1 // SPDX-License-Identifier: GPL-2.0-only
5 * Replacement code for mm functions to support CPU's that don't
6 * have any form of memory management unit (thus no virtual memory).
8 * See Documentation/nommu-mmap.txt
10 * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com>
11 * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
12 * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
13 * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com>
14 * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org>
17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19 #include <linux/export.h>
21 #include <linux/sched/mm.h>
22 #include <linux/vmacache.h>
23 #include <linux/mman.h>
24 #include <linux/swap.h>
25 #include <linux/file.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/slab.h>
29 #include <linux/vmalloc.h>
30 #include <linux/blkdev.h>
31 #include <linux/backing-dev.h>
32 #include <linux/compiler.h>
33 #include <linux/mount.h>
34 #include <linux/personality.h>
35 #include <linux/security.h>
36 #include <linux/syscalls.h>
37 #include <linux/audit.h>
38 #include <linux/printk.h>
40 #include <linux/uaccess.h>
42 #include <asm/tlbflush.h>
43 #include <asm/mmu_context.h>
47 EXPORT_SYMBOL(high_memory
);
49 unsigned long max_mapnr
;
50 EXPORT_SYMBOL(max_mapnr
);
51 unsigned long highest_memmap_pfn
;
52 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
53 int heap_stack_gap
= 0;
55 atomic_long_t mmap_pages_allocated
;
57 EXPORT_SYMBOL(mem_map
);
59 /* list of mapped, potentially shareable regions */
60 static struct kmem_cache
*vm_region_jar
;
61 struct rb_root nommu_region_tree
= RB_ROOT
;
62 DECLARE_RWSEM(nommu_region_sem
);
64 const struct vm_operations_struct generic_file_vm_ops
= {
68 * Return the total memory allocated for this pointer, not
69 * just what the caller asked for.
71 * Doesn't have to be accurate, i.e. may have races.
73 unsigned int kobjsize(const void *objp
)
78 * If the object we have should not have ksize performed on it,
81 if (!objp
|| !virt_addr_valid(objp
))
84 page
= virt_to_head_page(objp
);
87 * If the allocator sets PageSlab, we know the pointer came from
94 * If it's not a compound page, see if we have a matching VMA
95 * region. This test is intentionally done in reverse order,
96 * so if there's no VMA, we still fall through and hand back
97 * PAGE_SIZE for 0-order pages.
99 if (!PageCompound(page
)) {
100 struct vm_area_struct
*vma
;
102 vma
= find_vma(current
->mm
, (unsigned long)objp
);
104 return vma
->vm_end
- vma
->vm_start
;
108 * The ksize() function is only guaranteed to work for pointers
109 * returned by kmalloc(). So handle arbitrary pointers here.
111 return page_size(page
);
115 * follow_pfn - look up PFN at a user virtual address
116 * @vma: memory mapping
117 * @address: user virtual address
118 * @pfn: location to store found PFN
120 * Only IO mappings and raw PFN mappings are allowed.
122 * Returns zero and the pfn at @pfn on success, -ve otherwise.
124 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
127 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
130 *pfn
= address
>> PAGE_SHIFT
;
133 EXPORT_SYMBOL(follow_pfn
);
135 LIST_HEAD(vmap_area_list
);
137 void vfree(const void *addr
)
141 EXPORT_SYMBOL(vfree
);
143 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
)
146 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
147 * returns only a logical address.
149 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
151 EXPORT_SYMBOL(__vmalloc
);
153 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
154 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
155 pgprot_t prot
, unsigned long vm_flags
, int node
,
158 return __vmalloc(size
, gfp_mask
);
161 void *__vmalloc_node(unsigned long size
, unsigned long align
, gfp_t gfp_mask
,
162 int node
, const void *caller
)
164 return __vmalloc(size
, gfp_mask
);
167 static void *__vmalloc_user_flags(unsigned long size
, gfp_t flags
)
171 ret
= __vmalloc(size
, flags
);
173 struct vm_area_struct
*vma
;
175 down_write(¤t
->mm
->mmap_sem
);
176 vma
= find_vma(current
->mm
, (unsigned long)ret
);
178 vma
->vm_flags
|= VM_USERMAP
;
179 up_write(¤t
->mm
->mmap_sem
);
185 void *vmalloc_user(unsigned long size
)
187 return __vmalloc_user_flags(size
, GFP_KERNEL
| __GFP_ZERO
);
189 EXPORT_SYMBOL(vmalloc_user
);
191 struct page
*vmalloc_to_page(const void *addr
)
193 return virt_to_page(addr
);
195 EXPORT_SYMBOL(vmalloc_to_page
);
197 unsigned long vmalloc_to_pfn(const void *addr
)
199 return page_to_pfn(virt_to_page(addr
));
201 EXPORT_SYMBOL(vmalloc_to_pfn
);
203 long vread(char *buf
, char *addr
, unsigned long count
)
205 /* Don't allow overflow */
206 if ((unsigned long) buf
+ count
< count
)
207 count
= -(unsigned long) buf
;
209 memcpy(buf
, addr
, count
);
213 long vwrite(char *buf
, char *addr
, unsigned long count
)
215 /* Don't allow overflow */
216 if ((unsigned long) addr
+ count
< count
)
217 count
= -(unsigned long) addr
;
219 memcpy(addr
, buf
, count
);
224 * vmalloc - allocate virtually contiguous memory
226 * @size: allocation size
228 * Allocate enough pages to cover @size from the page level
229 * allocator and map them into contiguous kernel virtual space.
231 * For tight control over page level allocator and protection flags
232 * use __vmalloc() instead.
234 void *vmalloc(unsigned long size
)
236 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
);
238 EXPORT_SYMBOL(vmalloc
);
241 * vzalloc - allocate virtually contiguous memory with zero fill
243 * @size: allocation size
245 * Allocate enough pages to cover @size from the page level
246 * allocator and map them into contiguous kernel virtual space.
247 * The memory allocated is set to zero.
249 * For tight control over page level allocator and protection flags
250 * use __vmalloc() instead.
252 void *vzalloc(unsigned long size
)
254 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
| __GFP_ZERO
);
256 EXPORT_SYMBOL(vzalloc
);
259 * vmalloc_node - allocate memory on a specific node
260 * @size: allocation size
263 * Allocate enough pages to cover @size from the page level
264 * allocator and map them into contiguous kernel virtual space.
266 * For tight control over page level allocator and protection flags
267 * use __vmalloc() instead.
269 void *vmalloc_node(unsigned long size
, int node
)
271 return vmalloc(size
);
273 EXPORT_SYMBOL(vmalloc_node
);
276 * vzalloc_node - allocate memory on a specific node with zero fill
277 * @size: allocation size
280 * Allocate enough pages to cover @size from the page level
281 * allocator and map them into contiguous kernel virtual space.
282 * The memory allocated is set to zero.
284 * For tight control over page level allocator and protection flags
285 * use __vmalloc() instead.
287 void *vzalloc_node(unsigned long size
, int node
)
289 return vzalloc(size
);
291 EXPORT_SYMBOL(vzalloc_node
);
294 * vmalloc_exec - allocate virtually contiguous, executable memory
295 * @size: allocation size
297 * Kernel-internal function to allocate enough pages to cover @size
298 * the page level allocator and map them into contiguous and
299 * executable kernel virtual space.
301 * For tight control over page level allocator and protection flags
302 * use __vmalloc() instead.
305 void *vmalloc_exec(unsigned long size
)
307 return __vmalloc(size
, GFP_KERNEL
| __GFP_HIGHMEM
);
311 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
312 * @size: allocation size
314 * Allocate enough 32bit PA addressable pages to cover @size from the
315 * page level allocator and map them into contiguous kernel virtual space.
317 void *vmalloc_32(unsigned long size
)
319 return __vmalloc(size
, GFP_KERNEL
);
321 EXPORT_SYMBOL(vmalloc_32
);
324 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
325 * @size: allocation size
327 * The resulting memory area is 32bit addressable and zeroed so it can be
328 * mapped to userspace without leaking data.
330 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
331 * remap_vmalloc_range() are permissible.
333 void *vmalloc_32_user(unsigned long size
)
336 * We'll have to sort out the ZONE_DMA bits for 64-bit,
337 * but for now this can simply use vmalloc_user() directly.
339 return vmalloc_user(size
);
341 EXPORT_SYMBOL(vmalloc_32_user
);
343 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
350 void vunmap(const void *addr
)
354 EXPORT_SYMBOL(vunmap
);
356 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
)
361 EXPORT_SYMBOL(vm_map_ram
);
363 void vm_unmap_ram(const void *mem
, unsigned int count
)
367 EXPORT_SYMBOL(vm_unmap_ram
);
369 void vm_unmap_aliases(void)
372 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
375 * Implement a stub for vmalloc_sync_[un]mapping() if the architecture
376 * chose not to have one.
378 void __weak
vmalloc_sync_mappings(void)
382 void __weak
vmalloc_sync_unmappings(void)
386 struct vm_struct
*alloc_vm_area(size_t size
, pte_t
**ptes
)
391 EXPORT_SYMBOL_GPL(alloc_vm_area
);
393 void free_vm_area(struct vm_struct
*area
)
397 EXPORT_SYMBOL_GPL(free_vm_area
);
399 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
404 EXPORT_SYMBOL(vm_insert_page
);
406 int vm_map_pages(struct vm_area_struct
*vma
, struct page
**pages
,
411 EXPORT_SYMBOL(vm_map_pages
);
413 int vm_map_pages_zero(struct vm_area_struct
*vma
, struct page
**pages
,
418 EXPORT_SYMBOL(vm_map_pages_zero
);
421 * sys_brk() for the most part doesn't need the global kernel
422 * lock, except when an application is doing something nasty
423 * like trying to un-brk an area that has already been mapped
424 * to a regular file. in this case, the unmapping will need
425 * to invoke file system routines that need the global lock.
427 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
429 struct mm_struct
*mm
= current
->mm
;
431 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
438 * Always allow shrinking brk
440 if (brk
<= mm
->brk
) {
446 * Ok, looks good - let it rip.
448 flush_icache_range(mm
->brk
, brk
);
449 return mm
->brk
= brk
;
453 * initialise the percpu counter for VM and region record slabs
455 void __init
mmap_init(void)
459 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
461 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
465 * validate the region tree
466 * - the caller must hold the region lock
468 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
469 static noinline
void validate_nommu_regions(void)
471 struct vm_region
*region
, *last
;
472 struct rb_node
*p
, *lastp
;
474 lastp
= rb_first(&nommu_region_tree
);
478 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
479 BUG_ON(last
->vm_end
<= last
->vm_start
);
480 BUG_ON(last
->vm_top
< last
->vm_end
);
482 while ((p
= rb_next(lastp
))) {
483 region
= rb_entry(p
, struct vm_region
, vm_rb
);
484 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
486 BUG_ON(region
->vm_end
<= region
->vm_start
);
487 BUG_ON(region
->vm_top
< region
->vm_end
);
488 BUG_ON(region
->vm_start
< last
->vm_top
);
494 static void validate_nommu_regions(void)
500 * add a region into the global tree
502 static void add_nommu_region(struct vm_region
*region
)
504 struct vm_region
*pregion
;
505 struct rb_node
**p
, *parent
;
507 validate_nommu_regions();
510 p
= &nommu_region_tree
.rb_node
;
513 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
514 if (region
->vm_start
< pregion
->vm_start
)
516 else if (region
->vm_start
> pregion
->vm_start
)
518 else if (pregion
== region
)
524 rb_link_node(®ion
->vm_rb
, parent
, p
);
525 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
527 validate_nommu_regions();
531 * delete a region from the global tree
533 static void delete_nommu_region(struct vm_region
*region
)
535 BUG_ON(!nommu_region_tree
.rb_node
);
537 validate_nommu_regions();
538 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
539 validate_nommu_regions();
543 * free a contiguous series of pages
545 static void free_page_series(unsigned long from
, unsigned long to
)
547 for (; from
< to
; from
+= PAGE_SIZE
) {
548 struct page
*page
= virt_to_page(from
);
550 atomic_long_dec(&mmap_pages_allocated
);
556 * release a reference to a region
557 * - the caller must hold the region semaphore for writing, which this releases
558 * - the region may not have been added to the tree yet, in which case vm_top
559 * will equal vm_start
561 static void __put_nommu_region(struct vm_region
*region
)
562 __releases(nommu_region_sem
)
564 BUG_ON(!nommu_region_tree
.rb_node
);
566 if (--region
->vm_usage
== 0) {
567 if (region
->vm_top
> region
->vm_start
)
568 delete_nommu_region(region
);
569 up_write(&nommu_region_sem
);
572 fput(region
->vm_file
);
574 /* IO memory and memory shared directly out of the pagecache
575 * from ramfs/tmpfs mustn't be released here */
576 if (region
->vm_flags
& VM_MAPPED_COPY
)
577 free_page_series(region
->vm_start
, region
->vm_top
);
578 kmem_cache_free(vm_region_jar
, region
);
580 up_write(&nommu_region_sem
);
585 * release a reference to a region
587 static void put_nommu_region(struct vm_region
*region
)
589 down_write(&nommu_region_sem
);
590 __put_nommu_region(region
);
594 * add a VMA into a process's mm_struct in the appropriate place in the list
595 * and tree and add to the address space's page tree also if not an anonymous
597 * - should be called with mm->mmap_sem held writelocked
599 static void add_vma_to_mm(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
601 struct vm_area_struct
*pvma
, *prev
;
602 struct address_space
*mapping
;
603 struct rb_node
**p
, *parent
, *rb_prev
;
605 BUG_ON(!vma
->vm_region
);
610 /* add the VMA to the mapping */
612 mapping
= vma
->vm_file
->f_mapping
;
614 i_mmap_lock_write(mapping
);
615 flush_dcache_mmap_lock(mapping
);
616 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
617 flush_dcache_mmap_unlock(mapping
);
618 i_mmap_unlock_write(mapping
);
621 /* add the VMA to the tree */
622 parent
= rb_prev
= NULL
;
623 p
= &mm
->mm_rb
.rb_node
;
626 pvma
= rb_entry(parent
, struct vm_area_struct
, vm_rb
);
628 /* sort by: start addr, end addr, VMA struct addr in that order
629 * (the latter is necessary as we may get identical VMAs) */
630 if (vma
->vm_start
< pvma
->vm_start
)
632 else if (vma
->vm_start
> pvma
->vm_start
) {
635 } else if (vma
->vm_end
< pvma
->vm_end
)
637 else if (vma
->vm_end
> pvma
->vm_end
) {
640 } else if (vma
< pvma
)
642 else if (vma
> pvma
) {
649 rb_link_node(&vma
->vm_rb
, parent
, p
);
650 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
652 /* add VMA to the VMA list also */
655 prev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
657 __vma_link_list(mm
, vma
, prev
);
661 * delete a VMA from its owning mm_struct and address space
663 static void delete_vma_from_mm(struct vm_area_struct
*vma
)
666 struct address_space
*mapping
;
667 struct mm_struct
*mm
= vma
->vm_mm
;
668 struct task_struct
*curr
= current
;
671 for (i
= 0; i
< VMACACHE_SIZE
; i
++) {
672 /* if the vma is cached, invalidate the entire cache */
673 if (curr
->vmacache
.vmas
[i
] == vma
) {
674 vmacache_invalidate(mm
);
679 /* remove the VMA from the mapping */
681 mapping
= vma
->vm_file
->f_mapping
;
683 i_mmap_lock_write(mapping
);
684 flush_dcache_mmap_lock(mapping
);
685 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
686 flush_dcache_mmap_unlock(mapping
);
687 i_mmap_unlock_write(mapping
);
690 /* remove from the MM's tree and list */
691 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
693 __vma_unlink_list(mm
, vma
);
697 * destroy a VMA record
699 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
701 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
702 vma
->vm_ops
->close(vma
);
705 put_nommu_region(vma
->vm_region
);
710 * look up the first VMA in which addr resides, NULL if none
711 * - should be called with mm->mmap_sem at least held readlocked
713 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
715 struct vm_area_struct
*vma
;
717 /* check the cache first */
718 vma
= vmacache_find(mm
, addr
);
722 /* trawl the list (there may be multiple mappings in which addr
724 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
725 if (vma
->vm_start
> addr
)
727 if (vma
->vm_end
> addr
) {
728 vmacache_update(addr
, vma
);
735 EXPORT_SYMBOL(find_vma
);
739 * - we don't extend stack VMAs under NOMMU conditions
741 struct vm_area_struct
*find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
743 return find_vma(mm
, addr
);
747 * expand a stack to a given address
748 * - not supported under NOMMU conditions
750 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
756 * look up the first VMA exactly that exactly matches addr
757 * - should be called with mm->mmap_sem at least held readlocked
759 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
763 struct vm_area_struct
*vma
;
764 unsigned long end
= addr
+ len
;
766 /* check the cache first */
767 vma
= vmacache_find_exact(mm
, addr
, end
);
771 /* trawl the list (there may be multiple mappings in which addr
773 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
774 if (vma
->vm_start
< addr
)
776 if (vma
->vm_start
> addr
)
778 if (vma
->vm_end
== end
) {
779 vmacache_update(addr
, vma
);
788 * determine whether a mapping should be permitted and, if so, what sort of
789 * mapping we're capable of supporting
791 static int validate_mmap_request(struct file
*file
,
797 unsigned long *_capabilities
)
799 unsigned long capabilities
, rlen
;
802 /* do the simple checks first */
803 if (flags
& MAP_FIXED
)
806 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
807 (flags
& MAP_TYPE
) != MAP_SHARED
)
813 /* Careful about overflows.. */
814 rlen
= PAGE_ALIGN(len
);
815 if (!rlen
|| rlen
> TASK_SIZE
)
818 /* offset overflow? */
819 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
823 /* files must support mmap */
824 if (!file
->f_op
->mmap
)
827 /* work out if what we've got could possibly be shared
828 * - we support chardevs that provide their own "memory"
829 * - we support files/blockdevs that are memory backed
831 if (file
->f_op
->mmap_capabilities
) {
832 capabilities
= file
->f_op
->mmap_capabilities(file
);
834 /* no explicit capabilities set, so assume some
836 switch (file_inode(file
)->i_mode
& S_IFMT
) {
839 capabilities
= NOMMU_MAP_COPY
;
854 /* eliminate any capabilities that we can't support on this
856 if (!file
->f_op
->get_unmapped_area
)
857 capabilities
&= ~NOMMU_MAP_DIRECT
;
858 if (!(file
->f_mode
& FMODE_CAN_READ
))
859 capabilities
&= ~NOMMU_MAP_COPY
;
861 /* The file shall have been opened with read permission. */
862 if (!(file
->f_mode
& FMODE_READ
))
865 if (flags
& MAP_SHARED
) {
866 /* do checks for writing, appending and locking */
867 if ((prot
& PROT_WRITE
) &&
868 !(file
->f_mode
& FMODE_WRITE
))
871 if (IS_APPEND(file_inode(file
)) &&
872 (file
->f_mode
& FMODE_WRITE
))
875 if (locks_verify_locked(file
))
878 if (!(capabilities
& NOMMU_MAP_DIRECT
))
881 /* we mustn't privatise shared mappings */
882 capabilities
&= ~NOMMU_MAP_COPY
;
884 /* we're going to read the file into private memory we
886 if (!(capabilities
& NOMMU_MAP_COPY
))
889 /* we don't permit a private writable mapping to be
890 * shared with the backing device */
891 if (prot
& PROT_WRITE
)
892 capabilities
&= ~NOMMU_MAP_DIRECT
;
895 if (capabilities
& NOMMU_MAP_DIRECT
) {
896 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
897 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
898 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
900 capabilities
&= ~NOMMU_MAP_DIRECT
;
901 if (flags
& MAP_SHARED
) {
902 pr_warn("MAP_SHARED not completely supported on !MMU\n");
908 /* handle executable mappings and implied executable
910 if (path_noexec(&file
->f_path
)) {
911 if (prot
& PROT_EXEC
)
913 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
914 /* handle implication of PROT_EXEC by PROT_READ */
915 if (current
->personality
& READ_IMPLIES_EXEC
) {
916 if (capabilities
& NOMMU_MAP_EXEC
)
919 } else if ((prot
& PROT_READ
) &&
920 (prot
& PROT_EXEC
) &&
921 !(capabilities
& NOMMU_MAP_EXEC
)
923 /* backing file is not executable, try to copy */
924 capabilities
&= ~NOMMU_MAP_DIRECT
;
927 /* anonymous mappings are always memory backed and can be
930 capabilities
= NOMMU_MAP_COPY
;
932 /* handle PROT_EXEC implication by PROT_READ */
933 if ((prot
& PROT_READ
) &&
934 (current
->personality
& READ_IMPLIES_EXEC
))
938 /* allow the security API to have its say */
939 ret
= security_mmap_addr(addr
);
944 *_capabilities
= capabilities
;
949 * we've determined that we can make the mapping, now translate what we
950 * now know into VMA flags
952 static unsigned long determine_vm_flags(struct file
*file
,
955 unsigned long capabilities
)
957 unsigned long vm_flags
;
959 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
960 /* vm_flags |= mm->def_flags; */
962 if (!(capabilities
& NOMMU_MAP_DIRECT
)) {
963 /* attempt to share read-only copies of mapped file chunks */
964 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
965 if (file
&& !(prot
& PROT_WRITE
))
966 vm_flags
|= VM_MAYSHARE
;
968 /* overlay a shareable mapping on the backing device or inode
969 * if possible - used for chardevs, ramfs/tmpfs/shmfs and
971 vm_flags
|= VM_MAYSHARE
| (capabilities
& NOMMU_VMFLAGS
);
972 if (flags
& MAP_SHARED
)
973 vm_flags
|= VM_SHARED
;
976 /* refuse to let anyone share private mappings with this process if
977 * it's being traced - otherwise breakpoints set in it may interfere
978 * with another untraced process
980 if ((flags
& MAP_PRIVATE
) && current
->ptrace
)
981 vm_flags
&= ~VM_MAYSHARE
;
987 * set up a shared mapping on a file (the driver or filesystem provides and
990 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
994 ret
= call_mmap(vma
->vm_file
, vma
);
996 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1002 /* getting -ENOSYS indicates that direct mmap isn't possible (as
1003 * opposed to tried but failed) so we can only give a suitable error as
1004 * it's not possible to make a private copy if MAP_SHARED was given */
1009 * set up a private mapping or an anonymous shared mapping
1011 static int do_mmap_private(struct vm_area_struct
*vma
,
1012 struct vm_region
*region
,
1014 unsigned long capabilities
)
1016 unsigned long total
, point
;
1020 /* invoke the file's mapping function so that it can keep track of
1021 * shared mappings on devices or memory
1022 * - VM_MAYSHARE will be set if it may attempt to share
1024 if (capabilities
& NOMMU_MAP_DIRECT
) {
1025 ret
= call_mmap(vma
->vm_file
, vma
);
1027 /* shouldn't return success if we're not sharing */
1028 BUG_ON(!(vma
->vm_flags
& VM_MAYSHARE
));
1029 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
1035 /* getting an ENOSYS error indicates that direct mmap isn't
1036 * possible (as opposed to tried but failed) so we'll try to
1037 * make a private copy of the data and map that instead */
1041 /* allocate some memory to hold the mapping
1042 * - note that this may not return a page-aligned address if the object
1043 * we're allocating is smaller than a page
1045 order
= get_order(len
);
1047 point
= len
>> PAGE_SHIFT
;
1049 /* we don't want to allocate a power-of-2 sized page set */
1050 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
1053 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
1057 atomic_long_add(total
, &mmap_pages_allocated
);
1059 region
->vm_flags
= vma
->vm_flags
|= VM_MAPPED_COPY
;
1060 region
->vm_start
= (unsigned long) base
;
1061 region
->vm_end
= region
->vm_start
+ len
;
1062 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
1064 vma
->vm_start
= region
->vm_start
;
1065 vma
->vm_end
= region
->vm_start
+ len
;
1068 /* read the contents of a file into the copy */
1071 fpos
= vma
->vm_pgoff
;
1072 fpos
<<= PAGE_SHIFT
;
1074 ret
= kernel_read(vma
->vm_file
, base
, len
, &fpos
);
1078 /* clear the last little bit */
1080 memset(base
+ ret
, 0, len
- ret
);
1083 vma_set_anonymous(vma
);
1089 free_page_series(region
->vm_start
, region
->vm_top
);
1090 region
->vm_start
= vma
->vm_start
= 0;
1091 region
->vm_end
= vma
->vm_end
= 0;
1096 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1097 len
, current
->pid
, current
->comm
);
1098 show_free_areas(0, NULL
);
1103 * handle mapping creation for uClinux
1105 unsigned long do_mmap(struct file
*file
,
1109 unsigned long flags
,
1110 vm_flags_t vm_flags
,
1111 unsigned long pgoff
,
1112 unsigned long *populate
,
1113 struct list_head
*uf
)
1115 struct vm_area_struct
*vma
;
1116 struct vm_region
*region
;
1118 unsigned long capabilities
, result
;
1123 /* decide whether we should attempt the mapping, and if so what sort of
1125 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1130 /* we ignore the address hint */
1132 len
= PAGE_ALIGN(len
);
1134 /* we've determined that we can make the mapping, now translate what we
1135 * now know into VMA flags */
1136 vm_flags
|= determine_vm_flags(file
, prot
, flags
, capabilities
);
1138 /* we're going to need to record the mapping */
1139 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1141 goto error_getting_region
;
1143 vma
= vm_area_alloc(current
->mm
);
1145 goto error_getting_vma
;
1147 region
->vm_usage
= 1;
1148 region
->vm_flags
= vm_flags
;
1149 region
->vm_pgoff
= pgoff
;
1151 vma
->vm_flags
= vm_flags
;
1152 vma
->vm_pgoff
= pgoff
;
1155 region
->vm_file
= get_file(file
);
1156 vma
->vm_file
= get_file(file
);
1159 down_write(&nommu_region_sem
);
1161 /* if we want to share, we need to check for regions created by other
1162 * mmap() calls that overlap with our proposed mapping
1163 * - we can only share with a superset match on most regular files
1164 * - shared mappings on character devices and memory backed files are
1165 * permitted to overlap inexactly as far as we are concerned for in
1166 * these cases, sharing is handled in the driver or filesystem rather
1169 if (vm_flags
& VM_MAYSHARE
) {
1170 struct vm_region
*pregion
;
1171 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1173 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1174 pgend
= pgoff
+ pglen
;
1176 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1177 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1179 if (!(pregion
->vm_flags
& VM_MAYSHARE
))
1182 /* search for overlapping mappings on the same file */
1183 if (file_inode(pregion
->vm_file
) !=
1187 if (pregion
->vm_pgoff
>= pgend
)
1190 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1191 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1192 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1193 if (pgoff
>= rpgend
)
1196 /* handle inexactly overlapping matches between
1198 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1199 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1200 /* new mapping is not a subset of the region */
1201 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1202 goto sharing_violation
;
1206 /* we've found a region we can share */
1207 pregion
->vm_usage
++;
1208 vma
->vm_region
= pregion
;
1209 start
= pregion
->vm_start
;
1210 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1211 vma
->vm_start
= start
;
1212 vma
->vm_end
= start
+ len
;
1214 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1215 vma
->vm_flags
|= VM_MAPPED_COPY
;
1217 ret
= do_mmap_shared_file(vma
);
1219 vma
->vm_region
= NULL
;
1222 pregion
->vm_usage
--;
1224 goto error_just_free
;
1227 fput(region
->vm_file
);
1228 kmem_cache_free(vm_region_jar
, region
);
1234 /* obtain the address at which to make a shared mapping
1235 * - this is the hook for quasi-memory character devices to
1236 * tell us the location of a shared mapping
1238 if (capabilities
& NOMMU_MAP_DIRECT
) {
1239 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1241 if (IS_ERR_VALUE(addr
)) {
1244 goto error_just_free
;
1246 /* the driver refused to tell us where to site
1247 * the mapping so we'll have to attempt to copy
1250 if (!(capabilities
& NOMMU_MAP_COPY
))
1251 goto error_just_free
;
1253 capabilities
&= ~NOMMU_MAP_DIRECT
;
1255 vma
->vm_start
= region
->vm_start
= addr
;
1256 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1261 vma
->vm_region
= region
;
1263 /* set up the mapping
1264 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1266 if (file
&& vma
->vm_flags
& VM_SHARED
)
1267 ret
= do_mmap_shared_file(vma
);
1269 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1271 goto error_just_free
;
1272 add_nommu_region(region
);
1274 /* clear anonymous mappings that don't ask for uninitialized data */
1275 if (!vma
->vm_file
&&
1276 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED
) ||
1277 !(flags
& MAP_UNINITIALIZED
)))
1278 memset((void *)region
->vm_start
, 0,
1279 region
->vm_end
- region
->vm_start
);
1281 /* okay... we have a mapping; now we have to register it */
1282 result
= vma
->vm_start
;
1284 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1287 add_vma_to_mm(current
->mm
, vma
);
1289 /* we flush the region from the icache only when the first executable
1290 * mapping of it is made */
1291 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1292 flush_icache_range(region
->vm_start
, region
->vm_end
);
1293 region
->vm_icache_flushed
= true;
1296 up_write(&nommu_region_sem
);
1301 up_write(&nommu_region_sem
);
1303 if (region
->vm_file
)
1304 fput(region
->vm_file
);
1305 kmem_cache_free(vm_region_jar
, region
);
1312 up_write(&nommu_region_sem
);
1313 pr_warn("Attempt to share mismatched mappings\n");
1318 kmem_cache_free(vm_region_jar
, region
);
1319 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1321 show_free_areas(0, NULL
);
1324 error_getting_region
:
1325 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1327 show_free_areas(0, NULL
);
1331 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1332 unsigned long prot
, unsigned long flags
,
1333 unsigned long fd
, unsigned long pgoff
)
1335 struct file
*file
= NULL
;
1336 unsigned long retval
= -EBADF
;
1338 audit_mmap_fd(fd
, flags
);
1339 if (!(flags
& MAP_ANONYMOUS
)) {
1345 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1347 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1355 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1356 unsigned long, prot
, unsigned long, flags
,
1357 unsigned long, fd
, unsigned long, pgoff
)
1359 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1362 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1363 struct mmap_arg_struct
{
1367 unsigned long flags
;
1369 unsigned long offset
;
1372 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1374 struct mmap_arg_struct a
;
1376 if (copy_from_user(&a
, arg
, sizeof(a
)))
1378 if (offset_in_page(a
.offset
))
1381 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1382 a
.offset
>> PAGE_SHIFT
);
1384 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1387 * split a vma into two pieces at address 'addr', a new vma is allocated either
1388 * for the first part or the tail.
1390 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1391 unsigned long addr
, int new_below
)
1393 struct vm_area_struct
*new;
1394 struct vm_region
*region
;
1395 unsigned long npages
;
1397 /* we're only permitted to split anonymous regions (these should have
1398 * only a single usage on the region) */
1402 if (mm
->map_count
>= sysctl_max_map_count
)
1405 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1409 new = vm_area_dup(vma
);
1411 kmem_cache_free(vm_region_jar
, region
);
1415 /* most fields are the same, copy all, and then fixup */
1416 *region
= *vma
->vm_region
;
1417 new->vm_region
= region
;
1419 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1422 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1424 region
->vm_start
= new->vm_start
= addr
;
1425 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1428 if (new->vm_ops
&& new->vm_ops
->open
)
1429 new->vm_ops
->open(new);
1431 delete_vma_from_mm(vma
);
1432 down_write(&nommu_region_sem
);
1433 delete_nommu_region(vma
->vm_region
);
1435 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1436 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1438 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1439 vma
->vm_region
->vm_top
= addr
;
1441 add_nommu_region(vma
->vm_region
);
1442 add_nommu_region(new->vm_region
);
1443 up_write(&nommu_region_sem
);
1444 add_vma_to_mm(mm
, vma
);
1445 add_vma_to_mm(mm
, new);
1450 * shrink a VMA by removing the specified chunk from either the beginning or
1453 static int shrink_vma(struct mm_struct
*mm
,
1454 struct vm_area_struct
*vma
,
1455 unsigned long from
, unsigned long to
)
1457 struct vm_region
*region
;
1459 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1461 delete_vma_from_mm(vma
);
1462 if (from
> vma
->vm_start
)
1466 add_vma_to_mm(mm
, vma
);
1468 /* cut the backing region down to size */
1469 region
= vma
->vm_region
;
1470 BUG_ON(region
->vm_usage
!= 1);
1472 down_write(&nommu_region_sem
);
1473 delete_nommu_region(region
);
1474 if (from
> region
->vm_start
) {
1475 to
= region
->vm_top
;
1476 region
->vm_top
= region
->vm_end
= from
;
1478 region
->vm_start
= to
;
1480 add_nommu_region(region
);
1481 up_write(&nommu_region_sem
);
1483 free_page_series(from
, to
);
1489 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1490 * VMA, though it need not cover the whole VMA
1492 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1494 struct vm_area_struct
*vma
;
1498 len
= PAGE_ALIGN(len
);
1504 /* find the first potentially overlapping VMA */
1505 vma
= find_vma(mm
, start
);
1509 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1510 current
->pid
, current
->comm
,
1511 start
, start
+ len
- 1);
1517 /* we're allowed to split an anonymous VMA but not a file-backed one */
1520 if (start
> vma
->vm_start
)
1522 if (end
== vma
->vm_end
)
1523 goto erase_whole_vma
;
1528 /* the chunk must be a subset of the VMA found */
1529 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1530 goto erase_whole_vma
;
1531 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1533 if (offset_in_page(start
))
1535 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1537 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1538 ret
= split_vma(mm
, vma
, start
, 1);
1542 return shrink_vma(mm
, vma
, start
, end
);
1546 delete_vma_from_mm(vma
);
1547 delete_vma(mm
, vma
);
1550 EXPORT_SYMBOL(do_munmap
);
1552 int vm_munmap(unsigned long addr
, size_t len
)
1554 struct mm_struct
*mm
= current
->mm
;
1557 down_write(&mm
->mmap_sem
);
1558 ret
= do_munmap(mm
, addr
, len
, NULL
);
1559 up_write(&mm
->mmap_sem
);
1562 EXPORT_SYMBOL(vm_munmap
);
1564 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1566 return vm_munmap(addr
, len
);
1570 * release all the mappings made in a process's VM space
1572 void exit_mmap(struct mm_struct
*mm
)
1574 struct vm_area_struct
*vma
;
1581 while ((vma
= mm
->mmap
)) {
1582 mm
->mmap
= vma
->vm_next
;
1583 delete_vma_from_mm(vma
);
1584 delete_vma(mm
, vma
);
1589 int vm_brk(unsigned long addr
, unsigned long len
)
1595 * expand (or shrink) an existing mapping, potentially moving it at the same
1596 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1598 * under NOMMU conditions, we only permit changing a mapping's size, and only
1599 * as long as it stays within the region allocated by do_mmap_private() and the
1600 * block is not shareable
1602 * MREMAP_FIXED is not supported under NOMMU conditions
1604 static unsigned long do_mremap(unsigned long addr
,
1605 unsigned long old_len
, unsigned long new_len
,
1606 unsigned long flags
, unsigned long new_addr
)
1608 struct vm_area_struct
*vma
;
1610 /* insanity checks first */
1611 old_len
= PAGE_ALIGN(old_len
);
1612 new_len
= PAGE_ALIGN(new_len
);
1613 if (old_len
== 0 || new_len
== 0)
1614 return (unsigned long) -EINVAL
;
1616 if (offset_in_page(addr
))
1619 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1620 return (unsigned long) -EINVAL
;
1622 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1624 return (unsigned long) -EINVAL
;
1626 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1627 return (unsigned long) -EFAULT
;
1629 if (vma
->vm_flags
& VM_MAYSHARE
)
1630 return (unsigned long) -EPERM
;
1632 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1633 return (unsigned long) -ENOMEM
;
1635 /* all checks complete - do it */
1636 vma
->vm_end
= vma
->vm_start
+ new_len
;
1637 return vma
->vm_start
;
1640 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1641 unsigned long, new_len
, unsigned long, flags
,
1642 unsigned long, new_addr
)
1646 down_write(¤t
->mm
->mmap_sem
);
1647 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1648 up_write(¤t
->mm
->mmap_sem
);
1652 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1653 unsigned int foll_flags
)
1658 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1659 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1661 if (addr
!= (pfn
<< PAGE_SHIFT
))
1664 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1667 EXPORT_SYMBOL(remap_pfn_range
);
1669 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1671 unsigned long pfn
= start
>> PAGE_SHIFT
;
1672 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1674 pfn
+= vma
->vm_pgoff
;
1675 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1677 EXPORT_SYMBOL(vm_iomap_memory
);
1679 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1680 unsigned long pgoff
)
1682 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1684 if (!(vma
->vm_flags
& VM_USERMAP
))
1687 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1688 vma
->vm_end
= vma
->vm_start
+ size
;
1692 EXPORT_SYMBOL(remap_vmalloc_range
);
1694 unsigned long arch_get_unmapped_area(struct file
*file
, unsigned long addr
,
1695 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1700 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
1705 EXPORT_SYMBOL(filemap_fault
);
1707 void filemap_map_pages(struct vm_fault
*vmf
,
1708 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1712 EXPORT_SYMBOL(filemap_map_pages
);
1714 int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
1715 unsigned long addr
, void *buf
, int len
, unsigned int gup_flags
)
1717 struct vm_area_struct
*vma
;
1718 int write
= gup_flags
& FOLL_WRITE
;
1720 if (down_read_killable(&mm
->mmap_sem
))
1723 /* the access must start within one of the target process's mappings */
1724 vma
= find_vma(mm
, addr
);
1726 /* don't overrun this mapping */
1727 if (addr
+ len
>= vma
->vm_end
)
1728 len
= vma
->vm_end
- addr
;
1730 /* only read or write mappings where it is permitted */
1731 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1732 copy_to_user_page(vma
, NULL
, addr
,
1733 (void *) addr
, buf
, len
);
1734 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1735 copy_from_user_page(vma
, NULL
, addr
,
1736 buf
, (void *) addr
, len
);
1743 up_read(&mm
->mmap_sem
);
1749 * access_remote_vm - access another process' address space
1750 * @mm: the mm_struct of the target address space
1751 * @addr: start address to access
1752 * @buf: source or destination buffer
1753 * @len: number of bytes to transfer
1754 * @gup_flags: flags modifying lookup behaviour
1756 * The caller must hold a reference on @mm.
1758 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1759 void *buf
, int len
, unsigned int gup_flags
)
1761 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, gup_flags
);
1765 * Access another process' address space.
1766 * - source/target buffer must be kernel space
1768 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1769 unsigned int gup_flags
)
1771 struct mm_struct
*mm
;
1773 if (addr
+ len
< addr
)
1776 mm
= get_task_mm(tsk
);
1780 len
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, gup_flags
);
1785 EXPORT_SYMBOL_GPL(access_process_vm
);
1788 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1789 * @inode: The inode to check
1790 * @size: The current filesize of the inode
1791 * @newsize: The proposed filesize of the inode
1793 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1794 * make sure that that any outstanding VMAs aren't broken and then shrink the
1795 * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't
1796 * automatically grant mappings that are too large.
1798 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1801 struct vm_area_struct
*vma
;
1802 struct vm_region
*region
;
1804 size_t r_size
, r_top
;
1806 low
= newsize
>> PAGE_SHIFT
;
1807 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1809 down_write(&nommu_region_sem
);
1810 i_mmap_lock_read(inode
->i_mapping
);
1812 /* search for VMAs that fall within the dead zone */
1813 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1814 /* found one - only interested if it's shared out of the page
1816 if (vma
->vm_flags
& VM_SHARED
) {
1817 i_mmap_unlock_read(inode
->i_mapping
);
1818 up_write(&nommu_region_sem
);
1819 return -ETXTBSY
; /* not quite true, but near enough */
1823 /* reduce any regions that overlap the dead zone - if in existence,
1824 * these will be pointed to by VMAs that don't overlap the dead zone
1826 * we don't check for any regions that start beyond the EOF as there
1829 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1830 if (!(vma
->vm_flags
& VM_SHARED
))
1833 region
= vma
->vm_region
;
1834 r_size
= region
->vm_top
- region
->vm_start
;
1835 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1837 if (r_top
> newsize
) {
1838 region
->vm_top
-= r_top
- newsize
;
1839 if (region
->vm_end
> region
->vm_top
)
1840 region
->vm_end
= region
->vm_top
;
1844 i_mmap_unlock_read(inode
->i_mapping
);
1845 up_write(&nommu_region_sem
);
1850 * Initialise sysctl_user_reserve_kbytes.
1852 * This is intended to prevent a user from starting a single memory hogging
1853 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1856 * The default value is min(3% of free memory, 128MB)
1857 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1859 static int __meminit
init_user_reserve(void)
1861 unsigned long free_kbytes
;
1863 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1865 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1868 subsys_initcall(init_user_reserve
);
1871 * Initialise sysctl_admin_reserve_kbytes.
1873 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1874 * to log in and kill a memory hogging process.
1876 * Systems with more than 256MB will reserve 8MB, enough to recover
1877 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1878 * only reserve 3% of free pages by default.
1880 static int __meminit
init_admin_reserve(void)
1882 unsigned long free_kbytes
;
1884 free_kbytes
= global_zone_page_state(NR_FREE_PAGES
) << (PAGE_SHIFT
- 10);
1886 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1889 subsys_initcall(init_admin_reserve
);