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 <linux/ksm.h>
  27 #include <asm/uaccess.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/cacheflush.h>
  30 #include <asm/tlbflush.h>
  31
  32 #ifdef CONFIG_E2K
  33 #include <asm/process.h>
  34 #endif
  35
  36 #ifndef pgprot_modify
  37 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
  38 {
  39         return newprot;
  40 }
  41 #endif
  42
  43 static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
  44                 unsigned long addr, unsigned long end, pgprot_t newprot,
  45                 int dirty_accountable, int prot_numa)
  46 {
  47         struct mm_struct *mm = vma->vm_mm;
  48         pte_t *pte, oldpte;
  49         spinlock_t *ptl;
  50         unsigned long pages = 0;
  51
  52         pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  53         arch_enter_lazy_mmu_mode();
  54         do {
  55                 oldpte = *pte;
  56                 if (pte_present(oldpte)) {
  57                         pte_t ptent;
  58                         bool updated = false;
  59
  60                         if (!prot_numa) {
  61                                 ptent = ptep_modify_prot_start(mm, addr, pte);
  62                                 if (pte_numa(ptent))
  63                                         ptent = pte_mknonnuma(ptent);
  64                                 ptent = pte_modify(ptent, newprot);
  65                                 /*
  66                                  * Avoid taking write faults for pages we
  67                                  * know to be dirty.
  68                                  */
  69                                 if (dirty_accountable && pte_dirty(ptent))
  70                                         ptent = pte_mkwrite(ptent);
  71                                 ptep_modify_prot_commit(mm, addr, pte, ptent);
  72                                 updated = true;
  73                         } else {
  74                                 struct page *page;
  75
  76                                 page = vm_normal_page(vma, addr, oldpte);
  77                                 if (page && !PageKsm(page)) {
  78                                         if (!pte_numa(oldpte)) {
  79                                                 ptep_set_numa(mm, addr, pte);
  80                                                 updated = true;
  81                                         }
  82                                 }
  83                         }
  84                         if (updated)
  85                                 pages++;
  86                 } else if (IS_ENABLED(CONFIG_MIGRATION) && !pte_file(oldpte)) {
  87                         swp_entry_t entry = pte_to_swp_entry(oldpte);
  88
  89                         if (is_write_migration_entry(entry)) {
  90                                 pte_t newpte;
  91                                 /*
  92                                  * A protection check is difficult so
  93                                  * just be safe and disable write
  94                                  */
  95                                 make_migration_entry_read(&entry);
  96                                 newpte = swp_entry_to_pte(entry);
  97                                 if (pte_swp_soft_dirty(oldpte))
  98                                         newpte = pte_swp_mksoft_dirty(newpte);
  99                                 set_pte_at(mm, addr, pte, newpte);
 100
 101                                 pages++;
 102                         }
 103                 }
 104         } while (pte++, addr += PAGE_SIZE, addr != end);
 105         arch_leave_lazy_mmu_mode();
 106         pte_unmap_unlock(pte - 1, ptl);
 107
 108         return pages;
 109 }
 110
 111 static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 112                 pud_t *pud, unsigned long addr, unsigned long end,
 113                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 114 {
 115         pmd_t *pmd;
 116         unsigned long next;
 117         unsigned long pages = 0;
 118         unsigned long nr_huge_updates = 0;
 119
 120         pmd = pmd_offset(pud, addr);
 121         do {
 122                 unsigned long this_pages;
 123
 124                 next = pmd_addr_end(addr, end);
 125                 if (pmd_trans_huge(*pmd)) {
 126                         if (next - addr != HPAGE_PMD_SIZE)
 127                                 split_huge_page_pmd(vma, addr, pmd);
 128                         else {
 129                                 int nr_ptes = change_huge_pmd(vma, pmd, addr,
 130                                                 newprot, prot_numa);
 131
 132                                 if (nr_ptes) {
 133                                         if (nr_ptes == HPAGE_PMD_NR) {
 134                                                 pages += HPAGE_PMD_NR;
 135                                                 nr_huge_updates++;
 136                                         }
 137                                         continue;
 138                                 }
 139                         }
 140                         /* fall through */
 141                 }
 142                 if (pmd_none_or_clear_bad(pmd))
 143                         continue;
 144                 this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 145                                  dirty_accountable, prot_numa);
 146                 pages += this_pages;
 147         } while (pmd++, addr = next, addr != end);
 148
 149         if (nr_huge_updates)
 150                 count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates);
 151         return pages;
 152 }
 153
 154 static inline unsigned long change_pud_range(struct vm_area_struct *vma,
 155                 pgd_t *pgd, unsigned long addr, unsigned long end,
 156                 pgprot_t newprot, int dirty_accountable, int prot_numa)
 157 {
 158         pud_t *pud;
 159         unsigned long next;
 160         unsigned long pages = 0;
 161
 162         pud = pud_offset(pgd, addr);
 163         do {
 164                 next = pud_addr_end(addr, end);
 165                 if (pud_none_or_clear_bad(pud))
 166                         continue;
 167                 pages += change_pmd_range(vma, pud, addr, next, newprot,
 168                                  dirty_accountable, prot_numa);
 169         } while (pud++, addr = next, addr != end);
 170
 171         return pages;
 172 }
 173
 174 static unsigned long change_protection_range(struct vm_area_struct *vma,
 175                 unsigned long addr, unsigned long end, pgprot_t newprot,
 176                 int dirty_accountable, int prot_numa)
 177 {
 178         struct mm_struct *mm = vma->vm_mm;
 179         pgd_t *pgd;
 180         unsigned long next;
 181         unsigned long start = addr;
 182         unsigned long pages = 0;
 183
 184         BUG_ON(addr >= end);
 185         pgd = pgd_offset(mm, addr);
 186         flush_cache_range(vma, addr, end);
 187         set_tlb_flush_pending(mm);
 188         do {
 189                 next = pgd_addr_end(addr, end);
 190                 if (pgd_none_or_clear_bad(pgd))
 191                         continue;
 192                 pages += change_pud_range(vma, pgd, addr, next, newprot,
 193                                  dirty_accountable, prot_numa);
 194         } while (pgd++, addr = next, addr != end);
 195
 196         /* Only flush the TLB if we actually modified any entries: */
 197         if (pages)
 198                 flush_tlb_range(vma, start, end);
 199         clear_tlb_flush_pending(mm);
 200
 201         return pages;
 202 }
 203
 204 unsigned long change_protection(struct vm_area_struct *vma, unsigned long start,
 205                        unsigned long end, pgprot_t newprot,
 206                        int dirty_accountable, int prot_numa)
 207 {
 208         struct mm_struct *mm = vma->vm_mm;
 209         unsigned long pages;
 210
 211         mmu_notifier_invalidate_range_start(mm, start, end);
 212         if (is_vm_hugetlb_page(vma))
 213                 pages = hugetlb_change_protection(vma, start, end, newprot);
 214         else
 215                 pages = change_protection_range(vma, start, end, newprot, dirty_accountable, prot_numa);
 216         mmu_notifier_invalidate_range_end(mm, start, end);
 217
 218         return pages;
 219 }
 220
 221 int
 222 mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 223         unsigned long start, unsigned long end, unsigned long newflags)
 224 {
 225         struct mm_struct *mm = vma->vm_mm;
 226         unsigned long oldflags = vma->vm_flags;
 227         long nrpages = (end - start) >> PAGE_SHIFT;
 228         unsigned long charged = 0;
 229         pgoff_t pgoff;
 230         int error;
 231         int dirty_accountable = 0;
 232 #if defined(CONFIG_E2K) && defined(CONFIG_MAKE_ALL_PAGES_VALID)
 233         pgprot_t oldprot = vma->vm_page_prot;
 234 #endif
 235
 236         if (newflags == oldflags) {
 237                 *pprev = vma;
 238                 return 0;
 239         }
 240
 241         /*
 242          * If we make a private mapping writable we increase our commit;
 243          * but (without finer accounting) cannot reduce our commit if we
 244          * make it unwritable again. hugetlb mapping were accounted for
 245          * even if read-only so there is no need to account for them here
 246          */
 247         if (newflags & VM_WRITE) {
 248                 if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB|
 249                                                 VM_SHARED|VM_NORESERVE))) {
 250                         charged = nrpages;
 251                         if (security_vm_enough_memory_mm(mm, charged))
 252                                 return -ENOMEM;
 253                         newflags |= VM_ACCOUNT;
 254                 }
 255         }
 256
 257         /*
 258          * First try to merge with previous and/or next vma.
 259          */
 260         pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
 261         *pprev = vma_merge(mm, *pprev, start, end, newflags,
 262                         vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 263         if (*pprev) {
 264                 vma = *pprev;
 265                 goto success;
 266         }
 267
 268         *pprev = vma;
 269
 270         if (start != vma->vm_start) {
 271                 error = split_vma(mm, vma, start, 1);
 272                 if (error)
 273                         goto fail;
 274         }
 275
 276         if (end != vma->vm_end) {
 277                 error = split_vma(mm, vma, end, 0);
 278                 if (error)
 279                         goto fail;
 280         }
 281
 282 success:
 283         /*
 284          * vm_flags and vm_page_prot are protected by the mmap_sem
 285          * held in write mode.
 286          */
 287         vma->vm_flags = newflags;
 288         vma->vm_page_prot = pgprot_modify(vma->vm_page_prot,
 289                                           vm_get_page_prot(newflags));
 290
 291         if (vma_wants_writenotify(vma)) {
 292                 vma->vm_page_prot = vm_get_page_prot(newflags & ~VM_SHARED);
 293                 dirty_accountable = 1;
 294         }
 295
 296         change_protection(vma, start, end, vma->vm_page_prot,
 297                           dirty_accountable, 0);
 298
 299 #if defined(CONFIG_E2K) && defined(CONFIG_MAKE_ALL_PAGES_VALID)
 300         /*
 301          * We may need to change valid bits in 2 cases:
 302          *
 303          * 1) !PROT_NONE -> PROT_NONE - must remove the valid bit to avoid
 304          * performance loss when semispeculative loads hit this area.
 305          *
 306          * 2) PROT_NONE -> !PROT_NONE - must set the valid bit for
 307          * semispeculative loads to work.
 308          *
 309          * We must do the TLB flush _after_ changing the valid bit
 310          * regardless of whether the flush has been done before.
 311          *
 312          * Also change_protection() function does not flush TLB in
 313          * the second case above.
 314          */
 315         if ((pgprot_val(oldprot) ^ pgprot_val(vma->vm_page_prot)) &
 316                         _PAGE_VALID) {
 317                 if (make_all_vma_pages_valid(vma, 1, 1))
 318                         printk_once(KERN_WARNING "make_all_vma_pages_valid() failed in change_protection()\n");
 319         }
 320 #endif
 321
 322         vm_stat_account(mm, oldflags, vma->vm_file, -nrpages);
 323         vm_stat_account(mm, newflags, vma->vm_file, nrpages);
 324         perf_event_mmap(vma);
 325         return 0;
 326
 327 fail:
 328         vm_unacct_memory(charged);
 329         return error;
 330 }
 331
 332 SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len,
 333                 unsigned long, prot)
 334 {
 335         unsigned long vm_flags, nstart, end, tmp, reqprot;
 336         struct vm_area_struct *vma, *prev;
 337         int error = -EINVAL;
 338         const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP);
 339         prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP);
 340         if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */
 341                 return -EINVAL;
 342
 343         if (start & ~PAGE_MASK)
 344                 return -EINVAL;
 345         if (!len)
 346                 return 0;
 347         len = PAGE_ALIGN(len);
 348         end = start + len;
 349         if (end <= start)
 350                 return -ENOMEM;
 351         if (!arch_validate_prot(prot))
 352                 return -EINVAL;
 353
 354         reqprot = prot;
 355         /*
 356          * Does the application expect PROT_READ to imply PROT_EXEC:
 357          */
 358         if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 359                 prot |= PROT_EXEC;
 360
 361         vm_flags = calc_vm_prot_bits(prot);
 362
 363         down_write(&current->mm->mmap_sem);
 364
 365         vma = find_vma(current->mm, start);
 366         error = -ENOMEM;
 367         if (!vma)
 368                 goto out;
 369 #ifdef CONFIG_E2K
 370         if (!test_ts_flag(TS_KERNEL_SYSCALL) &&
 371                         __is_u_hw_stack_range(vma, start, start + len)) {
 372                 error = -EPERM;
 373                 goto out;
 374         }
 375 #endif
 376         prev = vma->vm_prev;
 377         if (unlikely(grows & PROT_GROWSDOWN)) {
 378                 if (vma->vm_start >= end)
 379                         goto out;
 380                 start = vma->vm_start;
 381                 error = -EINVAL;
 382                 if (!(vma->vm_flags & VM_GROWSDOWN))
 383                         goto out;
 384         } else {
 385                 if (vma->vm_start > start)
 386                         goto out;
 387                 if (unlikely(grows & PROT_GROWSUP)) {
 388                         end = vma->vm_end;
 389                         error = -EINVAL;
 390                         if (!(vma->vm_flags & VM_GROWSUP))
 391                                 goto out;
 392                 }
 393         }
 394         if (start > vma->vm_start)
 395                 prev = vma;
 396
 397         for (nstart = start ; ; ) {
 398                 unsigned long newflags;
 399
 400                 /* Here we know that vma->vm_start <= nstart < vma->vm_end. */
 401
 402                 newflags = vm_flags;
 403                 newflags |= (vma->vm_flags & ~(VM_READ | VM_WRITE | VM_EXEC));
 404
 405                 /* newflags >> 4 shift VM_MAY% in place of VM_% */
 406                 if ((newflags & ~(newflags >> 4)) & (VM_READ | VM_WRITE | VM_EXEC)) {
 407                         error = -EACCES;
 408                         goto out;
 409                 }
 410
 411                 error = security_file_mprotect(vma, reqprot, prot);
 412                 if (error)
 413                         goto out;
 414
 415                 tmp = vma->vm_end;
 416                 if (tmp > end)
 417                         tmp = end;
 418                 error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
 419                 if (error)
 420                         goto out;
 421                 nstart = tmp;
 422
 423                 if (nstart < prev->vm_end)
 424                         nstart = prev->vm_end;
 425                 if (nstart >= end)
 426                         goto out;
 427
 428                 vma = prev->vm_next;
 429                 if (!vma || vma->vm_start != nstart) {
 430                         error = -ENOMEM;
 431                         goto out;
 432                 }
 433         }
 434 out:
 435         up_write(&current->mm->mmap_sem);
 436         return error;
 437 }