Merge branch 'topic/drop-l3' into for-linus
[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@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/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 #include <linux/swap.h>
23 #include <linux/swapops.h>
24 #include <linux/mmu_notifier.h>
25 #include <linux/migrate.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 mm_struct *mm, 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_none_or_clear_bad(pmd))
92                         continue;
93                 change_pte_range(mm, pmd, addr, next, newprot, dirty_accountable);
94         } while (pmd++, addr = next, addr != end);
95 }
96
97 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
98                 unsigned long addr, unsigned long end, pgprot_t newprot,
99                 int dirty_accountable)
100 {
101         pud_t *pud;
102         unsigned long next;
103
104         pud = pud_offset(pgd, addr);
105         do {
106                 next = pud_addr_end(addr, end);
107                 if (pud_none_or_clear_bad(pud))
108                         continue;
109                 change_pmd_range(mm, pud, addr, next, newprot, dirty_accountable);
110         } while (pud++, addr = next, addr != end);
111 }
112
113 static void change_protection(struct vm_area_struct *vma,
114                 unsigned long addr, unsigned long end, pgprot_t newprot,
115                 int dirty_accountable)
116 {
117         struct mm_struct *mm = vma->vm_mm;
118         pgd_t *pgd;
119         unsigned long next;
120         unsigned long start = addr;
121
122         BUG_ON(addr >= end);
123         pgd = pgd_offset(mm, addr);
124         flush_cache_range(vma, addr, end);
125         do {
126                 next = pgd_addr_end(addr, end);
127                 if (pgd_none_or_clear_bad(pgd))
128                         continue;
129                 change_pud_range(mm, pgd, addr, next, newprot, dirty_accountable);
130         } while (pgd++, addr = next, addr != end);
131         flush_tlb_range(vma, start, end);
132 }
133
134 int
135 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
136         unsigned long start, unsigned long end, unsigned long newflags)
137 {
138         struct mm_struct *mm = vma->vm_mm;
139         unsigned long oldflags = vma->vm_flags;
140         long nrpages = (end - start) >> PAGE_SHIFT;
141         unsigned long charged = 0;
142         pgoff_t pgoff;
143         int error;
144         int dirty_accountable = 0;
145
146         if (newflags == oldflags) {
147                 *pprev = vma;
148                 return 0;
149         }
150
151         /*
152          * If we make a private mapping writable we increase our commit;
153          * but (without finer accounting) cannot reduce our commit if we
154          * make it unwritable again. hugetlb mapping were accounted for
155          * even if read-only so there is no need to account for them here
156          */
157         if (newflags & VM_WRITE) {
158                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
159                                                 VM_SHARED|VM_NORESERVE))) {
160                         charged = nrpages;
161                         if (security_vm_enough_memory(charged))
162                                 return -ENOMEM;
163                         newflags |= VM_ACCOUNT;
164                 }
165         }
166
167         /*
168          * First try to merge with previous and/or next vma.
169          */
170         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
171         *pprev = vma_merge(mm, *pprev, start, end, newflags,
172                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
173         if (*pprev) {
174                 vma = *pprev;
175                 goto success;
176         }
177
178         *pprev = vma;
179
180         if (start != vma->vm_start) {
181                 error = split_vma(mm, vma, start, 1);
182                 if (error)
183                         goto fail;
184         }
185
186         if (end != vma->vm_end) {
187                 error = split_vma(mm, vma, end, 0);
188                 if (error)
189                         goto fail;
190         }
191
192 success:
193         /*
194          * vm_flags and vm_page_prot are protected by the mmap_sem
195          * held in write mode.
196          */
197         vma->vm_flags = newflags;
198         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
199                                           vm_get_page_prot(newflags));
200
201         if (vma_wants_writenotify(vma)) {
202                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
203                 dirty_accountable = 1;
204         }
205
206         mmu_notifier_invalidate_range_start(mm, start, end);
207         if (is_vm_hugetlb_page(vma))
208                 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
209         else
210                 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
211         mmu_notifier_invalidate_range_end(mm, start, end);
212         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
213         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
214         return 0;
215
216 fail:
217         vm_unacct_memory(charged);
218         return error;
219 }
220
221 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
222                 unsigned long, prot)
223 {
224         unsigned long vm_flags, nstart, end, tmp, reqprot;
225         struct vm_area_struct *vma, *prev;
226         int error = -EINVAL;
227         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
228         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
229         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
230                 return -EINVAL;
231
232         if (start & ~PAGE_MASK)
233                 return -EINVAL;
234         if (!len)
235                 return 0;
236         len = PAGE_ALIGN(len);
237         end = start + len;
238         if (end <= start)
239                 return -ENOMEM;
240         if (!arch_validate_prot(prot))
241                 return -EINVAL;
242
243         reqprot = prot;
244         /*
245          * Does the application expect PROT_READ to imply PROT_EXEC:
246          */
247         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
248                 prot |= PROT_EXEC;
249
250         vm_flags = calc_vm_prot_bits(prot);
251
252         down_write(&current->mm->mmap_sem);
253
254         vma = find_vma_prev(current->mm, start, &prev);
255         error = -ENOMEM;
256         if (!vma)
257                 goto out;
258         if (unlikely(grows & PROT_GROWSDOWN)) {
259                 if (vma->vm_start >= end)
260                         goto out;
261                 start = vma->vm_start;
262                 error = -EINVAL;
263                 if (!(vma->vm_flags & VM_GROWSDOWN))
264                         goto out;
265         }
266         else {
267                 if (vma->vm_start > start)
268                         goto out;
269                 if (unlikely(grows & PROT_GROWSUP)) {
270                         end = vma->vm_end;
271                         error = -EINVAL;
272                         if (!(vma->vm_flags & VM_GROWSUP))
273                                 goto out;
274                 }
275         }
276         if (start > vma->vm_start)
277                 prev = vma;
278
279         for (nstart = start ; ; ) {
280                 unsigned long newflags;
281
282                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
283
284                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
285
286                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
287                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
288                         error = -EACCES;
289                         goto out;
290                 }
291
292                 error = security_file_mprotect(vma, reqprot, prot);
293                 if (error)
294                         goto out;
295
296                 tmp = vma->vm_end;
297                 if (tmp > end)
298                         tmp = end;
299                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
300                 if (error)
301                         goto out;
302                 nstart = tmp;
303
304                 if (nstart < prev->vm_end)
305                         nstart = prev->vm_end;
306                 if (nstart >= end)
307                         goto out;
308
309                 vma = prev->vm_next;
310                 if (!vma || vma->vm_start != nstart) {
311                         error = -ENOMEM;
312                         goto out;
313                 }
314         }
315 out:
316         up_write(&current->mm->mmap_sem);
317         return error;
318 }