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
 116         pmd = pmd_offset(pud, addr);
 117         do {
 118                 unsigned long this_pages;
 119
 120                 next = pmd_addr_end(addr, end);
 121                 if (pmd_trans_huge(*pmd)) {
 122                         if (next - addr != HPAGE_PMD_SIZE)
 123                                 split_huge_page_pmd(vma, addr, pmd);
 124                         else {
 125                                 int nr_ptes = change_huge_pmd(vma, pmd, addr,
 126                                                 newprot, prot_numa);
 127
 128                                 if (nr_ptes) {
 129                                         if (nr_ptes == HPAGE_PMD_NR)
 130                                                 pages++;
 131
 132                                         continue;
 133                                 }
 134                         }
 135                         /* fall through */
 136                 }
 137                 if (pmd_none_or_clear_bad(pmd))
 138                         continue;
 139                 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 140                                  dirty_accountable, prot_numa);
 141                 pages += this_pages;
 142         } while (pmd++, addr = next, addr != end);
 143
 144         return pages;
 145 }
 146
 147 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
 148                 pgd_t *pgd, unsigned long addr, unsigned long end,
 149                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 150 {
 151         pud_t *pud;
 152         unsigned long next;
 153         unsigned long pages = 0;
 154
 155         pud = pud_offset(pgd, addr);
 156         do {
 157                 next = pud_addr_end(addr, end);
 158                 if (pud_none_or_clear_bad(pud))
 159                         continue;
 160                 pages += change_pmd_range(vma, pud, addr, next, newprot,
 161                                  dirty_accountable, prot_numa);
 162         } while (pud++, addr = next, addr != end);
 163
 164         return pages;
 165 }
 166
 167 static unsigned long change_protection_range(struct vm_area_struct *vma,
 168                 unsigned long addr, unsigned long end, pgprot_t newprot,
 169                 int dirty_accountable, int prot_numa)
 170 {
 171         struct mm_struct *mm = vma->vm_mm;
 172         pgd_t *pgd;
 173         unsigned long next;
 174         unsigned long start = addr;
 175         unsigned long pages = 0;
 176
 177         BUG_ON(addr >= end);
 178         pgd = pgd_offset(mm, addr);
 179         flush_cache_range(vma, addr, end);
 180         do {
 181                 next = pgd_addr_end(addr, end);
 182                 if (pgd_none_or_clear_bad(pgd))
 183                         continue;
 184                 pages += change_pud_range(vma, pgd, addr, next, newprot,
 185                                  dirty_accountable, prot_numa);
 186         } while (pgd++, addr = next, addr != end);
 187
 188         /* Only flush the TLB if we actually modified any entries: */
 189         if (pages)
 190                 flush_tlb_range(vma, start, end);
 191
 192         return pages;
 193 }
 194
 195 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
 196                        unsigned long end, pgprot_t newprot,
 197                        int dirty_accountable, int prot_numa)
 198 {
 199         struct mm_struct *mm = vma->vm_mm;
 200         unsigned long pages;
 201
 202         mmu_notifier_invalidate_range_start(mm, start, end);
 203         if (is_vm_hugetlb_page(vma))
 204                 pages = hugetlb_change_protection(vma, start, end, newprot);
 205         else
 206                 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
 207         mmu_notifier_invalidate_range_end(mm, start, end);
 208
 209         return pages;
 210 }
 211
 212 int
 213 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 214         unsigned long start, unsigned long end, unsigned long newflags)
 215 {
 216         struct mm_struct *mm = vma->vm_mm;
 217         unsigned long oldflags = vma->vm_flags;
 218         long nrpages = (end - start) >> PAGE_SHIFT;
 219         unsigned long charged = 0;
 220         pgoff_t pgoff;
 221         int error;
 222         int dirty_accountable = 0;
 223
 224         if (newflags == oldflags) {
 225                 *pprev = vma;
 226                 return 0;
 227         }
 228
 229         /*
 230          * If we make a private mapping writable we increase our commit;
 231          * but (without finer accounting) cannot reduce our commit if we
 232          * make it unwritable again. hugetlb mapping were accounted for
 233          * even if read-only so there is no need to account for them here
 234          */
 235         if (newflags & VM_WRITE) {
 236                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 237                                                 VM_SHARED|VM_NORESERVE))) {
 238                         charged = nrpages;
 239                         if (security_vm_enough_memory_mm(mm, charged))
 240                                 return -ENOMEM;
 241                         newflags |= VM_ACCOUNT;
 242                 }
 243         }
 244
 245         /*
 246          * First try to merge with previous and/or next vma.
 247          */
 248         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 249         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 250                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 251         if (*pprev) {
 252                 vma = *pprev;
 253                 goto success;
 254         }
 255
 256         *pprev = vma;
 257
 258         if (start != vma->vm_start) {
 259                 error = split_vma(mm, vma, start, 1);
 260                 if (error)
 261                         goto fail;
 262         }
 263
 264         if (end != vma->vm_end) {
 265                 error = split_vma(mm, vma, end, 0);
 266                 if (error)
 267                         goto fail;
 268         }
 269
 270 success:
 271         /*
 272          * vm_flags and vm_page_prot are protected by the mmap_sem
 273          * held in write mode.
 274          */
 275         vma->vm_flags = newflags;
 276         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
 277                                           vm_get_page_prot(newflags));
 278
 279         if (vma_wants_writenotify(vma)) {
 280                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
 281                 dirty_accountable = 1;
 282         }
 283
 284         change_protection(vma, start, end, vma->vm_page_prot,
 285                           dirty_accountable, 0);
 286
 287         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 288         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 289         perf_event_mmap(vma);
 290         return 0;
 291
 292 fail:
 293         vm_unacct_memory(charged);
 294         return error;
 295 }
 296
 297 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 298                 unsigned long, prot)
 299 {
 300         unsigned long vm_flags, nstart, end, tmp, reqprot;
 301         struct vm_area_struct *vma, *prev;
 302         int error = -EINVAL;
 303         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 304         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 305         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 306                 return -EINVAL;
 307
 308         if (start & ~PAGE_MASK)
 309                 return -EINVAL;
 310         if (!len)
 311                 return 0;
 312         len = PAGE_ALIGN(len);
 313         end = start + len;
 314         if (end <= start)
 315                 return -ENOMEM;
 316         if (!arch_validate_prot(prot))
 317                 return -EINVAL;
 318
 319         reqprot = prot;
 320         /*
 321          * Does the application expect PROT_READ to imply PROT_EXEC:
 322          */
 323         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 324                 prot |= PROT_EXEC;
 325
 326         vm_flags = calc_vm_prot_bits(prot);
 327
 328         down_write(&current->mm->mmap_sem);
 329
 330         vma = find_vma(current->mm, start);
 331         error = -ENOMEM;
 332         if (!vma)
 333                 goto out;
 334         prev = vma->vm_prev;
 335         if (unlikely(grows & PROT_GROWSDOWN)) {
 336                 if (vma->vm_start >= end)
 337                         goto out;
 338                 start = vma->vm_start;
 339                 error = -EINVAL;
 340                 if (!(vma->vm_flags & VM_GROWSDOWN))
 341                         goto out;
 342         } else {
 343                 if (vma->vm_start > start)
 344                         goto out;
 345                 if (unlikely(grows & PROT_GROWSUP)) {
 346                         end = vma->vm_end;
 347                         error = -EINVAL;
 348                         if (!(vma->vm_flags & VM_GROWSUP))
 349                                 goto out;
 350                 }
 351         }
 352         if (start > vma->vm_start)
 353                 prev = vma;
 354
 355         for (nstart = start ; ; ) {
 356                 unsigned long newflags;
 357
 358                 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
 359
 360                 newflags = vm_flags;
 361                 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 362
 363                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 364                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 365                         error = -EACCES;
 366                         goto out;
 367                 }
 368
 369                 error = security_file_mprotect(vma, reqprot, prot);
 370                 if (error)
 371                         goto out;
 372
 373                 tmp = vma->vm_end;
 374                 if (tmp > end)
 375                         tmp = end;
 376                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 377                 if (error)
 378                         goto out;
 379                 nstart = tmp;
 380
 381                 if (nstart < prev->vm_end)
 382                         nstart = prev->vm_end;
 383                 if (nstart >= end)
 384                         goto out;
 385
 386                 vma = prev->vm_next;
 387                 if (!vma || vma->vm_start != nstart) {
 388                         error = -ENOMEM;
 389                         goto out;
 390                 }
 391         }
 392 out:
 393         up_write(&current->mm->mmap_sem);
 394         return error;
 395 }