mm,numa: reorganize change_pmd_range()
[linux-2.6-block.git] / mm / mprotect.c
... / ...
CommitLineData
1/*
2 * mm/mprotect.c
3 *
4 * (C) Copyright 1994 Linus Torvalds
5 * (C) Copyright 2002 Christoph Hellwig
6 *
7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9 */
10
11#include <linux/mm.h>
12#include <linux/hugetlb.h>
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>
21#include <linux/swap.h>
22#include <linux/swapops.h>
23#include <linux/mmu_notifier.h>
24#include <linux/migrate.h>
25#include <linux/perf_event.h>
26#include <linux/ksm.h>
27#include <asm/uaccess.h>
28#include <asm/pgtable.h>
29#include <asm/cacheflush.h>
30#include <asm/tlbflush.h>
31
32#ifndef pgprot_modify
33static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
34{
35 return newprot;
36}
37#endif
38
39static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
40 unsigned long addr, unsigned long end, pgprot_t newprot,
41 int dirty_accountable, int prot_numa)
42{
43 struct mm_struct *mm = vma->vm_mm;
44 pte_t *pte, oldpte;
45 spinlock_t *ptl;
46 unsigned long pages = 0;
47
48 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
49 arch_enter_lazy_mmu_mode();
50 do {
51 oldpte = *pte;
52 if (pte_present(oldpte)) {
53 pte_t ptent;
54 bool updated = false;
55
56 if (!prot_numa) {
57 ptent = ptep_modify_prot_start(mm, addr, pte);
58 if (pte_numa(ptent))
59 ptent = pte_mknonnuma(ptent);
60 ptent = pte_modify(ptent, newprot);
61 /*
62 * Avoid taking write faults for pages we
63 * know to be dirty.
64 */
65 if (dirty_accountable && pte_dirty(ptent))
66 ptent = pte_mkwrite(ptent);
67 ptep_modify_prot_commit(mm, addr, pte, ptent);
68 updated = true;
69 } else {
70 struct page *page;
71
72 page = vm_normal_page(vma, addr, oldpte);
73 if (page && !PageKsm(page)) {
74 if (!pte_numa(oldpte)) {
75 ptep_set_numa(mm, addr, pte);
76 updated = true;
77 }
78 }
79 }
80 if (updated)
81 pages++;
82 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
83 swp_entry_t entry = pte_to_swp_entry(oldpte);
84
85 if (is_write_migration_entry(entry)) {
86 pte_t newpte;
87 /*
88 * A protection check is difficult so
89 * just be safe and disable write
90 */
91 make_migration_entry_read(&entry);
92 newpte = swp_entry_to_pte(entry);
93 if (pte_swp_soft_dirty(oldpte))
94 newpte = pte_swp_mksoft_dirty(newpte);
95 set_pte_at(mm, addr, pte, newpte);
96
97 pages++;
98 }
99 }
100 } while (pte++, addr += PAGE_SIZE, addr != end);
101 arch_leave_lazy_mmu_mode();
102 pte_unmap_unlock(pte - 1, ptl);
103
104 return pages;
105}
106
107static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
108 pud_t *pud, unsigned long addr, unsigned long end,
109 pgprot_t newprot, int dirty_accountable, int prot_numa)
110{
111 pmd_t *pmd;
112 unsigned long next;
113 unsigned long pages = 0;
114 unsigned long nr_huge_updates = 0;
115
116 pmd = pmd_offset(pud, addr);
117 do {
118 unsigned long this_pages;
119
120 next = pmd_addr_end(addr, end);
121 if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd))
122 continue;
123 if (pmd_trans_huge(*pmd)) {
124 if (next - addr != HPAGE_PMD_SIZE)
125 split_huge_page_pmd(vma, addr, pmd);
126 else {
127 int nr_ptes = change_huge_pmd(vma, pmd, addr,
128 newprot, prot_numa);
129
130 if (nr_ptes) {
131 if (nr_ptes == HPAGE_PMD_NR) {
132 pages += HPAGE_PMD_NR;
133 nr_huge_updates++;
134 }
135 continue;
136 }
137 }
138 /* fall through, the trans huge pmd just split */
139 }
140 VM_BUG_ON(pmd_trans_huge(*pmd));
141 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
142 dirty_accountable, prot_numa);
143 pages += this_pages;
144 } while (pmd++, addr = next, addr != end);
145
146 if (nr_huge_updates)
147 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
148 return pages;
149}
150
151static inline unsigned long change_pud_range(struct vm_area_struct *vma,
152 pgd_t *pgd, unsigned long addr, unsigned long end,
153 pgprot_t newprot, int dirty_accountable, int prot_numa)
154{
155 pud_t *pud;
156 unsigned long next;
157 unsigned long pages = 0;
158
159 pud = pud_offset(pgd, addr);
160 do {
161 next = pud_addr_end(addr, end);
162 if (pud_none_or_clear_bad(pud))
163 continue;
164 pages += change_pmd_range(vma, pud, addr, next, newprot,
165 dirty_accountable, prot_numa);
166 } while (pud++, addr = next, addr != end);
167
168 return pages;
169}
170
171static unsigned long change_protection_range(struct vm_area_struct *vma,
172 unsigned long addr, unsigned long end, pgprot_t newprot,
173 int dirty_accountable, int prot_numa)
174{
175 struct mm_struct *mm = vma->vm_mm;
176 pgd_t *pgd;
177 unsigned long next;
178 unsigned long start = addr;
179 unsigned long pages = 0;
180
181 BUG_ON(addr >= end);
182 pgd = pgd_offset(mm, addr);
183 flush_cache_range(vma, addr, end);
184 set_tlb_flush_pending(mm);
185 do {
186 next = pgd_addr_end(addr, end);
187 if (pgd_none_or_clear_bad(pgd))
188 continue;
189 pages += change_pud_range(vma, pgd, addr, next, newprot,
190 dirty_accountable, prot_numa);
191 } while (pgd++, addr = next, addr != end);
192
193 /* Only flush the TLB if we actually modified any entries: */
194 if (pages)
195 flush_tlb_range(vma, start, end);
196 clear_tlb_flush_pending(mm);
197
198 return pages;
199}
200
201unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
202 unsigned long end, pgprot_t newprot,
203 int dirty_accountable, int prot_numa)
204{
205 struct mm_struct *mm = vma->vm_mm;
206 unsigned long pages;
207
208 mmu_notifier_invalidate_range_start(mm, start, end);
209 if (is_vm_hugetlb_page(vma))
210 pages = hugetlb_change_protection(vma, start, end, newprot);
211 else
212 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
213 mmu_notifier_invalidate_range_end(mm, start, end);
214
215 return pages;
216}
217
218int
219mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
220 unsigned long start, unsigned long end, unsigned long newflags)
221{
222 struct mm_struct *mm = vma->vm_mm;
223 unsigned long oldflags = vma->vm_flags;
224 long nrpages = (end - start) >> PAGE_SHIFT;
225 unsigned long charged = 0;
226 pgoff_t pgoff;
227 int error;
228 int dirty_accountable = 0;
229
230 if (newflags == oldflags) {
231 *pprev = vma;
232 return 0;
233 }
234
235 /*
236 * If we make a private mapping writable we increase our commit;
237 * but (without finer accounting) cannot reduce our commit if we
238 * make it unwritable again. hugetlb mapping were accounted for
239 * even if read-only so there is no need to account for them here
240 */
241 if (newflags & VM_WRITE) {
242 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
243 VM_SHARED|VM_NORESERVE))) {
244 charged = nrpages;
245 if (security_vm_enough_memory_mm(mm, charged))
246 return -ENOMEM;
247 newflags |= VM_ACCOUNT;
248 }
249 }
250
251 /*
252 * First try to merge with previous and/or next vma.
253 */
254 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
255 *pprev = vma_merge(mm, *pprev, start, end, newflags,
256 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
257 if (*pprev) {
258 vma = *pprev;
259 goto success;
260 }
261
262 *pprev = vma;
263
264 if (start != vma->vm_start) {
265 error = split_vma(mm, vma, start, 1);
266 if (error)
267 goto fail;
268 }
269
270 if (end != vma->vm_end) {
271 error = split_vma(mm, vma, end, 0);
272 if (error)
273 goto fail;
274 }
275
276success:
277 /*
278 * vm_flags and vm_page_prot are protected by the mmap_sem
279 * held in write mode.
280 */
281 vma->vm_flags = newflags;
282 vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
283 vm_get_page_prot(newflags));
284
285 if (vma_wants_writenotify(vma)) {
286 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
287 dirty_accountable = 1;
288 }
289
290 change_protection(vma, start, end, vma->vm_page_prot,
291 dirty_accountable, 0);
292
293 vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
294 vm_stat_account(mm, newflags, vma->vm_file, nrpages);
295 perf_event_mmap(vma);
296 return 0;
297
298fail:
299 vm_unacct_memory(charged);
300 return error;
301}
302
303SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
304 unsigned long, prot)
305{
306 unsigned long vm_flags, nstart, end, tmp, reqprot;
307 struct vm_area_struct *vma, *prev;
308 int error = -EINVAL;
309 const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
310 prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
311 if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
312 return -EINVAL;
313
314 if (start & ~PAGE_MASK)
315 return -EINVAL;
316 if (!len)
317 return 0;
318 len = PAGE_ALIGN(len);
319 end = start + len;
320 if (end <= start)
321 return -ENOMEM;
322 if (!arch_validate_prot(prot))
323 return -EINVAL;
324
325 reqprot = prot;
326 /*
327 * Does the application expect PROT_READ to imply PROT_EXEC:
328 */
329 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
330 prot |= PROT_EXEC;
331
332 vm_flags = calc_vm_prot_bits(prot);
333
334 down_write(&current->mm->mmap_sem);
335
336 vma = find_vma(current->mm, start);
337 error = -ENOMEM;
338 if (!vma)
339 goto out;
340 prev = vma->vm_prev;
341 if (unlikely(grows & PROT_GROWSDOWN)) {
342 if (vma->vm_start >= end)
343 goto out;
344 start = vma->vm_start;
345 error = -EINVAL;
346 if (!(vma->vm_flags & VM_GROWSDOWN))
347 goto out;
348 } else {
349 if (vma->vm_start > start)
350 goto out;
351 if (unlikely(grows & PROT_GROWSUP)) {
352 end = vma->vm_end;
353 error = -EINVAL;
354 if (!(vma->vm_flags & VM_GROWSUP))
355 goto out;
356 }
357 }
358 if (start > vma->vm_start)
359 prev = vma;
360
361 for (nstart = start ; ; ) {
362 unsigned long newflags;
363
364 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
365
366 newflags = vm_flags;
367 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
368
369 /* newflags >> 4 shift VM_MAY% in place of VM_% */
370 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
371 error = -EACCES;
372 goto out;
373 }
374
375 error = security_file_mprotect(vma, reqprot, prot);
376 if (error)
377 goto out;
378
379 tmp = vma->vm_end;
380 if (tmp > end)
381 tmp = end;
382 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
383 if (error)
384 goto out;
385 nstart = tmp;
386
387 if (nstart < prev->vm_end)
388 nstart = prev->vm_end;
389 if (nstart >= end)
390 goto out;
391
392 vma = prev->vm_next;
393 if (!vma || vma->vm_start != nstart) {
394 error = -ENOMEM;
395 goto out;
396 }
397 }
398out:
399 up_write(&current->mm->mmap_sem);
400 return error;
401}