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