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> |
a1a3a2fc | 32 | #include <linux/sched/sysctl.h> |
1da177e4 | 33 | #include <asm/cacheflush.h> |
e8c24d3a | 34 | #include <asm/mmu_context.h> |
1da177e4 LT |
35 | #include <asm/tlbflush.h> |
36 | ||
36f88188 KS |
37 | #include "internal.h" |
38 | ||
4b10e7d5 | 39 | static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
c1e6098b | 40 | unsigned long addr, unsigned long end, pgprot_t newprot, |
58705444 | 41 | unsigned long cp_flags) |
1da177e4 | 42 | { |
0697212a | 43 | pte_t *pte, oldpte; |
705e87c0 | 44 | spinlock_t *ptl; |
7da4d641 | 45 | unsigned long pages = 0; |
3e321587 | 46 | int target_node = NUMA_NO_NODE; |
58705444 PX |
47 | bool dirty_accountable = cp_flags & MM_CP_DIRTY_ACCT; |
48 | bool prot_numa = cp_flags & MM_CP_PROT_NUMA; | |
292924b2 PX |
49 | bool uffd_wp = cp_flags & MM_CP_UFFD_WP; |
50 | bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE; | |
1da177e4 | 51 | |
175ad4f1 | 52 | /* |
c1e8d7c6 | 53 | * Can be called with only the mmap_lock for reading by |
175ad4f1 AA |
54 | * prot_numa so we must check the pmd isn't constantly |
55 | * changing from under us from pmd_none to pmd_trans_huge | |
56 | * and/or the other way around. | |
57 | */ | |
58 | if (pmd_trans_unstable(pmd)) | |
59 | return 0; | |
60 | ||
61 | /* | |
62 | * The pmd points to a regular pte so the pmd can't change | |
c1e8d7c6 | 63 | * from under us even if the mmap_lock is only hold for |
175ad4f1 AA |
64 | * reading. |
65 | */ | |
66 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
1ad9f620 | 67 | |
3e321587 AK |
68 | /* Get target node for single threaded private VMAs */ |
69 | if (prot_numa && !(vma->vm_flags & VM_SHARED) && | |
70 | atomic_read(&vma->vm_mm->mm_users) == 1) | |
71 | target_node = numa_node_id(); | |
72 | ||
3ea27719 | 73 | flush_tlb_batched_pending(vma->vm_mm); |
6606c3e0 | 74 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 75 | do { |
0697212a CL |
76 | oldpte = *pte; |
77 | if (pte_present(oldpte)) { | |
1da177e4 | 78 | pte_t ptent; |
b191f9b1 | 79 | bool preserve_write = prot_numa && pte_write(oldpte); |
1da177e4 | 80 | |
e944fd67 MG |
81 | /* |
82 | * Avoid trapping faults against the zero or KSM | |
83 | * pages. See similar comment in change_huge_pmd. | |
84 | */ | |
85 | if (prot_numa) { | |
86 | struct page *page; | |
a1a3a2fc | 87 | int nid; |
e944fd67 | 88 | |
a818f536 HY |
89 | /* Avoid TLB flush if possible */ |
90 | if (pte_protnone(oldpte)) | |
91 | continue; | |
92 | ||
e944fd67 MG |
93 | page = vm_normal_page(vma, addr, oldpte); |
94 | if (!page || PageKsm(page)) | |
95 | continue; | |
10c1045f | 96 | |
859d4adc HW |
97 | /* Also skip shared copy-on-write pages */ |
98 | if (is_cow_mapping(vma->vm_flags) && | |
80d47f5d | 99 | page_count(page) != 1) |
859d4adc HW |
100 | continue; |
101 | ||
09a913a7 MG |
102 | /* |
103 | * While migration can move some dirty pages, | |
104 | * it cannot move them all from MIGRATE_ASYNC | |
105 | * context. | |
106 | */ | |
9de4f22a | 107 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
108 | continue; |
109 | ||
3e321587 AK |
110 | /* |
111 | * Don't mess with PTEs if page is already on the node | |
112 | * a single-threaded process is running on. | |
113 | */ | |
a1a3a2fc HY |
114 | nid = page_to_nid(page); |
115 | if (target_node == nid) | |
116 | continue; | |
117 | ||
118 | /* | |
119 | * Skip scanning top tier node if normal numa | |
120 | * balancing is disabled | |
121 | */ | |
122 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_NORMAL) && | |
123 | node_is_toptier(nid)) | |
3e321587 | 124 | continue; |
e944fd67 MG |
125 | } |
126 | ||
04a86453 AK |
127 | oldpte = ptep_modify_prot_start(vma, addr, pte); |
128 | ptent = pte_modify(oldpte, newprot); | |
b191f9b1 | 129 | if (preserve_write) |
288bc549 | 130 | ptent = pte_mk_savedwrite(ptent); |
4b10e7d5 | 131 | |
292924b2 PX |
132 | if (uffd_wp) { |
133 | ptent = pte_wrprotect(ptent); | |
134 | ptent = pte_mkuffd_wp(ptent); | |
135 | } else if (uffd_wp_resolve) { | |
136 | /* | |
137 | * Leave the write bit to be handled | |
138 | * by PF interrupt handler, then | |
139 | * things like COW could be properly | |
140 | * handled. | |
141 | */ | |
142 | ptent = pte_clear_uffd_wp(ptent); | |
143 | } | |
144 | ||
8a0516ed MG |
145 | /* Avoid taking write faults for known dirty pages */ |
146 | if (dirty_accountable && pte_dirty(ptent) && | |
147 | (pte_soft_dirty(ptent) || | |
148 | !(vma->vm_flags & VM_SOFTDIRTY))) { | |
149 | ptent = pte_mkwrite(ptent); | |
4b10e7d5 | 150 | } |
04a86453 | 151 | ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); |
8a0516ed | 152 | pages++; |
f45ec5ff | 153 | } else if (is_swap_pte(oldpte)) { |
0697212a | 154 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
6c287605 | 155 | struct page *page = pfn_swap_entry_to_page(entry); |
f45ec5ff | 156 | pte_t newpte; |
0697212a | 157 | |
4dd845b5 | 158 | if (is_writable_migration_entry(entry)) { |
0697212a CL |
159 | /* |
160 | * A protection check is difficult so | |
161 | * just be safe and disable write | |
162 | */ | |
6c287605 DH |
163 | if (PageAnon(page)) |
164 | entry = make_readable_exclusive_migration_entry( | |
165 | swp_offset(entry)); | |
166 | else | |
167 | entry = make_readable_migration_entry(swp_offset(entry)); | |
c3d16e16 CG |
168 | newpte = swp_entry_to_pte(entry); |
169 | if (pte_swp_soft_dirty(oldpte)) | |
170 | newpte = pte_swp_mksoft_dirty(newpte); | |
f45ec5ff PX |
171 | if (pte_swp_uffd_wp(oldpte)) |
172 | newpte = pte_swp_mkuffd_wp(newpte); | |
4dd845b5 | 173 | } else if (is_writable_device_private_entry(entry)) { |
5042db43 JG |
174 | /* |
175 | * We do not preserve soft-dirtiness. See | |
176 | * copy_one_pte() for explanation. | |
177 | */ | |
4dd845b5 AP |
178 | entry = make_readable_device_private_entry( |
179 | swp_offset(entry)); | |
5042db43 | 180 | newpte = swp_entry_to_pte(entry); |
f45ec5ff PX |
181 | if (pte_swp_uffd_wp(oldpte)) |
182 | newpte = pte_swp_mkuffd_wp(newpte); | |
b756a3b5 AP |
183 | } else if (is_writable_device_exclusive_entry(entry)) { |
184 | entry = make_readable_device_exclusive_entry( | |
185 | swp_offset(entry)); | |
186 | newpte = swp_entry_to_pte(entry); | |
187 | if (pte_swp_soft_dirty(oldpte)) | |
188 | newpte = pte_swp_mksoft_dirty(newpte); | |
189 | if (pte_swp_uffd_wp(oldpte)) | |
190 | newpte = pte_swp_mkuffd_wp(newpte); | |
f45ec5ff PX |
191 | } else { |
192 | newpte = oldpte; | |
193 | } | |
5042db43 | 194 | |
f45ec5ff PX |
195 | if (uffd_wp) |
196 | newpte = pte_swp_mkuffd_wp(newpte); | |
197 | else if (uffd_wp_resolve) | |
198 | newpte = pte_swp_clear_uffd_wp(newpte); | |
199 | ||
200 | if (!pte_same(oldpte, newpte)) { | |
201 | set_pte_at(vma->vm_mm, addr, pte, newpte); | |
5042db43 JG |
202 | pages++; |
203 | } | |
1da177e4 LT |
204 | } |
205 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 206 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 207 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
208 | |
209 | return pages; | |
1da177e4 LT |
210 | } |
211 | ||
8b272b3c MG |
212 | /* |
213 | * Used when setting automatic NUMA hinting protection where it is | |
214 | * critical that a numa hinting PMD is not confused with a bad PMD. | |
215 | */ | |
216 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) | |
217 | { | |
218 | pmd_t pmdval = pmd_read_atomic(pmd); | |
219 | ||
220 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ | |
221 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
222 | barrier(); | |
223 | #endif | |
224 | ||
225 | if (pmd_none(pmdval)) | |
226 | return 1; | |
227 | if (pmd_trans_huge(pmdval)) | |
228 | return 0; | |
229 | if (unlikely(pmd_bad(pmdval))) { | |
230 | pmd_clear_bad(pmd); | |
231 | return 1; | |
232 | } | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
7d12efae AM |
237 | static inline unsigned long change_pmd_range(struct vm_area_struct *vma, |
238 | pud_t *pud, unsigned long addr, unsigned long end, | |
58705444 | 239 | pgprot_t newprot, unsigned long cp_flags) |
1da177e4 LT |
240 | { |
241 | pmd_t *pmd; | |
242 | unsigned long next; | |
7da4d641 | 243 | unsigned long pages = 0; |
72403b4a | 244 | unsigned long nr_huge_updates = 0; |
ac46d4f3 JG |
245 | struct mmu_notifier_range range; |
246 | ||
247 | range.start = 0; | |
1da177e4 LT |
248 | |
249 | pmd = pmd_offset(pud, addr); | |
250 | do { | |
25cbbef1 MG |
251 | unsigned long this_pages; |
252 | ||
1da177e4 | 253 | next = pmd_addr_end(addr, end); |
8b272b3c MG |
254 | |
255 | /* | |
c1e8d7c6 | 256 | * Automatic NUMA balancing walks the tables with mmap_lock |
8b272b3c MG |
257 | * held for read. It's possible a parallel update to occur |
258 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() | |
259 | * check leading to a false positive and clearing. | |
260 | * Hence, it's necessary to atomically read the PMD value | |
261 | * for all the checks. | |
262 | */ | |
263 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && | |
264 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) | |
4991c09c | 265 | goto next; |
a5338093 RR |
266 | |
267 | /* invoke the mmu notifier if the pmd is populated */ | |
ac46d4f3 | 268 | if (!range.start) { |
7269f999 JG |
269 | mmu_notifier_range_init(&range, |
270 | MMU_NOTIFY_PROTECTION_VMA, 0, | |
271 | vma, vma->vm_mm, addr, end); | |
ac46d4f3 | 272 | mmu_notifier_invalidate_range_start(&range); |
a5338093 RR |
273 | } |
274 | ||
84c3fc4e | 275 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
6b9116a6 | 276 | if (next - addr != HPAGE_PMD_SIZE) { |
fd60775a | 277 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
6b9116a6 | 278 | } else { |
f123d74a | 279 | int nr_ptes = change_huge_pmd(vma, pmd, addr, |
58705444 | 280 | newprot, cp_flags); |
f123d74a MG |
281 | |
282 | if (nr_ptes) { | |
72403b4a MG |
283 | if (nr_ptes == HPAGE_PMD_NR) { |
284 | pages += HPAGE_PMD_NR; | |
285 | nr_huge_updates++; | |
286 | } | |
1ad9f620 MG |
287 | |
288 | /* huge pmd was handled */ | |
4991c09c | 289 | goto next; |
f123d74a | 290 | } |
7da4d641 | 291 | } |
88a9ab6e | 292 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 293 | } |
25cbbef1 | 294 | this_pages = change_pte_range(vma, pmd, addr, next, newprot, |
58705444 | 295 | cp_flags); |
25cbbef1 | 296 | pages += this_pages; |
4991c09c AK |
297 | next: |
298 | cond_resched(); | |
1da177e4 | 299 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 300 | |
ac46d4f3 JG |
301 | if (range.start) |
302 | mmu_notifier_invalidate_range_end(&range); | |
a5338093 | 303 | |
72403b4a MG |
304 | if (nr_huge_updates) |
305 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 306 | return pages; |
1da177e4 LT |
307 | } |
308 | ||
7d12efae | 309 | static inline unsigned long change_pud_range(struct vm_area_struct *vma, |
c2febafc | 310 | p4d_t *p4d, unsigned long addr, unsigned long end, |
58705444 | 311 | pgprot_t newprot, unsigned long cp_flags) |
1da177e4 LT |
312 | { |
313 | pud_t *pud; | |
314 | unsigned long next; | |
7da4d641 | 315 | unsigned long pages = 0; |
1da177e4 | 316 | |
c2febafc | 317 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
318 | do { |
319 | next = pud_addr_end(addr, end); | |
320 | if (pud_none_or_clear_bad(pud)) | |
321 | continue; | |
7da4d641 | 322 | pages += change_pmd_range(vma, pud, addr, next, newprot, |
58705444 | 323 | cp_flags); |
1da177e4 | 324 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
325 | |
326 | return pages; | |
1da177e4 LT |
327 | } |
328 | ||
c2febafc KS |
329 | static inline unsigned long change_p4d_range(struct vm_area_struct *vma, |
330 | pgd_t *pgd, unsigned long addr, unsigned long end, | |
58705444 | 331 | pgprot_t newprot, unsigned long cp_flags) |
c2febafc KS |
332 | { |
333 | p4d_t *p4d; | |
334 | unsigned long next; | |
335 | unsigned long pages = 0; | |
336 | ||
337 | p4d = p4d_offset(pgd, addr); | |
338 | do { | |
339 | next = p4d_addr_end(addr, end); | |
340 | if (p4d_none_or_clear_bad(p4d)) | |
341 | continue; | |
342 | pages += change_pud_range(vma, p4d, addr, next, newprot, | |
58705444 | 343 | cp_flags); |
c2febafc KS |
344 | } while (p4d++, addr = next, addr != end); |
345 | ||
346 | return pages; | |
347 | } | |
348 | ||
7da4d641 | 349 | static unsigned long change_protection_range(struct vm_area_struct *vma, |
c1e6098b | 350 | unsigned long addr, unsigned long end, pgprot_t newprot, |
58705444 | 351 | unsigned long cp_flags) |
1da177e4 LT |
352 | { |
353 | struct mm_struct *mm = vma->vm_mm; | |
354 | pgd_t *pgd; | |
355 | unsigned long next; | |
356 | unsigned long start = addr; | |
7da4d641 | 357 | unsigned long pages = 0; |
1da177e4 LT |
358 | |
359 | BUG_ON(addr >= end); | |
360 | pgd = pgd_offset(mm, addr); | |
361 | flush_cache_range(vma, addr, end); | |
16af97dc | 362 | inc_tlb_flush_pending(mm); |
1da177e4 LT |
363 | do { |
364 | next = pgd_addr_end(addr, end); | |
365 | if (pgd_none_or_clear_bad(pgd)) | |
366 | continue; | |
c2febafc | 367 | pages += change_p4d_range(vma, pgd, addr, next, newprot, |
58705444 | 368 | cp_flags); |
1da177e4 | 369 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 370 | |
1233d588 IM |
371 | /* Only flush the TLB if we actually modified any entries: */ |
372 | if (pages) | |
373 | flush_tlb_range(vma, start, end); | |
16af97dc | 374 | dec_tlb_flush_pending(mm); |
7da4d641 PZ |
375 | |
376 | return pages; | |
377 | } | |
378 | ||
379 | unsigned long change_protection(struct vm_area_struct *vma, unsigned long start, | |
380 | unsigned long end, pgprot_t newprot, | |
58705444 | 381 | unsigned long cp_flags) |
7da4d641 | 382 | { |
7da4d641 PZ |
383 | unsigned long pages; |
384 | ||
292924b2 PX |
385 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
386 | ||
7da4d641 PZ |
387 | if (is_vm_hugetlb_page(vma)) |
388 | pages = hugetlb_change_protection(vma, start, end, newprot); | |
389 | else | |
58705444 PX |
390 | pages = change_protection_range(vma, start, end, newprot, |
391 | cp_flags); | |
7da4d641 PZ |
392 | |
393 | return pages; | |
1da177e4 LT |
394 | } |
395 | ||
42e4089c AK |
396 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
397 | unsigned long next, struct mm_walk *walk) | |
398 | { | |
399 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
400 | 0 : -EACCES; | |
401 | } | |
402 | ||
403 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, | |
404 | unsigned long addr, unsigned long next, | |
405 | struct mm_walk *walk) | |
406 | { | |
407 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
408 | 0 : -EACCES; | |
409 | } | |
410 | ||
411 | static int prot_none_test(unsigned long addr, unsigned long next, | |
412 | struct mm_walk *walk) | |
413 | { | |
414 | return 0; | |
415 | } | |
416 | ||
7b86ac33 CH |
417 | static const struct mm_walk_ops prot_none_walk_ops = { |
418 | .pte_entry = prot_none_pte_entry, | |
419 | .hugetlb_entry = prot_none_hugetlb_entry, | |
420 | .test_walk = prot_none_test, | |
421 | }; | |
42e4089c | 422 | |
b6a2fea3 | 423 | int |
1da177e4 LT |
424 | mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, |
425 | unsigned long start, unsigned long end, unsigned long newflags) | |
426 | { | |
427 | struct mm_struct *mm = vma->vm_mm; | |
428 | unsigned long oldflags = vma->vm_flags; | |
429 | long nrpages = (end - start) >> PAGE_SHIFT; | |
430 | unsigned long charged = 0; | |
1da177e4 LT |
431 | pgoff_t pgoff; |
432 | int error; | |
c1e6098b | 433 | int dirty_accountable = 0; |
1da177e4 LT |
434 | |
435 | if (newflags == oldflags) { | |
436 | *pprev = vma; | |
437 | return 0; | |
438 | } | |
439 | ||
42e4089c AK |
440 | /* |
441 | * Do PROT_NONE PFN permission checks here when we can still | |
442 | * bail out without undoing a lot of state. This is a rather | |
443 | * uncommon case, so doesn't need to be very optimized. | |
444 | */ | |
445 | if (arch_has_pfn_modify_check() && | |
446 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && | |
6cb4d9a2 | 447 | (newflags & VM_ACCESS_FLAGS) == 0) { |
7b86ac33 CH |
448 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
449 | ||
450 | error = walk_page_range(current->mm, start, end, | |
451 | &prot_none_walk_ops, &new_pgprot); | |
42e4089c AK |
452 | if (error) |
453 | return error; | |
454 | } | |
455 | ||
1da177e4 LT |
456 | /* |
457 | * If we make a private mapping writable we increase our commit; | |
458 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
459 | * make it unwritable again. hugetlb mapping were accounted for |
460 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
461 | */ |
462 | if (newflags & VM_WRITE) { | |
84638335 KK |
463 | /* Check space limits when area turns into data. */ |
464 | if (!may_expand_vm(mm, newflags, nrpages) && | |
465 | may_expand_vm(mm, oldflags, nrpages)) | |
466 | return -ENOMEM; | |
5a6fe125 | 467 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 468 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 469 | charged = nrpages; |
191c5424 | 470 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
471 | return -ENOMEM; |
472 | newflags |= VM_ACCOUNT; | |
473 | } | |
474 | } | |
475 | ||
1da177e4 LT |
476 | /* |
477 | * First try to merge with previous and/or next vma. | |
478 | */ | |
479 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
480 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
19a809af | 481 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 482 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
483 | if (*pprev) { |
484 | vma = *pprev; | |
e86f15ee | 485 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
1da177e4 LT |
486 | goto success; |
487 | } | |
488 | ||
489 | *pprev = vma; | |
490 | ||
491 | if (start != vma->vm_start) { | |
492 | error = split_vma(mm, vma, start, 1); | |
493 | if (error) | |
494 | goto fail; | |
495 | } | |
496 | ||
497 | if (end != vma->vm_end) { | |
498 | error = split_vma(mm, vma, end, 0); | |
499 | if (error) | |
500 | goto fail; | |
501 | } | |
502 | ||
503 | success: | |
504 | /* | |
c1e8d7c6 | 505 | * vm_flags and vm_page_prot are protected by the mmap_lock |
1da177e4 LT |
506 | * held in write mode. |
507 | */ | |
508 | vma->vm_flags = newflags; | |
6d2329f8 | 509 | dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot); |
64e45507 | 510 | vma_set_page_prot(vma); |
d08b3851 | 511 | |
7d12efae | 512 | change_protection(vma, start, end, vma->vm_page_prot, |
58705444 | 513 | dirty_accountable ? MM_CP_DIRTY_ACCT : 0); |
7da4d641 | 514 | |
36f88188 KS |
515 | /* |
516 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major | |
517 | * fault on access. | |
518 | */ | |
519 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && | |
520 | (newflags & VM_WRITE)) { | |
521 | populate_vma_page_range(vma, start, end, NULL); | |
522 | } | |
523 | ||
84638335 KK |
524 | vm_stat_account(mm, oldflags, -nrpages); |
525 | vm_stat_account(mm, newflags, nrpages); | |
63bfd738 | 526 | perf_event_mmap(vma); |
1da177e4 LT |
527 | return 0; |
528 | ||
529 | fail: | |
530 | vm_unacct_memory(charged); | |
531 | return error; | |
532 | } | |
533 | ||
7d06d9c9 DH |
534 | /* |
535 | * pkey==-1 when doing a legacy mprotect() | |
536 | */ | |
537 | static int do_mprotect_pkey(unsigned long start, size_t len, | |
538 | unsigned long prot, int pkey) | |
1da177e4 | 539 | { |
62b5f7d0 | 540 | unsigned long nstart, end, tmp, reqprot; |
1da177e4 LT |
541 | struct vm_area_struct *vma, *prev; |
542 | int error = -EINVAL; | |
543 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); | |
f138556d PK |
544 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
545 | (prot & PROT_READ); | |
546 | ||
057d3389 AK |
547 | start = untagged_addr(start); |
548 | ||
1da177e4 LT |
549 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
550 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
551 | return -EINVAL; | |
552 | ||
553 | if (start & ~PAGE_MASK) | |
554 | return -EINVAL; | |
555 | if (!len) | |
556 | return 0; | |
557 | len = PAGE_ALIGN(len); | |
558 | end = start + len; | |
559 | if (end <= start) | |
560 | return -ENOMEM; | |
9035cf9a | 561 | if (!arch_validate_prot(prot, start)) |
1da177e4 LT |
562 | return -EINVAL; |
563 | ||
564 | reqprot = prot; | |
1da177e4 | 565 | |
d8ed45c5 | 566 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 567 | return -EINTR; |
1da177e4 | 568 | |
e8c24d3a DH |
569 | /* |
570 | * If userspace did not allocate the pkey, do not let | |
571 | * them use it here. | |
572 | */ | |
573 | error = -EINVAL; | |
574 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) | |
575 | goto out; | |
576 | ||
097d5910 | 577 | vma = find_vma(current->mm, start); |
1da177e4 LT |
578 | error = -ENOMEM; |
579 | if (!vma) | |
580 | goto out; | |
6af5fa0d | 581 | |
1da177e4 LT |
582 | if (unlikely(grows & PROT_GROWSDOWN)) { |
583 | if (vma->vm_start >= end) | |
584 | goto out; | |
585 | start = vma->vm_start; | |
586 | error = -EINVAL; | |
587 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
588 | goto out; | |
7d12efae | 589 | } else { |
1da177e4 LT |
590 | if (vma->vm_start > start) |
591 | goto out; | |
592 | if (unlikely(grows & PROT_GROWSUP)) { | |
593 | end = vma->vm_end; | |
594 | error = -EINVAL; | |
595 | if (!(vma->vm_flags & VM_GROWSUP)) | |
596 | goto out; | |
597 | } | |
598 | } | |
6af5fa0d | 599 | |
1da177e4 LT |
600 | if (start > vma->vm_start) |
601 | prev = vma; | |
6af5fa0d LS |
602 | else |
603 | prev = vma->vm_prev; | |
1da177e4 LT |
604 | |
605 | for (nstart = start ; ; ) { | |
a8502b67 | 606 | unsigned long mask_off_old_flags; |
1da177e4 | 607 | unsigned long newflags; |
7d06d9c9 | 608 | int new_vma_pkey; |
1da177e4 | 609 | |
7d12efae | 610 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 | 611 | |
f138556d PK |
612 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
613 | if (rier && (vma->vm_flags & VM_MAYEXEC)) | |
614 | prot |= PROT_EXEC; | |
615 | ||
a8502b67 DH |
616 | /* |
617 | * Each mprotect() call explicitly passes r/w/x permissions. | |
618 | * If a permission is not passed to mprotect(), it must be | |
619 | * cleared from the VMA. | |
620 | */ | |
621 | mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC | | |
2c2d57b5 | 622 | VM_FLAGS_CLEAR; |
a8502b67 | 623 | |
7d06d9c9 DH |
624 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
625 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); | |
a8502b67 | 626 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
1da177e4 | 627 | |
7e2cff42 | 628 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
6cb4d9a2 | 629 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
1da177e4 LT |
630 | error = -EACCES; |
631 | goto out; | |
632 | } | |
633 | ||
c462ac28 CM |
634 | /* Allow architectures to sanity-check the new flags */ |
635 | if (!arch_validate_flags(newflags)) { | |
636 | error = -EINVAL; | |
637 | goto out; | |
638 | } | |
639 | ||
1da177e4 LT |
640 | error = security_file_mprotect(vma, reqprot, prot); |
641 | if (error) | |
642 | goto out; | |
643 | ||
644 | tmp = vma->vm_end; | |
645 | if (tmp > end) | |
646 | tmp = end; | |
95bb7c42 | 647 | |
dbf53f75 | 648 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
95bb7c42 | 649 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
dbf53f75 TZ |
650 | if (error) |
651 | goto out; | |
652 | } | |
95bb7c42 | 653 | |
1da177e4 LT |
654 | error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); |
655 | if (error) | |
656 | goto out; | |
95bb7c42 | 657 | |
1da177e4 LT |
658 | nstart = tmp; |
659 | ||
660 | if (nstart < prev->vm_end) | |
661 | nstart = prev->vm_end; | |
662 | if (nstart >= end) | |
663 | goto out; | |
664 | ||
665 | vma = prev->vm_next; | |
666 | if (!vma || vma->vm_start != nstart) { | |
667 | error = -ENOMEM; | |
668 | goto out; | |
669 | } | |
f138556d | 670 | prot = reqprot; |
1da177e4 LT |
671 | } |
672 | out: | |
d8ed45c5 | 673 | mmap_write_unlock(current->mm); |
1da177e4 LT |
674 | return error; |
675 | } | |
7d06d9c9 DH |
676 | |
677 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, | |
678 | unsigned long, prot) | |
679 | { | |
680 | return do_mprotect_pkey(start, len, prot, -1); | |
681 | } | |
682 | ||
c7142aea HC |
683 | #ifdef CONFIG_ARCH_HAS_PKEYS |
684 | ||
7d06d9c9 DH |
685 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
686 | unsigned long, prot, int, pkey) | |
687 | { | |
688 | return do_mprotect_pkey(start, len, prot, pkey); | |
689 | } | |
e8c24d3a DH |
690 | |
691 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) | |
692 | { | |
693 | int pkey; | |
694 | int ret; | |
695 | ||
696 | /* No flags supported yet. */ | |
697 | if (flags) | |
698 | return -EINVAL; | |
699 | /* check for unsupported init values */ | |
700 | if (init_val & ~PKEY_ACCESS_MASK) | |
701 | return -EINVAL; | |
702 | ||
d8ed45c5 | 703 | mmap_write_lock(current->mm); |
e8c24d3a DH |
704 | pkey = mm_pkey_alloc(current->mm); |
705 | ||
706 | ret = -ENOSPC; | |
707 | if (pkey == -1) | |
708 | goto out; | |
709 | ||
710 | ret = arch_set_user_pkey_access(current, pkey, init_val); | |
711 | if (ret) { | |
712 | mm_pkey_free(current->mm, pkey); | |
713 | goto out; | |
714 | } | |
715 | ret = pkey; | |
716 | out: | |
d8ed45c5 | 717 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
718 | return ret; |
719 | } | |
720 | ||
721 | SYSCALL_DEFINE1(pkey_free, int, pkey) | |
722 | { | |
723 | int ret; | |
724 | ||
d8ed45c5 | 725 | mmap_write_lock(current->mm); |
e8c24d3a | 726 | ret = mm_pkey_free(current->mm, pkey); |
d8ed45c5 | 727 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
728 | |
729 | /* | |
f0953a1b | 730 | * We could provide warnings or errors if any VMA still |
e8c24d3a DH |
731 | * has the pkey set here. |
732 | */ | |
733 | return ret; | |
734 | } | |
c7142aea HC |
735 | |
736 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |