4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
7 * Address space accounting code <alan@redhat.com>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
12 #include <linux/hugetlb.h>
13 #include <linux/slab.h>
14 #include <linux/shm.h>
15 #include <linux/mman.h>
17 #include <linux/highmem.h>
18 #include <linux/security.h>
19 #include <linux/mempolicy.h>
20 #include <linux/personality.h>
21 #include <linux/syscalls.h>
23 #include <asm/uaccess.h>
24 #include <asm/pgtable.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
28 static void change_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
29 unsigned long addr
, unsigned long end
, pgprot_t newprot
)
34 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
36 if (pte_present(*pte
)) {
39 /* Avoid an SMP race with hardware updated dirty/clean
40 * bits by wiping the pte and then setting the new pte
43 ptent
= pte_modify(ptep_get_and_clear(mm
, addr
, pte
), newprot
);
44 set_pte_at(mm
, addr
, pte
, ptent
);
45 lazy_mmu_prot_update(ptent
);
47 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
48 pte_unmap_unlock(pte
- 1, ptl
);
51 static inline void change_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
52 unsigned long addr
, unsigned long end
, pgprot_t newprot
)
57 pmd
= pmd_offset(pud
, addr
);
59 next
= pmd_addr_end(addr
, end
);
60 if (pmd_none_or_clear_bad(pmd
))
62 change_pte_range(mm
, pmd
, addr
, next
, newprot
);
63 } while (pmd
++, addr
= next
, addr
!= end
);
66 static inline void change_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
67 unsigned long addr
, unsigned long end
, pgprot_t newprot
)
72 pud
= pud_offset(pgd
, addr
);
74 next
= pud_addr_end(addr
, end
);
75 if (pud_none_or_clear_bad(pud
))
77 change_pmd_range(mm
, pud
, addr
, next
, newprot
);
78 } while (pud
++, addr
= next
, addr
!= end
);
81 static void change_protection(struct vm_area_struct
*vma
,
82 unsigned long addr
, unsigned long end
, pgprot_t newprot
)
84 struct mm_struct
*mm
= vma
->vm_mm
;
87 unsigned long start
= addr
;
90 pgd
= pgd_offset(mm
, addr
);
91 flush_cache_range(vma
, addr
, end
);
93 next
= pgd_addr_end(addr
, end
);
94 if (pgd_none_or_clear_bad(pgd
))
96 change_pud_range(mm
, pgd
, addr
, next
, newprot
);
97 } while (pgd
++, addr
= next
, addr
!= end
);
98 flush_tlb_range(vma
, start
, end
);
102 mprotect_fixup(struct vm_area_struct
*vma
, struct vm_area_struct
**pprev
,
103 unsigned long start
, unsigned long end
, unsigned long newflags
)
105 struct mm_struct
*mm
= vma
->vm_mm
;
106 unsigned long oldflags
= vma
->vm_flags
;
107 long nrpages
= (end
- start
) >> PAGE_SHIFT
;
108 unsigned long charged
= 0;
113 if (newflags
== oldflags
) {
119 * If we make a private mapping writable we increase our commit;
120 * but (without finer accounting) cannot reduce our commit if we
121 * make it unwritable again.
123 * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
124 * a MAP_NORESERVE private mapping to writable will now reserve.
126 if (newflags
& VM_WRITE
) {
127 if (!(oldflags
& (VM_ACCOUNT
|VM_WRITE
|VM_SHARED
|VM_HUGETLB
))) {
129 if (security_vm_enough_memory(charged
))
131 newflags
|= VM_ACCOUNT
;
135 newprot
= protection_map
[newflags
& 0xf];
138 * First try to merge with previous and/or next vma.
140 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
141 *pprev
= vma_merge(mm
, *pprev
, start
, end
, newflags
,
142 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
150 if (start
!= vma
->vm_start
) {
151 error
= split_vma(mm
, vma
, start
, 1);
156 if (end
!= vma
->vm_end
) {
157 error
= split_vma(mm
, vma
, end
, 0);
164 * vm_flags and vm_page_prot are protected by the mmap_sem
165 * held in write mode.
167 vma
->vm_flags
= newflags
;
168 vma
->vm_page_prot
= newprot
;
169 change_protection(vma
, start
, end
, newprot
);
170 vm_stat_account(mm
, oldflags
, vma
->vm_file
, -nrpages
);
171 vm_stat_account(mm
, newflags
, vma
->vm_file
, nrpages
);
175 vm_unacct_memory(charged
);
180 sys_mprotect(unsigned long start
, size_t len
, unsigned long prot
)
182 unsigned long vm_flags
, nstart
, end
, tmp
, reqprot
;
183 struct vm_area_struct
*vma
, *prev
;
185 const int grows
= prot
& (PROT_GROWSDOWN
|PROT_GROWSUP
);
186 prot
&= ~(PROT_GROWSDOWN
|PROT_GROWSUP
);
187 if (grows
== (PROT_GROWSDOWN
|PROT_GROWSUP
)) /* can't be both */
190 if (start
& ~PAGE_MASK
)
194 len
= PAGE_ALIGN(len
);
198 if (prot
& ~(PROT_READ
| PROT_WRITE
| PROT_EXEC
| PROT_SEM
))
203 * Does the application expect PROT_READ to imply PROT_EXEC:
205 if (unlikely((prot
& PROT_READ
) &&
206 (current
->personality
& READ_IMPLIES_EXEC
)))
209 vm_flags
= calc_vm_prot_bits(prot
);
211 down_write(¤t
->mm
->mmap_sem
);
213 vma
= find_vma_prev(current
->mm
, start
, &prev
);
217 if (unlikely(grows
& PROT_GROWSDOWN
)) {
218 if (vma
->vm_start
>= end
)
220 start
= vma
->vm_start
;
222 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
226 if (vma
->vm_start
> start
)
228 if (unlikely(grows
& PROT_GROWSUP
)) {
231 if (!(vma
->vm_flags
& VM_GROWSUP
))
235 if (start
> vma
->vm_start
)
238 for (nstart
= start
; ; ) {
239 unsigned long newflags
;
241 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
243 if (is_vm_hugetlb_page(vma
)) {
248 newflags
= vm_flags
| (vma
->vm_flags
& ~(VM_READ
| VM_WRITE
| VM_EXEC
));
250 /* newflags >> 4 shift VM_MAY% in place of VM_% */
251 if ((newflags
& ~(newflags
>> 4)) & (VM_READ
| VM_WRITE
| VM_EXEC
)) {
256 error
= security_file_mprotect(vma
, reqprot
, prot
);
263 error
= mprotect_fixup(vma
, &prev
, nstart
, tmp
, newflags
);
268 if (nstart
< prev
->vm_end
)
269 nstart
= prev
->vm_end
;
274 if (!vma
|| vma
->vm_start
!= nstart
) {
280 up_write(¤t
->mm
->mmap_sem
);