blob: acfe7c8d72fcff59dec990869a5fce9d63b56e55 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Christoph Lameter06972122006-06-23 02:03:35 -070022#include <linux/swap.h>
23#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/uaccess.h>
25#include <asm/pgtable.h>
26#include <asm/cacheflush.h>
27#include <asm/tlbflush.h>
28
Venki Pallipadi1c12c4c2008-05-14 16:05:51 -070029#ifndef pgprot_modify
30static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
31{
32 return newprot;
33}
34#endif
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070037 unsigned long addr, unsigned long end, pgprot_t newprot,
38 int dirty_accountable)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039{
Christoph Lameter06972122006-06-23 02:03:35 -070040 pte_t *pte, oldpte;
Hugh Dickins705e87c2005-10-29 18:16:27 -070041 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Hugh Dickins705e87c2005-10-29 18:16:27 -070043 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -070044 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 do {
Christoph Lameter06972122006-06-23 02:03:35 -070046 oldpte = *pte;
47 if (pte_present(oldpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 pte_t ptent;
49
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -070050 ptent = ptep_modify_prot_start(mm, addr, pte);
Peter Zijlstrac1e60982006-09-25 23:30:59 -070051 ptent = pte_modify(ptent, newprot);
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -070052
Peter Zijlstrac1e60982006-09-25 23:30:59 -070053 /*
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);
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -070059
60 ptep_modify_prot_commit(mm, addr, pte, ptent);
Christoph Lameter06972122006-06-23 02:03:35 -070061#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
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 }
Christoph Lameter06972122006-06-23 02:03:35 -070076
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -070078 arch_leave_lazy_mmu_mode();
Hugh Dickins705e87c2005-10-29 18:16:27 -070079 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080}
81
82static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070083 unsigned long addr, unsigned long end, pgprot_t newprot,
84 int dirty_accountable)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
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;
Peter Zijlstrac1e60982006-09-25 23:30:59 -070094 change_pte_range(mm, pmd, addr, next, newprot, dirty_accountable);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 } while (pmd++, addr = next, addr != end);
96}
97
98static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070099 unsigned long addr, unsigned long end, pgprot_t newprot,
100 int dirty_accountable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
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;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700110 change_pmd_range(mm, pud, addr, next, newprot, dirty_accountable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 } while (pud++, addr = next, addr != end);
112}
113
114static void change_protection(struct vm_area_struct *vma,
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700115 unsigned long addr, unsigned long end, pgprot_t newprot,
116 int dirty_accountable)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 do {
127 next = pgd_addr_end(addr, end);
128 if (pgd_none_or_clear_bad(pgd))
129 continue;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700130 change_pud_range(mm, pgd, addr, next, newprot, dirty_accountable);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 } while (pgd++, addr = next, addr != end);
132 flush_tlb_range(vma, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700135int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136mprotect_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 pgoff_t pgoff;
144 int error;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700145 int dirty_accountable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
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 * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
158 * a MAP_NORESERVE private mapping to writable will now reserve.
159 */
160 if (newflags & VM_WRITE) {
Zhang, Yanmin8f860592006-03-22 00:08:50 -0800161 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 charged = nrpages;
163 if (security_vm_enough_memory(charged))
164 return -ENOMEM;
165 newflags |= VM_ACCOUNT;
166 }
167 }
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 /*
170 * First try to merge with previous and/or next vma.
171 */
172 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
173 *pprev = vma_merge(mm, *pprev, start, end, newflags,
174 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
175 if (*pprev) {
176 vma = *pprev;
177 goto success;
178 }
179
180 *pprev = vma;
181
182 if (start != vma->vm_start) {
183 error = split_vma(mm, vma, start, 1);
184 if (error)
185 goto fail;
186 }
187
188 if (end != vma->vm_end) {
189 error = split_vma(mm, vma, end, 0);
190 if (error)
191 goto fail;
192 }
193
194success:
195 /*
196 * vm_flags and vm_page_prot are protected by the mmap_sem
197 * held in write mode.
198 */
199 vma->vm_flags = newflags;
Venki Pallipadi1c12c4c2008-05-14 16:05:51 -0700200 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
201 vm_get_page_prot(newflags));
202
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700203 if (vma_wants_writenotify(vma)) {
Hugh Dickins1ddd4392007-10-22 20:45:12 -0700204 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700205 dirty_accountable = 1;
206 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700207
Zhang, Yanmin8f860592006-03-22 00:08:50 -0800208 if (is_vm_hugetlb_page(vma))
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700209 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
Zhang, Yanmin8f860592006-03-22 00:08:50 -0800210 else
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700211 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700212 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
213 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 return 0;
215
216fail:
217 vm_unacct_memory(charged);
218 return error;
219}
220
221asmlinkage long
222sys_mprotect(unsigned long start, size_t len, 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 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
241 return -EINVAL;
242
243 reqprot = prot;
244 /*
245 * Does the application expect PROT_READ to imply PROT_EXEC:
246 */
Hua Zhongb344e052006-06-23 02:03:23 -0700247 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
285
Paolo 'Blaisorblade' Giarrusso7e2cff42005-09-21 09:55:39 -0700286 /* newflags >> 4 shift VM_MAY% in place of VM_% */
287 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 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 }
315out:
316 up_write(&current->mm->mmap_sem);
317 return error;
318}