]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * mm/mprotect.c | |
4 | * | |
5 | * (C) Copyright 1994 Linus Torvalds | |
6 | * (C) Copyright 2002 Christoph Hellwig | |
7 | * | |
046c6884 | 8 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
1da177e4 LT |
9 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved |
10 | */ | |
11 | ||
a520110e | 12 | #include <linux/pagewalk.h> |
1da177e4 | 13 | #include <linux/hugetlb.h> |
1da177e4 LT |
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> | |
0697212a CL |
22 | #include <linux/swap.h> |
23 | #include <linux/swapops.h> | |
cddb8a5c | 24 | #include <linux/mmu_notifier.h> |
64cdd548 | 25 | #include <linux/migrate.h> |
cdd6c482 | 26 | #include <linux/perf_event.h> |
e8c24d3a | 27 | #include <linux/pkeys.h> |
64a9a34e | 28 | #include <linux/ksm.h> |
7c0f6ba6 | 29 | #include <linux/uaccess.h> |
09a913a7 | 30 | #include <linux/mm_inline.h> |
ca5999fd | 31 | #include <linux/pgtable.h> |
1da177e4 | 32 | #include <asm/cacheflush.h> |
e8c24d3a | 33 | #include <asm/mmu_context.h> |
1da177e4 LT |
34 | #include <asm/tlbflush.h> |
35 | ||
36f88188 KS |
36 | #include "internal.h" |
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, |
58705444 | 40 | unsigned long cp_flags) |
1da177e4 | 41 | { |
0697212a | 42 | pte_t *pte, oldpte; |
705e87c0 | 43 | spinlock_t *ptl; |
7da4d641 | 44 | unsigned long pages = 0; |
3e321587 | 45 | int target_node = NUMA_NO_NODE; |
58705444 PX |
46 | bool dirty_accountable = cp_flags & MM_CP_DIRTY_ACCT; |
47 | bool prot_numa = cp_flags & MM_CP_PROT_NUMA; | |
292924b2 PX |
48 | bool uffd_wp = cp_flags & MM_CP_UFFD_WP; |
49 | bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE; | |
1da177e4 | 50 | |
175ad4f1 | 51 | /* |
c1e8d7c6 | 52 | * Can be called with only the mmap_lock for reading by |
175ad4f1 AA |
53 | * prot_numa so we must check the pmd isn't constantly |
54 | * changing from under us from pmd_none to pmd_trans_huge | |
55 | * and/or the other way around. | |
56 | */ | |
57 | if (pmd_trans_unstable(pmd)) | |
58 | return 0; | |
59 | ||
60 | /* | |
61 | * The pmd points to a regular pte so the pmd can't change | |
c1e8d7c6 | 62 | * from under us even if the mmap_lock is only hold for |
175ad4f1 AA |
63 | * reading. |
64 | */ | |
65 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
1ad9f620 | 66 | |
3e321587 AK |
67 | /* Get target node for single threaded private VMAs */ |
68 | if (prot_numa && !(vma->vm_flags & VM_SHARED) && | |
69 | atomic_read(&vma->vm_mm->mm_users) == 1) | |
70 | target_node = numa_node_id(); | |
71 | ||
3ea27719 | 72 | flush_tlb_batched_pending(vma->vm_mm); |
6606c3e0 | 73 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 74 | do { |
0697212a CL |
75 | oldpte = *pte; |
76 | if (pte_present(oldpte)) { | |
1da177e4 | 77 | pte_t ptent; |
b191f9b1 | 78 | bool preserve_write = prot_numa && pte_write(oldpte); |
1da177e4 | 79 | |
e944fd67 MG |
80 | /* |
81 | * Avoid trapping faults against the zero or KSM | |
82 | * pages. See similar comment in change_huge_pmd. | |
83 | */ | |
84 | if (prot_numa) { | |
85 | struct page *page; | |
86 | ||
a818f536 HY |
87 | /* Avoid TLB flush if possible */ |
88 | if (pte_protnone(oldpte)) | |
89 | continue; | |
90 | ||
e944fd67 MG |
91 | page = vm_normal_page(vma, addr, oldpte); |
92 | if (!page || PageKsm(page)) | |
93 | continue; | |
10c1045f | 94 | |
859d4adc HW |
95 | /* Also skip shared copy-on-write pages */ |
96 | if (is_cow_mapping(vma->vm_flags) && | |
97 | page_mapcount(page) != 1) | |
98 | continue; | |
99 | ||
09a913a7 MG |
100 | /* |
101 | * While migration can move some dirty pages, | |
102 | * it cannot move them all from MIGRATE_ASYNC | |
103 | * context. | |
104 | */ | |
9de4f22a | 105 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
106 | continue; |
107 | ||
3e321587 AK |
108 | /* |
109 | * Don't mess with PTEs if page is already on the node | |
110 | * a single-threaded process is running on. | |
111 | */ | |
112 | if (target_node == page_to_nid(page)) | |
113 | continue; | |
e944fd67 MG |
114 | } |
115 | ||
04a86453 AK |
116 | oldpte = ptep_modify_prot_start(vma, addr, pte); |
117 | ptent = pte_modify(oldpte, newprot); | |
b191f9b1 | 118 | if (preserve_write) |
288bc549 | 119 | ptent = pte_mk_savedwrite(ptent); |
4b10e7d5 | 120 | |
292924b2 PX |
121 | if (uffd_wp) { |
122 | ptent = pte_wrprotect(ptent); | |
123 | ptent = pte_mkuffd_wp(ptent); | |
124 | } else if (uffd_wp_resolve) { | |
125 | /* | |
126 | * Leave the write bit to be handled | |
127 | * by PF interrupt handler, then | |
128 | * things like COW could be properly | |
129 | * handled. | |
130 | */ | |
131 | ptent = pte_clear_uffd_wp(ptent); | |
132 | } | |
133 | ||
8a0516ed MG |
134 | /* Avoid taking write faults for known dirty pages */ |
135 | if (dirty_accountable && pte_dirty(ptent) && | |
136 | (pte_soft_dirty(ptent) || | |
137 | !(vma->vm_flags & VM_SOFTDIRTY))) { | |
138 | ptent = pte_mkwrite(ptent); | |
4b10e7d5 | 139 | } |
04a86453 | 140 | ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); |
8a0516ed | 141 | pages++; |
f45ec5ff | 142 | } else if (is_swap_pte(oldpte)) { |
0697212a | 143 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
f45ec5ff | 144 | pte_t newpte; |
0697212a CL |
145 | |
146 | if (is_write_migration_entry(entry)) { | |
147 | /* | |
148 | * A protection check is difficult so | |
149 | * just be safe and disable write | |
150 | */ | |
151 | make_migration_entry_read(&entry); | |
c3d16e16 CG |
152 | newpte = swp_entry_to_pte(entry); |
153 | if (pte_swp_soft_dirty(oldpte)) | |
154 | newpte = pte_swp_mksoft_dirty(newpte); | |
f45ec5ff PX |
155 | if (pte_swp_uffd_wp(oldpte)) |
156 | newpte = pte_swp_mkuffd_wp(newpte); | |
157 | } else if (is_write_device_private_entry(entry)) { | |
5042db43 JG |
158 | /* |
159 | * We do not preserve soft-dirtiness. See | |
160 | * copy_one_pte() for explanation. | |
161 | */ | |
162 | make_device_private_entry_read(&entry); | |
163 | newpte = swp_entry_to_pte(entry); | |
f45ec5ff PX |
164 | if (pte_swp_uffd_wp(oldpte)) |
165 | newpte = pte_swp_mkuffd_wp(newpte); | |
166 | } else { | |
167 | newpte = oldpte; | |
168 | } | |
5042db43 | 169 | |
f45ec5ff PX |
170 | if (uffd_wp) |
171 | newpte = pte_swp_mkuffd_wp(newpte); | |
172 | else if (uffd_wp_resolve) | |
173 | newpte = pte_swp_clear_uffd_wp(newpte); | |
174 | ||
175 | if (!pte_same(oldpte, newpte)) { | |
176 | set_pte_at(vma->vm_mm, addr, pte, newpte); | |
5042db43 JG |
177 | pages++; |
178 | } | |
1da177e4 LT |
179 | } |
180 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 181 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 182 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
183 | |
184 | return pages; | |
1da177e4 LT |
185 | } |
186 | ||
8b272b3c MG |
187 | /* |
188 | * Used when setting automatic NUMA hinting protection where it is | |
189 | * critical that a numa hinting PMD is not confused with a bad PMD. | |
190 | */ | |
191 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) | |
192 | { | |
193 | pmd_t pmdval = pmd_read_atomic(pmd); | |
194 | ||
195 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ | |
196 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
197 | barrier(); | |
198 | #endif | |
199 | ||
200 | if (pmd_none(pmdval)) | |
201 | return 1; | |
202 | if (pmd_trans_huge(pmdval)) | |
203 | return 0; | |
204 | if (unlikely(pmd_bad(pmdval))) { | |
205 | pmd_clear_bad(pmd); | |
206 | return 1; | |
207 | } | |
208 | ||
209 | return 0; | |
210 | } | |
211 | ||
7d12efae AM |
212 | static inline unsigned long change_pmd_range(struct vm_area_struct *vma, |
213 | pud_t *pud, unsigned long addr, unsigned long end, | |
58705444 | 214 | pgprot_t newprot, unsigned long cp_flags) |
1da177e4 LT |
215 | { |
216 | pmd_t *pmd; | |
217 | unsigned long next; | |
7da4d641 | 218 | unsigned long pages = 0; |
72403b4a | 219 | unsigned long nr_huge_updates = 0; |
ac46d4f3 JG |
220 | struct mmu_notifier_range range; |
221 | ||
222 | range.start = 0; | |
1da177e4 LT |
223 | |
224 | pmd = pmd_offset(pud, addr); | |
225 | do { | |
25cbbef1 MG |
226 | unsigned long this_pages; |
227 | ||
1da177e4 | 228 | next = pmd_addr_end(addr, end); |
8b272b3c MG |
229 | |
230 | /* | |
c1e8d7c6 | 231 | * Automatic NUMA balancing walks the tables with mmap_lock |
8b272b3c MG |
232 | * held for read. It's possible a parallel update to occur |
233 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() | |
234 | * check leading to a false positive and clearing. | |
235 | * Hence, it's necessary to atomically read the PMD value | |
236 | * for all the checks. | |
237 | */ | |
238 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && | |
239 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) | |
4991c09c | 240 | goto next; |
a5338093 RR |
241 | |
242 | /* invoke the mmu notifier if the pmd is populated */ | |
ac46d4f3 | 243 | if (!range.start) { |
7269f999 JG |
244 | mmu_notifier_range_init(&range, |
245 | MMU_NOTIFY_PROTECTION_VMA, 0, | |
246 | vma, vma->vm_mm, addr, end); | |
ac46d4f3 | 247 | mmu_notifier_invalidate_range_start(&range); |
a5338093 RR |
248 | } |
249 | ||
84c3fc4e | 250 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
6b9116a6 | 251 | if (next - addr != HPAGE_PMD_SIZE) { |
fd60775a | 252 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
6b9116a6 | 253 | } else { |
f123d74a | 254 | int nr_ptes = change_huge_pmd(vma, pmd, addr, |
58705444 | 255 | newprot, cp_flags); |
f123d74a MG |
256 | |
257 | if (nr_ptes) { | |
72403b4a MG |
258 | if (nr_ptes == HPAGE_PMD_NR) { |
259 | pages += HPAGE_PMD_NR; | |
260 | nr_huge_updates++; | |
261 | } | |
1ad9f620 MG |
262 | |
263 | /* huge pmd was handled */ | |
4991c09c | 264 | goto next; |
f123d74a | 265 | } |
7da4d641 | 266 | } |
88a9ab6e | 267 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 268 | } |
25cbbef1 | 269 | this_pages = change_pte_range(vma, pmd, addr, next, newprot, |
58705444 | 270 | cp_flags); |
25cbbef1 | 271 | pages += this_pages; |
4991c09c AK |
272 | next: |
273 | cond_resched(); | |
1da177e4 | 274 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 275 | |
ac46d4f3 JG |
276 | if (range.start) |
277 | mmu_notifier_invalidate_range_end(&range); | |
a5338093 | 278 | |
72403b4a MG |
279 | if (nr_huge_updates) |
280 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 281 | return pages; |
1da177e4 LT |
282 | } |
283 | ||
7d12efae | 284 | static inline unsigned long change_pud_range(struct vm_area_struct *vma, |
c2febafc | 285 | p4d_t *p4d, unsigned long addr, unsigned long end, |
58705444 | 286 | pgprot_t newprot, unsigned long cp_flags) |
1da177e4 LT |
287 | { |
288 | pud_t *pud; | |
289 | unsigned long next; | |
7da4d641 | 290 | unsigned long pages = 0; |
1da177e4 | 291 | |
c2febafc | 292 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
293 | do { |
294 | next = pud_addr_end(addr, end); | |
295 | if (pud_none_or_clear_bad(pud)) | |
296 | continue; | |
7da4d641 | 297 | pages += change_pmd_range(vma, pud, addr, next, newprot, |
58705444 | 298 | cp_flags); |
1da177e4 | 299 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
300 | |
301 | return pages; | |
1da177e4 LT |
302 | } |
303 | ||
c2febafc KS |
304 | static inline unsigned long change_p4d_range(struct vm_area_struct *vma, |
305 | pgd_t *pgd, unsigned long addr, unsigned long end, | |
58705444 | 306 | pgprot_t newprot, unsigned long cp_flags) |
c2febafc KS |
307 | { |
308 | p4d_t *p4d; | |
309 | unsigned long next; | |
310 | unsigned long pages = 0; | |
311 | ||
312 | p4d = p4d_offset(pgd, addr); | |
313 | do { | |
314 | next = p4d_addr_end(addr, end); | |
315 | if (p4d_none_or_clear_bad(p4d)) | |
316 | continue; | |
317 | pages += change_pud_range(vma, p4d, addr, next, newprot, | |
58705444 | 318 | cp_flags); |
c2febafc KS |
319 | } while (p4d++, addr = next, addr != end); |
320 | ||
321 | return pages; | |
322 | } | |
323 | ||
7da4d641 | 324 | static unsigned long change_protection_range(struct vm_area_struct *vma, |
c1e6098b | 325 | unsigned long addr, unsigned long end, pgprot_t newprot, |
58705444 | 326 | unsigned long cp_flags) |
1da177e4 LT |
327 | { |
328 | struct mm_struct *mm = vma->vm_mm; | |
329 | pgd_t *pgd; | |
330 | unsigned long next; | |
331 | unsigned long start = addr; | |
7da4d641 | 332 | unsigned long pages = 0; |
1da177e4 LT |
333 | |
334 | BUG_ON(addr >= end); | |
335 | pgd = pgd_offset(mm, addr); | |
336 | flush_cache_range(vma, addr, end); | |
16af97dc | 337 | inc_tlb_flush_pending(mm); |
1da177e4 LT |
338 | do { |
339 | next = pgd_addr_end(addr, end); | |
340 | if (pgd_none_or_clear_bad(pgd)) | |
341 | continue; | |
c2febafc | 342 | pages += change_p4d_range(vma, pgd, addr, next, newprot, |
58705444 | 343 | cp_flags); |
1da177e4 | 344 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 345 | |
1233d588 IM |
346 | /* Only flush the TLB if we actually modified any entries: */ |
347 | if (pages) | |
348 | flush_tlb_range(vma, start, end); | |
16af97dc | 349 | dec_tlb_flush_pending(mm); |
7da4d641 PZ |
350 | |
351 | return pages; | |
352 | } | |
353 | ||
354 | unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, | |
355 | unsigned long end, pgprot_t newprot, | |
58705444 | 356 | unsigned long cp_flags) |
7da4d641 | 357 | { |
7da4d641 PZ |
358 | unsigned long pages; |
359 | ||
292924b2 PX |
360 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
361 | ||
7da4d641 PZ |
362 | if (is_vm_hugetlb_page(vma)) |
363 | pages = hugetlb_change_protection(vma, start, end, newprot); | |
364 | else | |
58705444 PX |
365 | pages = change_protection_range(vma, start, end, newprot, |
366 | cp_flags); | |
7da4d641 PZ |
367 | |
368 | return pages; | |
1da177e4 LT |
369 | } |
370 | ||
42e4089c AK |
371 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
372 | unsigned long next, struct mm_walk *walk) | |
373 | { | |
374 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
375 | 0 : -EACCES; | |
376 | } | |
377 | ||
378 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, | |
379 | unsigned long addr, unsigned long next, | |
380 | struct mm_walk *walk) | |
381 | { | |
382 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
383 | 0 : -EACCES; | |
384 | } | |
385 | ||
386 | static int prot_none_test(unsigned long addr, unsigned long next, | |
387 | struct mm_walk *walk) | |
388 | { | |
389 | return 0; | |
390 | } | |
391 | ||
7b86ac33 CH |
392 | static const struct mm_walk_ops prot_none_walk_ops = { |
393 | .pte_entry = prot_none_pte_entry, | |
394 | .hugetlb_entry = prot_none_hugetlb_entry, | |
395 | .test_walk = prot_none_test, | |
396 | }; | |
42e4089c | 397 | |
b6a2fea3 | 398 | int |
1da177e4 LT |
399 | mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, |
400 | unsigned long start, unsigned long end, unsigned long newflags) | |
401 | { | |
402 | struct mm_struct *mm = vma->vm_mm; | |
403 | unsigned long oldflags = vma->vm_flags; | |
404 | long nrpages = (end - start) >> PAGE_SHIFT; | |
405 | unsigned long charged = 0; | |
1da177e4 LT |
406 | pgoff_t pgoff; |
407 | int error; | |
c1e6098b | 408 | int dirty_accountable = 0; |
1da177e4 LT |
409 | |
410 | if (newflags == oldflags) { | |
411 | *pprev = vma; | |
412 | return 0; | |
413 | } | |
414 | ||
42e4089c AK |
415 | /* |
416 | * Do PROT_NONE PFN permission checks here when we can still | |
417 | * bail out without undoing a lot of state. This is a rather | |
418 | * uncommon case, so doesn't need to be very optimized. | |
419 | */ | |
420 | if (arch_has_pfn_modify_check() && | |
421 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && | |
6cb4d9a2 | 422 | (newflags & VM_ACCESS_FLAGS) == 0) { |
7b86ac33 CH |
423 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
424 | ||
425 | error = walk_page_range(current->mm, start, end, | |
426 | &prot_none_walk_ops, &new_pgprot); | |
42e4089c AK |
427 | if (error) |
428 | return error; | |
429 | } | |
430 | ||
1da177e4 LT |
431 | /* |
432 | * If we make a private mapping writable we increase our commit; | |
433 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
434 | * make it unwritable again. hugetlb mapping were accounted for |
435 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
436 | */ |
437 | if (newflags & VM_WRITE) { | |
84638335 KK |
438 | /* Check space limits when area turns into data. */ |
439 | if (!may_expand_vm(mm, newflags, nrpages) && | |
440 | may_expand_vm(mm, oldflags, nrpages)) | |
441 | return -ENOMEM; | |
5a6fe125 | 442 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 443 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 444 | charged = nrpages; |
191c5424 | 445 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
446 | return -ENOMEM; |
447 | newflags |= VM_ACCOUNT; | |
448 | } | |
449 | } | |
450 | ||
1da177e4 LT |
451 | /* |
452 | * First try to merge with previous and/or next vma. | |
453 | */ | |
454 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
455 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
19a809af AA |
456 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
457 | vma->vm_userfaultfd_ctx); | |
1da177e4 LT |
458 | if (*pprev) { |
459 | vma = *pprev; | |
e86f15ee | 460 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
1da177e4 LT |
461 | goto success; |
462 | } | |
463 | ||
464 | *pprev = vma; | |
465 | ||
466 | if (start != vma->vm_start) { | |
467 | error = split_vma(mm, vma, start, 1); | |
468 | if (error) | |
469 | goto fail; | |
470 | } | |
471 | ||
472 | if (end != vma->vm_end) { | |
473 | error = split_vma(mm, vma, end, 0); | |
474 | if (error) | |
475 | goto fail; | |
476 | } | |
477 | ||
478 | success: | |
479 | /* | |
c1e8d7c6 | 480 | * vm_flags and vm_page_prot are protected by the mmap_lock |
1da177e4 LT |
481 | * held in write mode. |
482 | */ | |
483 | vma->vm_flags = newflags; | |
6d2329f8 | 484 | dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot); |
64e45507 | 485 | vma_set_page_prot(vma); |
d08b3851 | 486 | |
7d12efae | 487 | change_protection(vma, start, end, vma->vm_page_prot, |
58705444 | 488 | dirty_accountable ? MM_CP_DIRTY_ACCT : 0); |
7da4d641 | 489 | |
36f88188 KS |
490 | /* |
491 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major | |
492 | * fault on access. | |
493 | */ | |
494 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && | |
495 | (newflags & VM_WRITE)) { | |
496 | populate_vma_page_range(vma, start, end, NULL); | |
497 | } | |
498 | ||
84638335 KK |
499 | vm_stat_account(mm, oldflags, -nrpages); |
500 | vm_stat_account(mm, newflags, nrpages); | |
63bfd738 | 501 | perf_event_mmap(vma); |
1da177e4 LT |
502 | return 0; |
503 | ||
504 | fail: | |
505 | vm_unacct_memory(charged); | |
506 | return error; | |
507 | } | |
508 | ||
7d06d9c9 DH |
509 | /* |
510 | * pkey==-1 when doing a legacy mprotect() | |
511 | */ | |
512 | static int do_mprotect_pkey(unsigned long start, size_t len, | |
513 | unsigned long prot, int pkey) | |
1da177e4 | 514 | { |
62b5f7d0 | 515 | unsigned long nstart, end, tmp, reqprot; |
1da177e4 LT |
516 | struct vm_area_struct *vma, *prev; |
517 | int error = -EINVAL; | |
518 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); | |
f138556d PK |
519 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
520 | (prot & PROT_READ); | |
521 | ||
057d3389 AK |
522 | start = untagged_addr(start); |
523 | ||
1da177e4 LT |
524 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
525 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
526 | return -EINVAL; | |
527 | ||
528 | if (start & ~PAGE_MASK) | |
529 | return -EINVAL; | |
530 | if (!len) | |
531 | return 0; | |
532 | len = PAGE_ALIGN(len); | |
533 | end = start + len; | |
534 | if (end <= start) | |
535 | return -ENOMEM; | |
9035cf9a | 536 | if (!arch_validate_prot(prot, start)) |
1da177e4 LT |
537 | return -EINVAL; |
538 | ||
539 | reqprot = prot; | |
1da177e4 | 540 | |
d8ed45c5 | 541 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 542 | return -EINTR; |
1da177e4 | 543 | |
e8c24d3a DH |
544 | /* |
545 | * If userspace did not allocate the pkey, do not let | |
546 | * them use it here. | |
547 | */ | |
548 | error = -EINVAL; | |
549 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) | |
550 | goto out; | |
551 | ||
097d5910 | 552 | vma = find_vma(current->mm, start); |
1da177e4 LT |
553 | error = -ENOMEM; |
554 | if (!vma) | |
555 | goto out; | |
097d5910 | 556 | prev = vma->vm_prev; |
1da177e4 LT |
557 | if (unlikely(grows & PROT_GROWSDOWN)) { |
558 | if (vma->vm_start >= end) | |
559 | goto out; | |
560 | start = vma->vm_start; | |
561 | error = -EINVAL; | |
562 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
563 | goto out; | |
7d12efae | 564 | } else { |
1da177e4 LT |
565 | if (vma->vm_start > start) |
566 | goto out; | |
567 | if (unlikely(grows & PROT_GROWSUP)) { | |
568 | end = vma->vm_end; | |
569 | error = -EINVAL; | |
570 | if (!(vma->vm_flags & VM_GROWSUP)) | |
571 | goto out; | |
572 | } | |
573 | } | |
574 | if (start > vma->vm_start) | |
575 | prev = vma; | |
576 | ||
577 | for (nstart = start ; ; ) { | |
a8502b67 | 578 | unsigned long mask_off_old_flags; |
1da177e4 | 579 | unsigned long newflags; |
7d06d9c9 | 580 | int new_vma_pkey; |
1da177e4 | 581 | |
7d12efae | 582 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 | 583 | |
f138556d PK |
584 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
585 | if (rier && (vma->vm_flags & VM_MAYEXEC)) | |
586 | prot |= PROT_EXEC; | |
587 | ||
a8502b67 DH |
588 | /* |
589 | * Each mprotect() call explicitly passes r/w/x permissions. | |
590 | * If a permission is not passed to mprotect(), it must be | |
591 | * cleared from the VMA. | |
592 | */ | |
593 | mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC | | |
2c2d57b5 | 594 | VM_FLAGS_CLEAR; |
a8502b67 | 595 | |
7d06d9c9 DH |
596 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
597 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); | |
a8502b67 | 598 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
1da177e4 | 599 | |
7e2cff42 | 600 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
6cb4d9a2 | 601 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
1da177e4 LT |
602 | error = -EACCES; |
603 | goto out; | |
604 | } | |
605 | ||
c462ac28 CM |
606 | /* Allow architectures to sanity-check the new flags */ |
607 | if (!arch_validate_flags(newflags)) { | |
608 | error = -EINVAL; | |
609 | goto out; | |
610 | } | |
611 | ||
1da177e4 LT |
612 | error = security_file_mprotect(vma, reqprot, prot); |
613 | if (error) | |
614 | goto out; | |
615 | ||
616 | tmp = vma->vm_end; | |
617 | if (tmp > end) | |
618 | tmp = end; | |
95bb7c42 | 619 | |
dbf53f75 | 620 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
95bb7c42 | 621 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
dbf53f75 TZ |
622 | if (error) |
623 | goto out; | |
624 | } | |
95bb7c42 | 625 | |
1da177e4 LT |
626 | error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); |
627 | if (error) | |
628 | goto out; | |
95bb7c42 | 629 | |
1da177e4 LT |
630 | nstart = tmp; |
631 | ||
632 | if (nstart < prev->vm_end) | |
633 | nstart = prev->vm_end; | |
634 | if (nstart >= end) | |
635 | goto out; | |
636 | ||
637 | vma = prev->vm_next; | |
638 | if (!vma || vma->vm_start != nstart) { | |
639 | error = -ENOMEM; | |
640 | goto out; | |
641 | } | |
f138556d | 642 | prot = reqprot; |
1da177e4 LT |
643 | } |
644 | out: | |
d8ed45c5 | 645 | mmap_write_unlock(current->mm); |
1da177e4 LT |
646 | return error; |
647 | } | |
7d06d9c9 DH |
648 | |
649 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, | |
650 | unsigned long, prot) | |
651 | { | |
652 | return do_mprotect_pkey(start, len, prot, -1); | |
653 | } | |
654 | ||
c7142aea HC |
655 | #ifdef CONFIG_ARCH_HAS_PKEYS |
656 | ||
7d06d9c9 DH |
657 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
658 | unsigned long, prot, int, pkey) | |
659 | { | |
660 | return do_mprotect_pkey(start, len, prot, pkey); | |
661 | } | |
e8c24d3a DH |
662 | |
663 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) | |
664 | { | |
665 | int pkey; | |
666 | int ret; | |
667 | ||
668 | /* No flags supported yet. */ | |
669 | if (flags) | |
670 | return -EINVAL; | |
671 | /* check for unsupported init values */ | |
672 | if (init_val & ~PKEY_ACCESS_MASK) | |
673 | return -EINVAL; | |
674 | ||
d8ed45c5 | 675 | mmap_write_lock(current->mm); |
e8c24d3a DH |
676 | pkey = mm_pkey_alloc(current->mm); |
677 | ||
678 | ret = -ENOSPC; | |
679 | if (pkey == -1) | |
680 | goto out; | |
681 | ||
682 | ret = arch_set_user_pkey_access(current, pkey, init_val); | |
683 | if (ret) { | |
684 | mm_pkey_free(current->mm, pkey); | |
685 | goto out; | |
686 | } | |
687 | ret = pkey; | |
688 | out: | |
d8ed45c5 | 689 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
690 | return ret; |
691 | } | |
692 | ||
693 | SYSCALL_DEFINE1(pkey_free, int, pkey) | |
694 | { | |
695 | int ret; | |
696 | ||
d8ed45c5 | 697 | mmap_write_lock(current->mm); |
e8c24d3a | 698 | ret = mm_pkey_free(current->mm, pkey); |
d8ed45c5 | 699 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
700 | |
701 | /* | |
f0953a1b | 702 | * We could provide warnings or errors if any VMA still |
e8c24d3a DH |
703 | * has the pkey set here. |
704 | */ | |
705 | return ret; | |
706 | } | |
c7142aea HC |
707 | |
708 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |