config: tegra3: enable /dev mount with ACL
[linux-2.6.git] / mm / mprotect.c
1 /*
2  *  mm/mprotect.c
3  *
4  *  (C) Copyright 1994 Linus Torvalds
5  *  (C) Copyright 2002 Christoph Hellwig
6  *
7  *  Address space accounting code       <alan@lxorguk.ukuu.org.uk>
8  *  (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9  */
10
11 #include <linux/mm.h>
12 #include <linux/hugetlb.h>
13 #include <linux/shm.h>
14 #include <linux/mman.h>
15 #include <linux/fs.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>
30
31 #ifndef pgprot_modify
32 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
33 {
34         return newprot;
35 }
36 #endif
37
38 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
39                 unsigned long addr, unsigned long end, pgprot_t newprot,
40                 int dirty_accountable)
41 {
42         pte_t *pte, oldpte;
43         spinlock_t *ptl;
44
45         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
46         arch_enter_lazy_mmu_mode();
47         do {
48                 oldpte = *pte;
49                 if (pte_present(oldpte)) {
50                         pte_t ptent;
51
52                         ptent = ptep_modify_prot_start(mm, addr, pte);
53                         ptent = pte_modify(ptent, newprot);
54
55                         /*
56                          * Avoid taking write faults for pages we know to be
57                          * dirty.
58                          */
59                         if (dirty_accountable && pte_dirty(ptent))
60                                 ptent = pte_mkwrite(ptent);
61
62                         ptep_modify_prot_commit(mm, addr, pte, ptent);
63                 } else if (PAGE_MIGRATION && !pte_file(oldpte)) {
64                         swp_entry_t entry = pte_to_swp_entry(oldpte);
65
66                         if (is_write_migration_entry(entry)) {
67                                 /*
68                                  * A protection check is difficult so
69                                  * just be safe and disable write
70                                  */
71                                 make_migration_entry_read(&entry);
72                                 set_pte_at(mm, addr, pte,
73                                         swp_entry_to_pte(entry));
74                         }
75                 }
76         } while (pte++, addr += PAGE_SIZE, addr != end);
77         arch_leave_lazy_mmu_mode();
78         pte_unmap_unlock(pte - 1, ptl);
79 }
80
81 static inline void change_pmd_range(struct vm_area_struct *vma, pud_t *pud,
82                 unsigned long addr, unsigned long end, pgprot_t newprot,
83                 int dirty_accountable)
84 {
85         pmd_t *pmd;
86         unsigned long next;
87
88         pmd = pmd_offset(pud, addr);
89         do {
90                 next = pmd_addr_end(addr, end);
91                 if (pmd_trans_huge(*pmd)) {
92                         if (next - addr != HPAGE_PMD_SIZE)
93                                 split_huge_page_pmd(vma->vm_mm, pmd);
94                         else if (change_huge_pmd(vma, pmd, addr, newprot))
95                                 continue;
96                         /* fall through */
97                 }
98                 if (pmd_none_or_clear_bad(pmd))
99                         continue;
100                 change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
101                                  dirty_accountable);
102         } while (pmd++, addr = next, addr != end);
103 }
104
105 static inline void change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
106                 unsigned long addr, unsigned long end, pgprot_t newprot,
107                 int dirty_accountable)
108 {
109         pud_t *pud;
110         unsigned long next;
111
112         pud = pud_offset(pgd, addr);
113         do {
114                 next = pud_addr_end(addr, end);
115                 if (pud_none_or_clear_bad(pud))
116                         continue;
117                 change_pmd_range(vma, pud, addr, next, newprot,
118                                  dirty_accountable);
119         } while (pud++, addr = next, addr != end);
120 }
121
122 static void change_protection(struct vm_area_struct *vma,
123                 unsigned long addr, unsigned long end, pgprot_t newprot,
124                 int dirty_accountable)
125 {
126         struct mm_struct *mm = vma->vm_mm;
127         pgd_t *pgd;
128         unsigned long next;
129         unsigned long start = addr;
130
131         BUG_ON(addr >= end);
132         pgd = pgd_offset(mm, addr);
133         flush_cache_range(vma, addr, end);
134         do {
135                 next = pgd_addr_end(addr, end);
136                 if (pgd_none_or_clear_bad(pgd))
137                         continue;
138                 change_pud_range(vma, pgd, addr, next, newprot,
139                                  dirty_accountable);
140         } while (pgd++, addr = next, addr != end);
141         flush_tlb_range(vma, start, end);
142 }
143
144 int
145 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
146         unsigned long start, unsigned long end, unsigned long newflags)
147 {
148         struct mm_struct *mm = vma->vm_mm;
149         unsigned long oldflags = vma->vm_flags;
150         long nrpages = (end - start) >> PAGE_SHIFT;
151         unsigned long charged = 0;
152         pgoff_t pgoff;
153         int error;
154         int dirty_accountable = 0;
155
156         if (newflags == oldflags) {
157                 *pprev = vma;
158                 return 0;
159         }
160
161         /*
162          * If we make a private mapping writable we increase our commit;
163          * but (without finer accounting) cannot reduce our commit if we
164          * make it unwritable again. hugetlb mapping were accounted for
165          * even if read-only so there is no need to account for them here
166          */
167         if (newflags & VM_WRITE) {
168                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
169                                                 VM_SHARED|VM_NORESERVE))) {
170                         charged = nrpages;
171                         if (security_vm_enough_memory(charged))
172                                 return -ENOMEM;
173                         newflags |= VM_ACCOUNT;
174                 }
175         }
176
177         /*
178          * First try to merge with previous and/or next vma.
179          */
180         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
181         *pprev = vma_merge(mm, *pprev, start, end, newflags,
182                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
183         if (*pprev) {
184                 vma = *pprev;
185                 goto success;
186         }
187
188         *pprev = vma;
189
190         if (start != vma->vm_start) {
191                 error = split_vma(mm, vma, start, 1);
192                 if (error)
193                         goto fail;
194         }
195
196         if (end != vma->vm_end) {
197                 error = split_vma(mm, vma, end, 0);
198                 if (error)
199                         goto fail;
200         }
201
202 success:
203         /*
204          * vm_flags and vm_page_prot are protected by the mmap_sem
205          * held in write mode.
206          */
207         vma->vm_flags = newflags;
208         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
209                                           vm_get_page_prot(newflags));
210
211         if (vma_wants_writenotify(vma)) {
212                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
213                 dirty_accountable = 1;
214         }
215
216         mmu_notifier_invalidate_range_start(mm, start, end);
217         if (is_vm_hugetlb_page(vma))
218                 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
219         else
220                 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
221         mmu_notifier_invalidate_range_end(mm, start, end);
222         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
223         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
224         perf_event_mmap(vma);
225         return 0;
226
227 fail:
228         vm_unacct_memory(charged);
229         return error;
230 }
231
232 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
233                 unsigned long, prot)
234 {
235         unsigned long vm_flags, nstart, end, tmp, reqprot;
236         struct vm_area_struct *vma, *prev;
237         int error = -EINVAL;
238         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
239         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
240         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
241                 return -EINVAL;
242
243         if (start & ~PAGE_MASK)
244                 return -EINVAL;
245         if (!len)
246                 return 0;
247         len = PAGE_ALIGN(len);
248         end = start + len;
249         if (end <= start)
250                 return -ENOMEM;
251         if (!arch_validate_prot(prot))
252                 return -EINVAL;
253
254         reqprot = prot;
255         /*
256          * Does the application expect PROT_READ to imply PROT_EXEC:
257          */
258         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
259                 prot |= PROT_EXEC;
260
261         vm_flags = calc_vm_prot_bits(prot);
262
263         down_write(&current->mm->mmap_sem);
264
265         vma = find_vma_prev(current->mm, start, &prev);
266         error = -ENOMEM;
267         if (!vma)
268                 goto out;
269         if (unlikely(grows & PROT_GROWSDOWN)) {
270                 if (vma->vm_start >= end)
271                         goto out;
272                 start = vma->vm_start;
273                 error = -EINVAL;
274                 if (!(vma->vm_flags & VM_GROWSDOWN))
275                         goto out;
276         }
277         else {
278                 if (vma->vm_start > start)
279                         goto out;
280                 if (unlikely(grows & PROT_GROWSUP)) {
281                         end = vma->vm_end;
282                         error = -EINVAL;
283                         if (!(vma->vm_flags & VM_GROWSUP))
284                                 goto out;
285                 }
286         }
287         if (start > vma->vm_start)
288                 prev = vma;
289
290         for (nstart = start ; ; ) {
291                 unsigned long newflags;
292
293                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
294
295                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
296
297                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
298                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
299                         error = -EACCES;
300                         goto out;
301                 }
302
303                 error = security_file_mprotect(vma, reqprot, prot);
304                 if (error)
305                         goto out;
306
307                 tmp = vma->vm_end;
308                 if (tmp > end)
309                         tmp = end;
310                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
311                 if (error)
312                         goto out;
313                 nstart = tmp;
314
315                 if (nstart < prev->vm_end)
316                         nstart = prev->vm_end;
317                 if (nstart >= end)
318                         goto out;
319
320                 vma = prev->vm_next;
321                 if (!vma || vma->vm_start != nstart) {
322                         error = -ENOMEM;
323                         goto out;
324                 }
325         }
326 out:
327         up_write(&current->mm->mmap_sem);
328         return error;
329 }