1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. 4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar 5 */ 6 #include <linux/magic.h> /* STACK_END_MAGIC */ 7 #include <linux/sched.h> /* test_thread_flag(), ... */ 8 #include <linux/kdebug.h> /* oops_begin/end, ... */ 9 #include <linux/module.h> /* search_exception_table */ 10 #include <linux/bootmem.h> /* max_low_pfn */ 11 #include <linux/kprobes.h> /* __kprobes, ... */ 12 #include <linux/mmiotrace.h> /* kmmio_handler, ... */ 13 #include <linux/perf_event.h> /* perf_sw_event */ 14 #include <linux/hugetlb.h> /* hstate_index_to_shift */ 15 #include <linux/prefetch.h> /* prefetchw */ 16 17 #include <asm/traps.h> /* dotraplinkage, ... */ 18 #include <asm/pgalloc.h> /* pgd_*(), ... */ 19 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ 20 #include <asm/fixmap.h> /* VSYSCALL_START */ 21 #include <asm/rcu.h> /* exception_enter(), ... */ 22 23 /* 24 * Page fault error code bits: 25 * 26 * bit 0 == 0: no page found 1: protection fault 27 * bit 1 == 0: read access 1: write access 28 * bit 2 == 0: kernel-mode access 1: user-mode access 29 * bit 3 == 1: use of reserved bit detected 30 * bit 4 == 1: fault was an instruction fetch 31 */ 32 enum x86_pf_error_code { 33 34 PF_PROT = 1 << 0, 35 PF_WRITE = 1 << 1, 36 PF_USER = 1 << 2, 37 PF_RSVD = 1 << 3, 38 PF_INSTR = 1 << 4, 39 }; 40 41 /* 42 * Returns 0 if mmiotrace is disabled, or if the fault is not 43 * handled by mmiotrace: 44 */ 45 static inline int __kprobes 46 kmmio_fault(struct pt_regs *regs, unsigned long addr) 47 { 48 if (unlikely(is_kmmio_active())) 49 if (kmmio_handler(regs, addr) == 1) 50 return -1; 51 return 0; 52 } 53 54 static inline int __kprobes notify_page_fault(struct pt_regs *regs) 55 { 56 int ret = 0; 57 58 /* kprobe_running() needs smp_processor_id() */ 59 if (kprobes_built_in() && !user_mode_vm(regs)) { 60 preempt_disable(); 61 if (kprobe_running() && kprobe_fault_handler(regs, 14)) 62 ret = 1; 63 preempt_enable(); 64 } 65 66 return ret; 67 } 68 69 /* 70 * Prefetch quirks: 71 * 72 * 32-bit mode: 73 * 74 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 75 * Check that here and ignore it. 76 * 77 * 64-bit mode: 78 * 79 * Sometimes the CPU reports invalid exceptions on prefetch. 80 * Check that here and ignore it. 81 * 82 * Opcode checker based on code by Richard Brunner. 83 */ 84 static inline int 85 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 86 unsigned char opcode, int *prefetch) 87 { 88 unsigned char instr_hi = opcode & 0xf0; 89 unsigned char instr_lo = opcode & 0x0f; 90 91 switch (instr_hi) { 92 case 0x20: 93 case 0x30: 94 /* 95 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 96 * In X86_64 long mode, the CPU will signal invalid 97 * opcode if some of these prefixes are present so 98 * X86_64 will never get here anyway 99 */ 100 return ((instr_lo & 7) == 0x6); 101 #ifdef CONFIG_X86_64 102 case 0x40: 103 /* 104 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 105 * Need to figure out under what instruction mode the 106 * instruction was issued. Could check the LDT for lm, 107 * but for now it's good enough to assume that long 108 * mode only uses well known segments or kernel. 109 */ 110 return (!user_mode(regs) || user_64bit_mode(regs)); 111 #endif 112 case 0x60: 113 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 114 return (instr_lo & 0xC) == 0x4; 115 case 0xF0: 116 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 117 return !instr_lo || (instr_lo>>1) == 1; 118 case 0x00: 119 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 120 if (probe_kernel_address(instr, opcode)) 121 return 0; 122 123 *prefetch = (instr_lo == 0xF) && 124 (opcode == 0x0D || opcode == 0x18); 125 return 0; 126 default: 127 return 0; 128 } 129 } 130 131 static int 132 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 133 { 134 unsigned char *max_instr; 135 unsigned char *instr; 136 int prefetch = 0; 137 138 /* 139 * If it was a exec (instruction fetch) fault on NX page, then 140 * do not ignore the fault: 141 */ 142 if (error_code & PF_INSTR) 143 return 0; 144 145 instr = (void *)convert_ip_to_linear(current, regs); 146 max_instr = instr + 15; 147 148 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) 149 return 0; 150 151 while (instr < max_instr) { 152 unsigned char opcode; 153 154 if (probe_kernel_address(instr, opcode)) 155 break; 156 157 instr++; 158 159 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 160 break; 161 } 162 return prefetch; 163 } 164 165 static void 166 force_sig_info_fault(int si_signo, int si_code, unsigned long address, 167 struct task_struct *tsk, int fault) 168 { 169 unsigned lsb = 0; 170 siginfo_t info; 171 172 info.si_signo = si_signo; 173 info.si_errno = 0; 174 info.si_code = si_code; 175 info.si_addr = (void __user *)address; 176 if (fault & VM_FAULT_HWPOISON_LARGE) 177 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 178 if (fault & VM_FAULT_HWPOISON) 179 lsb = PAGE_SHIFT; 180 info.si_addr_lsb = lsb; 181 182 force_sig_info(si_signo, &info, tsk); 183 } 184 185 DEFINE_SPINLOCK(pgd_lock); 186 LIST_HEAD(pgd_list); 187 188 #ifdef CONFIG_X86_32 189 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 190 { 191 unsigned index = pgd_index(address); 192 pgd_t *pgd_k; 193 pud_t *pud, *pud_k; 194 pmd_t *pmd, *pmd_k; 195 196 pgd += index; 197 pgd_k = init_mm.pgd + index; 198 199 if (!pgd_present(*pgd_k)) 200 return NULL; 201 202 /* 203 * set_pgd(pgd, *pgd_k); here would be useless on PAE 204 * and redundant with the set_pmd() on non-PAE. As would 205 * set_pud. 206 */ 207 pud = pud_offset(pgd, address); 208 pud_k = pud_offset(pgd_k, address); 209 if (!pud_present(*pud_k)) 210 return NULL; 211 212 pmd = pmd_offset(pud, address); 213 pmd_k = pmd_offset(pud_k, address); 214 if (!pmd_present(*pmd_k)) 215 return NULL; 216 217 if (!pmd_present(*pmd)) 218 set_pmd(pmd, *pmd_k); 219 else 220 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 221 222 return pmd_k; 223 } 224 225 void vmalloc_sync_all(void) 226 { 227 unsigned long address; 228 229 if (SHARED_KERNEL_PMD) 230 return; 231 232 for (address = VMALLOC_START & PMD_MASK; 233 address >= TASK_SIZE && address < FIXADDR_TOP; 234 address += PMD_SIZE) { 235 struct page *page; 236 237 spin_lock(&pgd_lock); 238 list_for_each_entry(page, &pgd_list, lru) { 239 spinlock_t *pgt_lock; 240 pmd_t *ret; 241 242 /* the pgt_lock only for Xen */ 243 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 244 245 spin_lock(pgt_lock); 246 ret = vmalloc_sync_one(page_address(page), address); 247 spin_unlock(pgt_lock); 248 249 if (!ret) 250 break; 251 } 252 spin_unlock(&pgd_lock); 253 } 254 } 255 256 /* 257 * 32-bit: 258 * 259 * Handle a fault on the vmalloc or module mapping area 260 */ 261 static noinline __kprobes int vmalloc_fault(unsigned long address) 262 { 263 unsigned long pgd_paddr; 264 pmd_t *pmd_k; 265 pte_t *pte_k; 266 267 /* Make sure we are in vmalloc area: */ 268 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 269 return -1; 270 271 WARN_ON_ONCE(in_nmi()); 272 273 /* 274 * Synchronize this task's top level page-table 275 * with the 'reference' page table. 276 * 277 * Do _not_ use "current" here. We might be inside 278 * an interrupt in the middle of a task switch.. 279 */ 280 pgd_paddr = read_cr3(); 281 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 282 if (!pmd_k) 283 return -1; 284 285 pte_k = pte_offset_kernel(pmd_k, address); 286 if (!pte_present(*pte_k)) 287 return -1; 288 289 return 0; 290 } 291 292 /* 293 * Did it hit the DOS screen memory VA from vm86 mode? 294 */ 295 static inline void 296 check_v8086_mode(struct pt_regs *regs, unsigned long address, 297 struct task_struct *tsk) 298 { 299 unsigned long bit; 300 301 if (!v8086_mode(regs)) 302 return; 303 304 bit = (address - 0xA0000) >> PAGE_SHIFT; 305 if (bit < 32) 306 tsk->thread.screen_bitmap |= 1 << bit; 307 } 308 309 static bool low_pfn(unsigned long pfn) 310 { 311 return pfn < max_low_pfn; 312 } 313 314 static void dump_pagetable(unsigned long address) 315 { 316 pgd_t *base = __va(read_cr3()); 317 pgd_t *pgd = &base[pgd_index(address)]; 318 pmd_t *pmd; 319 pte_t *pte; 320 321 #ifdef CONFIG_X86_PAE 322 printk("*pdpt = %016Lx ", pgd_val(*pgd)); 323 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 324 goto out; 325 #endif 326 pmd = pmd_offset(pud_offset(pgd, address), address); 327 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 328 329 /* 330 * We must not directly access the pte in the highpte 331 * case if the page table is located in highmem. 332 * And let's rather not kmap-atomic the pte, just in case 333 * it's allocated already: 334 */ 335 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 336 goto out; 337 338 pte = pte_offset_kernel(pmd, address); 339 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 340 out: 341 printk("\n"); 342 } 343 344 #else /* CONFIG_X86_64: */ 345 346 void vmalloc_sync_all(void) 347 { 348 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); 349 } 350 351 /* 352 * 64-bit: 353 * 354 * Handle a fault on the vmalloc area 355 * 356 * This assumes no large pages in there. 357 */ 358 static noinline __kprobes int vmalloc_fault(unsigned long address) 359 { 360 pgd_t *pgd, *pgd_ref; 361 pud_t *pud, *pud_ref; 362 pmd_t *pmd, *pmd_ref; 363 pte_t *pte, *pte_ref; 364 365 /* Make sure we are in vmalloc area: */ 366 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 367 return -1; 368 369 WARN_ON_ONCE(in_nmi()); 370 371 /* 372 * Copy kernel mappings over when needed. This can also 373 * happen within a race in page table update. In the later 374 * case just flush: 375 */ 376 pgd = pgd_offset(current->active_mm, address); 377 pgd_ref = pgd_offset_k(address); 378 if (pgd_none(*pgd_ref)) 379 return -1; 380 381 if (pgd_none(*pgd)) 382 set_pgd(pgd, *pgd_ref); 383 else 384 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); 385 386 /* 387 * Below here mismatches are bugs because these lower tables 388 * are shared: 389 */ 390 391 pud = pud_offset(pgd, address); 392 pud_ref = pud_offset(pgd_ref, address); 393 if (pud_none(*pud_ref)) 394 return -1; 395 396 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) 397 BUG(); 398 399 pmd = pmd_offset(pud, address); 400 pmd_ref = pmd_offset(pud_ref, address); 401 if (pmd_none(*pmd_ref)) 402 return -1; 403 404 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) 405 BUG(); 406 407 pte_ref = pte_offset_kernel(pmd_ref, address); 408 if (!pte_present(*pte_ref)) 409 return -1; 410 411 pte = pte_offset_kernel(pmd, address); 412 413 /* 414 * Don't use pte_page here, because the mappings can point 415 * outside mem_map, and the NUMA hash lookup cannot handle 416 * that: 417 */ 418 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) 419 BUG(); 420 421 return 0; 422 } 423 424 #ifdef CONFIG_CPU_SUP_AMD 425 static const char errata93_warning[] = 426 KERN_ERR 427 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 428 "******* Working around it, but it may cause SEGVs or burn power.\n" 429 "******* Please consider a BIOS update.\n" 430 "******* Disabling USB legacy in the BIOS may also help.\n"; 431 #endif 432 433 /* 434 * No vm86 mode in 64-bit mode: 435 */ 436 static inline void 437 check_v8086_mode(struct pt_regs *regs, unsigned long address, 438 struct task_struct *tsk) 439 { 440 } 441 442 static int bad_address(void *p) 443 { 444 unsigned long dummy; 445 446 return probe_kernel_address((unsigned long *)p, dummy); 447 } 448 449 static void dump_pagetable(unsigned long address) 450 { 451 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); 452 pgd_t *pgd = base + pgd_index(address); 453 pud_t *pud; 454 pmd_t *pmd; 455 pte_t *pte; 456 457 if (bad_address(pgd)) 458 goto bad; 459 460 printk("PGD %lx ", pgd_val(*pgd)); 461 462 if (!pgd_present(*pgd)) 463 goto out; 464 465 pud = pud_offset(pgd, address); 466 if (bad_address(pud)) 467 goto bad; 468 469 printk("PUD %lx ", pud_val(*pud)); 470 if (!pud_present(*pud) || pud_large(*pud)) 471 goto out; 472 473 pmd = pmd_offset(pud, address); 474 if (bad_address(pmd)) 475 goto bad; 476 477 printk("PMD %lx ", pmd_val(*pmd)); 478 if (!pmd_present(*pmd) || pmd_large(*pmd)) 479 goto out; 480 481 pte = pte_offset_kernel(pmd, address); 482 if (bad_address(pte)) 483 goto bad; 484 485 printk("PTE %lx", pte_val(*pte)); 486 out: 487 printk("\n"); 488 return; 489 bad: 490 printk("BAD\n"); 491 } 492 493 #endif /* CONFIG_X86_64 */ 494 495 /* 496 * Workaround for K8 erratum #93 & buggy BIOS. 497 * 498 * BIOS SMM functions are required to use a specific workaround 499 * to avoid corruption of the 64bit RIP register on C stepping K8. 500 * 501 * A lot of BIOS that didn't get tested properly miss this. 502 * 503 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 504 * Try to work around it here. 505 * 506 * Note we only handle faults in kernel here. 507 * Does nothing on 32-bit. 508 */ 509 static int is_errata93(struct pt_regs *regs, unsigned long address) 510 { 511 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 512 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 513 || boot_cpu_data.x86 != 0xf) 514 return 0; 515 516 if (address != regs->ip) 517 return 0; 518 519 if ((address >> 32) != 0) 520 return 0; 521 522 address |= 0xffffffffUL << 32; 523 if ((address >= (u64)_stext && address <= (u64)_etext) || 524 (address >= MODULES_VADDR && address <= MODULES_END)) { 525 printk_once(errata93_warning); 526 regs->ip = address; 527 return 1; 528 } 529 #endif 530 return 0; 531 } 532 533 /* 534 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 535 * to illegal addresses >4GB. 536 * 537 * We catch this in the page fault handler because these addresses 538 * are not reachable. Just detect this case and return. Any code 539 * segment in LDT is compatibility mode. 540 */ 541 static int is_errata100(struct pt_regs *regs, unsigned long address) 542 { 543 #ifdef CONFIG_X86_64 544 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 545 return 1; 546 #endif 547 return 0; 548 } 549 550 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 551 { 552 #ifdef CONFIG_X86_F00F_BUG 553 unsigned long nr; 554 555 /* 556 * Pentium F0 0F C7 C8 bug workaround: 557 */ 558 if (boot_cpu_data.f00f_bug) { 559 nr = (address - idt_descr.address) >> 3; 560 561 if (nr == 6) { 562 do_invalid_op(regs, 0); 563 return 1; 564 } 565 } 566 #endif 567 return 0; 568 } 569 570 static const char nx_warning[] = KERN_CRIT 571 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; 572 573 static void 574 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 575 unsigned long address) 576 { 577 if (!oops_may_print()) 578 return; 579 580 if (error_code & PF_INSTR) { 581 unsigned int level; 582 583 pte_t *pte = lookup_address(address, &level); 584 585 if (pte && pte_present(*pte) && !pte_exec(*pte)) 586 printk(nx_warning, from_kuid(&init_user_ns, current_uid())); 587 } 588 589 printk(KERN_ALERT "BUG: unable to handle kernel "); 590 if (address < PAGE_SIZE) 591 printk(KERN_CONT "NULL pointer dereference"); 592 else 593 printk(KERN_CONT "paging request"); 594 595 printk(KERN_CONT " at %p\n", (void *) address); 596 printk(KERN_ALERT "IP:"); 597 printk_address(regs->ip, 1); 598 599 dump_pagetable(address); 600 } 601 602 static noinline void 603 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 604 unsigned long address) 605 { 606 struct task_struct *tsk; 607 unsigned long flags; 608 int sig; 609 610 flags = oops_begin(); 611 tsk = current; 612 sig = SIGKILL; 613 614 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 615 tsk->comm, address); 616 dump_pagetable(address); 617 618 tsk->thread.cr2 = address; 619 tsk->thread.trap_nr = X86_TRAP_PF; 620 tsk->thread.error_code = error_code; 621 622 if (__die("Bad pagetable", regs, error_code)) 623 sig = 0; 624 625 oops_end(flags, regs, sig); 626 } 627 628 static noinline void 629 no_context(struct pt_regs *regs, unsigned long error_code, 630 unsigned long address, int signal, int si_code) 631 { 632 struct task_struct *tsk = current; 633 unsigned long *stackend; 634 unsigned long flags; 635 int sig; 636 637 /* Are we prepared to handle this kernel fault? */ 638 if (fixup_exception(regs)) { 639 if (current_thread_info()->sig_on_uaccess_error && signal) { 640 tsk->thread.trap_nr = X86_TRAP_PF; 641 tsk->thread.error_code = error_code | PF_USER; 642 tsk->thread.cr2 = address; 643 644 /* XXX: hwpoison faults will set the wrong code. */ 645 force_sig_info_fault(signal, si_code, address, tsk, 0); 646 } 647 return; 648 } 649 650 /* 651 * 32-bit: 652 * 653 * Valid to do another page fault here, because if this fault 654 * had been triggered by is_prefetch fixup_exception would have 655 * handled it. 656 * 657 * 64-bit: 658 * 659 * Hall of shame of CPU/BIOS bugs. 660 */ 661 if (is_prefetch(regs, error_code, address)) 662 return; 663 664 if (is_errata93(regs, address)) 665 return; 666 667 /* 668 * Oops. The kernel tried to access some bad page. We'll have to 669 * terminate things with extreme prejudice: 670 */ 671 flags = oops_begin(); 672 673 show_fault_oops(regs, error_code, address); 674 675 stackend = end_of_stack(tsk); 676 if (tsk != &init_task && *stackend != STACK_END_MAGIC) 677 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 678 679 tsk->thread.cr2 = address; 680 tsk->thread.trap_nr = X86_TRAP_PF; 681 tsk->thread.error_code = error_code; 682 683 sig = SIGKILL; 684 if (__die("Oops", regs, error_code)) 685 sig = 0; 686 687 /* Executive summary in case the body of the oops scrolled away */ 688 printk(KERN_DEFAULT "CR2: %016lx\n", address); 689 690 oops_end(flags, regs, sig); 691 } 692 693 /* 694 * Print out info about fatal segfaults, if the show_unhandled_signals 695 * sysctl is set: 696 */ 697 static inline void 698 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 699 unsigned long address, struct task_struct *tsk) 700 { 701 if (!unhandled_signal(tsk, SIGSEGV)) 702 return; 703 704 if (!printk_ratelimit()) 705 return; 706 707 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", 708 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, 709 tsk->comm, task_pid_nr(tsk), address, 710 (void *)regs->ip, (void *)regs->sp, error_code); 711 712 print_vma_addr(KERN_CONT " in ", regs->ip); 713 714 printk(KERN_CONT "\n"); 715 } 716 717 static void 718 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 719 unsigned long address, int si_code) 720 { 721 struct task_struct *tsk = current; 722 723 /* User mode accesses just cause a SIGSEGV */ 724 if (error_code & PF_USER) { 725 /* 726 * It's possible to have interrupts off here: 727 */ 728 local_irq_enable(); 729 730 /* 731 * Valid to do another page fault here because this one came 732 * from user space: 733 */ 734 if (is_prefetch(regs, error_code, address)) 735 return; 736 737 if (is_errata100(regs, address)) 738 return; 739 740 #ifdef CONFIG_X86_64 741 /* 742 * Instruction fetch faults in the vsyscall page might need 743 * emulation. 744 */ 745 if (unlikely((error_code & PF_INSTR) && 746 ((address & ~0xfff) == VSYSCALL_START))) { 747 if (emulate_vsyscall(regs, address)) 748 return; 749 } 750 #endif 751 752 if (unlikely(show_unhandled_signals)) 753 show_signal_msg(regs, error_code, address, tsk); 754 755 /* Kernel addresses are always protection faults: */ 756 tsk->thread.cr2 = address; 757 tsk->thread.error_code = error_code | (address >= TASK_SIZE); 758 tsk->thread.trap_nr = X86_TRAP_PF; 759 760 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); 761 762 return; 763 } 764 765 if (is_f00f_bug(regs, address)) 766 return; 767 768 no_context(regs, error_code, address, SIGSEGV, si_code); 769 } 770 771 static noinline void 772 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 773 unsigned long address) 774 { 775 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); 776 } 777 778 static void 779 __bad_area(struct pt_regs *regs, unsigned long error_code, 780 unsigned long address, int si_code) 781 { 782 struct mm_struct *mm = current->mm; 783 784 /* 785 * Something tried to access memory that isn't in our memory map.. 786 * Fix it, but check if it's kernel or user first.. 787 */ 788 up_read(&mm->mmap_sem); 789 790 __bad_area_nosemaphore(regs, error_code, address, si_code); 791 } 792 793 static noinline void 794 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 795 { 796 __bad_area(regs, error_code, address, SEGV_MAPERR); 797 } 798 799 static noinline void 800 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 801 unsigned long address) 802 { 803 __bad_area(regs, error_code, address, SEGV_ACCERR); 804 } 805 806 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */ 807 static void 808 out_of_memory(struct pt_regs *regs, unsigned long error_code, 809 unsigned long address) 810 { 811 /* 812 * We ran out of memory, call the OOM killer, and return the userspace 813 * (which will retry the fault, or kill us if we got oom-killed): 814 */ 815 up_read(¤t->mm->mmap_sem); 816 817 pagefault_out_of_memory(); 818 } 819 820 static void 821 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 822 unsigned int fault) 823 { 824 struct task_struct *tsk = current; 825 struct mm_struct *mm = tsk->mm; 826 int code = BUS_ADRERR; 827 828 up_read(&mm->mmap_sem); 829 830 /* Kernel mode? Handle exceptions or die: */ 831 if (!(error_code & PF_USER)) { 832 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 833 return; 834 } 835 836 /* User-space => ok to do another page fault: */ 837 if (is_prefetch(regs, error_code, address)) 838 return; 839 840 tsk->thread.cr2 = address; 841 tsk->thread.error_code = error_code; 842 tsk->thread.trap_nr = X86_TRAP_PF; 843 844 #ifdef CONFIG_MEMORY_FAILURE 845 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 846 printk(KERN_ERR 847 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 848 tsk->comm, tsk->pid, address); 849 code = BUS_MCEERR_AR; 850 } 851 #endif 852 force_sig_info_fault(SIGBUS, code, address, tsk, fault); 853 } 854 855 static noinline int 856 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 857 unsigned long address, unsigned int fault) 858 { 859 /* 860 * Pagefault was interrupted by SIGKILL. We have no reason to 861 * continue pagefault. 862 */ 863 if (fatal_signal_pending(current)) { 864 if (!(fault & VM_FAULT_RETRY)) 865 up_read(¤t->mm->mmap_sem); 866 if (!(error_code & PF_USER)) 867 no_context(regs, error_code, address, 0, 0); 868 return 1; 869 } 870 if (!(fault & VM_FAULT_ERROR)) 871 return 0; 872 873 if (fault & VM_FAULT_OOM) { 874 /* Kernel mode? Handle exceptions or die: */ 875 if (!(error_code & PF_USER)) { 876 up_read(¤t->mm->mmap_sem); 877 no_context(regs, error_code, address, 878 SIGSEGV, SEGV_MAPERR); 879 return 1; 880 } 881 882 out_of_memory(regs, error_code, address); 883 } else { 884 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 885 VM_FAULT_HWPOISON_LARGE)) 886 do_sigbus(regs, error_code, address, fault); 887 else 888 BUG(); 889 } 890 return 1; 891 } 892 893 static int spurious_fault_check(unsigned long error_code, pte_t *pte) 894 { 895 if ((error_code & PF_WRITE) && !pte_write(*pte)) 896 return 0; 897 898 if ((error_code & PF_INSTR) && !pte_exec(*pte)) 899 return 0; 900 901 return 1; 902 } 903 904 /* 905 * Handle a spurious fault caused by a stale TLB entry. 906 * 907 * This allows us to lazily refresh the TLB when increasing the 908 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 909 * eagerly is very expensive since that implies doing a full 910 * cross-processor TLB flush, even if no stale TLB entries exist 911 * on other processors. 912 * 913 * There are no security implications to leaving a stale TLB when 914 * increasing the permissions on a page. 915 */ 916 static noinline __kprobes int 917 spurious_fault(unsigned long error_code, unsigned long address) 918 { 919 pgd_t *pgd; 920 pud_t *pud; 921 pmd_t *pmd; 922 pte_t *pte; 923 int ret; 924 925 /* Reserved-bit violation or user access to kernel space? */ 926 if (error_code & (PF_USER | PF_RSVD)) 927 return 0; 928 929 pgd = init_mm.pgd + pgd_index(address); 930 if (!pgd_present(*pgd)) 931 return 0; 932 933 pud = pud_offset(pgd, address); 934 if (!pud_present(*pud)) 935 return 0; 936 937 if (pud_large(*pud)) 938 return spurious_fault_check(error_code, (pte_t *) pud); 939 940 pmd = pmd_offset(pud, address); 941 if (!pmd_present(*pmd)) 942 return 0; 943 944 if (pmd_large(*pmd)) 945 return spurious_fault_check(error_code, (pte_t *) pmd); 946 947 /* 948 * Note: don't use pte_present() here, since it returns true 949 * if the _PAGE_PROTNONE bit is set. However, this aliases the 950 * _PAGE_GLOBAL bit, which for kernel pages give false positives 951 * when CONFIG_DEBUG_PAGEALLOC is used. 952 */ 953 pte = pte_offset_kernel(pmd, address); 954 if (!(pte_flags(*pte) & _PAGE_PRESENT)) 955 return 0; 956 957 ret = spurious_fault_check(error_code, pte); 958 if (!ret) 959 return 0; 960 961 /* 962 * Make sure we have permissions in PMD. 963 * If not, then there's a bug in the page tables: 964 */ 965 ret = spurious_fault_check(error_code, (pte_t *) pmd); 966 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 967 968 return ret; 969 } 970 971 int show_unhandled_signals = 1; 972 973 static inline int 974 access_error(unsigned long error_code, struct vm_area_struct *vma) 975 { 976 if (error_code & PF_WRITE) { 977 /* write, present and write, not present: */ 978 if (unlikely(!(vma->vm_flags & VM_WRITE))) 979 return 1; 980 return 0; 981 } 982 983 /* read, present: */ 984 if (unlikely(error_code & PF_PROT)) 985 return 1; 986 987 /* read, not present: */ 988 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 989 return 1; 990 991 return 0; 992 } 993 994 static int fault_in_kernel_space(unsigned long address) 995 { 996 return address >= TASK_SIZE_MAX; 997 } 998 999 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1000 { 1001 if (error_code & PF_USER) 1002 return false; 1003 1004 if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) 1005 return false; 1006 1007 return true; 1008 } 1009 1010 /* 1011 * This routine handles page faults. It determines the address, 1012 * and the problem, and then passes it off to one of the appropriate 1013 * routines. 1014 */ 1015 static void __kprobes 1016 __do_page_fault(struct pt_regs *regs, unsigned long error_code) 1017 { 1018 struct vm_area_struct *vma; 1019 struct task_struct *tsk; 1020 unsigned long address; 1021 struct mm_struct *mm; 1022 int fault; 1023 int write = error_code & PF_WRITE; 1024 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE | 1025 (write ? FAULT_FLAG_WRITE : 0); 1026 1027 tsk = current; 1028 mm = tsk->mm; 1029 1030 /* Get the faulting address: */ 1031 address = read_cr2(); 1032 1033 /* 1034 * Detect and handle instructions that would cause a page fault for 1035 * both a tracked kernel page and a userspace page. 1036 */ 1037 if (kmemcheck_active(regs)) 1038 kmemcheck_hide(regs); 1039 prefetchw(&mm->mmap_sem); 1040 1041 if (unlikely(kmmio_fault(regs, address))) 1042 return; 1043 1044 /* 1045 * We fault-in kernel-space virtual memory on-demand. The 1046 * 'reference' page table is init_mm.pgd. 1047 * 1048 * NOTE! We MUST NOT take any locks for this case. We may 1049 * be in an interrupt or a critical region, and should 1050 * only copy the information from the master page table, 1051 * nothing more. 1052 * 1053 * This verifies that the fault happens in kernel space 1054 * (error_code & 4) == 0, and that the fault was not a 1055 * protection error (error_code & 9) == 0. 1056 */ 1057 if (unlikely(fault_in_kernel_space(address))) { 1058 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { 1059 if (vmalloc_fault(address) >= 0) 1060 return; 1061 1062 if (kmemcheck_fault(regs, address, error_code)) 1063 return; 1064 } 1065 1066 /* Can handle a stale RO->RW TLB: */ 1067 if (spurious_fault(error_code, address)) 1068 return; 1069 1070 /* kprobes don't want to hook the spurious faults: */ 1071 if (notify_page_fault(regs)) 1072 return; 1073 /* 1074 * Don't take the mm semaphore here. If we fixup a prefetch 1075 * fault we could otherwise deadlock: 1076 */ 1077 bad_area_nosemaphore(regs, error_code, address); 1078 1079 return; 1080 } 1081 1082 /* kprobes don't want to hook the spurious faults: */ 1083 if (unlikely(notify_page_fault(regs))) 1084 return; 1085 /* 1086 * It's safe to allow irq's after cr2 has been saved and the 1087 * vmalloc fault has been handled. 1088 * 1089 * User-mode registers count as a user access even for any 1090 * potential system fault or CPU buglet: 1091 */ 1092 if (user_mode_vm(regs)) { 1093 local_irq_enable(); 1094 error_code |= PF_USER; 1095 } else { 1096 if (regs->flags & X86_EFLAGS_IF) 1097 local_irq_enable(); 1098 } 1099 1100 if (unlikely(error_code & PF_RSVD)) 1101 pgtable_bad(regs, error_code, address); 1102 1103 if (static_cpu_has(X86_FEATURE_SMAP)) { 1104 if (unlikely(smap_violation(error_code, regs))) { 1105 bad_area_nosemaphore(regs, error_code, address); 1106 return; 1107 } 1108 } 1109 1110 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1111 1112 /* 1113 * If we're in an interrupt, have no user context or are running 1114 * in an atomic region then we must not take the fault: 1115 */ 1116 if (unlikely(in_atomic() || !mm)) { 1117 bad_area_nosemaphore(regs, error_code, address); 1118 return; 1119 } 1120 1121 /* 1122 * When running in the kernel we expect faults to occur only to 1123 * addresses in user space. All other faults represent errors in 1124 * the kernel and should generate an OOPS. Unfortunately, in the 1125 * case of an erroneous fault occurring in a code path which already 1126 * holds mmap_sem we will deadlock attempting to validate the fault 1127 * against the address space. Luckily the kernel only validly 1128 * references user space from well defined areas of code, which are 1129 * listed in the exceptions table. 1130 * 1131 * As the vast majority of faults will be valid we will only perform 1132 * the source reference check when there is a possibility of a 1133 * deadlock. Attempt to lock the address space, if we cannot we then 1134 * validate the source. If this is invalid we can skip the address 1135 * space check, thus avoiding the deadlock: 1136 */ 1137 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1138 if ((error_code & PF_USER) == 0 && 1139 !search_exception_tables(regs->ip)) { 1140 bad_area_nosemaphore(regs, error_code, address); 1141 return; 1142 } 1143 retry: 1144 down_read(&mm->mmap_sem); 1145 } else { 1146 /* 1147 * The above down_read_trylock() might have succeeded in 1148 * which case we'll have missed the might_sleep() from 1149 * down_read(): 1150 */ 1151 might_sleep(); 1152 } 1153 1154 vma = find_vma(mm, address); 1155 if (unlikely(!vma)) { 1156 bad_area(regs, error_code, address); 1157 return; 1158 } 1159 if (likely(vma->vm_start <= address)) 1160 goto good_area; 1161 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1162 bad_area(regs, error_code, address); 1163 return; 1164 } 1165 if (error_code & PF_USER) { 1166 /* 1167 * Accessing the stack below %sp is always a bug. 1168 * The large cushion allows instructions like enter 1169 * and pusha to work. ("enter $65535, $31" pushes 1170 * 32 pointers and then decrements %sp by 65535.) 1171 */ 1172 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1173 bad_area(regs, error_code, address); 1174 return; 1175 } 1176 } 1177 if (unlikely(expand_stack(vma, address))) { 1178 bad_area(regs, error_code, address); 1179 return; 1180 } 1181 1182 /* 1183 * Ok, we have a good vm_area for this memory access, so 1184 * we can handle it.. 1185 */ 1186 good_area: 1187 if (unlikely(access_error(error_code, vma))) { 1188 bad_area_access_error(regs, error_code, address); 1189 return; 1190 } 1191 1192 /* 1193 * If for any reason at all we couldn't handle the fault, 1194 * make sure we exit gracefully rather than endlessly redo 1195 * the fault: 1196 */ 1197 fault = handle_mm_fault(mm, vma, address, flags); 1198 1199 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) { 1200 if (mm_fault_error(regs, error_code, address, fault)) 1201 return; 1202 } 1203 1204 /* 1205 * Major/minor page fault accounting is only done on the 1206 * initial attempt. If we go through a retry, it is extremely 1207 * likely that the page will be found in page cache at that point. 1208 */ 1209 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1210 if (fault & VM_FAULT_MAJOR) { 1211 tsk->maj_flt++; 1212 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 1213 regs, address); 1214 } else { 1215 tsk->min_flt++; 1216 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 1217 regs, address); 1218 } 1219 if (fault & VM_FAULT_RETRY) { 1220 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 1221 * of starvation. */ 1222 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1223 flags |= FAULT_FLAG_TRIED; 1224 goto retry; 1225 } 1226 } 1227 1228 check_v8086_mode(regs, address, tsk); 1229 1230 up_read(&mm->mmap_sem); 1231 } 1232 1233 dotraplinkage void __kprobes 1234 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1235 { 1236 exception_enter(regs); 1237 __do_page_fault(regs, error_code); 1238 exception_exit(regs); 1239 } 1240