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 unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
  39                 unsigned long addr, unsigned long end, pgprot_t newprot,
  40                 int dirty_accountable, int prot_numa)
  41 {
  42         struct mm_struct *mm = vma->vm_mm;
  43         pte_t *pte, oldpte;
  44         spinlock_t *ptl;
  45         unsigned long pages = 0;
  46
  47         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  48         arch_enter_lazy_mmu_mode();
  49         do {
  50                 oldpte = *pte;
  51                 if (pte_present(oldpte)) {
  52                         pte_t ptent;
  53                         bool updated = false;
  54
  55                         ptent = ptep_modify_prot_start(mm, addr, pte);
  56                         if (!prot_numa) {
  57                                 ptent = pte_modify(ptent, newprot);
  58                                 updated = true;
  59                         } else {
  60                                 struct page *page;
  61
  62                                 page = vm_normal_page(vma, addr, oldpte);
  63                                 if (page) {
  64                                         if (!pte_numa(oldpte)) {
  65                                                 ptent = pte_mknuma(ptent);
  66                                                 updated = true;
  67                                         }
  68                                 }
  69                         }
  70
  71                         /*
  72                          * Avoid taking write faults for pages we know to be
  73                          * dirty.
  74                          */
  75                         if (dirty_accountable && pte_dirty(ptent)) {
  76                                 ptent = pte_mkwrite(ptent);
  77                                 updated = true;
  78                         }
  79
  80                         if (updated)
  81                                 pages++;
  82                         ptep_modify_prot_commit(mm, addr, pte, ptent);
  83                 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
  84                         swp_entry_t entry = pte_to_swp_entry(oldpte);
  85
  86                         if (is_write_migration_entry(entry)) {
  87                                 pte_t newpte;
  88                                 /*
  89                                  * A protection check is difficult so
  90                                  * just be safe and disable write
  91                                  */
  92                                 make_migration_entry_read(&entry);
  93                                 newpte = swp_entry_to_pte(entry);
  94                                 if (pte_swp_soft_dirty(oldpte))
  95                                         newpte = pte_swp_mksoft_dirty(newpte);
  96                                 set_pte_at(mm, addr, pte, newpte);
  97
  98                                 pages++;
  99                         }
 100                 }
 101         } while (pte++, addr += PAGE_SIZE, addr != end);
 102         arch_leave_lazy_mmu_mode();
 103         pte_unmap_unlock(pte - 1, ptl);
 104
 105         return pages;
 106 }
 107
 108 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 109                 pud_t *pud, unsigned long addr, unsigned long end,
 110                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 111 {
 112         pmd_t *pmd;
 113         unsigned long next;
 114         unsigned long pages = 0;
 115         unsigned long nr_huge_updates = 0;
 116
 117         pmd = pmd_offset(pud, addr);
 118         do {
 119                 unsigned long this_pages;
 120
 121                 next = pmd_addr_end(addr, end);
 122                 if (pmd_trans_huge(*pmd)) {
 123                         if (next - addr != HPAGE_PMD_SIZE)
 124                                 split_huge_page_pmd(vma, addr, pmd);
 125                         else {
 126                                 int nr_ptes = change_huge_pmd(vma, pmd, addr,
 127                                                 newprot, prot_numa);
 128
 129                                 if (nr_ptes) {
 130                                         if (nr_ptes == HPAGE_PMD_NR) {
 131                                                 pages += HPAGE_PMD_NR;
 132                                                 nr_huge_updates++;
 133                                         }
 134                                         continue;
 135                                 }
 136                         }
 137                         /* fall through */
 138                 }
 139                 if (pmd_none_or_clear_bad(pmd))
 140                         continue;
 141                 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 142                                  dirty_accountable, prot_numa);
 143                 pages += this_pages;
 144         } while (pmd++, addr = next, addr != end);
 145
 146         if (nr_huge_updates)
 147                 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
 148         return pages;
 149 }
 150
 151 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
 152                 pgd_t *pgd, unsigned long addr, unsigned long end,
 153                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 154 {
 155         pud_t *pud;
 156         unsigned long next;
 157         unsigned long pages = 0;
 158
 159         pud = pud_offset(pgd, addr);
 160         do {
 161                 next = pud_addr_end(addr, end);
 162                 if (pud_none_or_clear_bad(pud))
 163                         continue;
 164                 pages += change_pmd_range(vma, pud, addr, next, newprot,
 165                                  dirty_accountable, prot_numa);
 166         } while (pud++, addr = next, addr != end);
 167
 168         return pages;
 169 }
 170
 171 static unsigned long change_protection_range(struct vm_area_struct *vma,
 172                 unsigned long addr, unsigned long end, pgprot_t newprot,
 173                 int dirty_accountable, int prot_numa)
 174 {
 175         struct mm_struct *mm = vma->vm_mm;
 176         pgd_t *pgd;
 177         unsigned long next;
 178         unsigned long start = addr;
 179         unsigned long pages = 0;
 180
 181         BUG_ON(addr >= end);
 182         pgd = pgd_offset(mm, addr);
 183         flush_cache_range(vma, addr, end);
 184         do {
 185                 next = pgd_addr_end(addr, end);
 186                 if (pgd_none_or_clear_bad(pgd))
 187                         continue;
 188                 pages += change_pud_range(vma, pgd, addr, next, newprot,
 189                                  dirty_accountable, prot_numa);
 190         } while (pgd++, addr = next, addr != end);
 191
 192         /* Only flush the TLB if we actually modified any entries: */
 193         if (pages)
 194                 flush_tlb_range(vma, start, end);
 195
 196         return pages;
 197 }
 198
 199 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
 200                        unsigned long end, pgprot_t newprot,
 201                        int dirty_accountable, int prot_numa)
 202 {
 203         struct mm_struct *mm = vma->vm_mm;
 204         unsigned long pages;
 205
 206         mmu_notifier_invalidate_range_start(mm, start, end);
 207         if (is_vm_hugetlb_page(vma))
 208                 pages = hugetlb_change_protection(vma, start, end, newprot);
 209         else
 210                 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
 211         mmu_notifier_invalidate_range_end(mm, start, end);
 212
 213         return pages;
 214 }
 215
 216 int
 217 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 218         unsigned long start, unsigned long end, unsigned long newflags)
 219 {
 220         struct mm_struct *mm = vma->vm_mm;
 221         unsigned long oldflags = vma->vm_flags;
 222         long nrpages = (end - start) >> PAGE_SHIFT;
 223         unsigned long charged = 0;
 224         pgoff_t pgoff;
 225         int error;
 226         int dirty_accountable = 0;
 227
 228         if (newflags == oldflags) {
 229                 *pprev = vma;
 230                 return 0;
 231         }
 232
 233         /*
 234          * If we make a private mapping writable we increase our commit;
 235          * but (without finer accounting) cannot reduce our commit if we
 236          * make it unwritable again. hugetlb mapping were accounted for
 237          * even if read-only so there is no need to account for them here
 238          */
 239         if (newflags & VM_WRITE) {
 240                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 241                                                 VM_SHARED|VM_NORESERVE))) {
 242                         charged = nrpages;
 243                         if (security_vm_enough_memory_mm(mm, charged))
 244                                 return -ENOMEM;
 245                         newflags |= VM_ACCOUNT;
 246                 }
 247         }
 248
 249         /*
 250          * First try to merge with previous and/or next vma.
 251          */
 252         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 253         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 254                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 255         if (*pprev) {
 256                 vma = *pprev;
 257                 goto success;
 258         }
 259
 260         *pprev = vma;
 261
 262         if (start != vma->vm_start) {
 263                 error = split_vma(mm, vma, start, 1);
 264                 if (error)
 265                         goto fail;
 266         }
 267
 268         if (end != vma->vm_end) {
 269                 error = split_vma(mm, vma, end, 0);
 270                 if (error)
 271                         goto fail;
 272         }
 273
 274 success:
 275         /*
 276          * vm_flags and vm_page_prot are protected by the mmap_sem
 277          * held in write mode.
 278          */
 279         vma->vm_flags = newflags;
 280         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
 281                                           vm_get_page_prot(newflags));
 282
 283         if (vma_wants_writenotify(vma)) {
 284                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
 285                 dirty_accountable = 1;
 286         }
 287
 288         change_protection(vma, start, end, vma->vm_page_prot,
 289                           dirty_accountable, 0);
 290
 291         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 292         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 293         perf_event_mmap(vma);
 294         return 0;
 295
 296 fail:
 297         vm_unacct_memory(charged);
 298         return error;
 299 }
 300
 301 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 302                 unsigned long, prot)
 303 {
 304         unsigned long vm_flags, nstart, end, tmp, reqprot;
 305         struct vm_area_struct *vma, *prev;
 306         int error = -EINVAL;
 307         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 308         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 309         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 310                 return -EINVAL;
 311
 312         if (start & ~PAGE_MASK)
 313                 return -EINVAL;
 314         if (!len)
 315                 return 0;
 316         len = PAGE_ALIGN(len);
 317         end = start + len;
 318         if (end <= start)
 319                 return -ENOMEM;
 320         if (!arch_validate_prot(prot))
 321                 return -EINVAL;
 322
 323         reqprot = prot;
 324         /*
 325          * Does the application expect PROT_READ to imply PROT_EXEC:
 326          */
 327         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 328                 prot |= PROT_EXEC;
 329
 330         vm_flags = calc_vm_prot_bits(prot);
 331
 332         down_write(&current->mm->mmap_sem);
 333
 334         vma = find_vma(current->mm, start);
 335         error = -ENOMEM;
 336         if (!vma)
 337                 goto out;
 338         prev = vma->vm_prev;
 339         if (unlikely(grows & PROT_GROWSDOWN)) {
 340                 if (vma->vm_start >= end)
 341                         goto out;
 342                 start = vma->vm_start;
 343                 error = -EINVAL;
 344                 if (!(vma->vm_flags & VM_GROWSDOWN))
 345                         goto out;
 346         } else {
 347                 if (vma->vm_start > start)
 348                         goto out;
 349                 if (unlikely(grows & PROT_GROWSUP)) {
 350                         end = vma->vm_end;
 351                         error = -EINVAL;
 352                         if (!(vma->vm_flags & VM_GROWSUP))
 353                                 goto out;
 354                 }
 355         }
 356         if (start > vma->vm_start)
 357                 prev = vma;
 358
 359         for (nstart = start ; ; ) {
 360                 unsigned long newflags;
 361
 362                 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
 363
 364                 newflags = vm_flags;
 365                 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 366
 367                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 368                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 369                         error = -EACCES;
 370                         goto out;
 371                 }
 372
 373                 error = security_file_mprotect(vma, reqprot, prot);
 374                 if (error)
 375                         goto out;
 376
 377                 tmp = vma->vm_end;
 378                 if (tmp > end)
 379                         tmp = end;
 380                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 381                 if (error)
 382                         goto out;
 383                 nstart = tmp;
 384
 385                 if (nstart < prev->vm_end)
 386                         nstart = prev->vm_end;
 387                 if (nstart >= end)
 388                         goto out;
 389
 390                 vma = prev->vm_next;
 391                 if (!vma || vma->vm_start != nstart) {
 392                         error = -ENOMEM;
 393                         goto out;
 394                 }
 395         }
 396 out:
 397         up_write(&current->mm->mmap_sem);
 398         return error;
 399 }