kernel/linux/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@redhat.com>
   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/slab.h>
  14 #include <linux/shm.h>
  15 #include <linux/mman.h>
  16 #include <linux/fs.h>
  17 #include <linux/highmem.h>
  18 #include <linux/security.h>
  19 #include <linux/mempolicy.h>
  20 #include <linux/personality.h>
  21
  22 #include <asm/uaccess.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/cacheflush.h>
  25 #include <asm/tlbflush.h>
  26
  27 static inline void
  28 change_pte_range(pmd_t *pmd, unsigned long address,
  29                 unsigned long size, pgprot_t newprot)
  30 {
  31         pte_t * pte;
  32         unsigned long end;
  33
  34         if (pmd_none(*pmd))
  35                 return;
  36         if (pmd_bad(*pmd)) {
  37                 pmd_ERROR(*pmd);
  38                 pmd_clear(pmd);
  39                 return;
  40         }
  41         pte = pte_offset_map(pmd, address);
  42         address &= ~PMD_MASK;
  43         end = address + size;
  44         if (end > PMD_SIZE)
  45                 end = PMD_SIZE;
  46         do {
  47                 if (pte_present(*pte)) {
  48                         pte_t entry;
  49
  50                         /* Avoid an SMP race with hardware updated dirty/clean
  51                          * bits by wiping the pte and then setting the new pte
  52                          * into place.
  53                          */
  54                         entry = ptep_get_and_clear(pte);
  55                         set_pte(pte, pte_modify(entry, newprot));
  56                 }
  57                 address += PAGE_SIZE;
  58                 pte++;
  59         } while (address && (address < end));
  60         pte_unmap(pte - 1);
  61 }
  62
  63 static inline void
  64 change_pmd_range(pgd_t *pgd, unsigned long address,
  65                 unsigned long size, pgprot_t newprot)
  66 {
  67         pmd_t * pmd;
  68         unsigned long end;
  69
  70         if (pgd_none(*pgd))
  71                 return;
  72         if (pgd_bad(*pgd)) {
  73                 pgd_ERROR(*pgd);
  74                 pgd_clear(pgd);
  75                 return;
  76         }
  77         pmd = pmd_offset(pgd, address);
  78         address &= ~PGDIR_MASK;
  79         end = address + size;
  80         if (end > PGDIR_SIZE)
  81                 end = PGDIR_SIZE;
  82         do {
  83                 change_pte_range(pmd, address, end - address, newprot);
  84                 address = (address + PMD_SIZE) & PMD_MASK;
  85                 pmd++;
  86         } while (address && (address < end));
  87 }
  88
  89 static void
  90 change_protection(struct vm_area_struct *vma, unsigned long start,
  91                 unsigned long end, pgprot_t newprot)
  92 {
  93         pgd_t *dir;
  94         unsigned long beg = start;
  95
  96         dir = pgd_offset(current->mm, start);
  97         flush_cache_range(vma, beg, end);
  98         if (start >= end)
  99                 BUG();
 100         spin_lock(&current->mm->page_table_lock);
 101         do {
 102                 change_pmd_range(dir, start, end - start, newprot);
 103                 start = (start + PGDIR_SIZE) & PGDIR_MASK;
 104                 dir++;
 105         } while (start && (start < end));
 106         flush_tlb_range(vma, beg, end);
 107         spin_unlock(&current->mm->page_table_lock);
 108         return;
 109 }
 110
 111 static int
 112 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 113         unsigned long start, unsigned long end, unsigned int newflags)
 114 {
 115         struct mm_struct * mm = vma->vm_mm;
 116         unsigned long charged = 0;
 117         pgprot_t newprot;
 118         pgoff_t pgoff;
 119         int error;
 120
 121         if (newflags == vma->vm_flags) {
 122                 *pprev = vma;
 123                 return 0;
 124         }
 125
 126         /*
 127          * If we make a private mapping writable we increase our commit;
 128          * but (without finer accounting) cannot reduce our commit if we
 129          * make it unwritable again.
 130          *
 131          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
 132          * a MAP_NORESERVE private mapping to writable will now reserve.
 133          */
 134         if (newflags & VM_WRITE) {
 135                 if (!(vma->vm_flags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
 136                         charged = (end - start) >> PAGE_SHIFT;
 137                         if (security_vm_enough_memory(charged))
 138                                 return -ENOMEM;
 139                         newflags |= VM_ACCOUNT;
 140                 }
 141         }
 142
 143         newprot = protection_map[newflags & 0xf];
 144
 145         /*
 146          * First try to merge with previous and/or next vma.
 147          */
 148         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 149         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 150                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 151         if (*pprev) {
 152                 vma = *pprev;
 153                 goto success;
 154         }
 155
 156         if (start != vma->vm_start) {
 157                 error = split_vma(mm, vma, start, 1);
 158                 if (error)
 159                         goto fail;
 160         }
 161         /*
 162          * Unless it returns an error, this function always sets *pprev to
 163          * the first vma for which vma->vm_end >= end.
 164          */
 165         *pprev = vma;
 166
 167         if (end != vma->vm_end) {
 168                 error = split_vma(mm, vma, end, 0);
 169                 if (error)
 170                         goto fail;
 171         }
 172
 173 success:
 174         /*
 175          * vm_flags and vm_page_prot are protected by the mmap_sem
 176          * held in write mode.
 177          */
 178         vma->vm_flags = newflags;
 179         vma->vm_page_prot = newprot;
 180         change_protection(vma, start, end, newprot);
 181         return 0;
 182
 183 fail:
 184         vm_unacct_memory(charged);
 185         return error;
 186 }
 187
 188 asmlinkage long
 189 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 190 {
 191         unsigned long vm_flags, nstart, end, tmp;
 192         struct vm_area_struct *vma, *prev;
 193         int error = -EINVAL;
 194         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 195         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 196         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 197                 return -EINVAL;
 198
 199         if (start & ~PAGE_MASK)
 200                 return -EINVAL;
 201         len = PAGE_ALIGN(len);
 202         end = start + len;
 203         if (end < start)
 204                 return -ENOMEM;
 205         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
 206                 return -EINVAL;
 207         if (end == start)
 208                 return 0;
 209         /*
 210          * Does the application expect PROT_READ to imply PROT_EXEC:
 211          */
 212         if (unlikely((prot & PROT_READ) &&
 213                         (current->personality & READ_IMPLIES_EXEC)))
 214                 prot |= PROT_EXEC;
 215
 216         vm_flags = calc_vm_prot_bits(prot);
 217
 218         down_write(&current->mm->mmap_sem);
 219
 220         vma = find_vma_prev(current->mm, start, &prev);
 221         error = -ENOMEM;
 222         if (!vma)
 223                 goto out;
 224         if (unlikely(grows & PROT_GROWSDOWN)) {
 225                 if (vma->vm_start >= end)
 226                         goto out;
 227                 start = vma->vm_start;
 228                 error = -EINVAL;
 229                 if (!(vma->vm_flags & VM_GROWSDOWN))
 230                         goto out;
 231         }
 232         else {
 233                 if (vma->vm_start > start)
 234                         goto out;
 235                 if (unlikely(grows & PROT_GROWSUP)) {
 236                         end = vma->vm_end;
 237                         error = -EINVAL;
 238                         if (!(vma->vm_flags & VM_GROWSUP))
 239                                 goto out;
 240                 }
 241         }
 242         if (start > vma->vm_start)
 243                 prev = vma;
 244
 245         for (nstart = start ; ; ) {
 246                 unsigned int newflags;
 247
 248                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 249
 250                 if (is_vm_hugetlb_page(vma)) {
 251                         error = -EACCES;
 252                         goto out;
 253                 }
 254
 255                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 256
 257                 if ((newflags & ~(newflags >> 4)) & 0xf) {
 258                         error = -EACCES;
 259                         goto out;
 260                 }
 261
 262                 error = security_file_mprotect(vma, prot);
 263                 if (error)
 264                         goto out;
 265
 266                 tmp = vma->vm_end;
 267                 if (tmp > end)
 268                         tmp = end;
 269                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 270                 if (error)
 271                         goto out;
 272                 nstart = tmp;
 273
 274                 if (nstart < prev->vm_end)
 275                         nstart = prev->vm_end;
 276                 if (nstart >= end)
 277                         goto out;
 278
 279                 vma = prev->vm_next;
 280                 if (!vma || vma->vm_start != nstart) {
 281                         error = -ENOMEM;
 282                         goto out;
 283                 }
 284         }
 285 out:
 286         up_write(&current->mm->mmap_sem);
 287         return error;
 288 }