 /* handle inexactly overlapping matches between
1296 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1297 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1298 /* new mapping is not a subset of the region */
1299 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1300 goto sharing_violation
;
1304 /* we've found a region we can share */
1305 pregion
->vm_usage
++;
1306 vma
->vm_region
= pregion
;
1307 start
= pregion
->vm_start
;
1308 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1309 vma
->vm_start
= start
;
1310 vma
->vm_end
= start
+ len
;
1312 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1313 vma
->vm_flags
|= VM_MAPPED_COPY
;
1315 ret
= do_mmap_shared_file(vma
);
1317 vma
->vm_region
= NULL
;
1320 pregion
->vm_usage
--;
1322 goto error_just_free
;
1325 fput(region
->vm_file
);
1326 kmem_cache_free(vm_region_jar
, region
);
1332 /* obtain the address at which to make a shared mapping
1333 * - this is the hook for quasi-memory character devices to
1334 * tell us the location of a shared mapping
1336 if (capabilities
& NOMMU_MAP_DIRECT
) {
1337 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1339 if (IS_ERR_VALUE(addr
)) {
1342 goto error_just_free
;
1344 /* the driver refused to tell us where to site
1345 * the mapping so we'll have to attempt to copy
1348 if (!(capabilities
& NOMMU_MAP_COPY
))
1349 goto error_just_free
;
1351 capabilities
&= ~NOMMU_MAP_DIRECT
;
1353 vma
->vm_start
= region
->vm_start
= addr
;
1354 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1359 vma
->vm_region
= region
;
1361 /* set up the mapping
1362 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1364 if (file
&& vma
->vm_flags
& VM_SHARED
)
1365 ret
= do_mmap_shared_file(vma
);
1367 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1369 goto error_just_free
;
1370 add_nommu_region(region
);
1372 /* clear anonymous mappings that don't ask for uninitialized data */
1373 if (!vma
->vm_file
&& !(flags
& MAP_UNINITIALIZED
))
1374 memset((void *)region
->vm_start
, 0,
1375 region
->vm_end
- region
->vm_start
);
1377 /* okay... we have a mapping; now we have to register it */
1378 result
= vma
->vm_start
;
1380 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1383 add_vma_to_mm(current
->mm
, vma
);
1385 /* we flush the region from the icache only when the first executable
1386 * mapping of it is made */
1387 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1388 flush_icache_range(region
->vm_start
, region
->vm_end
);
1389 region
->vm_icache_flushed
= true;
1392 up_write(&nommu_region_sem
);
1397 up_write(&nommu_region_sem
);
1399 if (region
->vm_file
)
1400 fput(region
->vm_file
);
1401 kmem_cache_free(vm_region_jar
, region
);
1404 kmem_cache_free(vm_area_cachep
, vma
);
1408 up_write(&nommu_region_sem
);
1409 pr_warn("Attempt to share mismatched mappings\n");
1414 kmem_cache_free(vm_region_jar
, region
);
1415 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1417 show_free_areas(0, NULL
);
1420 error_getting_region
:
1421 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1423 show_free_areas(0, NULL
);
1427 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1428 unsigned long, prot
, unsigned long, flags
,
1429 unsigned long, fd
, unsigned long, pgoff
)
1431 struct file
*file
= NULL
;
1432 unsigned long retval
= -EBADF
;
1434 audit_mmap_fd(fd
, flags
);
1435 if (!(flags
& MAP_ANONYMOUS
)) {
1441 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1443 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1451 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1452 struct mmap_arg_struct
{
1456 unsigned long flags
;
1458 unsigned long offset
;
1461 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1463 struct mmap_arg_struct a
;
1465 if (copy_from_user(&a
, arg
, sizeof(a
)))
1467 if (offset_in_page(a
.offset
))
1470 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1471 a
.offset
>> PAGE_SHIFT
);
1473 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1476 * split a vma into two pieces at address 'addr', a new vma is allocated either
1477 * for the first part or the tail.
1479 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1480 unsigned long addr
, int new_below
)
1482 struct vm_area_struct
*new;
1483 struct vm_region
*region
;
1484 unsigned long npages
;
1486 /* we're only permitted to split anonymous regions (these should have
1487 * only a single usage on the region) */
1491 if (mm
->map_count
>= sysctl_max_map_count
)
1494 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1498 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1500 kmem_cache_free(vm_region_jar
, region
);
1504 /* most fields are the same, copy all, and then fixup */
1506 *region
= *vma
->vm_region
;
1507 new->vm_region
= region
;
1509 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1512 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1514 region
->vm_start
= new->vm_start
= addr
;
1515 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1518 if (new->vm_ops
&& new->vm_ops
->open
)
1519 new->vm_ops
->open(new);
1521 delete_vma_from_mm(vma
);
1522 down_write(&nommu_region_sem
);
1523 delete_nommu_region(vma
->vm_region
);
1525 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1526 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1528 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1529 vma
->vm_region
->vm_top
= addr
;
1531 add_nommu_region(vma
->vm_region
);
1532 add_nommu_region(new->vm_region
);
1533 up_write(&nommu_region_sem
);
1534 add_vma_to_mm(mm
, vma
);
1535 add_vma_to_mm(mm
, new);
1540 * shrink a VMA by removing the specified chunk from either the beginning or
1543 static int shrink_vma(struct mm_struct
*mm
,
1544 struct vm_area_struct
*vma
,
1545 unsigned long from
, unsigned long to
)
1547 struct vm_region
*region
;
1549 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1551 delete_vma_from_mm(vma
);
1552 if (from
> vma
->vm_start
)
1556 add_vma_to_mm(mm
, vma
);
1558 /* cut the backing region down to size */
1559 region
= vma
->vm_region
;
1560 BUG_ON(region
->vm_usage
!= 1);
1562 down_write(&nommu_region_sem
);
1563 delete_nommu_region(region
);
1564 if (from
> region
->vm_start
) {
1565 to
= region
->vm_top
;
1566 region
->vm_top
= region
->vm_end
= from
;
1568 region
->vm_start
= to
;
1570 add_nommu_region(region
);
1571 up_write(&nommu_region_sem
);
1573 free_page_series(from
, to
);
1579 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1580 * VMA, though it need not cover the whole VMA
1582 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1584 struct vm_area_struct
*vma
;
1588 len
= PAGE_ALIGN(len
);
1594 /* find the first potentially overlapping VMA */
1595 vma
= find_vma(mm
, start
);
1599 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1600 current
->pid
, current
->comm
,
1601 start
, start
+ len
- 1);
1607 /* we're allowed to split an anonymous VMA but not a file-backed one */
1610 if (start
> vma
->vm_start
)
1612 if (end
== vma
->vm_end
)
1613 goto erase_whole_vma
;
1618 /* the chunk must be a subset of the VMA found */
1619 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1620 goto erase_whole_vma
;
1621 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1623 if (offset_in_page(start
))
1625 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1627 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1628 ret
= split_vma(mm
, vma
, start
, 1);
1632 return shrink_vma(mm
, vma
, start
, end
);
1636 delete_vma_from_mm(vma
);
1637 delete_vma(mm
, vma
);
1640 EXPORT_SYMBOL(do_munmap
);
1642 int vm_munmap(unsigned long addr
, size_t len
)
1644 struct mm_struct
*mm
= current
->mm
;
1647 down_write(&mm
->mmap_sem
);
1648 ret
= do_munmap(mm
, addr
, len
, NULL
);
1649 up_write(&mm
->mmap_sem
);
1652 EXPORT_SYMBOL(vm_munmap
);
1654 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1656 return vm_munmap(addr
, len
);
1660 * release all the mappings made in a process's VM space
1662 void exit_mmap(struct mm_struct
*mm
)
1664 struct vm_area_struct
*vma
;
1671 while ((vma
= mm
->mmap
)) {
1672 mm
->mmap
= vma
->vm_next
;
1673 delete_vma_from_mm(vma
);
1674 delete_vma(mm
, vma
);
1679 int vm_brk(unsigned long addr
, unsigned long len
)
1685 * expand (or shrink) an existing mapping, potentially moving it at the same
1686 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1688 * under NOMMU conditions, we only permit changing a mapping's size, and only
1689 * as long as it stays within the region allocated by do_mmap_private() and the
1690 * block is not shareable
1692 * MREMAP_FIXED is not supported under NOMMU conditions
1694 static unsigned long do_mremap(unsigned long addr
,
1695 unsigned long old_len
, unsigned long new_len
,
1696 unsigned long flags
, unsigned long new_addr
)
1698 struct vm_area_struct
*vma
;
1700 /* insanity checks first */
1701 old_len
= PAGE_ALIGN(old_len
);
1702 new_len
= PAGE_ALIGN(new_len
);
1703 if (old_len
== 0 || new_len
== 0)
1704 return (unsigned long) -EINVAL
;
1706 if (offset_in_page(addr
))
1709 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1710 return (unsigned long) -EINVAL
;
1712 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1714 return (unsigned long) -EINVAL
;
1716 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1717 return (unsigned long) -EFAULT
;
1719 if (vma
->vm_flags
& VM_MAYSHARE
)
1720 return (unsigned long) -EPERM
;
1722 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1723 return (unsigned long) -ENOMEM
;
1725 /* all checks complete - do it */
1726 vma
->vm_end
= vma
->vm_start
+ new_len
;
1727 return vma
->vm_start
;
1730 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1731 unsigned long, new_len
, unsigned long, flags
,
1732 unsigned long, new_addr
)
1736 down_write(¤t
->mm
->mmap_sem
);
1737 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1738 up_write(¤t
->mm
->mmap_sem
);
1742 struct page
*follow_page_mask(struct vm_area_struct
*vma
,
1743 unsigned long address
, unsigned int flags
,
1744 unsigned int *page_mask
)
1750 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1751 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1753 if (addr
!= (pfn
<< PAGE_SHIFT
))
1756 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1759 EXPORT_SYMBOL(remap_pfn_range
);
1761 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1763 unsigned long pfn
= start
>> PAGE_SHIFT
;
1764 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1766 pfn
+= vma
->vm_pgoff
;
1767 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1769 EXPORT_SYMBOL(vm_iomap_memory
);
1771 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1772 unsigned long pgoff
)
1774 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1776 if (!(vma
->vm_flags
& VM_USERMAP
))
1779 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1780 vma
->vm_end
= vma
->vm_start
+ size
;
1784 EXPORT_SYMBOL(remap_vmalloc_range
);
1786 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1787 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1792 void unmap_mapping_range(struct address_space
*mapping
,
1793 loff_t
const holebegin
, loff_t
const holelen
,
1797 EXPORT_SYMBOL(unmap_mapping_range
);
1799 int filemap_fault(struct vm_fault
*vmf
)
1804 EXPORT_SYMBOL(filemap_fault
);
1806 void filemap_map_pages(struct vm_fault
*vmf
,
1807 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1811 EXPORT_SYMBOL(filemap_map_pages
);
1813 int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
1814 unsigned long addr
, void *buf
, int len
, unsigned int gup_flags
)
1816 struct vm_area_struct
*vma
;
1817 int write
= gup_flags
& FOLL_WRITE
;
1819 down_read(&mm
->mmap_sem
);
1821 /* the access must start within one of the target process's mappings */
1822 vma
= find_vma(mm
, addr
);
1824 /* don't overrun this mapping */
1825 if (addr
+ len
>= vma
->vm_end
)
1826 len
= vma
->vm_end
- addr
;
1828 /* only read or write mappings where it is permitted */
1829 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1830 copy_to_user_page(vma
, NULL
, addr
,
1831 (void *) addr
, buf
, len
);
1832 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1833 copy_from_user_page(vma
, NULL
, addr
,
1834 buf
, (void *) addr
, len
);
1841 up_read(&mm
->mmap_sem
);
1847 * @access_remote_vm - access another process' address space
1848 * @mm: the mm_struct of the target address space
1849 * @addr: start address to access
1850 * @buf: source or destination buffer
1851 * @len: number of bytes to transfer
1852 * @gup_flags: flags modifying lookup behaviour
1854 * The caller must hold a reference on @mm.
1856 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1857 void *buf
, int len
, unsigned int gup_flags
)
1859 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, gup_flags
);
1863 * Access another process' address space.
1864 * - source/target buffer must be kernel space
1866 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1867 unsigned int gup_flags
)
1869 struct mm_struct
*mm
;
1871 if (addr
+ len
< addr
)
1874 mm
= get_task_mm(tsk
);
1878 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, gup_flags
);
1883 EXPORT_SYMBOL_GPL(access_process_vm
);
1886 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1887 * @inode: The inode to check
1888 * @size: The current filesize of the inode
1889 * @newsize: The proposed filesize of the inode
1891 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1892 * make sure that that any outstanding VMAs aren't broken and then shrink the
1893 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1894 * automatically grant mappings that are too large.
1896 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1899 struct vm_area_struct
*vma
;
1900 struct vm_region
*region
;
1902 size_t r_size
, r_top
;
1904 low
= newsize
>> PAGE_SHIFT
;
1905 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1907 down_write(&nommu_region_sem
);
1908 i_mmap_lock_read(inode
->i_mapping
);
1910 /* search for VMAs that fall within the dead zone */
1911 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1912 /* found one - only interested if it's shared out of the page
1914 if (vma
->vm_flags
& VM_SHARED
) {
1915 i_mmap_unlock_read(inode
->i_mapping
);
1916 up_write(&nommu_region_sem
);
1917 return -ETXTBSY
; /* not quite true, but near enough */
1921 /* reduce any regions that overlap the dead zone - if in existence,
1922 * these will be pointed to by VMAs that don't overlap the dead zone
1924 * we don't check for any regions that start beyond the EOF as there
1927 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1928 if (!(vma
->vm_flags
& VM_SHARED
))
1931 region
= vma
->vm_region
;
1932 r_size
= region
->vm_top
- region
->vm_start
;
1933 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1935 if (r_top
> newsize
) {
1936 region
->vm_top
-= r_top
- newsize
;
1937 if (region
->vm_end
> region
->vm_top
)
1938 region
->vm_end
= region
->vm_top
;
1942 i_mmap_unlock_read(inode
->i_mapping
);
1943 up_write(&nommu_region_sem
);
1948 * Initialise sysctl_user_reserve_kbytes.
1950 * This is intended to prevent a user from starting a single memory hogging
1951 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1954 * The default value is min(3% of free memory, 128MB)
1955 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1957 static int __meminit
init_user_reserve(void)
1959 unsigned long free_kbytes
;
1961 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1963 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1966 subsys_initcall(init_user_reserve
);
1969 * Initialise sysctl_admin_reserve_kbytes.
1971 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1972 * to log in and kill a memory hogging process.
1974 * Systems with more than 256MB will reserve 8MB, enough to recover
1975 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1976 * only reserve 3% of free pages by default.
1978 static int __meminit
init_admin_reserve(void)
1980 unsigned long free_kbytes
;
1982 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1984 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1987 subsys_initcall(init_admin_reserve
);