blob: 8edd0d576254d4c6a3974a42dd2a27eff17fa8bf [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 Torvalds7c0f6ba2016-12-24 11:46:01 -080028#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#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 Gorman4b10e7d2012-10-25 14:16:32 +020036static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070037 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +010038 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039{
Mel Gorman4b10e7d2012-10-25 14:16:32 +020040 struct mm_struct *mm = vma->vm_mm;
Christoph Lameter06972122006-06-23 02:03:35 -070041 pte_t *pte, oldpte;
Hugh Dickins705e87c2005-10-29 18:16:27 -070042 spinlock_t *ptl;
Peter Zijlstra7da4d642012-11-19 03:14:23 +010043 unsigned long pages = 0;
Andi Kleen3e321582016-12-12 16:41:47 -080044 int target_node = NUMA_NO_NODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Andrea Arcangeli175ad4f2017-02-22 15:44:12 -080046 /*
47 * Can be called with only the mmap_sem for reading by
48 * prot_numa so we must check the pmd isn't constantly
49 * changing from under us from pmd_none to pmd_trans_huge
50 * and/or the other way around.
51 */
52 if (pmd_trans_unstable(pmd))
53 return 0;
54
55 /*
56 * The pmd points to a regular pte so the pmd can't change
57 * from under us even if the mmap_sem is only hold for
58 * reading.
59 */
60 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Mel Gorman1ad9f622014-04-07 15:36:56 -070061 if (!pte)
62 return 0;
63
Andi Kleen3e321582016-12-12 16:41:47 -080064 /* Get target node for single threaded private VMAs */
65 if (prot_numa && !(vma->vm_flags & VM_SHARED) &&
66 atomic_read(&vma->vm_mm->mm_users) == 1)
67 target_node = numa_node_id();
68
Zachary Amsden6606c3e2006-09-30 23:29:33 -070069 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 do {
Christoph Lameter06972122006-06-23 02:03:35 -070071 oldpte = *pte;
72 if (pte_present(oldpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 pte_t ptent;
Mel Gormanb191f9b2015-03-25 15:55:40 -070074 bool preserve_write = prot_numa && pte_write(oldpte);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Mel Gormane944fd62015-02-12 14:58:35 -080076 /*
77 * Avoid trapping faults against the zero or KSM
78 * pages. See similar comment in change_huge_pmd.
79 */
80 if (prot_numa) {
81 struct page *page;
82
83 page = vm_normal_page(vma, addr, oldpte);
84 if (!page || PageKsm(page))
85 continue;
Mel Gorman10c10452015-02-12 14:58:44 -080086
87 /* Avoid TLB flush if possible */
88 if (pte_protnone(oldpte))
89 continue;
Andi Kleen3e321582016-12-12 16:41:47 -080090
91 /*
92 * Don't mess with PTEs if page is already on the node
93 * a single-threaded process is running on.
94 */
95 if (target_node == page_to_nid(page))
96 continue;
Mel Gormane944fd62015-02-12 14:58:35 -080097 }
98
Mel Gorman8a0516e2015-02-12 14:58:22 -080099 ptent = ptep_modify_prot_start(mm, addr, pte);
100 ptent = pte_modify(ptent, newprot);
Mel Gormanb191f9b2015-03-25 15:55:40 -0700101 if (preserve_write)
Aneesh Kumar K.V288bc542017-02-24 14:59:16 -0800102 ptent = pte_mk_savedwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200103
Mel Gorman8a0516e2015-02-12 14:58:22 -0800104 /* Avoid taking write faults for known dirty pages */
105 if (dirty_accountable && pte_dirty(ptent) &&
106 (pte_soft_dirty(ptent) ||
107 !(vma->vm_flags & VM_SOFTDIRTY))) {
108 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200109 }
Mel Gorman8a0516e2015-02-12 14:58:22 -0800110 ptep_modify_prot_commit(mm, addr, pte, ptent);
111 pages++;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800112 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700113 swp_entry_t entry = pte_to_swp_entry(oldpte);
114
115 if (is_write_migration_entry(entry)) {
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700116 pte_t newpte;
Christoph Lameter06972122006-06-23 02:03:35 -0700117 /*
118 * A protection check is difficult so
119 * just be safe and disable write
120 */
121 make_migration_entry_read(&entry);
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700122 newpte = swp_entry_to_pte(entry);
123 if (pte_swp_soft_dirty(oldpte))
124 newpte = pte_swp_mksoft_dirty(newpte);
125 set_pte_at(mm, addr, pte, newpte);
Mel Gormane920e142013-10-07 11:28:48 +0100126
127 pages++;
Christoph Lameter06972122006-06-23 02:03:35 -0700128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
130 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700131 arch_leave_lazy_mmu_mode();
Hugh Dickins705e87c2005-10-29 18:16:27 -0700132 pte_unmap_unlock(pte - 1, ptl);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100133
134 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
Andrew Morton7d12efa2012-12-18 14:23:17 -0800137static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
138 pud_t *pud, unsigned long addr, unsigned long end,
139 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
141 pmd_t *pmd;
Rik van Riela5338092014-04-07 15:36:57 -0700142 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100144 unsigned long pages = 0;
Mel Gorman72403b42013-11-12 15:08:32 -0800145 unsigned long nr_huge_updates = 0;
Rik van Riela5338092014-04-07 15:36:57 -0700146 unsigned long mni_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148 pmd = pmd_offset(pud, addr);
149 do {
Mel Gorman25cbbef2013-10-07 11:29:14 +0100150 unsigned long this_pages;
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 next = pmd_addr_end(addr, end);
Dan Williams5c7fb562016-01-15 16:56:52 -0800153 if (!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd)
154 && pmd_none_or_clear_bad(pmd))
Rik van Riel88a9ab62014-04-07 15:36:55 -0700155 continue;
Rik van Riela5338092014-04-07 15:36:57 -0700156
157 /* invoke the mmu notifier if the pmd is populated */
158 if (!mni_start) {
159 mni_start = addr;
160 mmu_notifier_invalidate_range_start(mm, mni_start, end);
161 }
162
Dan Williams5c7fb562016-01-15 16:56:52 -0800163 if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800164 if (next - addr != HPAGE_PMD_SIZE) {
David Rientjesfd607752016-12-12 16:42:20 -0800165 __split_huge_pmd(vma, pmd, addr, false, NULL);
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800166 } else {
Mel Gormanf123d742013-10-07 11:28:49 +0100167 int nr_ptes = change_huge_pmd(vma, pmd, addr,
Mel Gormane944fd62015-02-12 14:58:35 -0800168 newprot, prot_numa);
Mel Gormanf123d742013-10-07 11:28:49 +0100169
170 if (nr_ptes) {
Mel Gorman72403b42013-11-12 15:08:32 -0800171 if (nr_ptes == HPAGE_PMD_NR) {
172 pages += HPAGE_PMD_NR;
173 nr_huge_updates++;
174 }
Mel Gorman1ad9f622014-04-07 15:36:56 -0700175
176 /* huge pmd was handled */
Mel Gormanf123d742013-10-07 11:28:49 +0100177 continue;
178 }
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100179 }
Rik van Riel88a9ab62014-04-07 15:36:55 -0700180 /* fall through, the trans huge pmd just split */
Johannes Weinercd7548a2011-01-13 15:47:04 -0800181 }
Mel Gorman25cbbef2013-10-07 11:29:14 +0100182 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +0100183 dirty_accountable, prot_numa);
Mel Gorman25cbbef2013-10-07 11:29:14 +0100184 pages += this_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 } while (pmd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100186
Rik van Riela5338092014-04-07 15:36:57 -0700187 if (mni_start)
188 mmu_notifier_invalidate_range_end(mm, mni_start, end);
189
Mel Gorman72403b42013-11-12 15:08:32 -0800190 if (nr_huge_updates)
191 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100192 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
Andrew Morton7d12efa2012-12-18 14:23:17 -0800195static inline unsigned long change_pud_range(struct vm_area_struct *vma,
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300196 p4d_t *p4d, unsigned long addr, unsigned long end,
Andrew Morton7d12efa2012-12-18 14:23:17 -0800197 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
199 pud_t *pud;
200 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100201 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300203 pud = pud_offset(p4d, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 do {
205 next = pud_addr_end(addr, end);
206 if (pud_none_or_clear_bad(pud))
207 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100208 pages += change_pmd_range(vma, pud, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200209 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 } while (pud++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100211
212 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300215static inline unsigned long change_p4d_range(struct vm_area_struct *vma,
216 pgd_t *pgd, unsigned long addr, unsigned long end,
217 pgprot_t newprot, int dirty_accountable, int prot_numa)
218{
219 p4d_t *p4d;
220 unsigned long next;
221 unsigned long pages = 0;
222
223 p4d = p4d_offset(pgd, addr);
224 do {
225 next = p4d_addr_end(addr, end);
226 if (p4d_none_or_clear_bad(p4d))
227 continue;
228 pages += change_pud_range(vma, p4d, addr, next, newprot,
229 dirty_accountable, prot_numa);
230 } while (p4d++, addr = next, addr != end);
231
232 return pages;
233}
234
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100235static unsigned long change_protection_range(struct vm_area_struct *vma,
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700236 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200237 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
239 struct mm_struct *mm = vma->vm_mm;
240 pgd_t *pgd;
241 unsigned long next;
242 unsigned long start = addr;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100243 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 BUG_ON(addr >= end);
246 pgd = pgd_offset(mm, addr);
247 flush_cache_range(vma, addr, end);
Rik van Riel20841402013-12-18 17:08:44 -0800248 set_tlb_flush_pending(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 do {
250 next = pgd_addr_end(addr, end);
251 if (pgd_none_or_clear_bad(pgd))
252 continue;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300253 pages += change_p4d_range(vma, pgd, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200254 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 } while (pgd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100256
Ingo Molnar1233d582012-11-19 03:14:24 +0100257 /* Only flush the TLB if we actually modified any entries: */
258 if (pages)
259 flush_tlb_range(vma, start, end);
Rik van Riel20841402013-12-18 17:08:44 -0800260 clear_tlb_flush_pending(mm);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100261
262 return pages;
263}
264
265unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
266 unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200267 int dirty_accountable, int prot_numa)
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100268{
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100269 unsigned long pages;
270
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100271 if (is_vm_hugetlb_page(vma))
272 pages = hugetlb_change_protection(vma, start, end, newprot);
273 else
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200274 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100275
276 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700279int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
281 unsigned long start, unsigned long end, unsigned long newflags)
282{
283 struct mm_struct *mm = vma->vm_mm;
284 unsigned long oldflags = vma->vm_flags;
285 long nrpages = (end - start) >> PAGE_SHIFT;
286 unsigned long charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 pgoff_t pgoff;
288 int error;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700289 int dirty_accountable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 if (newflags == oldflags) {
292 *pprev = vma;
293 return 0;
294 }
295
296 /*
297 * If we make a private mapping writable we increase our commit;
298 * but (without finer accounting) cannot reduce our commit if we
Mel Gorman5a6fe122009-02-10 14:02:27 +0000299 * make it unwritable again. hugetlb mapping were accounted for
300 * even if read-only so there is no need to account for them here
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 */
302 if (newflags & VM_WRITE) {
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800303 /* Check space limits when area turns into data. */
304 if (!may_expand_vm(mm, newflags, nrpages) &&
305 may_expand_vm(mm, oldflags, nrpages))
306 return -ENOMEM;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000307 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
Andy Whitcroftcdfd4322008-07-23 21:27:28 -0700308 VM_SHARED|VM_NORESERVE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 charged = nrpages;
Al Viro191c5422012-02-13 03:58:52 +0000310 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 return -ENOMEM;
312 newflags |= VM_ACCOUNT;
313 }
314 }
315
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 /*
317 * First try to merge with previous and/or next vma.
318 */
319 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
320 *pprev = vma_merge(mm, *pprev, start, end, newflags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700321 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
322 vma->vm_userfaultfd_ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 if (*pprev) {
324 vma = *pprev;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700325 VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 goto success;
327 }
328
329 *pprev = vma;
330
331 if (start != vma->vm_start) {
332 error = split_vma(mm, vma, start, 1);
333 if (error)
334 goto fail;
335 }
336
337 if (end != vma->vm_end) {
338 error = split_vma(mm, vma, end, 0);
339 if (error)
340 goto fail;
341 }
342
343success:
344 /*
345 * vm_flags and vm_page_prot are protected by the mmap_sem
346 * held in write mode.
347 */
348 vma->vm_flags = newflags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700349 dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700350 vma_set_page_prot(vma);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700351
Andrew Morton7d12efa2012-12-18 14:23:17 -0800352 change_protection(vma, start, end, vma->vm_page_prot,
353 dirty_accountable, 0);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100354
Kirill A. Shutemov36f88182015-06-24 16:56:10 -0700355 /*
356 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
357 * fault on access.
358 */
359 if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
360 (newflags & VM_WRITE)) {
361 populate_vma_page_range(vma, start, end, NULL);
362 }
363
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800364 vm_stat_account(mm, oldflags, -nrpages);
365 vm_stat_account(mm, newflags, nrpages);
Pekka Enberg63bfd732010-11-08 21:29:07 +0200366 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 return 0;
368
369fail:
370 vm_unacct_memory(charged);
371 return error;
372}
373
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700374/*
375 * pkey==-1 when doing a legacy mprotect()
376 */
377static int do_mprotect_pkey(unsigned long start, size_t len,
378 unsigned long prot, int pkey)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Dave Hansen62b5f7d2016-02-12 13:02:40 -0800380 unsigned long nstart, end, tmp, reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 struct vm_area_struct *vma, *prev;
382 int error = -EINVAL;
383 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700384 const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
385 (prot & PROT_READ);
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
388 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
389 return -EINVAL;
390
391 if (start & ~PAGE_MASK)
392 return -EINVAL;
393 if (!len)
394 return 0;
395 len = PAGE_ALIGN(len);
396 end = start + len;
397 if (end <= start)
398 return -ENOMEM;
Dave Kleikampb845f312008-07-08 00:28:51 +1000399 if (!arch_validate_prot(prot))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return -EINVAL;
401
402 reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700404 if (down_write_killable(&current->mm->mmap_sem))
405 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Dave Hansene8c24d32016-07-29 09:30:15 -0700407 /*
408 * If userspace did not allocate the pkey, do not let
409 * them use it here.
410 */
411 error = -EINVAL;
412 if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
413 goto out;
414
Linus Torvalds097d5912012-03-06 18:23:36 -0800415 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 error = -ENOMEM;
417 if (!vma)
418 goto out;
Linus Torvalds097d5912012-03-06 18:23:36 -0800419 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (unlikely(grows & PROT_GROWSDOWN)) {
421 if (vma->vm_start >= end)
422 goto out;
423 start = vma->vm_start;
424 error = -EINVAL;
425 if (!(vma->vm_flags & VM_GROWSDOWN))
426 goto out;
Andrew Morton7d12efa2012-12-18 14:23:17 -0800427 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 if (vma->vm_start > start)
429 goto out;
430 if (unlikely(grows & PROT_GROWSUP)) {
431 end = vma->vm_end;
432 error = -EINVAL;
433 if (!(vma->vm_flags & VM_GROWSUP))
434 goto out;
435 }
436 }
437 if (start > vma->vm_start)
438 prev = vma;
439
440 for (nstart = start ; ; ) {
Dave Hansena8502b62016-07-29 09:30:13 -0700441 unsigned long mask_off_old_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 unsigned long newflags;
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700443 int new_vma_pkey;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Andrew Morton7d12efa2012-12-18 14:23:17 -0800445 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700447 /* Does the application expect PROT_READ to imply PROT_EXEC */
448 if (rier && (vma->vm_flags & VM_MAYEXEC))
449 prot |= PROT_EXEC;
450
Dave Hansena8502b62016-07-29 09:30:13 -0700451 /*
452 * Each mprotect() call explicitly passes r/w/x permissions.
453 * If a permission is not passed to mprotect(), it must be
454 * cleared from the VMA.
455 */
456 mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC |
457 ARCH_VM_PKEY_FLAGS;
458
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700459 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
460 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
Dave Hansena8502b62016-07-29 09:30:13 -0700461 newflags |= (vma->vm_flags & ~mask_off_old_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Paolo 'Blaisorblade' Giarrusso7e2cff42005-09-21 09:55:39 -0700463 /* newflags >> 4 shift VM_MAY% in place of VM_% */
464 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 error = -EACCES;
466 goto out;
467 }
468
469 error = security_file_mprotect(vma, reqprot, prot);
470 if (error)
471 goto out;
472
473 tmp = vma->vm_end;
474 if (tmp > end)
475 tmp = end;
476 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
477 if (error)
478 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 nstart = tmp;
480
481 if (nstart < prev->vm_end)
482 nstart = prev->vm_end;
483 if (nstart >= end)
484 goto out;
485
486 vma = prev->vm_next;
487 if (!vma || vma->vm_start != nstart) {
488 error = -ENOMEM;
489 goto out;
490 }
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700491 prot = reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 }
493out:
494 up_write(&current->mm->mmap_sem);
495 return error;
496}
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700497
498SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
499 unsigned long, prot)
500{
501 return do_mprotect_pkey(start, len, prot, -1);
502}
503
Heiko Carstensc7142ae2016-12-12 16:43:09 -0800504#ifdef CONFIG_ARCH_HAS_PKEYS
505
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700506SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
507 unsigned long, prot, int, pkey)
508{
509 return do_mprotect_pkey(start, len, prot, pkey);
510}
Dave Hansene8c24d32016-07-29 09:30:15 -0700511
512SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
513{
514 int pkey;
515 int ret;
516
517 /* No flags supported yet. */
518 if (flags)
519 return -EINVAL;
520 /* check for unsupported init values */
521 if (init_val & ~PKEY_ACCESS_MASK)
522 return -EINVAL;
523
524 down_write(&current->mm->mmap_sem);
525 pkey = mm_pkey_alloc(current->mm);
526
527 ret = -ENOSPC;
528 if (pkey == -1)
529 goto out;
530
531 ret = arch_set_user_pkey_access(current, pkey, init_val);
532 if (ret) {
533 mm_pkey_free(current->mm, pkey);
534 goto out;
535 }
536 ret = pkey;
537out:
538 up_write(&current->mm->mmap_sem);
539 return ret;
540}
541
542SYSCALL_DEFINE1(pkey_free, int, pkey)
543{
544 int ret;
545
546 down_write(&current->mm->mmap_sem);
547 ret = mm_pkey_free(current->mm, pkey);
548 up_write(&current->mm->mmap_sem);
549
550 /*
551 * We could provie warnings or errors if any VMA still
552 * has the pkey set here.
553 */
554 return ret;
555}
Heiko Carstensc7142ae2016-12-12 16:43:09 -0800556
557#endif /* CONFIG_ARCH_HAS_PKEYS */