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); | |
4dd845b5 | 226 | } else if (is_writable_device_private_entry(entry)) { |
5042db43 JG |
227 | /* |
228 | * We do not preserve soft-dirtiness. See | |
229 | * copy_one_pte() for explanation. | |
230 | */ | |
4dd845b5 AP |
231 | entry = make_readable_device_private_entry( |
232 | swp_offset(entry)); | |
5042db43 | 233 | newpte = swp_entry_to_pte(entry); |
f45ec5ff PX |
234 | if (pte_swp_uffd_wp(oldpte)) |
235 | newpte = pte_swp_mkuffd_wp(newpte); | |
b756a3b5 AP |
236 | } else if (is_writable_device_exclusive_entry(entry)) { |
237 | entry = make_readable_device_exclusive_entry( | |
238 | swp_offset(entry)); | |
239 | newpte = swp_entry_to_pte(entry); | |
240 | if (pte_swp_soft_dirty(oldpte)) | |
241 | newpte = pte_swp_mksoft_dirty(newpte); | |
242 | if (pte_swp_uffd_wp(oldpte)) | |
243 | newpte = pte_swp_mkuffd_wp(newpte); | |
7e3ce3f8 PX |
244 | } else if (is_pte_marker_entry(entry)) { |
245 | /* | |
246 | * Ignore swapin errors unconditionally, | |
247 | * because any access should sigbus anyway. | |
248 | */ | |
249 | if (is_swapin_error_entry(entry)) | |
250 | continue; | |
fe2567eb PX |
251 | /* |
252 | * If this is uffd-wp pte marker and we'd like | |
253 | * to unprotect it, drop it; the next page | |
254 | * fault will trigger without uffd trapping. | |
255 | */ | |
256 | if (uffd_wp_resolve) { | |
257 | pte_clear(vma->vm_mm, addr, pte); | |
258 | pages++; | |
259 | } | |
5c041f5d | 260 | continue; |
f45ec5ff PX |
261 | } else { |
262 | newpte = oldpte; | |
263 | } | |
5042db43 | 264 | |
f45ec5ff PX |
265 | if (uffd_wp) |
266 | newpte = pte_swp_mkuffd_wp(newpte); | |
267 | else if (uffd_wp_resolve) | |
268 | newpte = pte_swp_clear_uffd_wp(newpte); | |
269 | ||
270 | if (!pte_same(oldpte, newpte)) { | |
271 | set_pte_at(vma->vm_mm, addr, pte, newpte); | |
5042db43 JG |
272 | pages++; |
273 | } | |
fe2567eb PX |
274 | } else { |
275 | /* It must be an none page, or what else?.. */ | |
276 | WARN_ON_ONCE(!pte_none(oldpte)); | |
2bad466c PX |
277 | |
278 | /* | |
279 | * Nobody plays with any none ptes besides | |
280 | * userfaultfd when applying the protections. | |
281 | */ | |
282 | if (likely(!uffd_wp)) | |
283 | continue; | |
284 | ||
285 | if (userfaultfd_wp_use_markers(vma)) { | |
fe2567eb PX |
286 | /* |
287 | * For file-backed mem, we need to be able to | |
288 | * wr-protect a none pte, because even if the | |
289 | * pte is none, the page/swap cache could | |
290 | * exist. Doing that by install a marker. | |
291 | */ | |
292 | set_pte_at(vma->vm_mm, addr, pte, | |
293 | make_pte_marker(PTE_MARKER_UFFD_WP)); | |
294 | pages++; | |
295 | } | |
1da177e4 LT |
296 | } |
297 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
6606c3e0 | 298 | arch_leave_lazy_mmu_mode(); |
705e87c0 | 299 | pte_unmap_unlock(pte - 1, ptl); |
7da4d641 PZ |
300 | |
301 | return pages; | |
1da177e4 LT |
302 | } |
303 | ||
8b272b3c MG |
304 | /* |
305 | * Used when setting automatic NUMA hinting protection where it is | |
306 | * critical that a numa hinting PMD is not confused with a bad PMD. | |
307 | */ | |
308 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) | |
309 | { | |
dab6e717 | 310 | pmd_t pmdval = pmdp_get_lockless(pmd); |
8b272b3c MG |
311 | |
312 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ | |
313 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
314 | barrier(); | |
315 | #endif | |
316 | ||
317 | if (pmd_none(pmdval)) | |
318 | return 1; | |
319 | if (pmd_trans_huge(pmdval)) | |
320 | return 0; | |
321 | if (unlikely(pmd_bad(pmdval))) { | |
322 | pmd_clear_bad(pmd); | |
323 | return 1; | |
324 | } | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
2bad466c PX |
329 | /* |
330 | * Return true if we want to split THPs into PTE mappings in change | |
331 | * protection procedure, false otherwise. | |
332 | */ | |
fe2567eb | 333 | static inline bool |
2bad466c | 334 | pgtable_split_needed(struct vm_area_struct *vma, unsigned long cp_flags) |
fe2567eb | 335 | { |
2bad466c PX |
336 | /* |
337 | * pte markers only resides in pte level, if we need pte markers, | |
338 | * we need to split. We cannot wr-protect shmem thp because file | |
339 | * thp is handled differently when split by erasing the pmd so far. | |
340 | */ | |
fe2567eb PX |
341 | return (cp_flags & MM_CP_UFFD_WP) && !vma_is_anonymous(vma); |
342 | } | |
343 | ||
344 | /* | |
2bad466c PX |
345 | * Return true if we want to populate pgtables in change protection |
346 | * procedure, false otherwise | |
347 | */ | |
348 | static inline bool | |
349 | pgtable_populate_needed(struct vm_area_struct *vma, unsigned long cp_flags) | |
350 | { | |
351 | /* If not within ioctl(UFFDIO_WRITEPROTECT), then don't bother */ | |
352 | if (!(cp_flags & MM_CP_UFFD_WP)) | |
353 | return false; | |
354 | ||
355 | /* Populate if the userfaultfd mode requires pte markers */ | |
356 | return userfaultfd_wp_use_markers(vma); | |
357 | } | |
358 | ||
359 | /* | |
360 | * Populate the pgtable underneath for whatever reason if requested. | |
361 | * When {pte|pmd|...}_alloc() failed we treat it the same way as pgtable | |
362 | * allocation failures during page faults by kicking OOM and returning | |
363 | * error. | |
fe2567eb PX |
364 | */ |
365 | #define change_pmd_prepare(vma, pmd, cp_flags) \ | |
d1751118 PX |
366 | ({ \ |
367 | long err = 0; \ | |
2bad466c | 368 | if (unlikely(pgtable_populate_needed(vma, cp_flags))) { \ |
d1751118 PX |
369 | if (pte_alloc(vma->vm_mm, pmd)) \ |
370 | err = -ENOMEM; \ | |
fe2567eb | 371 | } \ |
d1751118 PX |
372 | err; \ |
373 | }) | |
374 | ||
fe2567eb PX |
375 | /* |
376 | * This is the general pud/p4d/pgd version of change_pmd_prepare(). We need to | |
377 | * have separate change_pmd_prepare() because pte_alloc() returns 0 on success, | |
378 | * while {pmd|pud|p4d}_alloc() returns the valid pointer on success. | |
379 | */ | |
380 | #define change_prepare(vma, high, low, addr, cp_flags) \ | |
d1751118 PX |
381 | ({ \ |
382 | long err = 0; \ | |
2bad466c | 383 | if (unlikely(pgtable_populate_needed(vma, cp_flags))) { \ |
fe2567eb | 384 | low##_t *p = low##_alloc(vma->vm_mm, high, addr); \ |
d1751118 PX |
385 | if (p == NULL) \ |
386 | err = -ENOMEM; \ | |
fe2567eb | 387 | } \ |
d1751118 PX |
388 | err; \ |
389 | }) | |
fe2567eb | 390 | |
a79390f5 | 391 | static inline long change_pmd_range(struct mmu_gather *tlb, |
4a18419f NA |
392 | struct vm_area_struct *vma, pud_t *pud, unsigned long addr, |
393 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
394 | { |
395 | pmd_t *pmd; | |
396 | unsigned long next; | |
a79390f5 | 397 | long pages = 0; |
72403b4a | 398 | unsigned long nr_huge_updates = 0; |
ac46d4f3 JG |
399 | struct mmu_notifier_range range; |
400 | ||
401 | range.start = 0; | |
1da177e4 LT |
402 | |
403 | pmd = pmd_offset(pud, addr); | |
404 | do { | |
d1751118 | 405 | long ret; |
25cbbef1 | 406 | |
1da177e4 | 407 | next = pmd_addr_end(addr, end); |
8b272b3c | 408 | |
d1751118 PX |
409 | ret = change_pmd_prepare(vma, pmd, cp_flags); |
410 | if (ret) { | |
411 | pages = ret; | |
412 | break; | |
413 | } | |
8b272b3c | 414 | /* |
c1e8d7c6 | 415 | * Automatic NUMA balancing walks the tables with mmap_lock |
8b272b3c MG |
416 | * held for read. It's possible a parallel update to occur |
417 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() | |
418 | * check leading to a false positive and clearing. | |
419 | * Hence, it's necessary to atomically read the PMD value | |
420 | * for all the checks. | |
421 | */ | |
422 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && | |
423 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) | |
4991c09c | 424 | goto next; |
a5338093 RR |
425 | |
426 | /* invoke the mmu notifier if the pmd is populated */ | |
ac46d4f3 | 427 | if (!range.start) { |
7269f999 JG |
428 | mmu_notifier_range_init(&range, |
429 | MMU_NOTIFY_PROTECTION_VMA, 0, | |
7d4a8be0 | 430 | vma->vm_mm, addr, end); |
ac46d4f3 | 431 | mmu_notifier_invalidate_range_start(&range); |
a5338093 RR |
432 | } |
433 | ||
84c3fc4e | 434 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
019c2d8b | 435 | if ((next - addr != HPAGE_PMD_SIZE) || |
2bad466c | 436 | pgtable_split_needed(vma, cp_flags)) { |
fd60775a | 437 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
019c2d8b PX |
438 | /* |
439 | * For file-backed, the pmd could have been | |
440 | * cleared; make sure pmd populated if | |
441 | * necessary, then fall-through to pte level. | |
442 | */ | |
d1751118 PX |
443 | ret = change_pmd_prepare(vma, pmd, cp_flags); |
444 | if (ret) { | |
445 | pages = ret; | |
446 | break; | |
447 | } | |
6b9116a6 | 448 | } else { |
4a18419f NA |
449 | /* |
450 | * change_huge_pmd() does not defer TLB flushes, | |
451 | * so no need to propagate the tlb argument. | |
452 | */ | |
453 | int nr_ptes = change_huge_pmd(tlb, vma, pmd, | |
454 | addr, newprot, cp_flags); | |
f123d74a MG |
455 | |
456 | if (nr_ptes) { | |
72403b4a MG |
457 | if (nr_ptes == HPAGE_PMD_NR) { |
458 | pages += HPAGE_PMD_NR; | |
459 | nr_huge_updates++; | |
460 | } | |
1ad9f620 MG |
461 | |
462 | /* huge pmd was handled */ | |
4991c09c | 463 | goto next; |
f123d74a | 464 | } |
7da4d641 | 465 | } |
88a9ab6e | 466 | /* fall through, the trans huge pmd just split */ |
cd7548ab | 467 | } |
d1751118 PX |
468 | pages += change_pte_range(tlb, vma, pmd, addr, next, |
469 | newprot, cp_flags); | |
4991c09c AK |
470 | next: |
471 | cond_resched(); | |
1da177e4 | 472 | } while (pmd++, addr = next, addr != end); |
7da4d641 | 473 | |
ac46d4f3 JG |
474 | if (range.start) |
475 | mmu_notifier_invalidate_range_end(&range); | |
a5338093 | 476 | |
72403b4a MG |
477 | if (nr_huge_updates) |
478 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); | |
7da4d641 | 479 | return pages; |
1da177e4 LT |
480 | } |
481 | ||
a79390f5 | 482 | static inline long change_pud_range(struct mmu_gather *tlb, |
4a18419f NA |
483 | struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr, |
484 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
485 | { |
486 | pud_t *pud; | |
487 | unsigned long next; | |
d1751118 | 488 | long pages = 0, ret; |
1da177e4 | 489 | |
c2febafc | 490 | pud = pud_offset(p4d, addr); |
1da177e4 LT |
491 | do { |
492 | next = pud_addr_end(addr, end); | |
d1751118 PX |
493 | ret = change_prepare(vma, pud, pmd, addr, cp_flags); |
494 | if (ret) | |
495 | return ret; | |
1da177e4 LT |
496 | if (pud_none_or_clear_bad(pud)) |
497 | continue; | |
4a18419f | 498 | pages += change_pmd_range(tlb, vma, pud, addr, next, newprot, |
58705444 | 499 | cp_flags); |
1da177e4 | 500 | } while (pud++, addr = next, addr != end); |
7da4d641 PZ |
501 | |
502 | return pages; | |
1da177e4 LT |
503 | } |
504 | ||
a79390f5 | 505 | static inline long change_p4d_range(struct mmu_gather *tlb, |
4a18419f NA |
506 | struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr, |
507 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
c2febafc KS |
508 | { |
509 | p4d_t *p4d; | |
510 | unsigned long next; | |
d1751118 | 511 | long pages = 0, ret; |
c2febafc KS |
512 | |
513 | p4d = p4d_offset(pgd, addr); | |
514 | do { | |
515 | next = p4d_addr_end(addr, end); | |
d1751118 PX |
516 | ret = change_prepare(vma, p4d, pud, addr, cp_flags); |
517 | if (ret) | |
518 | return ret; | |
c2febafc KS |
519 | if (p4d_none_or_clear_bad(p4d)) |
520 | continue; | |
4a18419f | 521 | pages += change_pud_range(tlb, vma, p4d, addr, next, newprot, |
58705444 | 522 | cp_flags); |
c2febafc KS |
523 | } while (p4d++, addr = next, addr != end); |
524 | ||
525 | return pages; | |
526 | } | |
527 | ||
a79390f5 | 528 | static long change_protection_range(struct mmu_gather *tlb, |
4a18419f NA |
529 | struct vm_area_struct *vma, unsigned long addr, |
530 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) | |
1da177e4 LT |
531 | { |
532 | struct mm_struct *mm = vma->vm_mm; | |
533 | pgd_t *pgd; | |
534 | unsigned long next; | |
d1751118 | 535 | long pages = 0, ret; |
1da177e4 LT |
536 | |
537 | BUG_ON(addr >= end); | |
538 | pgd = pgd_offset(mm, addr); | |
4a18419f | 539 | tlb_start_vma(tlb, vma); |
1da177e4 LT |
540 | do { |
541 | next = pgd_addr_end(addr, end); | |
d1751118 PX |
542 | ret = change_prepare(vma, pgd, p4d, addr, cp_flags); |
543 | if (ret) { | |
544 | pages = ret; | |
545 | break; | |
546 | } | |
1da177e4 LT |
547 | if (pgd_none_or_clear_bad(pgd)) |
548 | continue; | |
4a18419f | 549 | pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot, |
58705444 | 550 | cp_flags); |
1da177e4 | 551 | } while (pgd++, addr = next, addr != end); |
7da4d641 | 552 | |
4a18419f | 553 | tlb_end_vma(tlb, vma); |
7da4d641 PZ |
554 | |
555 | return pages; | |
556 | } | |
557 | ||
a79390f5 | 558 | long change_protection(struct mmu_gather *tlb, |
4a18419f | 559 | struct vm_area_struct *vma, unsigned long start, |
1ef488ed | 560 | unsigned long end, unsigned long cp_flags) |
7da4d641 | 561 | { |
1ef488ed | 562 | pgprot_t newprot = vma->vm_page_prot; |
a79390f5 | 563 | long pages; |
7da4d641 | 564 | |
292924b2 PX |
565 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
566 | ||
1ef488ed DH |
567 | #ifdef CONFIG_NUMA_BALANCING |
568 | /* | |
569 | * Ordinary protection updates (mprotect, uffd-wp, softdirty tracking) | |
570 | * are expected to reflect their requirements via VMA flags such that | |
571 | * vma_set_page_prot() will adjust vma->vm_page_prot accordingly. | |
572 | */ | |
573 | if (cp_flags & MM_CP_PROT_NUMA) | |
574 | newprot = PAGE_NONE; | |
575 | #else | |
576 | WARN_ON_ONCE(cp_flags & MM_CP_PROT_NUMA); | |
577 | #endif | |
578 | ||
7da4d641 | 579 | if (is_vm_hugetlb_page(vma)) |
5a90d5a1 PX |
580 | pages = hugetlb_change_protection(vma, start, end, newprot, |
581 | cp_flags); | |
7da4d641 | 582 | else |
4a18419f | 583 | pages = change_protection_range(tlb, vma, start, end, newprot, |
58705444 | 584 | cp_flags); |
7da4d641 PZ |
585 | |
586 | return pages; | |
1da177e4 LT |
587 | } |
588 | ||
42e4089c AK |
589 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
590 | unsigned long next, struct mm_walk *walk) | |
591 | { | |
592 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
593 | 0 : -EACCES; | |
594 | } | |
595 | ||
596 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, | |
597 | unsigned long addr, unsigned long next, | |
598 | struct mm_walk *walk) | |
599 | { | |
600 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? | |
601 | 0 : -EACCES; | |
602 | } | |
603 | ||
604 | static int prot_none_test(unsigned long addr, unsigned long next, | |
605 | struct mm_walk *walk) | |
606 | { | |
607 | return 0; | |
608 | } | |
609 | ||
7b86ac33 CH |
610 | static const struct mm_walk_ops prot_none_walk_ops = { |
611 | .pte_entry = prot_none_pte_entry, | |
612 | .hugetlb_entry = prot_none_hugetlb_entry, | |
613 | .test_walk = prot_none_test, | |
614 | }; | |
42e4089c | 615 | |
b6a2fea3 | 616 | int |
2286a691 LH |
617 | mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb, |
618 | struct vm_area_struct *vma, struct vm_area_struct **pprev, | |
619 | unsigned long start, unsigned long end, unsigned long newflags) | |
1da177e4 LT |
620 | { |
621 | struct mm_struct *mm = vma->vm_mm; | |
622 | unsigned long oldflags = vma->vm_flags; | |
623 | long nrpages = (end - start) >> PAGE_SHIFT; | |
eb309ec8 | 624 | unsigned int mm_cp_flags = 0; |
1da177e4 | 625 | unsigned long charged = 0; |
1da177e4 LT |
626 | pgoff_t pgoff; |
627 | int error; | |
628 | ||
629 | if (newflags == oldflags) { | |
630 | *pprev = vma; | |
631 | return 0; | |
632 | } | |
633 | ||
42e4089c AK |
634 | /* |
635 | * Do PROT_NONE PFN permission checks here when we can still | |
636 | * bail out without undoing a lot of state. This is a rather | |
637 | * uncommon case, so doesn't need to be very optimized. | |
638 | */ | |
639 | if (arch_has_pfn_modify_check() && | |
640 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && | |
6cb4d9a2 | 641 | (newflags & VM_ACCESS_FLAGS) == 0) { |
7b86ac33 CH |
642 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
643 | ||
644 | error = walk_page_range(current->mm, start, end, | |
645 | &prot_none_walk_ops, &new_pgprot); | |
42e4089c AK |
646 | if (error) |
647 | return error; | |
648 | } | |
649 | ||
1da177e4 LT |
650 | /* |
651 | * If we make a private mapping writable we increase our commit; | |
652 | * but (without finer accounting) cannot reduce our commit if we | |
5a6fe125 MG |
653 | * make it unwritable again. hugetlb mapping were accounted for |
654 | * even if read-only so there is no need to account for them here | |
1da177e4 LT |
655 | */ |
656 | if (newflags & VM_WRITE) { | |
84638335 KK |
657 | /* Check space limits when area turns into data. */ |
658 | if (!may_expand_vm(mm, newflags, nrpages) && | |
659 | may_expand_vm(mm, oldflags, nrpages)) | |
660 | return -ENOMEM; | |
5a6fe125 | 661 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
cdfd4325 | 662 | VM_SHARED|VM_NORESERVE))) { |
1da177e4 | 663 | charged = nrpages; |
191c5424 | 664 | if (security_vm_enough_memory_mm(mm, charged)) |
1da177e4 LT |
665 | return -ENOMEM; |
666 | newflags |= VM_ACCOUNT; | |
667 | } | |
668 | } | |
669 | ||
1da177e4 LT |
670 | /* |
671 | * First try to merge with previous and/or next vma. | |
672 | */ | |
673 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); | |
9760ebff | 674 | *pprev = vma_merge(vmi, mm, *pprev, start, end, newflags, |
19a809af | 675 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
5c26f6ac | 676 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
1da177e4 LT |
677 | if (*pprev) { |
678 | vma = *pprev; | |
e86f15ee | 679 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
1da177e4 LT |
680 | goto success; |
681 | } | |
682 | ||
683 | *pprev = vma; | |
684 | ||
685 | if (start != vma->vm_start) { | |
9760ebff | 686 | error = split_vma(vmi, vma, start, 1); |
1da177e4 LT |
687 | if (error) |
688 | goto fail; | |
689 | } | |
690 | ||
691 | if (end != vma->vm_end) { | |
9760ebff | 692 | error = split_vma(vmi, vma, end, 0); |
1da177e4 LT |
693 | if (error) |
694 | goto fail; | |
695 | } | |
696 | ||
697 | success: | |
698 | /* | |
c1e8d7c6 | 699 | * vm_flags and vm_page_prot are protected by the mmap_lock |
1da177e4 LT |
700 | * held in write mode. |
701 | */ | |
1c71222e | 702 | vm_flags_reset(vma, newflags); |
eb309ec8 DH |
703 | if (vma_wants_manual_pte_write_upgrade(vma)) |
704 | mm_cp_flags |= MM_CP_TRY_CHANGE_WRITABLE; | |
64e45507 | 705 | vma_set_page_prot(vma); |
d08b3851 | 706 | |
1ef488ed | 707 | change_protection(tlb, vma, start, end, mm_cp_flags); |
7da4d641 | 708 | |
36f88188 KS |
709 | /* |
710 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major | |
711 | * fault on access. | |
712 | */ | |
713 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && | |
714 | (newflags & VM_WRITE)) { | |
715 | populate_vma_page_range(vma, start, end, NULL); | |
716 | } | |
717 | ||
84638335 KK |
718 | vm_stat_account(mm, oldflags, -nrpages); |
719 | vm_stat_account(mm, newflags, nrpages); | |
63bfd738 | 720 | perf_event_mmap(vma); |
1da177e4 LT |
721 | return 0; |
722 | ||
723 | fail: | |
724 | vm_unacct_memory(charged); | |
725 | return error; | |
726 | } | |
727 | ||
7d06d9c9 DH |
728 | /* |
729 | * pkey==-1 when doing a legacy mprotect() | |
730 | */ | |
731 | static int do_mprotect_pkey(unsigned long start, size_t len, | |
732 | unsigned long prot, int pkey) | |
1da177e4 | 733 | { |
62b5f7d0 | 734 | unsigned long nstart, end, tmp, reqprot; |
1da177e4 | 735 | struct vm_area_struct *vma, *prev; |
48725bbc | 736 | int error; |
1da177e4 | 737 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); |
f138556d PK |
738 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
739 | (prot & PROT_READ); | |
4a18419f | 740 | struct mmu_gather tlb; |
2286a691 | 741 | struct vma_iterator vmi; |
f138556d | 742 | |
057d3389 AK |
743 | start = untagged_addr(start); |
744 | ||
1da177e4 LT |
745 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
746 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ | |
747 | return -EINVAL; | |
748 | ||
749 | if (start & ~PAGE_MASK) | |
750 | return -EINVAL; | |
751 | if (!len) | |
752 | return 0; | |
753 | len = PAGE_ALIGN(len); | |
754 | end = start + len; | |
755 | if (end <= start) | |
756 | return -ENOMEM; | |
9035cf9a | 757 | if (!arch_validate_prot(prot, start)) |
1da177e4 LT |
758 | return -EINVAL; |
759 | ||
760 | reqprot = prot; | |
1da177e4 | 761 | |
d8ed45c5 | 762 | if (mmap_write_lock_killable(current->mm)) |
dc0ef0df | 763 | return -EINTR; |
1da177e4 | 764 | |
e8c24d3a DH |
765 | /* |
766 | * If userspace did not allocate the pkey, do not let | |
767 | * them use it here. | |
768 | */ | |
769 | error = -EINVAL; | |
770 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) | |
771 | goto out; | |
772 | ||
2286a691 LH |
773 | vma_iter_init(&vmi, current->mm, start); |
774 | vma = vma_find(&vmi, end); | |
1da177e4 LT |
775 | error = -ENOMEM; |
776 | if (!vma) | |
777 | goto out; | |
6af5fa0d | 778 | |
1da177e4 LT |
779 | if (unlikely(grows & PROT_GROWSDOWN)) { |
780 | if (vma->vm_start >= end) | |
781 | goto out; | |
782 | start = vma->vm_start; | |
783 | error = -EINVAL; | |
784 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
785 | goto out; | |
7d12efae | 786 | } else { |
1da177e4 LT |
787 | if (vma->vm_start > start) |
788 | goto out; | |
789 | if (unlikely(grows & PROT_GROWSUP)) { | |
790 | end = vma->vm_end; | |
791 | error = -EINVAL; | |
792 | if (!(vma->vm_flags & VM_GROWSUP)) | |
793 | goto out; | |
794 | } | |
795 | } | |
6af5fa0d | 796 | |
2286a691 | 797 | prev = vma_prev(&vmi); |
1da177e4 LT |
798 | if (start > vma->vm_start) |
799 | prev = vma; | |
800 | ||
4a18419f | 801 | tlb_gather_mmu(&tlb, current->mm); |
2286a691 LH |
802 | nstart = start; |
803 | tmp = vma->vm_start; | |
804 | for_each_vma_range(vmi, vma, end) { | |
a8502b67 | 805 | unsigned long mask_off_old_flags; |
1da177e4 | 806 | unsigned long newflags; |
7d06d9c9 | 807 | int new_vma_pkey; |
1da177e4 | 808 | |
2286a691 LH |
809 | if (vma->vm_start != tmp) { |
810 | error = -ENOMEM; | |
811 | break; | |
812 | } | |
1da177e4 | 813 | |
f138556d PK |
814 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
815 | if (rier && (vma->vm_flags & VM_MAYEXEC)) | |
816 | prot |= PROT_EXEC; | |
817 | ||
a8502b67 DH |
818 | /* |
819 | * Each mprotect() call explicitly passes r/w/x permissions. | |
820 | * If a permission is not passed to mprotect(), it must be | |
821 | * cleared from the VMA. | |
822 | */ | |
e39ee675 | 823 | mask_off_old_flags = VM_ACCESS_FLAGS | VM_FLAGS_CLEAR; |
a8502b67 | 824 | |
7d06d9c9 DH |
825 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
826 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); | |
a8502b67 | 827 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
1da177e4 | 828 | |
7e2cff42 | 829 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
6cb4d9a2 | 830 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
1da177e4 | 831 | error = -EACCES; |
4a18419f | 832 | break; |
1da177e4 LT |
833 | } |
834 | ||
b507808e JG |
835 | if (map_deny_write_exec(vma, newflags)) { |
836 | error = -EACCES; | |
3d27a95b | 837 | break; |
b507808e JG |
838 | } |
839 | ||
c462ac28 CM |
840 | /* Allow architectures to sanity-check the new flags */ |
841 | if (!arch_validate_flags(newflags)) { | |
842 | error = -EINVAL; | |
4a18419f | 843 | break; |
c462ac28 CM |
844 | } |
845 | ||
1da177e4 LT |
846 | error = security_file_mprotect(vma, reqprot, prot); |
847 | if (error) | |
4a18419f | 848 | break; |
1da177e4 LT |
849 | |
850 | tmp = vma->vm_end; | |
851 | if (tmp > end) | |
852 | tmp = end; | |
95bb7c42 | 853 | |
dbf53f75 | 854 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
95bb7c42 | 855 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
dbf53f75 | 856 | if (error) |
4a18419f | 857 | break; |
dbf53f75 | 858 | } |
95bb7c42 | 859 | |
2286a691 | 860 | error = mprotect_fixup(&vmi, &tlb, vma, &prev, nstart, tmp, newflags); |
1da177e4 | 861 | if (error) |
4a18419f | 862 | break; |
95bb7c42 | 863 | |
2fcd07b7 | 864 | tmp = vma_iter_end(&vmi); |
1da177e4 | 865 | nstart = tmp; |
f138556d | 866 | prot = reqprot; |
1da177e4 | 867 | } |
4a18419f | 868 | tlb_finish_mmu(&tlb); |
2286a691 | 869 | |
82f95134 | 870 | if (!error && vma_iter_end(&vmi) < end) |
2286a691 LH |
871 | error = -ENOMEM; |
872 | ||
1da177e4 | 873 | out: |
d8ed45c5 | 874 | mmap_write_unlock(current->mm); |
1da177e4 LT |
875 | return error; |
876 | } | |
7d06d9c9 DH |
877 | |
878 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, | |
879 | unsigned long, prot) | |
880 | { | |
881 | return do_mprotect_pkey(start, len, prot, -1); | |
882 | } | |
883 | ||
c7142aea HC |
884 | #ifdef CONFIG_ARCH_HAS_PKEYS |
885 | ||
7d06d9c9 DH |
886 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
887 | unsigned long, prot, int, pkey) | |
888 | { | |
889 | return do_mprotect_pkey(start, len, prot, pkey); | |
890 | } | |
e8c24d3a DH |
891 | |
892 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) | |
893 | { | |
894 | int pkey; | |
895 | int ret; | |
896 | ||
897 | /* No flags supported yet. */ | |
898 | if (flags) | |
899 | return -EINVAL; | |
900 | /* check for unsupported init values */ | |
901 | if (init_val & ~PKEY_ACCESS_MASK) | |
902 | return -EINVAL; | |
903 | ||
d8ed45c5 | 904 | mmap_write_lock(current->mm); |
e8c24d3a DH |
905 | pkey = mm_pkey_alloc(current->mm); |
906 | ||
907 | ret = -ENOSPC; | |
908 | if (pkey == -1) | |
909 | goto out; | |
910 | ||
911 | ret = arch_set_user_pkey_access(current, pkey, init_val); | |
912 | if (ret) { | |
913 | mm_pkey_free(current->mm, pkey); | |
914 | goto out; | |
915 | } | |
916 | ret = pkey; | |
917 | out: | |
d8ed45c5 | 918 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
919 | return ret; |
920 | } | |
921 | ||
922 | SYSCALL_DEFINE1(pkey_free, int, pkey) | |
923 | { | |
924 | int ret; | |
925 | ||
d8ed45c5 | 926 | mmap_write_lock(current->mm); |
e8c24d3a | 927 | ret = mm_pkey_free(current->mm, pkey); |
d8ed45c5 | 928 | mmap_write_unlock(current->mm); |
e8c24d3a DH |
929 | |
930 | /* | |
f0953a1b | 931 | * We could provide warnings or errors if any VMA still |
e8c24d3a DH |
932 | * has the pkey set here. |
933 | */ | |
934 | return ret; | |
935 | } | |
c7142aea HC |
936 | |
937 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |