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/admin-guide/mm/nommu-mmap.rst
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/mman.h>
23 #include <linux/swap.h>
24 #include <linux/file.h>
25 #include <linux/highmem.h>
26 #include <linux/pagemap.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/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/printk.h>
38 #include <linux/uaccess.h>
39 #include <linux/uio.h>
41 #include <asm/tlbflush.h>
42 #include <asm/mmu_context.h>
46 EXPORT_SYMBOL(high_memory
);
48 unsigned long max_mapnr
;
49 EXPORT_SYMBOL(max_mapnr
);
50 unsigned long highest_memmap_pfn
;
51 int sysctl_nr_trim_pages
= CONFIG_NOMMU_INITIAL_TRIM_EXCESS
;
52 int heap_stack_gap
= 0;
54 atomic_long_t mmap_pages_allocated
;
56 EXPORT_SYMBOL(mem_map
);
58 /* list of mapped, potentially shareable regions */
59 static struct kmem_cache
*vm_region_jar
;
60 struct rb_root nommu_region_tree
= RB_ROOT
;
61 DECLARE_RWSEM(nommu_region_sem
);
63 const struct vm_operations_struct generic_file_vm_ops
= {
67 * Return the total memory allocated for this pointer, not
68 * just what the caller asked for.
70 * Doesn't have to be accurate, i.e. may have races.
72 unsigned int kobjsize(const void *objp
)
77 * If the object we have should not have ksize performed on it,
80 if (!objp
|| !virt_addr_valid(objp
))
83 page
= virt_to_head_page(objp
);
86 * If the allocator sets PageSlab, we know the pointer came from
93 * If it's not a compound page, see if we have a matching VMA
94 * region. This test is intentionally done in reverse order,
95 * so if there's no VMA, we still fall through and hand back
96 * PAGE_SIZE for 0-order pages.
98 if (!PageCompound(page
)) {
99 struct vm_area_struct
*vma
;
101 vma
= find_vma(current
->mm
, (unsigned long)objp
);
103 return vma
->vm_end
- vma
->vm_start
;
107 * The ksize() function is only guaranteed to work for pointers
108 * returned by kmalloc(). So handle arbitrary pointers here.
110 return page_size(page
);
114 * follow_pfn - look up PFN at a user virtual address
115 * @vma: memory mapping
116 * @address: user virtual address
117 * @pfn: location to store found PFN
119 * Only IO mappings and raw PFN mappings are allowed.
121 * Returns zero and the pfn at @pfn on success, -ve otherwise.
123 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
126 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
129 *pfn
= address
>> PAGE_SHIFT
;
132 EXPORT_SYMBOL(follow_pfn
);
134 void vfree(const void *addr
)
138 EXPORT_SYMBOL(vfree
);
140 void *__vmalloc(unsigned long size
, gfp_t gfp_mask
)
143 * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc()
144 * returns only a logical address.
146 return kmalloc(size
, (gfp_mask
| __GFP_COMP
) & ~__GFP_HIGHMEM
);
148 EXPORT_SYMBOL(__vmalloc
);
150 void *__vmalloc_node_range(unsigned long size
, unsigned long align
,
151 unsigned long start
, unsigned long end
, gfp_t gfp_mask
,
152 pgprot_t prot
, unsigned long vm_flags
, int node
,
155 return __vmalloc(size
, gfp_mask
);
158 void *__vmalloc_node(unsigned long size
, unsigned long align
, gfp_t gfp_mask
,
159 int node
, const void *caller
)
161 return __vmalloc(size
, gfp_mask
);
164 static void *__vmalloc_user_flags(unsigned long size
, gfp_t flags
)
168 ret
= __vmalloc(size
, flags
);
170 struct vm_area_struct
*vma
;
172 mmap_write_lock(current
->mm
);
173 vma
= find_vma(current
->mm
, (unsigned long)ret
);
175 vm_flags_set(vma
, VM_USERMAP
);
176 mmap_write_unlock(current
->mm
);
182 void *vmalloc_user(unsigned long size
)
184 return __vmalloc_user_flags(size
, GFP_KERNEL
| __GFP_ZERO
);
186 EXPORT_SYMBOL(vmalloc_user
);
188 struct page
*vmalloc_to_page(const void *addr
)
190 return virt_to_page(addr
);
192 EXPORT_SYMBOL(vmalloc_to_page
);
194 unsigned long vmalloc_to_pfn(const void *addr
)
196 return page_to_pfn(virt_to_page(addr
));
198 EXPORT_SYMBOL(vmalloc_to_pfn
);
200 long vread_iter(struct iov_iter
*iter
, const char *addr
, size_t count
)
202 /* Don't allow overflow */
203 if ((unsigned long) addr
+ count
< count
)
204 count
= -(unsigned long) addr
;
206 return copy_to_iter(addr
, count
, iter
);
210 * vmalloc - allocate virtually contiguous memory
212 * @size: allocation size
214 * Allocate enough pages to cover @size from the page level
215 * allocator and map them into contiguous kernel virtual space.
217 * For tight control over page level allocator and protection flags
218 * use __vmalloc() instead.
220 void *vmalloc(unsigned long size
)
222 return __vmalloc(size
, GFP_KERNEL
);
224 EXPORT_SYMBOL(vmalloc
);
226 void *vmalloc_huge(unsigned long size
, gfp_t gfp_mask
) __weak
__alias(__vmalloc
);
229 * vzalloc - allocate virtually contiguous memory with zero fill
231 * @size: allocation size
233 * Allocate enough pages to cover @size from the page level
234 * allocator and map them into contiguous kernel virtual space.
235 * The memory allocated is set to zero.
237 * For tight control over page level allocator and protection flags
238 * use __vmalloc() instead.
240 void *vzalloc(unsigned long size
)
242 return __vmalloc(size
, GFP_KERNEL
| __GFP_ZERO
);
244 EXPORT_SYMBOL(vzalloc
);
247 * vmalloc_node - allocate memory on a specific node
248 * @size: allocation size
251 * Allocate enough pages to cover @size from the page level
252 * allocator and map them into contiguous kernel virtual space.
254 * For tight control over page level allocator and protection flags
255 * use __vmalloc() instead.
257 void *vmalloc_node(unsigned long size
, int node
)
259 return vmalloc(size
);
261 EXPORT_SYMBOL(vmalloc_node
);
264 * vzalloc_node - allocate memory on a specific node with zero fill
265 * @size: allocation size
268 * Allocate enough pages to cover @size from the page level
269 * allocator and map them into contiguous kernel virtual space.
270 * The memory allocated is set to zero.
272 * For tight control over page level allocator and protection flags
273 * use __vmalloc() instead.
275 void *vzalloc_node(unsigned long size
, int node
)
277 return vzalloc(size
);
279 EXPORT_SYMBOL(vzalloc_node
);
282 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
283 * @size: allocation size
285 * Allocate enough 32bit PA addressable pages to cover @size from the
286 * page level allocator and map them into contiguous kernel virtual space.
288 void *vmalloc_32(unsigned long size
)
290 return __vmalloc(size
, GFP_KERNEL
);
292 EXPORT_SYMBOL(vmalloc_32
);
295 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
296 * @size: allocation size
298 * The resulting memory area is 32bit addressable and zeroed so it can be
299 * mapped to userspace without leaking data.
301 * VM_USERMAP is set on the corresponding VMA so that subsequent calls to
302 * remap_vmalloc_range() are permissible.
304 void *vmalloc_32_user(unsigned long size
)
307 * We'll have to sort out the ZONE_DMA bits for 64-bit,
308 * but for now this can simply use vmalloc_user() directly.
310 return vmalloc_user(size
);
312 EXPORT_SYMBOL(vmalloc_32_user
);
314 void *vmap(struct page
**pages
, unsigned int count
, unsigned long flags
, pgprot_t prot
)
321 void vunmap(const void *addr
)
325 EXPORT_SYMBOL(vunmap
);
327 void *vm_map_ram(struct page
**pages
, unsigned int count
, int node
)
332 EXPORT_SYMBOL(vm_map_ram
);
334 void vm_unmap_ram(const void *mem
, unsigned int count
)
338 EXPORT_SYMBOL(vm_unmap_ram
);
340 void vm_unmap_aliases(void)
343 EXPORT_SYMBOL_GPL(vm_unmap_aliases
);
345 void free_vm_area(struct vm_struct
*area
)
349 EXPORT_SYMBOL_GPL(free_vm_area
);
351 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
356 EXPORT_SYMBOL(vm_insert_page
);
358 int vm_map_pages(struct vm_area_struct
*vma
, struct page
**pages
,
363 EXPORT_SYMBOL(vm_map_pages
);
365 int vm_map_pages_zero(struct vm_area_struct
*vma
, struct page
**pages
,
370 EXPORT_SYMBOL(vm_map_pages_zero
);
373 * sys_brk() for the most part doesn't need the global kernel
374 * lock, except when an application is doing something nasty
375 * like trying to un-brk an area that has already been mapped
376 * to a regular file. in this case, the unmapping will need
377 * to invoke file system routines that need the global lock.
379 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
381 struct mm_struct
*mm
= current
->mm
;
383 if (brk
< mm
->start_brk
|| brk
> mm
->context
.end_brk
)
390 * Always allow shrinking brk
392 if (brk
<= mm
->brk
) {
398 * Ok, looks good - let it rip.
400 flush_icache_user_range(mm
->brk
, brk
);
401 return mm
->brk
= brk
;
405 * initialise the percpu counter for VM and region record slabs
407 void __init
mmap_init(void)
411 ret
= percpu_counter_init(&vm_committed_as
, 0, GFP_KERNEL
);
413 vm_region_jar
= KMEM_CACHE(vm_region
, SLAB_PANIC
|SLAB_ACCOUNT
);
417 * validate the region tree
418 * - the caller must hold the region lock
420 #ifdef CONFIG_DEBUG_NOMMU_REGIONS
421 static noinline
void validate_nommu_regions(void)
423 struct vm_region
*region
, *last
;
424 struct rb_node
*p
, *lastp
;
426 lastp
= rb_first(&nommu_region_tree
);
430 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
431 BUG_ON(last
->vm_end
<= last
->vm_start
);
432 BUG_ON(last
->vm_top
< last
->vm_end
);
434 while ((p
= rb_next(lastp
))) {
435 region
= rb_entry(p
, struct vm_region
, vm_rb
);
436 last
= rb_entry(lastp
, struct vm_region
, vm_rb
);
438 BUG_ON(region
->vm_end
<= region
->vm_start
);
439 BUG_ON(region
->vm_top
< region
->vm_end
);
440 BUG_ON(region
->vm_start
< last
->vm_top
);
446 static void validate_nommu_regions(void)
452 * add a region into the global tree
454 static void add_nommu_region(struct vm_region
*region
)
456 struct vm_region
*pregion
;
457 struct rb_node
**p
, *parent
;
459 validate_nommu_regions();
462 p
= &nommu_region_tree
.rb_node
;
465 pregion
= rb_entry(parent
, struct vm_region
, vm_rb
);
466 if (region
->vm_start
< pregion
->vm_start
)
468 else if (region
->vm_start
> pregion
->vm_start
)
470 else if (pregion
== region
)
476 rb_link_node(®ion
->vm_rb
, parent
, p
);
477 rb_insert_color(®ion
->vm_rb
, &nommu_region_tree
);
479 validate_nommu_regions();
483 * delete a region from the global tree
485 static void delete_nommu_region(struct vm_region
*region
)
487 BUG_ON(!nommu_region_tree
.rb_node
);
489 validate_nommu_regions();
490 rb_erase(®ion
->vm_rb
, &nommu_region_tree
);
491 validate_nommu_regions();
495 * free a contiguous series of pages
497 static void free_page_series(unsigned long from
, unsigned long to
)
499 for (; from
< to
; from
+= PAGE_SIZE
) {
500 struct page
*page
= virt_to_page((void *)from
);
502 atomic_long_dec(&mmap_pages_allocated
);
508 * release a reference to a region
509 * - the caller must hold the region semaphore for writing, which this releases
510 * - the region may not have been added to the tree yet, in which case vm_top
511 * will equal vm_start
513 static void __put_nommu_region(struct vm_region
*region
)
514 __releases(nommu_region_sem
)
516 BUG_ON(!nommu_region_tree
.rb_node
);
518 if (--region
->vm_usage
== 0) {
519 if (region
->vm_top
> region
->vm_start
)
520 delete_nommu_region(region
);
521 up_write(&nommu_region_sem
);
524 fput(region
->vm_file
);
526 /* IO memory and memory shared directly out of the pagecache
527 * from ramfs/tmpfs mustn't be released here */
528 if (region
->vm_flags
& VM_MAPPED_COPY
)
529 free_page_series(region
->vm_start
, region
->vm_top
);
530 kmem_cache_free(vm_region_jar
, region
);
532 up_write(&nommu_region_sem
);
537 * release a reference to a region
539 static void put_nommu_region(struct vm_region
*region
)
541 down_write(&nommu_region_sem
);
542 __put_nommu_region(region
);
545 static void setup_vma_to_mm(struct vm_area_struct
*vma
, struct mm_struct
*mm
)
549 /* add the VMA to the mapping */
551 struct address_space
*mapping
= vma
->vm_file
->f_mapping
;
553 i_mmap_lock_write(mapping
);
554 flush_dcache_mmap_lock(mapping
);
555 vma_interval_tree_insert(vma
, &mapping
->i_mmap
);
556 flush_dcache_mmap_unlock(mapping
);
557 i_mmap_unlock_write(mapping
);
561 static void cleanup_vma_from_mm(struct vm_area_struct
*vma
)
563 vma
->vm_mm
->map_count
--;
564 /* remove the VMA from the mapping */
566 struct address_space
*mapping
;
567 mapping
= vma
->vm_file
->f_mapping
;
569 i_mmap_lock_write(mapping
);
570 flush_dcache_mmap_lock(mapping
);
571 vma_interval_tree_remove(vma
, &mapping
->i_mmap
);
572 flush_dcache_mmap_unlock(mapping
);
573 i_mmap_unlock_write(mapping
);
578 * delete a VMA from its owning mm_struct and address space
580 static int delete_vma_from_mm(struct vm_area_struct
*vma
)
582 VMA_ITERATOR(vmi
, vma
->vm_mm
, vma
->vm_start
);
584 vma_iter_config(&vmi
, vma
->vm_start
, vma
->vm_end
);
585 if (vma_iter_prealloc(&vmi
, vma
)) {
586 pr_warn("Allocation of vma tree for process %d failed\n",
590 cleanup_vma_from_mm(vma
);
592 /* remove from the MM's tree and list */
593 vma_iter_clear(&vmi
);
597 * destroy a VMA record
599 static void delete_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
601 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
602 vma
->vm_ops
->close(vma
);
605 put_nommu_region(vma
->vm_region
);
609 struct vm_area_struct
*find_vma_intersection(struct mm_struct
*mm
,
610 unsigned long start_addr
,
611 unsigned long end_addr
)
613 unsigned long index
= start_addr
;
615 mmap_assert_locked(mm
);
616 return mt_find(&mm
->mm_mt
, &index
, end_addr
- 1);
618 EXPORT_SYMBOL(find_vma_intersection
);
621 * look up the first VMA in which addr resides, NULL if none
622 * - should be called with mm->mmap_lock at least held readlocked
624 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
626 VMA_ITERATOR(vmi
, mm
, addr
);
628 return vma_iter_load(&vmi
);
630 EXPORT_SYMBOL(find_vma
);
633 * At least xtensa ends up having protection faults even with no
634 * MMU.. No stack expansion, at least.
636 struct vm_area_struct
*lock_mm_and_find_vma(struct mm_struct
*mm
,
637 unsigned long addr
, struct pt_regs
*regs
)
639 struct vm_area_struct
*vma
;
642 vma
= vma_lookup(mm
, addr
);
644 mmap_read_unlock(mm
);
649 * expand a stack to a given address
650 * - not supported under NOMMU conditions
652 int expand_stack_locked(struct vm_area_struct
*vma
, unsigned long addr
)
657 struct vm_area_struct
*expand_stack(struct mm_struct
*mm
, unsigned long addr
)
659 mmap_read_unlock(mm
);
664 * look up the first VMA exactly that exactly matches addr
665 * - should be called with mm->mmap_lock at least held readlocked
667 static struct vm_area_struct
*find_vma_exact(struct mm_struct
*mm
,
671 struct vm_area_struct
*vma
;
672 unsigned long end
= addr
+ len
;
673 VMA_ITERATOR(vmi
, mm
, addr
);
675 vma
= vma_iter_load(&vmi
);
678 if (vma
->vm_start
!= addr
)
680 if (vma
->vm_end
!= end
)
687 * determine whether a mapping should be permitted and, if so, what sort of
688 * mapping we're capable of supporting
690 static int validate_mmap_request(struct file
*file
,
696 unsigned long *_capabilities
)
698 unsigned long capabilities
, rlen
;
701 /* do the simple checks first */
702 if (flags
& MAP_FIXED
)
705 if ((flags
& MAP_TYPE
) != MAP_PRIVATE
&&
706 (flags
& MAP_TYPE
) != MAP_SHARED
)
712 /* Careful about overflows.. */
713 rlen
= PAGE_ALIGN(len
);
714 if (!rlen
|| rlen
> TASK_SIZE
)
717 /* offset overflow? */
718 if ((pgoff
+ (rlen
>> PAGE_SHIFT
)) < pgoff
)
722 /* files must support mmap */
723 if (!file
->f_op
->mmap
)
726 /* work out if what we've got could possibly be shared
727 * - we support chardevs that provide their own "memory"
728 * - we support files/blockdevs that are memory backed
730 if (file
->f_op
->mmap_capabilities
) {
731 capabilities
= file
->f_op
->mmap_capabilities(file
);
733 /* no explicit capabilities set, so assume some
735 switch (file_inode(file
)->i_mode
& S_IFMT
) {
738 capabilities
= NOMMU_MAP_COPY
;
753 /* eliminate any capabilities that we can't support on this
755 if (!file
->f_op
->get_unmapped_area
)
756 capabilities
&= ~NOMMU_MAP_DIRECT
;
757 if (!(file
->f_mode
& FMODE_CAN_READ
))
758 capabilities
&= ~NOMMU_MAP_COPY
;
760 /* The file shall have been opened with read permission. */
761 if (!(file
->f_mode
& FMODE_READ
))
764 if (flags
& MAP_SHARED
) {
765 /* do checks for writing, appending and locking */
766 if ((prot
& PROT_WRITE
) &&
767 !(file
->f_mode
& FMODE_WRITE
))
770 if (IS_APPEND(file_inode(file
)) &&
771 (file
->f_mode
& FMODE_WRITE
))
774 if (!(capabilities
& NOMMU_MAP_DIRECT
))
777 /* we mustn't privatise shared mappings */
778 capabilities
&= ~NOMMU_MAP_COPY
;
780 /* we're going to read the file into private memory we
782 if (!(capabilities
& NOMMU_MAP_COPY
))
785 /* we don't permit a private writable mapping to be
786 * shared with the backing device */
787 if (prot
& PROT_WRITE
)
788 capabilities
&= ~NOMMU_MAP_DIRECT
;
791 if (capabilities
& NOMMU_MAP_DIRECT
) {
792 if (((prot
& PROT_READ
) && !(capabilities
& NOMMU_MAP_READ
)) ||
793 ((prot
& PROT_WRITE
) && !(capabilities
& NOMMU_MAP_WRITE
)) ||
794 ((prot
& PROT_EXEC
) && !(capabilities
& NOMMU_MAP_EXEC
))
796 capabilities
&= ~NOMMU_MAP_DIRECT
;
797 if (flags
& MAP_SHARED
) {
798 pr_warn("MAP_SHARED not completely supported on !MMU\n");
804 /* handle executable mappings and implied executable
806 if (path_noexec(&file
->f_path
)) {
807 if (prot
& PROT_EXEC
)
809 } else if ((prot
& PROT_READ
) && !(prot
& PROT_EXEC
)) {
810 /* handle implication of PROT_EXEC by PROT_READ */
811 if (current
->personality
& READ_IMPLIES_EXEC
) {
812 if (capabilities
& NOMMU_MAP_EXEC
)
815 } else if ((prot
& PROT_READ
) &&
816 (prot
& PROT_EXEC
) &&
817 !(capabilities
& NOMMU_MAP_EXEC
)
819 /* backing file is not executable, try to copy */
820 capabilities
&= ~NOMMU_MAP_DIRECT
;
823 /* anonymous mappings are always memory backed and can be
826 capabilities
= NOMMU_MAP_COPY
;
828 /* handle PROT_EXEC implication by PROT_READ */
829 if ((prot
& PROT_READ
) &&
830 (current
->personality
& READ_IMPLIES_EXEC
))
834 /* allow the security API to have its say */
835 ret
= security_mmap_addr(addr
);
840 *_capabilities
= capabilities
;
845 * we've determined that we can make the mapping, now translate what we
846 * now know into VMA flags
848 static unsigned long determine_vm_flags(struct file
*file
,
851 unsigned long capabilities
)
853 unsigned long vm_flags
;
855 vm_flags
= calc_vm_prot_bits(prot
, 0) | calc_vm_flag_bits(flags
);
859 * MAP_ANONYMOUS. MAP_SHARED is mapped to MAP_PRIVATE, because
860 * there is no fork().
862 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
863 } else if (flags
& MAP_PRIVATE
) {
864 /* MAP_PRIVATE file mapping */
865 if (capabilities
& NOMMU_MAP_DIRECT
)
866 vm_flags
|= (capabilities
& NOMMU_VMFLAGS
);
868 vm_flags
|= VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
870 if (!(prot
& PROT_WRITE
) && !current
->ptrace
)
872 * R/O private file mapping which cannot be used to
873 * modify memory, especially also not via active ptrace
874 * (e.g., set breakpoints) or later by upgrading
875 * permissions (no mprotect()). We can try overlaying
876 * the file mapping, which will work e.g., on chardevs,
877 * ramfs/tmpfs/shmfs and romfs/cramf.
879 vm_flags
|= VM_MAYOVERLAY
;
881 /* MAP_SHARED file mapping: NOMMU_MAP_DIRECT is set. */
882 vm_flags
|= VM_SHARED
| VM_MAYSHARE
|
883 (capabilities
& NOMMU_VMFLAGS
);
890 * set up a shared mapping on a file (the driver or filesystem provides and
893 static int do_mmap_shared_file(struct vm_area_struct
*vma
)
897 ret
= call_mmap(vma
->vm_file
, vma
);
899 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
905 /* getting -ENOSYS indicates that direct mmap isn't possible (as
906 * opposed to tried but failed) so we can only give a suitable error as
907 * it's not possible to make a private copy if MAP_SHARED was given */
912 * set up a private mapping or an anonymous shared mapping
914 static int do_mmap_private(struct vm_area_struct
*vma
,
915 struct vm_region
*region
,
917 unsigned long capabilities
)
919 unsigned long total
, point
;
924 * Invoke the file's mapping function so that it can keep track of
925 * shared mappings on devices or memory. VM_MAYOVERLAY will be set if
926 * it may attempt to share, which will make is_nommu_shared_mapping()
929 if (capabilities
& NOMMU_MAP_DIRECT
) {
930 ret
= call_mmap(vma
->vm_file
, vma
);
931 /* shouldn't return success if we're not sharing */
932 if (WARN_ON_ONCE(!is_nommu_shared_mapping(vma
->vm_flags
)))
935 vma
->vm_region
->vm_top
= vma
->vm_region
->vm_end
;
941 /* getting an ENOSYS error indicates that direct mmap isn't
942 * possible (as opposed to tried but failed) so we'll try to
943 * make a private copy of the data and map that instead */
947 /* allocate some memory to hold the mapping
948 * - note that this may not return a page-aligned address if the object
949 * we're allocating is smaller than a page
951 order
= get_order(len
);
953 point
= len
>> PAGE_SHIFT
;
955 /* we don't want to allocate a power-of-2 sized page set */
956 if (sysctl_nr_trim_pages
&& total
- point
>= sysctl_nr_trim_pages
)
959 base
= alloc_pages_exact(total
<< PAGE_SHIFT
, GFP_KERNEL
);
963 atomic_long_add(total
, &mmap_pages_allocated
);
965 vm_flags_set(vma
, VM_MAPPED_COPY
);
966 region
->vm_flags
= vma
->vm_flags
;
967 region
->vm_start
= (unsigned long) base
;
968 region
->vm_end
= region
->vm_start
+ len
;
969 region
->vm_top
= region
->vm_start
+ (total
<< PAGE_SHIFT
);
971 vma
->vm_start
= region
->vm_start
;
972 vma
->vm_end
= region
->vm_start
+ len
;
975 /* read the contents of a file into the copy */
978 fpos
= vma
->vm_pgoff
;
981 ret
= kernel_read(vma
->vm_file
, base
, len
, &fpos
);
985 /* clear the last little bit */
987 memset(base
+ ret
, 0, len
- ret
);
990 vma_set_anonymous(vma
);
996 free_page_series(region
->vm_start
, region
->vm_top
);
997 region
->vm_start
= vma
->vm_start
= 0;
998 region
->vm_end
= vma
->vm_end
= 0;
1003 pr_err("Allocation of length %lu from process %d (%s) failed\n",
1004 len
, current
->pid
, current
->comm
);
1010 * handle mapping creation for uClinux
1012 unsigned long do_mmap(struct file
*file
,
1016 unsigned long flags
,
1017 vm_flags_t vm_flags
,
1018 unsigned long pgoff
,
1019 unsigned long *populate
,
1020 struct list_head
*uf
)
1022 struct vm_area_struct
*vma
;
1023 struct vm_region
*region
;
1025 unsigned long capabilities
, result
;
1027 VMA_ITERATOR(vmi
, current
->mm
, 0);
1031 /* decide whether we should attempt the mapping, and if so what sort of
1033 ret
= validate_mmap_request(file
, addr
, len
, prot
, flags
, pgoff
,
1038 /* we ignore the address hint */
1040 len
= PAGE_ALIGN(len
);
1042 /* we've determined that we can make the mapping, now translate what we
1043 * now know into VMA flags */
1044 vm_flags
|= determine_vm_flags(file
, prot
, flags
, capabilities
);
1047 /* we're going to need to record the mapping */
1048 region
= kmem_cache_zalloc(vm_region_jar
, GFP_KERNEL
);
1050 goto error_getting_region
;
1052 vma
= vm_area_alloc(current
->mm
);
1054 goto error_getting_vma
;
1056 region
->vm_usage
= 1;
1057 region
->vm_flags
= vm_flags
;
1058 region
->vm_pgoff
= pgoff
;
1060 vm_flags_init(vma
, vm_flags
);
1061 vma
->vm_pgoff
= pgoff
;
1064 region
->vm_file
= get_file(file
);
1065 vma
->vm_file
= get_file(file
);
1068 down_write(&nommu_region_sem
);
1070 /* if we want to share, we need to check for regions created by other
1071 * mmap() calls that overlap with our proposed mapping
1072 * - we can only share with a superset match on most regular files
1073 * - shared mappings on character devices and memory backed files are
1074 * permitted to overlap inexactly as far as we are concerned for in
1075 * these cases, sharing is handled in the driver or filesystem rather
1078 if (is_nommu_shared_mapping(vm_flags
)) {
1079 struct vm_region
*pregion
;
1080 unsigned long pglen
, rpglen
, pgend
, rpgend
, start
;
1082 pglen
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1083 pgend
= pgoff
+ pglen
;
1085 for (rb
= rb_first(&nommu_region_tree
); rb
; rb
= rb_next(rb
)) {
1086 pregion
= rb_entry(rb
, struct vm_region
, vm_rb
);
1088 if (!is_nommu_shared_mapping(pregion
->vm_flags
))
1091 /* search for overlapping mappings on the same file */
1092 if (file_inode(pregion
->vm_file
) !=
1096 if (pregion
->vm_pgoff
>= pgend
)
1099 rpglen
= pregion
->vm_end
- pregion
->vm_start
;
1100 rpglen
= (rpglen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1101 rpgend
= pregion
->vm_pgoff
+ rpglen
;
1102 if (pgoff
>= rpgend
)
1105 /* handle inexactly overlapping matches between
1107 if ((pregion
->vm_pgoff
!= pgoff
|| rpglen
!= pglen
) &&
1108 !(pgoff
>= pregion
->vm_pgoff
&& pgend
<= rpgend
)) {
1109 /* new mapping is not a subset of the region */
1110 if (!(capabilities
& NOMMU_MAP_DIRECT
))
1111 goto sharing_violation
;
1115 /* we've found a region we can share */
1116 pregion
->vm_usage
++;
1117 vma
->vm_region
= pregion
;
1118 start
= pregion
->vm_start
;
1119 start
+= (pgoff
- pregion
->vm_pgoff
) << PAGE_SHIFT
;
1120 vma
->vm_start
= start
;
1121 vma
->vm_end
= start
+ len
;
1123 if (pregion
->vm_flags
& VM_MAPPED_COPY
)
1124 vm_flags_set(vma
, VM_MAPPED_COPY
);
1126 ret
= do_mmap_shared_file(vma
);
1128 vma
->vm_region
= NULL
;
1131 pregion
->vm_usage
--;
1133 goto error_just_free
;
1136 fput(region
->vm_file
);
1137 kmem_cache_free(vm_region_jar
, region
);
1143 /* obtain the address at which to make a shared mapping
1144 * - this is the hook for quasi-memory character devices to
1145 * tell us the location of a shared mapping
1147 if (capabilities
& NOMMU_MAP_DIRECT
) {
1148 addr
= file
->f_op
->get_unmapped_area(file
, addr
, len
,
1150 if (IS_ERR_VALUE(addr
)) {
1153 goto error_just_free
;
1155 /* the driver refused to tell us where to site
1156 * the mapping so we'll have to attempt to copy
1159 if (!(capabilities
& NOMMU_MAP_COPY
))
1160 goto error_just_free
;
1162 capabilities
&= ~NOMMU_MAP_DIRECT
;
1164 vma
->vm_start
= region
->vm_start
= addr
;
1165 vma
->vm_end
= region
->vm_end
= addr
+ len
;
1170 vma
->vm_region
= region
;
1172 /* set up the mapping
1173 * - the region is filled in if NOMMU_MAP_DIRECT is still set
1175 if (file
&& vma
->vm_flags
& VM_SHARED
)
1176 ret
= do_mmap_shared_file(vma
);
1178 ret
= do_mmap_private(vma
, region
, len
, capabilities
);
1180 goto error_just_free
;
1181 add_nommu_region(region
);
1183 /* clear anonymous mappings that don't ask for uninitialized data */
1184 if (!vma
->vm_file
&&
1185 (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED
) ||
1186 !(flags
& MAP_UNINITIALIZED
)))
1187 memset((void *)region
->vm_start
, 0,
1188 region
->vm_end
- region
->vm_start
);
1190 /* okay... we have a mapping; now we have to register it */
1191 result
= vma
->vm_start
;
1193 current
->mm
->total_vm
+= len
>> PAGE_SHIFT
;
1196 BUG_ON(!vma
->vm_region
);
1197 vma_iter_config(&vmi
, vma
->vm_start
, vma
->vm_end
);
1198 if (vma_iter_prealloc(&vmi
, vma
))
1199 goto error_just_free
;
1201 setup_vma_to_mm(vma
, current
->mm
);
1202 current
->mm
->map_count
++;
1203 /* add the VMA to the tree */
1204 vma_iter_store(&vmi
, vma
);
1206 /* we flush the region from the icache only when the first executable
1207 * mapping of it is made */
1208 if (vma
->vm_flags
& VM_EXEC
&& !region
->vm_icache_flushed
) {
1209 flush_icache_user_range(region
->vm_start
, region
->vm_end
);
1210 region
->vm_icache_flushed
= true;
1213 up_write(&nommu_region_sem
);
1218 up_write(&nommu_region_sem
);
1220 vma_iter_free(&vmi
);
1221 if (region
->vm_file
)
1222 fput(region
->vm_file
);
1223 kmem_cache_free(vm_region_jar
, region
);
1230 up_write(&nommu_region_sem
);
1231 pr_warn("Attempt to share mismatched mappings\n");
1236 kmem_cache_free(vm_region_jar
, region
);
1237 pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n",
1242 error_getting_region
:
1243 pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n",
1249 unsigned long ksys_mmap_pgoff(unsigned long addr
, unsigned long len
,
1250 unsigned long prot
, unsigned long flags
,
1251 unsigned long fd
, unsigned long pgoff
)
1253 struct file
*file
= NULL
;
1254 unsigned long retval
= -EBADF
;
1256 audit_mmap_fd(fd
, flags
);
1257 if (!(flags
& MAP_ANONYMOUS
)) {
1263 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1271 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1272 unsigned long, prot
, unsigned long, flags
,
1273 unsigned long, fd
, unsigned long, pgoff
)
1275 return ksys_mmap_pgoff(addr
, len
, prot
, flags
, fd
, pgoff
);
1278 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1279 struct mmap_arg_struct
{
1283 unsigned long flags
;
1285 unsigned long offset
;
1288 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1290 struct mmap_arg_struct a
;
1292 if (copy_from_user(&a
, arg
, sizeof(a
)))
1294 if (offset_in_page(a
.offset
))
1297 return ksys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1298 a
.offset
>> PAGE_SHIFT
);
1300 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1303 * split a vma into two pieces at address 'addr', a new vma is allocated either
1304 * for the first part or the tail.
1306 static int split_vma(struct vma_iterator
*vmi
, struct vm_area_struct
*vma
,
1307 unsigned long addr
, int new_below
)
1309 struct vm_area_struct
*new;
1310 struct vm_region
*region
;
1311 unsigned long npages
;
1312 struct mm_struct
*mm
;
1314 /* we're only permitted to split anonymous regions (these should have
1315 * only a single usage on the region) */
1320 if (mm
->map_count
>= sysctl_max_map_count
)
1323 region
= kmem_cache_alloc(vm_region_jar
, GFP_KERNEL
);
1327 new = vm_area_dup(vma
);
1331 /* most fields are the same, copy all, and then fixup */
1332 *region
= *vma
->vm_region
;
1333 new->vm_region
= region
;
1335 npages
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
1338 region
->vm_top
= region
->vm_end
= new->vm_end
= addr
;
1340 region
->vm_start
= new->vm_start
= addr
;
1341 region
->vm_pgoff
= new->vm_pgoff
+= npages
;
1344 vma_iter_config(vmi
, new->vm_start
, new->vm_end
);
1345 if (vma_iter_prealloc(vmi
, vma
)) {
1346 pr_warn("Allocation of vma tree for process %d failed\n",
1348 goto err_vmi_preallocate
;
1351 if (new->vm_ops
&& new->vm_ops
->open
)
1352 new->vm_ops
->open(new);
1354 down_write(&nommu_region_sem
);
1355 delete_nommu_region(vma
->vm_region
);
1357 vma
->vm_region
->vm_start
= vma
->vm_start
= addr
;
1358 vma
->vm_region
->vm_pgoff
= vma
->vm_pgoff
+= npages
;
1360 vma
->vm_region
->vm_end
= vma
->vm_end
= addr
;
1361 vma
->vm_region
->vm_top
= addr
;
1363 add_nommu_region(vma
->vm_region
);
1364 add_nommu_region(new->vm_region
);
1365 up_write(&nommu_region_sem
);
1367 setup_vma_to_mm(vma
, mm
);
1368 setup_vma_to_mm(new, mm
);
1369 vma_iter_store(vmi
, new);
1373 err_vmi_preallocate
:
1376 kmem_cache_free(vm_region_jar
, region
);
1381 * shrink a VMA by removing the specified chunk from either the beginning or
1384 static int vmi_shrink_vma(struct vma_iterator
*vmi
,
1385 struct vm_area_struct
*vma
,
1386 unsigned long from
, unsigned long to
)
1388 struct vm_region
*region
;
1390 /* adjust the VMA's pointers, which may reposition it in the MM's tree
1392 if (from
> vma
->vm_start
) {
1393 if (vma_iter_clear_gfp(vmi
, from
, vma
->vm_end
, GFP_KERNEL
))
1397 if (vma_iter_clear_gfp(vmi
, vma
->vm_start
, to
, GFP_KERNEL
))
1402 /* cut the backing region down to size */
1403 region
= vma
->vm_region
;
1404 BUG_ON(region
->vm_usage
!= 1);
1406 down_write(&nommu_region_sem
);
1407 delete_nommu_region(region
);
1408 if (from
> region
->vm_start
) {
1409 to
= region
->vm_top
;
1410 region
->vm_top
= region
->vm_end
= from
;
1412 region
->vm_start
= to
;
1414 add_nommu_region(region
);
1415 up_write(&nommu_region_sem
);
1417 free_page_series(from
, to
);
1423 * - under NOMMU conditions the chunk to be unmapped must be backed by a single
1424 * VMA, though it need not cover the whole VMA
1426 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
, struct list_head
*uf
)
1428 VMA_ITERATOR(vmi
, mm
, start
);
1429 struct vm_area_struct
*vma
;
1433 len
= PAGE_ALIGN(len
);
1439 /* find the first potentially overlapping VMA */
1440 vma
= vma_find(&vmi
, end
);
1444 pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n",
1445 current
->pid
, current
->comm
,
1446 start
, start
+ len
- 1);
1452 /* we're allowed to split an anonymous VMA but not a file-backed one */
1455 if (start
> vma
->vm_start
)
1457 if (end
== vma
->vm_end
)
1458 goto erase_whole_vma
;
1459 vma
= vma_find(&vmi
, end
);
1463 /* the chunk must be a subset of the VMA found */
1464 if (start
== vma
->vm_start
&& end
== vma
->vm_end
)
1465 goto erase_whole_vma
;
1466 if (start
< vma
->vm_start
|| end
> vma
->vm_end
)
1468 if (offset_in_page(start
))
1470 if (end
!= vma
->vm_end
&& offset_in_page(end
))
1472 if (start
!= vma
->vm_start
&& end
!= vma
->vm_end
) {
1473 ret
= split_vma(&vmi
, vma
, start
, 1);
1477 return vmi_shrink_vma(&vmi
, vma
, start
, end
);
1481 if (delete_vma_from_mm(vma
))
1484 delete_vma(mm
, vma
);
1488 int vm_munmap(unsigned long addr
, size_t len
)
1490 struct mm_struct
*mm
= current
->mm
;
1493 mmap_write_lock(mm
);
1494 ret
= do_munmap(mm
, addr
, len
, NULL
);
1495 mmap_write_unlock(mm
);
1498 EXPORT_SYMBOL(vm_munmap
);
1500 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1502 return vm_munmap(addr
, len
);
1506 * release all the mappings made in a process's VM space
1508 void exit_mmap(struct mm_struct
*mm
)
1510 VMA_ITERATOR(vmi
, mm
, 0);
1511 struct vm_area_struct
*vma
;
1519 * Lock the mm to avoid assert complaining even though this is the only
1522 mmap_write_lock(mm
);
1523 for_each_vma(vmi
, vma
) {
1524 cleanup_vma_from_mm(vma
);
1525 delete_vma(mm
, vma
);
1528 __mt_destroy(&mm
->mm_mt
);
1529 mmap_write_unlock(mm
);
1533 * expand (or shrink) an existing mapping, potentially moving it at the same
1534 * time (controlled by the MREMAP_MAYMOVE flag and available VM space)
1536 * under NOMMU conditions, we only permit changing a mapping's size, and only
1537 * as long as it stays within the region allocated by do_mmap_private() and the
1538 * block is not shareable
1540 * MREMAP_FIXED is not supported under NOMMU conditions
1542 static unsigned long do_mremap(unsigned long addr
,
1543 unsigned long old_len
, unsigned long new_len
,
1544 unsigned long flags
, unsigned long new_addr
)
1546 struct vm_area_struct
*vma
;
1548 /* insanity checks first */
1549 old_len
= PAGE_ALIGN(old_len
);
1550 new_len
= PAGE_ALIGN(new_len
);
1551 if (old_len
== 0 || new_len
== 0)
1552 return (unsigned long) -EINVAL
;
1554 if (offset_in_page(addr
))
1557 if (flags
& MREMAP_FIXED
&& new_addr
!= addr
)
1558 return (unsigned long) -EINVAL
;
1560 vma
= find_vma_exact(current
->mm
, addr
, old_len
);
1562 return (unsigned long) -EINVAL
;
1564 if (vma
->vm_end
!= vma
->vm_start
+ old_len
)
1565 return (unsigned long) -EFAULT
;
1567 if (is_nommu_shared_mapping(vma
->vm_flags
))
1568 return (unsigned long) -EPERM
;
1570 if (new_len
> vma
->vm_region
->vm_end
- vma
->vm_region
->vm_start
)
1571 return (unsigned long) -ENOMEM
;
1573 /* all checks complete - do it */
1574 vma
->vm_end
= vma
->vm_start
+ new_len
;
1575 return vma
->vm_start
;
1578 SYSCALL_DEFINE5(mremap
, unsigned long, addr
, unsigned long, old_len
,
1579 unsigned long, new_len
, unsigned long, flags
,
1580 unsigned long, new_addr
)
1584 mmap_write_lock(current
->mm
);
1585 ret
= do_mremap(addr
, old_len
, new_len
, flags
, new_addr
);
1586 mmap_write_unlock(current
->mm
);
1590 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1591 unsigned int foll_flags
)
1596 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1597 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1599 if (addr
!= (pfn
<< PAGE_SHIFT
))
1602 vm_flags_set(vma
, VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
);
1605 EXPORT_SYMBOL(remap_pfn_range
);
1607 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1609 unsigned long pfn
= start
>> PAGE_SHIFT
;
1610 unsigned long vm_len
= vma
->vm_end
- vma
->vm_start
;
1612 pfn
+= vma
->vm_pgoff
;
1613 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1615 EXPORT_SYMBOL(vm_iomap_memory
);
1617 int remap_vmalloc_range(struct vm_area_struct
*vma
, void *addr
,
1618 unsigned long pgoff
)
1620 unsigned int size
= vma
->vm_end
- vma
->vm_start
;
1622 if (!(vma
->vm_flags
& VM_USERMAP
))
1625 vma
->vm_start
= (unsigned long)(addr
+ (pgoff
<< PAGE_SHIFT
));
1626 vma
->vm_end
= vma
->vm_start
+ size
;
1630 EXPORT_SYMBOL(remap_vmalloc_range
);
1632 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
1637 EXPORT_SYMBOL(filemap_fault
);
1639 vm_fault_t
filemap_map_pages(struct vm_fault
*vmf
,
1640 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
1645 EXPORT_SYMBOL(filemap_map_pages
);
1647 static int __access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1648 void *buf
, int len
, unsigned int gup_flags
)
1650 struct vm_area_struct
*vma
;
1651 int write
= gup_flags
& FOLL_WRITE
;
1653 if (mmap_read_lock_killable(mm
))
1656 /* the access must start within one of the target process's mappings */
1657 vma
= find_vma(mm
, addr
);
1659 /* don't overrun this mapping */
1660 if (addr
+ len
>= vma
->vm_end
)
1661 len
= vma
->vm_end
- addr
;
1663 /* only read or write mappings where it is permitted */
1664 if (write
&& vma
->vm_flags
& VM_MAYWRITE
)
1665 copy_to_user_page(vma
, NULL
, addr
,
1666 (void *) addr
, buf
, len
);
1667 else if (!write
&& vma
->vm_flags
& VM_MAYREAD
)
1668 copy_from_user_page(vma
, NULL
, addr
,
1669 buf
, (void *) addr
, len
);
1676 mmap_read_unlock(mm
);
1682 * access_remote_vm - access another process' address space
1683 * @mm: the mm_struct of the target address space
1684 * @addr: start address to access
1685 * @buf: source or destination buffer
1686 * @len: number of bytes to transfer
1687 * @gup_flags: flags modifying lookup behaviour
1689 * The caller must hold a reference on @mm.
1691 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
1692 void *buf
, int len
, unsigned int gup_flags
)
1694 return __access_remote_vm(mm
, addr
, buf
, len
, gup_flags
);
1698 * Access another process' address space.
1699 * - source/target buffer must be kernel space
1701 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
,
1702 unsigned int gup_flags
)
1704 struct mm_struct
*mm
;
1706 if (addr
+ len
< addr
)
1709 mm
= get_task_mm(tsk
);
1713 len
= __access_remote_vm(mm
, addr
, buf
, len
, gup_flags
);
1718 EXPORT_SYMBOL_GPL(access_process_vm
);
1721 * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode
1722 * @inode: The inode to check
1723 * @size: The current filesize of the inode
1724 * @newsize: The proposed filesize of the inode
1726 * Check the shared mappings on an inode on behalf of a shrinking truncate to
1727 * make sure that any outstanding VMAs aren't broken and then shrink the
1728 * vm_regions that extend beyond so that do_mmap() doesn't
1729 * automatically grant mappings that are too large.
1731 int nommu_shrink_inode_mappings(struct inode
*inode
, size_t size
,
1734 struct vm_area_struct
*vma
;
1735 struct vm_region
*region
;
1737 size_t r_size
, r_top
;
1739 low
= newsize
>> PAGE_SHIFT
;
1740 high
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1742 down_write(&nommu_region_sem
);
1743 i_mmap_lock_read(inode
->i_mapping
);
1745 /* search for VMAs that fall within the dead zone */
1746 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, low
, high
) {
1747 /* found one - only interested if it's shared out of the page
1749 if (vma
->vm_flags
& VM_SHARED
) {
1750 i_mmap_unlock_read(inode
->i_mapping
);
1751 up_write(&nommu_region_sem
);
1752 return -ETXTBSY
; /* not quite true, but near enough */
1756 /* reduce any regions that overlap the dead zone - if in existence,
1757 * these will be pointed to by VMAs that don't overlap the dead zone
1759 * we don't check for any regions that start beyond the EOF as there
1762 vma_interval_tree_foreach(vma
, &inode
->i_mapping
->i_mmap
, 0, ULONG_MAX
) {
1763 if (!(vma
->vm_flags
& VM_SHARED
))
1766 region
= vma
->vm_region
;
1767 r_size
= region
->vm_top
- region
->vm_start
;
1768 r_top
= (region
->vm_pgoff
<< PAGE_SHIFT
) + r_size
;
1770 if (r_top
> newsize
) {
1771 region
->vm_top
-= r_top
- newsize
;
1772 if (region
->vm_end
> region
->vm_top
)
1773 region
->vm_end
= region
->vm_top
;
1777 i_mmap_unlock_read(inode
->i_mapping
);
1778 up_write(&nommu_region_sem
);
1783 * Initialise sysctl_user_reserve_kbytes.
1785 * This is intended to prevent a user from starting a single memory hogging
1786 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1789 * The default value is min(3% of free memory, 128MB)
1790 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1792 static int __meminit
init_user_reserve(void)
1794 unsigned long free_kbytes
;
1796 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
1798 sysctl_user_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 17);
1801 subsys_initcall(init_user_reserve
);
1804 * Initialise sysctl_admin_reserve_kbytes.
1806 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1807 * to log in and kill a memory hogging process.
1809 * Systems with more than 256MB will reserve 8MB, enough to recover
1810 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1811 * only reserve 3% of free pages by default.
1813 static int __meminit
init_admin_reserve(void)
1815 unsigned long free_kbytes
;
1817 free_kbytes
= K(global_zone_page_state(NR_FREE_PAGES
));
1819 sysctl_admin_reserve_kbytes
= min(free_kbytes
/ 32, 1UL << 13);
1822 subsys_initcall(init_admin_reserve
);