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 | ||
64fe24a3 DH |
42 | static inline bool can_change_pte_writable(struct vm_area_struct *vma, |
43 | unsigned long addr, pte_t pte) | |
44 | { | |
45 | struct page *page; | |
46 | ||
47 | VM_BUG_ON(!(vma->vm_flags & VM_WRITE) || pte_write(pte)); | |
48 | ||
d8488773 | 49 | if (pte_protnone(pte)) |
64fe24a3 DH |
50 | return false; |
51 | ||
52 | /* Do we need write faults for softdirty tracking? */ | |
76aefad6 | 53 | if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte)) |
64fe24a3 DH |
54 | return false; |
55 | ||
56 | /* Do we need write faults for uffd-wp tracking? */ | |
57 | if (userfaultfd_pte_wp(vma, pte)) | |
58 | return false; | |
59 | ||
60 | if (!(vma->vm_flags & VM_SHARED)) { | |
61 | /* | |
62 | * We can only special-case on exclusive anonymous pages, | |
63 | * because we know that our write-fault handler similarly would | |
64 | * map them writable without any additional checks while holding | |
65 | * the PT lock. | |
66 | */ | |
67 | page = vm_normal_page(vma, addr, pte); | |
d8488773 | 68 | return page && PageAnon(page) && PageAnonExclusive(page); |
64fe24a3 DH |
69 | } |
70 | ||
d8488773 | 71 | return pte_dirty(pte); |
64fe24a3 DH |
72 | } |
73 | ||
4a18419f NA |
74 | static unsigned long change_pte_range(struct mmu_gather *tlb, |
75 | struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr, | |
76 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 | 77 | { |
0697212a | 78 | pte_t *pte, oldpte; |
705e87c0 | 79 | spinlock_t *ptl; |
7da4d641 | 80 | unsigned long pages = 0; |
3e321587 | 81 | int target_node = NUMA_NO_NODE; |
58705444 | 82 | bool prot_numa = cp_flags & MM_CP_PROT_NUMA; |
292924b2 PX |
83 | bool uffd_wp = cp_flags & MM_CP_UFFD_WP; |
84 | bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE; | |
1da177e4 | 85 | |
4a18419f NA |
86 | tlb_change_page_size(tlb, PAGE_SIZE); |
87 | ||
175ad4f1 | 88 | /* |
c1e8d7c6 | 89 | * Can be called with only the mmap_lock for reading by |
175ad4f1 AA |
90 | * prot_numa so we must check the pmd isn't constantly |
91 | * changing from under us from pmd_none to pmd_trans_huge | |
92 | * and/or the other way around. | |
93 | */ | |
94 | if (pmd_trans_unstable(pmd)) | |
95 | return 0; | |
96 | ||
97 | /* | |
98 | * The pmd points to a regular pte so the pmd can't change | |
c1e8d7c6 | 99 | * from under us even if the mmap_lock is only hold for |
175ad4f1 AA |
100 | * reading. |
101 | */ | |
102 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
1ad9f620 | 103 | |
3e321587 AK |
104 | /* Get target node for single threaded private VMAs */ |
105 | if (prot_numa && !(vma->vm_flags & VM_SHARED) && | |
106 | atomic_read(&vma->vm_mm->mm_users) == 1) | |
107 | target_node = numa_node_id(); | |
108 | ||
3ea27719 | 109 | flush_tlb_batched_pending(vma->vm_mm); |
6606c3e0 | 110 | arch_enter_lazy_mmu_mode(); |
1da177e4 | 111 | do { |
0697212a CL |
112 | oldpte = *pte; |
113 | if (pte_present(oldpte)) { | |
1da177e4 | 114 | pte_t ptent; |
b191f9b1 | 115 | bool preserve_write = prot_numa && pte_write(oldpte); |
1da177e4 | 116 | |
e944fd67 MG |
117 | /* |
118 | * Avoid trapping faults against the zero or KSM | |
119 | * pages. See similar comment in change_huge_pmd. | |
120 | */ | |
121 | if (prot_numa) { | |
122 | struct page *page; | |
a1a3a2fc | 123 | int nid; |
33024536 | 124 | bool toptier; |
e944fd67 | 125 | |
a818f536 HY |
126 | /* Avoid TLB flush if possible */ |
127 | if (pte_protnone(oldpte)) | |
128 | continue; | |
129 | ||
e944fd67 | 130 | page = vm_normal_page(vma, addr, oldpte); |
3218f871 | 131 | if (!page || is_zone_device_page(page) || PageKsm(page)) |
e944fd67 | 132 | continue; |
10c1045f | 133 | |
859d4adc HW |
134 | /* Also skip shared copy-on-write pages */ |
135 | if (is_cow_mapping(vma->vm_flags) && | |
80d47f5d | 136 | page_count(page) != 1) |
859d4adc HW |
137 | continue; |
138 | ||
09a913a7 MG |
139 | /* |
140 | * While migration can move some dirty pages, | |
141 | * it cannot move them all from MIGRATE_ASYNC | |
142 | * context. | |
143 | */ | |
9de4f22a | 144 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
145 | continue; |
146 | ||
3e321587 AK |
147 | /* |
148 | * Don't mess with PTEs if page is already on the node | |
149 | * a single-threaded process is running on. | |
150 | */ | |
a1a3a2fc HY |
151 | nid = page_to_nid(page); |
152 | if (target_node == nid) | |
153 | continue; | |
33024536 | 154 | toptier = node_is_toptier(nid); |
a1a3a2fc HY |
155 | |
156 | /* | |
157 | * Skip scanning top tier node if normal numa | |
158 | * balancing is disabled | |
159 | */ | |
160 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_NORMAL) && | |
33024536 | 161 | toptier) |
3e321587 | 162 | continue; |
33024536 HY |
163 | if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING && |
164 | !toptier) | |
165 | xchg_page_access_time(page, | |
166 | jiffies_to_msecs(jiffies)); | |
e944fd67 MG |
167 | } |
168 | ||
04a86453 AK |
169 | oldpte = ptep_modify_prot_start(vma, addr, pte); |
170 | ptent = pte_modify(oldpte, newprot); | |
b191f9b1 | 171 | if (preserve_write) |
288bc549 | 172 | ptent = pte_mk_savedwrite(ptent); |
4b10e7d5 | 173 | |
292924b2 PX |
174 | if (uffd_wp) { |
175 | ptent = pte_wrprotect(ptent); | |
176 | ptent = pte_mkuffd_wp(ptent); | |
177 | } else if (uffd_wp_resolve) { | |
292924b2 PX |
178 | ptent = pte_clear_uffd_wp(ptent); |
179 | } | |
180 | ||
64fe24a3 DH |
181 | /* |
182 | * In some writable, shared mappings, we might want | |
183 | * to catch actual write access -- see | |
184 | * vma_wants_writenotify(). | |
185 | * | |
186 | * In all writable, private mappings, we have to | |
187 | * properly handle COW. | |
188 | * | |
189 | * In both cases, we can sometimes still change PTEs | |
190 | * writable and avoid the write-fault handler, for | |
191 | * example, if a PTE is already dirty and no other | |
192 | * COW or special handling is required. | |
193 | */ | |
194 | if ((cp_flags & MM_CP_TRY_CHANGE_WRITABLE) && | |
195 | !pte_write(ptent) && | |
196 | can_change_pte_writable(vma, addr, ptent)) | |
8a0516ed | 197 | ptent = pte_mkwrite(ptent); |
64fe24a3 | 198 | |
04a86453 | 199 | ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); |
c9fe6656 NA |
200 | if (pte_needs_flush(oldpte, ptent)) |
201 | tlb_flush_pte_range(tlb, addr, PAGE_SIZE); | |
8a0516ed | 202 | pages++; |
f45ec5ff | 203 | } else if (is_swap_pte(oldpte)) { |
0697212a | 204 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
f45ec5ff | 205 | pte_t newpte; |
0697212a | 206 | |
4dd845b5 | 207 | if (is_writable_migration_entry(entry)) { |
3d2f78f0 PX |
208 | struct page *page = pfn_swap_entry_to_page(entry); |
209 | ||
0697212a CL |
210 | /* |
211 | * A protection check is difficult so | |
212 | * just be safe and disable write | |
213 | */ | |
6c287605 DH |
214 | if (PageAnon(page)) |
215 | entry = make_readable_exclusive_migration_entry( | |
216 | swp_offset(entry)); | |
217 | else | |
218 | entry = make_readable_migration_entry(swp_offset(entry)); | |
c3d16e16 CG |
219 | newpte = swp_entry_to_pte(entry); |
220 | if (pte_swp_soft_dirty(oldpte)) | |
221 | newpte = pte_swp_mksoft_dirty(newpte); | |
f45ec5ff PX |
222 | if (pte_swp_uffd_wp(oldpte)) |
223 | newpte = pte_swp_mkuffd_wp(newpte); | |
4dd845b5 | 224 | } else if (is_writable_device_private_entry(entry)) { |
5042db43 JG |
225 | /* |
226 | * We do not preserve soft-dirtiness. See | |
227 | * copy_one_pte() for explanation. | |
228 | */ | |
4dd845b5 AP |
229 | entry = make_readable_device_private_entry( |
230 | swp_offset(entry)); | |
5042db43 | 231 | newpte = swp_entry_to_pte(entry); |
f45ec5ff PX |
232 | if (pte_swp_uffd_wp(oldpte)) |
233 | newpte = pte_swp_mkuffd_wp(newpte); | |
b756a3b5 AP |
234 | } else if (is_writable_device_exclusive_entry(entry)) { |
235 | entry = make_readable_device_exclusive_entry( | |
236 | swp_offset(entry)); | |
237 | newpte = swp_entry_to_pte(entry); | |
238 | if (pte_swp_soft_dirty(oldpte)) | |
239 | newpte = pte_swp_mksoft_dirty(newpte); | |
240 | if (pte_swp_uffd_wp(oldpte)) | |
241 | newpte = pte_swp_mkuffd_wp(newpte); | |
fe2567eb PX |
242 | } else if (pte_marker_entry_uffd_wp(entry)) { |
243 | /* | |
244 | * If this is uffd-wp pte marker and we'd like | |
245 | * to unprotect it, drop it; the next page | |
246 | * fault will trigger without uffd trapping. | |
247 | */ | |
248 | if (uffd_wp_resolve) { | |
249 | pte_clear(vma->vm_mm, addr, pte); | |
250 | pages++; | |
251 | } | |
5c041f5d | 252 | continue; |
f45ec5ff PX |
253 | } else { |
254 | newpte = oldpte; | |
255 | } | |
5042db43 | 256 | |
f45ec5ff PX |
257 | if (uffd_wp) |
258 | newpte = pte_swp_mkuffd_wp(newpte); | |
259 | else if (uffd_wp_resolve) | |
260 | newpte = pte_swp_clear_uffd_wp(newpte); | |
261 | ||
262 | if (!pte_same(oldpte, newpte)) { | |
263 | set_pte_at(vma->vm_mm, addr, pte, newpte); | |
5042db43 JG |
264 | pages++; |
265 | } | |
fe2567eb PX |
266 | } else { |
267 | /* It must be an none page, or what else?.. */ | |
268 | WARN_ON_ONCE(!pte_none(oldpte)); | |
269 | if (unlikely(uffd_wp && !vma_is_anonymous(vma))) { | |
270 | /* | |
271 | * For file-backed mem, we need to be able to | |
272 | * wr-protect a none pte, because even if the | |
273 | * pte is none, the page/swap cache could | |
274 | * exist. Doing that by install a marker. | |
275 | */ | |
276 | set_pte_at(vma->vm_mm, addr, pte, | |
277 | make_pte_marker(PTE_MARKER_UFFD_WP)); | |
278 | pages++; | |
279 | } | |
1da177e4 LT |
280 | } |
281 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 282 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 283 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
284 | |
285 | return pages; | |
1da177e4 LT |
286 | } |
287 | ||
8b272b3c MG |
288 | /* |
289 | * Used when setting automatic NUMA hinting protection where it is | |
290 | * critical that a numa hinting PMD is not confused with a bad PMD. | |
291 | */ | |
292 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) | |
293 | { | |
294 | pmd_t pmdval = pmd_read_atomic(pmd); | |
295 | ||
296 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ | |
297 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
298 | barrier(); | |
299 | #endif | |
300 | ||
301 | if (pmd_none(pmdval)) | |
302 | return 1; | |
303 | if (pmd_trans_huge(pmdval)) | |
304 | return 0; | |
305 | if (unlikely(pmd_bad(pmdval))) { | |
306 | pmd_clear_bad(pmd); | |
307 | return 1; | |
308 | } | |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
fe2567eb PX |
313 | /* Return true if we're uffd wr-protecting file-backed memory, or false */ |
314 | static inline bool | |
315 | uffd_wp_protect_file(struct vm_area_struct *vma, unsigned long cp_flags) | |
316 | { | |
317 | return (cp_flags & MM_CP_UFFD_WP) && !vma_is_anonymous(vma); | |
318 | } | |
319 | ||
320 | /* | |
321 | * If wr-protecting the range for file-backed, populate pgtable for the case | |
322 | * when pgtable is empty but page cache exists. When {pte|pmd|...}_alloc() | |
323 | * failed it means no memory, we don't have a better option but stop. | |
324 | */ | |
325 | #define change_pmd_prepare(vma, pmd, cp_flags) \ | |
326 | do { \ | |
327 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ | |
328 | if (WARN_ON_ONCE(pte_alloc(vma->vm_mm, pmd))) \ | |
329 | break; \ | |
330 | } \ | |
331 | } while (0) | |
332 | /* | |
333 | * This is the general pud/p4d/pgd version of change_pmd_prepare(). We need to | |
334 | * have separate change_pmd_prepare() because pte_alloc() returns 0 on success, | |
335 | * while {pmd|pud|p4d}_alloc() returns the valid pointer on success. | |
336 | */ | |
337 | #define change_prepare(vma, high, low, addr, cp_flags) \ | |
338 | do { \ | |
339 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ | |
340 | low##_t *p = low##_alloc(vma->vm_mm, high, addr); \ | |
341 | if (WARN_ON_ONCE(p == NULL)) \ | |
342 | break; \ | |
343 | } \ | |
344 | } while (0) | |
345 | ||
4a18419f NA |
346 | static inline unsigned long change_pmd_range(struct mmu_gather *tlb, |
347 | struct vm_area_struct *vma, pud_t *pud, unsigned long addr, | |
348 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
349 | { |
350 | pmd_t *pmd; | |
351 | unsigned long next; | |
7da4d641 | 352 | unsigned long pages = 0; |
72403b4a | 353 | unsigned long nr_huge_updates = 0; |
ac46d4f3 JG |
354 | struct mmu_notifier_range range; |
355 | ||
356 | range.start = 0; | |
1da177e4 LT |
357 | |
358 | pmd = pmd_offset(pud, addr); | |
359 | do { | |
25cbbef1 MG |
360 | unsigned long this_pages; |
361 | ||
1da177e4 | 362 | next = pmd_addr_end(addr, end); |
8b272b3c | 363 | |
fe2567eb | 364 | change_pmd_prepare(vma, pmd, cp_flags); |
8b272b3c | 365 | /* |
c1e8d7c6 | 366 | * Automatic NUMA balancing walks the tables with mmap_lock |
8b272b3c MG |
367 | * held for read. It's possible a parallel update to occur |
368 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() | |
369 | * check leading to a false positive and clearing. | |
370 | * Hence, it's necessary to atomically read the PMD value | |
371 | * for all the checks. | |
372 | */ | |
373 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && | |
374 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) | |
4991c09c | 375 | goto next; |
a5338093 RR |
376 | |
377 | /* invoke the mmu notifier if the pmd is populated */ | |
ac46d4f3 | 378 | if (!range.start) { |
7269f999 JG |
379 | mmu_notifier_range_init(&range, |
380 | MMU_NOTIFY_PROTECTION_VMA, 0, | |
381 | vma, vma->vm_mm, addr, end); | |
ac46d4f3 | 382 | mmu_notifier_invalidate_range_start(&range); |
a5338093 RR |
383 | } |
384 | ||
84c3fc4e | 385 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
019c2d8b PX |
386 | if ((next - addr != HPAGE_PMD_SIZE) || |
387 | uffd_wp_protect_file(vma, cp_flags)) { | |
fd60775a | 388 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
019c2d8b PX |
389 | /* |
390 | * For file-backed, the pmd could have been | |
391 | * cleared; make sure pmd populated if | |
392 | * necessary, then fall-through to pte level. | |
393 | */ | |
394 | change_pmd_prepare(vma, pmd, cp_flags); | |
6b9116a6 | 395 | } else { |
4a18419f NA |
396 | /* |
397 | * change_huge_pmd() does not defer TLB flushes, | |
398 | * so no need to propagate the tlb argument. | |
399 | */ | |
400 | int nr_ptes = change_huge_pmd(tlb, vma, pmd, | |
401 | addr, newprot, cp_flags); | |
f123d74a MG |
402 | |
403 | if (nr_ptes) { | |
72403b4a MG |
404 | if (nr_ptes == HPAGE_PMD_NR) { |
405 | pages += HPAGE_PMD_NR; | |
406 | nr_huge_updates++; | |
407 | } | |
1ad9f620 MG |
408 | |
409 | /* huge pmd was handled */ | |
4991c09c | 410 | goto next; |
f123d74a | 411 | } |
7da4d641 | 412 | } |
88a9ab6e | 413 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 414 | } |
4a18419f NA |
415 | this_pages = change_pte_range(tlb, vma, pmd, addr, next, |
416 | newprot, cp_flags); | |
25cbbef1 | 417 | pages += this_pages; |
4991c09c AK |
418 | next: |
419 | cond_resched(); | |
1da177e4 | 420 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 421 | |
ac46d4f3 JG |
422 | if (range.start) |
423 | mmu_notifier_invalidate_range_end(&range); | |
a5338093 | 424 | |
72403b4a MG |
425 | if (nr_huge_updates) |
426 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 427 | return pages; |
1da177e4 LT |
428 | } |
429 | ||
4a18419f NA |
430 | static inline unsigned long change_pud_range(struct mmu_gather *tlb, |
431 | struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr, | |
432 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
433 | { |
434 | pud_t *pud; | |
435 | unsigned long next; | |
7da4d641 | 436 | unsigned long pages = 0; |
1da177e4 | 437 | |
c2febafc | 438 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
439 | do { |
440 | next = pud_addr_end(addr, end); | |
fe2567eb | 441 | change_prepare(vma, pud, pmd, addr, cp_flags); |
1da177e4 LT |
442 | if (pud_none_or_clear_bad(pud)) |
443 | continue; | |
4a18419f | 444 | pages += change_pmd_range(tlb, vma, pud, addr, next, newprot, |
58705444 | 445 | cp_flags); |
1da177e4 | 446 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
447 | |
448 | return pages; | |
1da177e4 LT |
449 | } |
450 | ||
4a18419f NA |
451 | static inline unsigned long change_p4d_range(struct mmu_gather *tlb, |
452 | struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr, | |
453 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
c2febafc KS |
454 | { |
455 | p4d_t *p4d; | |
456 | unsigned long next; | |
457 | unsigned long pages = 0; | |
458 | ||
459 | p4d = p4d_offset(pgd, addr); | |
460 | do { | |
461 | next = p4d_addr_end(addr, end); | |
fe2567eb | 462 | change_prepare(vma, p4d, pud, addr, cp_flags); |
c2febafc KS |
463 | if (p4d_none_or_clear_bad(p4d)) |
464 | continue; | |
4a18419f | 465 | pages += change_pud_range(tlb, vma, p4d, addr, next, newprot, |
58705444 | 466 | cp_flags); |
c2febafc KS |
467 | } while (p4d++, addr = next, addr != end); |
468 | ||
469 | return pages; | |
470 | } | |
471 | ||
4a18419f NA |
472 | static unsigned long change_protection_range(struct mmu_gather *tlb, |
473 | struct vm_area_struct *vma, unsigned long addr, | |
474 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
475 | { |
476 | struct mm_struct *mm = vma->vm_mm; | |
477 | pgd_t *pgd; | |
478 | unsigned long next; | |
7da4d641 | 479 | unsigned long pages = 0; |
1da177e4 LT |
480 | |
481 | BUG_ON(addr >= end); | |
482 | pgd = pgd_offset(mm, addr); | |
4a18419f | 483 | tlb_start_vma(tlb, vma); |
1da177e4 LT |
484 | do { |
485 | next = pgd_addr_end(addr, end); | |
fe2567eb | 486 | change_prepare(vma, pgd, p4d, addr, cp_flags); |
1da177e4 LT |
487 | if (pgd_none_or_clear_bad(pgd)) |
488 | continue; | |
4a18419f | 489 | pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot, |
58705444 | 490 | cp_flags); |
1da177e4 | 491 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 492 | |
4a18419f | 493 | tlb_end_vma(tlb, vma); |
7da4d641 PZ |
494 | |
495 | return pages; | |
496 | } | |
497 | ||
4a18419f NA |
498 | unsigned long change_protection(struct mmu_gather *tlb, |
499 | struct vm_area_struct *vma, unsigned long start, | |
7da4d641 | 500 | unsigned long end, pgprot_t newprot, |
58705444 | 501 | unsigned long cp_flags) |
7da4d641 | 502 | { |
7da4d641 PZ |
503 | unsigned long pages; |
504 | ||
292924b2 PX |
505 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
506 | ||
7da4d641 | 507 | if (is_vm_hugetlb_page(vma)) |
5a90d5a1 PX |
508 | pages = hugetlb_change_protection(vma, start, end, newprot, |
509 | cp_flags); | |
7da4d641 | 510 | else |
4a18419f | 511 | pages = change_protection_range(tlb, vma, start, end, newprot, |
58705444 | 512 | cp_flags); |
7da4d641 PZ |
513 | |
514 | return pages; | |
1da177e4 LT |
515 | } |
516 | ||
42e4089c AK |
517 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
518 | unsigned long next, struct mm_walk *walk) | |
519 | { | |
520 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
521 | 0 : -EACCES; | |
522 | } | |
523 | ||
524 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, | |
525 | unsigned long addr, unsigned long next, | |
526 | struct mm_walk *walk) | |
527 | { | |
528 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
529 | 0 : -EACCES; | |
530 | } | |
531 | ||
532 | static int prot_none_test(unsigned long addr, unsigned long next, | |
533 | struct mm_walk *walk) | |
534 | { | |
535 | return 0; | |
536 | } | |
537 | ||
7b86ac33 CH |
538 | static const struct mm_walk_ops prot_none_walk_ops = { |
539 | .pte_entry = prot_none_pte_entry, | |
540 | .hugetlb_entry = prot_none_hugetlb_entry, | |
541 | .test_walk = prot_none_test, | |
542 | }; | |
42e4089c | 543 | |
b6a2fea3 | 544 | int |
4a18419f NA |
545 | mprotect_fixup(struct mmu_gather *tlb, struct vm_area_struct *vma, |
546 | struct vm_area_struct **pprev, unsigned long start, | |
547 | unsigned long end, unsigned long newflags) | |
1da177e4 LT |
548 | { |
549 | struct mm_struct *mm = vma->vm_mm; | |
550 | unsigned long oldflags = vma->vm_flags; | |
551 | long nrpages = (end - start) >> PAGE_SHIFT; | |
552 | unsigned long charged = 0; | |
64fe24a3 | 553 | bool try_change_writable; |
1da177e4 LT |
554 | pgoff_t pgoff; |
555 | int error; | |
556 | ||
557 | if (newflags == oldflags) { | |
558 | *pprev = vma; | |
559 | return 0; | |
560 | } | |
561 | ||
42e4089c AK |
562 | /* |
563 | * Do PROT_NONE PFN permission checks here when we can still | |
564 | * bail out without undoing a lot of state. This is a rather | |
565 | * uncommon case, so doesn't need to be very optimized. | |
566 | */ | |
567 | if (arch_has_pfn_modify_check() && | |
568 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && | |
6cb4d9a2 | 569 | (newflags & VM_ACCESS_FLAGS) == 0) { |
7b86ac33 CH |
570 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
571 | ||
572 | error = walk_page_range(current->mm, start, end, | |
573 | &prot_none_walk_ops, &new_pgprot); | |
42e4089c AK |
574 | if (error) |
575 | return error; | |
576 | } | |
577 | ||
1da177e4 LT |
578 | /* |
579 | * If we make a private mapping writable we increase our commit; | |
580 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
581 | * make it unwritable again. hugetlb mapping were accounted for |
582 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
583 | */ |
584 | if (newflags & VM_WRITE) { | |
84638335 KK |
585 | /* Check space limits when area turns into data. */ |
586 | if (!may_expand_vm(mm, newflags, nrpages) && | |
587 | may_expand_vm(mm, oldflags, nrpages)) | |
588 | return -ENOMEM; | |
5a6fe125 | 589 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 590 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 591 | charged = nrpages; |
191c5424 | 592 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
593 | return -ENOMEM; |
594 | newflags |= VM_ACCOUNT; | |
595 | } | |
596 | } | |
597 | ||
1da177e4 LT |
598 | /* |
599 | * First try to merge with previous and/or next vma. | |
600 | */ | |
601 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
602 | *pprev = vma_merge(mm, *pprev, start, end, newflags, | |
19a809af | 603 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 604 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
605 | if (*pprev) { |
606 | vma = *pprev; | |
e86f15ee | 607 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
1da177e4 LT |
608 | goto success; |
609 | } | |
610 | ||
611 | *pprev = vma; | |
612 | ||
613 | if (start != vma->vm_start) { | |
614 | error = split_vma(mm, vma, start, 1); | |
615 | if (error) | |
616 | goto fail; | |
617 | } | |
618 | ||
619 | if (end != vma->vm_end) { | |
620 | error = split_vma(mm, vma, end, 0); | |
621 | if (error) | |
622 | goto fail; | |
623 | } | |
624 | ||
625 | success: | |
626 | /* | |
c1e8d7c6 | 627 | * vm_flags and vm_page_prot are protected by the mmap_lock |
1da177e4 LT |
628 | * held in write mode. |
629 | */ | |
630 | vma->vm_flags = newflags; | |
64fe24a3 DH |
631 | /* |
632 | * We want to check manually if we can change individual PTEs writable | |
633 | * if we can't do that automatically for all PTEs in a mapping. For | |
634 | * private mappings, that's always the case when we have write | |
635 | * permissions as we properly have to handle COW. | |
636 | */ | |
637 | if (vma->vm_flags & VM_SHARED) | |
638 | try_change_writable = vma_wants_writenotify(vma, vma->vm_page_prot); | |
639 | else | |
640 | try_change_writable = !!(vma->vm_flags & VM_WRITE); | |
64e45507 | 641 | vma_set_page_prot(vma); |
d08b3851 | 642 | |
4a18419f | 643 | change_protection(tlb, vma, start, end, vma->vm_page_prot, |
64fe24a3 | 644 | try_change_writable ? MM_CP_TRY_CHANGE_WRITABLE : 0); |
7da4d641 | 645 | |
36f88188 KS |
646 | /* |
647 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major | |
648 | * fault on access. | |
649 | */ | |
650 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && | |
651 | (newflags & VM_WRITE)) { | |
652 | populate_vma_page_range(vma, start, end, NULL); | |
653 | } | |
654 | ||
84638335 KK |
655 | vm_stat_account(mm, oldflags, -nrpages); |
656 | vm_stat_account(mm, newflags, nrpages); | |
63bfd738 | 657 | perf_event_mmap(vma); |
1da177e4 LT |
658 | return 0; |
659 | ||
660 | fail: | |
661 | vm_unacct_memory(charged); | |
662 | return error; | |
663 | } | |
664 | ||
7d06d9c9 DH |
665 | /* |
666 | * pkey==-1 when doing a legacy mprotect() | |
667 | */ | |
668 | static int do_mprotect_pkey(unsigned long start, size_t len, | |
669 | unsigned long prot, int pkey) | |
1da177e4 | 670 | { |
62b5f7d0 | 671 | unsigned long nstart, end, tmp, reqprot; |
1da177e4 | 672 | struct vm_area_struct *vma, *prev; |
48725bbc | 673 | int error; |
1da177e4 | 674 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); |
f138556d PK |
675 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
676 | (prot & PROT_READ); | |
4a18419f | 677 | struct mmu_gather tlb; |
70821e0b | 678 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
f138556d | 679 | |
057d3389 AK |
680 | start = untagged_addr(start); |
681 | ||
1da177e4 LT |
682 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
683 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
684 | return -EINVAL; | |
685 | ||
686 | if (start & ~PAGE_MASK) | |
687 | return -EINVAL; | |
688 | if (!len) | |
689 | return 0; | |
690 | len = PAGE_ALIGN(len); | |
691 | end = start + len; | |
692 | if (end <= start) | |
693 | return -ENOMEM; | |
9035cf9a | 694 | if (!arch_validate_prot(prot, start)) |
1da177e4 LT |
695 | return -EINVAL; |
696 | ||
697 | reqprot = prot; | |
1da177e4 | 698 | |
d8ed45c5 | 699 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 700 | return -EINTR; |
1da177e4 | 701 | |
e8c24d3a DH |
702 | /* |
703 | * If userspace did not allocate the pkey, do not let | |
704 | * them use it here. | |
705 | */ | |
706 | error = -EINVAL; | |
707 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) | |
708 | goto out; | |
709 | ||
70821e0b LH |
710 | mas_set(&mas, start); |
711 | vma = mas_find(&mas, ULONG_MAX); | |
1da177e4 LT |
712 | error = -ENOMEM; |
713 | if (!vma) | |
714 | goto out; | |
6af5fa0d | 715 | |
1da177e4 LT |
716 | if (unlikely(grows & PROT_GROWSDOWN)) { |
717 | if (vma->vm_start >= end) | |
718 | goto out; | |
719 | start = vma->vm_start; | |
720 | error = -EINVAL; | |
721 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
722 | goto out; | |
7d12efae | 723 | } else { |
1da177e4 LT |
724 | if (vma->vm_start > start) |
725 | goto out; | |
726 | if (unlikely(grows & PROT_GROWSUP)) { | |
727 | end = vma->vm_end; | |
728 | error = -EINVAL; | |
729 | if (!(vma->vm_flags & VM_GROWSUP)) | |
730 | goto out; | |
731 | } | |
732 | } | |
6af5fa0d | 733 | |
1da177e4 LT |
734 | if (start > vma->vm_start) |
735 | prev = vma; | |
6af5fa0d | 736 | else |
70821e0b | 737 | prev = mas_prev(&mas, 0); |
1da177e4 | 738 | |
4a18419f | 739 | tlb_gather_mmu(&tlb, current->mm); |
1da177e4 | 740 | for (nstart = start ; ; ) { |
a8502b67 | 741 | unsigned long mask_off_old_flags; |
1da177e4 | 742 | unsigned long newflags; |
7d06d9c9 | 743 | int new_vma_pkey; |
1da177e4 | 744 | |
7d12efae | 745 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
1da177e4 | 746 | |
f138556d PK |
747 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
748 | if (rier && (vma->vm_flags & VM_MAYEXEC)) | |
749 | prot |= PROT_EXEC; | |
750 | ||
a8502b67 DH |
751 | /* |
752 | * Each mprotect() call explicitly passes r/w/x permissions. | |
753 | * If a permission is not passed to mprotect(), it must be | |
754 | * cleared from the VMA. | |
755 | */ | |
e39ee675 | 756 | mask_off_old_flags = VM_ACCESS_FLAGS | VM_FLAGS_CLEAR; |
a8502b67 | 757 | |
7d06d9c9 DH |
758 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
759 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); | |
a8502b67 | 760 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
1da177e4 | 761 | |
7e2cff42 | 762 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
6cb4d9a2 | 763 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
1da177e4 | 764 | error = -EACCES; |
4a18419f | 765 | break; |
1da177e4 LT |
766 | } |
767 | ||
c462ac28 CM |
768 | /* Allow architectures to sanity-check the new flags */ |
769 | if (!arch_validate_flags(newflags)) { | |
770 | error = -EINVAL; | |
4a18419f | 771 | break; |
c462ac28 CM |
772 | } |
773 | ||
1da177e4 LT |
774 | error = security_file_mprotect(vma, reqprot, prot); |
775 | if (error) | |
4a18419f | 776 | break; |
1da177e4 LT |
777 | |
778 | tmp = vma->vm_end; | |
779 | if (tmp > end) | |
780 | tmp = end; | |
95bb7c42 | 781 | |
dbf53f75 | 782 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
95bb7c42 | 783 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
dbf53f75 | 784 | if (error) |
4a18419f | 785 | break; |
dbf53f75 | 786 | } |
95bb7c42 | 787 | |
4a18419f | 788 | error = mprotect_fixup(&tlb, vma, &prev, nstart, tmp, newflags); |
1da177e4 | 789 | if (error) |
4a18419f | 790 | break; |
95bb7c42 | 791 | |
1da177e4 LT |
792 | nstart = tmp; |
793 | ||
794 | if (nstart < prev->vm_end) | |
795 | nstart = prev->vm_end; | |
796 | if (nstart >= end) | |
4a18419f | 797 | break; |
1da177e4 | 798 | |
70821e0b | 799 | vma = find_vma(current->mm, prev->vm_end); |
1da177e4 LT |
800 | if (!vma || vma->vm_start != nstart) { |
801 | error = -ENOMEM; | |
4a18419f | 802 | break; |
1da177e4 | 803 | } |
f138556d | 804 | prot = reqprot; |
1da177e4 | 805 | } |
4a18419f | 806 | tlb_finish_mmu(&tlb); |
1da177e4 | 807 | out: |
d8ed45c5 | 808 | mmap_write_unlock(current->mm); |
1da177e4 LT |
809 | return error; |
810 | } | |
7d06d9c9 DH |
811 | |
812 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, | |
813 | unsigned long, prot) | |
814 | { | |
815 | return do_mprotect_pkey(start, len, prot, -1); | |
816 | } | |
817 | ||
c7142aea HC |
818 | #ifdef CONFIG_ARCH_HAS_PKEYS |
819 | ||
7d06d9c9 DH |
820 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
821 | unsigned long, prot, int, pkey) | |
822 | { | |
823 | return do_mprotect_pkey(start, len, prot, pkey); | |
824 | } | |
e8c24d3a DH |
825 | |
826 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) | |
827 | { | |
828 | int pkey; | |
829 | int ret; | |
830 | ||
831 | /* No flags supported yet. */ | |
832 | if (flags) | |
833 | return -EINVAL; | |
834 | /* check for unsupported init values */ | |
835 | if (init_val & ~PKEY_ACCESS_MASK) | |
836 | return -EINVAL; | |
837 | ||
d8ed45c5 | 838 | mmap_write_lock(current->mm); |
e8c24d3a DH |
839 | pkey = mm_pkey_alloc(current->mm); |
840 | ||
841 | ret = -ENOSPC; | |
842 | if (pkey == -1) | |
843 | goto out; | |
844 | ||
845 | ret = arch_set_user_pkey_access(current, pkey, init_val); | |
846 | if (ret) { | |
847 | mm_pkey_free(current->mm, pkey); | |
848 | goto out; | |
849 | } | |
850 | ret = pkey; | |
851 | out: | |
d8ed45c5 | 852 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
853 | return ret; |
854 | } | |
855 | ||
856 | SYSCALL_DEFINE1(pkey_free, int, pkey) | |
857 | { | |
858 | int ret; | |
859 | ||
d8ed45c5 | 860 | mmap_write_lock(current->mm); |
e8c24d3a | 861 | ret = mm_pkey_free(current->mm, pkey); |
d8ed45c5 | 862 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
863 | |
864 | /* | |
f0953a1b | 865 | * We could provide warnings or errors if any VMA still |
e8c24d3a DH |
866 | * has the pkey set here. |
867 | */ | |
868 | return ret; | |
869 | } | |
c7142aea HC |
870 | |
871 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |