mm: record MAP_NORESERVE status on vmas and fix small page mprotect reservations
[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 #include <linux/swap.h>
23 #include <linux/swapops.h>
24 #include <asm/uaccess.h>
25 #include <asm/pgtable.h>
26 #include <asm/cacheflush.h>
27 #include <asm/tlbflush.h>
28
29 #ifndef pgprot_modify
30 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
31 {
32         return newprot;
33 }
34 #endif
35
36 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
37                 unsigned long addr, unsigned long end, pgprot_t newprot,
38                 int dirty_accountable)
39 {
40         pte_t *pte, oldpte;
41         spinlock_t *ptl;
42
43         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
44         arch_enter_lazy_mmu_mode();
45         do {
46                 oldpte = *pte;
47                 if (pte_present(oldpte)) {
48                         pte_t ptent;
49
50                         ptent = ptep_modify_prot_start(mm, addr, pte);
51                         ptent = pte_modify(ptent, newprot);
52
53                         /*
54                          * Avoid taking write faults for pages we know to be
55                          * dirty.
56                          */
57                         if (dirty_accountable && pte_dirty(ptent))
58                                 ptent = pte_mkwrite(ptent);
59
60                         ptep_modify_prot_commit(mm, addr, pte, ptent);
61 #ifdef CONFIG_MIGRATION
62                 } else if (!pte_file(oldpte)) {
63                         swp_entry_t entry = pte_to_swp_entry(oldpte);
64
65                         if (is_write_migration_entry(entry)) {
66                                 /*
67                                  * A protection check is difficult so
68                                  * just be safe and disable write
69                                  */
70                                 make_migration_entry_read(&entry);
71                                 set_pte_at(mm, addr, pte,
72                                         swp_entry_to_pte(entry));
73                         }
74 #endif
75                 }
76
77         } while (pte++, addr += PAGE_SIZE, addr != end);
78         arch_leave_lazy_mmu_mode();
79         pte_unmap_unlock(pte - 1, ptl);
80 }
81
82 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
83                 unsigned long addr, unsigned long end, pgprot_t newprot,
84                 int dirty_accountable)
85 {
86         pmd_t *pmd;
87         unsigned long next;
88
89         pmd = pmd_offset(pud, addr);
90         do {
91                 next = pmd_addr_end(addr, end);
92                 if (pmd_none_or_clear_bad(pmd))
93                         continue;
94                 change_pte_range(mm, pmd, addr, next, newprot, dirty_accountable);
95         } while (pmd++, addr = next, addr != end);
96 }
97
98 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
99                 unsigned long addr, unsigned long end, pgprot_t newprot,
100                 int dirty_accountable)
101 {
102         pud_t *pud;
103         unsigned long next;
104
105         pud = pud_offset(pgd, addr);
106         do {
107                 next = pud_addr_end(addr, end);
108                 if (pud_none_or_clear_bad(pud))
109                         continue;
110                 change_pmd_range(mm, pud, addr, next, newprot, dirty_accountable);
111         } while (pud++, addr = next, addr != end);
112 }
113
114 static void change_protection(struct vm_area_struct *vma,
115                 unsigned long addr, unsigned long end, pgprot_t newprot,
116                 int dirty_accountable)
117 {
118         struct mm_struct *mm = vma->vm_mm;
119         pgd_t *pgd;
120         unsigned long next;
121         unsigned long start = addr;
122
123         BUG_ON(addr >= end);
124         pgd = pgd_offset(mm, addr);
125         flush_cache_range(vma, addr, end);
126         do {
127                 next = pgd_addr_end(addr, end);
128                 if (pgd_none_or_clear_bad(pgd))
129                         continue;
130                 change_pud_range(mm, pgd, addr, next, newprot, dirty_accountable);
131         } while (pgd++, addr = next, addr != end);
132         flush_tlb_range(vma, start, end);
133 }
134
135 int
136 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
137         unsigned long start, unsigned long end, unsigned long newflags)
138 {
139         struct mm_struct *mm = vma->vm_mm;
140         unsigned long oldflags = vma->vm_flags;
141         long nrpages = (end - start) >> PAGE_SHIFT;
142         unsigned long charged = 0;
143         pgoff_t pgoff;
144         int error;
145         int dirty_accountable = 0;
146
147         if (newflags == oldflags) {
148                 *pprev = vma;
149                 return 0;
150         }
151
152         /*
153          * If we make a private mapping writable we increase our commit;
154          * but (without finer accounting) cannot reduce our commit if we
155          * make it unwritable again.
156          */
157         if (newflags & VM_WRITE) {
158                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|
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         if (is_vm_hugetlb_page(vma))
207                 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
208         else
209                 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
210         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
211         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
212         return 0;
213
214 fail:
215         vm_unacct_memory(charged);
216         return error;
217 }
218
219 asmlinkage long
220 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
221 {
222         unsigned long vm_flags, nstart, end, tmp, reqprot;
223         struct vm_area_struct *vma, *prev;
224         int error = -EINVAL;
225         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
226         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
227         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
228                 return -EINVAL;
229
230         if (start & ~PAGE_MASK)
231                 return -EINVAL;
232         if (!len)
233                 return 0;
234         len = PAGE_ALIGN(len);
235         end = start + len;
236         if (end <= start)
237                 return -ENOMEM;
238         if (!arch_validate_prot(prot))
239                 return -EINVAL;
240
241         reqprot = prot;
242         /*
243          * Does the application expect PROT_READ to imply PROT_EXEC:
244          */
245         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
246                 prot |= PROT_EXEC;
247
248         vm_flags = calc_vm_prot_bits(prot);
249
250         down_write(&current->mm->mmap_sem);
251
252         vma = find_vma_prev(current->mm, start, &prev);
253         error = -ENOMEM;
254         if (!vma)
255                 goto out;
256         if (unlikely(grows & PROT_GROWSDOWN)) {
257                 if (vma->vm_start >= end)
258                         goto out;
259                 start = vma->vm_start;
260                 error = -EINVAL;
261                 if (!(vma->vm_flags & VM_GROWSDOWN))
262                         goto out;
263         }
264         else {
265                 if (vma->vm_start > start)
266                         goto out;
267                 if (unlikely(grows & PROT_GROWSUP)) {
268                         end = vma->vm_end;
269                         error = -EINVAL;
270                         if (!(vma->vm_flags & VM_GROWSUP))
271                                 goto out;
272                 }
273         }
274         if (start > vma->vm_start)
275                 prev = vma;
276
277         for (nstart = start ; ; ) {
278                 unsigned long newflags;
279
280                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
281
282                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
283
284                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
285                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
286                         error = -EACCES;
287                         goto out;
288                 }
289
290                 error = security_file_mprotect(vma, reqprot, prot);
291                 if (error)
292                         goto out;
293
294                 tmp = vma->vm_end;
295                 if (tmp > end)
296                         tmp = end;
297                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
298                 if (error)
299                         goto out;
300                 nstart = tmp;
301
302                 if (nstart < prev->vm_end)
303                         nstart = prev->vm_end;
304                 if (nstart >= end)
305                         goto out;
306
307                 vma = prev->vm_next;
308                 if (!vma || vma->vm_start != nstart) {
309                         error = -ENOMEM;
310                         goto out;
311                 }
312         }
313 out:
314         up_write(&current->mm->mmap_sem);
315         return error;
316 }