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 #include <linux/swap.h>
  23 #include <linux/swapops.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/cacheflush.h>
  27 #include <asm/tlbflush.h>
  28
  29 static void change_pte_range(struct mm_struct *mm, pmd_t *pmd,
  30                 unsigned long addr, unsigned long end, pgprot_t newprot)
  31 {
  32         pte_t *pte, oldpte;
  33         spinlock_t *ptl;
  34
  35         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  36         do {
  37                 oldpte = *pte;
  38                 if (pte_present(oldpte)) {
  39                         pte_t ptent;
  40
  41                         /* Avoid an SMP race with hardware updated dirty/clean
  42                          * bits by wiping the pte and then setting the new pte
  43                          * into place.
  44                          */
  45                         ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
  46                         set_pte_at(mm, addr, pte, ptent);
  47                         lazy_mmu_prot_update(ptent);
  48 #ifdef CONFIG_MIGRATION
  49                 } else if (!pte_file(oldpte)) {
  50                         swp_entry_t entry = pte_to_swp_entry(oldpte);
  51
  52                         if (is_write_migration_entry(entry)) {
  53                                 /*
  54                                  * A protection check is difficult so
  55                                  * just be safe and disable write
  56                                  */
  57                                 make_migration_entry_read(&entry);
  58                                 set_pte_at(mm, addr, pte,
  59                                         swp_entry_to_pte(entry));
  60                         }
  61 #endif
  62                 }
  63
  64         } while (pte++, addr += PAGE_SIZE, addr != end);
  65         pte_unmap_unlock(pte - 1, ptl);
  66 }
  67
  68 static inline void change_pmd_range(struct mm_struct *mm, pud_t *pud,
  69                 unsigned long addr, unsigned long end, pgprot_t newprot)
  70 {
  71         pmd_t *pmd;
  72         unsigned long next;
  73
  74         pmd = pmd_offset(pud, addr);
  75         do {
  76                 next = pmd_addr_end(addr, end);
  77                 if (pmd_none_or_clear_bad(pmd))
  78                         continue;
  79                 change_pte_range(mm, pmd, addr, next, newprot);
  80         } while (pmd++, addr = next, addr != end);
  81 }
  82
  83 static inline void change_pud_range(struct mm_struct *mm, pgd_t *pgd,
  84                 unsigned long addr, unsigned long end, pgprot_t newprot)
  85 {
  86         pud_t *pud;
  87         unsigned long next;
  88
  89         pud = pud_offset(pgd, addr);
  90         do {
  91                 next = pud_addr_end(addr, end);
  92                 if (pud_none_or_clear_bad(pud))
  93                         continue;
  94                 change_pmd_range(mm, pud, addr, next, newprot);
  95         } while (pud++, addr = next, addr != end);
  96 }
  97
  98 static void change_protection(struct vm_area_struct *vma,
  99                 unsigned long addr, unsigned long end, pgprot_t newprot)
 100 {
 101         struct mm_struct *mm = vma->vm_mm;
 102         pgd_t *pgd;
 103         unsigned long next;
 104         unsigned long start = addr;
 105
 106         BUG_ON(addr >= end);
 107         pgd = pgd_offset(mm, addr);
 108         flush_cache_range(vma, addr, end);
 109         do {
 110                 next = pgd_addr_end(addr, end);
 111                 if (pgd_none_or_clear_bad(pgd))
 112                         continue;
 113                 change_pud_range(mm, pgd, addr, next, newprot);
 114         } while (pgd++, addr = next, addr != end);
 115         flush_tlb_range(vma, start, end);
 116 }
 117
 118 static int
 119 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 120         unsigned long start, unsigned long end, unsigned long newflags)
 121 {
 122         struct mm_struct *mm = vma->vm_mm;
 123         unsigned long oldflags = vma->vm_flags;
 124         long nrpages = (end - start) >> PAGE_SHIFT;
 125         unsigned long charged = 0;
 126         pgprot_t newprot;
 127         pgoff_t pgoff;
 128         int error;
 129
 130         if (newflags == oldflags) {
 131                 *pprev = vma;
 132                 return 0;
 133         }
 134
 135         /*
 136          * If we make a private mapping writable we increase our commit;
 137          * but (without finer accounting) cannot reduce our commit if we
 138          * make it unwritable again.
 139          *
 140          * FIXME? We haven't defined a VM_NORESERVE flag, so mprotecting
 141          * a MAP_NORESERVE private mapping to writable will now reserve.
 142          */
 143         if (newflags & VM_WRITE) {
 144                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_SHARED))) {
 145                         charged = nrpages;
 146                         if (security_vm_enough_memory(charged))
 147                                 return -ENOMEM;
 148                         newflags |= VM_ACCOUNT;
 149                 }
 150         }
 151
 152         newprot = protection_map[newflags & 0xf];
 153
 154         /*
 155          * First try to merge with previous and/or next vma.
 156          */
 157         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 158         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 159                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 160         if (*pprev) {
 161                 vma = *pprev;
 162                 goto success;
 163         }
 164
 165         *pprev = vma;
 166
 167         if (start != vma->vm_start) {
 168                 error = split_vma(mm, vma, start, 1);
 169                 if (error)
 170                         goto fail;
 171         }
 172
 173         if (end != vma->vm_end) {
 174                 error = split_vma(mm, vma, end, 0);
 175                 if (error)
 176                         goto fail;
 177         }
 178
 179 success:
 180         /*
 181          * vm_flags and vm_page_prot are protected by the mmap_sem
 182          * held in write mode.
 183          */
 184         vma->vm_flags = newflags;
 185         vma->vm_page_prot = newprot;
 186         if (is_vm_hugetlb_page(vma))
 187                 hugetlb_change_protection(vma, start, end, newprot);
 188         else
 189                 change_protection(vma, start, end, newprot);
 190         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 191         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 192         return 0;
 193
 194 fail:
 195         vm_unacct_memory(charged);
 196         return error;
 197 }
 198
 199 asmlinkage long
 200 sys_mprotect(unsigned long start, size_t len, unsigned long prot)
 201 {
 202         unsigned long vm_flags, nstart, end, tmp, reqprot;
 203         struct vm_area_struct *vma, *prev;
 204         int error = -EINVAL;
 205         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 206         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 207         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 208                 return -EINVAL;
 209
 210         if (start & ~PAGE_MASK)
 211                 return -EINVAL;
 212         if (!len)
 213                 return 0;
 214         len = PAGE_ALIGN(len);
 215         end = start + len;
 216         if (end <= start)
 217                 return -ENOMEM;
 218         if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_SEM))
 219                 return -EINVAL;
 220
 221         reqprot = prot;
 222         /*
 223          * Does the application expect PROT_READ to imply PROT_EXEC:
 224          */
 225         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 226                 prot |= PROT_EXEC;
 227
 228         vm_flags = calc_vm_prot_bits(prot);
 229
 230         down_write(&current->mm->mmap_sem);
 231
 232         vma = find_vma_prev(current->mm, start, &prev);
 233         error = -ENOMEM;
 234         if (!vma)
 235                 goto out;
 236         if (unlikely(grows & PROT_GROWSDOWN)) {
 237                 if (vma->vm_start >= end)
 238                         goto out;
 239                 start = vma->vm_start;
 240                 error = -EINVAL;
 241                 if (!(vma->vm_flags & VM_GROWSDOWN))
 242                         goto out;
 243         }
 244         else {
 245                 if (vma->vm_start > start)
 246                         goto out;
 247                 if (unlikely(grows & PROT_GROWSUP)) {
 248                         end = vma->vm_end;
 249                         error = -EINVAL;
 250                         if (!(vma->vm_flags & VM_GROWSUP))
 251                                 goto out;
 252                 }
 253         }
 254         if (start > vma->vm_start)
 255                 prev = vma;
 256
 257         for (nstart = start ; ; ) {
 258                 unsigned long newflags;
 259
 260                 /* Here we know that  vma->vm_start <= nstart < vma->vm_end. */
 261
 262                 newflags = vm_flags | (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 263
 264                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 265                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 266                         error = -EACCES;
 267                         goto out;
 268                 }
 269
 270                 error = security_file_mprotect(vma, reqprot, prot);
 271                 if (error)
 272                         goto out;
 273
 274                 tmp = vma->vm_end;
 275                 if (tmp > end)
 276                         tmp = end;
 277                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 278                 if (error)
 279                         goto out;
 280                 nstart = tmp;
 281
 282                 if (nstart < prev->vm_end)
 283                         nstart = prev->vm_end;
 284                 if (nstart >= end)
 285                         goto out;
 286
 287                 vma = prev->vm_next;
 288                 if (!vma || vma->vm_start != nstart) {
 289                         error = -ENOMEM;
 290                         goto out;
 291                 }
 292         }
 293 out:
 294         up_write(&current->mm->mmap_sem);
 295         return error;
 296 }