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/vmacache.h>
21 #include <linux/mman.h>
22 #include <linux/swap.h>
23 #include <linux/file.h>
24 #include <linux/highmem.h>
25 #include <linux/pagemap.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/blkdev.h>
29 #include <linux/backing-dev.h>
30 #include <linux/compiler.h>
31 #include <linux/mount.h>
32 #include <linux/personality.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/audit.h>
36 #include <linux/sched/sysctl.h>
37 #include <linux/printk.h>
39 #include <asm/uaccess.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
46 #define kenter(FMT, ...) \
47 printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
48 #define kleave(FMT, ...) \
49 printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
50 #define kdebug(FMT, ...) \
51 printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__)
53 #define kenter(FMT, ...) \
54 no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__)
55 #define kleave(FMT, ...) \
56 no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__)
57 #define kdebug(FMT, ...) \
58 no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__)
63 unsigned long max_mapnr
;
64 unsigned long highest_memmap_pfn
;
65 struct percpu_counter vm_committed_as
;
66 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
67 int sysctl_overcommit_ratio
= 50; /* default is 50% */
68 unsigned long sysctl_overcommit_kbytes __read_mostly
;
69 int sysctl_max_map_count
= DEFAULT_MAX_MAP_COUNT
;
70 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
71 unsigned long sysctl_user_reserve_kbytes __read_mostly
= 1UL << 17; /* 128MB */
72 unsigned long sysctl_admin_reserve_kbytes __read_mostly
= 1UL << 13; /* 8MB */
73 int heap_stack_gap
= 0;
75 atomic_long_t mmap_pages_allocated
;
78 * The global memory commitment made in the system can be a metric
79 * that can be used to drive ballooning decisions when Linux is hosted
80 * as a guest. On Hyper-V, the host implements a policy engine for dynamically
81 * balancing memory across competing virtual machines that are hosted.
82 * Several metrics drive this policy engine including the guest reported
85 unsigned long vm_memory_committed(void)
87 return percpu_counter_read_positive(&vm_committed_as
);
90 EXPORT_SYMBOL_GPL(vm_memory_committed
);
92 EXPORT_SYMBOL(mem_map
);
94 /* list of mapped, potentially shareable regions */
95 static struct kmem_cache
*vm_region_jar
;
96 struct rb_root nommu_region_tree
= RB_ROOT
;
97 DECLARE_RWSEM(nommu_region_sem
);
99 const struct vm_operations_struct generic_file_vm_ops
= {
103 * Return the total memory allocated for this pointer, not
104 * just what the caller asked for.
106 * Doesn't have to be accurate, i.e. may have races.
108 unsigned int kobjsize(const void *objp
)
113 * If the object we have should not have ksize performed on it,
116 if (!objp
|| !virt_addr_valid(objp
))
119 page
= virt_to_head_page(objp
);
122 * If the allocator sets PageSlab, we know the pointer came from
129 * If it's not a compound page, see if we have a matching VMA
130 * region. This test is intentionally done in reverse order,
131 * so if there's no VMA, we still fall through and hand back
132 * PAGE_SIZE for 0-order pages.
134 if (!PageCompound(page
)) {
135 struct vm_area_struct
*vma
;
137 vma
= find_vma(current
->mm
, (unsigned long)objp
);
139 return vma
->vm_end
- vma
->vm_start
;
143 * The ksize() function is only guaranteed to work for pointers
144 * returned by kmalloc(). So handle arbitrary pointers here.
146 return PAGE_SIZE
<< compound_order(page
);
149 long __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
150 unsigned long start
, unsigned long nr_pages
,
151 unsigned int foll_flags
, struct page
**pages
,
152 struct vm_area_struct
**vmas
, int *nonblocking
)
154 struct vm_area_struct
*vma
;
155 unsigned long vm_flags
;
158 /* calculate required read or write permissions.
159 * If FOLL_FORCE is set, we only require the "MAY" flags.
161 vm_flags
= (foll_flags
& FOLL_WRITE
) ?
162 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
163 vm_flags
&= (foll_flags
& FOLL_FORCE
) ?
164 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
166 for (i
= 0; i
< nr_pages
; i
++) {
167 vma
= find_vma(mm
, start
);
169 goto finish_or_fault
;
171 /* protect what we can, including chardevs */
172 if ((vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
173 !(vm_flags
& vma
->vm_flags
))
174 goto finish_or_fault
;
177 pages
[i
] = virt_to_page(start
);
179 page_cache_get(pages
[i
]);
183 start
= (start
+ PAGE_SIZE
) & PAGE_MASK
;
189 return i
? : -EFAULT
;
193 * get a list of pages in an address range belonging to the specified process
194 * and indicate the VMA that covers each page
195 * - this is potentially dodgy as we may end incrementing the page count of a
196 * slab page or a secondary page from a compound page
197 * - don't permit access to VMAs that don't support it, such as I/O mappings
199 long get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
200 unsigned long start
, unsigned long nr_pages
,
201 int write
, int force
, struct page
**pages
,
202 struct vm_area_struct
**vmas
)
211 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
,
214 EXPORT_SYMBOL(get_user_pages
);
217 * follow_pfn - look up PFN at a user virtual address
218 * @vma: memory mapping
219 * @address: user virtual address
220 * @pfn: location to store found PFN
222 * Only IO mappings and raw PFN mappings are allowed.
224 * Returns zero and the pfn at @pfn on success, -ve otherwise.
226 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
229 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
232 *pfn
= address
>> PAGE_SHIFT
;
235 EXPORT_SYMBOL(follow_pfn
);
237 LIST_HEAD(vmap_area_list
);
239 void vfree(const void *addr
)
243 EXPORT_SYMBOL(vfree
);
245 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
, pgprot_t prot
)
248 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
249 * returns only a logical address.
251 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
253 EXPORT_SYMBOL(__vmalloc
);
255 void *vmalloc_user(unsigned long size
)
259 ret
= __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
262 struct vm_area_struct
*vma
;
264 down_write(¤t
->mm
->mmap_sem
);
265 vma
= find_vma(current
->mm
, (unsigned long)ret
);
267 vma
->vm_flags
|= VM_USERMAP
;
268 up_write(¤t
->mm
->mmap_sem
);
273 EXPORT_SYMBOL(vmalloc_user
);
275 struct page
*vmalloc_to_page(const void *addr
)
277 return virt_to_page(addr
);
279 EXPORT_SYMBOL(vmalloc_to_page
);
281 unsigned long vmalloc_to_pfn(const void *addr
)
283 return page_to_pfn(virt_to_page(addr
));
285 EXPORT_SYMBOL(vmalloc_to_pfn
);
287 long vread(char *buf
, char *addr
, unsigned long count
)
289 /* Don't allow overflow */
290 if ((unsigned long) buf
+ count
< count
)
291 count
= -(unsigned long) buf
;
293 memcpy(buf
, addr
, count
);
297 long vwrite(char *buf
, char *addr
, unsigned long count
)
299 /* Don't allow overflow */
300 if ((unsigned long) addr
+ count
< count
)
301 count
= -(unsigned long) addr
;
303 memcpy(addr
, buf
, count
);
308 * vmalloc - allocate virtually continguos memory
310 * @size: allocation size
312 * Allocate enough pages to cover @size from the page level
313 * allocator and map them into continguos kernel virtual space.
315 * For tight control over page level allocator and protection flags
316 * use __vmalloc() instead.
318 void *vmalloc(unsigned long size
)
320 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL
);
322 EXPORT_SYMBOL(vmalloc
);
325 * vzalloc - allocate virtually continguos memory with zero fill
327 * @size: allocation size
329 * Allocate enough pages to cover @size from the page level
330 * allocator and map them into continguos kernel virtual space.
331 * The memory allocated is set to zero.
333 * For tight control over page level allocator and protection flags
334 * use __vmalloc() instead.
336 void *vzalloc(unsigned long size
)
338 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
,
341 EXPORT_SYMBOL(vzalloc
);
344 * vmalloc_node - allocate memory on a specific node
345 * @size: allocation size
348 * Allocate enough pages to cover @size from the page level
349 * allocator and map them into contiguous kernel virtual space.
351 * For tight control over page level allocator and protection flags
352 * use __vmalloc() instead.
354 void *vmalloc_node(unsigned long size
, int node
)
356 return vmalloc(size
);
358 EXPORT_SYMBOL(vmalloc_node
);
361 * vzalloc_node - allocate memory on a specific node with zero fill
362 * @size: allocation size
365 * Allocate enough pages to cover @size from the page level
366 * allocator and map them into contiguous kernel virtual space.
367 * The memory allocated is set to zero.
369 * For tight control over page level allocator and protection flags
370 * use __vmalloc() instead.
372 void *vzalloc_node(unsigned long size
, int node
)
374 return vzalloc(size
);
376 EXPORT_SYMBOL(vzalloc_node
);
378 #ifndef PAGE_KERNEL_EXEC
379 # define PAGE_KERNEL_EXEC PAGE_KERNEL
383 * vmalloc_exec - allocate virtually contiguous, executable memory
384 * @size: allocation size
386 * Kernel-internal function to allocate enough pages to cover @size
387 * the page level allocator and map them into contiguous and
388 * executable kernel virtual space.
390 * For tight control over page level allocator and protection flags
391 * use __vmalloc() instead.
394 void *vmalloc_exec(unsigned long size
)
396 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
, PAGE_KERNEL_EXEC
);
400 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
401 * @size: allocation size
403 * Allocate enough 32bit PA addressable pages to cover @size from the
404 * page level allocator and map them into continguos kernel virtual space.
406 void *vmalloc_32(unsigned long size
)
408 return __vmalloc(size
, GFP_KERNEL
, PAGE_KERNEL
);
410 EXPORT_SYMBOL(vmalloc_32
);
413 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
414 * @size: allocation size
416 * The resulting memory area is 32bit addressable and zeroed so it can be
417 * mapped to userspace without leaking data.
419 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
420 * remap_vmalloc_range() are permissible.
422 void *vmalloc_32_user(unsigned long size
)
425 * We'll have to sort out the ZONE_DMA bits for 64-bit,
426 * but for now this can simply use vmalloc_user() directly.
428 return vmalloc_user(size
);
430 EXPORT_SYMBOL(vmalloc_32_user
);
432 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
439 void vunmap(const void *addr
)
443 EXPORT_SYMBOL(vunmap
);
445 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
, pgprot_t prot
)
450 EXPORT_SYMBOL(vm_map_ram
);
452 void vm_unmap_ram(const void *mem
, unsigned int count
)
456 EXPORT_SYMBOL(vm_unmap_ram
);
458 void vm_unmap_aliases(void)
461 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
464 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
467 void __weak
vmalloc_sync_all(void)
472 * alloc_vm_area - allocate a range of kernel address space
473 * @size: size of the area
475 * Returns: NULL on failure, vm_struct on success
477 * This function reserves a range of kernel address space, and
478 * allocates pagetables to map that range. No actual mappings
479 * are created. If the kernel address space is not shared
480 * between processes, it syncs the pagetable across all
483 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
488 EXPORT_SYMBOL_GPL(alloc_vm_area
);
490 void free_vm_area(struct vm_struct
*area
)
494 EXPORT_SYMBOL_GPL(free_vm_area
);
496 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
501 EXPORT_SYMBOL(vm_insert_page
);
504 * sys_brk() for the most part doesn't need the global kernel
505 * lock, except when an application is doing something nasty
506 * like trying to un-brk an area that has already been mapped
507 * to a regular file. in this case, the unmapping will need
508 * to invoke file system routines that need the global lock.
510 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
512 struct mm_struct
*mm
= current
->mm
;
514 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
521 * Always allow shrinking brk
523 if (brk
<= mm
->brk
) {
529 * Ok, looks good - let it rip.
531 flush_icache_range(mm
->brk
, brk
);
532 return mm
->brk
= brk
;
536 * initialise the VMA and region record slabs
538 void __init
mmap_init(void)
542 ret
= percpu_counter_init(&vm_committed_as
, 0);
544 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
);
548 * validate the region tree
549 * - the caller must hold the region lock
551 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
552 static noinline
void validate_nommu_regions(void)
554 struct vm_region
*region
, *last
;
555 struct rb_node
*p
, *lastp
;
557 lastp
= rb_first(&nommu_region_tree
);
561 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
562 BUG_ON(unlikely(last
->vm_end
<= last
->vm_start
));
563 BUG_ON(unlikely(last
->vm_top
< last
->vm_end
));
565 while ((p
= rb_next(lastp
))) {
566 region
= rb_entry(p
, struct vm_region
, vm_rb
);
567 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
569 BUG_ON(unlikely(region
->vm_end
<= region
->vm_start
));
570 BUG_ON(unlikely(region
->vm_top
< region
->vm_end
));
571 BUG_ON(unlikely(region
->vm_start
< last
->vm_top
));
577 static void validate_nommu_regions(void)
583 * add a region into the global tree
585 static void add_nommu_region(struct vm_region
*region
)
587 struct vm_region
*pregion
;
588 struct rb_node
**p
, *parent
;
590 validate_nommu_regions();
593 p
= &nommu_region_tree
.rb_node
;
596 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
597 if (region
->vm_start
< pregion
->vm_start
)
599 else if (region
->vm_start
> pregion
->vm_start
)
601 else if (pregion
== region
)
607 rb_link_node(®ion
->vm_rb
, parent
, p
);
608 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
610 validate_nommu_regions();
614 * delete a region from the global tree
616 static void delete_nommu_region(struct vm_region
*region
)
618 BUG_ON(!nommu_region_tree
.rb_node
);
620 validate_nommu_regions();
621 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
622 validate_nommu_regions();
626 * free a contiguous series of pages
628 static void free_page_series(unsigned long from
, unsigned long to
)
630 for (; from
< to
; from
+= PAGE_SIZE
) {
631 struct page
*page
= virt_to_page(from
);
633 kdebug("- free %lx", from
);
634 atomic_long_dec(&mmap_pages_allocated
);
635 if (page_count(page
) != 1)
636 kdebug("free page %p: refcount not one: %d",
637 page
, page_count(page
));
643 * release a reference to a region
644 * - the caller must hold the region semaphore for writing, which this releases
645 * - the region may not have been added to the tree yet, in which case vm_top
646 * will equal vm_start
648 static void __put_nommu_region(struct vm_region
*region
)
649 __releases(nommu_region_sem
)
651 kenter("%p{%d}", region
, region
->vm_usage
);
653 BUG_ON(!nommu_region_tree
.rb_node
);
655 if (--region
->vm_usage
== 0) {
656 if (region
->vm_top
> region
->vm_start
)
657 delete_nommu_region(region
);
658 up_write(&nommu_region_sem
);
661 fput(region
->vm_file
);
663 /* IO memory and memory shared directly out of the pagecache
664 * from ramfs/tmpfs mustn't be released here */
665 if (region
->vm_flags
& VM_MAPPED_COPY
) {
666 kdebug("free series");
667 free_page_series(region
->vm_start
, region
->vm_top
);
669 kmem_cache_free(vm_region_jar
, region
);
671 up_write(&nommu_region_sem
);
676 * release a reference to a region
678 static void put_nommu_region(struct vm_region
*region
)
680 down_write(&nommu_region_sem
);
681 __put_nommu_region(region
);
685 * update protection on a vma
687 static void protect_vma(struct vm_area_struct
*vma
, unsigned long flags
)
690 struct mm_struct
*mm
= vma
->vm_mm
;
691 long start
= vma
->vm_start
& PAGE_MASK
;
692 while (start
< vma
->vm_end
) {
693 protect_page(mm
, start
, flags
);
696 update_protections(mm
);
701 * add a VMA into a process's mm_struct in the appropriate place in the list
702 * and tree and add to the address space's page tree also if not an anonymous
704 * - should be called with mm->mmap_sem held writelocked
706 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
708 struct vm_area_struct
*pvma
, *prev
;
709 struct address_space
*mapping
;
710 struct rb_node
**p
, *parent
, *rb_prev
;
714 BUG_ON(!vma
->vm_region
);
719 protect_vma(vma
, vma
->vm_flags
);
721 /* add the VMA to the mapping */
723 mapping
= vma
->vm_file
->f_mapping
;
725 mutex_lock(&mapping
->i_mmap_mutex
);
726 flush_dcache_mmap_lock(mapping
);
727 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
728 flush_dcache_mmap_unlock(mapping
);
729 mutex_unlock(&mapping
->i_mmap_mutex
);
732 /* add the VMA to the tree */
733 parent
= rb_prev
= NULL
;
734 p
= &mm
->mm_rb
.rb_node
;
737 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
739 /* sort by: start addr, end addr, VMA struct addr in that order
740 * (the latter is necessary as we may get identical VMAs) */
741 if (vma
->vm_start
< pvma
->vm_start
)
743 else if (vma
->vm_start
> pvma
->vm_start
) {
746 } else if (vma
->vm_end
< pvma
->vm_end
)
748 else if (vma
->vm_end
> pvma
->vm_end
) {
751 } else if (vma
< pvma
)
753 else if (vma
> pvma
) {
760 rb_link_node(&vma
->vm_rb
, parent
, p
);
761 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
763 /* add VMA to the VMA list also */
766 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
768 __vma_link_list(mm
, vma
, prev
, parent
);
772 * delete a VMA from its owning mm_struct and address space
774 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
777 struct address_space
*mapping
;
778 struct mm_struct
*mm
= vma
->vm_mm
;
779 struct task_struct
*curr
= current
;
786 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
787 /* if the vma is cached, invalidate the entire cache */
788 if (curr
->vmacache
[i
] == vma
) {
789 vmacache_invalidate(mm
);
794 /* remove the VMA from the mapping */
796 mapping
= vma
->vm_file
->f_mapping
;
798 mutex_lock(&mapping
->i_mmap_mutex
);
799 flush_dcache_mmap_lock(mapping
);
800 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
801 flush_dcache_mmap_unlock(mapping
);
802 mutex_unlock(&mapping
->i_mmap_mutex
);
805 /* remove from the MM's tree and list */
806 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
809 vma
->vm_prev
->vm_next
= vma
->vm_next
;
811 mm
->mmap
= vma
->vm_next
;
814 vma
->vm_next
->vm_prev
= vma
->vm_prev
;
818 * destroy a VMA record
820 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
823 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
824 vma
->vm_ops
->close(vma
);
827 put_nommu_region(vma
->vm_region
);
828 kmem_cache_free(vm_area_cachep
, vma
);
832 * look up the first VMA in which addr resides, NULL if none
833 * - should be called with mm->mmap_sem at least held readlocked
835 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
837 struct vm_area_struct
*vma
;
839 /* check the cache first */
840 vma
= vmacache_find(mm
, addr
);
844 /* trawl the list (there may be multiple mappings in which addr
846 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
847 if (vma
->vm_start
> addr
)
849 if (vma
->vm_end
> addr
) {
850 vmacache_update(addr
, vma
);
857 EXPORT_SYMBOL(find_vma
);
861 * - we don't extend stack VMAs under NOMMU conditions
863 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
865 return find_vma(mm
, addr
);
869 * expand a stack to a given address
870 * - not supported under NOMMU conditions
872 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
878 * look up the first VMA exactly that exactly matches addr
879 * - should be called with mm->mmap_sem at least held readlocked
881 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
885 struct vm_area_struct
*vma
;
886 unsigned long end
= addr
+ len
;
888 /* check the cache first */
889 vma
= vmacache_find_exact(mm
, addr
, end
);
893 /* trawl the list (there may be multiple mappings in which addr
895 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
896 if (vma
->vm_start
< addr
)
898 if (vma
->vm_start
> addr
)
900 if (vma
->vm_end
== end
) {
901 vmacache_update(addr
, vma
);
910 * determine whether a mapping should be permitted and, if so, what sort of
911 * mapping we're capable of supporting
913 static int validate_mmap_request(struct file
*file
,
919 unsigned long *_capabilities
)
921 unsigned long capabilities
, rlen
;
924 /* do the simple checks first */
925 if (flags
& MAP_FIXED
) {
927 "%d: Can't do fixed-address/overlay mmap of RAM\n",
932 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
933 (flags
& MAP_TYPE
) != MAP_SHARED
)
939 /* Careful about overflows.. */
940 rlen
= PAGE_ALIGN(len
);
941 if (!rlen
|| rlen
> TASK_SIZE
)
944 /* offset overflow? */
945 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
949 /* validate file mapping requests */
950 struct address_space
*mapping
;
952 /* files must support mmap */
953 if (!file
->f_op
->mmap
)
956 /* work out if what we've got could possibly be shared
957 * - we support chardevs that provide their own "memory"
958 * - we support files/blockdevs that are memory backed
960 mapping
= file
->f_mapping
;
962 mapping
= file_inode(file
)->i_mapping
;
965 if (mapping
&& mapping
->backing_dev_info
)
966 capabilities
= mapping
->backing_dev_info
->capabilities
;
969 /* no explicit capabilities set, so assume some
971 switch (file_inode(file
)->i_mode
& S_IFMT
) {
974 capabilities
= BDI_CAP_MAP_COPY
;
989 /* eliminate any capabilities that we can't support on this
991 if (!file
->f_op
->get_unmapped_area
)
992 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
993 if (!file
->f_op
->read
)
994 capabilities
&= ~BDI_CAP_MAP_COPY
;
996 /* The file shall have been opened with read permission. */
997 if (!(file
->f_mode
& FMODE_READ
))
1000 if (flags
& MAP_SHARED
) {
1001 /* do checks for writing, appending and locking */
1002 if ((prot
& PROT_WRITE
) &&
1003 !(file
->f_mode
& FMODE_WRITE
))
1006 if (IS_APPEND(file_inode(file
)) &&
1007 (file
->f_mode
& FMODE_WRITE
))
1010 if (locks_verify_locked(file
))
1013 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1016 /* we mustn't privatise shared mappings */
1017 capabilities
&= ~BDI_CAP_MAP_COPY
;
1019 /* we're going to read the file into private memory we
1021 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1024 /* we don't permit a private writable mapping to be
1025 * shared with the backing device */
1026 if (prot
& PROT_WRITE
)
1027 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1030 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1031 if (((prot
& PROT_READ
) && !(capabilities
& BDI_CAP_READ_MAP
)) ||
1032 ((prot
& PROT_WRITE
) && !(capabilities
& BDI_CAP_WRITE_MAP
)) ||
1033 ((prot
& PROT_EXEC
) && !(capabilities
& BDI_CAP_EXEC_MAP
))
1035 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1036 if (flags
& MAP_SHARED
) {
1038 "MAP_SHARED not completely supported on !MMU\n");
1044 /* handle executable mappings and implied executable
1046 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1047 if (prot
& PROT_EXEC
)
1049 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
1050 /* handle implication of PROT_EXEC by PROT_READ */
1051 if (current
->personality
& READ_IMPLIES_EXEC
) {
1052 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
;
1063 /* anonymous mappings are always memory backed and can be
1066 capabilities
= BDI_CAP_MAP_COPY
;
1068 /* handle PROT_EXEC implication by PROT_READ */
1069 if ((prot
& PROT_READ
) &&
1070 (current
->personality
& READ_IMPLIES_EXEC
))
1074 /* allow the security API to have its say */
1075 ret
= security_mmap_addr(addr
);
1080 *_capabilities
= capabilities
;
1085 * we've determined that we can make the mapping, now translate what we
1086 * now know into VMA flags
1088 static unsigned long determine_vm_flags(struct file
*file
,
1090 unsigned long flags
,
1091 unsigned long capabilities
)
1093 unsigned long vm_flags
;
1095 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
);
1096 /* vm_flags |= mm->def_flags; */
1098 if (!(capabilities
& BDI_CAP_MAP_DIRECT
)) {
1099 /* attempt to share read-only copies of mapped file chunks */
1100 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1101 if (file
&& !(prot
& PROT_WRITE
))
1102 vm_flags
|= VM_MAYSHARE
;
1104 /* overlay a shareable mapping on the backing device or inode
1105 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
1107 vm_flags
|= VM_MAYSHARE
| (capabilities
& BDI_CAP_VMFLAGS
);
1108 if (flags
& MAP_SHARED
)
1109 vm_flags
|= VM_SHARED
;
1112 /* refuse to let anyone share private mappings with this process if
1113 * it's being traced - otherwise breakpoints set in it may interfere
1114 * with another untraced process
1116 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
1117 vm_flags
&= ~VM_MAYSHARE
;
1123 * set up a shared mapping on a file (the driver or filesystem provides and
1126 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
1130 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1132 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1138 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1139 * opposed to tried but failed) so we can only give a suitable error as
1140 * it's not possible to make a private copy if MAP_SHARED was given */
1145 * set up a private mapping or an anonymous shared mapping
1147 static int do_mmap_private(struct vm_area_struct
*vma
,
1148 struct vm_region
*region
,
1150 unsigned long capabilities
)
1153 unsigned long total
, point
, n
;
1157 /* invoke the file's mapping function so that it can keep track of
1158 * shared mappings on devices or memory
1159 * - VM_MAYSHARE will be set if it may attempt to share
1161 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1162 ret
= vma
->vm_file
->f_op
->mmap(vma
->vm_file
, vma
);
1164 /* shouldn't return success if we're not sharing */
1165 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1166 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1172 /* getting an ENOSYS error indicates that direct mmap isn't
1173 * possible (as opposed to tried but failed) so we'll try to
1174 * make a private copy of the data and map that instead */
1178 /* allocate some memory to hold the mapping
1179 * - note that this may not return a page-aligned address if the object
1180 * we're allocating is smaller than a page
1182 order
= get_order(len
);
1183 kdebug("alloc order %d for %lx", order
, len
);
1185 pages
= alloc_pages(GFP_KERNEL
, order
);
1190 atomic_long_add(total
, &mmap_pages_allocated
);
1192 point
= len
>> PAGE_SHIFT
;
1194 /* we allocated a power-of-2 sized page set, so we may want to trim off
1196 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
) {
1197 while (total
> point
) {
1198 order
= ilog2(total
- point
);
1200 kdebug("shave %lu/%lu @%lu", n
, total
- point
, total
);
1201 atomic_long_sub(n
, &mmap_pages_allocated
);
1203 set_page_refcounted(pages
+ total
);
1204 __free_pages(pages
+ total
, order
);
1208 for (point
= 1; point
< total
; point
++)
1209 set_page_refcounted(&pages
[point
]);
1211 base
= page_address(pages
);
1212 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1213 region
->vm_start
= (unsigned long) base
;
1214 region
->vm_end
= region
->vm_start
+ len
;
1215 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1217 vma
->vm_start
= region
->vm_start
;
1218 vma
->vm_end
= region
->vm_start
+ len
;
1221 /* read the contents of a file into the copy */
1222 mm_segment_t old_fs
;
1225 fpos
= vma
->vm_pgoff
;
1226 fpos
<<= PAGE_SHIFT
;
1230 ret
= vma
->vm_file
->f_op
->read(vma
->vm_file
, base
, len
, &fpos
);
1236 /* clear the last little bit */
1238 memset(base
+ ret
, 0, len
- ret
);
1245 free_page_series(region
->vm_start
, region
->vm_top
);
1246 region
->vm_start
= vma
->vm_start
= 0;
1247 region
->vm_end
= vma
->vm_end
= 0;
1252 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1253 len
, current
->pid
, current
->comm
);
1259 * handle mapping creation for uClinux
1261 unsigned long do_mmap_pgoff(struct file
*file
,
1265 unsigned long flags
,
1266 unsigned long pgoff
,
1267 unsigned long *populate
)
1269 struct vm_area_struct
*vma
;
1270 struct vm_region
*region
;
1272 unsigned long capabilities
, vm_flags
, result
;
1275 kenter(",%lx,%lx,%lx,%lx,%lx", addr
, len
, prot
, flags
, pgoff
);
1279 /* decide whether we should attempt the mapping, and if so what sort of
1281 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1284 kleave(" = %d [val]", ret
);
1288 /* we ignore the address hint */
1290 len
= PAGE_ALIGN(len
);
1292 /* we've determined that we can make the mapping, now translate what we
1293 * now know into VMA flags */
1294 vm_flags
= determine_vm_flags(file
, prot
, flags
, capabilities
);
1296 /* we're going to need to record the mapping */
1297 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1299 goto error_getting_region
;
1301 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1303 goto error_getting_vma
;
1305 region
->vm_usage
= 1;
1306 region
->vm_flags
= vm_flags
;
1307 region
->vm_pgoff
= pgoff
;
1309 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1310 vma
->vm_flags
= vm_flags
;
1311 vma
->vm_pgoff
= pgoff
;
1314 region
->vm_file
= get_file(file
);
1315 vma
->vm_file
= get_file(file
);
1318 down_write(&nommu_region_sem
);
1320 /* if we want to share, we need to check for regions created by other
1321 * mmap() calls that overlap with our proposed mapping
1322 * - we can only share with a superset match on most regular files
1323 * - shared mappings on character devices and memory backed files are
1324 * permitted to overlap inexactly as far as we are concerned for in
1325 * these cases, sharing is handled in the driver or filesystem rather
1328 if (vm_flags
& VM_MAYSHARE
) {
1329 struct vm_region
*pregion
;
1330 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1332 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1333 pgend
= pgoff
+ pglen
;
1335 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1336 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1338 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1341 /* search for overlapping mappings on the same file */
1342 if (file_inode(pregion
->vm_file
) !=
1346 if (pregion
->vm_pgoff
>= pgend
)
1349 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1350 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1351 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1352 if (pgoff
>= rpgend
)
1355 /* handle inexactly overlapping matches between
1357 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1358 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1359 /* new mapping is not a subset of the region */
1360 if (!(capabilities
& BDI_CAP_MAP_DIRECT
))
1361 goto sharing_violation
;
1365 /* we've found a region we can share */
1366 pregion
->vm_usage
++;
1367 vma
->vm_region
= pregion
;
1368 start
= pregion
->vm_start
;
1369 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1370 vma
->vm_start
= start
;
1371 vma
->vm_end
= start
+ len
;
1373 if (pregion
->vm_flags
& VM_MAPPED_COPY
) {
1374 kdebug("share copy");
1375 vma
->vm_flags
|= VM_MAPPED_COPY
;
1377 kdebug("share mmap");
1378 ret
= do_mmap_shared_file(vma
);
1380 vma
->vm_region
= NULL
;
1383 pregion
->vm_usage
--;
1385 goto error_just_free
;
1388 fput(region
->vm_file
);
1389 kmem_cache_free(vm_region_jar
, region
);
1395 /* obtain the address at which to make a shared mapping
1396 * - this is the hook for quasi-memory character devices to
1397 * tell us the location of a shared mapping
1399 if (capabilities
& BDI_CAP_MAP_DIRECT
) {
1400 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1402 if (IS_ERR_VALUE(addr
)) {
1405 goto error_just_free
;
1407 /* the driver refused to tell us where to site
1408 * the mapping so we'll have to attempt to copy
1411 if (!(capabilities
& BDI_CAP_MAP_COPY
))
1412 goto error_just_free
;
1414 capabilities
&= ~BDI_CAP_MAP_DIRECT
;
1416 vma
->vm_start
= region
->vm_start
= addr
;
1417 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1422 vma
->vm_region
= region
;
1424 /* set up the mapping
1425 * - the region is filled in if BDI_CAP_MAP_DIRECT is still set
1427 if (file
&& vma
->vm_flags
& VM_SHARED
)
1428 ret
= do_mmap_shared_file(vma
);
1430 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1432 goto error_just_free
;
1433 add_nommu_region(region
);
1435 /* clear anonymous mappings that don't ask for uninitialized data */
1436 if (!vma
->vm_file
&& !(flags
& MAP_UNINITIALIZED
))
1437 memset((void *)region
->vm_start
, 0,
1438 region
->vm_end
- region
->vm_start
);
1440 /* okay... we have a mapping; now we have to register it */
1441 result
= vma
->vm_start
;
1443 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1446 add_vma_to_mm(current
->mm
, vma
);
1448 /* we flush the region from the icache only when the first executable
1449 * mapping of it is made */
1450 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1451 flush_icache_range(region
->vm_start
, region
->vm_end
);
1452 region
->vm_icache_flushed
= true;
1455 up_write(&nommu_region_sem
);
1457 kleave(" = %lx", result
);
1461 up_write(&nommu_region_sem
);
1463 if (region
->vm_file
)
1464 fput(region
->vm_file
);
1465 kmem_cache_free(vm_region_jar
, region
);
1468 kmem_cache_free(vm_area_cachep
, vma
);
1469 kleave(" = %d", ret
);
1473 up_write(&nommu_region_sem
);
1474 printk(KERN_WARNING
"Attempt to share mismatched mappings\n");
1479 kmem_cache_free(vm_region_jar
, region
);
1480 printk(KERN_WARNING
"Allocation of vma for %lu byte allocation"
1481 " from process %d failed\n",
1486 error_getting_region
:
1487 printk(KERN_WARNING
"Allocation of vm region for %lu byte allocation"
1488 " from process %d failed\n",
1494 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1495 unsigned long, prot
, unsigned long, flags
,
1496 unsigned long, fd
, unsigned long, pgoff
)
1498 struct file
*file
= NULL
;
1499 unsigned long retval
= -EBADF
;
1501 audit_mmap_fd(fd
, flags
);
1502 if (!(flags
& MAP_ANONYMOUS
)) {
1508 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1510 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1518 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1519 struct mmap_arg_struct
{
1523 unsigned long flags
;
1525 unsigned long offset
;
1528 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1530 struct mmap_arg_struct a
;
1532 if (copy_from_user(&a
, arg
, sizeof(a
)))
1534 if (a
.offset
& ~PAGE_MASK
)
1537 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1538 a
.offset
>> PAGE_SHIFT
);
1540 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1543 * split a vma into two pieces at address 'addr', a new vma is allocated either
1544 * for the first part or the tail.
1546 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1547 unsigned long addr
, int new_below
)
1549 struct vm_area_struct
*new;
1550 struct vm_region
*region
;
1551 unsigned long npages
;
1555 /* we're only permitted to split anonymous regions (these should have
1556 * only a single usage on the region) */
1560 if (mm
->map_count
>= sysctl_max_map_count
)
1563 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1567 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1569 kmem_cache_free(vm_region_jar
, region
);
1573 /* most fields are the same, copy all, and then fixup */
1575 *region
= *vma
->vm_region
;
1576 new->vm_region
= region
;
1578 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1581 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1583 region
->vm_start
= new->vm_start
= addr
;
1584 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1587 if (new->vm_ops
&& new->vm_ops
->open
)
1588 new->vm_ops
->open(new);
1590 delete_vma_from_mm(vma
);
1591 down_write(&nommu_region_sem
);
1592 delete_nommu_region(vma
->vm_region
);
1594 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1595 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1597 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1598 vma
->vm_region
->vm_top
= addr
;
1600 add_nommu_region(vma
->vm_region
);
1601 add_nommu_region(new->vm_region
);
1602 up_write(&nommu_region_sem
);
1603 add_vma_to_mm(mm
, vma
);
1604 add_vma_to_mm(mm
, new);
1609 * shrink a VMA by removing the specified chunk from either the beginning or
1612 static int shrink_vma(struct mm_struct
*mm
,
1613 struct vm_area_struct
*vma
,
1614 unsigned long from
, unsigned long to
)
1616 struct vm_region
*region
;
1620 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1622 delete_vma_from_mm(vma
);
1623 if (from
> vma
->vm_start
)
1627 add_vma_to_mm(mm
, vma
);
1629 /* cut the backing region down to size */
1630 region
= vma
->vm_region
;
1631 BUG_ON(region
->vm_usage
!= 1);
1633 down_write(&nommu_region_sem
);
1634 delete_nommu_region(region
);
1635 if (from
> region
->vm_start
) {
1636 to
= region
->vm_top
;
1637 region
->vm_top
= region
->vm_end
= from
;
1639 region
->vm_start
= to
;
1641 add_nommu_region(region
);
1642 up_write(&nommu_region_sem
);
1644 free_page_series(from
, to
);
1650 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1651 * VMA, though it need not cover the whole VMA
1653 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1655 struct vm_area_struct
*vma
;
1659 kenter(",%lx,%zx", start
, len
);
1661 len
= PAGE_ALIGN(len
);
1667 /* find the first potentially overlapping VMA */
1668 vma
= find_vma(mm
, start
);
1673 "munmap of memory not mmapped by process %d"
1674 " (%s): 0x%lx-0x%lx\n",
1675 current
->pid
, current
->comm
,
1676 start
, start
+ len
- 1);
1682 /* we're allowed to split an anonymous VMA but not a file-backed one */
1685 if (start
> vma
->vm_start
) {
1686 kleave(" = -EINVAL [miss]");
1689 if (end
== vma
->vm_end
)
1690 goto erase_whole_vma
;
1693 kleave(" = -EINVAL [split file]");
1696 /* the chunk must be a subset of the VMA found */
1697 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1698 goto erase_whole_vma
;
1699 if (start
< vma
->vm_start
|| end
> vma
->vm_end
) {
1700 kleave(" = -EINVAL [superset]");
1703 if (start
& ~PAGE_MASK
) {
1704 kleave(" = -EINVAL [unaligned start]");
1707 if (end
!= vma
->vm_end
&& end
& ~PAGE_MASK
) {
1708 kleave(" = -EINVAL [unaligned split]");
1711 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1712 ret
= split_vma(mm
, vma
, start
, 1);
1714 kleave(" = %d [split]", ret
);
1718 return shrink_vma(mm
, vma
, start
, end
);
1722 delete_vma_from_mm(vma
);
1723 delete_vma(mm
, vma
);
1727 EXPORT_SYMBOL(do_munmap
);
1729 int vm_munmap(unsigned long addr
, size_t len
)
1731 struct mm_struct
*mm
= current
->mm
;
1734 down_write(&mm
->mmap_sem
);
1735 ret
= do_munmap(mm
, addr
, len
);
1736 up_write(&mm
->mmap_sem
);
1739 EXPORT_SYMBOL(vm_munmap
);
1741 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1743 return vm_munmap(addr
, len
);
1747 * release all the mappings made in a process's VM space
1749 void exit_mmap(struct mm_struct
*mm
)
1751 struct vm_area_struct
*vma
;
1760 while ((vma
= mm
->mmap
)) {
1761 mm
->mmap
= vma
->vm_next
;
1762 delete_vma_from_mm(vma
);
1763 delete_vma(mm
, vma
);
1770 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
1776 * expand (or shrink) an existing mapping, potentially moving it at the same
1777 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1779 * under NOMMU conditions, we only permit changing a mapping's size, and only
1780 * as long as it stays within the region allocated by do_mmap_private() and the
1781 * block is not shareable
1783 * MREMAP_FIXED is not supported under NOMMU conditions
1785 static unsigned long do_mremap(unsigned long addr
,
1786 unsigned long old_len
, unsigned long new_len
,
1787 unsigned long flags
, unsigned long new_addr
)
1789 struct vm_area_struct
*vma
;
1791 /* insanity checks first */
1792 old_len
= PAGE_ALIGN(old_len
);
1793 new_len
= PAGE_ALIGN(new_len
);
1794 if (old_len
== 0 || new_len
== 0)
1795 return (unsigned long) -EINVAL
;
1797 if (addr
& ~PAGE_MASK
)
1800 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1801 return (unsigned long) -EINVAL
;
1803 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1805 return (unsigned long) -EINVAL
;
1807 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1808 return (unsigned long) -EFAULT
;
1810 if (vma
->vm_flags
& VM_MAYSHARE
)
1811 return (unsigned long) -EPERM
;
1813 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1814 return (unsigned long) -ENOMEM
;
1816 /* all checks complete - do it */
1817 vma
->vm_end
= vma
->vm_start
+ new_len
;
1818 return vma
->vm_start
;
1821 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1822 unsigned long, new_len
, unsigned long, flags
,
1823 unsigned long, new_addr
)
1827 down_write(¤t
->mm
->mmap_sem
);
1828 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1829 up_write(¤t
->mm
->mmap_sem
);
1833 struct page
*follow_page_mask(struct vm_area_struct
*vma
,
1834 unsigned long address
, unsigned int flags
,
1835 unsigned int *page_mask
)
1841 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1842 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1844 if (addr
!= (pfn
<< PAGE_SHIFT
))
1847 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1850 EXPORT_SYMBOL(remap_pfn_range
);
1852 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1854 unsigned long pfn
= start
>> PAGE_SHIFT
;
1855 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1857 pfn
+= vma
->vm_pgoff
;
1858 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1860 EXPORT_SYMBOL(vm_iomap_memory
);
1862 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1863 unsigned long pgoff
)
1865 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1867 if (!(vma
->vm_flags
& VM_USERMAP
))
1870 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1871 vma
->vm_end
= vma
->vm_start
+ size
;
1875 EXPORT_SYMBOL(remap_vmalloc_range
);
1877 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1878 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1883 void unmap_mapping_range(struct address_space
*mapping
,
1884 loff_t
const holebegin
, loff_t
const holelen
,
1888 EXPORT_SYMBOL(unmap_mapping_range
);
1891 * Check that a process has enough memory to allocate a new virtual
1892 * mapping. 0 means there is enough memory for the allocation to
1893 * succeed and -ENOMEM implies there is not.
1895 * We currently support three overcommit policies, which are set via the
1896 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
1898 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
1899 * Additional code 2002 Jul 20 by Robert Love.
1901 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
1903 * Note this is a helper function intended to be used by LSMs which
1904 * wish to use this logic.
1906 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
1908 unsigned long free
, allowed
, reserve
;
1910 vm_acct_memory(pages
);
1913 * Sometimes we want to use more memory than we have
1915 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
1918 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
1919 free
= global_page_state(NR_FREE_PAGES
);
1920 free
+= global_page_state(NR_FILE_PAGES
);
1923 * shmem pages shouldn't be counted as free in this
1924 * case, they can't be purged, only swapped out, and
1925 * that won't affect the overall amount of available
1926 * memory in the system.
1928 free
-= global_page_state(NR_SHMEM
);
1930 free
+= get_nr_swap_pages();
1933 * Any slabs which are created with the
1934 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
1935 * which are reclaimable, under pressure. The dentry
1936 * cache and most inode caches should fall into this
1938 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
1941 * Leave reserved pages. The pages are not for anonymous pages.
1943 if (free
<= totalreserve_pages
)
1946 free
-= totalreserve_pages
;
1949 * Reserve some for root
1952 free
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1960 allowed
= vm_commit_limit();
1962 * Reserve some 3% for root
1965 allowed
-= sysctl_admin_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1968 * Don't let a single process grow so big a user can't recover
1971 reserve
= sysctl_user_reserve_kbytes
>> (PAGE_SHIFT
- 10);
1972 allowed
-= min(mm
->total_vm
/ 32, reserve
);
1975 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
1979 vm_unacct_memory(pages
);
1984 int in_gate_area_no_mm(unsigned long addr
)
1989 int filemap_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1994 EXPORT_SYMBOL(filemap_fault
);
1996 void filemap_map_pages(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2000 EXPORT_SYMBOL(filemap_map_pages
);
2002 int generic_file_remap_pages(struct vm_area_struct
*vma
, unsigned long addr
,
2003 unsigned long size
, pgoff_t pgoff
)
2008 EXPORT_SYMBOL(generic_file_remap_pages
);
2010 static int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
2011 unsigned long addr
, void *buf
, int len
, int write
)
2013 struct vm_area_struct
*vma
;
2015 down_read(&mm
->mmap_sem
);
2017 /* the access must start within one of the target process's mappings */
2018 vma
= find_vma(mm
, addr
);
2020 /* don't overrun this mapping */
2021 if (addr
+ len
>= vma
->vm_end
)
2022 len
= vma
->vm_end
- addr
;
2024 /* only read or write mappings where it is permitted */
2025 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
2026 copy_to_user_page(vma
, NULL
, addr
,
2027 (void *) addr
, buf
, len
);
2028 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
2029 copy_from_user_page(vma
, NULL
, addr
,
2030 buf
, (void *) addr
, len
);
2037 up_read(&mm
->mmap_sem
);
2043 * @access_remote_vm - access another process' address space
2044 * @mm: the mm_struct of the target address space
2045 * @addr: start address to access
2046 * @buf: source or destination buffer
2047 * @len: number of bytes to transfer
2048 * @write: whether the access is a write
2050 * The caller must hold a reference on @mm.
2052 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
2053 void *buf
, int len
, int write
)
2055 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, write
);
2059 * Access another process' address space.
2060 * - source/target buffer must be kernel space
2062 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
2064 struct mm_struct
*mm
;
2066 if (addr
+ len
< addr
)
2069 mm
= get_task_mm(tsk
);
2073 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, write
);
2080 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
2081 * @inode: The inode to check
2082 * @size: The current filesize of the inode
2083 * @newsize: The proposed filesize of the inode
2085 * Check the shared mappings on an inode on behalf of a shrinking truncate to
2086 * make sure that that any outstanding VMAs aren't broken and then shrink the
2087 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
2088 * automatically grant mappings that are too large.
2090 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
2093 struct vm_area_struct
*vma
;
2094 struct vm_region
*region
;
2096 size_t r_size
, r_top
;
2098 low
= newsize
>> PAGE_SHIFT
;
2099 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2101 down_write(&nommu_region_sem
);
2102 mutex_lock(&inode
->i_mapping
->i_mmap_mutex
);
2104 /* search for VMAs that fall within the dead zone */
2105 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
2106 /* found one - only interested if it's shared out of the page
2108 if (vma
->vm_flags
& VM_SHARED
) {
2109 mutex_unlock(&inode
->i_mapping
->i_mmap_mutex
);
2110 up_write(&nommu_region_sem
);
2111 return -ETXTBSY
; /* not quite true, but near enough */
2115 /* reduce any regions that overlap the dead zone - if in existence,
2116 * these will be pointed to by VMAs that don't overlap the dead zone
2118 * we don't check for any regions that start beyond the EOF as there
2121 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
,
2123 if (!(vma
->vm_flags
& VM_SHARED
))
2126 region
= vma
->vm_region
;
2127 r_size
= region
->vm_top
- region
->vm_start
;
2128 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
2130 if (r_top
> newsize
) {
2131 region
->vm_top
-= r_top
- newsize
;
2132 if (region
->vm_end
> region
->vm_top
)
2133 region
->vm_end
= region
->vm_top
;
2137 mutex_unlock(&inode
->i_mapping
->i_mmap_mutex
);
2138 up_write(&nommu_region_sem
);
2143 * Initialise sysctl_user_reserve_kbytes.
2145 * This is intended to prevent a user from starting a single memory hogging
2146 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
2149 * The default value is min(3% of free memory, 128MB)
2150 * 128MB is enough to recover with sshd/login, bash, and top/kill.
2152 static int __meminit
init_user_reserve(void)
2154 unsigned long free_kbytes
;
2156 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
2158 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
2161 module_init(init_user_reserve
)
2164 * Initialise sysctl_admin_reserve_kbytes.
2166 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
2167 * to log in and kill a memory hogging process.
2169 * Systems with more than 256MB will reserve 8MB, enough to recover
2170 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
2171 * only reserve 3% of free pages by default.
2173 static int __meminit
init_admin_reserve(void)
2175 unsigned long free_kbytes
;
2177 free_kbytes
= global_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
2179 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
2182 module_init(init_admin_reserve
)