xref: /linux/arch/powerpc/mm/fault.c (revision 3dc12dfe74300febc568c3b530c0f9bee01f2821)
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 
36 #include <asm/firmware.h>
37 #include <asm/page.h>
38 #include <asm/mmu.h>
39 #include <asm/mmu_context.h>
40 #include <asm/siginfo.h>
41 #include <asm/debug.h>
42 #include <asm/kup.h>
43 #include <asm/inst.h>
44 
45 
46 /*
47  * do_page_fault error handling helpers
48  */
49 
50 static int
51 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long address, int si_code)
52 {
53 	/*
54 	 * If we are in kernel mode, bail out with a SEGV, this will
55 	 * be caught by the assembly which will restore the non-volatile
56 	 * registers before calling bad_page_fault()
57 	 */
58 	if (!user_mode(regs))
59 		return SIGSEGV;
60 
61 	_exception(SIGSEGV, regs, si_code, address);
62 
63 	return 0;
64 }
65 
66 static noinline int bad_area_nosemaphore(struct pt_regs *regs, unsigned long address)
67 {
68 	return __bad_area_nosemaphore(regs, address, SEGV_MAPERR);
69 }
70 
71 static int __bad_area(struct pt_regs *regs, unsigned long address, int si_code)
72 {
73 	struct mm_struct *mm = current->mm;
74 
75 	/*
76 	 * Something tried to access memory that isn't in our memory map..
77 	 * Fix it, but check if it's kernel or user first..
78 	 */
79 	mmap_read_unlock(mm);
80 
81 	return __bad_area_nosemaphore(regs, address, si_code);
82 }
83 
84 static noinline int bad_area(struct pt_regs *regs, unsigned long address)
85 {
86 	return __bad_area(regs, address, SEGV_MAPERR);
87 }
88 
89 #ifdef CONFIG_PPC_MEM_KEYS
90 static noinline int bad_access_pkey(struct pt_regs *regs, unsigned long address,
91 				    struct vm_area_struct *vma)
92 {
93 	struct mm_struct *mm = current->mm;
94 	int pkey;
95 
96 	/*
97 	 * We don't try to fetch the pkey from page table because reading
98 	 * page table without locking doesn't guarantee stable pte value.
99 	 * Hence the pkey value that we return to userspace can be different
100 	 * from the pkey that actually caused access error.
101 	 *
102 	 * It does *not* guarantee that the VMA we find here
103 	 * was the one that we faulted on.
104 	 *
105 	 * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
106 	 * 2. T1   : set AMR to deny access to pkey=4, touches, page
107 	 * 3. T1   : faults...
108 	 * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
109 	 * 5. T1   : enters fault handler, takes mmap_lock, etc...
110 	 * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
111 	 *	     faulted on a pte with its pkey=4.
112 	 */
113 	pkey = vma_pkey(vma);
114 
115 	mmap_read_unlock(mm);
116 
117 	/*
118 	 * If we are in kernel mode, bail out with a SEGV, this will
119 	 * be caught by the assembly which will restore the non-volatile
120 	 * registers before calling bad_page_fault()
121 	 */
122 	if (!user_mode(regs))
123 		return SIGSEGV;
124 
125 	_exception_pkey(regs, address, pkey);
126 
127 	return 0;
128 }
129 #endif
130 
131 static noinline int bad_access(struct pt_regs *regs, unsigned long address)
132 {
133 	return __bad_area(regs, address, SEGV_ACCERR);
134 }
135 
136 static int do_sigbus(struct pt_regs *regs, unsigned long address,
137 		     vm_fault_t fault)
138 {
139 	if (!user_mode(regs))
140 		return SIGBUS;
141 
142 	current->thread.trap_nr = BUS_ADRERR;
143 #ifdef CONFIG_MEMORY_FAILURE
144 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
145 		unsigned int lsb = 0; /* shutup gcc */
146 
147 		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
148 			current->comm, current->pid, address);
149 
150 		if (fault & VM_FAULT_HWPOISON_LARGE)
151 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
152 		if (fault & VM_FAULT_HWPOISON)
153 			lsb = PAGE_SHIFT;
154 
155 		force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
156 		return 0;
157 	}
158 
159 #endif
160 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
161 	return 0;
162 }
163 
164 static int mm_fault_error(struct pt_regs *regs, unsigned long addr,
165 				vm_fault_t fault)
166 {
167 	/*
168 	 * Kernel page fault interrupted by SIGKILL. We have no reason to
169 	 * continue processing.
170 	 */
171 	if (fatal_signal_pending(current) && !user_mode(regs))
172 		return SIGKILL;
173 
174 	/* Out of memory */
175 	if (fault & VM_FAULT_OOM) {
176 		/*
177 		 * We ran out of memory, or some other thing happened to us that
178 		 * made us unable to handle the page fault gracefully.
179 		 */
180 		if (!user_mode(regs))
181 			return SIGSEGV;
182 		pagefault_out_of_memory();
183 	} else {
184 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
185 			     VM_FAULT_HWPOISON_LARGE))
186 			return do_sigbus(regs, addr, fault);
187 		else if (fault & VM_FAULT_SIGSEGV)
188 			return bad_area_nosemaphore(regs, addr);
189 		else
190 			BUG();
191 	}
192 	return 0;
193 }
194 
195 /* Is this a bad kernel fault ? */
196 static bool bad_kernel_fault(struct pt_regs *regs, unsigned long error_code,
197 			     unsigned long address, bool is_write)
198 {
199 	int is_exec = TRAP(regs) == 0x400;
200 
201 	/* NX faults set DSISR_PROTFAULT on the 8xx, DSISR_NOEXEC_OR_G on others */
202 	if (is_exec && (error_code & (DSISR_NOEXEC_OR_G | DSISR_KEYFAULT |
203 				      DSISR_PROTFAULT))) {
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 	if (!is_exec && address < TASK_SIZE && (error_code & (DSISR_PROTFAULT | DSISR_KEYFAULT)) &&
214 	    !search_exception_tables(regs->nip)) {
215 		pr_crit_ratelimited("Kernel attempted to access user page (%lx) - exploit attempt? (uid: %d)\n",
216 				    address,
217 				    from_kuid(&init_user_ns, current_uid()));
218 	}
219 
220 	// Kernel fault on kernel address is bad
221 	if (address >= TASK_SIZE)
222 		return true;
223 
224 	// Fault on user outside of certain regions (eg. copy_tofrom_user()) is bad
225 	if (!search_exception_tables(regs->nip))
226 		return true;
227 
228 	// Read/write fault in a valid region (the exception table search passed
229 	// above), but blocked by KUAP is bad, it can never succeed.
230 	if (bad_kuap_fault(regs, address, is_write))
231 		return WARN(true, "Bug: %s fault blocked by KUAP!", is_write ? "Write" : "Read");
232 
233 	// What's left? Kernel fault on user in well defined regions (extable
234 	// matched), and allowed by KUAP in the faulting context.
235 	return false;
236 }
237 
238 #ifdef CONFIG_PPC_MEM_KEYS
239 static bool access_pkey_error(bool is_write, bool is_exec, bool is_pkey,
240 			      struct vm_area_struct *vma)
241 {
242 	/*
243 	 * Make sure to check the VMA so that we do not perform
244 	 * faults just to hit a pkey fault as soon as we fill in a
245 	 * page. Only called for current mm, hence foreign == 0
246 	 */
247 	if (!arch_vma_access_permitted(vma, is_write, is_exec, 0))
248 		return true;
249 
250 	return false;
251 }
252 #endif
253 
254 static bool access_error(bool is_write, bool is_exec, struct vm_area_struct *vma)
255 {
256 	/*
257 	 * Allow execution from readable areas if the MMU does not
258 	 * provide separate controls over reading and executing.
259 	 *
260 	 * Note: That code used to not be enabled for 4xx/BookE.
261 	 * It is now as I/D cache coherency for these is done at
262 	 * set_pte_at() time and I see no reason why the test
263 	 * below wouldn't be valid on those processors. This -may-
264 	 * break programs compiled with a really old ABI though.
265 	 */
266 	if (is_exec) {
267 		return !(vma->vm_flags & VM_EXEC) &&
268 			(cpu_has_feature(CPU_FTR_NOEXECUTE) ||
269 			 !(vma->vm_flags & (VM_READ | VM_WRITE)));
270 	}
271 
272 	if (is_write) {
273 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
274 			return true;
275 		return false;
276 	}
277 
278 	if (unlikely(!vma_is_accessible(vma)))
279 		return true;
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
290 static 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
303 static inline void cmo_account_page_fault(void) { }
304 #endif /* CONFIG_PPC_SMLPAR */
305 
306 static 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)
375 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_64S)
376 #else
377 #define page_fault_is_bad(__err)	((__err) & DSISR_BAD_FAULT_32S)
378 #endif
379 
380 /*
381  * For 600- and 800-family processors, the error_code parameter is DSISR
382  * for a data fault, SRR1 for an instruction fault. For 400-family processors
383  * the error_code parameter is ESR for a data fault, 0 for an instruction
384  * fault.
385  * For 64-bit processors, the error_code parameter is
386  *  - DSISR for a non-SLB data access fault,
387  *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
388  *  - 0 any SLB fault.
389  *
390  * The return value is 0 if the fault was handled, or the signal
391  * number if this is a kernel fault that can't be handled here.
392  */
393 static int __do_page_fault(struct pt_regs *regs, unsigned long address,
394 			   unsigned long error_code)
395 {
396 	struct vm_area_struct * vma;
397 	struct mm_struct *mm = current->mm;
398 	unsigned int flags = FAULT_FLAG_DEFAULT;
399  	int is_exec = TRAP(regs) == 0x400;
400 	int is_user = user_mode(regs);
401 	int is_write = page_fault_is_write(error_code);
402 	vm_fault_t fault, major = 0;
403 	bool kprobe_fault = kprobe_page_fault(regs, 11);
404 
405 	if (unlikely(debugger_fault_handler(regs) || kprobe_fault))
406 		return 0;
407 
408 	if (unlikely(page_fault_is_bad(error_code))) {
409 		if (is_user) {
410 			_exception(SIGBUS, regs, BUS_OBJERR, address);
411 			return 0;
412 		}
413 		return SIGBUS;
414 	}
415 
416 	/* Additional sanity check(s) */
417 	sanity_check_fault(is_write, is_user, error_code, address);
418 
419 	/*
420 	 * The kernel should never take an execute fault nor should it
421 	 * take a page fault to a kernel address or a page fault to a user
422 	 * address outside of dedicated places
423 	 */
424 	if (unlikely(!is_user && bad_kernel_fault(regs, error_code, address, is_write)))
425 		return SIGSEGV;
426 
427 	/*
428 	 * If we're in an interrupt, have no user context or are running
429 	 * in a region with pagefaults disabled then we must not take the fault
430 	 */
431 	if (unlikely(faulthandler_disabled() || !mm)) {
432 		if (is_user)
433 			printk_ratelimited(KERN_ERR "Page fault in user mode"
434 					   " with faulthandler_disabled()=%d"
435 					   " mm=%p\n",
436 					   faulthandler_disabled(), mm);
437 		return bad_area_nosemaphore(regs, address);
438 	}
439 
440 	/* We restore the interrupt state now */
441 	if (!arch_irq_disabled_regs(regs))
442 		local_irq_enable();
443 
444 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
445 
446 	/*
447 	 * We want to do this outside mmap_lock, because reading code around nip
448 	 * can result in fault, which will cause a deadlock when called with
449 	 * mmap_lock held
450 	 */
451 	if (is_user)
452 		flags |= FAULT_FLAG_USER;
453 	if (is_write)
454 		flags |= FAULT_FLAG_WRITE;
455 	if (is_exec)
456 		flags |= FAULT_FLAG_INSTRUCTION;
457 
458 	/* When running in the kernel we expect faults to occur only to
459 	 * addresses in user space.  All other faults represent errors in the
460 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
461 	 * erroneous fault occurring in a code path which already holds mmap_lock
462 	 * we will deadlock attempting to validate the fault against the
463 	 * address space.  Luckily the kernel only validly references user
464 	 * space from well defined areas of code, which are listed in the
465 	 * exceptions table.
466 	 *
467 	 * As the vast majority of faults will be valid we will only perform
468 	 * the source reference check when there is a possibility of a deadlock.
469 	 * Attempt to lock the address space, if we cannot we then validate the
470 	 * source.  If this is invalid we can skip the address space check,
471 	 * thus avoiding the deadlock.
472 	 */
473 	if (unlikely(!mmap_read_trylock(mm))) {
474 		if (!is_user && !search_exception_tables(regs->nip))
475 			return bad_area_nosemaphore(regs, address);
476 
477 retry:
478 		mmap_read_lock(mm);
479 	} else {
480 		/*
481 		 * The above down_read_trylock() might have succeeded in
482 		 * which case we'll have missed the might_sleep() from
483 		 * down_read():
484 		 */
485 		might_sleep();
486 	}
487 
488 	vma = find_vma(mm, address);
489 	if (unlikely(!vma))
490 		return bad_area(regs, address);
491 
492 	if (unlikely(vma->vm_start > address)) {
493 		if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
494 			return bad_area(regs, address);
495 
496 		if (unlikely(expand_stack(vma, address)))
497 			return bad_area(regs, address);
498 	}
499 
500 #ifdef CONFIG_PPC_MEM_KEYS
501 	if (unlikely(access_pkey_error(is_write, is_exec,
502 				       (error_code & DSISR_KEYFAULT), vma)))
503 		return bad_access_pkey(regs, address, vma);
504 #endif /* CONFIG_PPC_MEM_KEYS */
505 
506 	if (unlikely(access_error(is_write, is_exec, vma)))
507 		return bad_access(regs, address);
508 
509 	/*
510 	 * If for any reason at all we couldn't handle the fault,
511 	 * make sure we exit gracefully rather than endlessly redo
512 	 * the fault.
513 	 */
514 	fault = handle_mm_fault(vma, address, flags, regs);
515 
516 	major |= fault & VM_FAULT_MAJOR;
517 
518 	if (fault_signal_pending(fault, regs))
519 		return user_mode(regs) ? 0 : SIGBUS;
520 
521 	/*
522 	 * Handle the retry right now, the mmap_lock has been released in that
523 	 * case.
524 	 */
525 	if (unlikely(fault & VM_FAULT_RETRY)) {
526 		if (flags & FAULT_FLAG_ALLOW_RETRY) {
527 			flags |= FAULT_FLAG_TRIED;
528 			goto retry;
529 		}
530 	}
531 
532 	mmap_read_unlock(current->mm);
533 
534 	if (unlikely(fault & VM_FAULT_ERROR))
535 		return mm_fault_error(regs, address, fault);
536 
537 	/*
538 	 * Major/minor page fault accounting.
539 	 */
540 	if (major)
541 		cmo_account_page_fault();
542 
543 	return 0;
544 }
545 NOKPROBE_SYMBOL(__do_page_fault);
546 
547 int do_page_fault(struct pt_regs *regs, unsigned long address,
548 		  unsigned long error_code)
549 {
550 	enum ctx_state prev_state = exception_enter();
551 	int rc = __do_page_fault(regs, address, error_code);
552 	exception_exit(prev_state);
553 	return rc;
554 }
555 NOKPROBE_SYMBOL(do_page_fault);
556 
557 /*
558  * bad_page_fault is called when we have a bad access from the kernel.
559  * It is called from the DSI and ISI handlers in head.S and from some
560  * of the procedures in traps.c.
561  */
562 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
563 {
564 	const struct exception_table_entry *entry;
565 	int is_write = page_fault_is_write(regs->dsisr);
566 
567 	/* Are we prepared to handle this fault?  */
568 	if ((entry = search_exception_tables(regs->nip)) != NULL) {
569 		regs->nip = extable_fixup(entry);
570 		return;
571 	}
572 
573 	/* kernel has accessed a bad area */
574 
575 	switch (TRAP(regs)) {
576 	case 0x300:
577 	case 0x380:
578 	case 0xe00:
579 		pr_alert("BUG: %s on %s at 0x%08lx\n",
580 			 regs->dar < PAGE_SIZE ? "Kernel NULL pointer dereference" :
581 			 "Unable to handle kernel data access",
582 			 is_write ? "write" : "read", regs->dar);
583 		break;
584 	case 0x400:
585 	case 0x480:
586 		pr_alert("BUG: Unable to handle kernel instruction fetch%s",
587 			 regs->nip < PAGE_SIZE ? " (NULL pointer?)\n" : "\n");
588 		break;
589 	case 0x600:
590 		pr_alert("BUG: Unable to handle kernel unaligned access at 0x%08lx\n",
591 			 regs->dar);
592 		break;
593 	default:
594 		pr_alert("BUG: Unable to handle unknown paging fault at 0x%08lx\n",
595 			 regs->dar);
596 		break;
597 	}
598 	printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
599 		regs->nip);
600 
601 	if (task_stack_end_corrupted(current))
602 		printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
603 
604 	die("Kernel access of bad area", regs, sig);
605 }
606