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