]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - mm/mprotect.c
f8421722acb998c488fb45054d06e76b032c6d53
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 <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/cacheflush.h>
29 #include <asm/tlbflush.h>
32 static inline pgprot_t
pgprot_modify(pgprot_t oldprot
, pgprot_t newprot
)
38 static unsigned long change_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
39 unsigned long addr
, unsigned long end
, pgprot_t newprot
,
40 int dirty_accountable
, int prot_numa
)
42 struct mm_struct
*mm
= vma
->vm_mm
;
45 unsigned long pages
= 0;
47 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
48 arch_enter_lazy_mmu_mode();
51 if (pte_present(oldpte
)) {
56 ptent
= ptep_modify_prot_start(mm
, addr
, pte
);
58 ptent
= pte_mknonnuma(ptent
);
59 ptent
= pte_modify(ptent
, newprot
);
65 page
= vm_normal_page(vma
, addr
, oldpte
);
67 if (!pte_numa(oldpte
)) {
68 ptent
= pte_mknuma(ptent
);
69 set_pte_at(mm
, addr
, pte
, ptent
);
76 * Avoid taking write faults for pages we know to be
79 if (dirty_accountable
&& pte_dirty(ptent
)) {
80 ptent
= pte_mkwrite(ptent
);
87 /* Only !prot_numa always clears the pte */
89 ptep_modify_prot_commit(mm
, addr
, pte
, ptent
);
90 } else if (IS_ENABLED(CONFIG_MIGRATION
) && !pte_file(oldpte
)) {
91 swp_entry_t entry
= pte_to_swp_entry(oldpte
);
93 if (is_write_migration_entry(entry
)) {
96 * A protection check is difficult so
97 * just be safe and disable write
99 make_migration_entry_read(&entry
);
100 newpte
= swp_entry_to_pte(entry
);
101 if (pte_swp_soft_dirty(oldpte
))
102 newpte
= pte_swp_mksoft_dirty(newpte
);
103 set_pte_at(mm
, addr
, pte
, newpte
);
108 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
109 arch_leave_lazy_mmu_mode();
110 pte_unmap_unlock(pte
- 1, ptl
);
115 static inline unsigned long change_pmd_range(struct vm_area_struct
*vma
,
116 pud_t
*pud
, unsigned long addr
, unsigned long end
,
117 pgprot_t newprot
, int dirty_accountable
, int prot_numa
)
121 unsigned long pages
= 0;
122 unsigned long nr_huge_updates
= 0;
124 pmd
= pmd_offset(pud
, addr
);
126 unsigned long this_pages
;
128 next
= pmd_addr_end(addr
, end
);
129 if (pmd_trans_huge(*pmd
)) {
130 if (next
- addr
!= HPAGE_PMD_SIZE
)
131 split_huge_page_pmd(vma
, addr
, pmd
);
133 int nr_ptes
= change_huge_pmd(vma
, pmd
, addr
,
137 if (nr_ptes
== HPAGE_PMD_NR
) {
138 pages
+= HPAGE_PMD_NR
;
146 if (pmd_none_or_clear_bad(pmd
))
148 this_pages
= change_pte_range(vma
, pmd
, addr
, next
, newprot
,
149 dirty_accountable
, prot_numa
);
151 } while (pmd
++, addr
= next
, addr
!= end
);
154 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES
, nr_huge_updates
);
158 static inline unsigned long change_pud_range(struct vm_area_struct
*vma
,
159 pgd_t
*pgd
, unsigned long addr
, unsigned long end
,
160 pgprot_t newprot
, int dirty_accountable
, int prot_numa
)
164 unsigned long pages
= 0;
166 pud
= pud_offset(pgd
, addr
);
168 next
= pud_addr_end(addr
, end
);
169 if (pud_none_or_clear_bad(pud
))
171 pages
+= change_pmd_range(vma
, pud
, addr
, next
, newprot
,
172 dirty_accountable
, prot_numa
);
173 } while (pud
++, addr
= next
, addr
!= end
);
178 static unsigned long change_protection_range(struct vm_area_struct
*vma
,
179 unsigned long addr
, unsigned long end
, pgprot_t newprot
,
180 int dirty_accountable
, int prot_numa
)
182 struct mm_struct
*mm
= vma
->vm_mm
;
185 unsigned long start
= addr
;
186 unsigned long pages
= 0;
189 pgd
= pgd_offset(mm
, addr
);
190 flush_cache_range(vma
, addr
, end
);
192 next
= pgd_addr_end(addr
, end
);
193 if (pgd_none_or_clear_bad(pgd
))
195 pages
+= change_pud_range(vma
, pgd
, addr
, next
, newprot
,
196 dirty_accountable
, prot_numa
);
197 } while (pgd
++, addr
= next
, addr
!= end
);
199 /* Only flush the TLB if we actually modified any entries: */
201 flush_tlb_range(vma
, start
, end
);
206 unsigned long change_protection(struct vm_area_struct
*vma
, unsigned long start
,
207 unsigned long end
, pgprot_t newprot
,
208 int dirty_accountable
, int prot_numa
)
210 struct mm_struct
*mm
= vma
->vm_mm
;
213 mmu_notifier_invalidate_range_start(mm
, start
, end
);
214 if (is_vm_hugetlb_page(vma
))
215 pages
= hugetlb_change_protection(vma
, start
, end
, newprot
);
217 pages
= change_protection_range(vma
, start
, end
, newprot
, dirty_accountable
, prot_numa
);
218 mmu_notifier_invalidate_range_end(mm
, start
, end
);
224 mprotect_fixup(struct vm_area_struct
*vma
, struct vm_area_struct
**pprev
,
225 unsigned long start
, unsigned long end
, unsigned long newflags
)
227 struct mm_struct
*mm
= vma
->vm_mm
;
228 unsigned long oldflags
= vma
->vm_flags
;
229 long nrpages
= (end
- start
) >> PAGE_SHIFT
;
230 unsigned long charged
= 0;
233 int dirty_accountable
= 0;
235 if (newflags
== oldflags
) {
241 * If we make a private mapping writable we increase our commit;
242 * but (without finer accounting) cannot reduce our commit if we
243 * make it unwritable again. hugetlb mapping were accounted for
244 * even if read-only so there is no need to account for them here
246 if (newflags
& VM_WRITE
) {
247 if (!(oldflags
& (VM_ACCOUNT
|VM_WRITE
|VM_HUGETLB
|
248 VM_SHARED
|VM_NORESERVE
))) {
250 if (security_vm_enough_memory_mm(mm
, charged
))
252 newflags
|= VM_ACCOUNT
;
257 * First try to merge with previous and/or next vma.
259 pgoff
= vma
->vm_pgoff
+ ((start
- vma
->vm_start
) >> PAGE_SHIFT
);
260 *pprev
= vma_merge(mm
, *pprev
, start
, end
, newflags
,
261 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
269 if (start
!= vma
->vm_start
) {
270 error
= split_vma(mm
, vma
, start
, 1);
275 if (end
!= vma
->vm_end
) {
276 error
= split_vma(mm
, vma
, end
, 0);
283 * vm_flags and vm_page_prot are protected by the mmap_sem
284 * held in write mode.
286 vma
->vm_flags
= newflags
;
287 vma
->vm_page_prot
= pgprot_modify(vma
->vm_page_prot
,
288 vm_get_page_prot(newflags
));
290 if (vma_wants_writenotify(vma
)) {
291 vma
->vm_page_prot
= vm_get_page_prot(newflags
& ~VM_SHARED
);
292 dirty_accountable
= 1;
295 change_protection(vma
, start
, end
, vma
->vm_page_prot
,
296 dirty_accountable
, 0);
298 vm_stat_account(mm
, oldflags
, vma
->vm_file
, -nrpages
);
299 vm_stat_account(mm
, newflags
, vma
->vm_file
, nrpages
);
300 perf_event_mmap(vma
);
304 vm_unacct_memory(charged
);
308 SYSCALL_DEFINE3(mprotect
, unsigned long, start
, size_t, len
,
311 unsigned long vm_flags
, nstart
, end
, tmp
, reqprot
;
312 struct vm_area_struct
*vma
, *prev
;
314 const int grows
= prot
& (PROT_GROWSDOWN
|PROT_GROWSUP
);
315 prot
&= ~(PROT_GROWSDOWN
|PROT_GROWSUP
);
316 if (grows
== (PROT_GROWSDOWN
|PROT_GROWSUP
)) /* can't be both */
319 if (start
& ~PAGE_MASK
)
323 len
= PAGE_ALIGN(len
);
327 if (!arch_validate_prot(prot
))
332 * Does the application expect PROT_READ to imply PROT_EXEC:
334 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
337 vm_flags
= calc_vm_prot_bits(prot
);
339 down_write(¤t
->mm
->mmap_sem
);
341 vma
= find_vma(current
->mm
, start
);
346 if (unlikely(grows
& PROT_GROWSDOWN
)) {
347 if (vma
->vm_start
>= end
)
349 start
= vma
->vm_start
;
351 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
354 if (vma
->vm_start
> start
)
356 if (unlikely(grows
& PROT_GROWSUP
)) {
359 if (!(vma
->vm_flags
& VM_GROWSUP
))
363 if (start
> vma
->vm_start
)
366 for (nstart
= start
; ; ) {
367 unsigned long newflags
;
369 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
372 newflags
|= (vma
->vm_flags
& ~(VM_READ
| VM_WRITE
| VM_EXEC
));
374 /* newflags >> 4 shift VM_MAY% in place of VM_% */
375 if ((newflags
& ~(newflags
>> 4)) & (VM_READ
| VM_WRITE
| VM_EXEC
)) {
380 error
= security_file_mprotect(vma
, reqprot
, prot
);
387 error
= mprotect_fixup(vma
, &prev
, nstart
, tmp
, newflags
);
392 if (nstart
< prev
->vm_end
)
393 nstart
= prev
->vm_end
;
398 if (!vma
|| vma
->vm_start
!= nstart
) {
404 up_write(¤t
->mm
->mmap_sem
);