Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/mprotect.c | |
3 | * | |
4 | * (C) Copyright 1994 Linus Torvalds | |
5 | * (C) Copyright 2002 Christoph Hellwig | |
6 | * | |
046c6884 | 7 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
8 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved |
9 | */ | |
10 | ||
11 | #include <linux/mm.h> | |
12 | #include <linux/hugetlb.h> | |
1da177e4 LT |
13 | #include <linux/shm.h> |
14 | #include <linux/mman.h> | |
15 | #include <linux/fs.h> | |
16 | #include <linux/highmem.h> | |
17 | #include <linux/security.h> | |
18 | #include <linux/mempolicy.h> | |
19 | #include <linux/personality.h> | |
20 | #include <linux/syscalls.h> | |
0697212a CL |
21 | #include <linux/swap.h> |
22 | #include <linux/swapops.h> | |
cddb8a5c | 23 | #include <linux/mmu_notifier.h> |
64cdd548 | 24 | #include <linux/migrate.h> |
cdd6c482 | 25 | #include <linux/perf_event.h> |
1da177e4 LT |
26 | #include <asm/uaccess.h> |
27 | #include <asm/pgtable.h> | |
28 | #include <asm/cacheflush.h> | |
29 | #include <asm/tlbflush.h> | |
30 | ||
1c12c4cf VP |
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 | ||
4b10e7d5 | 38 | static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
c1e6098b | 39 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 40 | int dirty_accountable, int prot_numa) |
1da177e4 | 41 | { |
4b10e7d5 | 42 | struct mm_struct *mm = vma->vm_mm; |
0697212a | 43 | pte_t *pte, oldpte; |
705e87c0 | 44 | spinlock_t *ptl; |
7da4d641 | 45 | unsigned long pages = 0; |
1da177e4 | 46 | |
705e87c0 | 47 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 48 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 49 | do { |
0697212a CL |
50 | oldpte = *pte; |
51 | if (pte_present(oldpte)) { | |
1da177e4 | 52 | pte_t ptent; |
4b10e7d5 | 53 | bool updated = false; |
1da177e4 | 54 | |
1ea0704e | 55 | ptent = ptep_modify_prot_start(mm, addr, pte); |
4b10e7d5 MG |
56 | if (!prot_numa) { |
57 | ptent = pte_modify(ptent, newprot); | |
58 | updated = true; | |
59 | } else { | |
60 | struct page *page; | |
61 | ||
62 | page = vm_normal_page(vma, addr, oldpte); | |
63 | if (page) { | |
64 | /* only check non-shared pages */ | |
65 | if (!pte_numa(oldpte) && | |
66 | page_mapcount(page) == 1) { | |
67 | ptent = pte_mknuma(ptent); | |
68 | updated = true; | |
69 | } | |
70 | } | |
71 | } | |
1ea0704e | 72 | |
c1e6098b PZ |
73 | /* |
74 | * Avoid taking write faults for pages we know to be | |
75 | * dirty. | |
76 | */ | |
4b10e7d5 | 77 | if (dirty_accountable && pte_dirty(ptent)) { |
c1e6098b | 78 | ptent = pte_mkwrite(ptent); |
4b10e7d5 MG |
79 | updated = true; |
80 | } | |
81 | ||
82 | if (updated) | |
83 | pages++; | |
1ea0704e JF |
84 | |
85 | ptep_modify_prot_commit(mm, addr, pte, ptent); | |
ce1744f4 | 86 | } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) { |
0697212a CL |
87 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
88 | ||
89 | if (is_write_migration_entry(entry)) { | |
90 | /* | |
91 | * A protection check is difficult so | |
92 | * just be safe and disable write | |
93 | */ | |
94 | make_migration_entry_read(&entry); | |
95 | set_pte_at(mm, addr, pte, | |
96 | swp_entry_to_pte(entry)); | |
97 | } | |
7da4d641 | 98 | pages++; |
1da177e4 LT |
99 | } |
100 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 101 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 102 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
103 | |
104 | return pages; | |
1da177e4 LT |
105 | } |
106 | ||
4b10e7d5 MG |
107 | #ifdef CONFIG_NUMA_BALANCING |
108 | static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr, | |
109 | pmd_t *pmd) | |
110 | { | |
111 | spin_lock(&mm->page_table_lock); | |
112 | set_pmd_at(mm, addr & PMD_MASK, pmd, pmd_mknuma(*pmd)); | |
113 | spin_unlock(&mm->page_table_lock); | |
114 | } | |
115 | #else | |
116 | static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr, | |
117 | pmd_t *pmd) | |
118 | { | |
119 | BUG(); | |
120 | } | |
121 | #endif /* CONFIG_NUMA_BALANCING */ | |
122 | ||
7da4d641 | 123 | static inline unsigned long change_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
c1e6098b | 124 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 125 | int dirty_accountable, int prot_numa) |
1da177e4 LT |
126 | { |
127 | pmd_t *pmd; | |
128 | unsigned long next; | |
7da4d641 | 129 | unsigned long pages = 0; |
1da177e4 LT |
130 | |
131 | pmd = pmd_offset(pud, addr); | |
132 | do { | |
133 | next = pmd_addr_end(addr, end); | |
cd7548ab JW |
134 | if (pmd_trans_huge(*pmd)) { |
135 | if (next - addr != HPAGE_PMD_SIZE) | |
136 | split_huge_page_pmd(vma->vm_mm, pmd); | |
4b10e7d5 | 137 | else if (change_huge_pmd(vma, pmd, addr, newprot, prot_numa)) { |
7da4d641 | 138 | pages += HPAGE_PMD_NR; |
cd7548ab | 139 | continue; |
7da4d641 | 140 | } |
cd7548ab JW |
141 | /* fall through */ |
142 | } | |
1da177e4 LT |
143 | if (pmd_none_or_clear_bad(pmd)) |
144 | continue; | |
4b10e7d5 MG |
145 | pages += change_pte_range(vma, pmd, addr, next, newprot, |
146 | dirty_accountable, prot_numa); | |
147 | ||
148 | if (prot_numa) | |
149 | change_pmd_protnuma(vma->vm_mm, addr, pmd); | |
1da177e4 | 150 | } while (pmd++, addr = next, addr != end); |
7da4d641 PZ |
151 | |
152 | return pages; | |
1da177e4 LT |
153 | } |
154 | ||
7da4d641 | 155 | static inline unsigned long change_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
c1e6098b | 156 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 157 | int dirty_accountable, int prot_numa) |
1da177e4 LT |
158 | { |
159 | pud_t *pud; | |
160 | unsigned long next; | |
7da4d641 | 161 | unsigned long pages = 0; |
1da177e4 LT |
162 | |
163 | pud = pud_offset(pgd, addr); | |
164 | do { | |
165 | next = pud_addr_end(addr, end); | |
166 | if (pud_none_or_clear_bad(pud)) | |
167 | continue; | |
7da4d641 | 168 | pages += change_pmd_range(vma, pud, addr, next, newprot, |
4b10e7d5 | 169 | dirty_accountable, prot_numa); |
1da177e4 | 170 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
171 | |
172 | return pages; | |
1da177e4 LT |
173 | } |
174 | ||
7da4d641 | 175 | static unsigned long change_protection_range(struct vm_area_struct *vma, |
c1e6098b | 176 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 177 | int dirty_accountable, int prot_numa) |
1da177e4 LT |
178 | { |
179 | struct mm_struct *mm = vma->vm_mm; | |
180 | pgd_t *pgd; | |
181 | unsigned long next; | |
182 | unsigned long start = addr; | |
7da4d641 | 183 | unsigned long pages = 0; |
1da177e4 LT |
184 | |
185 | BUG_ON(addr >= end); | |
186 | pgd = pgd_offset(mm, addr); | |
187 | flush_cache_range(vma, addr, end); | |
1da177e4 LT |
188 | do { |
189 | next = pgd_addr_end(addr, end); | |
190 | if (pgd_none_or_clear_bad(pgd)) | |
191 | continue; | |
7da4d641 | 192 | pages += change_pud_range(vma, pgd, addr, next, newprot, |
4b10e7d5 | 193 | dirty_accountable, prot_numa); |
1da177e4 | 194 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 195 | |
1233d588 IM |
196 | /* Only flush the TLB if we actually modified any entries: */ |
197 | if (pages) | |
198 | flush_tlb_range(vma, start, end); | |
7da4d641 PZ |
199 | |
200 | return pages; | |
201 | } | |
202 | ||
203 | unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, | |
204 | unsigned long end, pgprot_t newprot, | |
4b10e7d5 | 205 | int dirty_accountable, int prot_numa) |
7da4d641 PZ |
206 | { |
207 | struct mm_struct *mm = vma->vm_mm; | |
208 | unsigned long pages; | |
209 | ||
210 | mmu_notifier_invalidate_range_start(mm, start, end); | |
211 | if (is_vm_hugetlb_page(vma)) | |
212 | pages = hugetlb_change_protection(vma, start, end, newprot); | |
213 | else | |
4b10e7d5 | 214 | pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa); |
7da4d641 PZ |
215 | mmu_notifier_invalidate_range_end(mm, start, end); |
216 | ||
217 | return pages; | |
1da177e4 LT |
218 | } |
219 | ||
b6a2fea3 | 220 | int |
1da177e4 LT |
221 | mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, |
222 | unsigned long start, unsigned long end, unsigned long newflags) | |
223 | { | |
224 | struct mm_struct *mm = vma->vm_mm; | |
225 | unsigned long oldflags = vma->vm_flags; | |
226 | long nrpages = (end - start) >> PAGE_SHIFT; | |
227 | unsigned long charged = 0; | |
1da177e4 LT |
228 | pgoff_t pgoff; |
229 | int error; | |
c1e6098b | 230 | int dirty_accountable = 0; |
1da177e4 LT |
231 | |
232 | if (newflags == oldflags) { | |
233 | *pprev = vma; | |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* | |
238 | * If we make a private mapping writable we increase our commit; | |
239 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
240 | * make it unwritable again. hugetlb mapping were accounted for |
241 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
242 | */ |
243 | if (newflags & VM_WRITE) { | |
5a6fe125 | 244 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 245 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 246 | charged = nrpages; |
191c5424 | 247 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
248 | return -ENOMEM; |
249 | newflags |= VM_ACCOUNT; | |
250 | } | |
251 | } | |
252 | ||
1da177e4 LT |
253 | /* |
254 | * First try to merge with previous and/or next vma. | |
255 | */ | |
256 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
257 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
258 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
259 | if (*pprev) { | |
260 | vma = *pprev; | |
261 | goto success; | |
262 | } | |
263 | ||
264 | *pprev = vma; | |
265 | ||
266 | if (start != vma->vm_start) { | |
267 | error = split_vma(mm, vma, start, 1); | |
268 | if (error) | |
269 | goto fail; | |
270 | } | |
271 | ||
272 | if (end != vma->vm_end) { | |
273 | error = split_vma(mm, vma, end, 0); | |
274 | if (error) | |
275 | goto fail; | |
276 | } | |
277 | ||
278 | success: | |
279 | /* | |
280 | * vm_flags and vm_page_prot are protected by the mmap_sem | |
281 | * held in write mode. | |
282 | */ | |
283 | vma->vm_flags = newflags; | |
1c12c4cf VP |
284 | vma->vm_page_prot = pgprot_modify(vma->vm_page_prot, |
285 | vm_get_page_prot(newflags)); | |
286 | ||
c1e6098b | 287 | if (vma_wants_writenotify(vma)) { |
1ddd439e | 288 | vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED); |
c1e6098b PZ |
289 | dirty_accountable = 1; |
290 | } | |
d08b3851 | 291 | |
4b10e7d5 | 292 | change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable, 0); |
7da4d641 | 293 | |
ab50b8ed HD |
294 | vm_stat_account(mm, oldflags, vma->vm_file, -nrpages); |
295 | vm_stat_account(mm, newflags, vma->vm_file, nrpages); | |
63bfd738 | 296 | perf_event_mmap(vma); |
1da177e4 LT |
297 | return 0; |
298 | ||
299 | fail: | |
300 | vm_unacct_memory(charged); | |
301 | return error; | |
302 | } | |
303 | ||
6a6160a7 HC |
304 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, |
305 | unsigned long, prot) | |
1da177e4 LT |
306 | { |
307 | unsigned long vm_flags, nstart, end, tmp, reqprot; | |
308 | struct vm_area_struct *vma, *prev; | |
309 | int error = -EINVAL; | |
310 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); | |
311 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); | |
312 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
313 | return -EINVAL; | |
314 | ||
315 | if (start & ~PAGE_MASK) | |
316 | return -EINVAL; | |
317 | if (!len) | |
318 | return 0; | |
319 | len = PAGE_ALIGN(len); | |
320 | end = start + len; | |
321 | if (end <= start) | |
322 | return -ENOMEM; | |
b845f313 | 323 | if (!arch_validate_prot(prot)) |
1da177e4 LT |
324 | return -EINVAL; |
325 | ||
326 | reqprot = prot; | |
327 | /* | |
328 | * Does the application expect PROT_READ to imply PROT_EXEC: | |
329 | */ | |
b344e05c | 330 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) |
1da177e4 LT |
331 | prot |= PROT_EXEC; |
332 | ||
333 | vm_flags = calc_vm_prot_bits(prot); | |
334 | ||
335 | down_write(¤t->mm->mmap_sem); | |
336 | ||
097d5910 | 337 | vma = find_vma(current->mm, start); |
1da177e4 LT |
338 | error = -ENOMEM; |
339 | if (!vma) | |
340 | goto out; | |
097d5910 | 341 | prev = vma->vm_prev; |
1da177e4 LT |
342 | if (unlikely(grows & PROT_GROWSDOWN)) { |
343 | if (vma->vm_start >= end) | |
344 | goto out; | |
345 | start = vma->vm_start; | |
346 | error = -EINVAL; | |
347 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
348 | goto out; | |
349 | } | |
350 | else { | |
351 | if (vma->vm_start > start) | |
352 | goto out; | |
353 | if (unlikely(grows & PROT_GROWSUP)) { | |
354 | end = vma->vm_end; | |
355 | error = -EINVAL; | |
356 | if (!(vma->vm_flags & VM_GROWSUP)) | |
357 | goto out; | |
358 | } | |
359 | } | |
360 | if (start > vma->vm_start) | |
361 | prev = vma; | |
362 | ||
363 | for (nstart = start ; ; ) { | |
364 | unsigned long newflags; | |
365 | ||
366 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ | |
367 | ||
1da177e4 LT |
368 | newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC)); |
369 | ||
7e2cff42 PBG |
370 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
371 | if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) { | |
1da177e4 LT |
372 | error = -EACCES; |
373 | goto out; | |
374 | } | |
375 | ||
376 | error = security_file_mprotect(vma, reqprot, prot); | |
377 | if (error) | |
378 | goto out; | |
379 | ||
380 | tmp = vma->vm_end; | |
381 | if (tmp > end) | |
382 | tmp = end; | |
383 | error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); | |
384 | if (error) | |
385 | goto out; | |
386 | nstart = tmp; | |
387 | ||
388 | if (nstart < prev->vm_end) | |
389 | nstart = prev->vm_end; | |
390 | if (nstart >= end) | |
391 | goto out; | |
392 | ||
393 | vma = prev->vm_next; | |
394 | if (!vma || vma->vm_start != nstart) { | |
395 | error = -ENOMEM; | |
396 | goto out; | |
397 | } | |
398 | } | |
399 | out: | |
400 | up_write(¤t->mm->mmap_sem); | |
401 | return error; | |
402 | } |