]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - mm/mprotect.c
Btrfs: fix device replace of a missing RAID 5/6 device
[mirror_ubuntu-artful-kernel.git] / mm / mprotect.c
CommitLineData
1da177e4
LT
1/*
2 * mm/mprotect.c
3 *
4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 *
046c6884 7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
1da177e4
LT
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9 */
10
11#include <linux/mm.h>
12#include <linux/hugetlb.h>
1da177e4
LT
13#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>
0697212a
CL
21#include <linux/swap.h>
22#include <linux/swapops.h>
cddb8a5c 23#include <linux/mmu_notifier.h>
64cdd548 24#include <linux/migrate.h>
cdd6c482 25#include <linux/perf_event.h>
64a9a34e 26#include <linux/ksm.h>
1da177e4
LT
27#include <asm/uaccess.h>
28#include <asm/pgtable.h>
29#include <asm/cacheflush.h>
30#include <asm/tlbflush.h>
31
36f88188
KS
32#include "internal.h"
33
1ad9f620
MG
34/*
35 * For a prot_numa update we only hold mmap_sem for read so there is a
36 * potential race with faulting where a pmd was temporarily none. This
37 * function checks for a transhuge pmd under the appropriate lock. It
38 * returns a pte if it was successfully locked or NULL if it raced with
39 * a transhuge insertion.
40 */
41static pte_t *lock_pte_protection(struct vm_area_struct *vma, pmd_t *pmd,
42 unsigned long addr, int prot_numa, spinlock_t **ptl)
43{
44 pte_t *pte;
45 spinlock_t *pmdl;
46
47 /* !prot_numa is protected by mmap_sem held for write */
48 if (!prot_numa)
49 return pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
50
51 pmdl = pmd_lock(vma->vm_mm, pmd);
52 if (unlikely(pmd_trans_huge(*pmd) || pmd_none(*pmd))) {
53 spin_unlock(pmdl);
54 return NULL;
55 }
56
57 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, ptl);
58 spin_unlock(pmdl);
59 return pte;
60}
61
4b10e7d5 62static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
c1e6098b 63 unsigned long addr, unsigned long end, pgprot_t newprot,
0f19c179 64 int dirty_accountable, int prot_numa)
1da177e4 65{
4b10e7d5 66 struct mm_struct *mm = vma->vm_mm;
0697212a 67 pte_t *pte, oldpte;
705e87c0 68 spinlock_t *ptl;
7da4d641 69 unsigned long pages = 0;
1da177e4 70
1ad9f620
MG
71 pte = lock_pte_protection(vma, pmd, addr, prot_numa, &ptl);
72 if (!pte)
73 return 0;
74
6606c3e0 75 arch_enter_lazy_mmu_mode();
1da177e4 76 do {
0697212a
CL
77 oldpte = *pte;
78 if (pte_present(oldpte)) {
1da177e4 79 pte_t ptent;
b191f9b1 80 bool preserve_write = prot_numa && pte_write(oldpte);
1da177e4 81
e944fd67
MG
82 /*
83 * Avoid trapping faults against the zero or KSM
84 * pages. See similar comment in change_huge_pmd.
85 */
86 if (prot_numa) {
87 struct page *page;
88
89 page = vm_normal_page(vma, addr, oldpte);
90 if (!page || PageKsm(page))
91 continue;
10c1045f
MG
92
93 /* Avoid TLB flush if possible */
94 if (pte_protnone(oldpte))
95 continue;
e944fd67
MG
96 }
97
8a0516ed
MG
98 ptent = ptep_modify_prot_start(mm, addr, pte);
99 ptent = pte_modify(ptent, newprot);
b191f9b1
MG
100 if (preserve_write)
101 ptent = pte_mkwrite(ptent);
4b10e7d5 102
8a0516ed
MG
103 /* Avoid taking write faults for known dirty pages */
104 if (dirty_accountable && pte_dirty(ptent) &&
105 (pte_soft_dirty(ptent) ||
106 !(vma->vm_flags & VM_SOFTDIRTY))) {
107 ptent = pte_mkwrite(ptent);
4b10e7d5 108 }
8a0516ed
MG
109 ptep_modify_prot_commit(mm, addr, pte, ptent);
110 pages++;
0661a336 111 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
0697212a
CL
112 swp_entry_t entry = pte_to_swp_entry(oldpte);
113
114 if (is_write_migration_entry(entry)) {
c3d16e16 115 pte_t newpte;
0697212a
CL
116 /*
117 * A protection check is difficult so
118 * just be safe and disable write
119 */
120 make_migration_entry_read(&entry);
c3d16e16
CG
121 newpte = swp_entry_to_pte(entry);
122 if (pte_swp_soft_dirty(oldpte))
123 newpte = pte_swp_mksoft_dirty(newpte);
124 set_pte_at(mm, addr, pte, newpte);
e920e14c
MG
125
126 pages++;
0697212a 127 }
1da177e4
LT
128 }
129 } while (pte++, addr += PAGE_SIZE, addr != end);
6606c3e0 130 arch_leave_lazy_mmu_mode();
705e87c0 131 pte_unmap_unlock(pte - 1, ptl);
7da4d641
PZ
132
133 return pages;
1da177e4
LT
134}
135
7d12efae
AM
136static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
137 pud_t *pud, unsigned long addr, unsigned long end,
138 pgprot_t newprot, int dirty_accountable, int prot_numa)
1da177e4
LT
139{
140 pmd_t *pmd;
a5338093 141 struct mm_struct *mm = vma->vm_mm;
1da177e4 142 unsigned long next;
7da4d641 143 unsigned long pages = 0;
72403b4a 144 unsigned long nr_huge_updates = 0;
a5338093 145 unsigned long mni_start = 0;
1da177e4
LT
146
147 pmd = pmd_offset(pud, addr);
148 do {
25cbbef1
MG
149 unsigned long this_pages;
150
1da177e4 151 next = pmd_addr_end(addr, end);
88a9ab6e
RR
152 if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd))
153 continue;
a5338093
RR
154
155 /* invoke the mmu notifier if the pmd is populated */
156 if (!mni_start) {
157 mni_start = addr;
158 mmu_notifier_invalidate_range_start(mm, mni_start, end);
159 }
160
cd7548ab
JW
161 if (pmd_trans_huge(*pmd)) {
162 if (next - addr != HPAGE_PMD_SIZE)
e180377f 163 split_huge_page_pmd(vma, addr, pmd);
f123d74a
MG
164 else {
165 int nr_ptes = change_huge_pmd(vma, pmd, addr,
e944fd67 166 newprot, prot_numa);
f123d74a
MG
167
168 if (nr_ptes) {
72403b4a
MG
169 if (nr_ptes == HPAGE_PMD_NR) {
170 pages += HPAGE_PMD_NR;
171 nr_huge_updates++;
172 }
1ad9f620
MG
173
174 /* huge pmd was handled */
f123d74a
MG
175 continue;
176 }
7da4d641 177 }
88a9ab6e 178 /* fall through, the trans huge pmd just split */
cd7548ab 179 }
25cbbef1 180 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
0f19c179 181 dirty_accountable, prot_numa);
25cbbef1 182 pages += this_pages;
1da177e4 183 } while (pmd++, addr = next, addr != end);
7da4d641 184
a5338093
RR
185 if (mni_start)
186 mmu_notifier_invalidate_range_end(mm, mni_start, end);
187
72403b4a
MG
188 if (nr_huge_updates)
189 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
7da4d641 190 return pages;
1da177e4
LT
191}
192
7d12efae
AM
193static inline unsigned long change_pud_range(struct vm_area_struct *vma,
194 pgd_t *pgd, unsigned long addr, unsigned long end,
195 pgprot_t newprot, int dirty_accountable, int prot_numa)
1da177e4
LT
196{
197 pud_t *pud;
198 unsigned long next;
7da4d641 199 unsigned long pages = 0;
1da177e4
LT
200
201 pud = pud_offset(pgd, addr);
202 do {
203 next = pud_addr_end(addr, end);
204 if (pud_none_or_clear_bad(pud))
205 continue;
7da4d641 206 pages += change_pmd_range(vma, pud, addr, next, newprot,
4b10e7d5 207 dirty_accountable, prot_numa);
1da177e4 208 } while (pud++, addr = next, addr != end);
7da4d641
PZ
209
210 return pages;
1da177e4
LT
211}
212
7da4d641 213static unsigned long change_protection_range(struct vm_area_struct *vma,
c1e6098b 214 unsigned long addr, unsigned long end, pgprot_t newprot,
4b10e7d5 215 int dirty_accountable, int prot_numa)
1da177e4
LT
216{
217 struct mm_struct *mm = vma->vm_mm;
218 pgd_t *pgd;
219 unsigned long next;
220 unsigned long start = addr;
7da4d641 221 unsigned long pages = 0;
1da177e4
LT
222
223 BUG_ON(addr >= end);
224 pgd = pgd_offset(mm, addr);
225 flush_cache_range(vma, addr, end);
20841405 226 set_tlb_flush_pending(mm);
1da177e4
LT
227 do {
228 next = pgd_addr_end(addr, end);
229 if (pgd_none_or_clear_bad(pgd))
230 continue;
7da4d641 231 pages += change_pud_range(vma, pgd, addr, next, newprot,
4b10e7d5 232 dirty_accountable, prot_numa);
1da177e4 233 } while (pgd++, addr = next, addr != end);
7da4d641 234
1233d588
IM
235 /* Only flush the TLB if we actually modified any entries: */
236 if (pages)
237 flush_tlb_range(vma, start, end);
20841405 238 clear_tlb_flush_pending(mm);
7da4d641
PZ
239
240 return pages;
241}
242
243unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
244 unsigned long end, pgprot_t newprot,
4b10e7d5 245 int dirty_accountable, int prot_numa)
7da4d641 246{
7da4d641
PZ
247 unsigned long pages;
248
7da4d641
PZ
249 if (is_vm_hugetlb_page(vma))
250 pages = hugetlb_change_protection(vma, start, end, newprot);
251 else
4b10e7d5 252 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
7da4d641
PZ
253
254 return pages;
1da177e4
LT
255}
256
b6a2fea3 257int
1da177e4
LT
258mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
259 unsigned long start, unsigned long end, unsigned long newflags)
260{
261 struct mm_struct *mm = vma->vm_mm;
262 unsigned long oldflags = vma->vm_flags;
263 long nrpages = (end - start) >> PAGE_SHIFT;
264 unsigned long charged = 0;
1da177e4
LT
265 pgoff_t pgoff;
266 int error;
c1e6098b 267 int dirty_accountable = 0;
1da177e4
LT
268
269 if (newflags == oldflags) {
270 *pprev = vma;
271 return 0;
272 }
273
274 /*
275 * If we make a private mapping writable we increase our commit;
276 * but (without finer accounting) cannot reduce our commit if we
5a6fe125
MG
277 * make it unwritable again. hugetlb mapping were accounted for
278 * even if read-only so there is no need to account for them here
1da177e4
LT
279 */
280 if (newflags & VM_WRITE) {
5a6fe125 281 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
cdfd4325 282 VM_SHARED|VM_NORESERVE))) {
1da177e4 283 charged = nrpages;
191c5424 284 if (security_vm_enough_memory_mm(mm, charged))
1da177e4
LT
285 return -ENOMEM;
286 newflags |= VM_ACCOUNT;
287 }
288 }
289
1da177e4
LT
290 /*
291 * First try to merge with previous and/or next vma.
292 */
293 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
294 *pprev = vma_merge(mm, *pprev, start, end, newflags,
295 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
296 if (*pprev) {
297 vma = *pprev;
298 goto success;
299 }
300
301 *pprev = vma;
302
303 if (start != vma->vm_start) {
304 error = split_vma(mm, vma, start, 1);
305 if (error)
306 goto fail;
307 }
308
309 if (end != vma->vm_end) {
310 error = split_vma(mm, vma, end, 0);
311 if (error)
312 goto fail;
313 }
314
315success:
316 /*
317 * vm_flags and vm_page_prot are protected by the mmap_sem
318 * held in write mode.
319 */
320 vma->vm_flags = newflags;
64e45507
PF
321 dirty_accountable = vma_wants_writenotify(vma);
322 vma_set_page_prot(vma);
d08b3851 323
7d12efae
AM
324 change_protection(vma, start, end, vma->vm_page_prot,
325 dirty_accountable, 0);
7da4d641 326
36f88188
KS
327 /*
328 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
329 * fault on access.
330 */
331 if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED &&
332 (newflags & VM_WRITE)) {
333 populate_vma_page_range(vma, start, end, NULL);
334 }
335
ab50b8ed
HD
336 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
337 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
63bfd738 338 perf_event_mmap(vma);
1da177e4
LT
339 return 0;
340
341fail:
342 vm_unacct_memory(charged);
343 return error;
344}
345
6a6160a7
HC
346SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
347 unsigned long, prot)
1da177e4
LT
348{
349 unsigned long vm_flags, nstart, end, tmp, reqprot;
350 struct vm_area_struct *vma, *prev;
351 int error = -EINVAL;
352 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
353 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
354 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
355 return -EINVAL;
356
357 if (start & ~PAGE_MASK)
358 return -EINVAL;
359 if (!len)
360 return 0;
361 len = PAGE_ALIGN(len);
362 end = start + len;
363 if (end <= start)
364 return -ENOMEM;
b845f313 365 if (!arch_validate_prot(prot))
1da177e4
LT
366 return -EINVAL;
367
368 reqprot = prot;
369 /*
370 * Does the application expect PROT_READ to imply PROT_EXEC:
371 */
b344e05c 372 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1da177e4
LT
373 prot |= PROT_EXEC;
374
375 vm_flags = calc_vm_prot_bits(prot);
376
377 down_write(&current->mm->mmap_sem);
378
097d5910 379 vma = find_vma(current->mm, start);
1da177e4
LT
380 error = -ENOMEM;
381 if (!vma)
382 goto out;
097d5910 383 prev = vma->vm_prev;
1da177e4
LT
384 if (unlikely(grows & PROT_GROWSDOWN)) {
385 if (vma->vm_start >= end)
386 goto out;
387 start = vma->vm_start;
388 error = -EINVAL;
389 if (!(vma->vm_flags & VM_GROWSDOWN))
390 goto out;
7d12efae 391 } else {
1da177e4
LT
392 if (vma->vm_start > start)
393 goto out;
394 if (unlikely(grows & PROT_GROWSUP)) {
395 end = vma->vm_end;
396 error = -EINVAL;
397 if (!(vma->vm_flags & VM_GROWSUP))
398 goto out;
399 }
400 }
401 if (start > vma->vm_start)
402 prev = vma;
403
404 for (nstart = start ; ; ) {
405 unsigned long newflags;
406
7d12efae 407 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
1da177e4 408
7d12efae
AM
409 newflags = vm_flags;
410 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
1da177e4 411
7e2cff42
PBG
412 /* newflags >> 4 shift VM_MAY% in place of VM_% */
413 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
1da177e4
LT
414 error = -EACCES;
415 goto out;
416 }
417
418 error = security_file_mprotect(vma, reqprot, prot);
419 if (error)
420 goto out;
421
422 tmp = vma->vm_end;
423 if (tmp > end)
424 tmp = end;
425 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
426 if (error)
427 goto out;
428 nstart = tmp;
429
430 if (nstart < prev->vm_end)
431 nstart = prev->vm_end;
432 if (nstart >= end)
433 goto out;
434
435 vma = prev->vm_next;
436 if (!vma || vma->vm_start != nstart) {
437 error = -ENOMEM;
438 goto out;
439 }
440 }
441out:
442 up_write(&current->mm->mmap_sem);
443 return error;
444}