blob: 1f2c969f785cd32bbe4152936be282dad54b7d19 [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
Mel Gorman5a1eef72017-08-02 13:31:52 -070077 flush_tlb_batched_pending(vma->vm_mm);
Zachary Amsden6606c3e2006-09-30 23:29:33 -070078 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 do {
Christoph Lameter06972122006-06-23 02:03:35 -070080 oldpte = *pte;
81 if (pte_present(oldpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 pte_t ptent;
Mel Gormanb191f9b2015-03-25 15:55:40 -070083 bool preserve_write = prot_numa && pte_write(oldpte);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Mel Gormane944fd62015-02-12 14:58:35 -080085 /*
86 * Avoid trapping faults against the zero or KSM
87 * pages. See similar comment in change_huge_pmd.
88 */
89 if (prot_numa) {
90 struct page *page;
91
92 page = vm_normal_page(vma, addr, oldpte);
93 if (!page || PageKsm(page))
94 continue;
Mel Gorman10c10452015-02-12 14:58:44 -080095
96 /* Avoid TLB flush if possible */
97 if (pte_protnone(oldpte))
98 continue;
Mel Gormane944fd62015-02-12 14:58:35 -080099 }
100
Mel Gorman8a0516e2015-02-12 14:58:22 -0800101 ptent = ptep_modify_prot_start(mm, addr, pte);
102 ptent = pte_modify(ptent, newprot);
Mel Gormanb191f9b2015-03-25 15:55:40 -0700103 if (preserve_write)
104 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200105
Mel Gorman8a0516e2015-02-12 14:58:22 -0800106 /* Avoid taking write faults for known dirty pages */
107 if (dirty_accountable && pte_dirty(ptent) &&
108 (pte_soft_dirty(ptent) ||
109 !(vma->vm_flags & VM_SOFTDIRTY))) {
110 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200111 }
Mel Gorman8a0516e2015-02-12 14:58:22 -0800112 ptep_modify_prot_commit(mm, addr, pte, ptent);
113 pages++;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -0800114 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
Christoph Lameter06972122006-06-23 02:03:35 -0700115 swp_entry_t entry = pte_to_swp_entry(oldpte);
116
117 if (is_write_migration_entry(entry)) {
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700118 pte_t newpte;
Christoph Lameter06972122006-06-23 02:03:35 -0700119 /*
120 * A protection check is difficult so
121 * just be safe and disable write
122 */
123 make_migration_entry_read(&entry);
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700124 newpte = swp_entry_to_pte(entry);
125 if (pte_swp_soft_dirty(oldpte))
126 newpte = pte_swp_mksoft_dirty(newpte);
127 set_pte_at(mm, addr, pte, newpte);
Mel Gormane920e142013-10-07 11:28:48 +0100128
129 pages++;
Christoph Lameter06972122006-06-23 02:03:35 -0700130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 }
132 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700133 arch_leave_lazy_mmu_mode();
Hugh Dickins705e87c2005-10-29 18:16:27 -0700134 pte_unmap_unlock(pte - 1, ptl);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100135
136 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Andrew Morton7d12efa2012-12-18 14:23:17 -0800139static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
140 pud_t *pud, unsigned long addr, unsigned long end,
141 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
143 pmd_t *pmd;
Rik van Riela5338092014-04-07 15:36:57 -0700144 struct mm_struct *mm = vma->vm_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100146 unsigned long pages = 0;
Mel Gorman72403b42013-11-12 15:08:32 -0800147 unsigned long nr_huge_updates = 0;
Rik van Riela5338092014-04-07 15:36:57 -0700148 unsigned long mni_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150 pmd = pmd_offset(pud, addr);
151 do {
Mel Gorman25cbbef2013-10-07 11:29:14 +0100152 unsigned long this_pages;
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 next = pmd_addr_end(addr, end);
Dan Williams5c7fb562016-01-15 16:56:52 -0800155 if (!pmd_trans_huge(*pmd) && !pmd_devmap(*pmd)
156 && pmd_none_or_clear_bad(pmd))
Rik van Riel88a9ab62014-04-07 15:36:55 -0700157 continue;
Rik van Riela5338092014-04-07 15:36:57 -0700158
159 /* invoke the mmu notifier if the pmd is populated */
160 if (!mni_start) {
161 mni_start = addr;
162 mmu_notifier_invalidate_range_start(mm, mni_start, end);
163 }
164
Dan Williams5c7fb562016-01-15 16:56:52 -0800165 if (pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) {
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800166 if (next - addr != HPAGE_PMD_SIZE) {
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -0800167 split_huge_pmd(vma, pmd, addr);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700168 if (pmd_trans_unstable(pmd))
Kirill A. Shutemov6b9116a2016-02-11 16:13:03 -0800169 continue;
170 } else {
Mel Gormanf123d742013-10-07 11:28:49 +0100171 int nr_ptes = change_huge_pmd(vma, pmd, addr,
Mel Gormane944fd62015-02-12 14:58:35 -0800172 newprot, prot_numa);
Mel Gormanf123d742013-10-07 11:28:49 +0100173
174 if (nr_ptes) {
Mel Gorman72403b42013-11-12 15:08:32 -0800175 if (nr_ptes == HPAGE_PMD_NR) {
176 pages += HPAGE_PMD_NR;
177 nr_huge_updates++;
178 }
Mel Gorman1ad9f622014-04-07 15:36:56 -0700179
180 /* huge pmd was handled */
Mel Gormanf123d742013-10-07 11:28:49 +0100181 continue;
182 }
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100183 }
Rik van Riel88a9ab62014-04-07 15:36:55 -0700184 /* fall through, the trans huge pmd just split */
Johannes Weinercd7548a2011-01-13 15:47:04 -0800185 }
Mel Gorman25cbbef2013-10-07 11:29:14 +0100186 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +0100187 dirty_accountable, prot_numa);
Mel Gorman25cbbef2013-10-07 11:29:14 +0100188 pages += this_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 } while (pmd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100190
Rik van Riela5338092014-04-07 15:36:57 -0700191 if (mni_start)
192 mmu_notifier_invalidate_range_end(mm, mni_start, end);
193
Mel Gorman72403b42013-11-12 15:08:32 -0800194 if (nr_huge_updates)
195 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100196 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Andrew Morton7d12efa2012-12-18 14:23:17 -0800199static inline unsigned long change_pud_range(struct vm_area_struct *vma,
200 pgd_t *pgd, unsigned long addr, unsigned long end,
201 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
203 pud_t *pud;
204 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100205 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 pud = pud_offset(pgd, addr);
208 do {
209 next = pud_addr_end(addr, end);
210 if (pud_none_or_clear_bad(pud))
211 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100212 pages += change_pmd_range(vma, pud, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200213 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 } while (pud++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100215
216 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100219static unsigned long change_protection_range(struct vm_area_struct *vma,
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700220 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200221 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
223 struct mm_struct *mm = vma->vm_mm;
224 pgd_t *pgd;
225 unsigned long next;
226 unsigned long start = addr;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100227 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 BUG_ON(addr >= end);
230 pgd = pgd_offset(mm, addr);
231 flush_cache_range(vma, addr, end);
Rik van Riel20841402013-12-18 17:08:44 -0800232 set_tlb_flush_pending(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 do {
234 next = pgd_addr_end(addr, end);
235 if (pgd_none_or_clear_bad(pgd))
236 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100237 pages += change_pud_range(vma, pgd, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200238 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 } while (pgd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100240
Ingo Molnar1233d582012-11-19 03:14:24 +0100241 /* Only flush the TLB if we actually modified any entries: */
242 if (pages)
243 flush_tlb_range(vma, start, end);
Rik van Riel20841402013-12-18 17:08:44 -0800244 clear_tlb_flush_pending(mm);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100245
246 return pages;
247}
248
249unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
250 unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200251 int dirty_accountable, int prot_numa)
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100252{
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100253 unsigned long pages;
254
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100255 if (is_vm_hugetlb_page(vma))
256 pages = hugetlb_change_protection(vma, start, end, newprot);
257 else
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200258 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100259
260 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700263int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
265 unsigned long start, unsigned long end, unsigned long newflags)
266{
267 struct mm_struct *mm = vma->vm_mm;
268 unsigned long oldflags = vma->vm_flags;
269 long nrpages = (end - start) >> PAGE_SHIFT;
270 unsigned long charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 pgoff_t pgoff;
272 int error;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700273 int dirty_accountable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 if (newflags == oldflags) {
276 *pprev = vma;
277 return 0;
278 }
279
280 /*
281 * If we make a private mapping writable we increase our commit;
282 * but (without finer accounting) cannot reduce our commit if we
Mel Gorman5a6fe122009-02-10 14:02:27 +0000283 * make it unwritable again. hugetlb mapping were accounted for
284 * even if read-only so there is no need to account for them here
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 */
286 if (newflags & VM_WRITE) {
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800287 /* Check space limits when area turns into data. */
288 if (!may_expand_vm(mm, newflags, nrpages) &&
289 may_expand_vm(mm, oldflags, nrpages))
290 return -ENOMEM;
Mel Gorman5a6fe122009-02-10 14:02:27 +0000291 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
Andy Whitcroftcdfd4322008-07-23 21:27:28 -0700292 VM_SHARED|VM_NORESERVE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 charged = nrpages;
Al Viro191c5422012-02-13 03:58:52 +0000294 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return -ENOMEM;
296 newflags |= VM_ACCOUNT;
297 }
298 }
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 /*
301 * First try to merge with previous and/or next vma.
302 */
303 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
304 *pprev = vma_merge(mm, *pprev, start, end, newflags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700305 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
Colin Cross3e4578f2015-10-27 16:42:08 -0700306 vma->vm_userfaultfd_ctx, vma_get_anon_name(vma));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 if (*pprev) {
308 vma = *pprev;
Andrea Arcangelie86f15e2016-10-07 17:01:28 -0700309 VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 goto success;
311 }
312
313 *pprev = vma;
314
315 if (start != vma->vm_start) {
316 error = split_vma(mm, vma, start, 1);
317 if (error)
318 goto fail;
319 }
320
321 if (end != vma->vm_end) {
322 error = split_vma(mm, vma, end, 0);
323 if (error)
324 goto fail;
325 }
326
327success:
328 /*
329 * vm_flags and vm_page_prot are protected by the mmap_sem
330 * held in write mode.
331 */
332 vma->vm_flags = newflags;
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700333 dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
Peter Feiner64e45502014-10-13 15:55:46 -0700334 vma_set_page_prot(vma);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700335
Andrew Morton7d12efa2012-12-18 14:23:17 -0800336 change_protection(vma, start, end, vma->vm_page_prot,
337 dirty_accountable, 0);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100338
Kirill A. Shutemov36f88182015-06-24 16:56:10 -0700339 /*
340 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
341 * fault on access.
342 */
343 if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
344 (newflags & VM_WRITE)) {
345 populate_vma_page_range(vma, start, end, NULL);
346 }
347
Konstantin Khlebnikov84638332016-01-14 15:22:07 -0800348 vm_stat_account(mm, oldflags, -nrpages);
349 vm_stat_account(mm, newflags, nrpages);
Pekka Enberg63bfd732010-11-08 21:29:07 +0200350 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 return 0;
352
353fail:
354 vm_unacct_memory(charged);
355 return error;
356}
357
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700358/*
359 * pkey==-1 when doing a legacy mprotect()
360 */
361static int do_mprotect_pkey(unsigned long start, size_t len,
362 unsigned long prot, int pkey)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Dave Hansen62b5f7d2016-02-12 13:02:40 -0800364 unsigned long nstart, end, tmp, reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 struct vm_area_struct *vma, *prev;
366 int error = -EINVAL;
367 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700368 const bool rier = (current->personality & READ_IMPLIES_EXEC) &&
369 (prot & PROT_READ);
370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
372 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
373 return -EINVAL;
374
375 if (start & ~PAGE_MASK)
376 return -EINVAL;
377 if (!len)
378 return 0;
379 len = PAGE_ALIGN(len);
380 end = start + len;
381 if (end <= start)
382 return -ENOMEM;
Dave Kleikampb845f312008-07-08 00:28:51 +1000383 if (!arch_validate_prot(prot))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return -EINVAL;
385
386 reqprot = prot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
Michal Hockodc0ef0d2016-05-23 16:25:27 -0700388 if (down_write_killable(&current->mm->mmap_sem))
389 return -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Dave Hansene8c24d32016-07-29 09:30:15 -0700391 /*
392 * If userspace did not allocate the pkey, do not let
393 * them use it here.
394 */
395 error = -EINVAL;
396 if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey))
397 goto out;
398
Linus Torvalds097d5912012-03-06 18:23:36 -0800399 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 error = -ENOMEM;
401 if (!vma)
402 goto out;
Linus Torvalds097d5912012-03-06 18:23:36 -0800403 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 if (unlikely(grows & PROT_GROWSDOWN)) {
405 if (vma->vm_start >= end)
406 goto out;
407 start = vma->vm_start;
408 error = -EINVAL;
409 if (!(vma->vm_flags & VM_GROWSDOWN))
410 goto out;
Andrew Morton7d12efa2012-12-18 14:23:17 -0800411 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 if (vma->vm_start > start)
413 goto out;
414 if (unlikely(grows & PROT_GROWSUP)) {
415 end = vma->vm_end;
416 error = -EINVAL;
417 if (!(vma->vm_flags & VM_GROWSUP))
418 goto out;
419 }
420 }
421 if (start > vma->vm_start)
422 prev = vma;
423
424 for (nstart = start ; ; ) {
Dave Hansena8502b62016-07-29 09:30:13 -0700425 unsigned long mask_off_old_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 unsigned long newflags;
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700427 int new_vma_pkey;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Andrew Morton7d12efa2012-12-18 14:23:17 -0800429 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700431 /* Does the application expect PROT_READ to imply PROT_EXEC */
432 if (rier && (vma->vm_flags & VM_MAYEXEC))
433 prot |= PROT_EXEC;
434
Dave Hansena8502b62016-07-29 09:30:13 -0700435 /*
436 * Each mprotect() call explicitly passes r/w/x permissions.
437 * If a permission is not passed to mprotect(), it must be
438 * cleared from the VMA.
439 */
440 mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC |
441 ARCH_VM_PKEY_FLAGS;
442
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700443 new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey);
444 newflags = calc_vm_prot_bits(prot, new_vma_pkey);
Dave Hansena8502b62016-07-29 09:30:13 -0700445 newflags |= (vma->vm_flags & ~mask_off_old_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Paolo 'Blaisorblade' Giarrusso7e2cff42005-09-21 09:55:39 -0700447 /* newflags >> 4 shift VM_MAY% in place of VM_% */
448 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 error = -EACCES;
450 goto out;
451 }
452
453 error = security_file_mprotect(vma, reqprot, prot);
454 if (error)
455 goto out;
456
457 tmp = vma->vm_end;
458 if (tmp > end)
459 tmp = end;
460 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
461 if (error)
462 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 nstart = tmp;
464
465 if (nstart < prev->vm_end)
466 nstart = prev->vm_end;
467 if (nstart >= end)
468 goto out;
469
470 vma = prev->vm_next;
471 if (!vma || vma->vm_start != nstart) {
472 error = -ENOMEM;
473 goto out;
474 }
Piotr Kwapulinskif1385562016-03-22 14:27:51 -0700475 prot = reqprot;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
477out:
478 up_write(&current->mm->mmap_sem);
479 return error;
480}
Dave Hansen7d06d9c2016-07-29 09:30:12 -0700481
482SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
483 unsigned long, prot)
484{
485 return do_mprotect_pkey(start, len, prot, -1);
486}
487
488SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len,
489 unsigned long, prot, int, pkey)
490{
491 return do_mprotect_pkey(start, len, prot, pkey);
492}
Dave Hansene8c24d32016-07-29 09:30:15 -0700493
494SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val)
495{
496 int pkey;
497 int ret;
498
499 /* No flags supported yet. */
500 if (flags)
501 return -EINVAL;
502 /* check for unsupported init values */
503 if (init_val & ~PKEY_ACCESS_MASK)
504 return -EINVAL;
505
506 down_write(&current->mm->mmap_sem);
507 pkey = mm_pkey_alloc(current->mm);
508
509 ret = -ENOSPC;
510 if (pkey == -1)
511 goto out;
512
513 ret = arch_set_user_pkey_access(current, pkey, init_val);
514 if (ret) {
515 mm_pkey_free(current->mm, pkey);
516 goto out;
517 }
518 ret = pkey;
519out:
520 up_write(&current->mm->mmap_sem);
521 return ret;
522}
523
524SYSCALL_DEFINE1(pkey_free, int, pkey)
525{
526 int ret;
527
528 down_write(&current->mm->mmap_sem);
529 ret = mm_pkey_free(current->mm, pkey);
530 up_write(&current->mm->mmap_sem);
531
532 /*
533 * We could provie warnings or errors if any VMA still
534 * has the pkey set here.
535 */
536 return ret;
537}