4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
12 #include <linux/hugetlb.h>
13 #include <linux/shm.h>
14 #include <linux/mman.h>
16 #include <linux/highmem.h>
17 #include <linux/security.h>
18 #include <linux/mempolicy.h>
19 #include <linux/personality.h>
20 #include <linux/syscalls.h>
21 #include <linux/swap.h>
22 #include <linux/swapops.h>
23 #include <linux/mmu_notifier.h>
24 #include <linux/migrate.h>
25 #include <linux/perf_event.h>
26 #include <linux/ksm.h>
27 #include <asm/uaccess.h>
28 #include <asm/pgtable.h>
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
33 static inline pgprot_t
pgprot_modify(pgprot_t oldprot
, pgprot_t newprot
)
39 static unsigned long change_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
40 unsigned long addr
, unsigned long end
, pgprot_t newprot
,
41 int dirty_accountable
, int prot_numa
)
43 struct mm_struct
*mm
= vma
->vm_mm
;
46 unsigned long pages
= 0;
48 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
49 arch_enter_lazy_mmu_mode();
52 if (pte_present(oldpte
)) {
57 ptent
= ptep_modify_prot_start(mm
, addr
, pte
);
59 ptent
= pte_mknonnuma(ptent
);
60 ptent
= pte_modify(ptent
, newprot
);
62 * Avoid taking write faults for pages we
65 if (dirty_accountable
&& pte_dirty(ptent
))
66 ptent
= pte_mkwrite(ptent
);
67 ptep_modify_prot_commit(mm
, addr
, pte
, ptent
);
72 page
= vm_normal_page(vma
, addr
, oldpte
);
73 if (page
&& !PageKsm(page
)) {
74 if (!pte_numa(oldpte
)) {
75 ptep_set_numa(mm
, addr
, pte
);
82 } else if (IS_ENABLED(CONFIG_MIGRATION
) && !pte_file(oldpte
)) {
83 swp_entry_t entry
= pte_to_swp_entry(oldpte
);
85 if (is_write_migration_entry(entry
)) {
88 * A protection check is difficult so
89 * just be safe and disable write
91 make_migration_entry_read(&entry
);
92 newpte
= swp_entry_to_pte(entry
);
93 if (pte_swp_soft_dirty(oldpte
))
94 newpte
= pte_swp_mksoft_dirty(newpte
);
95 set_pte_at(mm
, addr
, pte
, newpte
);
100 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
101 arch_leave_lazy_mmu_mode();
102 pte_unmap_unlock(pte
- 1, ptl
);
107 static inline unsigned long change_pmd_range(struct vm_area_struct
*vma
,
108 pud_t
*pud
, unsigned long addr
, unsigned long end
,
109 pgprot_t newprot
, int dirty_accountable
, int prot_numa
)
113 unsigned long pages
= 0;
114 unsigned long nr_huge_updates
= 0;
116 pmd
= pmd_offset(pud
, addr
);
118 unsigned long this_pages
;
120 next
= pmd_addr_end(addr
, end
);
121 if (pmd_trans_huge(*pmd
)) {
122 if (next
- addr
!= HPAGE_PMD_SIZE
)
123 split_huge_page_pmd(vma
, addr
, pmd
);
125 int nr_ptes
= change_huge_pmd(vma
, pmd
, addr
,
129 if (nr_ptes
== HPAGE_PMD_NR
) {
130 pages
+= HPAGE_PMD_NR
;
138 if (pmd_none_or_clear_bad(pmd
))
140 this_pages
= change_pte_range(vma
, pmd
, addr
, next
, newprot
,
141 dirty_accountable
, prot_numa
);
143 } while (pmd
++, addr
= next
, addr
!= end
);
146 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES
, nr_huge_updates
);
150 static inline unsigned long change_pud_range(struct vm_area_struct
*vma
,
151 pgd_t
*pgd
, unsigned long addr
, unsigned long end
,
152 pgprot_t newprot
, int dirty_accountable
, int prot_numa
)
156 unsigned long pages
= 0;
158 pud
= pud_offset(pgd
, addr
);
160 next
= pud_addr_end(addr
, end
);
161 if (pud_none_or_clear_bad(pud
))
163 pages
+= change_pmd_range(vma
, pud
, addr
, next
, newprot
,
164 dirty_accountable
, prot_numa
);
165 } while (pud
++, addr
= next
, addr
!= end
);
170 static unsigned long change_protection_range(struct vm_area_struct
*vma
,
171 unsigned long addr
, unsigned long end
, pgprot_t newprot
,
172 int dirty_accountable
, int prot_numa
)
174 struct mm_struct
*mm
= vma
->vm_mm
;
177 unsigned long start
= addr
;
178 unsigned long pages
= 0;
181 pgd
= pgd_offset(mm
, addr
);
182 flush_cache_range(vma
, addr
, end
);
183 set_tlb_flush_pending(mm
);
185 next
= pgd_addr_end(addr
, end
);
186 if (pgd_none_or_clear_bad(pgd
))
188 pages
+= change_pud_range(vma
, pgd
, addr
, next
, newprot
,
189 dirty_accountable
, prot_numa
);
190 } while (pgd
++, addr
= next
, addr
!= end
);
192 /* Only flush the TLB if we actually modified any entries: */
194 flush_tlb_range(vma
, start
, end
);
195 clear_tlb_flush_pending(mm
);
200 unsigned long change_protection(struct vm_area_struct
*vma
, unsigned long start
,
201 unsigned long end
, pgprot_t newprot
,
202 int dirty_accountable
, int prot_numa
)
204 struct mm_struct
*mm
= vma
->vm_mm
;
207 mmu_notifier_invalidate_range_start(mm
, start
, end
);
208 if (is_vm_hugetlb_page(vma
))
209 pages
= hugetlb_change_protection(vma
, start
, end
, newprot
);
211 pages
= change_protection_range(vma
, start
, end
, newprot
, dirty_accountable
, prot_numa
);
212 mmu_notifier_invalidate_range_end(mm
, start
, end
);
218 mprotect_fixup(struct vm_area_struct
*vma
, struct vm_area_struct
**pprev
,
219 unsigned long start
, unsigned long end
, unsigned long newflags
)
221 struct mm_struct
*mm
= vma
->vm_mm
;
222 unsigned long oldflags
= vma
->vm_flags
;
223 long nrpages
= (end
- start
) >> PAGE_SHIFT
;
224 unsigned long charged
= 0;
227 int dirty_accountable
= 0;
229 if (newflags
== oldflags
) {
235 * If we make a private mapping writable we increase our commit;
236 * but (without finer accounting) cannot reduce our commit if we
237 * make it unwritable again. hugetlb mapping were accounted for
238 * even if read-only so there is no need to account for them here
240 if (newflags
& VM_WRITE
) {
241 if (!(oldflags
& (VM_ACCOUNT
|VM_WRITE
|VM_HUGETLB
|
242 VM_SHARED
|VM_NORESERVE
))) {
244 if (security_vm_enough_memory_mm(mm
, charged
))
246 newflags
|= VM_ACCOUNT
;
251 * First try to merge with previous and/or next vma.
253 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
254 *pprev
= vma_merge(mm
, *pprev
, start
, end
, newflags
,
255 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
263 if (start
!= vma
->vm_start
) {
264 error
= split_vma(mm
, vma
, start
, 1);
269 if (end
!= vma
->vm_end
) {
270 error
= split_vma(mm
, vma
, end
, 0);
277 * vm_flags and vm_page_prot are protected by the mmap_sem
278 * held in write mode.
280 vma
->vm_flags
= newflags
;
281 vma
->vm_page_prot
= pgprot_modify(vma
->vm_page_prot
,
282 vm_get_page_prot(newflags
));
284 if (vma_wants_writenotify(vma
)) {
285 vma
->vm_page_prot
= vm_get_page_prot(newflags
& ~VM_SHARED
);
286 dirty_accountable
= 1;
289 change_protection(vma
, start
, end
, vma
->vm_page_prot
,
290 dirty_accountable
, 0);
292 vm_stat_account(mm
, oldflags
, vma
->vm_file
, -nrpages
);
293 vm_stat_account(mm
, newflags
, vma
->vm_file
, nrpages
);
294 perf_event_mmap(vma
);
298 vm_unacct_memory(charged
);
302 SYSCALL_DEFINE3(mprotect
, unsigned long, start
, size_t, len
,
305 unsigned long vm_flags
, nstart
, end
, tmp
, reqprot
;
306 struct vm_area_struct
*vma
, *prev
;
308 const int grows
= prot
& (PROT_GROWSDOWN
|PROT_GROWSUP
);
309 prot
&= ~(PROT_GROWSDOWN
|PROT_GROWSUP
);
310 if (grows
== (PROT_GROWSDOWN
|PROT_GROWSUP
)) /* can't be both */
313 if (start
& ~PAGE_MASK
)
317 len
= PAGE_ALIGN(len
);
321 if (!arch_validate_prot(prot
))
326 * Does the application expect PROT_READ to imply PROT_EXEC:
328 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
331 vm_flags
= calc_vm_prot_bits(prot
);
333 down_write(¤t
->mm
->mmap_sem
);
335 vma
= find_vma(current
->mm
, start
);
340 if (unlikely(grows
& PROT_GROWSDOWN
)) {
341 if (vma
->vm_start
>= end
)
343 start
= vma
->vm_start
;
345 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
348 if (vma
->vm_start
> start
)
350 if (unlikely(grows
& PROT_GROWSUP
)) {
353 if (!(vma
->vm_flags
& VM_GROWSUP
))
357 if (start
> vma
->vm_start
)
360 for (nstart
= start
; ; ) {
361 unsigned long newflags
;
363 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
366 newflags
|= (vma
->vm_flags
& ~(VM_READ
| VM_WRITE
| VM_EXEC
));
368 /* newflags >> 4 shift VM_MAY% in place of VM_% */
369 if ((newflags
& ~(newflags
>> 4)) & (VM_READ
| VM_WRITE
| VM_EXEC
)) {
374 error
= security_file_mprotect(vma
, reqprot
, prot
);
381 error
= mprotect_fixup(vma
, &prev
, nstart
, tmp
, newflags
);
386 if (nstart
< prev
->vm_end
)
387 nstart
= prev
->vm_end
;
392 if (!vma
|| vma
->vm_start
!= nstart
) {
398 up_write(¤t
->mm
->mmap_sem
);