blob: bb53a6591aea1373d6bc74d5bb3e847650af8d24 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/uaccess.h>
27#include <asm/pgtable.h>
28#include <asm/cacheflush.h>
29#include <asm/tlbflush.h>
30
Venki Pallipadi1c12c4c2008-05-14 16:05:51 -070031#ifndef pgprot_modify
32static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
33{
34 return newprot;
35}
36#endif
37
Mel Gorman4b10e7d2012-10-25 14:16:32 +020038static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Peter Zijlstrac1e60982006-09-25 23:30:59 -070039 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +010040 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -070041{
Mel Gorman4b10e7d2012-10-25 14:16:32 +020042 struct mm_struct *mm = vma->vm_mm;
Christoph Lameter06972122006-06-23 02:03:35 -070043 pte_t *pte, oldpte;
Hugh Dickins705e87c2005-10-29 18:16:27 -070044 spinlock_t *ptl;
Peter Zijlstra7da4d642012-11-19 03:14:23 +010045 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Hugh Dickins705e87c2005-10-29 18:16:27 -070047 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
Zachary Amsden6606c3e2006-09-30 23:29:33 -070048 arch_enter_lazy_mmu_mode();
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 do {
Christoph Lameter06972122006-06-23 02:03:35 -070050 oldpte = *pte;
51 if (pte_present(oldpte)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 pte_t ptent;
Mel Gorman4b10e7d2012-10-25 14:16:32 +020053 bool updated = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Mel Gorman4b10e7d2012-10-25 14:16:32 +020055 if (!prot_numa) {
Mel Gorman0c5f83c2013-12-18 17:08:37 -080056 ptent = ptep_modify_prot_start(mm, addr, pte);
Mel Gorman16679182013-12-18 17:08:41 -080057 if (pte_numa(ptent))
58 ptent = pte_mknonnuma(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +020059 ptent = pte_modify(ptent, newprot);
60 updated = true;
61 } else {
62 struct page *page;
63
Mel Gorman0c5f83c2013-12-18 17:08:37 -080064 ptent = *pte;
Mel Gorman4b10e7d2012-10-25 14:16:32 +020065 page = vm_normal_page(vma, addr, oldpte);
66 if (page) {
Mel Gorman1bc115d2013-10-07 11:29:05 +010067 if (!pte_numa(oldpte)) {
Mel Gorman4b10e7d2012-10-25 14:16:32 +020068 ptent = pte_mknuma(ptent);
Mel Gorman0c5f83c2013-12-18 17:08:37 -080069 set_pte_at(mm, addr, pte, ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +020070 updated = true;
71 }
72 }
73 }
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -070074
Peter Zijlstrac1e60982006-09-25 23:30:59 -070075 /*
76 * Avoid taking write faults for pages we know to be
77 * dirty.
78 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +020079 if (dirty_accountable && pte_dirty(ptent)) {
Peter Zijlstrac1e60982006-09-25 23:30:59 -070080 ptent = pte_mkwrite(ptent);
Mel Gorman4b10e7d2012-10-25 14:16:32 +020081 updated = true;
82 }
Jeremy Fitzhardinge1ea07042008-06-16 04:30:00 -070083
Mel Gorman4b10e7d2012-10-25 14:16:32 +020084 if (updated)
85 pages++;
Mel Gorman0c5f83c2013-12-18 17:08:37 -080086
87 /* Only !prot_numa always clears the pte */
88 if (!prot_numa)
89 ptep_modify_prot_commit(mm, addr, pte, ptent);
Konstantin Khlebnikovce1744f2012-03-21 16:33:59 -070090 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
Christoph Lameter06972122006-06-23 02:03:35 -070091 swp_entry_t entry = pte_to_swp_entry(oldpte);
92
93 if (is_write_migration_entry(entry)) {
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -070094 pte_t newpte;
Christoph Lameter06972122006-06-23 02:03:35 -070095 /*
96 * A protection check is difficult so
97 * just be safe and disable write
98 */
99 make_migration_entry_read(&entry);
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700100 newpte = swp_entry_to_pte(entry);
101 if (pte_swp_soft_dirty(oldpte))
102 newpte = pte_swp_mksoft_dirty(newpte);
103 set_pte_at(mm, addr, pte, newpte);
Mel Gormane920e142013-10-07 11:28:48 +0100104
105 pages++;
Christoph Lameter06972122006-06-23 02:03:35 -0700106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 }
108 } while (pte++, addr += PAGE_SIZE, addr != end);
Zachary Amsden6606c3e2006-09-30 23:29:33 -0700109 arch_leave_lazy_mmu_mode();
Hugh Dickins705e87c2005-10-29 18:16:27 -0700110 pte_unmap_unlock(pte - 1, ptl);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100111
112 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
Andrew Morton7d12efa2012-12-18 14:23:17 -0800115static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
116 pud_t *pud, unsigned long addr, unsigned long end,
117 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
119 pmd_t *pmd;
120 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100121 unsigned long pages = 0;
Mel Gorman72403b42013-11-12 15:08:32 -0800122 unsigned long nr_huge_updates = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 pmd = pmd_offset(pud, addr);
125 do {
Mel Gorman25cbbef2013-10-07 11:29:14 +0100126 unsigned long this_pages;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 next = pmd_addr_end(addr, end);
Johannes Weinercd7548a2011-01-13 15:47:04 -0800129 if (pmd_trans_huge(*pmd)) {
130 if (next - addr != HPAGE_PMD_SIZE)
Kirill A. Shutemove1803772012-12-12 13:50:59 -0800131 split_huge_page_pmd(vma, addr, pmd);
Mel Gormanf123d742013-10-07 11:28:49 +0100132 else {
133 int nr_ptes = change_huge_pmd(vma, pmd, addr,
134 newprot, prot_numa);
135
136 if (nr_ptes) {
Mel Gorman72403b42013-11-12 15:08:32 -0800137 if (nr_ptes == HPAGE_PMD_NR) {
138 pages += HPAGE_PMD_NR;
139 nr_huge_updates++;
140 }
Mel Gormanf123d742013-10-07 11:28:49 +0100141 continue;
142 }
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100143 }
Johannes Weinercd7548a2011-01-13 15:47:04 -0800144 /* fall through */
145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 if (pmd_none_or_clear_bad(pmd))
147 continue;
Mel Gorman25cbbef2013-10-07 11:29:14 +0100148 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
Mel Gorman0f19c172013-10-07 11:29:25 +0100149 dirty_accountable, prot_numa);
Mel Gorman25cbbef2013-10-07 11:29:14 +0100150 pages += this_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 } while (pmd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100152
Mel Gorman72403b42013-11-12 15:08:32 -0800153 if (nr_huge_updates)
154 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100155 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
Andrew Morton7d12efa2012-12-18 14:23:17 -0800158static inline unsigned long change_pud_range(struct vm_area_struct *vma,
159 pgd_t *pgd, unsigned long addr, unsigned long end,
160 pgprot_t newprot, int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
162 pud_t *pud;
163 unsigned long next;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100164 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166 pud = pud_offset(pgd, addr);
167 do {
168 next = pud_addr_end(addr, end);
169 if (pud_none_or_clear_bad(pud))
170 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100171 pages += change_pmd_range(vma, pud, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200172 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 } while (pud++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100174
175 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100178static unsigned long change_protection_range(struct vm_area_struct *vma,
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700179 unsigned long addr, unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200180 int dirty_accountable, int prot_numa)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181{
182 struct mm_struct *mm = vma->vm_mm;
183 pgd_t *pgd;
184 unsigned long next;
185 unsigned long start = addr;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100186 unsigned long pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 BUG_ON(addr >= end);
189 pgd = pgd_offset(mm, addr);
190 flush_cache_range(vma, addr, end);
Rik van Riel20841402013-12-18 17:08:44 -0800191 set_tlb_flush_pending(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 do {
193 next = pgd_addr_end(addr, end);
194 if (pgd_none_or_clear_bad(pgd))
195 continue;
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100196 pages += change_pud_range(vma, pgd, addr, next, newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200197 dirty_accountable, prot_numa);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 } while (pgd++, addr = next, addr != end);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100199
Ingo Molnar1233d582012-11-19 03:14:24 +0100200 /* Only flush the TLB if we actually modified any entries: */
201 if (pages)
202 flush_tlb_range(vma, start, end);
Rik van Riel20841402013-12-18 17:08:44 -0800203 clear_tlb_flush_pending(mm);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100204
205 return pages;
206}
207
208unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
209 unsigned long end, pgprot_t newprot,
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200210 int dirty_accountable, int prot_numa)
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100211{
212 struct mm_struct *mm = vma->vm_mm;
213 unsigned long pages;
214
215 mmu_notifier_invalidate_range_start(mm, start, end);
216 if (is_vm_hugetlb_page(vma))
217 pages = hugetlb_change_protection(vma, start, end, newprot);
218 else
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200219 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100220 mmu_notifier_invalidate_range_end(mm, start, end);
221
222 return pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700225int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
227 unsigned long start, unsigned long end, unsigned long newflags)
228{
229 struct mm_struct *mm = vma->vm_mm;
230 unsigned long oldflags = vma->vm_flags;
231 long nrpages = (end - start) >> PAGE_SHIFT;
232 unsigned long charged = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 pgoff_t pgoff;
234 int error;
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700235 int dirty_accountable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237 if (newflags == oldflags) {
238 *pprev = vma;
239 return 0;
240 }
241
242 /*
243 * If we make a private mapping writable we increase our commit;
244 * but (without finer accounting) cannot reduce our commit if we
Mel Gorman5a6fe122009-02-10 14:02:27 +0000245 * make it unwritable again. hugetlb mapping were accounted for
246 * even if read-only so there is no need to account for them here
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 */
248 if (newflags & VM_WRITE) {
Mel Gorman5a6fe122009-02-10 14:02:27 +0000249 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
Andy Whitcroftcdfd4322008-07-23 21:27:28 -0700250 VM_SHARED|VM_NORESERVE))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 charged = nrpages;
Al Viro191c5422012-02-13 03:58:52 +0000252 if (security_vm_enough_memory_mm(mm, charged))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return -ENOMEM;
254 newflags |= VM_ACCOUNT;
255 }
256 }
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 /*
259 * First try to merge with previous and/or next vma.
260 */
261 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
262 *pprev = vma_merge(mm, *pprev, start, end, newflags,
263 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
264 if (*pprev) {
265 vma = *pprev;
266 goto success;
267 }
268
269 *pprev = vma;
270
271 if (start != vma->vm_start) {
272 error = split_vma(mm, vma, start, 1);
273 if (error)
274 goto fail;
275 }
276
277 if (end != vma->vm_end) {
278 error = split_vma(mm, vma, end, 0);
279 if (error)
280 goto fail;
281 }
282
283success:
284 /*
285 * vm_flags and vm_page_prot are protected by the mmap_sem
286 * held in write mode.
287 */
288 vma->vm_flags = newflags;
Venki Pallipadi1c12c4c2008-05-14 16:05:51 -0700289 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
290 vm_get_page_prot(newflags));
291
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700292 if (vma_wants_writenotify(vma)) {
Hugh Dickins1ddd4392007-10-22 20:45:12 -0700293 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
Peter Zijlstrac1e60982006-09-25 23:30:59 -0700294 dirty_accountable = 1;
295 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700296
Andrew Morton7d12efa2012-12-18 14:23:17 -0800297 change_protection(vma, start, end, vma->vm_page_prot,
298 dirty_accountable, 0);
Peter Zijlstra7da4d642012-11-19 03:14:23 +0100299
Hugh Dickinsab50b8e2005-10-29 18:15:56 -0700300 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
301 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
Pekka Enberg63bfd732010-11-08 21:29:07 +0200302 perf_event_mmap(vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 return 0;
304
305fail:
306 vm_unacct_memory(charged);
307 return error;
308}
309
Heiko Carstens6a6160a2009-01-14 14:14:15 +0100310SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
311 unsigned long, prot)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
313 unsigned long vm_flags, nstart, end, tmp, reqprot;
314 struct vm_area_struct *vma, *prev;
315 int error = -EINVAL;
316 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
317 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
318 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
319 return -EINVAL;
320
321 if (start & ~PAGE_MASK)
322 return -EINVAL;
323 if (!len)
324 return 0;
325 len = PAGE_ALIGN(len);
326 end = start + len;
327 if (end <= start)
328 return -ENOMEM;
Dave Kleikampb845f312008-07-08 00:28:51 +1000329 if (!arch_validate_prot(prot))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 return -EINVAL;
331
332 reqprot = prot;
333 /*
334 * Does the application expect PROT_READ to imply PROT_EXEC:
335 */
Hua Zhongb344e052006-06-23 02:03:23 -0700336 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 prot |= PROT_EXEC;
338
339 vm_flags = calc_vm_prot_bits(prot);
340
341 down_write(&current->mm->mmap_sem);
342
Linus Torvalds097d5912012-03-06 18:23:36 -0800343 vma = find_vma(current->mm, start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 error = -ENOMEM;
345 if (!vma)
346 goto out;
Linus Torvalds097d5912012-03-06 18:23:36 -0800347 prev = vma->vm_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 if (unlikely(grows & PROT_GROWSDOWN)) {
349 if (vma->vm_start >= end)
350 goto out;
351 start = vma->vm_start;
352 error = -EINVAL;
353 if (!(vma->vm_flags & VM_GROWSDOWN))
354 goto out;
Andrew Morton7d12efa2012-12-18 14:23:17 -0800355 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 if (vma->vm_start > start)
357 goto out;
358 if (unlikely(grows & PROT_GROWSUP)) {
359 end = vma->vm_end;
360 error = -EINVAL;
361 if (!(vma->vm_flags & VM_GROWSUP))
362 goto out;
363 }
364 }
365 if (start > vma->vm_start)
366 prev = vma;
367
368 for (nstart = start ; ; ) {
369 unsigned long newflags;
370
Andrew Morton7d12efa2012-12-18 14:23:17 -0800371 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Andrew Morton7d12efa2012-12-18 14:23:17 -0800373 newflags = vm_flags;
374 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Paolo 'Blaisorblade' Giarrusso7e2cff42005-09-21 09:55:39 -0700376 /* newflags >> 4 shift VM_MAY% in place of VM_% */
377 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 error = -EACCES;
379 goto out;
380 }
381
382 error = security_file_mprotect(vma, reqprot, prot);
383 if (error)
384 goto out;
385
386 tmp = vma->vm_end;
387 if (tmp > end)
388 tmp = end;
389 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
390 if (error)
391 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 nstart = tmp;
393
394 if (nstart < prev->vm_end)
395 nstart = prev->vm_end;
396 if (nstart >= end)
397 goto out;
398
399 vma = prev->vm_next;
400 if (!vma || vma->vm_start != nstart) {
401 error = -ENOMEM;
402 goto out;
403 }
404 }
405out:
406 up_write(&current->mm->mmap_sem);
407 return error;
408}