net/mlx4_en: Fix mixed PFC and Global pause user control requests
[pandora-kernel.git] / mm / mprotect.c
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 <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/cacheflush.h>
29 #include <asm/tlbflush.h>
30
31 #ifndef pgprot_modify
32 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
33 {
34         return newprot;
35 }
36 #endif
37
38 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
39                 unsigned long addr, unsigned long end, pgprot_t newprot,
40                 int dirty_accountable)
41 {
42         pte_t *pte, oldpte;
43         spinlock_t *ptl;
44
45         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
46         arch_enter_lazy_mmu_mode();
47         do {
48                 oldpte = *pte;
49                 if (pte_present(oldpte)) {
50                         pte_t ptent;
51
52                         ptent = ptep_modify_prot_start(mm, addr, pte);
53                         ptent = pte_modify(ptent, newprot);
54
55                         /*
56                          * Avoid taking write faults for pages we know to be
57                          * dirty.
58                          */
59                         if (dirty_accountable && pte_dirty(ptent))
60                                 ptent = pte_mkwrite(ptent);
61
62                         ptep_modify_prot_commit(mm, addr, pte, ptent);
63                 } else if (PAGE_MIGRATION && !pte_file(oldpte)) {
64                         swp_entry_t entry = pte_to_swp_entry(oldpte);
65
66                         if (is_write_migration_entry(entry)) {
67                                 /*
68                                  * A protection check is difficult so
69                                  * just be safe and disable write
70                                  */
71                                 make_migration_entry_read(&entry);
72                                 set_pte_at(mm, addr, pte,
73                                         swp_entry_to_pte(entry));
74                         }
75                 }
76         } while (pte++, addr += PAGE_SIZE, addr != end);
77         arch_leave_lazy_mmu_mode();
78         pte_unmap_unlock(pte - 1, ptl);
79 }
80
81 static inline void change_pmd_range(struct vm_area_struct *vma, pud_t *pud,
82                 unsigned long addr, unsigned long end, pgprot_t newprot,
83                 int dirty_accountable)
84 {
85         pmd_t *pmd;
86         unsigned long next;
87
88         pmd = pmd_offset(pud, addr);
89         do {
90                 next = pmd_addr_end(addr, end);
91                 if (pmd_trans_huge(*pmd)) {
92                         if (next - addr != HPAGE_PMD_SIZE)
93                                 split_huge_page_pmd(vma->vm_mm, pmd);
94                         else if (change_huge_pmd(vma, pmd, addr, newprot))
95                                 goto next;
96                         /* fall through */
97                 }
98                 if (pmd_none_or_clear_bad(pmd))
99                         goto next;
100                 change_pte_range(vma->vm_mm, pmd, addr, next, newprot,
101                                  dirty_accountable);
102 next:
103                 cond_resched();
104         } while (pmd++, addr = next, addr != end);
105 }
106
107 static inline void change_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
108                 unsigned long addr, unsigned long end, pgprot_t newprot,
109                 int dirty_accountable)
110 {
111         pud_t *pud;
112         unsigned long next;
113
114         pud = pud_offset(pgd, addr);
115         do {
116                 next = pud_addr_end(addr, end);
117                 if (pud_none_or_clear_bad(pud))
118                         continue;
119                 change_pmd_range(vma, pud, addr, next, newprot,
120                                  dirty_accountable);
121         } while (pud++, addr = next, addr != end);
122 }
123
124 static void change_protection(struct vm_area_struct *vma,
125                 unsigned long addr, unsigned long end, pgprot_t newprot,
126                 int dirty_accountable)
127 {
128         struct mm_struct *mm = vma->vm_mm;
129         pgd_t *pgd;
130         unsigned long next;
131         unsigned long start = addr;
132
133         BUG_ON(addr >= end);
134         pgd = pgd_offset(mm, addr);
135         flush_cache_range(vma, addr, end);
136         do {
137                 next = pgd_addr_end(addr, end);
138                 if (pgd_none_or_clear_bad(pgd))
139                         continue;
140                 change_pud_range(vma, pgd, addr, next, newprot,
141                                  dirty_accountable);
142         } while (pgd++, addr = next, addr != end);
143         flush_tlb_range(vma, start, end);
144 }
145
146 int
147 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
148         unsigned long start, unsigned long end, unsigned long newflags)
149 {
150         struct mm_struct *mm = vma->vm_mm;
151         unsigned long oldflags = vma->vm_flags;
152         long nrpages = (end - start) >> PAGE_SHIFT;
153         unsigned long charged = 0;
154         pgoff_t pgoff;
155         int error;
156         int dirty_accountable = 0;
157
158         if (newflags == oldflags) {
159                 *pprev = vma;
160                 return 0;
161         }
162
163         /*
164          * If we make a private mapping writable we increase our commit;
165          * but (without finer accounting) cannot reduce our commit if we
166          * make it unwritable again. hugetlb mapping were accounted for
167          * even if read-only so there is no need to account for them here
168          */
169         if (newflags & VM_WRITE) {
170                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
171                                                 VM_SHARED|VM_NORESERVE))) {
172                         charged = nrpages;
173                         if (security_vm_enough_memory(charged))
174                                 return -ENOMEM;
175                         newflags |= VM_ACCOUNT;
176                 }
177         }
178
179         /*
180          * First try to merge with previous and/or next vma.
181          */
182         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
183         *pprev = vma_merge(mm, *pprev, start, end, newflags,
184                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
185         if (*pprev) {
186                 vma = *pprev;
187                 goto success;
188         }
189
190         *pprev = vma;
191
192         if (start != vma->vm_start) {
193                 error = split_vma(mm, vma, start, 1);
194                 if (error)
195                         goto fail;
196         }
197
198         if (end != vma->vm_end) {
199                 error = split_vma(mm, vma, end, 0);
200                 if (error)
201                         goto fail;
202         }
203
204 success:
205         /*
206          * vm_flags and vm_page_prot are protected by the mmap_sem
207          * held in write mode.
208          */
209         vma->vm_flags = newflags;
210         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
211                                           vm_get_page_prot(newflags));
212
213         if (vma_wants_writenotify(vma)) {
214                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
215                 dirty_accountable = 1;
216         }
217
218         mmu_notifier_invalidate_range_start(mm, start, end);
219         if (is_vm_hugetlb_page(vma))
220                 hugetlb_change_protection(vma, start, end, vma->vm_page_prot);
221         else
222                 change_protection(vma, start, end, vma->vm_page_prot, dirty_accountable);
223         mmu_notifier_invalidate_range_end(mm, start, end);
224         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
225         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
226         perf_event_mmap(vma);
227         return 0;
228
229 fail:
230         vm_unacct_memory(charged);
231         return error;
232 }
233
234 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
235                 unsigned long, prot)
236 {
237         unsigned long vm_flags, nstart, end, tmp, reqprot;
238         struct vm_area_struct *vma, *prev;
239         int error = -EINVAL;
240         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
241         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
242         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
243                 return -EINVAL;
244
245         if (start & ~PAGE_MASK)
246                 return -EINVAL;
247         if (!len)
248                 return 0;
249         len = PAGE_ALIGN(len);
250         end = start + len;
251         if (end <= start)
252                 return -ENOMEM;
253         if (!arch_validate_prot(prot))
254                 return -EINVAL;
255
256         reqprot = prot;
257         /*
258          * Does the application expect PROT_READ to imply PROT_EXEC:
259          */
260         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
261                 prot |= PROT_EXEC;
262
263         vm_flags = calc_vm_prot_bits(prot);
264
265         down_write(&current->mm->mmap_sem);
266
267         vma = find_vma_prev(current->mm, start, &prev);
268         error = -ENOMEM;
269         if (!vma)
270                 goto out;
271         if (unlikely(grows & PROT_GROWSDOWN)) {
272                 if (vma->vm_start >= end)
273                         goto out;
274                 start = vma->vm_start;
275                 error = -EINVAL;
276                 if (!(vma->vm_flags & VM_GROWSDOWN))
277                         goto out;
278         }
279         else {
280                 if (vma->vm_start > start)
281                         goto out;
282                 if (unlikely(grows & PROT_GROWSUP)) {
283                         end = vma->vm_end;
284                         error = -EINVAL;
285                         if (!(vma->vm_flags & VM_GROWSUP))
286                                 goto out;
287                 }
288         }
289         if (start > vma->vm_start)
290                 prev = vma;
291
292         for (nstart = start ; ; ) {
293                 unsigned long newflags;
294
295                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
296
297                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
298
299                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
300                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
301                         error = -EACCES;
302                         goto out;
303                 }
304
305                 error = security_file_mprotect(vma, reqprot, prot);
306                 if (error)
307                         goto out;
308
309                 tmp = vma->vm_end;
310                 if (tmp > end)
311                         tmp = end;
312                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
313                 if (error)
314                         goto out;
315                 nstart = tmp;
316
317                 if (nstart < prev->vm_end)
318                         nstart = prev->vm_end;
319                 if (nstart >= end)
320                         goto out;
321
322                 vma = prev->vm_next;
323                 if (!vma || vma->vm_start != nstart) {
324                         error = -ENOMEM;
325                         goto out;
326                 }
327         }
328 out:
329         up_write(&current->mm->mmap_sem);
330         return error;
331 }