upgrade to fedora-2.6.12-1.1398.FC4 + vserver 2.0.rc7
[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@redhat.com>
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/slab.h>
14 #include <linux/shm.h>
15 #include <linux/mman.h>
16 #include <linux/fs.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>
22
23 #include <asm/uaccess.h>
24 #include <asm/pgtable.h>
25 #include <asm/pgalloc.h>
26 #include <asm/cacheflush.h>
27 #include <asm/tlbflush.h>
28
29 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
30                 unsigned long addr, unsigned long end, pgprot_t newprot)
31 {
32         pte_t *pte;
33
34         pte = pte_offset_map(pmd, addr);
35         do {
36                 if (pte_present(*pte)) {
37                         pte_t ptent;
38
39                         /* Avoid an SMP race with hardware updated dirty/clean
40                          * bits by wiping the pte and then setting the new pte
41                          * into place.
42                          */
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);
46                 }
47         } while (pte++, addr += PAGE_SIZE, addr != end);
48         pte_unmap(pte - 1);
49 }
50
51 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
52                 unsigned long addr, unsigned long end, pgprot_t newprot)
53 {
54         pmd_t *pmd;
55         unsigned long next;
56
57         pmd = pmd_offset(pud, addr);
58         do {
59                 next = pmd_addr_end(addr, end);
60                 if (pmd_none_or_clear_bad(pmd))
61                         continue;
62                 change_pte_range(mm, pmd, addr, next, newprot);
63         } while (pmd++, addr = next, addr != end);
64 }
65
66 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
67                 unsigned long addr, unsigned long end, pgprot_t newprot)
68 {
69         pud_t *pud;
70         unsigned long next;
71
72         pud = pud_offset(pgd, addr);
73         do {
74                 next = pud_addr_end(addr, end);
75                 if (pud_none_or_clear_bad(pud))
76                         continue;
77                 change_pmd_range(mm, pud, addr, next, newprot);
78         } while (pud++, addr = next, addr != end);
79 }
80
81 static void change_protection(struct vm_area_struct *vma,
82                 unsigned long addr, unsigned long end, pgprot_t newprot)
83 {
84         struct mm_struct *mm = vma->vm_mm;
85         pgd_t *pgd;
86         unsigned long next;
87         unsigned long start = addr;
88
89         BUG_ON(addr >= end);
90         pgd = pgd_offset(mm, addr);
91         flush_cache_range(vma, addr, end);
92         spin_lock(&mm->page_table_lock);
93         do {
94                 next = pgd_addr_end(addr, end);
95                 if (pgd_none_or_clear_bad(pgd))
96                         continue;
97                 change_pud_range(mm, pgd, addr, next, newprot);
98         } while (pgd++, addr = next, addr != end);
99         flush_tlb_range(vma, start, end);
100         spin_unlock(&mm->page_table_lock);
101 }
102
103 static int
104 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
105         unsigned long start, unsigned long end, unsigned long newflags)
106 {
107         struct mm_struct *mm = vma->vm_mm;
108         unsigned long oldflags = vma->vm_flags;
109         long nrpages = (end - start) >> PAGE_SHIFT;
110         unsigned long charged = 0, old_end = vma->vm_end;
111         pgprot_t newprot;
112         pgoff_t pgoff;
113         int error;
114
115         if (newflags == oldflags) {
116                 *pprev = vma;
117                 return 0;
118         }
119
120         /*
121          * If we make a private mapping writable we increase our commit;
122          * but (without finer accounting) cannot reduce our commit if we
123          * make it unwritable again.
124          *
125          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
126          * a MAP_NORESERVE private mapping to writable will now reserve.
127          */
128         if (newflags & VM_WRITE) {
129                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
130                         charged = nrpages;
131                         if (security_vm_enough_memory(charged))
132                                 return -ENOMEM;
133                         newflags |= VM_ACCOUNT;
134                 }
135         }
136
137         newprot = protection_map[newflags & 0xf];
138
139         /*
140          * First try to merge with previous and/or next vma.
141          */
142         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
143         *pprev = vma_merge(mm, *pprev, start, end, newflags,
144                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
145         if (*pprev) {
146                 vma = *pprev;
147                 goto success;
148         }
149
150         *pprev = vma;
151
152         if (start != vma->vm_start) {
153                 error = split_vma(mm, vma, start, 1);
154                 if (error)
155                         goto fail;
156         }
157
158         if (end != vma->vm_end) {
159                 error = split_vma(mm, vma, end, 0);
160                 if (error)
161                         goto fail;
162         }
163
164 success:
165         /*
166          * vm_flags and vm_page_prot are protected by the mmap_sem
167          * held in write mode.
168          */
169         vma->vm_flags = newflags;
170         vma->vm_page_prot = newprot;
171         if (oldflags & VM_EXEC)
172                 arch_remove_exec_range(current->mm, old_end);
173         change_protection(vma, start, end, newprot);
174         __vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
175         __vm_stat_account(mm, newflags, vma->vm_file, nrpages);
176         return 0;
177
178 fail:
179         vm_unacct_memory(charged);
180         return error;
181 }
182
183 asmlinkage long
184 do_mprotect(struct mm_struct *mm, unsigned long start, size_t len, 
185              unsigned long prot)
186 {
187         unsigned long vm_flags, nstart, end, tmp, reqprot;
188         struct vm_area_struct *vma, *prev;
189         int error = -EINVAL;
190         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
191         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
192         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
193                 return -EINVAL;
194
195         if (start & ~PAGE_MASK)
196                 return -EINVAL;
197         if (!len)
198                 return 0;
199         len = PAGE_ALIGN(len);
200         end = start + len;
201         if (end <= start)
202                 return -ENOMEM;
203         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
204                 return -EINVAL;
205
206         reqprot = prot;
207         /*
208          * Does the application expect PROT_READ to imply PROT_EXEC:
209          */
210         if (unlikely((prot & PROT_READ) &&
211                         (current->personality & READ_IMPLIES_EXEC)))
212                 prot |= PROT_EXEC;
213
214         vm_flags = calc_vm_prot_bits(prot);
215
216         down_write(&mm->mmap_sem);
217
218         vma = find_vma_prev(mm, start, &prev);
219         error = -ENOMEM;
220         if (!vma)
221                 goto out;
222         if (unlikely(grows & PROT_GROWSDOWN)) {
223                 if (vma->vm_start >= end)
224                         goto out;
225                 start = vma->vm_start;
226                 error = -EINVAL;
227                 if (!(vma->vm_flags & VM_GROWSDOWN))
228                         goto out;
229         }
230         else {
231                 if (vma->vm_start > start)
232                         goto out;
233                 if (unlikely(grows & PROT_GROWSUP)) {
234                         end = vma->vm_end;
235                         error = -EINVAL;
236                         if (!(vma->vm_flags & VM_GROWSUP))
237                                 goto out;
238                 }
239         }
240         if (start > vma->vm_start)
241                 prev = vma;
242
243         for (nstart = start ; ; ) {
244                 unsigned long newflags;
245
246                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
247
248                 if (is_vm_hugetlb_page(vma)) {
249                         error = -EACCES;
250                         goto out;
251                 }
252
253                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
254
255                 if ((newflags & ~(newflags >> 4)) & 0xf) {
256                         error = -EACCES;
257                         goto out;
258                 }
259
260                 error = security_file_mprotect(vma, reqprot, prot);
261                 if (error)
262                         goto out;
263
264                 tmp = vma->vm_end;
265                 if (tmp > end)
266                         tmp = end;
267                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
268                 if (error)
269                         goto out;
270                 nstart = tmp;
271
272                 if (nstart < prev->vm_end)
273                         nstart = prev->vm_end;
274                 if (nstart >= end)
275                         goto out;
276
277                 vma = prev->vm_next;
278                 if (!vma || vma->vm_start != nstart) {
279                         error = -ENOMEM;
280                         goto out;
281                 }
282         }
283 out:
284         up_write(&mm->mmap_sem);
285         return error;
286 }
287
288 asmlinkage long sys_mprotect(unsigned long start, size_t len, unsigned long prot)
289 {
290         return(do_mprotect(current->mm, start, len, prot));
291 }