[PATCH] x86: entry.S trap return fixes
[linux-2.6] / 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/cacheflush.h>
26 #include <asm/tlbflush.h>
27
28 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
29                 unsigned long addr, unsigned long end, pgprot_t newprot)
30 {
31         pte_t *pte;
32
33         pte = pte_offset_map(pmd, addr);
34         do {
35                 if (pte_present(*pte)) {
36                         pte_t ptent;
37
38                         /* Avoid an SMP race with hardware updated dirty/clean
39                          * bits by wiping the pte and then setting the new pte
40                          * into place.
41                          */
42                         ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
43                         set_pte_at(mm, addr, pte, ptent);
44                         lazy_mmu_prot_update(ptent);
45                 }
46         } while (pte++, addr += PAGE_SIZE, addr != end);
47         pte_unmap(pte - 1);
48 }
49
50 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
51                 unsigned long addr, unsigned long end, pgprot_t newprot)
52 {
53         pmd_t *pmd;
54         unsigned long next;
55
56         pmd = pmd_offset(pud, addr);
57         do {
58                 next = pmd_addr_end(addr, end);
59                 if (pmd_none_or_clear_bad(pmd))
60                         continue;
61                 change_pte_range(mm, pmd, addr, next, newprot);
62         } while (pmd++, addr = next, addr != end);
63 }
64
65 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
66                 unsigned long addr, unsigned long end, pgprot_t newprot)
67 {
68         pud_t *pud;
69         unsigned long next;
70
71         pud = pud_offset(pgd, addr);
72         do {
73                 next = pud_addr_end(addr, end);
74                 if (pud_none_or_clear_bad(pud))
75                         continue;
76                 change_pmd_range(mm, pud, addr, next, newprot);
77         } while (pud++, addr = next, addr != end);
78 }
79
80 static void change_protection(struct vm_area_struct *vma,
81                 unsigned long addr, unsigned long end, pgprot_t newprot)
82 {
83         struct mm_struct *mm = vma->vm_mm;
84         pgd_t *pgd;
85         unsigned long next;
86         unsigned long start = addr;
87
88         BUG_ON(addr >= end);
89         pgd = pgd_offset(mm, addr);
90         flush_cache_range(vma, addr, end);
91         spin_lock(&mm->page_table_lock);
92         do {
93                 next = pgd_addr_end(addr, end);
94                 if (pgd_none_or_clear_bad(pgd))
95                         continue;
96                 change_pud_range(mm, pgd, addr, next, newprot);
97         } while (pgd++, addr = next, addr != end);
98         flush_tlb_range(vma, start, end);
99         spin_unlock(&mm->page_table_lock);
100 }
101
102 static int
103 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
104         unsigned long start, unsigned long end, unsigned long newflags)
105 {
106         struct mm_struct *mm = vma->vm_mm;
107         unsigned long oldflags = vma->vm_flags;
108         long nrpages = (end - start) >> PAGE_SHIFT;
109         unsigned long charged = 0;
110         pgprot_t newprot;
111         pgoff_t pgoff;
112         int error;
113
114         if (newflags == oldflags) {
115                 *pprev = vma;
116                 return 0;
117         }
118
119         /*
120          * If we make a private mapping writable we increase our commit;
121          * but (without finer accounting) cannot reduce our commit if we
122          * make it unwritable again.
123          *
124          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
125          * a MAP_NORESERVE private mapping to writable will now reserve.
126          */
127         if (newflags & VM_WRITE) {
128                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
129                         charged = nrpages;
130                         if (security_vm_enough_memory(charged))
131                                 return -ENOMEM;
132                         newflags |= VM_ACCOUNT;
133                 }
134         }
135
136         newprot = protection_map[newflags & 0xf];
137
138         /*
139          * First try to merge with previous and/or next vma.
140          */
141         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
142         *pprev = vma_merge(mm, *pprev, start, end, newflags,
143                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
144         if (*pprev) {
145                 vma = *pprev;
146                 goto success;
147         }
148
149         *pprev = vma;
150
151         if (start != vma->vm_start) {
152                 error = split_vma(mm, vma, start, 1);
153                 if (error)
154                         goto fail;
155         }
156
157         if (end != vma->vm_end) {
158                 error = split_vma(mm, vma, end, 0);
159                 if (error)
160                         goto fail;
161         }
162
163 success:
164         /*
165          * vm_flags and vm_page_prot are protected by the mmap_sem
166          * held in write mode.
167          */
168         vma->vm_flags = newflags;
169         vma->vm_page_prot = newprot;
170         change_protection(vma, start, end, newprot);
171         __vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
172         __vm_stat_account(mm, newflags, vma->vm_file, nrpages);
173         return 0;
174
175 fail:
176         vm_unacct_memory(charged);
177         return error;
178 }
179
180 asmlinkage long
181 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
182 {
183         unsigned long vm_flags, nstart, end, tmp, reqprot;
184         struct vm_area_struct *vma, *prev;
185         int error = -EINVAL;
186         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
187         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
188         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
189                 return -EINVAL;
190
191         if (start & ~PAGE_MASK)
192                 return -EINVAL;
193         if (!len)
194                 return 0;
195         len = PAGE_ALIGN(len);
196         end = start + len;
197         if (end <= start)
198                 return -ENOMEM;
199         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
200                 return -EINVAL;
201
202         reqprot = prot;
203         /*
204          * Does the application expect PROT_READ to imply PROT_EXEC:
205          */
206         if (unlikely((prot & PROT_READ) &&
207                         (current->personality & READ_IMPLIES_EXEC)))
208                 prot |= PROT_EXEC;
209
210         vm_flags = calc_vm_prot_bits(prot);
211
212         down_write(&current->mm->mmap_sem);
213
214         vma = find_vma_prev(current->mm, start, &prev);
215         error = -ENOMEM;
216         if (!vma)
217                 goto out;
218         if (unlikely(grows & PROT_GROWSDOWN)) {
219                 if (vma->vm_start >= end)
220                         goto out;
221                 start = vma->vm_start;
222                 error = -EINVAL;
223                 if (!(vma->vm_flags & VM_GROWSDOWN))
224                         goto out;
225         }
226         else {
227                 if (vma->vm_start > start)
228                         goto out;
229                 if (unlikely(grows & PROT_GROWSUP)) {
230                         end = vma->vm_end;
231                         error = -EINVAL;
232                         if (!(vma->vm_flags & VM_GROWSUP))
233                                 goto out;
234                 }
235         }
236         if (start > vma->vm_start)
237                 prev = vma;
238
239         for (nstart = start ; ; ) {
240                 unsigned long newflags;
241
242                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
243
244                 if (is_vm_hugetlb_page(vma)) {
245                         error = -EACCES;
246                         goto out;
247                 }
248
249                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
250
251                 if ((newflags & ~(newflags >> 4)) & 0xf) {
252                         error = -EACCES;
253                         goto out;
254                 }
255
256                 error = security_file_mprotect(vma, reqprot, prot);
257                 if (error)
258                         goto out;
259
260                 tmp = vma->vm_end;
261                 if (tmp > end)
262                         tmp = end;
263                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
264                 if (error)
265                         goto out;
266                 nstart = tmp;
267
268                 if (nstart < prev->vm_end)
269                         nstart = prev->vm_end;
270                 if (nstart >= end)
271                         goto out;
272
273                 vma = prev->vm_next;
274                 if (!vma || vma->vm_start != nstart) {
275                         error = -ENOMEM;
276                         goto out;
277                 }
278         }
279 out:
280         up_write(&current->mm->mmap_sem);
281         return error;
282 }