blob: 11936526b08b8c5f5c5d0454f6789008a0c6f313 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/mprotect.c
3 *
4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 *
Alan Cox046c6882009-01-05 14:06:29 +00007 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9 */
10
11#include <linux/mm.h>
12#include <linux/hugetlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/shm.h>
14#include <linux/mman.h>
15#include <linux/fs.h>
16#include <linux/highmem.h>
17#include <linux/security.h>
18#include <linux/mempolicy.h>
19#include <linux/personality.h>
20#include <linux/syscalls.h>
Christoph Lameter06972122006-06-23 02:03:35 -070021#include <linux/swap.h>
22#include <linux/swapops.h>
Andrea Arcangelicddb8a52008-07-28 15:46:29 -070023#include <linux/mmu_notifier.h>
KOSAKI Motohiro64cdd542009-01-06 14:39:16 -080024#include <linux/migrate.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020025#include <linux/perf_event.h>
Dave Hansene8c24d32016-07-29 09:30:15 -070026#include <linux/pkeys.h>
Mel Gorman64a9a342014-01-21 15:51:02 -080027#include <linux/ksm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
29#include <asm/pgtable.h>
30#include <asm/cacheflush.h>
Dave Hansene8c24d32016-07-29 09:30:15 -070031#include <asm/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/tlbflush.h>
33
Kirill A. Shutemov36f88182015-06-24 16:56:10 -070034#include "internal.h"
35
Mel Gorman1ad9f622014-04-07 15:36:56 -070036/*
37 * For a prot_numa update we only hold mmap_sem for read so there is a
38 * potential race with faulting where a pmd was temporarily none. This
39 * function checks for a transhuge pmd under the appropriate lock. It
40 * returns a pte if it was successfully locked or NULL if it raced with
41 * a transhuge insertion.
42 */
43static pte_t *lock_pte_protection(struct vm_area_struct *vma, pmd_t *pmd,
44 unsigned long addr, int prot_numa, spinlock_t **ptl)
45{
46 pte_t *pte;
47 spinlock_t *pmdl;
48
49 /* !prot_numa is protected by mmap_sem held for write */
50 if (!prot_numa)
51 return pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
52
53 pmdl = pmd_lock(vma->vm_mm, pmd);
54 if (unlikely(pmd_trans_huge(*pmd) || pmd_none(*pmd))) {
55 spin_unlock(pmdl);
56 return NULL;
57 }
58
59 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
60 spin_unlock(pmdl);
61 return pte;
62}
63
Mel Gorman4b10e7d2012-10-25 14:16:32 +020064static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070065 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +010066 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -070067{
Mel Gorman4b10e7d2012-10-25 14:16:32 +020068 struct mm_struct *mm = vma->vm_mm;
Christoph Lameter06972122006-06-23 02:03:35 -070069 pte_t *pte, oldpte;
Hugh Dickins705e87c2005-10-29 18:16:27 -070070 spinlock_t *ptl;
Peter Zijlstra7da4d642012-11-19 03:14:23 +010071 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Mel Gorman1ad9f622014-04-07 15:36:56 -070073 pte = lock_pte_protection(vma, pmd, addr, prot_numa, &ptl);
74 if (!pte)
75 return 0;
76
Zachary Amsden6606c3e2006-09-30 23:29:33 -070077 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 do {
Christoph Lameter06972122006-06-23 02:03:35 -070079 oldpte = *pte;
80 if (pte_present(oldpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 pte_t ptent;
Mel Gormanb191f9b2015-03-25 15:55:40 -070082 bool preserve_write = prot_numa && pte_write(oldpte);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Mel Gormane944fd62015-02-12 14:58:35 -080084 /*
85 * Avoid trapping faults against the zero or KSM
86 * pages. See similar comment in change_huge_pmd.
87 */
88 if (prot_numa) {
89 struct page *page;
90
91 page = vm_normal_page(vma, addr, oldpte);
92 if (!page || PageKsm(page))
93 continue;
Mel Gorman10c10452015-02-12 14:58:44 -080094
95 /* Avoid TLB flush if possible */
96 if (pte_protnone(oldpte))
97 continue;
Mel Gormane944fd62015-02-12 14:58:35 -080098 }
99
Mel Gorman8a0516e2015-02-12 14:58:22 -0800100 ptent = ptep_modify_prot_start(mm, addr, pte);
101 ptent = pte_modify(ptent, newprot);
Mel Gormanb191f9b2015-03-25 15:55:40 -0700102 if (preserve_write)
103 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200104
Mel Gorman8a0516e2015-02-12 14:58:22 -0800105 /* Avoid taking write faults for known dirty pages */
106 if (dirty_accountable && pte_dirty(ptent) &&
107 (pte_soft_dirty(ptent) ||
108 !(vma->vm_flags & VM_SOFTDIRTY))) {
109 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200110 }
Mel Gorman8a0516e2015-02-12 14:58:22 -0800111 ptep_modify_prot_commit(mm, addr, pte, ptent);
112 pages++;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800113 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700114 swp_entry_t entry = pte_to_swp_entry(oldpte);
115
116 if (is_write_migration_entry(entry)) {
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700117 pte_t newpte;
Christoph Lameter06972122006-06-23 02:03:35 -0700118 /*
119 * A protection check is difficult so
120 * just be safe and disable write
121 */
122 make_migration_entry_read(&entry);
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700123 newpte = swp_entry_to_pte(entry);
124 if (pte_swp_soft_dirty(oldpte))
125 newpte = pte_swp_mksoft_dirty(newpte);
126 set_pte_at(mm, addr, pte, newpte);
Mel Gormane920e142013-10-07 11:28:48 +0100127
128 pages++;
Christoph Lameter06972122006-06-23 02:03:35 -0700129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 }
131 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700132 arch_leave_lazy_mmu_mode();
Hugh Dickins705e87c2005-10-29 18:16:27 -0700133 pte_unmap_unlock(pte - 1, ptl);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100134
135 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
Andrew Morton7d12efa2012-12-18 14:23:17 -0800138static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
139 pud_t *pud, unsigned long addr, unsigned long end,
140 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 pmd_t *pmd;
Rik van Riela5338092014-04-07 15:36:57 -0700143 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100145 unsigned long pages = 0;
Mel Gorman72403b42013-11-12 15:08:32 -0800146 unsigned long nr_huge_updates = 0;
Rik van Riela5338092014-04-07 15:36:57 -0700147 unsigned long mni_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
149 pmd = pmd_offset(pud, addr);
150 do {
Mel Gorman25cbbef2013-10-07 11:29:14 +0100151 unsigned long this_pages;
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 next = pmd_addr_end(addr, end);
Dan Williams5c7fb562016-01-15 16:56:52 -0800154 if (!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd)
155 && pmd_none_or_clear_bad(pmd))
Rik van Riel88a9ab62014-04-07 15:36:55 -0700156 continue;
Rik van Riela5338092014-04-07 15:36:57 -0700157
158 /* invoke the mmu notifier if the pmd is populated */
159 if (!mni_start) {
160 mni_start = addr;
161 mmu_notifier_invalidate_range_start(mm, mni_start, end);
162 }
163
Dan Williams5c7fb562016-01-15 16:56:52 -0800164 if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800165 if (next - addr != HPAGE_PMD_SIZE) {
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -0800166 split_huge_pmd(vma, pmd, addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700167 if (pmd_trans_unstable(pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800168 continue;
169 } else {
Mel Gormanf123d742013-10-07 11:28:49 +0100170 int nr_ptes = change_huge_pmd(vma, pmd, addr,
Mel Gormane944fd62015-02-12 14:58:35 -0800171 newprot, prot_numa);
Mel Gormanf123d742013-10-07 11:28:49 +0100172
173 if (nr_ptes) {
Mel Gorman72403b42013-11-12 15:08:32 -0800174 if (nr_ptes == HPAGE_PMD_NR) {
175 pages += HPAGE_PMD_NR;
176 nr_huge_updates++;
177 }
Mel Gorman1ad9f622014-04-07 15:36:56 -0700178
179 /* huge pmd was handled */
Mel Gormanf123d742013-10-07 11:28:49 +0100180 continue;
181 }
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100182 }
Rik van Riel88a9ab62014-04-07 15:36:55 -0700183 /* fall through, the trans huge pmd just split */
Johannes Weinercd7548a2011-01-13 15:47:04 -0800184 }
Mel Gorman25cbbef2013-10-07 11:29:14 +0100185 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +0100186 dirty_accountable, prot_numa);
Mel Gorman25cbbef2013-10-07 11:29:14 +0100187 pages += this_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 } while (pmd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100189
Rik van Riela5338092014-04-07 15:36:57 -0700190 if (mni_start)
191 mmu_notifier_invalidate_range_end(mm, mni_start, end);
192
Mel Gorman72403b42013-11-12 15:08:32 -0800193 if (nr_huge_updates)
194 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100195 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196}
197
Andrew Morton7d12efa2012-12-18 14:23:17 -0800198static inline unsigned long change_pud_range(struct vm_area_struct *vma,
199 pgd_t *pgd, unsigned long addr, unsigned long end,
200 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
202 pud_t *pud;
203 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100204 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206 pud = pud_offset(pgd, addr);
207 do {
208 next = pud_addr_end(addr, end);
209 if (pud_none_or_clear_bad(pud))
210 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100211 pages += change_pmd_range(vma, pud, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200212 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 } while (pud++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100214
215 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100218static unsigned long change_protection_range(struct vm_area_struct *vma,
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700219 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200220 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221{
222 struct mm_struct *mm = vma->vm_mm;
223 pgd_t *pgd;
224 unsigned long next;
225 unsigned long start = addr;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100226 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 BUG_ON(addr >= end);
229 pgd = pgd_offset(mm, addr);
230 flush_cache_range(vma, addr, end);
Rik van Riel20841402013-12-18 17:08:44 -0800231 set_tlb_flush_pending(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 do {
233 next = pgd_addr_end(addr, end);
234 if (pgd_none_or_clear_bad(pgd))
235 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100236 pages += change_pud_range(vma, pgd, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200237 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 } while (pgd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100239
Ingo Molnar1233d582012-11-19 03:14:24 +0100240 /* Only flush the TLB if we actually modified any entries: */
241 if (pages)
242 flush_tlb_range(vma, start, end);
Rik van Riel20841402013-12-18 17:08:44 -0800243 clear_tlb_flush_pending(mm);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100244
245 return pages;
246}
247
248unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
249 unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200250 int dirty_accountable, int prot_numa)
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100251{
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100252 unsigned long pages;
253
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100254 if (is_vm_hugetlb_page(vma))
255 pages = hugetlb_change_protection(vma, start, end, newprot);
256 else
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200257 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100258
259 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
261
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700262int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
264 unsigned long start, unsigned long end, unsigned long newflags)
265{
266 struct mm_struct *mm = vma->vm_mm;
267 unsigned long oldflags = vma->vm_flags;
268 long nrpages = (end - start) >> PAGE_SHIFT;
269 unsigned long charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 pgoff_t pgoff;
271 int error;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700272 int dirty_accountable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 if (newflags == oldflags) {
275 *pprev = vma;
276 return 0;
277 }
278
279 /*
280 * If we make a private mapping writable we increase our commit;
281 * but (without finer accounting) cannot reduce our commit if we
Mel Gorman5a6fe122009-02-10 14:02:27 +0000282 * make it unwritable again. hugetlb mapping were accounted for
283 * even if read-only so there is no need to account for them here
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
285 if (newflags & VM_WRITE) {
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800286 /* Check space limits when area turns into data. */
287 if (!may_expand_vm(mm, newflags, nrpages) &&
288 may_expand_vm(mm, oldflags, nrpages))
289 return -ENOMEM;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000290 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
Andy Whitcroftcdfd4322008-07-23 21:27:28 -0700291 VM_SHARED|VM_NORESERVE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 charged = nrpages;
Al Viro191c5422012-02-13 03:58:52 +0000293 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 return -ENOMEM;
295 newflags |= VM_ACCOUNT;
296 }
297 }
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 /*
300 * First try to merge with previous and/or next vma.
301 */
302 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
303 *pprev = vma_merge(mm, *pprev, start, end, newflags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700304 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
305 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 if (*pprev) {
307 vma = *pprev;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700308 VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 goto success;
310 }
311
312 *pprev = vma;
313
314 if (start != vma->vm_start) {
315 error = split_vma(mm, vma, start, 1);
316 if (error)
317 goto fail;
318 }
319
320 if (end != vma->vm_end) {
321 error = split_vma(mm, vma, end, 0);
322 if (error)
323 goto fail;
324 }
325
326success:
327 /*
328 * vm_flags and vm_page_prot are protected by the mmap_sem
329 * held in write mode.
330 */
331 vma->vm_flags = newflags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700332 dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700333 vma_set_page_prot(vma);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700334
Andrew Morton7d12efa2012-12-18 14:23:17 -0800335 change_protection(vma, start, end, vma->vm_page_prot,
336 dirty_accountable, 0);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100337
Kirill A. Shutemov36f88182015-06-24 16:56:10 -0700338 /*
339 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
340 * fault on access.
341 */
342 if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
343 (newflags & VM_WRITE)) {
344 populate_vma_page_range(vma, start, end, NULL);
345 }
346
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800347 vm_stat_account(mm, oldflags, -nrpages);
348 vm_stat_account(mm, newflags, nrpages);
Pekka Enberg63bfd732010-11-08 21:29:07 +0200349 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 return 0;
351
352fail:
353 vm_unacct_memory(charged);
354 return error;
355}
356
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700357/*
358 * pkey==-1 when doing a legacy mprotect()
359 */
360static int do_mprotect_pkey(unsigned long start, size_t len,
361 unsigned long prot, int pkey)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Dave Hansen62b5f7d2016-02-12 13:02:40 -0800363 unsigned long nstart, end, tmp, reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 struct vm_area_struct *vma, *prev;
365 int error = -EINVAL;
366 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700367 const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
368 (prot & PROT_READ);
369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
371 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
372 return -EINVAL;
373
374 if (start & ~PAGE_MASK)
375 return -EINVAL;
376 if (!len)
377 return 0;
378 len = PAGE_ALIGN(len);
379 end = start + len;
380 if (end <= start)
381 return -ENOMEM;
Dave Kleikampb845f312008-07-08 00:28:51 +1000382 if (!arch_validate_prot(prot))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return -EINVAL;
384
385 reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700387 if (down_write_killable(&current->mm->mmap_sem))
388 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Dave Hansene8c24d32016-07-29 09:30:15 -0700390 /*
391 * If userspace did not allocate the pkey, do not let
392 * them use it here.
393 */
394 error = -EINVAL;
395 if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
396 goto out;
397
Linus Torvalds097d5912012-03-06 18:23:36 -0800398 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 error = -ENOMEM;
400 if (!vma)
401 goto out;
Linus Torvalds097d5912012-03-06 18:23:36 -0800402 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (unlikely(grows & PROT_GROWSDOWN)) {
404 if (vma->vm_start >= end)
405 goto out;
406 start = vma->vm_start;
407 error = -EINVAL;
408 if (!(vma->vm_flags & VM_GROWSDOWN))
409 goto out;
Andrew Morton7d12efa2012-12-18 14:23:17 -0800410 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (vma->vm_start > start)
412 goto out;
413 if (unlikely(grows & PROT_GROWSUP)) {
414 end = vma->vm_end;
415 error = -EINVAL;
416 if (!(vma->vm_flags & VM_GROWSUP))
417 goto out;
418 }
419 }
420 if (start > vma->vm_start)
421 prev = vma;
422
423 for (nstart = start ; ; ) {
Dave Hansena8502b62016-07-29 09:30:13 -0700424 unsigned long mask_off_old_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 unsigned long newflags;
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700426 int new_vma_pkey;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Andrew Morton7d12efa2012-12-18 14:23:17 -0800428 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700430 /* Does the application expect PROT_READ to imply PROT_EXEC */
431 if (rier && (vma->vm_flags & VM_MAYEXEC))
432 prot |= PROT_EXEC;
433
Dave Hansena8502b62016-07-29 09:30:13 -0700434 /*
435 * Each mprotect() call explicitly passes r/w/x permissions.
436 * If a permission is not passed to mprotect(), it must be
437 * cleared from the VMA.
438 */
439 mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC |
440 ARCH_VM_PKEY_FLAGS;
441
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700442 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
443 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
Dave Hansena8502b62016-07-29 09:30:13 -0700444 newflags |= (vma->vm_flags & ~mask_off_old_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Paolo 'Blaisorblade' Giarrusso7e2cff42005-09-21 09:55:39 -0700446 /* newflags >> 4 shift VM_MAY% in place of VM_% */
447 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 error = -EACCES;
449 goto out;
450 }
451
452 error = security_file_mprotect(vma, reqprot, prot);
453 if (error)
454 goto out;
455
456 tmp = vma->vm_end;
457 if (tmp > end)
458 tmp = end;
459 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
460 if (error)
461 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 nstart = tmp;
463
464 if (nstart < prev->vm_end)
465 nstart = prev->vm_end;
466 if (nstart >= end)
467 goto out;
468
469 vma = prev->vm_next;
470 if (!vma || vma->vm_start != nstart) {
471 error = -ENOMEM;
472 goto out;
473 }
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700474 prot = reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 }
476out:
477 up_write(&current->mm->mmap_sem);
478 return error;
479}
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700480
481SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
482 unsigned long, prot)
483{
484 return do_mprotect_pkey(start, len, prot, -1);
485}
486
487SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
488 unsigned long, prot, int, pkey)
489{
490 return do_mprotect_pkey(start, len, prot, pkey);
491}
Dave Hansene8c24d32016-07-29 09:30:15 -0700492
493SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
494{
495 int pkey;
496 int ret;
497
498 /* No flags supported yet. */
499 if (flags)
500 return -EINVAL;
501 /* check for unsupported init values */
502 if (init_val & ~PKEY_ACCESS_MASK)
503 return -EINVAL;
504
505 down_write(&current->mm->mmap_sem);
506 pkey = mm_pkey_alloc(current->mm);
507
508 ret = -ENOSPC;
509 if (pkey == -1)
510 goto out;
511
512 ret = arch_set_user_pkey_access(current, pkey, init_val);
513 if (ret) {
514 mm_pkey_free(current->mm, pkey);
515 goto out;
516 }
517 ret = pkey;
518out:
519 up_write(&current->mm->mmap_sem);
520 return ret;
521}
522
523SYSCALL_DEFINE1(pkey_free, int, pkey)
524{
525 int ret;
526
527 down_write(&current->mm->mmap_sem);
528 ret = mm_pkey_free(current->mm, pkey);
529 up_write(&current->mm->mmap_sem);
530
531 /*
532 * We could provie warnings or errors if any VMA still
533 * has the pkey set here.
534 */
535 return ret;
536}