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/firmware.h> 39#include <asm/interrupt.h> 40#include <asm/page.h> 41#include <asm/mmu.h> 42#include <asm/mmu_context.h> 43#include <asm/siginfo.h> 44#include <asm/debug.h> 45#include <asm/kup.h> 46#include <asm/inst.h> 47 48 49/* 50 * do_page_fault error handling helpers 51 */ 52 53static int 54__bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code) 55{ 56 /* 57 * If we are in kernel mode, bail out with a SEGV, this will 58 * be caught by the assembly which will restore the non-volatile 59 * registers before calling bad_page_fault() 60 */ 61 if (!user_mode(regs)) 62 return SIGSEGV; 63 64 _exception(SIGSEGV, regs, si_code, address); 65 66 return 0; 67} 68 69static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address) 70{ 71 return __bad_area_nosemaphore(regs, address, SEGV_MAPERR); 72} 73 74static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code) 75{ 76 struct mm_struct *mm = current->mm; 77 78 /* 79 * Something tried to access memory that isn't in our memory map.. 80 * Fix it, but check if it's kernel or user first.. 81 */ 82 mmap_read_unlock(mm); 83 84 return __bad_area_nosemaphore(regs, address, si_code); 85} 86 87static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address, 88 struct vm_area_struct *vma) 89{ 90 struct mm_struct *mm = current->mm; 91 int pkey; 92 93 /* 94 * We don't try to fetch the pkey from page table because reading 95 * page table without locking doesn't guarantee stable pte value. 96 * Hence the pkey value that we return to userspace can be different 97 * from the pkey that actually caused access error. 98 * 99 * It does *not* guarantee that the VMA we find here 100 * was the one that we faulted on. 101 * 102 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 103 * 2. T1 : set AMR to deny access to pkey=4, touches, page 104 * 3. T1 : faults... 105 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 106 * 5. T1 : enters fault handler, takes mmap_lock, etc... 107 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 108 * faulted on a pte with its pkey=4. 109 */ 110 pkey = vma_pkey(vma); 111 112 mmap_read_unlock(mm); 113 114 /* 115 * If we are in kernel mode, bail out with a SEGV, this will 116 * be caught by the assembly which will restore the non-volatile 117 * registers before calling bad_page_fault() 118 */ 119 if (!user_mode(regs)) 120 return SIGSEGV; 121 122 _exception_pkey(regs, address, pkey); 123 124 return 0; 125} 126 127static noinline int bad_access(struct pt_regs *regs, unsigned long address) 128{ 129 return __bad_area(regs, address, SEGV_ACCERR); 130} 131 132static int do_sigbus(struct pt_regs *regs, unsigned long address, 133 vm_fault_t fault) 134{ 135 if (!user_mode(regs)) 136 return SIGBUS; 137 138 current->thread.trap_nr = BUS_ADRERR; 139#ifdef CONFIG_MEMORY_FAILURE 140 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 141 unsigned int lsb = 0; /* shutup gcc */ 142 143 pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 144 current->comm, current->pid, address); 145 146 if (fault & VM_FAULT_HWPOISON_LARGE) 147 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 148 if (fault & VM_FAULT_HWPOISON) 149 lsb = PAGE_SHIFT; 150 151 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); 152 return 0; 153 } 154 155#endif 156 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); 157 return 0; 158} 159 160static int mm_fault_error(struct pt_regs *regs, unsigned long addr, 161 vm_fault_t fault) 162{ 163 /* 164 * Kernel page fault interrupted by SIGKILL. We have no reason to 165 * continue processing. 166 */ 167 if (fatal_signal_pending(current) && !user_mode(regs)) 168 return SIGKILL; 169 170 /* Out of memory */ 171 if (fault & VM_FAULT_OOM) { 172 /* 173 * We ran out of memory, or some other thing happened to us that 174 * made us unable to handle the page fault gracefully. 175 */ 176 if (!user_mode(regs)) 177 return SIGSEGV; 178 pagefault_out_of_memory(); 179 } else { 180 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 181 VM_FAULT_HWPOISON_LARGE)) 182 return do_sigbus(regs, addr, fault); 183 else if (fault & VM_FAULT_SIGSEGV) 184 return bad_area_nosemaphore(regs, addr); 185 else 186 BUG(); 187 } 188 return 0; 189} 190 191/* Is this a bad kernel fault ? */ 192static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code, 193 unsigned long address, bool is_write) 194{ 195 int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE; 196 197 if (is_exec) { 198 pr_crit_ratelimited("kernel tried to execute %s page (%lx) - exploit attempt? (uid: %d)\n", 199 address >= TASK_SIZE ? "exec-protected" : "user", 200 address, 201 from_kuid(&init_user_ns, current_uid())); 202 203 // Kernel exec fault is always bad 204 return true; 205 } 206 207 // Kernel fault on kernel address is bad 208 if (address >= TASK_SIZE) 209 return true; 210 211 // Read/write fault blocked by KUAP is bad, it can never succeed. 212 if (bad_kuap_fault(regs, address, is_write)) { 213 pr_crit_ratelimited("Kernel attempted to %s user page (%lx) - exploit attempt? (uid: %d)\n", 214 is_write ? "write" : "read", address, 215 from_kuid(&init_user_ns, current_uid())); 216 217 // Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad 218 if (!search_exception_tables(regs->nip)) 219 return true; 220 221 // Read/write fault in a valid region (the exception table search passed 222 // above), but blocked by KUAP is bad, it can never succeed. 223 return WARN(true, "Bug: %s fault blocked by KUAP!", is_write ? "Write" : "Read"); 224 } 225 226 // What's left? Kernel fault on user and allowed by KUAP in the faulting context. 227 return false; 228} 229 230static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey, 231 struct vm_area_struct *vma) 232{ 233 /* 234 * Make sure to check the VMA so that we do not perform 235 * faults just to hit a pkey fault as soon as we fill in a 236 * page. Only called for current mm, hence foreign == 0 237 */ 238 if (!arch_vma_access_permitted(vma, is_write, is_exec, 0)) 239 return true; 240 241 return false; 242} 243 244static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma) 245{ 246 /* 247 * Allow execution from readable areas if the MMU does not 248 * provide separate controls over reading and executing. 249 * 250 * Note: That code used to not be enabled for 4xx/BookE. 251 * It is now as I/D cache coherency for these is done at 252 * set_pte_at() time and I see no reason why the test 253 * below wouldn't be valid on those processors. This -may- 254 * break programs compiled with a really old ABI though. 255 */ 256 if (is_exec) { 257 return !(vma->vm_flags & VM_EXEC) && 258 (cpu_has_feature(CPU_FTR_NOEXECUTE) || 259 !(vma->vm_flags & (VM_READ | VM_WRITE))); 260 } 261 262 if (is_write) { 263 if (unlikely(!(vma->vm_flags & VM_WRITE))) 264 return true; 265 return false; 266 } 267 268 /* 269 * VM_READ, VM_WRITE and VM_EXEC may imply read permissions, as 270 * defined in protection_map[]. In that case Read faults can only be 271 * caused by a PROT_NONE mapping. However a non exec access on a 272 * VM_EXEC only mapping is invalid anyway, so report it as such. 273 */ 274 if (unlikely(!vma_is_accessible(vma))) 275 return true; 276 277 if ((vma->vm_flags & VM_ACCESS_FLAGS) == VM_EXEC) 278 return true; 279 280 /* 281 * We should ideally do the vma pkey access check here. But in the 282 * fault path, handle_mm_fault() also does the same check. To avoid 283 * these multiple checks, we skip it here and handle access error due 284 * to pkeys later. 285 */ 286 return false; 287} 288 289#ifdef CONFIG_PPC_SMLPAR 290static inline void cmo_account_page_fault(void) 291{ 292 if (firmware_has_feature(FW_FEATURE_CMO)) { 293 u32 page_ins; 294 295 preempt_disable(); 296 page_ins = be32_to_cpu(get_lppaca()->page_ins); 297 page_ins += 1 << PAGE_FACTOR; 298 get_lppaca()->page_ins = cpu_to_be32(page_ins); 299 preempt_enable(); 300 } 301} 302#else 303static inline void cmo_account_page_fault(void) { } 304#endif /* CONFIG_PPC_SMLPAR */ 305 306static void sanity_check_fault(bool is_write, bool is_user, 307 unsigned long error_code, unsigned long address) 308{ 309 /* 310 * Userspace trying to access kernel address, we get PROTFAULT for that. 311 */ 312 if (is_user && address >= TASK_SIZE) { 313 if ((long)address == -1) 314 return; 315 316 pr_crit_ratelimited("%s[%d]: User access of kernel address (%lx) - exploit attempt? (uid: %d)\n", 317 current->comm, current->pid, address, 318 from_kuid(&init_user_ns, current_uid())); 319 return; 320 } 321 322 if (!IS_ENABLED(CONFIG_PPC_BOOK3S)) 323 return; 324 325 /* 326 * For hash translation mode, we should never get a 327 * PROTFAULT. Any update to pte to reduce access will result in us 328 * removing the hash page table entry, thus resulting in a DSISR_NOHPTE 329 * fault instead of DSISR_PROTFAULT. 330 * 331 * A pte update to relax the access will not result in a hash page table 332 * entry invalidate and hence can result in DSISR_PROTFAULT. 333 * ptep_set_access_flags() doesn't do a hpte flush. This is why we have 334 * the special !is_write in the below conditional. 335 * 336 * For platforms that doesn't supports coherent icache and do support 337 * per page noexec bit, we do setup things such that we do the 338 * sync between D/I cache via fault. But that is handled via low level 339 * hash fault code (hash_page_do_lazy_icache()) and we should not reach 340 * here in such case. 341 * 342 * For wrong access that can result in PROTFAULT, the above vma->vm_flags 343 * check should handle those and hence we should fall to the bad_area 344 * handling correctly. 345 * 346 * For embedded with per page exec support that doesn't support coherent 347 * icache we do get PROTFAULT and we handle that D/I cache sync in 348 * set_pte_at while taking the noexec/prot fault. Hence this is WARN_ON 349 * is conditional for server MMU. 350 * 351 * For radix, we can get prot fault for autonuma case, because radix 352 * page table will have them marked noaccess for user. 353 */ 354 if (radix_enabled() || is_write) 355 return; 356 357 WARN_ON_ONCE(error_code & DSISR_PROTFAULT); 358} 359 360/* 361 * Define the correct "is_write" bit in error_code based 362 * on the processor family 363 */ 364#if (defined(CONFIG_4xx) || defined(CONFIG_BOOKE)) 365#define page_fault_is_write(__err) ((__err) & ESR_DST) 366#else 367#define page_fault_is_write(__err) ((__err) & DSISR_ISSTORE) 368#endif 369 370#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE) 371#define page_fault_is_bad(__err) (0) 372#elif defined(CONFIG_PPC_8xx) 373#define page_fault_is_bad(__err) ((__err) & DSISR_NOEXEC_OR_G) 374#elif defined(CONFIG_PPC64) 375static int page_fault_is_bad(unsigned long err) 376{ 377 unsigned long flag = DSISR_BAD_FAULT_64S; 378 379 /* 380 * PAPR+ v2.11 �� 14.15.3.4.1 (unreleased) 381 * If byte 0, bit 3 of pi-attribute-specifier-type in 382 * ibm,pi-features property is defined, ignore the DSI error 383 * which is caused by the paste instruction on the 384 * suspended NX window. 385 */ 386 if (mmu_has_feature(MMU_FTR_NX_DSI)) 387 flag &= ~DSISR_BAD_COPYPASTE; 388 389 return err & flag; 390} 391#else 392#define page_fault_is_bad(__err) ((__err) & DSISR_BAD_FAULT_32S) 393#endif 394 395/* 396 * For 600- and 800-family processors, the error_code parameter is DSISR 397 * for a data fault, SRR1 for an instruction fault. 398 * For 400-family processors the error_code parameter is ESR for a data fault, 399 * 0 for an instruction fault. 400 * For 64-bit processors, the error_code parameter is DSISR for a data access 401 * fault, SRR1 & 0x08000000 for an instruction access fault. 402 * 403 * The return value is 0 if the fault was handled, or the signal 404 * number if this is a kernel fault that can't be handled here. 405 */ 406static int ___do_page_fault(struct pt_regs *regs, unsigned long address, 407 unsigned long error_code) 408{ 409 struct vm_area_struct * vma; 410 struct mm_struct *mm = current->mm; 411 unsigned int flags = FAULT_FLAG_DEFAULT; 412 int is_exec = TRAP(regs) == INTERRUPT_INST_STORAGE; 413 int is_user = user_mode(regs); 414 int is_write = page_fault_is_write(error_code); 415 vm_fault_t fault, major = 0; 416 bool kprobe_fault = kprobe_page_fault(regs, 11); 417 418 if (unlikely(debugger_fault_handler(regs) || kprobe_fault)) 419 return 0; 420 421 if (unlikely(page_fault_is_bad(error_code))) { 422 if (is_user) { 423 _exception(SIGBUS, regs, BUS_OBJERR, address); 424 return 0; 425 } 426 return SIGBUS; 427 } 428 429 /* Additional sanity check(s) */ 430 sanity_check_fault(is_write, is_user, error_code, address); 431 432 /* 433 * The kernel should never take an execute fault nor should it 434 * take a page fault to a kernel address or a page fault to a user 435 * address outside of dedicated places 436 */ 437 if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write))) { 438 if (kfence_handle_page_fault(address, is_write, regs)) 439 return 0; 440 441 return SIGSEGV; 442 } 443 444 /* 445 * If we're in an interrupt, have no user context or are running 446 * in a region with pagefaults disabled then we must not take the fault 447 */ 448 if (unlikely(faulthandler_disabled() || !mm)) { 449 if (is_user) 450 printk_ratelimited(KERN_ERR "Page fault in user mode" 451 " with faulthandler_disabled()=%d" 452 " mm=%p\n", 453 faulthandler_disabled(), mm); 454 return bad_area_nosemaphore(regs, address); 455 } 456 457 interrupt_cond_local_irq_enable(regs); 458 459 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 460 461 /* 462 * We want to do this outside mmap_lock, because reading code around nip 463 * can result in fault, which will cause a deadlock when called with 464 * mmap_lock held 465 */ 466 if (is_user) 467 flags |= FAULT_FLAG_USER; 468 if (is_write) 469 flags |= FAULT_FLAG_WRITE; 470 if (is_exec) 471 flags |= FAULT_FLAG_INSTRUCTION; 472 473 if (!(flags & FAULT_FLAG_USER)) 474 goto lock_mmap; 475 476 vma = lock_vma_under_rcu(mm, address); 477 if (!vma) 478 goto lock_mmap; 479 480 if (unlikely(access_pkey_error(is_write, is_exec, 481 (error_code & DSISR_KEYFAULT), vma))) { 482 vma_end_read(vma); 483 goto lock_mmap; 484 } 485 486 if (unlikely(access_error(is_write, is_exec, vma))) { 487 vma_end_read(vma); 488 goto lock_mmap; 489 } 490 491 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs); 492 if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED))) 493 vma_end_read(vma); 494 495 if (!(fault & VM_FAULT_RETRY)) { 496 count_vm_vma_lock_event(VMA_LOCK_SUCCESS); 497 goto done; 498 } 499 count_vm_vma_lock_event(VMA_LOCK_RETRY); 500 if (fault & VM_FAULT_MAJOR) 501 flags |= FAULT_FLAG_TRIED; 502 503 if (fault_signal_pending(fault, regs)) 504 return user_mode(regs) ? 0 : SIGBUS; 505 506lock_mmap: 507 508 /* When running in the kernel we expect faults to occur only to 509 * addresses in user space. All other faults represent errors in the 510 * kernel and should generate an OOPS. Unfortunately, in the case of an 511 * erroneous fault occurring in a code path which already holds mmap_lock 512 * we will deadlock attempting to validate the fault against the 513 * address space. Luckily the kernel only validly references user 514 * space from well defined areas of code, which are listed in the 515 * exceptions table. lock_mm_and_find_vma() handles that logic. 516 */ 517retry: 518 vma = lock_mm_and_find_vma(mm, address, regs); 519 if (unlikely(!vma)) 520 return bad_area_nosemaphore(regs, address); 521 522 if (unlikely(access_pkey_error(is_write, is_exec, 523 (error_code & DSISR_KEYFAULT), vma))) 524 return bad_access_pkey(regs, address, vma); 525 526 if (unlikely(access_error(is_write, is_exec, vma))) 527 return bad_access(regs, address); 528 529 /* 530 * If for any reason at all we couldn't handle the fault, 531 * make sure we exit gracefully rather than endlessly redo 532 * the fault. 533 */ 534 fault = handle_mm_fault(vma, address, flags, regs); 535 536 major |= fault & VM_FAULT_MAJOR; 537 538 if (fault_signal_pending(fault, regs)) 539 return user_mode(regs) ? 0 : SIGBUS; 540 541 /* The fault is fully completed (including releasing mmap lock) */ 542 if (fault & VM_FAULT_COMPLETED) 543 goto out; 544 545 /* 546 * Handle the retry right now, the mmap_lock has been released in that 547 * case. 548 */ 549 if (unlikely(fault & VM_FAULT_RETRY)) { 550 flags |= FAULT_FLAG_TRIED; 551 goto retry; 552 } 553 554 mmap_read_unlock(current->mm); 555 556done: 557 if (unlikely(fault & VM_FAULT_ERROR)) 558 return mm_fault_error(regs, address, fault); 559 560out: 561 /* 562 * Major/minor page fault accounting. 563 */ 564 if (major) 565 cmo_account_page_fault(); 566 567 return 0; 568} 569NOKPROBE_SYMBOL(___do_page_fault); 570 571static __always_inline void __do_page_fault(struct pt_regs *regs) 572{ 573 long err; 574 575 err = ___do_page_fault(regs, regs->dar, regs->dsisr); 576 if (unlikely(err)) 577 bad_page_fault(regs, err); 578} 579 580DEFINE_INTERRUPT_HANDLER(do_page_fault) 581{ 582 __do_page_fault(regs); 583} 584 585#ifdef CONFIG_PPC_BOOK3S_64 586/* Same as do_page_fault but interrupt entry has already run in do_hash_fault */ 587void hash__do_page_fault(struct pt_regs *regs) 588{ 589 __do_page_fault(regs); 590} 591NOKPROBE_SYMBOL(hash__do_page_fault); 592#endif 593 594/* 595 * bad_page_fault is called when we have a bad access from the kernel. 596 * It is called from the DSI and ISI handlers in head.S and from some 597 * of the procedures in traps.c. 598 */ 599static void __bad_page_fault(struct pt_regs *regs, int sig) 600{ 601 int is_write = page_fault_is_write(regs->dsisr); 602 const char *msg; 603 604 /* kernel has accessed a bad area */ 605 606 if (regs->dar < PAGE_SIZE) 607 msg = "Kernel NULL pointer dereference"; 608 else 609 msg = "Unable to handle kernel data access"; 610 611 switch (TRAP(regs)) { 612 case INTERRUPT_DATA_STORAGE: 613 case INTERRUPT_H_DATA_STORAGE: 614 pr_alert("BUG: %s on %s at 0x%08lx\n", msg, 615 is_write ? "write" : "read", regs->dar); 616 break; 617 case INTERRUPT_DATA_SEGMENT: 618 pr_alert("BUG: %s at 0x%08lx\n", msg, regs->dar); 619 break; 620 case INTERRUPT_INST_STORAGE: 621 case INTERRUPT_INST_SEGMENT: 622 pr_alert("BUG: Unable to handle kernel instruction fetch%s", 623 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n"); 624 break; 625 case INTERRUPT_ALIGNMENT: 626 pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n", 627 regs->dar); 628 break; 629 default: 630 pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n", 631 regs->dar); 632 break; 633 } 634 printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n", 635 regs->nip); 636 637 if (task_stack_end_corrupted(current)) 638 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n"); 639 640 die("Kernel access of bad area", regs, sig); 641} 642 643void bad_page_fault(struct pt_regs *regs, int sig) 644{ 645 const struct exception_table_entry *entry; 646 647 /* Are we prepared to handle this fault? */ 648 entry = search_exception_tables(instruction_pointer(regs)); 649 if (entry) 650 instruction_pointer_set(regs, extable_fixup(entry)); 651 else 652 __bad_page_fault(regs, sig); 653} 654 655#ifdef CONFIG_PPC_BOOK3S_64 656DEFINE_INTERRUPT_HANDLER(do_bad_page_fault_segv) 657{ 658 bad_page_fault(regs, SIGSEGV); 659} 660 661/* 662 * In radix, segment interrupts indicate the EA is not addressable by the 663 * page table geometry, so they are always sent here. 664 * 665 * In hash, this is called if do_slb_fault returns error. Typically it is 666 * because the EA was outside the region allowed by software. 667 */ 668DEFINE_INTERRUPT_HANDLER(do_bad_segment_interrupt) 669{ 670 int err = regs->result; 671 672 if (err == -EFAULT) { 673 if (user_mode(regs)) 674 _exception(SIGSEGV, regs, SEGV_BNDERR, regs->dar); 675 else 676 bad_page_fault(regs, SIGSEGV); 677 } else if (err == -EINVAL) { 678 unrecoverable_exception(regs); 679 } else { 680 BUG(); 681 } 682} 683#endif 684