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 void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
  29                 unsigned long addr, unsigned long end, pgprot_t newprot)
  30 {
  31         pte_t *pte;
  32
  33         pte = pte_offset_map(pmd, addr);
  34         do {
  35                 if (pte_present(*pte)) {
  36                         pte_t ptent;
  37
  38                         /* Avoid an SMP race with hardware updated dirty/clean
  39                          * bits by wiping the pte and then setting the new pte
  40                          * into place.
  41                          */
  42                         ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
  43                         set_pte_at(mm, addr, pte, ptent);
  44                         lazy_mmu_prot_update(ptent);
  45                 }
  46         } while (pte++, addr += PAGE_SIZE, addr != end);
  47         pte_unmap(pte - 1);
  48 }
  49
  50 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
  51                 unsigned long addr, unsigned long end, pgprot_t newprot)
  52 {
  53         pmd_t *pmd;
  54         unsigned long next;
  55
  56         pmd = pmd_offset(pud, addr);
  57         do {
  58                 next = pmd_addr_end(addr, end);
  59                 if (pmd_none_or_clear_bad(pmd))
  60                         continue;
  61                 change_pte_range(mm, pmd, addr, next, newprot);
  62         } while (pmd++, addr = next, addr != end);
  63 }
  64
  65 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
  66                 unsigned long addr, unsigned long end, pgprot_t newprot)
  67 {
  68         pud_t *pud;
  69         unsigned long next;
  70
  71         pud = pud_offset(pgd, addr);
  72         do {
  73                 next = pud_addr_end(addr, end);
  74                 if (pud_none_or_clear_bad(pud))
  75                         continue;
  76                 change_pmd_range(mm, pud, addr, next, newprot);
  77         } while (pud++, addr = next, addr != end);
  78 }
  79
  80 static void change_protection(struct vm_area_struct *vma,
  81                 unsigned long addr, unsigned long end, pgprot_t newprot)
  82 {
  83         struct mm_struct *mm = vma->vm_mm;
  84         pgd_t *pgd;
  85         unsigned long next;
  86         unsigned long start = addr;
  87
  88         BUG_ON(addr >= end);
  89         pgd = pgd_offset(mm, addr);
  90         flush_cache_range(vma, addr, end);
  91         spin_lock(&mm->page_table_lock);
  92         do {
  93                 next = pgd_addr_end(addr, end);
  94                 if (pgd_none_or_clear_bad(pgd))
  95                         continue;
  96                 change_pud_range(mm, pgd, addr, next, newprot);
  97         } while (pgd++, addr = next, addr != end);
  98         flush_tlb_range(vma, start, end);
  99         spin_unlock(&mm->page_table_lock);
 100 }
 101
 102 static int
 103 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 104         unsigned long start, unsigned long end, unsigned long newflags)
 105 {
 106         struct mm_struct *mm = vma->vm_mm;
 107         unsigned long oldflags = vma->vm_flags;
 108         long nrpages = (end - start) >> PAGE_SHIFT;
 109         unsigned long charged = 0;
 110         pgprot_t newprot;
 111         pgoff_t pgoff;
 112         int error;
 113
 114         if (newflags == oldflags) {
 115                 *pprev = vma;
 116                 return 0;
 117         }
 118
 119         /*
 120          * If we make a private mapping writable we increase our commit;
 121          * but (without finer accounting) cannot reduce our commit if we
 122          * make it unwritable again.
 123          *
 124          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
 125          * a MAP_NORESERVE private mapping to writable will now reserve.
 126          */
 127         if (newflags & VM_WRITE) {
 128                 if (oldflags & VM_RESERVED) {
 129                         BUG_ON(oldflags & VM_WRITE);
 130                         printk(KERN_WARNING "program %s is using MAP_PRIVATE, "
 131                                 "PROT_WRITE mprotect of VM_RESERVED memory, "
 132                                 "which is deprecated. Please report this to "
 133                                 "linux-kernel@vger.kernel.org\n",current->comm);
 134                         return -EACCES;
 135                 }
 136                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED|VM_HUGETLB))) {
 137                         charged = nrpages;
 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         *pprev = vma;
 158
 159         if (start != vma->vm_start) {
 160                 error = split_vma(mm, vma, start, 1);
 161                 if (error)
 162                         goto fail;
 163         }
 164
 165         if (end != vma->vm_end) {
 166                 error = split_vma(mm, vma, end, 0);
 167                 if (error)
 168                         goto fail;
 169         }
 170
 171 success:
 172         /*
 173          * vm_flags and vm_page_prot are protected by the mmap_sem
 174          * held in write mode.
 175          */
 176         vma->vm_flags = newflags;
 177         vma->vm_page_prot = newprot;
 178         change_protection(vma, start, end, newprot);
 179         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 180         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 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, reqprot;
 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         if (!len)
 202                 return 0;
 203         len = PAGE_ALIGN(len);
 204         end = start + len;
 205         if (end <= start)
 206                 return -ENOMEM;
 207         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
 208                 return -EINVAL;
 209
 210         reqprot = prot;
 211         /*
 212          * Does the application expect PROT_READ to imply PROT_EXEC:
 213          */
 214         if (unlikely((prot & PROT_READ) &&
 215                         (current->personality & READ_IMPLIES_EXEC)))
 216                 prot |= PROT_EXEC;
 217
 218         vm_flags = calc_vm_prot_bits(prot);
 219
 220         down_write(&current->mm->mmap_sem);
 221
 222         vma = find_vma_prev(current->mm, start, &prev);
 223         error = -ENOMEM;
 224         if (!vma)
 225                 goto out;
 226         if (unlikely(grows & PROT_GROWSDOWN)) {
 227                 if (vma->vm_start >= end)
 228                         goto out;
 229                 start = vma->vm_start;
 230                 error = -EINVAL;
 231                 if (!(vma->vm_flags & VM_GROWSDOWN))
 232                         goto out;
 233         }
 234         else {
 235                 if (vma->vm_start > start)
 236                         goto out;
 237                 if (unlikely(grows & PROT_GROWSUP)) {
 238                         end = vma->vm_end;
 239                         error = -EINVAL;
 240                         if (!(vma->vm_flags & VM_GROWSUP))
 241                                 goto out;
 242                 }
 243         }
 244         if (start > vma->vm_start)
 245                 prev = vma;
 246
 247         for (nstart = start ; ; ) {
 248                 unsigned long newflags;
 249
 250                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 251
 252                 if (is_vm_hugetlb_page(vma)) {
 253                         error = -EACCES;
 254                         goto out;
 255                 }
 256
 257                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 258
 259                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 260                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 261                         error = -EACCES;
 262                         goto out;
 263                 }
 264
 265                 error = security_file_mprotect(vma, reqprot, 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 }