| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * mm/mprotect.c |
| 4 | * |
| 5 | * (C) Copyright 1994 Linus Torvalds |
| 6 | * (C) Copyright 2002 Christoph Hellwig |
| 7 | * |
| 8 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
| 9 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved |
| 10 | */ |
| 11 | |
| 12 | #include <linux/pagewalk.h> |
| 13 | #include <linux/hugetlb.h> |
| 14 | #include <linux/shm.h> |
| 15 | #include <linux/mman.h> |
| 16 | #include <linux/fs.h> |
| 17 | #include <linux/highmem.h> |
| 18 | #include <linux/security.h> |
| 19 | #include <linux/mempolicy.h> |
| 20 | #include <linux/personality.h> |
| 21 | #include <linux/syscalls.h> |
| 22 | #include <linux/swap.h> |
| 23 | #include <linux/swapops.h> |
| 24 | #include <linux/mmu_notifier.h> |
| 25 | #include <linux/migrate.h> |
| 26 | #include <linux/perf_event.h> |
| 27 | #include <linux/pkeys.h> |
| 28 | #include <linux/ksm.h> |
| 29 | #include <linux/uaccess.h> |
| 30 | #include <linux/mm_inline.h> |
| 31 | #include <linux/pgtable.h> |
| 32 | #include <linux/sched/sysctl.h> |
| 33 | #include <linux/userfaultfd_k.h> |
| 34 | #include <linux/memory-tiers.h> |
| 35 | #include <asm/cacheflush.h> |
| 36 | #include <asm/mmu_context.h> |
| 37 | #include <asm/tlbflush.h> |
| 38 | #include <asm/tlb.h> |
| 39 | |
| 40 | #include "internal.h" |
| 41 | |
| 42 | static inline bool can_change_pte_writable(struct vm_area_struct *vma, |
| 43 | unsigned long addr, pte_t pte) |
| 44 | { |
| 45 | struct page *page; |
| 46 | |
| 47 | VM_BUG_ON(!(vma->vm_flags & VM_WRITE) || pte_write(pte)); |
| 48 | |
| 49 | if (pte_protnone(pte) || !pte_dirty(pte)) |
| 50 | return false; |
| 51 | |
| 52 | /* Do we need write faults for softdirty tracking? */ |
| 53 | if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte)) |
| 54 | return false; |
| 55 | |
| 56 | /* Do we need write faults for uffd-wp tracking? */ |
| 57 | if (userfaultfd_pte_wp(vma, pte)) |
| 58 | return false; |
| 59 | |
| 60 | if (!(vma->vm_flags & VM_SHARED)) { |
| 61 | /* |
| 62 | * We can only special-case on exclusive anonymous pages, |
| 63 | * because we know that our write-fault handler similarly would |
| 64 | * map them writable without any additional checks while holding |
| 65 | * the PT lock. |
| 66 | */ |
| 67 | page = vm_normal_page(vma, addr, pte); |
| 68 | if (!page || !PageAnon(page) || !PageAnonExclusive(page)) |
| 69 | return false; |
| 70 | } |
| 71 | |
| 72 | return true; |
| 73 | } |
| 74 | |
| 75 | static unsigned long change_pte_range(struct mmu_gather *tlb, |
| 76 | struct vm_area_struct *vma, pmd_t *pmd, unsigned long addr, |
| 77 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) |
| 78 | { |
| 79 | pte_t *pte, oldpte; |
| 80 | spinlock_t *ptl; |
| 81 | unsigned long pages = 0; |
| 82 | int target_node = NUMA_NO_NODE; |
| 83 | bool prot_numa = cp_flags & MM_CP_PROT_NUMA; |
| 84 | bool uffd_wp = cp_flags & MM_CP_UFFD_WP; |
| 85 | bool uffd_wp_resolve = cp_flags & MM_CP_UFFD_WP_RESOLVE; |
| 86 | |
| 87 | tlb_change_page_size(tlb, PAGE_SIZE); |
| 88 | |
| 89 | /* |
| 90 | * Can be called with only the mmap_lock for reading by |
| 91 | * prot_numa so we must check the pmd isn't constantly |
| 92 | * changing from under us from pmd_none to pmd_trans_huge |
| 93 | * and/or the other way around. |
| 94 | */ |
| 95 | if (pmd_trans_unstable(pmd)) |
| 96 | return 0; |
| 97 | |
| 98 | /* |
| 99 | * The pmd points to a regular pte so the pmd can't change |
| 100 | * from under us even if the mmap_lock is only hold for |
| 101 | * reading. |
| 102 | */ |
| 103 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 104 | |
| 105 | /* Get target node for single threaded private VMAs */ |
| 106 | if (prot_numa && !(vma->vm_flags & VM_SHARED) && |
| 107 | atomic_read(&vma->vm_mm->mm_users) == 1) |
| 108 | target_node = numa_node_id(); |
| 109 | |
| 110 | flush_tlb_batched_pending(vma->vm_mm); |
| 111 | arch_enter_lazy_mmu_mode(); |
| 112 | do { |
| 113 | oldpte = *pte; |
| 114 | if (pte_present(oldpte)) { |
| 115 | pte_t ptent; |
| 116 | bool preserve_write = prot_numa && pte_write(oldpte); |
| 117 | |
| 118 | /* |
| 119 | * Avoid trapping faults against the zero or KSM |
| 120 | * pages. See similar comment in change_huge_pmd. |
| 121 | */ |
| 122 | if (prot_numa) { |
| 123 | struct page *page; |
| 124 | int nid; |
| 125 | bool toptier; |
| 126 | |
| 127 | /* Avoid TLB flush if possible */ |
| 128 | if (pte_protnone(oldpte)) |
| 129 | continue; |
| 130 | |
| 131 | page = vm_normal_page(vma, addr, oldpte); |
| 132 | if (!page || is_zone_device_page(page) || PageKsm(page)) |
| 133 | continue; |
| 134 | |
| 135 | /* Also skip shared copy-on-write pages */ |
| 136 | if (is_cow_mapping(vma->vm_flags) && |
| 137 | page_count(page) != 1) |
| 138 | continue; |
| 139 | |
| 140 | /* |
| 141 | * While migration can move some dirty pages, |
| 142 | * it cannot move them all from MIGRATE_ASYNC |
| 143 | * context. |
| 144 | */ |
| 145 | if (page_is_file_lru(page) && PageDirty(page)) |
| 146 | continue; |
| 147 | |
| 148 | /* |
| 149 | * Don't mess with PTEs if page is already on the node |
| 150 | * a single-threaded process is running on. |
| 151 | */ |
| 152 | nid = page_to_nid(page); |
| 153 | if (target_node == nid) |
| 154 | continue; |
| 155 | toptier = node_is_toptier(nid); |
| 156 | |
| 157 | /* |
| 158 | * Skip scanning top tier node if normal numa |
| 159 | * balancing is disabled |
| 160 | */ |
| 161 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_NORMAL) && |
| 162 | toptier) |
| 163 | continue; |
| 164 | if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING && |
| 165 | !toptier) |
| 166 | xchg_page_access_time(page, |
| 167 | jiffies_to_msecs(jiffies)); |
| 168 | } |
| 169 | |
| 170 | oldpte = ptep_modify_prot_start(vma, addr, pte); |
| 171 | ptent = pte_modify(oldpte, newprot); |
| 172 | if (preserve_write) |
| 173 | ptent = pte_mk_savedwrite(ptent); |
| 174 | |
| 175 | if (uffd_wp) { |
| 176 | ptent = pte_wrprotect(ptent); |
| 177 | ptent = pte_mkuffd_wp(ptent); |
| 178 | } else if (uffd_wp_resolve) { |
| 179 | ptent = pte_clear_uffd_wp(ptent); |
| 180 | } |
| 181 | |
| 182 | /* |
| 183 | * In some writable, shared mappings, we might want |
| 184 | * to catch actual write access -- see |
| 185 | * vma_wants_writenotify(). |
| 186 | * |
| 187 | * In all writable, private mappings, we have to |
| 188 | * properly handle COW. |
| 189 | * |
| 190 | * In both cases, we can sometimes still change PTEs |
| 191 | * writable and avoid the write-fault handler, for |
| 192 | * example, if a PTE is already dirty and no other |
| 193 | * COW or special handling is required. |
| 194 | */ |
| 195 | if ((cp_flags & MM_CP_TRY_CHANGE_WRITABLE) && |
| 196 | !pte_write(ptent) && |
| 197 | can_change_pte_writable(vma, addr, ptent)) |
| 198 | ptent = pte_mkwrite(ptent); |
| 199 | |
| 200 | ptep_modify_prot_commit(vma, addr, pte, oldpte, ptent); |
| 201 | if (pte_needs_flush(oldpte, ptent)) |
| 202 | tlb_flush_pte_range(tlb, addr, PAGE_SIZE); |
| 203 | pages++; |
| 204 | } else if (is_swap_pte(oldpte)) { |
| 205 | swp_entry_t entry = pte_to_swp_entry(oldpte); |
| 206 | pte_t newpte; |
| 207 | |
| 208 | if (is_writable_migration_entry(entry)) { |
| 209 | struct page *page = pfn_swap_entry_to_page(entry); |
| 210 | |
| 211 | /* |
| 212 | * A protection check is difficult so |
| 213 | * just be safe and disable write |
| 214 | */ |
| 215 | if (PageAnon(page)) |
| 216 | entry = make_readable_exclusive_migration_entry( |
| 217 | swp_offset(entry)); |
| 218 | else |
| 219 | entry = make_readable_migration_entry(swp_offset(entry)); |
| 220 | newpte = swp_entry_to_pte(entry); |
| 221 | if (pte_swp_soft_dirty(oldpte)) |
| 222 | newpte = pte_swp_mksoft_dirty(newpte); |
| 223 | if (pte_swp_uffd_wp(oldpte)) |
| 224 | newpte = pte_swp_mkuffd_wp(newpte); |
| 225 | } else if (is_writable_device_private_entry(entry)) { |
| 226 | /* |
| 227 | * We do not preserve soft-dirtiness. See |
| 228 | * copy_one_pte() for explanation. |
| 229 | */ |
| 230 | entry = make_readable_device_private_entry( |
| 231 | swp_offset(entry)); |
| 232 | newpte = swp_entry_to_pte(entry); |
| 233 | if (pte_swp_uffd_wp(oldpte)) |
| 234 | newpte = pte_swp_mkuffd_wp(newpte); |
| 235 | } else if (is_writable_device_exclusive_entry(entry)) { |
| 236 | entry = make_readable_device_exclusive_entry( |
| 237 | swp_offset(entry)); |
| 238 | newpte = swp_entry_to_pte(entry); |
| 239 | if (pte_swp_soft_dirty(oldpte)) |
| 240 | newpte = pte_swp_mksoft_dirty(newpte); |
| 241 | if (pte_swp_uffd_wp(oldpte)) |
| 242 | newpte = pte_swp_mkuffd_wp(newpte); |
| 243 | } else if (pte_marker_entry_uffd_wp(entry)) { |
| 244 | /* |
| 245 | * If this is uffd-wp pte marker and we'd like |
| 246 | * to unprotect it, drop it; the next page |
| 247 | * fault will trigger without uffd trapping. |
| 248 | */ |
| 249 | if (uffd_wp_resolve) { |
| 250 | pte_clear(vma->vm_mm, addr, pte); |
| 251 | pages++; |
| 252 | } |
| 253 | continue; |
| 254 | } else { |
| 255 | newpte = oldpte; |
| 256 | } |
| 257 | |
| 258 | if (uffd_wp) |
| 259 | newpte = pte_swp_mkuffd_wp(newpte); |
| 260 | else if (uffd_wp_resolve) |
| 261 | newpte = pte_swp_clear_uffd_wp(newpte); |
| 262 | |
| 263 | if (!pte_same(oldpte, newpte)) { |
| 264 | set_pte_at(vma->vm_mm, addr, pte, newpte); |
| 265 | pages++; |
| 266 | } |
| 267 | } else { |
| 268 | /* It must be an none page, or what else?.. */ |
| 269 | WARN_ON_ONCE(!pte_none(oldpte)); |
| 270 | #ifdef CONFIG_PTE_MARKER_UFFD_WP |
| 271 | if (unlikely(uffd_wp && !vma_is_anonymous(vma))) { |
| 272 | /* |
| 273 | * For file-backed mem, we need to be able to |
| 274 | * wr-protect a none pte, because even if the |
| 275 | * pte is none, the page/swap cache could |
| 276 | * exist. Doing that by install a marker. |
| 277 | */ |
| 278 | set_pte_at(vma->vm_mm, addr, pte, |
| 279 | make_pte_marker(PTE_MARKER_UFFD_WP)); |
| 280 | pages++; |
| 281 | } |
| 282 | #endif |
| 283 | } |
| 284 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 285 | arch_leave_lazy_mmu_mode(); |
| 286 | pte_unmap_unlock(pte - 1, ptl); |
| 287 | |
| 288 | return pages; |
| 289 | } |
| 290 | |
| 291 | /* |
| 292 | * Used when setting automatic NUMA hinting protection where it is |
| 293 | * critical that a numa hinting PMD is not confused with a bad PMD. |
| 294 | */ |
| 295 | static inline int pmd_none_or_clear_bad_unless_trans_huge(pmd_t *pmd) |
| 296 | { |
| 297 | pmd_t pmdval = pmd_read_atomic(pmd); |
| 298 | |
| 299 | /* See pmd_none_or_trans_huge_or_clear_bad for info on barrier */ |
| 300 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 301 | barrier(); |
| 302 | #endif |
| 303 | |
| 304 | if (pmd_none(pmdval)) |
| 305 | return 1; |
| 306 | if (pmd_trans_huge(pmdval)) |
| 307 | return 0; |
| 308 | if (unlikely(pmd_bad(pmdval))) { |
| 309 | pmd_clear_bad(pmd); |
| 310 | return 1; |
| 311 | } |
| 312 | |
| 313 | return 0; |
| 314 | } |
| 315 | |
| 316 | /* Return true if we're uffd wr-protecting file-backed memory, or false */ |
| 317 | static inline bool |
| 318 | uffd_wp_protect_file(struct vm_area_struct *vma, unsigned long cp_flags) |
| 319 | { |
| 320 | return (cp_flags & MM_CP_UFFD_WP) && !vma_is_anonymous(vma); |
| 321 | } |
| 322 | |
| 323 | /* |
| 324 | * If wr-protecting the range for file-backed, populate pgtable for the case |
| 325 | * when pgtable is empty but page cache exists. When {pte|pmd|...}_alloc() |
| 326 | * failed it means no memory, we don't have a better option but stop. |
| 327 | */ |
| 328 | #define change_pmd_prepare(vma, pmd, cp_flags) \ |
| 329 | do { \ |
| 330 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ |
| 331 | if (WARN_ON_ONCE(pte_alloc(vma->vm_mm, pmd))) \ |
| 332 | break; \ |
| 333 | } \ |
| 334 | } while (0) |
| 335 | /* |
| 336 | * This is the general pud/p4d/pgd version of change_pmd_prepare(). We need to |
| 337 | * have separate change_pmd_prepare() because pte_alloc() returns 0 on success, |
| 338 | * while {pmd|pud|p4d}_alloc() returns the valid pointer on success. |
| 339 | */ |
| 340 | #define change_prepare(vma, high, low, addr, cp_flags) \ |
| 341 | do { \ |
| 342 | if (unlikely(uffd_wp_protect_file(vma, cp_flags))) { \ |
| 343 | low##_t *p = low##_alloc(vma->vm_mm, high, addr); \ |
| 344 | if (WARN_ON_ONCE(p == NULL)) \ |
| 345 | break; \ |
| 346 | } \ |
| 347 | } while (0) |
| 348 | |
| 349 | static inline unsigned long change_pmd_range(struct mmu_gather *tlb, |
| 350 | struct vm_area_struct *vma, pud_t *pud, unsigned long addr, |
| 351 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) |
| 352 | { |
| 353 | pmd_t *pmd; |
| 354 | unsigned long next; |
| 355 | unsigned long pages = 0; |
| 356 | unsigned long nr_huge_updates = 0; |
| 357 | struct mmu_notifier_range range; |
| 358 | |
| 359 | range.start = 0; |
| 360 | |
| 361 | pmd = pmd_offset(pud, addr); |
| 362 | do { |
| 363 | unsigned long this_pages; |
| 364 | |
| 365 | next = pmd_addr_end(addr, end); |
| 366 | |
| 367 | change_pmd_prepare(vma, pmd, cp_flags); |
| 368 | /* |
| 369 | * Automatic NUMA balancing walks the tables with mmap_lock |
| 370 | * held for read. It's possible a parallel update to occur |
| 371 | * between pmd_trans_huge() and a pmd_none_or_clear_bad() |
| 372 | * check leading to a false positive and clearing. |
| 373 | * Hence, it's necessary to atomically read the PMD value |
| 374 | * for all the checks. |
| 375 | */ |
| 376 | if (!is_swap_pmd(*pmd) && !pmd_devmap(*pmd) && |
| 377 | pmd_none_or_clear_bad_unless_trans_huge(pmd)) |
| 378 | goto next; |
| 379 | |
| 380 | /* invoke the mmu notifier if the pmd is populated */ |
| 381 | if (!range.start) { |
| 382 | mmu_notifier_range_init(&range, |
| 383 | MMU_NOTIFY_PROTECTION_VMA, 0, |
| 384 | vma, vma->vm_mm, addr, end); |
| 385 | mmu_notifier_invalidate_range_start(&range); |
| 386 | } |
| 387 | |
| 388 | if (is_swap_pmd(*pmd) || pmd_trans_huge(*pmd) || pmd_devmap(*pmd)) { |
| 389 | if ((next - addr != HPAGE_PMD_SIZE) || |
| 390 | uffd_wp_protect_file(vma, cp_flags)) { |
| 391 | __split_huge_pmd(vma, pmd, addr, false, NULL); |
| 392 | /* |
| 393 | * For file-backed, the pmd could have been |
| 394 | * cleared; make sure pmd populated if |
| 395 | * necessary, then fall-through to pte level. |
| 396 | */ |
| 397 | change_pmd_prepare(vma, pmd, cp_flags); |
| 398 | } else { |
| 399 | /* |
| 400 | * change_huge_pmd() does not defer TLB flushes, |
| 401 | * so no need to propagate the tlb argument. |
| 402 | */ |
| 403 | int nr_ptes = change_huge_pmd(tlb, vma, pmd, |
| 404 | addr, newprot, cp_flags); |
| 405 | |
| 406 | if (nr_ptes) { |
| 407 | if (nr_ptes == HPAGE_PMD_NR) { |
| 408 | pages += HPAGE_PMD_NR; |
| 409 | nr_huge_updates++; |
| 410 | } |
| 411 | |
| 412 | /* huge pmd was handled */ |
| 413 | goto next; |
| 414 | } |
| 415 | } |
| 416 | /* fall through, the trans huge pmd just split */ |
| 417 | } |
| 418 | this_pages = change_pte_range(tlb, vma, pmd, addr, next, |
| 419 | newprot, cp_flags); |
| 420 | pages += this_pages; |
| 421 | next: |
| 422 | cond_resched(); |
| 423 | } while (pmd++, addr = next, addr != end); |
| 424 | |
| 425 | if (range.start) |
| 426 | mmu_notifier_invalidate_range_end(&range); |
| 427 | |
| 428 | if (nr_huge_updates) |
| 429 | count_vm_numa_events(NUMA_HUGE_PTE_UPDATES, nr_huge_updates); |
| 430 | return pages; |
| 431 | } |
| 432 | |
| 433 | static inline unsigned long change_pud_range(struct mmu_gather *tlb, |
| 434 | struct vm_area_struct *vma, p4d_t *p4d, unsigned long addr, |
| 435 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) |
| 436 | { |
| 437 | pud_t *pud; |
| 438 | unsigned long next; |
| 439 | unsigned long pages = 0; |
| 440 | |
| 441 | pud = pud_offset(p4d, addr); |
| 442 | do { |
| 443 | next = pud_addr_end(addr, end); |
| 444 | change_prepare(vma, pud, pmd, addr, cp_flags); |
| 445 | if (pud_none_or_clear_bad(pud)) |
| 446 | continue; |
| 447 | pages += change_pmd_range(tlb, vma, pud, addr, next, newprot, |
| 448 | cp_flags); |
| 449 | } while (pud++, addr = next, addr != end); |
| 450 | |
| 451 | return pages; |
| 452 | } |
| 453 | |
| 454 | static inline unsigned long change_p4d_range(struct mmu_gather *tlb, |
| 455 | struct vm_area_struct *vma, pgd_t *pgd, unsigned long addr, |
| 456 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) |
| 457 | { |
| 458 | p4d_t *p4d; |
| 459 | unsigned long next; |
| 460 | unsigned long pages = 0; |
| 461 | |
| 462 | p4d = p4d_offset(pgd, addr); |
| 463 | do { |
| 464 | next = p4d_addr_end(addr, end); |
| 465 | change_prepare(vma, p4d, pud, addr, cp_flags); |
| 466 | if (p4d_none_or_clear_bad(p4d)) |
| 467 | continue; |
| 468 | pages += change_pud_range(tlb, vma, p4d, addr, next, newprot, |
| 469 | cp_flags); |
| 470 | } while (p4d++, addr = next, addr != end); |
| 471 | |
| 472 | return pages; |
| 473 | } |
| 474 | |
| 475 | static unsigned long change_protection_range(struct mmu_gather *tlb, |
| 476 | struct vm_area_struct *vma, unsigned long addr, |
| 477 | unsigned long end, pgprot_t newprot, unsigned long cp_flags) |
| 478 | { |
| 479 | struct mm_struct *mm = vma->vm_mm; |
| 480 | pgd_t *pgd; |
| 481 | unsigned long next; |
| 482 | unsigned long pages = 0; |
| 483 | |
| 484 | BUG_ON(addr >= end); |
| 485 | pgd = pgd_offset(mm, addr); |
| 486 | tlb_start_vma(tlb, vma); |
| 487 | do { |
| 488 | next = pgd_addr_end(addr, end); |
| 489 | change_prepare(vma, pgd, p4d, addr, cp_flags); |
| 490 | if (pgd_none_or_clear_bad(pgd)) |
| 491 | continue; |
| 492 | pages += change_p4d_range(tlb, vma, pgd, addr, next, newprot, |
| 493 | cp_flags); |
| 494 | } while (pgd++, addr = next, addr != end); |
| 495 | |
| 496 | tlb_end_vma(tlb, vma); |
| 497 | |
| 498 | return pages; |
| 499 | } |
| 500 | |
| 501 | unsigned long change_protection(struct mmu_gather *tlb, |
| 502 | struct vm_area_struct *vma, unsigned long start, |
| 503 | unsigned long end, pgprot_t newprot, |
| 504 | unsigned long cp_flags) |
| 505 | { |
| 506 | unsigned long pages; |
| 507 | |
| 508 | BUG_ON((cp_flags & MM_CP_UFFD_WP_ALL) == MM_CP_UFFD_WP_ALL); |
| 509 | |
| 510 | if (is_vm_hugetlb_page(vma)) |
| 511 | pages = hugetlb_change_protection(vma, start, end, newprot, |
| 512 | cp_flags); |
| 513 | else |
| 514 | pages = change_protection_range(tlb, vma, start, end, newprot, |
| 515 | cp_flags); |
| 516 | |
| 517 | return pages; |
| 518 | } |
| 519 | |
| 520 | static int prot_none_pte_entry(pte_t *pte, unsigned long addr, |
| 521 | unsigned long next, struct mm_walk *walk) |
| 522 | { |
| 523 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? |
| 524 | 0 : -EACCES; |
| 525 | } |
| 526 | |
| 527 | static int prot_none_hugetlb_entry(pte_t *pte, unsigned long hmask, |
| 528 | unsigned long addr, unsigned long next, |
| 529 | struct mm_walk *walk) |
| 530 | { |
| 531 | return pfn_modify_allowed(pte_pfn(*pte), *(pgprot_t *)(walk->private)) ? |
| 532 | 0 : -EACCES; |
| 533 | } |
| 534 | |
| 535 | static int prot_none_test(unsigned long addr, unsigned long next, |
| 536 | struct mm_walk *walk) |
| 537 | { |
| 538 | return 0; |
| 539 | } |
| 540 | |
| 541 | static const struct mm_walk_ops prot_none_walk_ops = { |
| 542 | .pte_entry = prot_none_pte_entry, |
| 543 | .hugetlb_entry = prot_none_hugetlb_entry, |
| 544 | .test_walk = prot_none_test, |
| 545 | }; |
| 546 | |
| 547 | int |
| 548 | mprotect_fixup(struct mmu_gather *tlb, struct vm_area_struct *vma, |
| 549 | struct vm_area_struct **pprev, unsigned long start, |
| 550 | unsigned long end, unsigned long newflags) |
| 551 | { |
| 552 | struct mm_struct *mm = vma->vm_mm; |
| 553 | unsigned long oldflags = vma->vm_flags; |
| 554 | long nrpages = (end - start) >> PAGE_SHIFT; |
| 555 | unsigned long charged = 0; |
| 556 | bool try_change_writable; |
| 557 | pgoff_t pgoff; |
| 558 | int error; |
| 559 | |
| 560 | if (newflags == oldflags) { |
| 561 | *pprev = vma; |
| 562 | return 0; |
| 563 | } |
| 564 | |
| 565 | /* |
| 566 | * Do PROT_NONE PFN permission checks here when we can still |
| 567 | * bail out without undoing a lot of state. This is a rather |
| 568 | * uncommon case, so doesn't need to be very optimized. |
| 569 | */ |
| 570 | if (arch_has_pfn_modify_check() && |
| 571 | (vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && |
| 572 | (newflags & VM_ACCESS_FLAGS) == 0) { |
| 573 | pgprot_t new_pgprot = vm_get_page_prot(newflags); |
| 574 | |
| 575 | error = walk_page_range(current->mm, start, end, |
| 576 | &prot_none_walk_ops, &new_pgprot); |
| 577 | if (error) |
| 578 | return error; |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * If we make a private mapping writable we increase our commit; |
| 583 | * but (without finer accounting) cannot reduce our commit if we |
| 584 | * make it unwritable again. hugetlb mapping were accounted for |
| 585 | * even if read-only so there is no need to account for them here |
| 586 | */ |
| 587 | if (newflags & VM_WRITE) { |
| 588 | /* Check space limits when area turns into data. */ |
| 589 | if (!may_expand_vm(mm, newflags, nrpages) && |
| 590 | may_expand_vm(mm, oldflags, nrpages)) |
| 591 | return -ENOMEM; |
| 592 | if (!(oldflags & (VM_ACCOUNT|VM_WRITE|VM_HUGETLB| |
| 593 | VM_SHARED|VM_NORESERVE))) { |
| 594 | charged = nrpages; |
| 595 | if (security_vm_enough_memory_mm(mm, charged)) |
| 596 | return -ENOMEM; |
| 597 | newflags |= VM_ACCOUNT; |
| 598 | } |
| 599 | } |
| 600 | |
| 601 | /* |
| 602 | * First try to merge with previous and/or next vma. |
| 603 | */ |
| 604 | pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); |
| 605 | *pprev = vma_merge(mm, *pprev, start, end, newflags, |
| 606 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
| 607 | vma->vm_userfaultfd_ctx, anon_vma_name(vma)); |
| 608 | if (*pprev) { |
| 609 | vma = *pprev; |
| 610 | VM_WARN_ON((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY); |
| 611 | goto success; |
| 612 | } |
| 613 | |
| 614 | *pprev = vma; |
| 615 | |
| 616 | if (start != vma->vm_start) { |
| 617 | error = split_vma(mm, vma, start, 1); |
| 618 | if (error) |
| 619 | goto fail; |
| 620 | } |
| 621 | |
| 622 | if (end != vma->vm_end) { |
| 623 | error = split_vma(mm, vma, end, 0); |
| 624 | if (error) |
| 625 | goto fail; |
| 626 | } |
| 627 | |
| 628 | success: |
| 629 | /* |
| 630 | * vm_flags and vm_page_prot are protected by the mmap_lock |
| 631 | * held in write mode. |
| 632 | */ |
| 633 | vma->vm_flags = newflags; |
| 634 | /* |
| 635 | * We want to check manually if we can change individual PTEs writable |
| 636 | * if we can't do that automatically for all PTEs in a mapping. For |
| 637 | * private mappings, that's always the case when we have write |
| 638 | * permissions as we properly have to handle COW. |
| 639 | */ |
| 640 | if (vma->vm_flags & VM_SHARED) |
| 641 | try_change_writable = vma_wants_writenotify(vma, vma->vm_page_prot); |
| 642 | else |
| 643 | try_change_writable = !!(vma->vm_flags & VM_WRITE); |
| 644 | vma_set_page_prot(vma); |
| 645 | |
| 646 | change_protection(tlb, vma, start, end, vma->vm_page_prot, |
| 647 | try_change_writable ? MM_CP_TRY_CHANGE_WRITABLE : 0); |
| 648 | |
| 649 | /* |
| 650 | * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major |
| 651 | * fault on access. |
| 652 | */ |
| 653 | if ((oldflags & (VM_WRITE | VM_SHARED | VM_LOCKED)) == VM_LOCKED && |
| 654 | (newflags & VM_WRITE)) { |
| 655 | populate_vma_page_range(vma, start, end, NULL); |
| 656 | } |
| 657 | |
| 658 | vm_stat_account(mm, oldflags, -nrpages); |
| 659 | vm_stat_account(mm, newflags, nrpages); |
| 660 | perf_event_mmap(vma); |
| 661 | return 0; |
| 662 | |
| 663 | fail: |
| 664 | vm_unacct_memory(charged); |
| 665 | return error; |
| 666 | } |
| 667 | |
| 668 | /* |
| 669 | * pkey==-1 when doing a legacy mprotect() |
| 670 | */ |
| 671 | static int do_mprotect_pkey(unsigned long start, size_t len, |
| 672 | unsigned long prot, int pkey) |
| 673 | { |
| 674 | unsigned long nstart, end, tmp, reqprot; |
| 675 | struct vm_area_struct *vma, *prev; |
| 676 | int error; |
| 677 | const int grows = prot & (PROT_GROWSDOWN|PROT_GROWSUP); |
| 678 | const bool rier = (current->personality & READ_IMPLIES_EXEC) && |
| 679 | (prot & PROT_READ); |
| 680 | struct mmu_gather tlb; |
| 681 | MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); |
| 682 | |
| 683 | start = untagged_addr(start); |
| 684 | |
| 685 | prot &= ~(PROT_GROWSDOWN|PROT_GROWSUP); |
| 686 | if (grows == (PROT_GROWSDOWN|PROT_GROWSUP)) /* can't be both */ |
| 687 | return -EINVAL; |
| 688 | |
| 689 | if (start & ~PAGE_MASK) |
| 690 | return -EINVAL; |
| 691 | if (!len) |
| 692 | return 0; |
| 693 | len = PAGE_ALIGN(len); |
| 694 | end = start + len; |
| 695 | if (end <= start) |
| 696 | return -ENOMEM; |
| 697 | if (!arch_validate_prot(prot, start)) |
| 698 | return -EINVAL; |
| 699 | |
| 700 | reqprot = prot; |
| 701 | |
| 702 | if (mmap_write_lock_killable(current->mm)) |
| 703 | return -EINTR; |
| 704 | |
| 705 | /* |
| 706 | * If userspace did not allocate the pkey, do not let |
| 707 | * them use it here. |
| 708 | */ |
| 709 | error = -EINVAL; |
| 710 | if ((pkey != -1) && !mm_pkey_is_allocated(current->mm, pkey)) |
| 711 | goto out; |
| 712 | |
| 713 | mas_set(&mas, start); |
| 714 | vma = mas_find(&mas, ULONG_MAX); |
| 715 | error = -ENOMEM; |
| 716 | if (!vma) |
| 717 | goto out; |
| 718 | |
| 719 | if (unlikely(grows & PROT_GROWSDOWN)) { |
| 720 | if (vma->vm_start >= end) |
| 721 | goto out; |
| 722 | start = vma->vm_start; |
| 723 | error = -EINVAL; |
| 724 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 725 | goto out; |
| 726 | } else { |
| 727 | if (vma->vm_start > start) |
| 728 | goto out; |
| 729 | if (unlikely(grows & PROT_GROWSUP)) { |
| 730 | end = vma->vm_end; |
| 731 | error = -EINVAL; |
| 732 | if (!(vma->vm_flags & VM_GROWSUP)) |
| 733 | goto out; |
| 734 | } |
| 735 | } |
| 736 | |
| 737 | if (start > vma->vm_start) |
| 738 | prev = vma; |
| 739 | else |
| 740 | prev = mas_prev(&mas, 0); |
| 741 | |
| 742 | tlb_gather_mmu(&tlb, current->mm); |
| 743 | for (nstart = start ; ; ) { |
| 744 | unsigned long mask_off_old_flags; |
| 745 | unsigned long newflags; |
| 746 | int new_vma_pkey; |
| 747 | |
| 748 | /* Here we know that vma->vm_start <= nstart < vma->vm_end. */ |
| 749 | |
| 750 | /* Does the application expect PROT_READ to imply PROT_EXEC */ |
| 751 | if (rier && (vma->vm_flags & VM_MAYEXEC)) |
| 752 | prot |= PROT_EXEC; |
| 753 | |
| 754 | /* |
| 755 | * Each mprotect() call explicitly passes r/w/x permissions. |
| 756 | * If a permission is not passed to mprotect(), it must be |
| 757 | * cleared from the VMA. |
| 758 | */ |
| 759 | mask_off_old_flags = VM_READ | VM_WRITE | VM_EXEC | |
| 760 | VM_FLAGS_CLEAR; |
| 761 | |
| 762 | new_vma_pkey = arch_override_mprotect_pkey(vma, prot, pkey); |
| 763 | newflags = calc_vm_prot_bits(prot, new_vma_pkey); |
| 764 | newflags |= (vma->vm_flags & ~mask_off_old_flags); |
| 765 | |
| 766 | /* newflags >> 4 shift VM_MAY% in place of VM_% */ |
| 767 | if ((newflags & ~(newflags >> 4)) & VM_ACCESS_FLAGS) { |
| 768 | error = -EACCES; |
| 769 | break; |
| 770 | } |
| 771 | |
| 772 | /* Allow architectures to sanity-check the new flags */ |
| 773 | if (!arch_validate_flags(newflags)) { |
| 774 | error = -EINVAL; |
| 775 | break; |
| 776 | } |
| 777 | |
| 778 | error = security_file_mprotect(vma, reqprot, prot); |
| 779 | if (error) |
| 780 | break; |
| 781 | |
| 782 | tmp = vma->vm_end; |
| 783 | if (tmp > end) |
| 784 | tmp = end; |
| 785 | |
| 786 | if (vma->vm_ops && vma->vm_ops->mprotect) { |
| 787 | error = vma->vm_ops->mprotect(vma, nstart, tmp, newflags); |
| 788 | if (error) |
| 789 | break; |
| 790 | } |
| 791 | |
| 792 | error = mprotect_fixup(&tlb, vma, &prev, nstart, tmp, newflags); |
| 793 | if (error) |
| 794 | break; |
| 795 | |
| 796 | nstart = tmp; |
| 797 | |
| 798 | if (nstart < prev->vm_end) |
| 799 | nstart = prev->vm_end; |
| 800 | if (nstart >= end) |
| 801 | break; |
| 802 | |
| 803 | vma = find_vma(current->mm, prev->vm_end); |
| 804 | if (!vma || vma->vm_start != nstart) { |
| 805 | error = -ENOMEM; |
| 806 | break; |
| 807 | } |
| 808 | prot = reqprot; |
| 809 | } |
| 810 | tlb_finish_mmu(&tlb); |
| 811 | out: |
| 812 | mmap_write_unlock(current->mm); |
| 813 | return error; |
| 814 | } |
| 815 | |
| 816 | SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, |
| 817 | unsigned long, prot) |
| 818 | { |
| 819 | return do_mprotect_pkey(start, len, prot, -1); |
| 820 | } |
| 821 | |
| 822 | #ifdef CONFIG_ARCH_HAS_PKEYS |
| 823 | |
| 824 | SYSCALL_DEFINE4(pkey_mprotect, unsigned long, start, size_t, len, |
| 825 | unsigned long, prot, int, pkey) |
| 826 | { |
| 827 | return do_mprotect_pkey(start, len, prot, pkey); |
| 828 | } |
| 829 | |
| 830 | SYSCALL_DEFINE2(pkey_alloc, unsigned long, flags, unsigned long, init_val) |
| 831 | { |
| 832 | int pkey; |
| 833 | int ret; |
| 834 | |
| 835 | /* No flags supported yet. */ |
| 836 | if (flags) |
| 837 | return -EINVAL; |
| 838 | /* check for unsupported init values */ |
| 839 | if (init_val & ~PKEY_ACCESS_MASK) |
| 840 | return -EINVAL; |
| 841 | |
| 842 | mmap_write_lock(current->mm); |
| 843 | pkey = mm_pkey_alloc(current->mm); |
| 844 | |
| 845 | ret = -ENOSPC; |
| 846 | if (pkey == -1) |
| 847 | goto out; |
| 848 | |
| 849 | ret = arch_set_user_pkey_access(current, pkey, init_val); |
| 850 | if (ret) { |
| 851 | mm_pkey_free(current->mm, pkey); |
| 852 | goto out; |
| 853 | } |
| 854 | ret = pkey; |
| 855 | out: |
| 856 | mmap_write_unlock(current->mm); |
| 857 | return ret; |
| 858 | } |
| 859 | |
| 860 | SYSCALL_DEFINE1(pkey_free, int, pkey) |
| 861 | { |
| 862 | int ret; |
| 863 | |
| 864 | mmap_write_lock(current->mm); |
| 865 | ret = mm_pkey_free(current->mm, pkey); |
| 866 | mmap_write_unlock(current->mm); |
| 867 | |
| 868 | /* |
| 869 | * We could provide warnings or errors if any VMA still |
| 870 | * has the pkey set here. |
| 871 | */ |
| 872 | return ret; |
| 873 | } |
| 874 | |
| 875 | #endif /* CONFIG_ARCH_HAS_PKEYS */ |