1 /*
2  *  Copyright (C) 1991, 1992  Linus Torvalds
3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4  */
5 #include <linux/kallsyms.h>
6 #include <linux/kprobes.h>
7 #include <linux/uaccess.h>
8 #include <linux/utsname.h>
9 #include <linux/hardirq.h>
10 #include <linux/kdebug.h>
11 #include <linux/module.h>
12 #include <linux/ptrace.h>
13 #include <linux/sched/debug.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/ftrace.h>
16 #include <linux/kexec.h>
17 #include <linux/bug.h>
18 #include <linux/nmi.h>
19 #include <linux/sysfs.h>
20 #include <linux/kasan.h>
21 
22 #include <asm/cpu_entry_area.h>
23 #include <asm/stacktrace.h>
24 #include <asm/unwind.h>
25 
26 int panic_on_unrecovered_nmi;
27 int panic_on_io_nmi;
28 static int die_counter;
29 
30 static struct pt_regs exec_summary_regs;
31 
in_task_stack(unsigned long * stack,struct task_struct * task,struct stack_info * info)32 bool noinstr in_task_stack(unsigned long *stack, struct task_struct *task,
33 			   struct stack_info *info)
34 {
35 	unsigned long *begin = task_stack_page(task);
36 	unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
37 
38 	if (stack < begin || stack >= end)
39 		return false;
40 
41 	info->type	= STACK_TYPE_TASK;
42 	info->begin	= begin;
43 	info->end	= end;
44 	info->next_sp	= NULL;
45 
46 	return true;
47 }
48 
49 /* Called from get_stack_info_noinstr - so must be noinstr too */
in_entry_stack(unsigned long * stack,struct stack_info * info)50 bool noinstr in_entry_stack(unsigned long *stack, struct stack_info *info)
51 {
52 	struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
53 
54 	void *begin = ss;
55 	void *end = ss + 1;
56 
57 	if ((void *)stack < begin || (void *)stack >= end)
58 		return false;
59 
60 	info->type	= STACK_TYPE_ENTRY;
61 	info->begin	= begin;
62 	info->end	= end;
63 	info->next_sp	= NULL;
64 
65 	return true;
66 }
67 
printk_stack_address(unsigned long address,int reliable,const char * log_lvl)68 static void printk_stack_address(unsigned long address, int reliable,
69 				 const char *log_lvl)
70 {
71 	touch_nmi_watchdog();
72 	printk("%s %s%pBb\n", log_lvl, reliable ? "" : "? ", (void *)address);
73 }
74 
copy_code(struct pt_regs * regs,u8 * buf,unsigned long src,unsigned int nbytes)75 static int copy_code(struct pt_regs *regs, u8 *buf, unsigned long src,
76 		     unsigned int nbytes)
77 {
78 	if (!user_mode(regs))
79 		return copy_from_kernel_nofault(buf, (u8 *)src, nbytes);
80 
81 	/* The user space code from other tasks cannot be accessed. */
82 	if (regs != task_pt_regs(current))
83 		return -EPERM;
84 
85 	/*
86 	 * Even if named copy_from_user_nmi() this can be invoked from
87 	 * other contexts and will not try to resolve a pagefault, which is
88 	 * the correct thing to do here as this code can be called from any
89 	 * context.
90 	 */
91 	return copy_from_user_nmi(buf, (void __user *)src, nbytes);
92 }
93 
94 /*
95  * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
96  *
97  * In case where we don't have the exact kernel image (which, if we did, we can
98  * simply disassemble and navigate to the RIP), the purpose of the bigger
99  * prologue is to have more context and to be able to correlate the code from
100  * the different toolchains better.
101  *
102  * In addition, it helps in recreating the register allocation of the failing
103  * kernel and thus make sense of the register dump.
104  *
105  * What is more, the additional complication of a variable length insn arch like
106  * x86 warrants having longer byte sequence before rIP so that the disassembler
107  * can "sync" up properly and find instruction boundaries when decoding the
108  * opcode bytes.
109  *
110  * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
111  * guesstimate in attempt to achieve all of the above.
112  */
show_opcodes(struct pt_regs * regs,const char * loglvl)113 void show_opcodes(struct pt_regs *regs, const char *loglvl)
114 {
115 #define PROLOGUE_SIZE 42
116 #define EPILOGUE_SIZE 21
117 #define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
118 	u8 opcodes[OPCODE_BUFSIZE];
119 	unsigned long prologue = regs->ip - PROLOGUE_SIZE;
120 
121 	switch (copy_code(regs, opcodes, prologue, sizeof(opcodes))) {
122 	case 0:
123 		printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
124 		       __stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
125 		       opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
126 		break;
127 	case -EPERM:
128 		/* No access to the user space stack of other tasks. Ignore. */
129 		break;
130 	default:
131 		printk("%sCode: Unable to access opcode bytes at 0x%lx.\n",
132 		       loglvl, prologue);
133 		break;
134 	}
135 }
136 
show_ip(struct pt_regs * regs,const char * loglvl)137 void show_ip(struct pt_regs *regs, const char *loglvl)
138 {
139 #ifdef CONFIG_X86_32
140 	printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
141 #else
142 	printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
143 #endif
144 	show_opcodes(regs, loglvl);
145 }
146 
show_iret_regs(struct pt_regs * regs,const char * log_lvl)147 void show_iret_regs(struct pt_regs *regs, const char *log_lvl)
148 {
149 	show_ip(regs, log_lvl);
150 	printk("%sRSP: %04x:%016lx EFLAGS: %08lx", log_lvl, (int)regs->ss,
151 		regs->sp, regs->flags);
152 }
153 
show_regs_if_on_stack(struct stack_info * info,struct pt_regs * regs,bool partial,const char * log_lvl)154 static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
155 				  bool partial, const char *log_lvl)
156 {
157 	/*
158 	 * These on_stack() checks aren't strictly necessary: the unwind code
159 	 * has already validated the 'regs' pointer.  The checks are done for
160 	 * ordering reasons: if the registers are on the next stack, we don't
161 	 * want to print them out yet.  Otherwise they'll be shown as part of
162 	 * the wrong stack.  Later, when show_trace_log_lvl() switches to the
163 	 * next stack, this function will be called again with the same regs so
164 	 * they can be printed in the right context.
165 	 */
166 	if (!partial && on_stack(info, regs, sizeof(*regs))) {
167 		__show_regs(regs, SHOW_REGS_SHORT, log_lvl);
168 
169 	} else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
170 				       IRET_FRAME_SIZE)) {
171 		/*
172 		 * When an interrupt or exception occurs in entry code, the
173 		 * full pt_regs might not have been saved yet.  In that case
174 		 * just print the iret frame.
175 		 */
176 		show_iret_regs(regs, log_lvl);
177 	}
178 }
179 
180 /*
181  * This function reads pointers from the stack and dereferences them. The
182  * pointers may not have their KMSAN shadow set up properly, which may result
183  * in false positive reports. Disable instrumentation to avoid those.
184  */
185 __no_kmsan_checks
show_trace_log_lvl(struct task_struct * task,struct pt_regs * regs,unsigned long * stack,const char * log_lvl)186 static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
187 			unsigned long *stack, const char *log_lvl)
188 {
189 	struct unwind_state state;
190 	struct stack_info stack_info = {0};
191 	unsigned long visit_mask = 0;
192 	int graph_idx = 0;
193 	bool partial = false;
194 
195 	printk("%sCall Trace:\n", log_lvl);
196 
197 	unwind_start(&state, task, regs, stack);
198 	regs = unwind_get_entry_regs(&state, &partial);
199 
200 	/*
201 	 * Iterate through the stacks, starting with the current stack pointer.
202 	 * Each stack has a pointer to the next one.
203 	 *
204 	 * x86-64 can have several stacks:
205 	 * - task stack
206 	 * - interrupt stack
207 	 * - HW exception stacks (double fault, nmi, debug, mce)
208 	 * - entry stack
209 	 *
210 	 * x86-32 can have up to four stacks:
211 	 * - task stack
212 	 * - softirq stack
213 	 * - hardirq stack
214 	 * - entry stack
215 	 */
216 	for (stack = stack ?: get_stack_pointer(task, regs);
217 	     stack;
218 	     stack = stack_info.next_sp) {
219 		const char *stack_name;
220 
221 		stack = PTR_ALIGN(stack, sizeof(long));
222 
223 		if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
224 			/*
225 			 * We weren't on a valid stack.  It's possible that
226 			 * we overflowed a valid stack into a guard page.
227 			 * See if the next page up is valid so that we can
228 			 * generate some kind of backtrace if this happens.
229 			 */
230 			stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
231 			if (get_stack_info(stack, task, &stack_info, &visit_mask))
232 				break;
233 		}
234 
235 		stack_name = stack_type_name(stack_info.type);
236 		if (stack_name)
237 			printk("%s <%s>\n", log_lvl, stack_name);
238 
239 		if (regs)
240 			show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
241 
242 		/*
243 		 * Scan the stack, printing any text addresses we find.  At the
244 		 * same time, follow proper stack frames with the unwinder.
245 		 *
246 		 * Addresses found during the scan which are not reported by
247 		 * the unwinder are considered to be additional clues which are
248 		 * sometimes useful for debugging and are prefixed with '?'.
249 		 * This also serves as a failsafe option in case the unwinder
250 		 * goes off in the weeds.
251 		 */
252 		for (; stack < stack_info.end; stack++) {
253 			unsigned long real_addr;
254 			int reliable = 0;
255 			unsigned long addr = READ_ONCE_NOCHECK(*stack);
256 			unsigned long *ret_addr_p =
257 				unwind_get_return_address_ptr(&state);
258 
259 			if (!__kernel_text_address(addr))
260 				continue;
261 
262 			/*
263 			 * Don't print regs->ip again if it was already printed
264 			 * by show_regs_if_on_stack().
265 			 */
266 			if (regs && stack == &regs->ip)
267 				goto next;
268 
269 			if (stack == ret_addr_p)
270 				reliable = 1;
271 
272 			/*
273 			 * When function graph tracing is enabled for a
274 			 * function, its return address on the stack is
275 			 * replaced with the address of an ftrace handler
276 			 * (return_to_handler).  In that case, before printing
277 			 * the "real" address, we want to print the handler
278 			 * address as an "unreliable" hint that function graph
279 			 * tracing was involved.
280 			 */
281 			real_addr = ftrace_graph_ret_addr(task, &graph_idx,
282 							  addr, stack);
283 			if (real_addr != addr)
284 				printk_stack_address(addr, 0, log_lvl);
285 			printk_stack_address(real_addr, reliable, log_lvl);
286 
287 			if (!reliable)
288 				continue;
289 
290 next:
291 			/*
292 			 * Get the next frame from the unwinder.  No need to
293 			 * check for an error: if anything goes wrong, the rest
294 			 * of the addresses will just be printed as unreliable.
295 			 */
296 			unwind_next_frame(&state);
297 
298 			/* if the frame has entry regs, print them */
299 			regs = unwind_get_entry_regs(&state, &partial);
300 			if (regs)
301 				show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
302 		}
303 
304 		if (stack_name)
305 			printk("%s </%s>\n", log_lvl, stack_name);
306 	}
307 }
308 
show_stack(struct task_struct * task,unsigned long * sp,const char * loglvl)309 void show_stack(struct task_struct *task, unsigned long *sp,
310 		       const char *loglvl)
311 {
312 	task = task ? : current;
313 
314 	/*
315 	 * Stack frames below this one aren't interesting.  Don't show them
316 	 * if we're printing for %current.
317 	 */
318 	if (!sp && task == current)
319 		sp = get_stack_pointer(current, NULL);
320 
321 	show_trace_log_lvl(task, NULL, sp, loglvl);
322 }
323 
show_stack_regs(struct pt_regs * regs)324 void show_stack_regs(struct pt_regs *regs)
325 {
326 	show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
327 }
328 
329 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
330 static int die_owner = -1;
331 static unsigned int die_nest_count;
332 
oops_begin(void)333 unsigned long oops_begin(void)
334 {
335 	int cpu;
336 	unsigned long flags;
337 
338 	oops_enter();
339 
340 	/* racy, but better than risking deadlock. */
341 	raw_local_irq_save(flags);
342 	cpu = smp_processor_id();
343 	if (!arch_spin_trylock(&die_lock)) {
344 		if (cpu == die_owner)
345 			/* nested oops. should stop eventually */;
346 		else
347 			arch_spin_lock(&die_lock);
348 	}
349 	die_nest_count++;
350 	die_owner = cpu;
351 	console_verbose();
352 	bust_spinlocks(1);
353 	return flags;
354 }
355 NOKPROBE_SYMBOL(oops_begin);
356 
357 void __noreturn rewind_stack_and_make_dead(int signr);
358 
oops_end(unsigned long flags,struct pt_regs * regs,int signr)359 void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
360 {
361 	if (regs && kexec_should_crash(current))
362 		crash_kexec(regs);
363 
364 	bust_spinlocks(0);
365 	die_owner = -1;
366 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
367 	die_nest_count--;
368 	if (!die_nest_count)
369 		/* Nest count reaches zero, release the lock. */
370 		arch_spin_unlock(&die_lock);
371 	raw_local_irq_restore(flags);
372 	oops_exit();
373 
374 	/* Executive summary in case the oops scrolled away */
375 	__show_regs(&exec_summary_regs, SHOW_REGS_ALL, KERN_DEFAULT);
376 
377 	if (!signr)
378 		return;
379 	if (in_interrupt())
380 		panic("Fatal exception in interrupt");
381 	if (panic_on_oops)
382 		panic("Fatal exception");
383 
384 	/*
385 	 * We're not going to return, but we might be on an IST stack or
386 	 * have very little stack space left.  Rewind the stack and kill
387 	 * the task.
388 	 * Before we rewind the stack, we have to tell KASAN that we're going to
389 	 * reuse the task stack and that existing poisons are invalid.
390 	 */
391 	kasan_unpoison_task_stack(current);
392 	rewind_stack_and_make_dead(signr);
393 }
394 NOKPROBE_SYMBOL(oops_end);
395 
__die_header(const char * str,struct pt_regs * regs,long err)396 static void __die_header(const char *str, struct pt_regs *regs, long err)
397 {
398 	const char *pr = "";
399 
400 	/* Save the regs of the first oops for the executive summary later. */
401 	if (!die_counter)
402 		exec_summary_regs = *regs;
403 
404 	if (IS_ENABLED(CONFIG_PREEMPTION))
405 		pr = IS_ENABLED(CONFIG_PREEMPT_RT) ? " PREEMPT_RT" : " PREEMPT";
406 
407 	printk(KERN_DEFAULT
408 	       "Oops: %s: %04lx [#%d]%s%s%s%s%s\n", str, err & 0xffff,
409 	       ++die_counter, pr,
410 	       IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
411 	       debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
412 	       IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "",
413 	       IS_ENABLED(CONFIG_MITIGATION_PAGE_TABLE_ISOLATION) ?
414 	       (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
415 }
416 NOKPROBE_SYMBOL(__die_header);
417 
__die_body(const char * str,struct pt_regs * regs,long err)418 static int __die_body(const char *str, struct pt_regs *regs, long err)
419 {
420 	show_regs(regs);
421 	print_modules();
422 
423 	if (notify_die(DIE_OOPS, str, regs, err,
424 			current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
425 		return 1;
426 
427 	return 0;
428 }
429 NOKPROBE_SYMBOL(__die_body);
430 
__die(const char * str,struct pt_regs * regs,long err)431 int __die(const char *str, struct pt_regs *regs, long err)
432 {
433 	__die_header(str, regs, err);
434 	return __die_body(str, regs, err);
435 }
436 NOKPROBE_SYMBOL(__die);
437 
438 /*
439  * This is gone through when something in the kernel has done something bad
440  * and is about to be terminated:
441  */
die(const char * str,struct pt_regs * regs,long err)442 void die(const char *str, struct pt_regs *regs, long err)
443 {
444 	unsigned long flags = oops_begin();
445 	int sig = SIGSEGV;
446 
447 	if (__die(str, regs, err))
448 		sig = 0;
449 	oops_end(flags, regs, sig);
450 }
451 
die_addr(const char * str,struct pt_regs * regs,long err,long gp_addr)452 void die_addr(const char *str, struct pt_regs *regs, long err, long gp_addr)
453 {
454 	unsigned long flags = oops_begin();
455 	int sig = SIGSEGV;
456 
457 	__die_header(str, regs, err);
458 	if (gp_addr)
459 		kasan_non_canonical_hook(gp_addr);
460 	if (__die_body(str, regs, err))
461 		sig = 0;
462 	oops_end(flags, regs, sig);
463 }
464 
show_regs(struct pt_regs * regs)465 void show_regs(struct pt_regs *regs)
466 {
467 	enum show_regs_mode print_kernel_regs;
468 
469 	show_regs_print_info(KERN_DEFAULT);
470 
471 	print_kernel_regs = user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL;
472 	__show_regs(regs, print_kernel_regs, KERN_DEFAULT);
473 
474 	/*
475 	 * When in-kernel, we also print out the stack at the time of the fault..
476 	 */
477 	if (!user_mode(regs))
478 		show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
479 }
480