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> |
64a9a34e | 26 | #include <linux/ksm.h> |
1da177e4 LT |
27 | #include <asm/uaccess.h> |
28 | #include <asm/pgtable.h> | |
29 | #include <asm/cacheflush.h> | |
30 | #include <asm/tlbflush.h> | |
31 | ||
1ad9f620 MG |
32 | /* |
33 | * For a prot_numa update we only hold mmap_sem for read so there is a | |
34 | * potential race with faulting where a pmd was temporarily none. This | |
35 | * function checks for a transhuge pmd under the appropriate lock. It | |
36 | * returns a pte if it was successfully locked or NULL if it raced with | |
37 | * a transhuge insertion. | |
38 | */ | |
39 | static pte_t *lock_pte_protection(struct vm_area_struct *vma, pmd_t *pmd, | |
40 | unsigned long addr, int prot_numa, spinlock_t **ptl) | |
41 | { | |
42 | pte_t *pte; | |
43 | spinlock_t *pmdl; | |
44 | ||
45 | /* !prot_numa is protected by mmap_sem held for write */ | |
46 | if (!prot_numa) | |
47 | return pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl); | |
48 | ||
49 | pmdl = pmd_lock(vma->vm_mm, pmd); | |
50 | if (unlikely(pmd_trans_huge(*pmd) || pmd_none(*pmd))) { | |
51 | spin_unlock(pmdl); | |
52 | return NULL; | |
53 | } | |
54 | ||
55 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl); | |
56 | spin_unlock(pmdl); | |
57 | return pte; | |
58 | } | |
59 | ||
4b10e7d5 | 60 | static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
c1e6098b | 61 | unsigned long addr, unsigned long end, pgprot_t newprot, |
0f19c179 | 62 | int dirty_accountable, int prot_numa) |
1da177e4 | 63 | { |
4b10e7d5 | 64 | struct mm_struct *mm = vma->vm_mm; |
0697212a | 65 | pte_t *pte, oldpte; |
705e87c0 | 66 | spinlock_t *ptl; |
7da4d641 | 67 | unsigned long pages = 0; |
1da177e4 | 68 | |
1ad9f620 MG |
69 | pte = lock_pte_protection(vma, pmd, addr, prot_numa, &ptl); |
70 | if (!pte) | |
71 | return 0; | |
72 | ||
6606c3e0 | 73 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 74 | do { |
0697212a CL |
75 | oldpte = *pte; |
76 | if (pte_present(oldpte)) { | |
1da177e4 LT |
77 | pte_t ptent; |
78 | ||
e944fd67 MG |
79 | /* |
80 | * Avoid trapping faults against the zero or KSM | |
81 | * pages. See similar comment in change_huge_pmd. | |
82 | */ | |
83 | if (prot_numa) { | |
84 | struct page *page; | |
85 | ||
86 | page = vm_normal_page(vma, addr, oldpte); | |
87 | if (!page || PageKsm(page)) | |
88 | continue; | |
89 | } | |
90 | ||
8a0516ed MG |
91 | ptent = ptep_modify_prot_start(mm, addr, pte); |
92 | ptent = pte_modify(ptent, newprot); | |
4b10e7d5 | 93 | |
8a0516ed MG |
94 | /* Avoid taking write faults for known dirty pages */ |
95 | if (dirty_accountable && pte_dirty(ptent) && | |
96 | (pte_soft_dirty(ptent) || | |
97 | !(vma->vm_flags & VM_SOFTDIRTY))) { | |
98 | ptent = pte_mkwrite(ptent); | |
4b10e7d5 | 99 | } |
8a0516ed MG |
100 | ptep_modify_prot_commit(mm, addr, pte, ptent); |
101 | pages++; | |
0661a336 | 102 | } else if (IS_ENABLED(CONFIG_MIGRATION)) { |
0697212a CL |
103 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
104 | ||
105 | if (is_write_migration_entry(entry)) { | |
c3d16e16 | 106 | pte_t newpte; |
0697212a CL |
107 | /* |
108 | * A protection check is difficult so | |
109 | * just be safe and disable write | |
110 | */ | |
111 | make_migration_entry_read(&entry); | |
c3d16e16 CG |
112 | newpte = swp_entry_to_pte(entry); |
113 | if (pte_swp_soft_dirty(oldpte)) | |
114 | newpte = pte_swp_mksoft_dirty(newpte); | |
115 | set_pte_at(mm, addr, pte, newpte); | |
e920e14c MG |
116 | |
117 | pages++; | |
0697212a | 118 | } |
1da177e4 LT |
119 | } |
120 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 121 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 122 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
123 | |
124 | return pages; | |
1da177e4 LT |
125 | } |
126 | ||
7d12efae AM |
127 | static inline unsigned long change_pmd_range(struct vm_area_struct *vma, |
128 | pud_t *pud, unsigned long addr, unsigned long end, | |
129 | pgprot_t newprot, int dirty_accountable, int prot_numa) | |
1da177e4 LT |
130 | { |
131 | pmd_t *pmd; | |
a5338093 | 132 | struct mm_struct *mm = vma->vm_mm; |
1da177e4 | 133 | unsigned long next; |
7da4d641 | 134 | unsigned long pages = 0; |
72403b4a | 135 | unsigned long nr_huge_updates = 0; |
a5338093 | 136 | unsigned long mni_start = 0; |
1da177e4 LT |
137 | |
138 | pmd = pmd_offset(pud, addr); | |
139 | do { | |
25cbbef1 MG |
140 | unsigned long this_pages; |
141 | ||
1da177e4 | 142 | next = pmd_addr_end(addr, end); |
88a9ab6e RR |
143 | if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd)) |
144 | continue; | |
a5338093 RR |
145 | |
146 | /* invoke the mmu notifier if the pmd is populated */ | |
147 | if (!mni_start) { | |
148 | mni_start = addr; | |
149 | mmu_notifier_invalidate_range_start(mm, mni_start, end); | |
150 | } | |
151 | ||
cd7548ab JW |
152 | if (pmd_trans_huge(*pmd)) { |
153 | if (next - addr != HPAGE_PMD_SIZE) | |
e180377f | 154 | split_huge_page_pmd(vma, addr, pmd); |
f123d74a MG |
155 | else { |
156 | int nr_ptes = change_huge_pmd(vma, pmd, addr, | |
e944fd67 | 157 | newprot, prot_numa); |
f123d74a MG |
158 | |
159 | if (nr_ptes) { | |
72403b4a MG |
160 | if (nr_ptes == HPAGE_PMD_NR) { |
161 | pages += HPAGE_PMD_NR; | |
162 | nr_huge_updates++; | |
163 | } | |
1ad9f620 MG |
164 | |
165 | /* huge pmd was handled */ | |
f123d74a MG |
166 | continue; |
167 | } | |
7da4d641 | 168 | } |
88a9ab6e | 169 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 170 | } |
25cbbef1 | 171 | this_pages = change_pte_range(vma, pmd, addr, next, newprot, |
0f19c179 | 172 | dirty_accountable, prot_numa); |
25cbbef1 | 173 | pages += this_pages; |
1da177e4 | 174 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 175 | |
a5338093 RR |
176 | if (mni_start) |
177 | mmu_notifier_invalidate_range_end(mm, mni_start, end); | |
178 | ||
72403b4a MG |
179 | if (nr_huge_updates) |
180 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 181 | return pages; |
1da177e4 LT |
182 | } |
183 | ||
7d12efae AM |
184 | static inline unsigned long change_pud_range(struct vm_area_struct *vma, |
185 | pgd_t *pgd, unsigned long addr, unsigned long end, | |
186 | pgprot_t newprot, int dirty_accountable, int prot_numa) | |
1da177e4 LT |
187 | { |
188 | pud_t *pud; | |
189 | unsigned long next; | |
7da4d641 | 190 | unsigned long pages = 0; |
1da177e4 LT |
191 | |
192 | pud = pud_offset(pgd, addr); | |
193 | do { | |
194 | next = pud_addr_end(addr, end); | |
195 | if (pud_none_or_clear_bad(pud)) | |
196 | continue; | |
7da4d641 | 197 | pages += change_pmd_range(vma, pud, addr, next, newprot, |
4b10e7d5 | 198 | dirty_accountable, prot_numa); |
1da177e4 | 199 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
200 | |
201 | return pages; | |
1da177e4 LT |
202 | } |
203 | ||
7da4d641 | 204 | static unsigned long change_protection_range(struct vm_area_struct *vma, |
c1e6098b | 205 | unsigned long addr, unsigned long end, pgprot_t newprot, |
4b10e7d5 | 206 | int dirty_accountable, int prot_numa) |
1da177e4 LT |
207 | { |
208 | struct mm_struct *mm = vma->vm_mm; | |
209 | pgd_t *pgd; | |
210 | unsigned long next; | |
211 | unsigned long start = addr; | |
7da4d641 | 212 | unsigned long pages = 0; |
1da177e4 LT |
213 | |
214 | BUG_ON(addr >= end); | |
215 | pgd = pgd_offset(mm, addr); | |
216 | flush_cache_range(vma, addr, end); | |
20841405 | 217 | set_tlb_flush_pending(mm); |
1da177e4 LT |
218 | do { |
219 | next = pgd_addr_end(addr, end); | |
220 | if (pgd_none_or_clear_bad(pgd)) | |
221 | continue; | |
7da4d641 | 222 | pages += change_pud_range(vma, pgd, addr, next, newprot, |
4b10e7d5 | 223 | dirty_accountable, prot_numa); |
1da177e4 | 224 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 225 | |
1233d588 IM |
226 | /* Only flush the TLB if we actually modified any entries: */ |
227 | if (pages) | |
228 | flush_tlb_range(vma, start, end); | |
20841405 | 229 | clear_tlb_flush_pending(mm); |
7da4d641 PZ |
230 | |
231 | return pages; | |
232 | } | |
233 | ||
234 | unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, | |
235 | unsigned long end, pgprot_t newprot, | |
4b10e7d5 | 236 | int dirty_accountable, int prot_numa) |
7da4d641 | 237 | { |
7da4d641 PZ |
238 | unsigned long pages; |
239 | ||
7da4d641 PZ |
240 | if (is_vm_hugetlb_page(vma)) |
241 | pages = hugetlb_change_protection(vma, start, end, newprot); | |
242 | else | |
4b10e7d5 | 243 | pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa); |
7da4d641 PZ |
244 | |
245 | return pages; | |
1da177e4 LT |
246 | } |
247 | ||
b6a2fea3 | 248 | int |
1da177e4 LT |
249 | mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, |
250 | unsigned long start, unsigned long end, unsigned long newflags) | |
251 | { | |
252 | struct mm_struct *mm = vma->vm_mm; | |
253 | unsigned long oldflags = vma->vm_flags; | |
254 | long nrpages = (end - start) >> PAGE_SHIFT; | |
255 | unsigned long charged = 0; | |
1da177e4 LT |
256 | pgoff_t pgoff; |
257 | int error; | |
c1e6098b | 258 | int dirty_accountable = 0; |
1da177e4 LT |
259 | |
260 | if (newflags == oldflags) { | |
261 | *pprev = vma; | |
262 | return 0; | |
263 | } | |
264 | ||
265 | /* | |
266 | * If we make a private mapping writable we increase our commit; | |
267 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
268 | * make it unwritable again. hugetlb mapping were accounted for |
269 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
270 | */ |
271 | if (newflags & VM_WRITE) { | |
5a6fe125 | 272 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 273 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 274 | charged = nrpages; |
191c5424 | 275 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
276 | return -ENOMEM; |
277 | newflags |= VM_ACCOUNT; | |
278 | } | |
279 | } | |
280 | ||
1da177e4 LT |
281 | /* |
282 | * First try to merge with previous and/or next vma. | |
283 | */ | |
284 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
285 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
286 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); | |
287 | if (*pprev) { | |
288 | vma = *pprev; | |
289 | goto success; | |
290 | } | |
291 | ||
292 | *pprev = vma; | |
293 | ||
294 | if (start != vma->vm_start) { | |
295 | error = split_vma(mm, vma, start, 1); | |
296 | if (error) | |
297 | goto fail; | |
298 | } | |
299 | ||
300 | if (end != vma->vm_end) { | |
301 | error = split_vma(mm, vma, end, 0); | |
302 | if (error) | |
303 | goto fail; | |
304 | } | |
305 | ||
306 | success: | |
307 | /* | |
308 | * vm_flags and vm_page_prot are protected by the mmap_sem | |
309 | * held in write mode. | |
310 | */ | |
311 | vma->vm_flags = newflags; | |
64e45507 PF |
312 | dirty_accountable = vma_wants_writenotify(vma); |
313 | vma_set_page_prot(vma); | |
d08b3851 | 314 | |
7d12efae AM |
315 | change_protection(vma, start, end, vma->vm_page_prot, |
316 | dirty_accountable, 0); | |
7da4d641 | 317 | |
ab50b8ed HD |
318 | vm_stat_account(mm, oldflags, vma->vm_file, -nrpages); |
319 | vm_stat_account(mm, newflags, vma->vm_file, nrpages); | |
63bfd738 | 320 | perf_event_mmap(vma); |
1da177e4 LT |
321 | return 0; |
322 | ||
323 | fail: | |
324 | vm_unacct_memory(charged); | |
325 | return error; | |
326 | } | |
327 | ||
6a6160a7 HC |
328 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, |
329 | unsigned long, prot) | |
1da177e4 LT |
330 | { |
331 | unsigned long vm_flags, nstart, end, tmp, reqprot; | |
332 | struct vm_area_struct *vma, *prev; | |
333 | int error = -EINVAL; | |
334 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); | |
335 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); | |
336 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
337 | return -EINVAL; | |
338 | ||
339 | if (start & ~PAGE_MASK) | |
340 | return -EINVAL; | |
341 | if (!len) | |
342 | return 0; | |
343 | len = PAGE_ALIGN(len); | |
344 | end = start + len; | |
345 | if (end <= start) | |
346 | return -ENOMEM; | |
b845f313 | 347 | if (!arch_validate_prot(prot)) |
1da177e4 LT |
348 | return -EINVAL; |
349 | ||
350 | reqprot = prot; | |
351 | /* | |
352 | * Does the application expect PROT_READ to imply PROT_EXEC: | |
353 | */ | |
b344e05c | 354 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) |
1da177e4 LT |
355 | prot |= PROT_EXEC; |
356 | ||
357 | vm_flags = calc_vm_prot_bits(prot); | |
358 | ||
359 | down_write(¤t->mm->mmap_sem); | |
360 | ||
097d5910 | 361 | vma = find_vma(current->mm, start); |
1da177e4 LT |
362 | error = -ENOMEM; |
363 | if (!vma) | |
364 | goto out; | |
097d5910 | 365 | prev = vma->vm_prev; |
1da177e4 LT |
366 | if (unlikely(grows & PROT_GROWSDOWN)) { |
367 | if (vma->vm_start >= end) | |
368 | goto out; | |
369 | start = vma->vm_start; | |
370 | error = -EINVAL; | |
371 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
372 | goto out; | |
7d12efae | 373 | } else { |
1da177e4 LT |
374 | if (vma->vm_start > start) |
375 | goto out; | |
376 | if (unlikely(grows & PROT_GROWSUP)) { | |
377 | end = vma->vm_end; | |
378 | error = -EINVAL; | |
379 | if (!(vma->vm_flags & VM_GROWSUP)) | |
380 | goto out; | |
381 | } | |
382 | } | |
383 | if (start > vma->vm_start) | |
384 | prev = vma; | |
385 | ||
386 | for (nstart = start ; ; ) { | |
387 | unsigned long newflags; | |
388 | ||
7d12efae | 389 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 | 390 | |
7d12efae AM |
391 | newflags = vm_flags; |
392 | newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC)); | |
1da177e4 | 393 | |
7e2cff42 PBG |
394 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
395 | if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) { | |
1da177e4 LT |
396 | error = -EACCES; |
397 | goto out; | |
398 | } | |
399 | ||
400 | error = security_file_mprotect(vma, reqprot, prot); | |
401 | if (error) | |
402 | goto out; | |
403 | ||
404 | tmp = vma->vm_end; | |
405 | if (tmp > end) | |
406 | tmp = end; | |
407 | error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); | |
408 | if (error) | |
409 | goto out; | |
410 | nstart = tmp; | |
411 | ||
412 | if (nstart < prev->vm_end) | |
413 | nstart = prev->vm_end; | |
414 | if (nstart >= end) | |
415 | goto out; | |
416 | ||
417 | vma = prev->vm_next; | |
418 | if (!vma || vma->vm_start != nstart) { | |
419 | error = -ENOMEM; | |
420 | goto out; | |
421 | } | |
422 | } | |
423 | out: | |
424 | up_write(¤t->mm->mmap_sem); | |
425 | return error; | |
426 | } |