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 #include <linux/syscalls.h>
  22
  23 #include <asm/uaccess.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/cacheflush.h>
  26 #include <asm/tlbflush.h>
  27
  28 static inline void
  29 change_pte_range(pmd_t *pmd, unsigned long address,
  30                 unsigned long size, pgprot_t newprot)
  31 {
  32         pte_t * pte;
  33         unsigned long end;
  34
  35         if (pmd_none(*pmd))
  36                 return;
  37         if (pmd_bad(*pmd)) {
  38                 pmd_ERROR(*pmd);
  39                 pmd_clear(pmd);
  40                 return;
  41         }
  42         pte = pte_offset_map(pmd, address);
  43         address &= ~PMD_MASK;
  44         end = address + size;
  45         if (end > PMD_SIZE)
  46                 end = PMD_SIZE;
  47         do {
  48                 if (pte_present(*pte)) {
  49                         pte_t entry;
  50
  51                         /* Avoid an SMP race with hardware updated dirty/clean
  52                          * bits by wiping the pte and then setting the new pte
  53                          * into place.
  54                          */
  55                         entry = ptep_get_and_clear(pte);
  56                         set_pte(pte, pte_modify(entry, newprot));
  57                 }
  58                 address += PAGE_SIZE;
  59                 pte++;
  60         } while (address && (address < end));
  61         pte_unmap(pte - 1);
  62 }
  63
  64 static inline void
  65 change_pmd_range(pgd_t *pgd, unsigned long address,
  66                 unsigned long size, pgprot_t newprot)
  67 {
  68         pmd_t * pmd;
  69         unsigned long end;
  70
  71         if (pgd_none(*pgd))
  72                 return;
  73         if (pgd_bad(*pgd)) {
  74                 pgd_ERROR(*pgd);
  75                 pgd_clear(pgd);
  76                 return;
  77         }
  78         pmd = pmd_offset(pgd, address);
  79         address &= ~PGDIR_MASK;
  80         end = address + size;
  81         if (end > PGDIR_SIZE)
  82                 end = PGDIR_SIZE;
  83         do {
  84                 change_pte_range(pmd, address, end - address, newprot);
  85                 address = (address + PMD_SIZE) & PMD_MASK;
  86                 pmd++;
  87         } while (address && (address < end));
  88 }
  89
  90 static void
  91 change_protection(struct vm_area_struct *vma, unsigned long start,
  92                 unsigned long end, pgprot_t newprot)
  93 {
  94         pgd_t *dir;
  95         unsigned long beg = start;
  96
  97         dir = pgd_offset(current->mm, start);
  98         flush_cache_range(vma, beg, end);
  99         if (start >= end)
 100                 BUG();
 101         spin_lock(&current->mm->page_table_lock);
 102         do {
 103                 change_pmd_range(dir, start, end - start, newprot);
 104                 start = (start + PGDIR_SIZE) & PGDIR_MASK;
 105                 dir++;
 106         } while (start && (start < end));
 107         flush_tlb_range(vma, beg, end);
 108         spin_unlock(&current->mm->page_table_lock);
 109         return;
 110 }
 111
 112 static int
 113 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 114         unsigned long start, unsigned long end, unsigned int newflags)
 115 {
 116         struct mm_struct * mm = vma->vm_mm;
 117         unsigned long charged = 0;
 118         pgprot_t newprot;
 119         pgoff_t pgoff;
 120         int error;
 121
 122         if (newflags == vma->vm_flags) {
 123                 *pprev = vma;
 124                 return 0;
 125         }
 126
 127         /*
 128          * If we make a private mapping writable we increase our commit;
 129          * but (without finer accounting) cannot reduce our commit if we
 130          * make it unwritable again.
 131          *
 132          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
 133          * a MAP_NORESERVE private mapping to writable will now reserve.
 134          */
 135         if (newflags & VM_WRITE) {
 136                 if (!(vma->vm_flags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
 137                         charged = (end - start) >> PAGE_SHIFT;
 138                         if (security_vm_enough_memory(charged))
 139                                 return -ENOMEM;
 140                         newflags |= VM_ACCOUNT;
 141                 }
 142         }
 143
 144         newprot = protection_map[newflags & 0xf];
 145
 146         /*
 147          * First try to merge with previous and/or next vma.
 148          */
 149         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 150         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 151                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 152         if (*pprev) {
 153                 vma = *pprev;
 154                 goto success;
 155         }
 156
 157         if (start != vma->vm_start) {
 158                 error = split_vma(mm, vma, start, 1);
 159                 if (error)
 160                         goto fail;
 161         }
 162         /*
 163          * Unless it returns an error, this function always sets *pprev to
 164          * the first vma for which vma->vm_end >= end.
 165          */
 166         *pprev = vma;
 167
 168         if (end != vma->vm_end) {
 169                 error = split_vma(mm, vma, end, 0);
 170                 if (error)
 171                         goto fail;
 172         }
 173
 174 success:
 175         /*
 176          * vm_flags and vm_page_prot are protected by the mmap_sem
 177          * held in write mode.
 178          */
 179         vm_stat_unaccount(vma);
 180         vma->vm_flags = newflags;
 181         vma->vm_page_prot = newprot;
 182         change_protection(vma, start, end, newprot);
 183         vm_stat_account(vma);
 184         return 0;
 185
 186 fail:
 187         vm_unacct_memory(charged);
 188         return error;
 189 }
 190
 191 asmlinkage long
 192 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 193 {
 194         unsigned long vm_flags, nstart, end, tmp;
 195         struct vm_area_struct *vma, *prev;
 196         int error = -EINVAL;
 197         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 198         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 199         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 200                 return -EINVAL;
 201
 202         if (start & ~PAGE_MASK)
 203                 return -EINVAL;
 204         len = PAGE_ALIGN(len);
 205         end = start + len;
 206         if (end < start)
 207                 return -ENOMEM;
 208         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
 209                 return -EINVAL;
 210         if (end == start)
 211                 return 0;
 212         /*
 213          * Does the application expect PROT_READ to imply PROT_EXEC:
 214          */
 215         if (unlikely((prot & PROT_READ) &&
 216                         (current->personality & READ_IMPLIES_EXEC)))
 217                 prot |= PROT_EXEC;
 218
 219         vm_flags = calc_vm_prot_bits(prot);
 220
 221         down_write(&current->mm->mmap_sem);
 222
 223         vma = find_vma_prev(current->mm, start, &prev);
 224         error = -ENOMEM;
 225         if (!vma)
 226                 goto out;
 227         if (unlikely(grows & PROT_GROWSDOWN)) {
 228                 if (vma->vm_start >= end)
 229                         goto out;
 230                 start = vma->vm_start;
 231                 error = -EINVAL;
 232                 if (!(vma->vm_flags & VM_GROWSDOWN))
 233                         goto out;
 234         }
 235         else {
 236                 if (vma->vm_start > start)
 237                         goto out;
 238                 if (unlikely(grows & PROT_GROWSUP)) {
 239                         end = vma->vm_end;
 240                         error = -EINVAL;
 241                         if (!(vma->vm_flags & VM_GROWSUP))
 242                                 goto out;
 243                 }
 244         }
 245         if (start > vma->vm_start)
 246                 prev = vma;
 247
 248         for (nstart = start ; ; ) {
 249                 unsigned int newflags;
 250
 251                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 252
 253                 if (is_vm_hugetlb_page(vma)) {
 254                         error = -EACCES;
 255                         goto out;
 256                 }
 257
 258                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 259
 260                 if ((newflags & ~(newflags >> 4)) & 0xf) {
 261                         error = -EACCES;
 262                         goto out;
 263                 }
 264
 265                 error = security_file_mprotect(vma, prot);
 266                 if (error)
 267                         goto out;
 268
 269                 tmp = vma->vm_end;
 270                 if (tmp > end)
 271                         tmp = end;
 272                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 273                 if (error)
 274                         goto out;
 275                 nstart = tmp;
 276
 277                 if (nstart < prev->vm_end)
 278                         nstart = prev->vm_end;
 279                 if (nstart >= end)
 280                         goto out;
 281
 282                 vma = prev->vm_next;
 283                 if (!vma || vma->vm_start != nstart) {
 284                         error = -ENOMEM;
 285                         goto out;
 286                 }
 287         }
 288 out:
 289         up_write(&current->mm->mmap_sem);
 290         return error;
 291 }