1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * mm/mprotect.c
4 *
5 * (C) Copyright 1994 Linus Torvalds
6 * (C) Copyright 2002 Christoph Hellwig
7 *
8 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
9 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
10 */
11
12 #include <linux/pagewalk.h>
13 #include <linux/hugetlb.h>
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>
22 #include <linux/swap.h>
23 #include <linux/swapops.h>
24 #include <linux/mmu_notifier.h>
25 #include <linux/migrate.h>
26 #include <linux/perf_event.h>
27 #include <linux/pkeys.h>
28 #include <linux/ksm.h>
29 #include <linux/uaccess.h>
30 #include <linux/mm_inline.h>
31 #include <linux/pgtable.h>
32 #include <asm/cacheflush.h>
33 #include <asm/mmu_context.h>
34 #include <asm/tlbflush.h>
35 #include <asm/tlb.h>
36
37 #include "internal.h"
38
change_pte_range(struct mmu_gather * tlb,struct vm_area_struct * vma,pmd_t * pmd,unsigned long addr,unsigned long end,pgprot_t newprot,unsigned long cp_flags)39 static unsigned long change_pte_range(struct mmu_gather *tlb,
40 struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr,
41 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
42 {
43 pte_t *pte, oldpte;
44 spinlock_t *ptl;
45 unsigned long pages = 0;
46 int target_node = NUMA_NO_NODE;
47 bool dirty_accountable = cp_flags & MM_CP_DIRTY_ACCT;
48 bool prot_numa = cp_flags & MM_CP_PROT_NUMA;
49 bool uffd_wp = cp_flags & MM_CP_UFFD_WP;
50 bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE;
51
52 tlb_change_page_size(tlb, PAGE_SIZE);
53
54 /*
55 * Can be called with only the mmap_lock for reading by
56 * prot_numa so we must check the pmd isn't constantly
57 * changing from under us from pmd_none to pmd_trans_huge
58 * and/or the other way around.
59 */
60 if (pmd_trans_unstable(pmd))
61 return 0;
62
63 /*
64 * The pmd points to a regular pte so the pmd can't change
65 * from under us even if the mmap_lock is only hold for
66 * reading.
67 */
68 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
69
70 /* Get target node for single threaded private VMAs */
71 if (prot_numa && !(vma->vm_flags & VM_SHARED) &&
72 atomic_read(&vma->vm_mm->mm_users) == 1)
73 target_node = numa_node_id();
74
75 flush_tlb_batched_pending(vma->vm_mm);
76 arch_enter_lazy_mmu_mode();
77 do {
78 oldpte = *pte;
79 if (pte_present(oldpte)) {
80 pte_t ptent;
81 bool preserve_write = prot_numa && pte_write(oldpte);
82
83 /*
84 * Avoid trapping faults against the zero or KSM
85 * pages. See similar comment in change_huge_pmd.
86 */
87 if (prot_numa) {
88 struct page *page;
89
90 /* Avoid TLB flush if possible */
91 if (pte_protnone(oldpte))
92 continue;
93
94 page = vm_normal_page(vma, addr, oldpte);
95 if (!page || PageKsm(page))
96 continue;
97
98 /* Also skip shared copy-on-write pages */
99 if (is_cow_mapping(vma->vm_flags) &&
100 page_count(page) != 1)
101 continue;
102
103 /*
104 * While migration can move some dirty pages,
105 * it cannot move them all from MIGRATE_ASYNC
106 * context.
107 */
108 if (page_is_file_lru(page) && PageDirty(page))
109 continue;
110
111 /*
112 * Don't mess with PTEs if page is already on the node
113 * a single-threaded process is running on.
114 */
115 if (target_node == page_to_nid(page))
116 continue;
117 }
118
119 oldpte = ptep_modify_prot_start(vma, addr, pte);
120 ptent = pte_modify(oldpte, newprot);
121 if (preserve_write)
122 ptent = pte_mk_savedwrite(ptent);
123
124 if (uffd_wp) {
125 ptent = pte_wrprotect(ptent);
126 ptent = pte_mkuffd_wp(ptent);
127 } else if (uffd_wp_resolve) {
128 /*
129 * Leave the write bit to be handled
130 * by PF interrupt handler, then
131 * things like COW could be properly
132 * handled.
133 */
134 ptent = pte_clear_uffd_wp(ptent);
135 }
136
137 /* Avoid taking write faults for known dirty pages */
138 if (dirty_accountable && pte_dirty(ptent) &&
139 (pte_soft_dirty(ptent) ||
140 !(vma->vm_flags & VM_SOFTDIRTY))) {
141 ptent = pte_mkwrite(ptent);
142 }
143 ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent);
144 if (pte_needs_flush(oldpte, ptent))
145 tlb_flush_pte_range(tlb, addr, PAGE_SIZE);
146 pages++;
147 } else if (is_swap_pte(oldpte)) {
148 swp_entry_t entry = pte_to_swp_entry(oldpte);
149 pte_t newpte;
150
151 if (is_writable_migration_entry(entry)) {
152 /*
153 * A protection check is difficult so
154 * just be safe and disable write
155 */
156 entry = make_readable_migration_entry(
157 swp_offset(entry));
158 newpte = swp_entry_to_pte(entry);
159 if (pte_swp_soft_dirty(oldpte))
160 newpte = pte_swp_mksoft_dirty(newpte);
161 if (pte_swp_uffd_wp(oldpte))
162 newpte = pte_swp_mkuffd_wp(newpte);
163 } else if (is_writable_device_private_entry(entry)) {
164 /*
165 * We do not preserve soft-dirtiness. See
166 * copy_one_pte() for explanation.
167 */
168 entry = make_readable_device_private_entry(
169 swp_offset(entry));
170 newpte = swp_entry_to_pte(entry);
171 if (pte_swp_uffd_wp(oldpte))
172 newpte = pte_swp_mkuffd_wp(newpte);
173 } else if (is_writable_device_exclusive_entry(entry)) {
174 entry = make_readable_device_exclusive_entry(
175 swp_offset(entry));
176 newpte = swp_entry_to_pte(entry);
177 if (pte_swp_soft_dirty(oldpte))
178 newpte = pte_swp_mksoft_dirty(newpte);
179 if (pte_swp_uffd_wp(oldpte))
180 newpte = pte_swp_mkuffd_wp(newpte);
181 } else {
182 newpte = oldpte;
183 }
184
185 if (uffd_wp)
186 newpte = pte_swp_mkuffd_wp(newpte);
187 else if (uffd_wp_resolve)
188 newpte = pte_swp_clear_uffd_wp(newpte);
189
190 if (!pte_same(oldpte, newpte)) {
191 set_pte_at(vma->vm_mm, addr, pte, newpte);
192 pages++;
193 }
194 }
195 } while (pte++, addr += PAGE_SIZE, addr != end);
196 arch_leave_lazy_mmu_mode();
197 pte_unmap_unlock(pte - 1, ptl);
198
199 return pages;
200 }
201
202 /*
203 * Used when setting automatic NUMA hinting protection where it is
204 * critical that a numa hinting PMD is not confused with a bad PMD.
205 */
pmd_none_or_clear_bad_unless_trans_huge(pmd_t * pmd)206 static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd)
207 {
208 pmd_t pmdval = pmd_read_atomic(pmd);
209
210 /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */
211 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
212 barrier();
213 #endif
214
215 if (pmd_none(pmdval))
216 return 1;
217 if (pmd_trans_huge(pmdval))
218 return 0;
219 if (unlikely(pmd_bad(pmdval))) {
220 pmd_clear_bad(pmd);
221 return 1;
222 }
223
224 return 0;
225 }
226
change_pmd_range(struct mmu_gather * tlb,struct vm_area_struct * vma,pud_t * pud,unsigned long addr,unsigned long end,pgprot_t newprot,unsigned long cp_flags)227 static inline unsigned long change_pmd_range(struct mmu_gather *tlb,
228 struct vm_area_struct *vma, pud_t *pud, unsigned long addr,
229 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
230 {
231 pmd_t *pmd;
232 unsigned long next;
233 unsigned long pages = 0;
234 unsigned long nr_huge_updates = 0;
235 struct mmu_notifier_range range;
236
237 range.start = 0;
238
239 pmd = pmd_offset(pud, addr);
240 do {
241 unsigned long this_pages;
242
243 next = pmd_addr_end(addr, end);
244
245 /*
246 * Automatic NUMA balancing walks the tables with mmap_lock
247 * held for read. It's possible a parallel update to occur
248 * between pmd_trans_huge() and a pmd_none_or_clear_bad()
249 * check leading to a false positive and clearing.
250 * Hence, it's necessary to atomically read the PMD value
251 * for all the checks.
252 */
253 if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) &&
254 pmd_none_or_clear_bad_unless_trans_huge(pmd))
255 goto next;
256
257 /* invoke the mmu notifier if the pmd is populated */
258 if (!range.start) {
259 mmu_notifier_range_init(&range,
260 MMU_NOTIFY_PROTECTION_VMA, 0,
261 vma, vma->vm_mm, addr, end);
262 mmu_notifier_invalidate_range_start(&range);
263 }
264
265 if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
266 if (next - addr != HPAGE_PMD_SIZE) {
267 __split_huge_pmd(vma, pmd, addr, false, NULL);
268 } else {
269 /*
270 * change_huge_pmd() does not defer TLB flushes,
271 * so no need to propagate the tlb argument.
272 */
273 int nr_ptes = change_huge_pmd(tlb, vma, pmd,
274 addr, newprot, cp_flags);
275
276 if (nr_ptes) {
277 if (nr_ptes == HPAGE_PMD_NR) {
278 pages += HPAGE_PMD_NR;
279 nr_huge_updates++;
280 }
281
282 /* huge pmd was handled */
283 goto next;
284 }
285 }
286 /* fall through, the trans huge pmd just split */
287 }
288 this_pages = change_pte_range(tlb, vma, pmd, addr, next,
289 newprot, cp_flags);
290 pages += this_pages;
291 next:
292 cond_resched();
293 } while (pmd++, addr = next, addr != end);
294
295 if (range.start)
296 mmu_notifier_invalidate_range_end(&range);
297
298 if (nr_huge_updates)
299 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
300 return pages;
301 }
302
change_pud_range(struct mmu_gather * tlb,struct vm_area_struct * vma,p4d_t * p4d,unsigned long addr,unsigned long end,pgprot_t newprot,unsigned long cp_flags)303 static inline unsigned long change_pud_range(struct mmu_gather *tlb,
304 struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr,
305 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
306 {
307 pud_t *pud;
308 unsigned long next;
309 unsigned long pages = 0;
310
311 pud = pud_offset(p4d, addr);
312 do {
313 next = pud_addr_end(addr, end);
314 if (pud_none_or_clear_bad(pud))
315 continue;
316 pages += change_pmd_range(tlb, vma, pud, addr, next, newprot,
317 cp_flags);
318 } while (pud++, addr = next, addr != end);
319
320 return pages;
321 }
322
change_p4d_range(struct mmu_gather * tlb,struct vm_area_struct * vma,pgd_t * pgd,unsigned long addr,unsigned long end,pgprot_t newprot,unsigned long cp_flags)323 static inline unsigned long change_p4d_range(struct mmu_gather *tlb,
324 struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr,
325 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
326 {
327 p4d_t *p4d;
328 unsigned long next;
329 unsigned long pages = 0;
330
331 p4d = p4d_offset(pgd, addr);
332 do {
333 next = p4d_addr_end(addr, end);
334 if (p4d_none_or_clear_bad(p4d))
335 continue;
336 pages += change_pud_range(tlb, vma, p4d, addr, next, newprot,
337 cp_flags);
338 } while (p4d++, addr = next, addr != end);
339
340 return pages;
341 }
342
change_protection_range(struct mmu_gather * tlb,struct vm_area_struct * vma,unsigned long addr,unsigned long end,pgprot_t newprot,unsigned long cp_flags)343 static unsigned long change_protection_range(struct mmu_gather *tlb,
344 struct vm_area_struct *vma, unsigned long addr,
345 unsigned long end, pgprot_t newprot, unsigned long cp_flags)
346 {
347 struct mm_struct *mm = vma->vm_mm;
348 pgd_t *pgd;
349 unsigned long next;
350 unsigned long pages = 0;
351
352 BUG_ON(addr >= end);
353 pgd = pgd_offset(mm, addr);
354 tlb_start_vma(tlb, vma);
355 do {
356 next = pgd_addr_end(addr, end);
357 if (pgd_none_or_clear_bad(pgd))
358 continue;
359 pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot,
360 cp_flags);
361 } while (pgd++, addr = next, addr != end);
362
363 tlb_end_vma(tlb, vma);
364
365 return pages;
366 }
367
change_protection(struct mmu_gather * tlb,struct vm_area_struct * vma,unsigned long start,unsigned long end,pgprot_t newprot,unsigned long cp_flags)368 unsigned long change_protection(struct mmu_gather *tlb,
369 struct vm_area_struct *vma, unsigned long start,
370 unsigned long end, pgprot_t newprot,
371 unsigned long cp_flags)
372 {
373 unsigned long pages;
374
375 BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL);
376
377 if (is_vm_hugetlb_page(vma))
378 pages = hugetlb_change_protection(vma, start, end, newprot);
379 else
380 pages = change_protection_range(tlb, vma, start, end, newprot,
381 cp_flags);
382
383 return pages;
384 }
385
prot_none_pte_entry(pte_t * pte,unsigned long addr,unsigned long next,struct mm_walk * walk)386 static int prot_none_pte_entry(pte_t *pte, unsigned long addr,
387 unsigned long next, struct mm_walk *walk)
388 {
389 return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ?
390 0 : -EACCES;
391 }
392
prot_none_hugetlb_entry(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long next,struct mm_walk * walk)393 static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask,
394 unsigned long addr, unsigned long next,
395 struct mm_walk *walk)
396 {
397 return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ?
398 0 : -EACCES;
399 }
400
prot_none_test(unsigned long addr,unsigned long next,struct mm_walk * walk)401 static int prot_none_test(unsigned long addr, unsigned long next,
402 struct mm_walk *walk)
403 {
404 return 0;
405 }
406
407 static const struct mm_walk_ops prot_none_walk_ops = {
408 .pte_entry = prot_none_pte_entry,
409 .hugetlb_entry = prot_none_hugetlb_entry,
410 .test_walk = prot_none_test,
411 };
412
413 int
mprotect_fixup(struct mmu_gather * tlb,struct vm_area_struct * vma,struct vm_area_struct ** pprev,unsigned long start,unsigned long end,unsigned long newflags)414 mprotect_fixup(struct mmu_gather *tlb, struct vm_area_struct *vma,
415 struct vm_area_struct **pprev, unsigned long start,
416 unsigned long end, unsigned long newflags)
417 {
418 struct mm_struct *mm = vma->vm_mm;
419 unsigned long oldflags = vma->vm_flags;
420 long nrpages = (end - start) >> PAGE_SHIFT;
421 unsigned long charged = 0;
422 pgoff_t pgoff;
423 int error;
424 int dirty_accountable = 0;
425
426 if (newflags == oldflags) {
427 *pprev = vma;
428 return 0;
429 }
430
431 /*
432 * Do PROT_NONE PFN permission checks here when we can still
433 * bail out without undoing a lot of state. This is a rather
434 * uncommon case, so doesn't need to be very optimized.
435 */
436 if (arch_has_pfn_modify_check() &&
437 (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) &&
438 (newflags & VM_ACCESS_FLAGS) == 0) {
439 pgprot_t new_pgprot = vm_get_page_prot(newflags);
440
441 error = walk_page_range(current->mm, start, end,
442 &prot_none_walk_ops, &new_pgprot);
443 if (error)
444 return error;
445 }
446
447 /*
448 * If we make a private mapping writable we increase our commit;
449 * but (without finer accounting) cannot reduce our commit if we
450 * make it unwritable again. hugetlb mapping were accounted for
451 * even if read-only so there is no need to account for them here
452 */
453 if (newflags & VM_WRITE) {
454 /* Check space limits when area turns into data. */
455 if (!may_expand_vm(mm, newflags, nrpages) &&
456 may_expand_vm(mm, oldflags, nrpages))
457 return -ENOMEM;
458 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
459 VM_SHARED|VM_NORESERVE))) {
460 charged = nrpages;
461 if (security_vm_enough_memory_mm(mm, charged))
462 return -ENOMEM;
463 newflags |= VM_ACCOUNT;
464 }
465 }
466
467 /*
468 * First try to merge with previous and/or next vma.
469 */
470 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
471 *pprev = vma_merge(mm, *pprev, start, end, newflags,
472 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
473 vma->vm_userfaultfd_ctx, anon_vma_name(vma));
474 if (*pprev) {
475 vma = *pprev;
476 VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY);
477 goto success;
478 }
479
480 *pprev = vma;
481
482 if (start != vma->vm_start) {
483 error = split_vma(mm, vma, start, 1);
484 if (error)
485 goto fail;
486 }
487
488 if (end != vma->vm_end) {
489 error = split_vma(mm, vma, end, 0);
490 if (error)
491 goto fail;
492 }
493
494 success:
495 /*
496 * vm_flags and vm_page_prot are protected by the mmap_lock
497 * held in write mode.
498 */
499 vma->vm_flags = newflags;
500 dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
501 vma_set_page_prot(vma);
502
503 change_protection(tlb, vma, start, end, vma->vm_page_prot,
504 dirty_accountable ? MM_CP_DIRTY_ACCT : 0);
505
506 /*
507 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
508 * fault on access.
509 */
510 if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
511 (newflags & VM_WRITE)) {
512 populate_vma_page_range(vma, start, end, NULL);
513 }
514
515 vm_stat_account(mm, oldflags, -nrpages);
516 vm_stat_account(mm, newflags, nrpages);
517 perf_event_mmap(vma);
518 return 0;
519
520 fail:
521 vm_unacct_memory(charged);
522 return error;
523 }
524
525 /*
526 * pkey==-1 when doing a legacy mprotect()
527 */
do_mprotect_pkey(unsigned long start,size_t len,unsigned long prot,int pkey)528 static int do_mprotect_pkey(unsigned long start, size_t len,
529 unsigned long prot, int pkey)
530 {
531 unsigned long nstart, end, tmp, reqprot;
532 struct vm_area_struct *vma, *prev;
533 int error = -EINVAL;
534 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
535 const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
536 (prot & PROT_READ);
537 struct mmu_gather tlb;
538
539 start = untagged_addr(start);
540
541 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
542 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
543 return -EINVAL;
544
545 if (start & ~PAGE_MASK)
546 return -EINVAL;
547 if (!len)
548 return 0;
549 len = PAGE_ALIGN(len);
550 end = start + len;
551 if (end <= start)
552 return -ENOMEM;
553 if (!arch_validate_prot(prot, start))
554 return -EINVAL;
555
556 reqprot = prot;
557
558 if (mmap_write_lock_killable(current->mm))
559 return -EINTR;
560
561 /*
562 * If userspace did not allocate the pkey, do not let
563 * them use it here.
564 */
565 error = -EINVAL;
566 if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
567 goto out;
568
569 vma = find_vma(current->mm, start);
570 error = -ENOMEM;
571 if (!vma)
572 goto out;
573 prev = vma->vm_prev;
574 if (unlikely(grows & PROT_GROWSDOWN)) {
575 if (vma->vm_start >= end)
576 goto out;
577 start = vma->vm_start;
578 error = -EINVAL;
579 if (!(vma->vm_flags & VM_GROWSDOWN))
580 goto out;
581 } else {
582 if (vma->vm_start > start)
583 goto out;
584 if (unlikely(grows & PROT_GROWSUP)) {
585 end = vma->vm_end;
586 error = -EINVAL;
587 if (!(vma->vm_flags & VM_GROWSUP))
588 goto out;
589 }
590 }
591 if (start > vma->vm_start)
592 prev = vma;
593
594 tlb_gather_mmu(&tlb, current->mm);
595 for (nstart = start ; ; ) {
596 unsigned long mask_off_old_flags;
597 unsigned long newflags;
598 int new_vma_pkey;
599
600 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
601
602 /* Does the application expect PROT_READ to imply PROT_EXEC */
603 if (rier && (vma->vm_flags & VM_MAYEXEC))
604 prot |= PROT_EXEC;
605
606 /*
607 * Each mprotect() call explicitly passes r/w/x permissions.
608 * If a permission is not passed to mprotect(), it must be
609 * cleared from the VMA.
610 */
611 mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC |
612 VM_FLAGS_CLEAR;
613
614 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
615 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
616 newflags |= (vma->vm_flags & ~mask_off_old_flags);
617
618 /* newflags >> 4 shift VM_MAY% in place of VM_% */
619 if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) {
620 error = -EACCES;
621 break;
622 }
623
624 /* Allow architectures to sanity-check the new flags */
625 if (!arch_validate_flags(newflags)) {
626 error = -EINVAL;
627 break;
628 }
629
630 error = security_file_mprotect(vma, reqprot, prot);
631 if (error)
632 break;
633
634 tmp = vma->vm_end;
635 if (tmp > end)
636 tmp = end;
637
638 if (vma->vm_ops && vma->vm_ops->mprotect) {
639 error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags);
640 if (error)
641 break;
642 }
643
644 error = mprotect_fixup(&tlb, vma, &prev, nstart, tmp, newflags);
645 if (error)
646 break;
647
648 nstart = tmp;
649
650 if (nstart < prev->vm_end)
651 nstart = prev->vm_end;
652 if (nstart >= end)
653 break;
654
655 vma = prev->vm_next;
656 if (!vma || vma->vm_start != nstart) {
657 error = -ENOMEM;
658 break;
659 }
660 prot = reqprot;
661 }
662 tlb_finish_mmu(&tlb);
663 out:
664 mmap_write_unlock(current->mm);
665 return error;
666 }
667
SYSCALL_DEFINE3(mprotect,unsigned long,start,size_t,len,unsigned long,prot)668 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
669 unsigned long, prot)
670 {
671 return do_mprotect_pkey(start, len, prot, -1);
672 }
673
674 #ifdef CONFIG_ARCH_HAS_PKEYS
675
SYSCALL_DEFINE4(pkey_mprotect,unsigned long,start,size_t,len,unsigned long,prot,int,pkey)676 SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
677 unsigned long, prot, int, pkey)
678 {
679 return do_mprotect_pkey(start, len, prot, pkey);
680 }
681
SYSCALL_DEFINE2(pkey_alloc,unsigned long,flags,unsigned long,init_val)682 SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
683 {
684 int pkey;
685 int ret;
686
687 /* No flags supported yet. */
688 if (flags)
689 return -EINVAL;
690 /* check for unsupported init values */
691 if (init_val & ~PKEY_ACCESS_MASK)
692 return -EINVAL;
693
694 mmap_write_lock(current->mm);
695 pkey = mm_pkey_alloc(current->mm);
696
697 ret = -ENOSPC;
698 if (pkey == -1)
699 goto out;
700
701 ret = arch_set_user_pkey_access(current, pkey, init_val);
702 if (ret) {
703 mm_pkey_free(current->mm, pkey);
704 goto out;
705 }
706 ret = pkey;
707 out:
708 mmap_write_unlock(current->mm);
709 return ret;
710 }
711
SYSCALL_DEFINE1(pkey_free,int,pkey)712 SYSCALL_DEFINE1(pkey_free, int, pkey)
713 {
714 int ret;
715
716 mmap_write_lock(current->mm);
717 ret = mm_pkey_free(current->mm, pkey);
718 mmap_write_unlock(current->mm);
719
720 /*
721 * We could provide warnings or errors if any VMA still
722 * has the pkey set here.
723 */
724 return ret;
725 }
726
727 #endif /* CONFIG_ARCH_HAS_PKEYS */
728