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> |
fe2567eb | 33 | #include <linux/userfaultfd_k.h> |
467b171a | 34 | #include <linux/memory-tiers.h> |
1da177e4 | 35 | #include <asm/cacheflush.h> |
e8c24d3a | 36 | #include <asm/mmu_context.h> |
1da177e4 | 37 | #include <asm/tlbflush.h> |
4a18419f | 38 | #include <asm/tlb.h> |
1da177e4 | 39 | |
36f88188 KS |
40 | #include "internal.h" |
41 | ||
6a56ccbc DH |
42 | bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr, |
43 | pte_t pte) | |
64fe24a3 DH |
44 | { |
45 | struct page *page; | |
46 | ||
7ea7e333 DH |
47 | if (WARN_ON_ONCE(!(vma->vm_flags & VM_WRITE))) |
48 | return false; | |
64fe24a3 | 49 | |
7ea7e333 | 50 | /* Don't touch entries that are not even readable. */ |
d8488773 | 51 | if (pte_protnone(pte)) |
64fe24a3 DH |
52 | return false; |
53 | ||
54 | /* Do we need write faults for softdirty tracking? */ | |
76aefad6 | 55 | if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte)) |
64fe24a3 DH |
56 | return false; |
57 | ||
58 | /* Do we need write faults for uffd-wp tracking? */ | |
59 | if (userfaultfd_pte_wp(vma, pte)) | |
60 | return false; | |
61 | ||
62 | if (!(vma->vm_flags & VM_SHARED)) { | |
63 | /* | |
7ea7e333 DH |
64 | * Writable MAP_PRIVATE mapping: We can only special-case on |
65 | * exclusive anonymous pages, because we know that our | |
66 | * write-fault handler similarly would map them writable without | |
67 | * any additional checks while holding the PT lock. | |
64fe24a3 DH |
68 | */ |
69 | page = vm_normal_page(vma, addr, pte); | |
d8488773 | 70 | return page && PageAnon(page) && PageAnonExclusive(page); |
64fe24a3 DH |
71 | } |
72 | ||
7ea7e333 DH |
73 | /* |
74 | * Writable MAP_SHARED mapping: "clean" might indicate that the FS still | |
75 | * needs a real write-fault for writenotify | |
76 | * (see vma_wants_writenotify()). If "dirty", the assumption is that the | |
77 | * FS was already notified and we can simply mark the PTE writable | |
78 | * just like the write-fault handler would do. | |
79 | */ | |
d8488773 | 80 | return pte_dirty(pte); |
64fe24a3 DH |
81 | } |
82 | ||
a79390f5 | 83 | static long change_pte_range(struct mmu_gather *tlb, |
4a18419f NA |
84 | struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr, |
85 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 | 86 | { |
0697212a | 87 | pte_t *pte, oldpte; |
705e87c0 | 88 | spinlock_t *ptl; |
a79390f5 | 89 | long pages = 0; |
3e321587 | 90 | int target_node = NUMA_NO_NODE; |
58705444 | 91 | bool prot_numa = cp_flags & MM_CP_PROT_NUMA; |
292924b2 PX |
92 | bool uffd_wp = cp_flags & MM_CP_UFFD_WP; |
93 | bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE; | |
1da177e4 | 94 | |
4a18419f NA |
95 | tlb_change_page_size(tlb, PAGE_SIZE); |
96 | ||
175ad4f1 | 97 | /* |
c1e8d7c6 | 98 | * Can be called with only the mmap_lock for reading by |
175ad4f1 AA |
99 | * prot_numa so we must check the pmd isn't constantly |
100 | * changing from under us from pmd_none to pmd_trans_huge | |
101 | * and/or the other way around. | |
102 | */ | |
103 | if (pmd_trans_unstable(pmd)) | |
104 | return 0; | |
105 | ||
106 | /* | |
107 | * The pmd points to a regular pte so the pmd can't change | |
c1e8d7c6 | 108 | * from under us even if the mmap_lock is only hold for |
175ad4f1 AA |
109 | * reading. |
110 | */ | |
111 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
1ad9f620 | 112 | |
3e321587 AK |
113 | /* Get target node for single threaded private VMAs */ |
114 | if (prot_numa && !(vma->vm_flags & VM_SHARED) && | |
115 | atomic_read(&vma->vm_mm->mm_users) == 1) | |
116 | target_node = numa_node_id(); | |
117 | ||
3ea27719 | 118 | flush_tlb_batched_pending(vma->vm_mm); |
6606c3e0 | 119 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 120 | do { |
0697212a CL |
121 | oldpte = *pte; |
122 | if (pte_present(oldpte)) { | |
1da177e4 LT |
123 | pte_t ptent; |
124 | ||
e944fd67 MG |
125 | /* |
126 | * Avoid trapping faults against the zero or KSM | |
127 | * pages. See similar comment in change_huge_pmd. | |
128 | */ | |
129 | if (prot_numa) { | |
130 | struct page *page; | |
a1a3a2fc | 131 | int nid; |
33024536 | 132 | bool toptier; |
e944fd67 | 133 | |
a818f536 HY |
134 | /* Avoid TLB flush if possible */ |
135 | if (pte_protnone(oldpte)) | |
136 | continue; | |
137 | ||
e944fd67 | 138 | page = vm_normal_page(vma, addr, oldpte); |
3218f871 | 139 | if (!page || is_zone_device_page(page) || PageKsm(page)) |
e944fd67 | 140 | continue; |
10c1045f | 141 | |
859d4adc HW |
142 | /* Also skip shared copy-on-write pages */ |
143 | if (is_cow_mapping(vma->vm_flags) && | |
80d47f5d | 144 | page_count(page) != 1) |
859d4adc HW |
145 | continue; |
146 | ||
09a913a7 MG |
147 | /* |
148 | * While migration can move some dirty pages, | |
149 | * it cannot move them all from MIGRATE_ASYNC | |
150 | * context. | |
151 | */ | |
9de4f22a | 152 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
153 | continue; |
154 | ||
3e321587 AK |
155 | /* |
156 | * Don't mess with PTEs if page is already on the node | |
157 | * a single-threaded process is running on. | |
158 | */ | |
a1a3a2fc HY |
159 | nid = page_to_nid(page); |
160 | if (target_node == nid) | |
161 | continue; | |
33024536 | 162 | toptier = node_is_toptier(nid); |
a1a3a2fc HY |
163 | |
164 | /* | |
165 | * Skip scanning top tier node if normal numa | |
166 | * balancing is disabled | |
167 | */ | |
168 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_NORMAL) && | |
33024536 | 169 | toptier) |
3e321587 | 170 | continue; |
33024536 HY |
171 | if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING && |
172 | !toptier) | |
173 | xchg_page_access_time(page, | |
174 | jiffies_to_msecs(jiffies)); | |
e944fd67 MG |
175 | } |
176 | ||
04a86453 AK |
177 | oldpte = ptep_modify_prot_start(vma, addr, pte); |
178 | ptent = pte_modify(oldpte, newprot); | |
4b10e7d5 | 179 | |
f1eb1bac | 180 | if (uffd_wp) |
292924b2 | 181 | ptent = pte_mkuffd_wp(ptent); |
f1eb1bac | 182 | else if (uffd_wp_resolve) |
292924b2 | 183 | ptent = pte_clear_uffd_wp(ptent); |
292924b2 | 184 | |
64fe24a3 DH |
185 | /* |
186 | * In some writable, shared mappings, we might want | |
187 | * to catch actual write access -- see | |
188 | * vma_wants_writenotify(). | |
189 | * | |
190 | * In all writable, private mappings, we have to | |
191 | * properly handle COW. | |
192 | * | |
193 | * In both cases, we can sometimes still change PTEs | |
194 | * writable and avoid the write-fault handler, for | |
195 | * example, if a PTE is already dirty and no other | |
196 | * COW or special handling is required. | |
197 | */ | |
198 | if ((cp_flags & MM_CP_TRY_CHANGE_WRITABLE) && | |
199 | !pte_write(ptent) && | |
200 | can_change_pte_writable(vma, addr, ptent)) | |
8a0516ed | 201 | ptent = pte_mkwrite(ptent); |
64fe24a3 | 202 | |
04a86453 | 203 | ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); |
c9fe6656 NA |
204 | if (pte_needs_flush(oldpte, ptent)) |
205 | tlb_flush_pte_range(tlb, addr, PAGE_SIZE); | |
8a0516ed | 206 | pages++; |
f45ec5ff | 207 | } else if (is_swap_pte(oldpte)) { |
0697212a | 208 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
f45ec5ff | 209 | pte_t newpte; |
0697212a | 210 | |
4dd845b5 | 211 | if (is_writable_migration_entry(entry)) { |
3d2f78f0 PX |
212 | struct page *page = pfn_swap_entry_to_page(entry); |
213 | ||
0697212a CL |
214 | /* |
215 | * A protection check is difficult so | |
216 | * just be safe and disable write | |
217 | */ | |
6c287605 DH |
218 | if (PageAnon(page)) |
219 | entry = make_readable_exclusive_migration_entry( | |
220 | swp_offset(entry)); | |
221 | else | |
222 | entry = make_readable_migration_entry(swp_offset(entry)); | |
c3d16e16 CG |
223 | newpte = swp_entry_to_pte(entry); |
224 | if (pte_swp_soft_dirty(oldpte)) | |
225 | newpte = pte_swp_mksoft_dirty(newpte); | |
f45ec5ff PX |
226 | if (pte_swp_uffd_wp(oldpte)) |
227 | newpte = pte_swp_mkuffd_wp(newpte); | |
4dd845b5 | 228 | } else if (is_writable_device_private_entry(entry)) { |
5042db43 JG |
229 | /* |
230 | * We do not preserve soft-dirtiness. See | |
231 | * copy_one_pte() for explanation. | |
232 | */ | |
4dd845b5 AP |
233 | entry = make_readable_device_private_entry( |
234 | swp_offset(entry)); | |
5042db43 | 235 | newpte = swp_entry_to_pte(entry); |
f45ec5ff PX |
236 | if (pte_swp_uffd_wp(oldpte)) |
237 | newpte = pte_swp_mkuffd_wp(newpte); | |
b756a3b5 AP |
238 | } else if (is_writable_device_exclusive_entry(entry)) { |
239 | entry = make_readable_device_exclusive_entry( | |
240 | swp_offset(entry)); | |
241 | newpte = swp_entry_to_pte(entry); | |
242 | if (pte_swp_soft_dirty(oldpte)) | |
243 | newpte = pte_swp_mksoft_dirty(newpte); | |
244 | if (pte_swp_uffd_wp(oldpte)) | |
245 | newpte = pte_swp_mkuffd_wp(newpte); | |
7e3ce3f8 PX |
246 | } else if (is_pte_marker_entry(entry)) { |
247 | /* | |
248 | * Ignore swapin errors unconditionally, | |
249 | * because any access should sigbus anyway. | |
250 | */ | |
251 | if (is_swapin_error_entry(entry)) | |
252 | continue; | |
fe2567eb PX |
253 | /* |
254 | * If this is uffd-wp pte marker and we'd like | |
255 | * to unprotect it, drop it; the next page | |
256 | * fault will trigger without uffd trapping. | |
257 | */ | |
258 | if (uffd_wp_resolve) { | |
259 | pte_clear(vma->vm_mm, addr, pte); | |
260 | pages++; | |
261 | } | |
5c041f5d | 262 | continue; |
f45ec5ff PX |
263 | } else { |
264 | newpte = oldpte; | |
265 | } | |
5042db43 | 266 | |
f45ec5ff PX |
267 | if (uffd_wp) |
268 | newpte = pte_swp_mkuffd_wp(newpte); | |
269 | else if (uffd_wp_resolve) | |
270 | newpte = pte_swp_clear_uffd_wp(newpte); | |
271 | ||
272 | if (!pte_same(oldpte, newpte)) { | |
273 | set_pte_at(vma->vm_mm, addr, pte, newpte); | |
5042db43 JG |
274 | pages++; |
275 | } | |
fe2567eb PX |
276 | } else { |
277 | /* It must be an none page, or what else?.. */ | |
278 | WARN_ON_ONCE(!pte_none(oldpte)); | |
279 | if (unlikely(uffd_wp && !vma_is_anonymous(vma))) { | |
280 | /* | |
281 | * For file-backed mem, we need to be able to | |
282 | * wr-protect a none pte, because even if the | |
283 | * pte is none, the page/swap cache could | |
284 | * exist. Doing that by install a marker. | |
285 | */ | |
286 | set_pte_at(vma->vm_mm, addr, pte, | |
287 | make_pte_marker(PTE_MARKER_UFFD_WP)); | |
288 | pages++; | |
289 | } | |
1da177e4 LT |
290 | } |
291 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 292 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 293 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
294 | |
295 | return pages; | |
1da177e4 LT |
296 | } |
297 | ||
8b272b3c MG |
298 | /* |
299 | * Used when setting automatic NUMA hinting protection where it is | |
300 | * critical that a numa hinting PMD is not confused with a bad PMD. | |
301 | */ | |
302 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) | |
303 | { | |
dab6e717 | 304 | pmd_t pmdval = pmdp_get_lockless(pmd); |
8b272b3c MG |
305 | |
306 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ | |
307 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
308 | barrier(); | |
309 | #endif | |
310 | ||
311 | if (pmd_none(pmdval)) | |
312 | return 1; | |
313 | if (pmd_trans_huge(pmdval)) | |
314 | return 0; | |
315 | if (unlikely(pmd_bad(pmdval))) { | |
316 | pmd_clear_bad(pmd); | |
317 | return 1; | |
318 | } | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
fe2567eb PX |
323 | /* Return true if we're uffd wr-protecting file-backed memory, or false */ |
324 | static inline bool | |
325 | uffd_wp_protect_file(struct vm_area_struct *vma, unsigned long cp_flags) | |
326 | { | |
327 | return (cp_flags & MM_CP_UFFD_WP) && !vma_is_anonymous(vma); | |
328 | } | |
329 | ||
330 | /* | |
331 | * If wr-protecting the range for file-backed, populate pgtable for the case | |
332 | * when pgtable is empty but page cache exists. When {pte|pmd|...}_alloc() | |
d1751118 PX |
333 | * failed we treat it the same way as pgtable allocation failures during |
334 | * page faults by kicking OOM and returning error. | |
fe2567eb PX |
335 | */ |
336 | #define change_pmd_prepare(vma, pmd, cp_flags) \ | |
d1751118 PX |
337 | ({ \ |
338 | long err = 0; \ | |
fe2567eb | 339 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ |
d1751118 PX |
340 | if (pte_alloc(vma->vm_mm, pmd)) \ |
341 | err = -ENOMEM; \ | |
fe2567eb | 342 | } \ |
d1751118 PX |
343 | err; \ |
344 | }) | |
345 | ||
fe2567eb PX |
346 | /* |
347 | * This is the general pud/p4d/pgd version of change_pmd_prepare(). We need to | |
348 | * have separate change_pmd_prepare() because pte_alloc() returns 0 on success, | |
349 | * while {pmd|pud|p4d}_alloc() returns the valid pointer on success. | |
350 | */ | |
351 | #define change_prepare(vma, high, low, addr, cp_flags) \ | |
d1751118 PX |
352 | ({ \ |
353 | long err = 0; \ | |
fe2567eb PX |
354 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ |
355 | low##_t *p = low##_alloc(vma->vm_mm, high, addr); \ | |
d1751118 PX |
356 | if (p == NULL) \ |
357 | err = -ENOMEM; \ | |
fe2567eb | 358 | } \ |
d1751118 PX |
359 | err; \ |
360 | }) | |
fe2567eb | 361 | |
a79390f5 | 362 | static inline long change_pmd_range(struct mmu_gather *tlb, |
4a18419f NA |
363 | struct vm_area_struct *vma, pud_t *pud, unsigned long addr, |
364 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
365 | { |
366 | pmd_t *pmd; | |
367 | unsigned long next; | |
a79390f5 | 368 | long pages = 0; |
72403b4a | 369 | unsigned long nr_huge_updates = 0; |
ac46d4f3 JG |
370 | struct mmu_notifier_range range; |
371 | ||
372 | range.start = 0; | |
1da177e4 LT |
373 | |
374 | pmd = pmd_offset(pud, addr); | |
375 | do { | |
d1751118 | 376 | long ret; |
25cbbef1 | 377 | |
1da177e4 | 378 | next = pmd_addr_end(addr, end); |
8b272b3c | 379 | |
d1751118 PX |
380 | ret = change_pmd_prepare(vma, pmd, cp_flags); |
381 | if (ret) { | |
382 | pages = ret; | |
383 | break; | |
384 | } | |
8b272b3c | 385 | /* |
c1e8d7c6 | 386 | * Automatic NUMA balancing walks the tables with mmap_lock |
8b272b3c MG |
387 | * held for read. It's possible a parallel update to occur |
388 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() | |
389 | * check leading to a false positive and clearing. | |
390 | * Hence, it's necessary to atomically read the PMD value | |
391 | * for all the checks. | |
392 | */ | |
393 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && | |
394 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) | |
4991c09c | 395 | goto next; |
a5338093 RR |
396 | |
397 | /* invoke the mmu notifier if the pmd is populated */ | |
ac46d4f3 | 398 | if (!range.start) { |
7269f999 JG |
399 | mmu_notifier_range_init(&range, |
400 | MMU_NOTIFY_PROTECTION_VMA, 0, | |
7d4a8be0 | 401 | vma->vm_mm, addr, end); |
ac46d4f3 | 402 | mmu_notifier_invalidate_range_start(&range); |
a5338093 RR |
403 | } |
404 | ||
84c3fc4e | 405 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
019c2d8b PX |
406 | if ((next - addr != HPAGE_PMD_SIZE) || |
407 | uffd_wp_protect_file(vma, cp_flags)) { | |
fd60775a | 408 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
019c2d8b PX |
409 | /* |
410 | * For file-backed, the pmd could have been | |
411 | * cleared; make sure pmd populated if | |
412 | * necessary, then fall-through to pte level. | |
413 | */ | |
d1751118 PX |
414 | ret = change_pmd_prepare(vma, pmd, cp_flags); |
415 | if (ret) { | |
416 | pages = ret; | |
417 | break; | |
418 | } | |
6b9116a6 | 419 | } else { |
4a18419f NA |
420 | /* |
421 | * change_huge_pmd() does not defer TLB flushes, | |
422 | * so no need to propagate the tlb argument. | |
423 | */ | |
424 | int nr_ptes = change_huge_pmd(tlb, vma, pmd, | |
425 | addr, newprot, cp_flags); | |
f123d74a MG |
426 | |
427 | if (nr_ptes) { | |
72403b4a MG |
428 | if (nr_ptes == HPAGE_PMD_NR) { |
429 | pages += HPAGE_PMD_NR; | |
430 | nr_huge_updates++; | |
431 | } | |
1ad9f620 MG |
432 | |
433 | /* huge pmd was handled */ | |
4991c09c | 434 | goto next; |
f123d74a | 435 | } |
7da4d641 | 436 | } |
88a9ab6e | 437 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 438 | } |
d1751118 PX |
439 | pages += change_pte_range(tlb, vma, pmd, addr, next, |
440 | newprot, cp_flags); | |
4991c09c AK |
441 | next: |
442 | cond_resched(); | |
1da177e4 | 443 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 444 | |
ac46d4f3 JG |
445 | if (range.start) |
446 | mmu_notifier_invalidate_range_end(&range); | |
a5338093 | 447 | |
72403b4a MG |
448 | if (nr_huge_updates) |
449 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 450 | return pages; |
1da177e4 LT |
451 | } |
452 | ||
a79390f5 | 453 | static inline long change_pud_range(struct mmu_gather *tlb, |
4a18419f NA |
454 | struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr, |
455 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
456 | { |
457 | pud_t *pud; | |
458 | unsigned long next; | |
d1751118 | 459 | long pages = 0, ret; |
1da177e4 | 460 | |
c2febafc | 461 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
462 | do { |
463 | next = pud_addr_end(addr, end); | |
d1751118 PX |
464 | ret = change_prepare(vma, pud, pmd, addr, cp_flags); |
465 | if (ret) | |
466 | return ret; | |
1da177e4 LT |
467 | if (pud_none_or_clear_bad(pud)) |
468 | continue; | |
4a18419f | 469 | pages += change_pmd_range(tlb, vma, pud, addr, next, newprot, |
58705444 | 470 | cp_flags); |
1da177e4 | 471 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
472 | |
473 | return pages; | |
1da177e4 LT |
474 | } |
475 | ||
a79390f5 | 476 | static inline long change_p4d_range(struct mmu_gather *tlb, |
4a18419f NA |
477 | struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr, |
478 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
c2febafc KS |
479 | { |
480 | p4d_t *p4d; | |
481 | unsigned long next; | |
d1751118 | 482 | long pages = 0, ret; |
c2febafc KS |
483 | |
484 | p4d = p4d_offset(pgd, addr); | |
485 | do { | |
486 | next = p4d_addr_end(addr, end); | |
d1751118 PX |
487 | ret = change_prepare(vma, p4d, pud, addr, cp_flags); |
488 | if (ret) | |
489 | return ret; | |
c2febafc KS |
490 | if (p4d_none_or_clear_bad(p4d)) |
491 | continue; | |
4a18419f | 492 | pages += change_pud_range(tlb, vma, p4d, addr, next, newprot, |
58705444 | 493 | cp_flags); |
c2febafc KS |
494 | } while (p4d++, addr = next, addr != end); |
495 | ||
496 | return pages; | |
497 | } | |
498 | ||
a79390f5 | 499 | static long change_protection_range(struct mmu_gather *tlb, |
4a18419f NA |
500 | struct vm_area_struct *vma, unsigned long addr, |
501 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
502 | { |
503 | struct mm_struct *mm = vma->vm_mm; | |
504 | pgd_t *pgd; | |
505 | unsigned long next; | |
d1751118 | 506 | long pages = 0, ret; |
1da177e4 LT |
507 | |
508 | BUG_ON(addr >= end); | |
509 | pgd = pgd_offset(mm, addr); | |
4a18419f | 510 | tlb_start_vma(tlb, vma); |
1da177e4 LT |
511 | do { |
512 | next = pgd_addr_end(addr, end); | |
d1751118 PX |
513 | ret = change_prepare(vma, pgd, p4d, addr, cp_flags); |
514 | if (ret) { | |
515 | pages = ret; | |
516 | break; | |
517 | } | |
1da177e4 LT |
518 | if (pgd_none_or_clear_bad(pgd)) |
519 | continue; | |
4a18419f | 520 | pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot, |
58705444 | 521 | cp_flags); |
1da177e4 | 522 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 523 | |
4a18419f | 524 | tlb_end_vma(tlb, vma); |
7da4d641 PZ |
525 | |
526 | return pages; | |
527 | } | |
528 | ||
a79390f5 | 529 | long change_protection(struct mmu_gather *tlb, |
4a18419f | 530 | struct vm_area_struct *vma, unsigned long start, |
1ef488ed | 531 | unsigned long end, unsigned long cp_flags) |
7da4d641 | 532 | { |
1ef488ed | 533 | pgprot_t newprot = vma->vm_page_prot; |
a79390f5 | 534 | long pages; |
7da4d641 | 535 | |
292924b2 PX |
536 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
537 | ||
1ef488ed DH |
538 | #ifdef CONFIG_NUMA_BALANCING |
539 | /* | |
540 | * Ordinary protection updates (mprotect, uffd-wp, softdirty tracking) | |
541 | * are expected to reflect their requirements via VMA flags such that | |
542 | * vma_set_page_prot() will adjust vma->vm_page_prot accordingly. | |
543 | */ | |
544 | if (cp_flags & MM_CP_PROT_NUMA) | |
545 | newprot = PAGE_NONE; | |
546 | #else | |
547 | WARN_ON_ONCE(cp_flags & MM_CP_PROT_NUMA); | |
548 | #endif | |
549 | ||
7da4d641 | 550 | if (is_vm_hugetlb_page(vma)) |
5a90d5a1 PX |
551 | pages = hugetlb_change_protection(vma, start, end, newprot, |
552 | cp_flags); | |
7da4d641 | 553 | else |
4a18419f | 554 | pages = change_protection_range(tlb, vma, start, end, newprot, |
58705444 | 555 | cp_flags); |
7da4d641 PZ |
556 | |
557 | return pages; | |
1da177e4 LT |
558 | } |
559 | ||
42e4089c AK |
560 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
561 | unsigned long next, struct mm_walk *walk) | |
562 | { | |
563 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
564 | 0 : -EACCES; | |
565 | } | |
566 | ||
567 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, | |
568 | unsigned long addr, unsigned long next, | |
569 | struct mm_walk *walk) | |
570 | { | |
571 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
572 | 0 : -EACCES; | |
573 | } | |
574 | ||
575 | static int prot_none_test(unsigned long addr, unsigned long next, | |
576 | struct mm_walk *walk) | |
577 | { | |
578 | return 0; | |
579 | } | |
580 | ||
7b86ac33 CH |
581 | static const struct mm_walk_ops prot_none_walk_ops = { |
582 | .pte_entry = prot_none_pte_entry, | |
583 | .hugetlb_entry = prot_none_hugetlb_entry, | |
584 | .test_walk = prot_none_test, | |
585 | }; | |
42e4089c | 586 | |
b6a2fea3 | 587 | int |
2286a691 LH |
588 | mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb, |
589 | struct vm_area_struct *vma, struct vm_area_struct **pprev, | |
590 | unsigned long start, unsigned long end, unsigned long newflags) | |
1da177e4 LT |
591 | { |
592 | struct mm_struct *mm = vma->vm_mm; | |
593 | unsigned long oldflags = vma->vm_flags; | |
594 | long nrpages = (end - start) >> PAGE_SHIFT; | |
eb309ec8 | 595 | unsigned int mm_cp_flags = 0; |
1da177e4 | 596 | unsigned long charged = 0; |
1da177e4 LT |
597 | pgoff_t pgoff; |
598 | int error; | |
599 | ||
600 | if (newflags == oldflags) { | |
601 | *pprev = vma; | |
602 | return 0; | |
603 | } | |
604 | ||
42e4089c AK |
605 | /* |
606 | * Do PROT_NONE PFN permission checks here when we can still | |
607 | * bail out without undoing a lot of state. This is a rather | |
608 | * uncommon case, so doesn't need to be very optimized. | |
609 | */ | |
610 | if (arch_has_pfn_modify_check() && | |
611 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && | |
6cb4d9a2 | 612 | (newflags & VM_ACCESS_FLAGS) == 0) { |
7b86ac33 CH |
613 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
614 | ||
615 | error = walk_page_range(current->mm, start, end, | |
616 | &prot_none_walk_ops, &new_pgprot); | |
42e4089c AK |
617 | if (error) |
618 | return error; | |
619 | } | |
620 | ||
1da177e4 LT |
621 | /* |
622 | * If we make a private mapping writable we increase our commit; | |
623 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
624 | * make it unwritable again. hugetlb mapping were accounted for |
625 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
626 | */ |
627 | if (newflags & VM_WRITE) { | |
84638335 KK |
628 | /* Check space limits when area turns into data. */ |
629 | if (!may_expand_vm(mm, newflags, nrpages) && | |
630 | may_expand_vm(mm, oldflags, nrpages)) | |
631 | return -ENOMEM; | |
5a6fe125 | 632 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 633 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 634 | charged = nrpages; |
191c5424 | 635 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
636 | return -ENOMEM; |
637 | newflags |= VM_ACCOUNT; | |
638 | } | |
639 | } | |
640 | ||
1da177e4 LT |
641 | /* |
642 | * First try to merge with previous and/or next vma. | |
643 | */ | |
644 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
9760ebff | 645 | *pprev = vma_merge(vmi, mm, *pprev, start, end, newflags, |
19a809af | 646 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 647 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
648 | if (*pprev) { |
649 | vma = *pprev; | |
e86f15ee | 650 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
1da177e4 LT |
651 | goto success; |
652 | } | |
653 | ||
654 | *pprev = vma; | |
655 | ||
656 | if (start != vma->vm_start) { | |
9760ebff | 657 | error = split_vma(vmi, vma, start, 1); |
1da177e4 LT |
658 | if (error) |
659 | goto fail; | |
660 | } | |
661 | ||
662 | if (end != vma->vm_end) { | |
9760ebff | 663 | error = split_vma(vmi, vma, end, 0); |
1da177e4 LT |
664 | if (error) |
665 | goto fail; | |
666 | } | |
667 | ||
668 | success: | |
669 | /* | |
c1e8d7c6 | 670 | * vm_flags and vm_page_prot are protected by the mmap_lock |
1da177e4 LT |
671 | * held in write mode. |
672 | */ | |
1c71222e | 673 | vm_flags_reset(vma, newflags); |
eb309ec8 DH |
674 | if (vma_wants_manual_pte_write_upgrade(vma)) |
675 | mm_cp_flags |= MM_CP_TRY_CHANGE_WRITABLE; | |
64e45507 | 676 | vma_set_page_prot(vma); |
d08b3851 | 677 | |
1ef488ed | 678 | change_protection(tlb, vma, start, end, mm_cp_flags); |
7da4d641 | 679 | |
36f88188 KS |
680 | /* |
681 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major | |
682 | * fault on access. | |
683 | */ | |
684 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && | |
685 | (newflags & VM_WRITE)) { | |
686 | populate_vma_page_range(vma, start, end, NULL); | |
687 | } | |
688 | ||
84638335 KK |
689 | vm_stat_account(mm, oldflags, -nrpages); |
690 | vm_stat_account(mm, newflags, nrpages); | |
63bfd738 | 691 | perf_event_mmap(vma); |
1da177e4 LT |
692 | return 0; |
693 | ||
694 | fail: | |
695 | vm_unacct_memory(charged); | |
696 | return error; | |
697 | } | |
698 | ||
7d06d9c9 DH |
699 | /* |
700 | * pkey==-1 when doing a legacy mprotect() | |
701 | */ | |
702 | static int do_mprotect_pkey(unsigned long start, size_t len, | |
703 | unsigned long prot, int pkey) | |
1da177e4 | 704 | { |
62b5f7d0 | 705 | unsigned long nstart, end, tmp, reqprot; |
1da177e4 | 706 | struct vm_area_struct *vma, *prev; |
48725bbc | 707 | int error; |
1da177e4 | 708 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); |
f138556d PK |
709 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
710 | (prot & PROT_READ); | |
4a18419f | 711 | struct mmu_gather tlb; |
2286a691 | 712 | struct vma_iterator vmi; |
f138556d | 713 | |
057d3389 AK |
714 | start = untagged_addr(start); |
715 | ||
1da177e4 LT |
716 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
717 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
718 | return -EINVAL; | |
719 | ||
720 | if (start & ~PAGE_MASK) | |
721 | return -EINVAL; | |
722 | if (!len) | |
723 | return 0; | |
724 | len = PAGE_ALIGN(len); | |
725 | end = start + len; | |
726 | if (end <= start) | |
727 | return -ENOMEM; | |
9035cf9a | 728 | if (!arch_validate_prot(prot, start)) |
1da177e4 LT |
729 | return -EINVAL; |
730 | ||
731 | reqprot = prot; | |
1da177e4 | 732 | |
d8ed45c5 | 733 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 734 | return -EINTR; |
1da177e4 | 735 | |
e8c24d3a DH |
736 | /* |
737 | * If userspace did not allocate the pkey, do not let | |
738 | * them use it here. | |
739 | */ | |
740 | error = -EINVAL; | |
741 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) | |
742 | goto out; | |
743 | ||
2286a691 LH |
744 | vma_iter_init(&vmi, current->mm, start); |
745 | vma = vma_find(&vmi, end); | |
1da177e4 LT |
746 | error = -ENOMEM; |
747 | if (!vma) | |
748 | goto out; | |
6af5fa0d | 749 | |
1da177e4 LT |
750 | if (unlikely(grows & PROT_GROWSDOWN)) { |
751 | if (vma->vm_start >= end) | |
752 | goto out; | |
753 | start = vma->vm_start; | |
754 | error = -EINVAL; | |
755 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
756 | goto out; | |
7d12efae | 757 | } else { |
1da177e4 LT |
758 | if (vma->vm_start > start) |
759 | goto out; | |
760 | if (unlikely(grows & PROT_GROWSUP)) { | |
761 | end = vma->vm_end; | |
762 | error = -EINVAL; | |
763 | if (!(vma->vm_flags & VM_GROWSUP)) | |
764 | goto out; | |
765 | } | |
766 | } | |
6af5fa0d | 767 | |
2286a691 | 768 | prev = vma_prev(&vmi); |
1da177e4 LT |
769 | if (start > vma->vm_start) |
770 | prev = vma; | |
771 | ||
4a18419f | 772 | tlb_gather_mmu(&tlb, current->mm); |
2286a691 LH |
773 | nstart = start; |
774 | tmp = vma->vm_start; | |
775 | for_each_vma_range(vmi, vma, end) { | |
a8502b67 | 776 | unsigned long mask_off_old_flags; |
1da177e4 | 777 | unsigned long newflags; |
7d06d9c9 | 778 | int new_vma_pkey; |
1da177e4 | 779 | |
2286a691 LH |
780 | if (vma->vm_start != tmp) { |
781 | error = -ENOMEM; | |
782 | break; | |
783 | } | |
1da177e4 | 784 | |
f138556d PK |
785 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
786 | if (rier && (vma->vm_flags & VM_MAYEXEC)) | |
787 | prot |= PROT_EXEC; | |
788 | ||
a8502b67 DH |
789 | /* |
790 | * Each mprotect() call explicitly passes r/w/x permissions. | |
791 | * If a permission is not passed to mprotect(), it must be | |
792 | * cleared from the VMA. | |
793 | */ | |
e39ee675 | 794 | mask_off_old_flags = VM_ACCESS_FLAGS | VM_FLAGS_CLEAR; |
a8502b67 | 795 | |
7d06d9c9 DH |
796 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
797 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); | |
a8502b67 | 798 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
1da177e4 | 799 | |
7e2cff42 | 800 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
6cb4d9a2 | 801 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
1da177e4 | 802 | error = -EACCES; |
4a18419f | 803 | break; |
1da177e4 LT |
804 | } |
805 | ||
b507808e JG |
806 | if (map_deny_write_exec(vma, newflags)) { |
807 | error = -EACCES; | |
808 | goto out; | |
809 | } | |
810 | ||
c462ac28 CM |
811 | /* Allow architectures to sanity-check the new flags */ |
812 | if (!arch_validate_flags(newflags)) { | |
813 | error = -EINVAL; | |
4a18419f | 814 | break; |
c462ac28 CM |
815 | } |
816 | ||
1da177e4 LT |
817 | error = security_file_mprotect(vma, reqprot, prot); |
818 | if (error) | |
4a18419f | 819 | break; |
1da177e4 LT |
820 | |
821 | tmp = vma->vm_end; | |
822 | if (tmp > end) | |
823 | tmp = end; | |
95bb7c42 | 824 | |
dbf53f75 | 825 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
95bb7c42 | 826 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
dbf53f75 | 827 | if (error) |
4a18419f | 828 | break; |
dbf53f75 | 829 | } |
95bb7c42 | 830 | |
2286a691 | 831 | error = mprotect_fixup(&vmi, &tlb, vma, &prev, nstart, tmp, newflags); |
1da177e4 | 832 | if (error) |
4a18419f | 833 | break; |
95bb7c42 | 834 | |
2fcd07b7 | 835 | tmp = vma_iter_end(&vmi); |
1da177e4 | 836 | nstart = tmp; |
f138556d | 837 | prot = reqprot; |
1da177e4 | 838 | } |
4a18419f | 839 | tlb_finish_mmu(&tlb); |
2286a691 LH |
840 | |
841 | if (vma_iter_end(&vmi) < end) | |
842 | error = -ENOMEM; | |
843 | ||
1da177e4 | 844 | out: |
d8ed45c5 | 845 | mmap_write_unlock(current->mm); |
1da177e4 LT |
846 | return error; |
847 | } | |
7d06d9c9 DH |
848 | |
849 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, | |
850 | unsigned long, prot) | |
851 | { | |
852 | return do_mprotect_pkey(start, len, prot, -1); | |
853 | } | |
854 | ||
c7142aea HC |
855 | #ifdef CONFIG_ARCH_HAS_PKEYS |
856 | ||
7d06d9c9 DH |
857 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
858 | unsigned long, prot, int, pkey) | |
859 | { | |
860 | return do_mprotect_pkey(start, len, prot, pkey); | |
861 | } | |
e8c24d3a DH |
862 | |
863 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) | |
864 | { | |
865 | int pkey; | |
866 | int ret; | |
867 | ||
868 | /* No flags supported yet. */ | |
869 | if (flags) | |
870 | return -EINVAL; | |
871 | /* check for unsupported init values */ | |
872 | if (init_val & ~PKEY_ACCESS_MASK) | |
873 | return -EINVAL; | |
874 | ||
d8ed45c5 | 875 | mmap_write_lock(current->mm); |
e8c24d3a DH |
876 | pkey = mm_pkey_alloc(current->mm); |
877 | ||
878 | ret = -ENOSPC; | |
879 | if (pkey == -1) | |
880 | goto out; | |
881 | ||
882 | ret = arch_set_user_pkey_access(current, pkey, init_val); | |
883 | if (ret) { | |
884 | mm_pkey_free(current->mm, pkey); | |
885 | goto out; | |
886 | } | |
887 | ret = pkey; | |
888 | out: | |
d8ed45c5 | 889 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
890 | return ret; |
891 | } | |
892 | ||
893 | SYSCALL_DEFINE1(pkey_free, int, pkey) | |
894 | { | |
895 | int ret; | |
896 | ||
d8ed45c5 | 897 | mmap_write_lock(current->mm); |
e8c24d3a | 898 | ret = mm_pkey_free(current->mm, pkey); |
d8ed45c5 | 899 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
900 | |
901 | /* | |
f0953a1b | 902 | * We could provide warnings or errors if any VMA still |
e8c24d3a DH |
903 | * has the pkey set here. |
904 | */ | |
905 | return ret; | |
906 | } | |
c7142aea HC |
907 | |
908 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |