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, |
9532fec1 | 40 | int dirty_accountable, int prot_numa, bool *ret_all_same_node) |
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; |
9532fec1 MG |
46 | bool all_same_node = true; |
47 | int last_nid = -1; | |
1da177e4 | 48 | |
705e87c0 | 49 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
6606c3e0 | 50 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 51 | do { |
0697212a CL |
52 | oldpte = *pte; |
53 | if (pte_present(oldpte)) { | |
1da177e4 | 54 | pte_t ptent; |
4b10e7d5 | 55 | bool updated = false; |
1da177e4 | 56 | |
1ea0704e | 57 | ptent = ptep_modify_prot_start(mm, addr, pte); |
4b10e7d5 MG |
58 | if (!prot_numa) { |
59 | ptent = pte_modify(ptent, newprot); | |
60 | updated = true; | |
61 | } else { | |
62 | struct page *page; | |
63 | ||
64 | page = vm_normal_page(vma, addr, oldpte); | |
65 | if (page) { | |
9532fec1 MG |
66 | int this_nid = page_to_nid(page); |
67 | if (last_nid == -1) | |
68 | last_nid = this_nid; | |
69 | if (last_nid != this_nid) | |
70 | all_same_node = false; | |
71 | ||
4b10e7d5 MG |
72 | /* only check non-shared pages */ |
73 | if (!pte_numa(oldpte) && | |
74 | page_mapcount(page) == 1) { | |
75 | ptent = pte_mknuma(ptent); | |
76 | updated = true; | |
77 | } | |
78 | } | |
79 | } | |
1ea0704e | 80 | |
c1e6098b PZ |
81 | /* |
82 | * Avoid taking write faults for pages we know to be | |
83 | * dirty. | |
84 | */ | |
4b10e7d5 | 85 | if (dirty_accountable && pte_dirty(ptent)) { |
c1e6098b | 86 | ptent = pte_mkwrite(ptent); |
4b10e7d5 MG |
87 | updated = true; |
88 | } | |
1ea0704e | 89 | |
4b10e7d5 MG |
90 | if (updated) |
91 | pages++; | |
1ea0704e | 92 | ptep_modify_prot_commit(mm, addr, pte, ptent); |
ce1744f4 | 93 | } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) { |
0697212a CL |
94 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
95 | ||
96 | if (is_write_migration_entry(entry)) { | |
97 | /* | |
98 | * A protection check is difficult so | |
99 | * just be safe and disable write | |
100 | */ | |
101 | make_migration_entry_read(&entry); | |
102 | set_pte_at(mm, addr, pte, | |
103 | swp_entry_to_pte(entry)); | |
104 | } | |
7da4d641 | 105 | pages++; |
1da177e4 LT |
106 | } |
107 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 108 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 109 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 | 110 | |
9532fec1 | 111 | *ret_all_same_node = all_same_node; |
7da4d641 | 112 | return pages; |
1da177e4 LT |
113 | } |
114 | ||
4b10e7d5 MG |
115 | #ifdef CONFIG_NUMA_BALANCING |
116 | static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr, | |
7d12efae | 117 | pmd_t *pmd) |
4b10e7d5 MG |
118 | { |
119 | spin_lock(&mm->page_table_lock); | |
120 | set_pmd_at(mm, addr & PMD_MASK, pmd, pmd_mknuma(*pmd)); | |
121 | spin_unlock(&mm->page_table_lock); | |
122 | } | |
123 | #else | |
124 | static inline void change_pmd_protnuma(struct mm_struct *mm, unsigned long addr, | |
7d12efae | 125 | pmd_t *pmd) |
4b10e7d5 MG |
126 | { |
127 | BUG(); | |
1da177e4 | 128 | } |
4b10e7d5 | 129 | #endif /* CONFIG_NUMA_BALANCING */ |
1da177e4 | 130 | |
7d12efae AM |
131 | static inline unsigned long change_pmd_range(struct vm_area_struct *vma, |
132 | pud_t *pud, unsigned long addr, unsigned long end, | |
133 | pgprot_t newprot, int dirty_accountable, int prot_numa) | |
1da177e4 LT |
134 | { |
135 | pmd_t *pmd; | |
136 | unsigned long next; | |
7da4d641 | 137 | unsigned long pages = 0; |
f99510dc | 138 | unsigned long nr_huge_updates = 0; |
9532fec1 | 139 | bool all_same_node; |
1da177e4 LT |
140 | |
141 | pmd = pmd_offset(pud, addr); | |
142 | do { | |
143 | next = pmd_addr_end(addr, end); | |
cd7548ab JW |
144 | if (pmd_trans_huge(*pmd)) { |
145 | if (next - addr != HPAGE_PMD_SIZE) | |
e180377f | 146 | split_huge_page_pmd(vma, addr, pmd); |
7d12efae AM |
147 | else if (change_huge_pmd(vma, pmd, addr, newprot, |
148 | prot_numa)) { | |
f99510dc MG |
149 | pages += HPAGE_PMD_NR; |
150 | nr_huge_updates++; | |
cd7548ab | 151 | continue; |
7da4d641 | 152 | } |
cd7548ab JW |
153 | /* fall through */ |
154 | } | |
1da177e4 LT |
155 | if (pmd_none_or_clear_bad(pmd)) |
156 | continue; | |
4b10e7d5 | 157 | pages += change_pte_range(vma, pmd, addr, next, newprot, |
9532fec1 | 158 | dirty_accountable, prot_numa, &all_same_node); |
4b10e7d5 | 159 | |
9532fec1 MG |
160 | /* |
161 | * If we are changing protections for NUMA hinting faults then | |
162 | * set pmd_numa if the examined pages were all on the same | |
163 | * node. This allows a regular PMD to be handled as one fault | |
164 | * and effectively batches the taking of the PTL | |
165 | */ | |
166 | if (prot_numa && all_same_node) | |
4b10e7d5 | 167 | change_pmd_protnuma(vma->vm_mm, addr, pmd); |
1da177e4 | 168 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 169 | |
f99510dc MG |
170 | if (nr_huge_updates) |
171 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
172 | ||
7da4d641 | 173 | return pages; |
1da177e4 LT |
174 | } |
175 | ||
7d12efae AM |
176 | static inline unsigned long change_pud_range(struct vm_area_struct *vma, |
177 | pgd_t *pgd, unsigned long addr, unsigned long end, | |
178 | pgprot_t newprot, int dirty_accountable, int prot_numa) | |
1da177e4 LT |
179 | { |
180 | pud_t *pud; | |
181 | unsigned long next; | |
7da4d641 | 182 | unsigned long pages = 0; |
1da177e4 LT |
183 | |
184 | pud = pud_offset(pgd, addr); | |
185 | do { | |
186 | next = pud_addr_end(addr, end); | |
187 | if (pud_none_or_clear_bad(pud)) | |
188 | continue; | |
7da4d641 | 189 | pages += change_pmd_range(vma, pud, addr, next, newprot, |
4b10e7d5 | 190 | dirty_accountable, prot_numa); |
1da177e4 | 191 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
192 | |
193 | return pages; | |
1da177e4 LT |
194 | } |
195 | ||
7da4d641 | 196 | static unsigned long change_protection_range(struct vm_area_struct *vma, |
c1e6098b | 197 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 198 | int dirty_accountable, int prot_numa) |
1da177e4 LT |
199 | { |
200 | struct mm_struct *mm = vma->vm_mm; | |
201 | pgd_t *pgd; | |
202 | unsigned long next; | |
203 | unsigned long start = addr; | |
7da4d641 | 204 | unsigned long pages = 0; |
1da177e4 LT |
205 | |
206 | BUG_ON(addr >= end); | |
207 | pgd = pgd_offset(mm, addr); | |
208 | flush_cache_range(vma, addr, end); | |
d303cf46 | 209 | set_tlb_flush_pending(mm); |
1da177e4 LT |
210 | do { |
211 | next = pgd_addr_end(addr, end); | |
212 | if (pgd_none_or_clear_bad(pgd)) | |
213 | continue; | |
7da4d641 | 214 | pages += change_pud_range(vma, pgd, addr, next, newprot, |
4b10e7d5 | 215 | dirty_accountable, prot_numa); |
1da177e4 | 216 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 217 | |
1233d588 IM |
218 | /* Only flush the TLB if we actually modified any entries: */ |
219 | if (pages) | |
220 | flush_tlb_range(vma, start, end); | |
d303cf46 | 221 | clear_tlb_flush_pending(mm); |
7da4d641 PZ |
222 | |
223 | return pages; | |
224 | } | |
225 | ||
226 | unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, | |
227 | unsigned long end, pgprot_t newprot, | |
4b10e7d5 | 228 | int dirty_accountable, int prot_numa) |
7da4d641 PZ |
229 | { |
230 | struct mm_struct *mm = vma->vm_mm; | |
231 | unsigned long pages; | |
232 | ||
233 | mmu_notifier_invalidate_range_start(mm, start, end); | |
234 | if (is_vm_hugetlb_page(vma)) | |
235 | pages = hugetlb_change_protection(vma, start, end, newprot); | |
236 | else | |
4b10e7d5 | 237 | pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa); |
7da4d641 PZ |
238 | mmu_notifier_invalidate_range_end(mm, start, end); |
239 | ||
240 | return pages; | |
1da177e4 LT |
241 | } |
242 | ||
b6a2fea3 | 243 | int |
1da177e4 LT |
244 | mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, |
245 | unsigned long start, unsigned long end, unsigned long newflags) | |
246 | { | |
247 | struct mm_struct *mm = vma->vm_mm; | |
248 | unsigned long oldflags = vma->vm_flags; | |
249 | long nrpages = (end - start) >> PAGE_SHIFT; | |
250 | unsigned long charged = 0; | |
1da177e4 LT |
251 | pgoff_t pgoff; |
252 | int error; | |
c1e6098b | 253 | int dirty_accountable = 0; |
1da177e4 LT |
254 | |
255 | if (newflags == oldflags) { | |
256 | *pprev = vma; | |
257 | return 0; | |
258 | } | |
259 | ||
260 | /* | |
261 | * If we make a private mapping writable we increase our commit; | |
262 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
263 | * make it unwritable again. hugetlb mapping were accounted for |
264 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
265 | */ |
266 | if (newflags & VM_WRITE) { | |
5a6fe125 | 267 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 268 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 269 | charged = nrpages; |
191c5424 | 270 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
271 | return -ENOMEM; |
272 | newflags |= VM_ACCOUNT; | |
273 | } | |
274 | } | |
275 | ||
1da177e4 LT |
276 | /* |
277 | * First try to merge with previous and/or next vma. | |
278 | */ | |
279 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
280 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
6fa3eb70 S |
281 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
282 | vma_get_anon_name(vma)); | |
1da177e4 LT |
283 | if (*pprev) { |
284 | vma = *pprev; | |
285 | goto success; | |
286 | } | |
287 | ||
288 | *pprev = vma; | |
289 | ||
290 | if (start != vma->vm_start) { | |
291 | error = split_vma(mm, vma, start, 1); | |
292 | if (error) | |
293 | goto fail; | |
294 | } | |
295 | ||
296 | if (end != vma->vm_end) { | |
297 | error = split_vma(mm, vma, end, 0); | |
298 | if (error) | |
299 | goto fail; | |
300 | } | |
301 | ||
302 | success: | |
303 | /* | |
304 | * vm_flags and vm_page_prot are protected by the mmap_sem | |
305 | * held in write mode. | |
306 | */ | |
307 | vma->vm_flags = newflags; | |
1c12c4cf VP |
308 | vma->vm_page_prot = pgprot_modify(vma->vm_page_prot, |
309 | vm_get_page_prot(newflags)); | |
310 | ||
c1e6098b | 311 | if (vma_wants_writenotify(vma)) { |
1ddd439e | 312 | vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED); |
c1e6098b PZ |
313 | dirty_accountable = 1; |
314 | } | |
d08b3851 | 315 | |
7d12efae AM |
316 | change_protection(vma, start, end, vma->vm_page_prot, |
317 | dirty_accountable, 0); | |
7da4d641 | 318 | |
ab50b8ed HD |
319 | vm_stat_account(mm, oldflags, vma->vm_file, -nrpages); |
320 | vm_stat_account(mm, newflags, vma->vm_file, nrpages); | |
63bfd738 | 321 | perf_event_mmap(vma); |
1da177e4 LT |
322 | return 0; |
323 | ||
324 | fail: | |
325 | vm_unacct_memory(charged); | |
326 | return error; | |
327 | } | |
328 | ||
6a6160a7 HC |
329 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, |
330 | unsigned long, prot) | |
1da177e4 LT |
331 | { |
332 | unsigned long vm_flags, nstart, end, tmp, reqprot; | |
333 | struct vm_area_struct *vma, *prev; | |
334 | int error = -EINVAL; | |
335 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); | |
336 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); | |
337 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
338 | return -EINVAL; | |
339 | ||
340 | if (start & ~PAGE_MASK) | |
341 | return -EINVAL; | |
342 | if (!len) | |
343 | return 0; | |
344 | len = PAGE_ALIGN(len); | |
345 | end = start + len; | |
346 | if (end <= start) | |
347 | return -ENOMEM; | |
b845f313 | 348 | if (!arch_validate_prot(prot)) |
1da177e4 LT |
349 | return -EINVAL; |
350 | ||
351 | reqprot = prot; | |
352 | /* | |
353 | * Does the application expect PROT_READ to imply PROT_EXEC: | |
354 | */ | |
b344e05c | 355 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) |
1da177e4 LT |
356 | prot |= PROT_EXEC; |
357 | ||
358 | vm_flags = calc_vm_prot_bits(prot); | |
359 | ||
360 | down_write(¤t->mm->mmap_sem); | |
361 | ||
097d5910 | 362 | vma = find_vma(current->mm, start); |
1da177e4 LT |
363 | error = -ENOMEM; |
364 | if (!vma) | |
365 | goto out; | |
097d5910 | 366 | prev = vma->vm_prev; |
1da177e4 LT |
367 | if (unlikely(grows & PROT_GROWSDOWN)) { |
368 | if (vma->vm_start >= end) | |
369 | goto out; | |
370 | start = vma->vm_start; | |
371 | error = -EINVAL; | |
372 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
373 | goto out; | |
7d12efae | 374 | } else { |
1da177e4 LT |
375 | if (vma->vm_start > start) |
376 | goto out; | |
377 | if (unlikely(grows & PROT_GROWSUP)) { | |
378 | end = vma->vm_end; | |
379 | error = -EINVAL; | |
380 | if (!(vma->vm_flags & VM_GROWSUP)) | |
381 | goto out; | |
382 | } | |
383 | } | |
384 | if (start > vma->vm_start) | |
385 | prev = vma; | |
386 | ||
387 | for (nstart = start ; ; ) { | |
388 | unsigned long newflags; | |
389 | ||
7d12efae | 390 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 | 391 | |
7d12efae AM |
392 | newflags = vm_flags; |
393 | newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC)); | |
1da177e4 | 394 | |
7e2cff42 PBG |
395 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
396 | if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) { | |
1da177e4 LT |
397 | error = -EACCES; |
398 | goto out; | |
399 | } | |
400 | ||
401 | error = security_file_mprotect(vma, reqprot, prot); | |
402 | if (error) | |
403 | goto out; | |
404 | ||
405 | tmp = vma->vm_end; | |
406 | if (tmp > end) | |
407 | tmp = end; | |
408 | error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); | |
409 | if (error) | |
410 | goto out; | |
411 | nstart = tmp; | |
412 | ||
413 | if (nstart < prev->vm_end) | |
414 | nstart = prev->vm_end; | |
415 | if (nstart >= end) | |
416 | goto out; | |
417 | ||
418 | vma = prev->vm_next; | |
419 | if (!vma || vma->vm_start != nstart) { | |
420 | error = -ENOMEM; | |
421 | goto out; | |
422 | } | |
423 | } | |
424 | out: | |
425 | up_write(¤t->mm->mmap_sem); | |
426 | return error; | |
427 | } |