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 #include <linux/export.h>
18 #include <linux/vmacache.h>
19 #include <linux/mman.h>
20 #include <linux/swap.h>
21 #include <linux/file.h>
22 #include <linux/highmem.h>
23 #include <linux/pagemap.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/compiler.h>
29 #include <linux/mount.h>
30 #include <linux/personality.h>
31 #include <linux/security.h>
32 #include <linux/syscalls.h>
33 #include <linux/audit.h>
34 #include <linux/sched/sysctl.h>
36 #include <asm/uaccess.h>
38 #include <asm/tlbflush.h>
39 #include <asm/mmu_context.h>
43 #define kenter(FMT, ...) \
44 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
45 #define kleave(FMT, ...) \
46 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
47 #define kdebug(FMT, ...) \
48 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
50 #define kenter(FMT, ...) \
51 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
52 #define kleave(FMT, ...) \
53 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
54 #define kdebug(FMT, ...) \
55 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
60 unsigned long max_mapnr
;
61 unsigned long highest_memmap_pfn
;
62 struct percpu_counter vm_committed_as
;
63 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
64 int sysctl_overcommit_ratio
= 50; /* default is 50% */
65 unsigned long sysctl_overcommit_kbytes __read_mostly
;
66 int sysctl_max_map_count
= DEFAULT_MAX_MAP_COUNT
;
67 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
68 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
69 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
70 int heap_stack_gap
= 0;
72 atomic_long_t mmap_pages_allocated
;
75 * The global memory commitment made in the system can be a metric
76 * that can be used to drive ballooning decisions when Linux is hosted
77 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
78 * balancing memory across competing virtual machines that are hosted.
79 * Several metrics drive this policy engine including the guest reported
82 unsigned long vm_memory_committed(void)
84 return percpu_counter_read_positive(&vm_committed_as
);
87 EXPORT_SYMBOL_GPL(vm_memory_committed
);
89 EXPORT_SYMBOL(mem_map
);
91 /* list of mapped, potentially shareable regions */
92 static struct kmem_cache
*vm_region_jar
;
93 struct rb_root nommu_region_tree
= RB_ROOT
;
94 DECLARE_RWSEM(nommu_region_sem
);
96 const struct vm_operations_struct generic_file_vm_ops
= {
100 * Return the total memory allocated for this pointer, not
101 * just what the caller asked for.
103 * Doesn't have to be accurate, i.e. may have races.
105 unsigned int kobjsize(const void *objp
)
110 * If the object we have should not have ksize performed on it,
113 if (!objp
|| !virt_addr_valid(objp
))
116 page
= virt_to_head_page(objp
);
119 * If the allocator sets PageSlab, we know the pointer came from
126 * If it's not a compound page, see if we have a matching VMA
127 * region. This test is intentionally done in reverse order,
128 * so if there's no VMA, we still fall through and hand back
129 * PAGE_SIZE for 0-order pages.
131 if (!PageCompound(page
)) {
132 struct vm_area_struct
*vma
;
134 vma
= find_vma(current
->mm
, (unsigned long)objp
);
136 return vma
->vm_end
- vma
->vm_start
;
140 * The ksize() function is only guaranteed to work for pointers
141 * returned by kmalloc(). So handle arbitrary pointers here.
143 return PAGE_SIZE
<< compound_order(page
);
146 long __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
147 unsigned long start
, unsigned long nr_pages
,
148 unsigned int foll_flags
, struct page
**pages
,
149 struct vm_area_struct
**vmas
, int *nonblocking
)
151 struct vm_area_struct
*vma
;
152 unsigned long vm_flags
;
155 /* calculate required read or write permissions.
156 * If FOLL_FORCE is set, we only require the "MAY" flags.
158 vm_flags
= (foll_flags
& FOLL_WRITE
) ?
159 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
160 vm_flags
&= (foll_flags
& FOLL_FORCE
) ?
161 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
163 for (i
= 0; i
< nr_pages
; i
++) {
164 vma
= find_vma(mm
, start
);
166 goto finish_or_fault
;
168 /* protect what we can, including chardevs */
169 if ((vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
170 !(vm_flags
& vma
->vm_flags
))
171 goto finish_or_fault
;
174 pages
[i
] = virt_to_page(start
);
176 page_cache_get(pages
[i
]);
180 start
= (start
+ PAGE_SIZE
) & PAGE_MASK
;
186 return i
? : -EFAULT
;
190 * get a list of pages in an address range belonging to the specified process
191 * and indicate the VMA that covers each page
192 * - this is potentially dodgy as we may end incrementing the page count of a
193 * slab page or a secondary page from a compound page
194 * - don't permit access to VMAs that don't support it, such as I/O mappings
196 long get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
197 unsigned long start
, unsigned long nr_pages
,
198 int write
, int force
, struct page
**pages
,
199 struct vm_area_struct
**vmas
)
208 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
,
211 EXPORT_SYMBOL(get_user_pages
);
214 * follow_pfn - look up PFN at a user virtual address
215 * @vma: memory mapping
216 * @address: user virtual address
217 * @pfn: location to store found PFN
219 * Only IO mappings and raw PFN mappings are allowed.
221 * Returns zero and the pfn at @pfn on success, -ve otherwise.
223 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
226 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
229 *pfn
= address
>> PAGE_SHIFT
;
232 EXPORT_SYMBOL(follow_pfn
);
234 LIST_HEAD(vmap_area_list
);
236 void vfree(const void *addr
)
240 EXPORT_SYMBOL(vfree
);
242 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
245 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
246 * returns only a logical address.
248 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
250 EXPORT_SYMBOL(__vmalloc
);
252 void *vmalloc_user(unsigned long size
)
256 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
259 struct vm_area_struct
*vma
;
261 down_write(¤t
->mm
->mmap_sem
);
262 vma
= find_vma(current
->mm
, (unsigned long)ret
);
264 vma
->vm_flags
|= VM_USERMAP
;
265 up_write(¤t
->mm
->mmap_sem
);
270 EXPORT_SYMBOL(vmalloc_user
);
272 struct page
*vmalloc_to_page(const void *addr
)
274 return virt_to_page(addr
);
276 EXPORT_SYMBOL(vmalloc_to_page
);
278 unsigned long vmalloc_to_pfn(const void *addr
)
280 return page_to_pfn(virt_to_page(addr
));
282 EXPORT_SYMBOL(vmalloc_to_pfn
);
284 long vread(char *buf
, char *addr
, unsigned long count
)
286 /* Don't allow overflow */
287 if ((unsigned long) buf
+ count
< count
)
288 count
= -(unsigned long) buf
;
290 memcpy(buf
, addr
, count
);
294 long vwrite(char *buf
, char *addr
, unsigned long count
)
296 /* Don't allow overflow */
297 if ((unsigned long) addr
+ count
< count
)
298 count
= -(unsigned long) addr
;
300 memcpy(addr
, buf
, count
);
305 * vmalloc - allocate virtually continguos memory
307 * @size: allocation size
309 * Allocate enough pages to cover @size from the page level
310 * allocator and map them into continguos kernel virtual space.
312 * For tight control over page level allocator and protection flags
313 * use __vmalloc() instead.
315 void *vmalloc(unsigned long size
)
317 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
319 EXPORT_SYMBOL(vmalloc
);
322 * vzalloc - allocate virtually continguos memory with zero fill
324 * @size: allocation size
326 * Allocate enough pages to cover @size from the page level
327 * allocator and map them into continguos kernel virtual space.
328 * The memory allocated is set to zero.
330 * For tight control over page level allocator and protection flags
331 * use __vmalloc() instead.
333 void *vzalloc(unsigned long size
)
335 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
338 EXPORT_SYMBOL(vzalloc
);
341 * vmalloc_node - allocate memory on a specific node
342 * @size: allocation size
345 * Allocate enough pages to cover @size from the page level
346 * allocator and map them into contiguous kernel virtual space.
348 * For tight control over page level allocator and protection flags
349 * use __vmalloc() instead.
351 void *vmalloc_node(unsigned long size
, int node
)
353 return vmalloc(size
);
355 EXPORT_SYMBOL(vmalloc_node
);
358 * vzalloc_node - allocate memory on a specific node with zero fill
359 * @size: allocation size
362 * Allocate enough pages to cover @size from the page level
363 * allocator and map them into contiguous kernel virtual space.
364 * The memory allocated is set to zero.
366 * For tight control over page level allocator and protection flags
367 * use __vmalloc() instead.
369 void *vzalloc_node(unsigned long size
, int node
)
371 return vzalloc(size
);
373 EXPORT_SYMBOL(vzalloc_node
);
375 #ifndef PAGE_KERNEL_EXEC
376 # define PAGE_KERNEL_EXEC PAGE_KERNEL
380 * vmalloc_exec - allocate virtually contiguous, executable memory
381 * @size: allocation size
383 * Kernel-internal function to allocate enough pages to cover @size
384 * the page level allocator and map them into contiguous and
385 * executable kernel virtual space.
387 * For tight control over page level allocator and protection flags
388 * use __vmalloc() instead.
391 void *vmalloc_exec(unsigned long size
)
393 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
397 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
398 * @size: allocation size
400 * Allocate enough 32bit PA addressable pages to cover @size from the
401 * page level allocator and map them into continguos kernel virtual space.
403 void *vmalloc_32(unsigned long size
)
405 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
407 EXPORT_SYMBOL(vmalloc_32
);
410 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
411 * @size: allocation size
413 * The resulting memory area is 32bit addressable and zeroed so it can be
414 * mapped to userspace without leaking data.
416 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
417 * remap_vmalloc_range() are permissible.
419 void *vmalloc_32_user(unsigned long size
)
422 * We'll have to sort out the ZONE_DMA bits for 64-bit,
423 * but for now this can simply use vmalloc_user() directly.
425 return vmalloc_user(size
);
427 EXPORT_SYMBOL(vmalloc_32_user
);
429 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
436 void vunmap(const void *addr
)
440 EXPORT_SYMBOL(vunmap
);
442 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
447 EXPORT_SYMBOL(vm_map_ram
);
449 void vm_unmap_ram(const void *mem
, unsigned int count
)
453 EXPORT_SYMBOL(vm_unmap_ram
);
455 void vm_unmap_aliases(void)
458 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
461 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
464 void __weak
vmalloc_sync_all(void)
469 * alloc_vm_area - allocate a range of kernel address space
470 * @size: size of the area
472 * Returns: NULL on failure, vm_struct on success
474 * This function reserves a range of kernel address space, and
475 * allocates pagetables to map that range. No actual mappings
476 * are created. If the kernel address space is not shared
477 * between processes, it syncs the pagetable across all
480 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
485 EXPORT_SYMBOL_GPL(alloc_vm_area
);
487 void free_vm_area(struct vm_struct
*area
)
491 EXPORT_SYMBOL_GPL(free_vm_area
);
493 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
498 EXPORT_SYMBOL(vm_insert_page
);
501 * sys_brk() for the most part doesn't need the global kernel
502 * lock, except when an application is doing something nasty
503 * like trying to un-brk an area that has already been mapped
504 * to a regular file. in this case, the unmapping will need
505 * to invoke file system routines that need the global lock.
507 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
509 struct mm_struct
*mm
= current
->mm
;
511 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
518 * Always allow shrinking brk
520 if (brk
<= mm
->brk
) {
526 * Ok, looks good - let it rip.
528 flush_icache_range(mm
->brk
, brk
);
529 return mm
->brk
= brk
;
533 * initialise the VMA and region record slabs
535 void __init
mmap_init(void)
539 ret
= percpu_counter_init(&vm_committed_as
, 0);
541 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
);
545 * validate the region tree
546 * - the caller must hold the region lock
548 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
549 static noinline
void validate_nommu_regions(void)
551 struct vm_region
*region
, *last
;
552 struct rb_node
*p
, *lastp
;
554 lastp
= rb_first(&nommu_region_tree
);
558 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
559 BUG_ON(unlikely(last
->vm_end
<= last
->vm_start
));
560 BUG_ON(unlikely(last
->vm_top
< last
->vm_end
));
562 while ((p
= rb_next(lastp
))) {
563 region
= rb_entry(p
, struct vm_region
, vm_rb
);
564 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
566 BUG_ON(unlikely(region
->vm_end
<= region
->vm_start
));
567 BUG_ON(unlikely(region
->vm_top
< region
->vm_end
));
568 BUG_ON(unlikely(region
->vm_start
< last
->vm_top
));
574 static void validate_nommu_regions(void)
580 * add a region into the global tree
582 static void add_nommu_region(struct vm_region
*region
)
584 struct vm_region
*pregion
;
585 struct rb_node
**p
, *parent
;
587 validate_nommu_regions();
590 p
= &nommu_region_tree
.rb_node
;
593 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
594 if (region
->vm_start
< pregion
->vm_start
)
596 else if (region
->vm_start
> pregion
->vm_start
)
598 else if (pregion
== region
)
604 rb_link_node(®ion
->vm_rb
, parent
, p
);
605 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
607 validate_nommu_regions();
611 * delete a region from the global tree
613 static void delete_nommu_region(struct vm_region
*region
)
615 BUG_ON(!nommu_region_tree
.rb_node
);
617 validate_nommu_regions();
618 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
619 validate_nommu_regions();
623 * free a contiguous series of pages
625 static void free_page_series(unsigned long from
, unsigned long to
)
627 for (; from
< to
; from
+= PAGE_SIZE
) {
628 struct page
*page
= virt_to_page(from
);
630 kdebug("- free %lx", from
);
631 atomic_long_dec(&mmap_pages_allocated
);
632 if (page_count(page
) != 1)
633 kdebug("free page %p: refcount not one: %d",
634 page
, page_count(page
));
640 * release a reference to a region
641 * - the caller must hold the region semaphore for writing, which this releases
642 * - the region may not have been added to the tree yet, in which case vm_top
643 * will equal vm_start
645 static void __put_nommu_region(struct vm_region
*region
)
646 __releases(nommu_region_sem
)
648 kenter("%p{%d}", region
, region
->vm_usage
);
650 BUG_ON(!nommu_region_tree
.rb_node
);
652 if (--region
->vm_usage
== 0) {
653 if (region
->vm_top
> region
->vm_start
)
654 delete_nommu_region(region
);
655 up_write(&nommu_region_sem
);
658 fput(region
->vm_file
);
660 /* IO memory and memory shared directly out of the pagecache
661 * from ramfs/tmpfs mustn't be released here */
662 if (region
->vm_flags
& VM_MAPPED_COPY
) {
663 kdebug("free series");
664 free_page_series(region
->vm_start
, region
->vm_top
);
666 kmem_cache_free(vm_region_jar
, region
);
668 up_write(&nommu_region_sem
);
673 * release a reference to a region
675 static void put_nommu_region(struct vm_region
*region
)
677 down_write(&nommu_region_sem
);
678 __put_nommu_region(region
);
682 * update protection on a vma
684 static void protect_vma(struct vm_area_struct
*vma
, unsigned long flags
)
687 struct mm_struct
*mm
= vma
->vm_mm
;
688 long start
= vma
->vm_start
& PAGE_MASK
;
689 while (start
< vma
->vm_end
) {
690 protect_page(mm
, start
, flags
);
693 update_protections(mm
);
698 * add a VMA into a process's mm_struct in the appropriate place in the list
699 * and tree and add to the address space's page tree also if not an anonymous
701 * - should be called with mm->mmap_sem held writelocked
703 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
705 struct vm_area_struct
*pvma
, *prev
;
706 struct address_space
*mapping
;
707 struct rb_node
**p
, *parent
, *rb_prev
;
711 BUG_ON(!vma
->vm_region
);
716 protect_vma(vma
, vma
->vm_flags
);
718 /* add the VMA to the mapping */
720 mapping
= vma
->vm_file
->f_mapping
;
722 mutex_lock(&mapping
->i_mmap_mutex
);
723 flush_dcache_mmap_lock(mapping
);
724 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
725 flush_dcache_mmap_unlock(mapping
);
726 mutex_unlock(&mapping
->i_mmap_mutex
);
729 /* add the VMA to the tree */
730 parent
= rb_prev
= NULL
;
731 p
= &mm
->mm_rb
.rb_node
;
734 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
736 /* sort by: start addr, end addr, VMA struct addr in that order
737 * (the latter is necessary as we may get identical VMAs) */
738 if (vma
->vm_start
< pvma
->vm_start
)
740 else if (vma
->vm_start
> pvma
->vm_start
) {
743 } else if (vma
->vm_end
< pvma
->vm_end
)
745 else if (vma
->vm_end
> pvma
->vm_end
) {
748 } else if (vma
< pvma
)
750 else if (vma
> pvma
) {
757 rb_link_node(&vma
->vm_rb
, parent
, p
);
758 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
760 /* add VMA to the VMA list also */
763 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
765 __vma_link_list(mm
, vma
, prev
, parent
);
769 * delete a VMA from its owning mm_struct and address space
771 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
774 struct address_space
*mapping
;
775 struct mm_struct
*mm
= vma
->vm_mm
;
776 struct task_struct
*curr
= current
;
783 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
784 /* if the vma is cached, invalidate the entire cache */
785 if (curr
->vmacache
[i
] == vma
) {
786 vmacache_invalidate(curr
->mm
);
791 /* remove the VMA from the mapping */
793 mapping
= vma
->vm_file
->f_mapping
;
795 mutex_lock(&mapping
->i_mmap_mutex
);
796 flush_dcache_mmap_lock(mapping
);
797 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
798 flush_dcache_mmap_unlock(mapping
);
799 mutex_unlock(&mapping
->i_mmap_mutex
);
802 /* remove from the MM's tree and list */
803 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
806 vma
->vm_prev
->vm_next
= vma
->vm_next
;
808 mm
->mmap
= vma
->vm_next
;
811 vma
->vm_next
->vm_prev
= vma
->vm_prev
;
815 * destroy a VMA record
817 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
820 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
821 vma
->vm_ops
->close(vma
);
824 put_nommu_region(vma
->vm_region
);
825 kmem_cache_free(vm_area_cachep
, vma
);
829 * look up the first VMA in which addr resides, NULL if none
830 * - should be called with mm->mmap_sem at least held readlocked
832 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
834 struct vm_area_struct
*vma
;
836 /* check the cache first */
837 vma
= vmacache_find(mm
, addr
);
841 /* trawl the list (there may be multiple mappings in which addr
843 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
844 if (vma
->vm_start
> addr
)
846 if (vma
->vm_end
> addr
) {
847 vmacache_update(addr
, vma
);
854 EXPORT_SYMBOL(find_vma
);
858 * - we don't extend stack VMAs under NOMMU conditions
860 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
862 return find_vma(mm
, addr
);
866 * expand a stack to a given address
867 * - not supported under NOMMU conditions
869 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
875 * look up the first VMA exactly that exactly matches addr
876 * - should be called with mm->mmap_sem at least held readlocked
878 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
882 struct vm_area_struct
*vma
;
883 unsigned long end
= addr
+ len
;
885 /* check the cache first */
886 vma
= vmacache_find_exact(mm
, addr
, end
);
890 /* trawl the list (there may be multiple mappings in which addr
892 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
893 if (vma
->vm_start
< addr
)
895 if (vma
->vm_start
> addr
)
897 if (vma
->vm_end
== end
) {
898 vmacache_update(addr
, vma
);
907 * determine whether a mapping should be permitted and, if so, what sort of
908 * mapping we're capable of supporting
910 static int validate_mmap_request(struct file
*file
,
916 unsigned long *_capabilities
)
918 unsigned long capabilities
, rlen
;
921 /* do the simple checks first */
922 if (flags
& MAP_FIXED
) {
924 "%d: Can't do fixed-address/overlay mmap of RAM\n",
929 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
930 (flags
& MAP_TYPE
) != MAP_SHARED
)
936 /* Careful about overflows.. */
937 rlen
= PAGE_ALIGN(len
);
938 if (!rlen
|| rlen
> TASK_SIZE
)
941 /* offset overflow? */
942 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
946 /* validate file mapping requests */
947 struct address_space
*mapping
;
949 /* files must support mmap */
950 if (!file
->f_op
->mmap
)
953 /* work out if what we've got could possibly be shared
954 * - we support chardevs that provide their own "memory"
955 * - we support files/blockdevs that are memory backed
957 mapping
= file
->f_mapping
;
959 mapping
= file_inode(file
)->i_mapping
;
962 if (mapping
&& mapping
->backing_dev_info
)
963 capabilities
= mapping
->backing_dev_info
->capabilities
;
966 /* no explicit capabilities set, so assume some
968 switch (file_inode(file
)->i_mode
& S_IFMT
) {
971 capabilities
= BDI_CAP_MAP_COPY
;
986 /* eliminate any capabilities that we can't support on this
988 if (!file
->f_op
->get_unmapped_area
)
989 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
990 if (!file
->f_op
->read
)
991 capabilities
&= ~BDI_CAP_MAP_COPY
;
993 /* The file shall have been opened with read permission. */
994 if (!(file
->f_mode
& FMODE_READ
))
997 if (flags
& MAP_SHARED
) {
998 /* do checks for writing, appending and locking */
999 if ((prot
& PROT_WRITE
) &&
1000 !(file
->f_mode
& FMODE_WRITE
))
1003 if (IS_APPEND(file_inode(file
)) &&
1004 (file
->f_mode
& FMODE_WRITE
))
1007 if (locks_verify_locked(file
))
1010 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1013 /* we mustn't privatise shared mappings */
1014 capabilities
&= ~BDI_CAP_MAP_COPY
;
1017 /* we're going to read the file into private memory we
1019 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1022 /* we don't permit a private writable mapping to be
1023 * shared with the backing device */
1024 if (prot
& PROT_WRITE
)
1025 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1028 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1029 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
1030 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
1031 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
1033 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1034 if (flags
& MAP_SHARED
) {
1036 "MAP_SHARED not completely supported on !MMU\n");
1042 /* handle executable mappings and implied executable
1044 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1045 if (prot
& PROT_EXEC
)
1048 else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
1049 /* handle implication of PROT_EXEC by PROT_READ */
1050 if (current
->personality
& READ_IMPLIES_EXEC
) {
1051 if (capabilities
& BDI_CAP_EXEC_MAP
)
1055 else if ((prot
& PROT_READ
) &&
1056 (prot
& PROT_EXEC
) &&
1057 !(capabilities
& BDI_CAP_EXEC_MAP
)
1059 /* backing file is not executable, try to copy */
1060 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1064 /* anonymous mappings are always memory backed and can be
1067 capabilities
= BDI_CAP_MAP_COPY
;
1069 /* handle PROT_EXEC implication by PROT_READ */
1070 if ((prot
& PROT_READ
) &&
1071 (current
->personality
& READ_IMPLIES_EXEC
))
1075 /* allow the security API to have its say */
1076 ret
= security_mmap_addr(addr
);
1081 *_capabilities
= capabilities
;
1086 * we've determined that we can make the mapping, now translate what we
1087 * now know into VMA flags
1089 static unsigned long determine_vm_flags(struct file
*file
,
1091 unsigned long flags
,
1092 unsigned long capabilities
)
1094 unsigned long vm_flags
;
1096 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
1097 /* vm_flags |= mm->def_flags; */
1099 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
1100 /* attempt to share read-only copies of mapped file chunks */
1101 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1102 if (file
&& !(prot
& PROT_WRITE
))
1103 vm_flags
|= VM_MAYSHARE
;
1105 /* overlay a shareable mapping on the backing device or inode
1106 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1108 vm_flags
|= VM_MAYSHARE
| (capabilities
& BDI_CAP_VMFLAGS
);
1109 if (flags
& MAP_SHARED
)
1110 vm_flags
|= VM_SHARED
;
1113 /* refuse to let anyone share private mappings with this process if
1114 * it's being traced - otherwise breakpoints set in it may interfere
1115 * with another untraced process
1117 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
1118 vm_flags
&= ~VM_MAYSHARE
;
1124 * set up a shared mapping on a file (the driver or filesystem provides and
1127 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1131 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1133 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1139 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1140 * opposed to tried but failed) so we can only give a suitable error as
1141 * it's not possible to make a private copy if MAP_SHARED was given */
1146 * set up a private mapping or an anonymous shared mapping
1148 static int do_mmap_private(struct vm_area_struct
*vma
,
1149 struct vm_region
*region
,
1151 unsigned long capabilities
)
1154 unsigned long total
, point
, n
;
1158 /* invoke the file's mapping function so that it can keep track of
1159 * shared mappings on devices or memory
1160 * - VM_MAYSHARE will be set if it may attempt to share
1162 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1163 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1165 /* shouldn't return success if we're not sharing */
1166 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1167 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1173 /* getting an ENOSYS error indicates that direct mmap isn't
1174 * possible (as opposed to tried but failed) so we'll try to
1175 * make a private copy of the data and map that instead */
1179 /* allocate some memory to hold the mapping
1180 * - note that this may not return a page-aligned address if the object
1181 * we're allocating is smaller than a page
1183 order
= get_order(len
);
1184 kdebug("alloc order %d for %lx", order
, len
);
1186 pages
= alloc_pages(GFP_KERNEL
, order
);
1191 atomic_long_add(total
, &mmap_pages_allocated
);
1193 point
= len
>> PAGE_SHIFT
;
1195 /* we allocated a power-of-2 sized page set, so we may want to trim off
1197 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
) {
1198 while (total
> point
) {
1199 order
= ilog2(total
- point
);
1201 kdebug("shave %lu/%lu @%lu", n
, total
- point
, total
);
1202 atomic_long_sub(n
, &mmap_pages_allocated
);
1204 set_page_refcounted(pages
+ total
);
1205 __free_pages(pages
+ total
, order
);
1209 for (point
= 1; point
< total
; point
++)
1210 set_page_refcounted(&pages
[point
]);
1212 base
= page_address(pages
);
1213 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1214 region
->vm_start
= (unsigned long) base
;
1215 region
->vm_end
= region
->vm_start
+ len
;
1216 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1218 vma
->vm_start
= region
->vm_start
;
1219 vma
->vm_end
= region
->vm_start
+ len
;
1222 /* read the contents of a file into the copy */
1223 mm_segment_t old_fs
;
1226 fpos
= vma
->vm_pgoff
;
1227 fpos
<<= PAGE_SHIFT
;
1231 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, len
, &fpos
);
1237 /* clear the last little bit */
1239 memset(base
+ ret
, 0, len
- ret
);
1246 free_page_series(region
->vm_start
, region
->vm_top
);
1247 region
->vm_start
= vma
->vm_start
= 0;
1248 region
->vm_end
= vma
->vm_end
= 0;
1253 printk("Allocation of length %lu from process %d (%s) failed\n",
1254 len
, current
->pid
, current
->comm
);
1260 * handle mapping creation for uClinux
1262 unsigned long do_mmap_pgoff(struct file
*file
,
1266 unsigned long flags
,
1267 unsigned long pgoff
,
1268 unsigned long *populate
)
1270 struct vm_area_struct
*vma
;
1271 struct vm_region
*region
;
1273 unsigned long capabilities
, vm_flags
, result
;
1276 kenter(",%lx,%lx,%lx,%lx,%lx", addr
, len
, prot
, flags
, pgoff
);
1280 /* decide whether we should attempt the mapping, and if so what sort of
1282 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1285 kleave(" = %d [val]", ret
);
1289 /* we ignore the address hint */
1291 len
= PAGE_ALIGN(len
);
1293 /* we've determined that we can make the mapping, now translate what we
1294 * now know into VMA flags */
1295 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1297 /* we're going to need to record the mapping */
1298 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1300 goto error_getting_region
;
1302 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1304 goto error_getting_vma
;
1306 region
->vm_usage
= 1;
1307 region
->vm_flags
= vm_flags
;
1308 region
->vm_pgoff
= pgoff
;
1310 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1311 vma
->vm_flags
= vm_flags
;
1312 vma
->vm_pgoff
= pgoff
;
1315 region
->vm_file
= get_file(file
);
1316 vma
->vm_file
= get_file(file
);
1319 down_write(&nommu_region_sem
);
1321 /* if we want to share, we need to check for regions created by other
1322 * mmap() calls that overlap with our proposed mapping
1323 * - we can only share with a superset match on most regular files
1324 * - shared mappings on character devices and memory backed files are
1325 * permitted to overlap inexactly as far as we are concerned for in
1326 * these cases, sharing is handled in the driver or filesystem rather
1329 if (vm_flags
& VM_MAYSHARE
) {
1330 struct vm_region
*pregion
;
1331 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1333 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1334 pgend
= pgoff
+ pglen
;
1336 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1337 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1339 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1342 /* search for overlapping mappings on the same file */
1343 if (file_inode(pregion
->vm_file
) !=
1347 if (pregion
->vm_pgoff
>= pgend
)
1350 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1351 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1352 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1353 if (pgoff
>= rpgend
)
1356 /* handle inexactly overlapping matches between
1358 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1359 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1360 /* new mapping is not a subset of the region */
1361 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1362 goto sharing_violation
;
1366 /* we've found a region we can share */
1367 pregion
->vm_usage
++;
1368 vma
->vm_region
= pregion
;
1369 start
= pregion
->vm_start
;
1370 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1371 vma
->vm_start
= start
;
1372 vma
->vm_end
= start
+ len
;
1374 if (pregion
->vm_flags
& VM_MAPPED_COPY
) {
1375 kdebug("share copy");
1376 vma
->vm_flags
|= VM_MAPPED_COPY
;
1378 kdebug("share mmap");
1379 ret
= do_mmap_shared_file(vma
);
1381 vma
->vm_region
= NULL
;
1384 pregion
->vm_usage
--;
1386 goto error_just_free
;
1389 fput(region
->vm_file
);
1390 kmem_cache_free(vm_region_jar
, region
);
1396 /* obtain the address at which to make a shared mapping
1397 * - this is the hook for quasi-memory character devices to
1398 * tell us the location of a shared mapping
1400 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1401 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1403 if (IS_ERR_VALUE(addr
)) {
1406 goto error_just_free
;
1408 /* the driver refused to tell us where to site
1409 * the mapping so we'll have to attempt to copy
1412 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1413 goto error_just_free
;
1415 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1417 vma
->vm_start
= region
->vm_start
= addr
;
1418 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1423 vma
->vm_region
= region
;
1425 /* set up the mapping
1426 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1428 if (file
&& vma
->vm_flags
& VM_SHARED
)
1429 ret
= do_mmap_shared_file(vma
);
1431 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1433 goto error_just_free
;
1434 add_nommu_region(region
);
1436 /* clear anonymous mappings that don't ask for uninitialized data */
1437 if (!vma
->vm_file
&& !(flags
& MAP_UNINITIALIZED
))
1438 memset((void *)region
->vm_start
, 0,
1439 region
->vm_end
- region
->vm_start
);
1441 /* okay... we have a mapping; now we have to register it */
1442 result
= vma
->vm_start
;
1444 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1447 add_vma_to_mm(current
->mm
, vma
);
1449 /* we flush the region from the icache only when the first executable
1450 * mapping of it is made */
1451 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1452 flush_icache_range(region
->vm_start
, region
->vm_end
);
1453 region
->vm_icache_flushed
= true;
1456 up_write(&nommu_region_sem
);
1458 kleave(" = %lx", result
);
1462 up_write(&nommu_region_sem
);
1464 if (region
->vm_file
)
1465 fput(region
->vm_file
);
1466 kmem_cache_free(vm_region_jar
, region
);
1469 kmem_cache_free(vm_area_cachep
, vma
);
1470 kleave(" = %d", ret
);
1474 up_write(&nommu_region_sem
);
1475 printk(KERN_WARNING
"Attempt to share mismatched mappings\n");
1480 kmem_cache_free(vm_region_jar
, region
);
1481 printk(KERN_WARNING
"Allocation of vma for %lu byte allocation"
1482 " from process %d failed\n",
1487 error_getting_region
:
1488 printk(KERN_WARNING
"Allocation of vm region for %lu byte allocation"
1489 " from process %d failed\n",
1495 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1496 unsigned long, prot
, unsigned long, flags
,
1497 unsigned long, fd
, unsigned long, pgoff
)
1499 struct file
*file
= NULL
;
1500 unsigned long retval
= -EBADF
;
1502 audit_mmap_fd(fd
, flags
);
1503 if (!(flags
& MAP_ANONYMOUS
)) {
1509 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1511 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1519 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1520 struct mmap_arg_struct
{
1524 unsigned long flags
;
1526 unsigned long offset
;
1529 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1531 struct mmap_arg_struct a
;
1533 if (copy_from_user(&a
, arg
, sizeof(a
)))
1535 if (a
.offset
& ~PAGE_MASK
)
1538 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1539 a
.offset
>> PAGE_SHIFT
);
1541 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1544 * split a vma into two pieces at address 'addr', a new vma is allocated either
1545 * for the first part or the tail.
1547 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1548 unsigned long addr
, int new_below
)
1550 struct vm_area_struct
*new;
1551 struct vm_region
*region
;
1552 unsigned long npages
;
1556 /* we're only permitted to split anonymous regions (these should have
1557 * only a single usage on the region) */
1561 if (mm
->map_count
>= sysctl_max_map_count
)
1564 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1568 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1570 kmem_cache_free(vm_region_jar
, region
);
1574 /* most fields are the same, copy all, and then fixup */
1576 *region
= *vma
->vm_region
;
1577 new->vm_region
= region
;
1579 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1582 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1584 region
->vm_start
= new->vm_start
= addr
;
1585 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1588 if (new->vm_ops
&& new->vm_ops
->open
)
1589 new->vm_ops
->open(new);
1591 delete_vma_from_mm(vma
);
1592 down_write(&nommu_region_sem
);
1593 delete_nommu_region(vma
->vm_region
);
1595 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1596 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1598 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1599 vma
->vm_region
->vm_top
= addr
;
1601 add_nommu_region(vma
->vm_region
);
1602 add_nommu_region(new->vm_region
);
1603 up_write(&nommu_region_sem
);
1604 add_vma_to_mm(mm
, vma
);
1605 add_vma_to_mm(mm
, new);
1610 * shrink a VMA by removing the specified chunk from either the beginning or
1613 static int shrink_vma(struct mm_struct
*mm
,
1614 struct vm_area_struct
*vma
,
1615 unsigned long from
, unsigned long to
)
1617 struct vm_region
*region
;
1621 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1623 delete_vma_from_mm(vma
);
1624 if (from
> vma
->vm_start
)
1628 add_vma_to_mm(mm
, vma
);
1630 /* cut the backing region down to size */
1631 region
= vma
->vm_region
;
1632 BUG_ON(region
->vm_usage
!= 1);
1634 down_write(&nommu_region_sem
);
1635 delete_nommu_region(region
);
1636 if (from
> region
->vm_start
) {
1637 to
= region
->vm_top
;
1638 region
->vm_top
= region
->vm_end
= from
;
1640 region
->vm_start
= to
;
1642 add_nommu_region(region
);
1643 up_write(&nommu_region_sem
);
1645 free_page_series(from
, to
);
1651 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1652 * VMA, though it need not cover the whole VMA
1654 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1656 struct vm_area_struct
*vma
;
1660 kenter(",%lx,%zx", start
, len
);
1662 len
= PAGE_ALIGN(len
);
1668 /* find the first potentially overlapping VMA */
1669 vma
= find_vma(mm
, start
);
1671 static int limit
= 0;
1674 "munmap of memory not mmapped by process %d"
1675 " (%s): 0x%lx-0x%lx\n",
1676 current
->pid
, current
->comm
,
1677 start
, start
+ len
- 1);
1683 /* we're allowed to split an anonymous VMA but not a file-backed one */
1686 if (start
> vma
->vm_start
) {
1687 kleave(" = -EINVAL [miss]");
1690 if (end
== vma
->vm_end
)
1691 goto erase_whole_vma
;
1694 kleave(" = -EINVAL [split file]");
1697 /* the chunk must be a subset of the VMA found */
1698 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1699 goto erase_whole_vma
;
1700 if (start
< vma
->vm_start
|| end
> vma
->vm_end
) {
1701 kleave(" = -EINVAL [superset]");
1704 if (start
& ~PAGE_MASK
) {
1705 kleave(" = -EINVAL [unaligned start]");
1708 if (end
!= vma
->vm_end
&& end
& ~PAGE_MASK
) {
1709 kleave(" = -EINVAL [unaligned split]");
1712 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1713 ret
= split_vma(mm
, vma
, start
, 1);
1715 kleave(" = %d [split]", ret
);
1719 return shrink_vma(mm
, vma
, start
, end
);
1723 delete_vma_from_mm(vma
);
1724 delete_vma(mm
, vma
);
1728 EXPORT_SYMBOL(do_munmap
);
1730 int vm_munmap(unsigned long addr
, size_t len
)
1732 struct mm_struct
*mm
= current
->mm
;
1735 down_write(&mm
->mmap_sem
);
1736 ret
= do_munmap(mm
, addr
, len
);
1737 up_write(&mm
->mmap_sem
);
1740 EXPORT_SYMBOL(vm_munmap
);
1742 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1744 return vm_munmap(addr
, len
);
1748 * release all the mappings made in a process's VM space
1750 void exit_mmap(struct mm_struct
*mm
)
1752 struct vm_area_struct
*vma
;
1761 while ((vma
= mm
->mmap
)) {
1762 mm
->mmap
= vma
->vm_next
;
1763 delete_vma_from_mm(vma
);
1764 delete_vma(mm
, vma
);
1771 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
1777 * expand (or shrink) an existing mapping, potentially moving it at the same
1778 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1780 * under NOMMU conditions, we only permit changing a mapping's size, and only
1781 * as long as it stays within the region allocated by do_mmap_private() and the
1782 * block is not shareable
1784 * MREMAP_FIXED is not supported under NOMMU conditions
1786 static unsigned long do_mremap(unsigned long addr
,
1787 unsigned long old_len
, unsigned long new_len
,
1788 unsigned long flags
, unsigned long new_addr
)
1790 struct vm_area_struct
*vma
;
1792 /* insanity checks first */
1793 old_len
= PAGE_ALIGN(old_len
);
1794 new_len
= PAGE_ALIGN(new_len
);
1795 if (old_len
== 0 || new_len
== 0)
1796 return (unsigned long) -EINVAL
;
1798 if (addr
& ~PAGE_MASK
)
1801 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1802 return (unsigned long) -EINVAL
;
1804 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1806 return (unsigned long) -EINVAL
;
1808 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1809 return (unsigned long) -EFAULT
;
1811 if (vma
->vm_flags
& VM_MAYSHARE
)
1812 return (unsigned long) -EPERM
;
1814 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1815 return (unsigned long) -ENOMEM
;
1817 /* all checks complete - do it */
1818 vma
->vm_end
= vma
->vm_start
+ new_len
;
1819 return vma
->vm_start
;
1822 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1823 unsigned long, new_len
, unsigned long, flags
,
1824 unsigned long, new_addr
)
1828 down_write(¤t
->mm
->mmap_sem
);
1829 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1830 up_write(¤t
->mm
->mmap_sem
);
1834 struct page
*follow_page_mask(struct vm_area_struct
*vma
,
1835 unsigned long address
, unsigned int flags
,
1836 unsigned int *page_mask
)
1842 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1843 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1845 if (addr
!= (pfn
<< PAGE_SHIFT
))
1848 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1851 EXPORT_SYMBOL(remap_pfn_range
);
1853 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1855 unsigned long pfn
= start
>> PAGE_SHIFT
;
1856 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1858 pfn
+= vma
->vm_pgoff
;
1859 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1861 EXPORT_SYMBOL(vm_iomap_memory
);
1863 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1864 unsigned long pgoff
)
1866 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1868 if (!(vma
->vm_flags
& VM_USERMAP
))
1871 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1872 vma
->vm_end
= vma
->vm_start
+ size
;
1876 EXPORT_SYMBOL(remap_vmalloc_range
);
1878 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1879 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1884 void unmap_mapping_range(struct address_space
*mapping
,
1885 loff_t
const holebegin
, loff_t
const holelen
,
1889 EXPORT_SYMBOL(unmap_mapping_range
);
1892 * Check that a process has enough memory to allocate a new virtual
1893 * mapping. 0 means there is enough memory for the allocation to
1894 * succeed and -ENOMEM implies there is not.
1896 * We currently support three overcommit policies, which are set via the
1897 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1899 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1900 * Additional code 2002 Jul 20 by Robert Love.
1902 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1904 * Note this is a helper function intended to be used by LSMs which
1905 * wish to use this logic.
1907 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1909 unsigned long free
, allowed
, reserve
;
1911 vm_acct_memory(pages
);
1914 * Sometimes we want to use more memory than we have
1916 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1919 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1920 free
= global_page_state(NR_FREE_PAGES
);
1921 free
+= global_page_state(NR_FILE_PAGES
);
1924 * shmem pages shouldn't be counted as free in this
1925 * case, they can't be purged, only swapped out, and
1926 * that won't affect the overall amount of available
1927 * memory in the system.
1929 free
-= global_page_state(NR_SHMEM
);
1931 free
+= get_nr_swap_pages();
1934 * Any slabs which are created with the
1935 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1936 * which are reclaimable, under pressure. The dentry
1937 * cache and most inode caches should fall into this
1939 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1942 * Leave reserved pages. The pages are not for anonymous pages.
1944 if (free
<= totalreserve_pages
)
1947 free
-= totalreserve_pages
;
1950 * Reserve some for root
1953 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1961 allowed
= vm_commit_limit();
1963 * Reserve some 3% for root
1966 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1969 * Don't let a single process grow so big a user can't recover
1972 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1973 allowed
-= min(mm
->total_vm
/ 32, reserve
);
1976 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
1980 vm_unacct_memory(pages
);
1985 int in_gate_area_no_mm(unsigned long addr
)
1990 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1995 EXPORT_SYMBOL(filemap_fault
);
1997 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2001 EXPORT_SYMBOL(filemap_map_pages
);
2003 int generic_file_remap_pages(struct vm_area_struct
*vma
, unsigned long addr
,
2004 unsigned long size
, pgoff_t pgoff
)
2009 EXPORT_SYMBOL(generic_file_remap_pages
);
2011 static int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
2012 unsigned long addr
, void *buf
, int len
, int write
)
2014 struct vm_area_struct
*vma
;
2016 down_read(&mm
->mmap_sem
);
2018 /* the access must start within one of the target process's mappings */
2019 vma
= find_vma(mm
, addr
);
2021 /* don't overrun this mapping */
2022 if (addr
+ len
>= vma
->vm_end
)
2023 len
= vma
->vm_end
- addr
;
2025 /* only read or write mappings where it is permitted */
2026 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
2027 copy_to_user_page(vma
, NULL
, addr
,
2028 (void *) addr
, buf
, len
);
2029 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
2030 copy_from_user_page(vma
, NULL
, addr
,
2031 buf
, (void *) addr
, len
);
2038 up_read(&mm
->mmap_sem
);
2044 * @access_remote_vm - access another process' address space
2045 * @mm: the mm_struct of the target address space
2046 * @addr: start address to access
2047 * @buf: source or destination buffer
2048 * @len: number of bytes to transfer
2049 * @write: whether the access is a write
2051 * The caller must hold a reference on @mm.
2053 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
2054 void *buf
, int len
, int write
)
2056 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, write
);
2060 * Access another process' address space.
2061 * - source/target buffer must be kernel space
2063 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
2065 struct mm_struct
*mm
;
2067 if (addr
+ len
< addr
)
2070 mm
= get_task_mm(tsk
);
2074 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, write
);
2081 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2082 * @inode: The inode to check
2083 * @size: The current filesize of the inode
2084 * @newsize: The proposed filesize of the inode
2086 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2087 * make sure that that any outstanding VMAs aren't broken and then shrink the
2088 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2089 * automatically grant mappings that are too large.
2091 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
2094 struct vm_area_struct
*vma
;
2095 struct vm_region
*region
;
2097 size_t r_size
, r_top
;
2099 low
= newsize
>> PAGE_SHIFT
;
2100 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2102 down_write(&nommu_region_sem
);
2103 mutex_lock(&inode
->i_mapping
->i_mmap_mutex
);
2105 /* search for VMAs that fall within the dead zone */
2106 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
2107 /* found one - only interested if it's shared out of the page
2109 if (vma
->vm_flags
& VM_SHARED
) {
2110 mutex_unlock(&inode
->i_mapping
->i_mmap_mutex
);
2111 up_write(&nommu_region_sem
);
2112 return -ETXTBSY
; /* not quite true, but near enough */
2116 /* reduce any regions that overlap the dead zone - if in existence,
2117 * these will be pointed to by VMAs that don't overlap the dead zone
2119 * we don't check for any regions that start beyond the EOF as there
2122 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
,
2124 if (!(vma
->vm_flags
& VM_SHARED
))
2127 region
= vma
->vm_region
;
2128 r_size
= region
->vm_top
- region
->vm_start
;
2129 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
2131 if (r_top
> newsize
) {
2132 region
->vm_top
-= r_top
- newsize
;
2133 if (region
->vm_end
> region
->vm_top
)
2134 region
->vm_end
= region
->vm_top
;
2138 mutex_unlock(&inode
->i_mapping
->i_mmap_mutex
);
2139 up_write(&nommu_region_sem
);
2144 * Initialise sysctl_user_reserve_kbytes.
2146 * This is intended to prevent a user from starting a single memory hogging
2147 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2150 * The default value is min(3% of free memory, 128MB)
2151 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2153 static int __meminit
init_user_reserve(void)
2155 unsigned long free_kbytes
;
2157 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
2159 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
2162 module_init(init_user_reserve
)
2165 * Initialise sysctl_admin_reserve_kbytes.
2167 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2168 * to log in and kill a memory hogging process.
2170 * Systems with more than 256MB will reserve 8MB, enough to recover
2171 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2172 * only reserve 3% of free pages by default.
2174 static int __meminit
init_admin_reserve(void)
2176 unsigned long free_kbytes
;
2178 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
2180 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
2183 module_init(init_admin_reserve
)