1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * PowerPC version 4 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) 5 * 6 * Derived from "arch/i386/mm/fault.c" 7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 8 * 9 * Modified by Cort Dougan and Paul Mackerras. 10 * 11 * Modified for PPC64 by Dave Engebretsen (engebret@ibm.com) 12 */ 13 14 #include <linux/signal.h> 15 #include <linux/sched.h> 16 #include <linux/sched/task_stack.h> 17 #include <linux/kernel.h> 18 #include <linux/errno.h> 19 #include <linux/string.h> 20 #include <linux/types.h> 21 #include <linux/pagemap.h> 22 #include <linux/ptrace.h> 23 #include <linux/mman.h> 24 #include <linux/mm.h> 25 #include <linux/interrupt.h> 26 #include <linux/highmem.h> 27 #include <linux/extable.h> 28 #include <linux/kprobes.h> 29 #include <linux/kdebug.h> 30 #include <linux/perf_event.h> 31 #include <linux/ratelimit.h> 32 #include <linux/context_tracking.h> 33 #include <linux/hugetlb.h> 34 #include <linux/uaccess.h> 35 #include <linux/kfence.h> 36 #include <linux/pkeys.h> 37 38 #include <asm/asm-prototypes.h> 39 #include <asm/firmware.h> 40 #include <asm/interrupt.h> 41 #include <asm/page.h> 42 #include <asm/mmu.h> 43 #include <asm/mmu_context.h> 44 #include <asm/siginfo.h> 45 #include <asm/debug.h> 46 #include <asm/kup.h> 47 #include <asm/inst.h> 48 49 50 /* 51 * do_page_fault error handling helpers 52 */ 53 54 static int 55 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code) 56 { 57 /* 58 * If we are in kernel mode, bail out with a SEGV, this will 59 * be caught by the assembly which will restore the non-volatile 60 * registers before calling bad_page_fault() 61 */ 62 if (!user_mode(regs)) 63 return SIGSEGV; 64 65 _exception(SIGSEGV, regs, si_code, address); 66 67 return 0; 68 } 69 70 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address) 71 { 72 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR); 73 } 74 75 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code) 76 { 77 struct mm_struct *mm = current->mm; 78 79 /* 80 * Something tried to access memory that isn't in our memory map.. 81 * Fix it, but check if it's kernel or user first.. 82 */ 83 mmap_read_unlock(mm); 84 85 return __bad_area_nosemaphore(regs, address, si_code); 86 } 87 88 static noinline int bad_area(struct pt_regs *regs, unsigned long address) 89 { 90 return __bad_area(regs, address, SEGV_MAPERR); 91 } 92 93 static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address, 94 struct vm_area_struct *vma) 95 { 96 struct mm_struct *mm = current->mm; 97 int pkey; 98 99 /* 100 * We don't try to fetch the pkey from page table because reading 101 * page table without locking doesn't guarantee stable pte value. 102 * Hence the pkey value that we return to userspace can be different 103 * from the pkey that actually caused access error. 104 * 105 * It does *not* guarantee that the VMA we find here 106 * was the one that we faulted on. 107 * 108 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 109 * 2. T1 : set AMR to deny access to pkey=4, touches, page 110 * 3. T1 : faults... 111 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 112 * 5. T1 : enters fault handler, takes mmap_lock, etc... 113 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 114 * faulted on a pte with its pkey=4. 115 */ 116 pkey = vma_pkey(vma); 117 118 mmap_read_unlock(mm); 119 120 /* 121 * If we are in kernel mode, bail out with a SEGV, this will 122 * be caught by the assembly which will restore the non-volatile 123 * registers before calling bad_page_fault() 124 */ 125 if (!user_mode(regs)) 126 return SIGSEGV; 127 128 _exception_pkey(regs, address, pkey); 129 130 return 0; 131 } 132 133 static noinline int bad_access(struct pt_regs *regs, unsigned long address) 134 { 135 return __bad_area(regs, address, SEGV_ACCERR); 136 } 137 138 static int do_sigbus(struct pt_regs *regs, unsigned long address, 139 vm_fault_t fault) 140 { 141 if (!user_mode(regs)) 142 return SIGBUS; 143 144 current->thread.trap_nr = BUS_ADRERR; 145 #ifdef CONFIG_MEMORY_FAILURE 146 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 147 unsigned int lsb = 0; /* shutup gcc */ 148 149 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 150 current->comm, current->pid, address); 151 152 if (fault & VM_FAULT_HWPOISON_LARGE) 153 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 154 if (fault & VM_FAULT_HWPOISON) 155 lsb = PAGE_SHIFT; 156 157 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); 158 return 0; 159 } 160 161 #endif 162 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); 163 return 0; 164 } 165 166 static int mm_fault_error(struct pt_regs *regs, unsigned long addr, 167 vm_fault_t fault) 168 { 169 /* 170 * Kernel page fault interrupted by SIGKILL. We have no reason to 171 * continue processing. 172 */ 173 if (fatal_signal_pending(current) && !user_mode(regs)) 174 return SIGKILL; 175 176 /* Out of memory */ 177 if (fault & VM_FAULT_OOM) { 178 /* 179 * We ran out of memory, or some other thing happened to us that 180 * made us unable to handle the page fault gracefully. 181 */ 182 if (!user_mode(regs)) 183 return SIGSEGV; 184 pagefault_out_of_memory(); 185 } else { 186 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 187 VM_FAULT_HWPOISON_LARGE)) 188 return do_sigbus(regs, addr, fault); 189 else if (fault & VM_FAULT_SIGSEGV) 190 return bad_area_nosemaphore(regs, addr); 191 else 192 BUG(); 193 } 194 return 0; 195 } 196 197 /* Is this a bad kernel fault ? */ 198 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code, 199 unsigned long address, bool is_write) 200 { 201 int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE; 202 203 if (is_exec) { 204 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n", 205 address >= TASK_SIZE ? "exec-protected" : "user", 206 address, 207 from_kuid(&init_user_ns, current_uid())); 208 209 // Kernel exec fault is always bad 210 return true; 211 } 212 213 // Kernel fault on kernel address is bad 214 if (address >= TASK_SIZE) 215 return true; 216 217 // Read/write fault blocked by KUAP is bad, it can never succeed. 218 if (bad_kuap_fault(regs, address, is_write)) { 219 pr_crit_ratelimited("Kernel attempted to %s user page (%lx) - exploit attempt? (uid: %d)\n", 220 is_write ? "write" : "read", address, 221 from_kuid(&init_user_ns, current_uid())); 222 223 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad 224 if (!search_exception_tables(regs->nip)) 225 return true; 226 227 // Read/write fault in a valid region (the exception table search passed 228 // above), but blocked by KUAP is bad, it can never succeed. 229 return WARN(true, "Bug: %s fault blocked by KUAP!", is_write ? "Write" : "Read"); 230 } 231 232 // What's left? Kernel fault on user and allowed by KUAP in the faulting context. 233 return false; 234 } 235 236 static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey, 237 struct vm_area_struct *vma) 238 { 239 /* 240 * Make sure to check the VMA so that we do not perform 241 * faults just to hit a pkey fault as soon as we fill in a 242 * page. Only called for current mm, hence foreign == 0 243 */ 244 if (!arch_vma_access_permitted(vma, is_write, is_exec, 0)) 245 return true; 246 247 return false; 248 } 249 250 static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma) 251 { 252 /* 253 * Allow execution from readable areas if the MMU does not 254 * provide separate controls over reading and executing. 255 * 256 * Note: That code used to not be enabled for 4xx/BookE. 257 * It is now as I/D cache coherency for these is done at 258 * set_pte_at() time and I see no reason why the test 259 * below wouldn't be valid on those processors. This -may- 260 * break programs compiled with a really old ABI though. 261 */ 262 if (is_exec) { 263 return !(vma->vm_flags & VM_EXEC) && 264 (cpu_has_feature(CPU_FTR_NOEXECUTE) || 265 !(vma->vm_flags & (VM_READ | VM_WRITE))); 266 } 267 268 if (is_write) { 269 if (unlikely(!(vma->vm_flags & VM_WRITE))) 270 return true; 271 return false; 272 } 273 274 if (unlikely(!vma_is_accessible(vma))) 275 return true; 276 /* 277 * We should ideally do the vma pkey access check here. But in the 278 * fault path, handle_mm_fault() also does the same check. To avoid 279 * these multiple checks, we skip it here and handle access error due 280 * to pkeys later. 281 */ 282 return false; 283 } 284 285 #ifdef CONFIG_PPC_SMLPAR 286 static inline void cmo_account_page_fault(void) 287 { 288 if (firmware_has_feature(FW_FEATURE_CMO)) { 289 u32 page_ins; 290 291 preempt_disable(); 292 page_ins = be32_to_cpu(get_lppaca()->page_ins); 293 page_ins += 1 << PAGE_FACTOR; 294 get_lppaca()->page_ins = cpu_to_be32(page_ins); 295 preempt_enable(); 296 } 297 } 298 #else 299 static inline void cmo_account_page_fault(void) { } 300 #endif /* CONFIG_PPC_SMLPAR */ 301 302 static void sanity_check_fault(bool is_write, bool is_user, 303 unsigned long error_code, unsigned long address) 304 { 305 /* 306 * Userspace trying to access kernel address, we get PROTFAULT for that. 307 */ 308 if (is_user && address >= TASK_SIZE) { 309 if ((long)address == -1) 310 return; 311 312 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n", 313 current->comm, current->pid, address, 314 from_kuid(&init_user_ns, current_uid())); 315 return; 316 } 317 318 if (!IS_ENABLED(CONFIG_PPC_BOOK3S)) 319 return; 320 321 /* 322 * For hash translation mode, we should never get a 323 * PROTFAULT. Any update to pte to reduce access will result in us 324 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE 325 * fault instead of DSISR_PROTFAULT. 326 * 327 * A pte update to relax the access will not result in a hash page table 328 * entry invalidate and hence can result in DSISR_PROTFAULT. 329 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have 330 * the special !is_write in the below conditional. 331 * 332 * For platforms that doesn't supports coherent icache and do support 333 * per page noexec bit, we do setup things such that we do the 334 * sync between D/I cache via fault. But that is handled via low level 335 * hash fault code (hash_page_do_lazy_icache()) and we should not reach 336 * here in such case. 337 * 338 * For wrong access that can result in PROTFAULT, the above vma->vm_flags 339 * check should handle those and hence we should fall to the bad_area 340 * handling correctly. 341 * 342 * For embedded with per page exec support that doesn't support coherent 343 * icache we do get PROTFAULT and we handle that D/I cache sync in 344 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON 345 * is conditional for server MMU. 346 * 347 * For radix, we can get prot fault for autonuma case, because radix 348 * page table will have them marked noaccess for user. 349 */ 350 if (radix_enabled() || is_write) 351 return; 352 353 WARN_ON_ONCE(error_code & DSISR_PROTFAULT); 354 } 355 356 /* 357 * Define the correct "is_write" bit in error_code based 358 * on the processor family 359 */ 360 #if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) 361 #define page_fault_is_write(__err) ((__err) & ESR_DST) 362 #else 363 #define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE) 364 #endif 365 366 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE) 367 #define page_fault_is_bad(__err) (0) 368 #elif defined(CONFIG_PPC_8xx) 369 #define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G) 370 #elif defined(CONFIG_PPC64) 371 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_64S) 372 #else 373 #define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S) 374 #endif 375 376 /* 377 * For 600- and 800-family processors, the error_code parameter is DSISR 378 * for a data fault, SRR1 for an instruction fault. 379 * For 400-family processors the error_code parameter is ESR for a data fault, 380 * 0 for an instruction fault. 381 * For 64-bit processors, the error_code parameter is DSISR for a data access 382 * fault, SRR1 & 0x08000000 for an instruction access fault. 383 * 384 * The return value is 0 if the fault was handled, or the signal 385 * number if this is a kernel fault that can't be handled here. 386 */ 387 static int ___do_page_fault(struct pt_regs *regs, unsigned long address, 388 unsigned long error_code) 389 { 390 struct vm_area_struct * vma; 391 struct mm_struct *mm = current->mm; 392 unsigned int flags = FAULT_FLAG_DEFAULT; 393 int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE; 394 int is_user = user_mode(regs); 395 int is_write = page_fault_is_write(error_code); 396 vm_fault_t fault, major = 0; 397 bool kprobe_fault = kprobe_page_fault(regs, 11); 398 399 if (unlikely(debugger_fault_handler(regs) || kprobe_fault)) 400 return 0; 401 402 if (unlikely(page_fault_is_bad(error_code))) { 403 if (is_user) { 404 _exception(SIGBUS, regs, BUS_OBJERR, address); 405 return 0; 406 } 407 return SIGBUS; 408 } 409 410 /* Additional sanity check(s) */ 411 sanity_check_fault(is_write, is_user, error_code, address); 412 413 /* 414 * The kernel should never take an execute fault nor should it 415 * take a page fault to a kernel address or a page fault to a user 416 * address outside of dedicated places 417 */ 418 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write))) { 419 if (kfence_handle_page_fault(address, is_write, regs)) 420 return 0; 421 422 return SIGSEGV; 423 } 424 425 /* 426 * If we're in an interrupt, have no user context or are running 427 * in a region with pagefaults disabled then we must not take the fault 428 */ 429 if (unlikely(faulthandler_disabled() || !mm)) { 430 if (is_user) 431 printk_ratelimited(KERN_ERR "Page fault in user mode" 432 " with faulthandler_disabled()=%d" 433 " mm=%p\n", 434 faulthandler_disabled(), mm); 435 return bad_area_nosemaphore(regs, address); 436 } 437 438 interrupt_cond_local_irq_enable(regs); 439 440 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 441 442 /* 443 * We want to do this outside mmap_lock, because reading code around nip 444 * can result in fault, which will cause a deadlock when called with 445 * mmap_lock held 446 */ 447 if (is_user) 448 flags |= FAULT_FLAG_USER; 449 if (is_write) 450 flags |= FAULT_FLAG_WRITE; 451 if (is_exec) 452 flags |= FAULT_FLAG_INSTRUCTION; 453 454 /* When running in the kernel we expect faults to occur only to 455 * addresses in user space. All other faults represent errors in the 456 * kernel and should generate an OOPS. Unfortunately, in the case of an 457 * erroneous fault occurring in a code path which already holds mmap_lock 458 * we will deadlock attempting to validate the fault against the 459 * address space. Luckily the kernel only validly references user 460 * space from well defined areas of code, which are listed in the 461 * exceptions table. 462 * 463 * As the vast majority of faults will be valid we will only perform 464 * the source reference check when there is a possibility of a deadlock. 465 * Attempt to lock the address space, if we cannot we then validate the 466 * source. If this is invalid we can skip the address space check, 467 * thus avoiding the deadlock. 468 */ 469 if (unlikely(!mmap_read_trylock(mm))) { 470 if (!is_user && !search_exception_tables(regs->nip)) 471 return bad_area_nosemaphore(regs, address); 472 473 retry: 474 mmap_read_lock(mm); 475 } else { 476 /* 477 * The above down_read_trylock() might have succeeded in 478 * which case we'll have missed the might_sleep() from 479 * down_read(): 480 */ 481 might_sleep(); 482 } 483 484 vma = find_vma(mm, address); 485 if (unlikely(!vma)) 486 return bad_area(regs, address); 487 488 if (unlikely(vma->vm_start > address)) { 489 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) 490 return bad_area(regs, address); 491 492 if (unlikely(expand_stack(vma, address))) 493 return bad_area(regs, address); 494 } 495 496 if (unlikely(access_pkey_error(is_write, is_exec, 497 (error_code & DSISR_KEYFAULT), vma))) 498 return bad_access_pkey(regs, address, vma); 499 500 if (unlikely(access_error(is_write, is_exec, vma))) 501 return bad_access(regs, address); 502 503 /* 504 * If for any reason at all we couldn't handle the fault, 505 * make sure we exit gracefully rather than endlessly redo 506 * the fault. 507 */ 508 fault = handle_mm_fault(vma, address, flags, regs); 509 510 major |= fault & VM_FAULT_MAJOR; 511 512 if (fault_signal_pending(fault, regs)) 513 return user_mode(regs) ? 0 : SIGBUS; 514 515 /* 516 * Handle the retry right now, the mmap_lock has been released in that 517 * case. 518 */ 519 if (unlikely(fault & VM_FAULT_RETRY)) { 520 flags |= FAULT_FLAG_TRIED; 521 goto retry; 522 } 523 524 mmap_read_unlock(current->mm); 525 526 if (unlikely(fault & VM_FAULT_ERROR)) 527 return mm_fault_error(regs, address, fault); 528 529 /* 530 * Major/minor page fault accounting. 531 */ 532 if (major) 533 cmo_account_page_fault(); 534 535 return 0; 536 } 537 NOKPROBE_SYMBOL(___do_page_fault); 538 539 static __always_inline void __do_page_fault(struct pt_regs *regs) 540 { 541 long err; 542 543 err = ___do_page_fault(regs, regs->dar, regs->dsisr); 544 if (unlikely(err)) 545 bad_page_fault(regs, err); 546 } 547 548 DEFINE_INTERRUPT_HANDLER(do_page_fault) 549 { 550 __do_page_fault(regs); 551 } 552 553 #ifdef CONFIG_PPC_BOOK3S_64 554 /* Same as do_page_fault but interrupt entry has already run in do_hash_fault */ 555 void hash__do_page_fault(struct pt_regs *regs) 556 { 557 __do_page_fault(regs); 558 } 559 NOKPROBE_SYMBOL(hash__do_page_fault); 560 #endif 561 562 /* 563 * bad_page_fault is called when we have a bad access from the kernel. 564 * It is called from the DSI and ISI handlers in head.S and from some 565 * of the procedures in traps.c. 566 */ 567 static void __bad_page_fault(struct pt_regs *regs, int sig) 568 { 569 int is_write = page_fault_is_write(regs->dsisr); 570 571 /* kernel has accessed a bad area */ 572 573 switch (TRAP(regs)) { 574 case INTERRUPT_DATA_STORAGE: 575 case INTERRUPT_DATA_SEGMENT: 576 case INTERRUPT_H_DATA_STORAGE: 577 pr_alert("BUG: %s on %s at 0x%08lx\n", 578 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" : 579 "Unable to handle kernel data access", 580 is_write ? "write" : "read", regs->dar); 581 break; 582 case INTERRUPT_INST_STORAGE: 583 case INTERRUPT_INST_SEGMENT: 584 pr_alert("BUG: Unable to handle kernel instruction fetch%s", 585 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n"); 586 break; 587 case INTERRUPT_ALIGNMENT: 588 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n", 589 regs->dar); 590 break; 591 default: 592 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n", 593 regs->dar); 594 break; 595 } 596 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", 597 regs->nip); 598 599 if (task_stack_end_corrupted(current)) 600 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); 601 602 die("Kernel access of bad area", regs, sig); 603 } 604 605 void bad_page_fault(struct pt_regs *regs, int sig) 606 { 607 const struct exception_table_entry *entry; 608 609 /* Are we prepared to handle this fault? */ 610 entry = search_exception_tables(instruction_pointer(regs)); 611 if (entry) 612 instruction_pointer_set(regs, extable_fixup(entry)); 613 else 614 __bad_page_fault(regs, sig); 615 } 616 617 #ifdef CONFIG_PPC_BOOK3S_64 618 DEFINE_INTERRUPT_HANDLER(do_bad_page_fault_segv) 619 { 620 bad_page_fault(regs, SIGSEGV); 621 } 622 623 /* 624 * In radix, segment interrupts indicate the EA is not addressable by the 625 * page table geometry, so they are always sent here. 626 * 627 * In hash, this is called if do_slb_fault returns error. Typically it is 628 * because the EA was outside the region allowed by software. 629 */ 630 DEFINE_INTERRUPT_HANDLER(do_bad_segment_interrupt) 631 { 632 int err = regs->result; 633 634 if (err == -EFAULT) { 635 if (user_mode(regs)) 636 _exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar); 637 else 638 bad_page_fault(regs, SIGSEGV); 639 } else if (err == -EINVAL) { 640 unrecoverable_exception(regs); 641 } else { 642 BUG(); 643 } 644 } 645 #endif 646