dumpstack.c

  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
 26static int die_counter;
 27
 28static struct pt_regs exec_summary_regs;
 29
 30bool noinstr in_task_stack(unsigned long *stack, struct task_struct *task,
 31			   struct stack_info *info)
 32{
 33	unsigned long *begin = task_stack_page(task);
 34	unsigned long *end   = task_stack_page(task) + THREAD_SIZE;
 35
 36	if (stack < begin || stack >= end)
 37		return false;
 38
 39	info->type	= STACK_TYPE_TASK;
 40	info->begin	= begin;
 41	info->end	= end;
 42	info->next_sp	= NULL;
 43
 44	return true;
 45}
 46
 47/* Called from get_stack_info_noinstr - so must be noinstr too */
 48bool noinstr in_entry_stack(unsigned long *stack, struct stack_info *info)
 49{
 50	struct entry_stack *ss = cpu_entry_stack(smp_processor_id());
 51
 52	void *begin = ss;
 53	void *end = ss + 1;
 54
 55	if ((void *)stack < begin || (void *)stack >= end)
 56		return false;
 57
 58	info->type	= STACK_TYPE_ENTRY;
 59	info->begin	= begin;
 60	info->end	= end;
 61	info->next_sp	= NULL;
 62
 63	return true;
 64}
 65
 66static void printk_stack_address(unsigned long address, int reliable,
 67				 const char *log_lvl)
 68{
 69	touch_nmi_watchdog();
 70	printk("%s %s%pBb\n", log_lvl, reliable ? "" : "? ", (void *)address);
 71}
 72
 73static int copy_code(struct pt_regs *regs, u8 *buf, unsigned long src,
 74		     unsigned int nbytes)
 75{
 76	if (!user_mode(regs))
 77		return copy_from_kernel_nofault(buf, (u8 *)src, nbytes);
 78
 79	/* The user space code from other tasks cannot be accessed. */
 80	if (regs != task_pt_regs(current))
 81		return -EPERM;
 82
 83	/*
 84	 * Even if named copy_from_user_nmi() this can be invoked from
 85	 * other contexts and will not try to resolve a pagefault, which is
 86	 * the correct thing to do here as this code can be called from any
 87	 * context.
 88	 */
 89	return copy_from_user_nmi(buf, (void __user *)src, nbytes);
 90}
 91
 92/*
 93 * There are a couple of reasons for the 2/3rd prologue, courtesy of Linus:
 94 *
 95 * In case where we don't have the exact kernel image (which, if we did, we can
 96 * simply disassemble and navigate to the RIP), the purpose of the bigger
 97 * prologue is to have more context and to be able to correlate the code from
 98 * the different toolchains better.
 99 *
100 * In addition, it helps in recreating the register allocation of the failing
101 * kernel and thus make sense of the register dump.
102 *
103 * What is more, the additional complication of a variable length insn arch like
104 * x86 warrants having longer byte sequence before rIP so that the disassembler
105 * can "sync" up properly and find instruction boundaries when decoding the
106 * opcode bytes.
107 *
108 * Thus, the 2/3rds prologue and 64 byte OPCODE_BUFSIZE is just a random
109 * guesstimate in attempt to achieve all of the above.
110 */
111void show_opcodes(struct pt_regs *regs, const char *loglvl)
112{
113#define PROLOGUE_SIZE 42
114#define EPILOGUE_SIZE 21
115#define OPCODE_BUFSIZE (PROLOGUE_SIZE + 1 + EPILOGUE_SIZE)
116	u8 opcodes[OPCODE_BUFSIZE];
117	unsigned long prologue = regs->ip - PROLOGUE_SIZE;
118
119	switch (copy_code(regs, opcodes, prologue, sizeof(opcodes))) {
120	case 0:
121		printk("%sCode: %" __stringify(PROLOGUE_SIZE) "ph <%02x> %"
122		       __stringify(EPILOGUE_SIZE) "ph\n", loglvl, opcodes,
123		       opcodes[PROLOGUE_SIZE], opcodes + PROLOGUE_SIZE + 1);
124		break;
125	case -EPERM:
126		/* No access to the user space stack of other tasks. Ignore. */
127		break;
128	default:
129		printk("%sCode: Unable to access opcode bytes at 0x%lx.\n",
130		       loglvl, prologue);
131		break;
132	}
133}
134
135void show_ip(struct pt_regs *regs, const char *loglvl)
136{
137#ifdef CONFIG_X86_32
138	printk("%sEIP: %pS\n", loglvl, (void *)regs->ip);
139#else
140	printk("%sRIP: %04x:%pS\n", loglvl, (int)regs->cs, (void *)regs->ip);
141#endif
142	show_opcodes(regs, loglvl);
143}
144
145void show_iret_regs(struct pt_regs *regs, const char *log_lvl)
146{
147	show_ip(regs, log_lvl);
148	printk("%sRSP: %04x:%016lx EFLAGS: %08lx", log_lvl, (int)regs->ss,
149		regs->sp, regs->flags);
150}
151
152static void show_regs_if_on_stack(struct stack_info *info, struct pt_regs *regs,
153				  bool partial, const char *log_lvl)
154{
155	/*
156	 * These on_stack() checks aren't strictly necessary: the unwind code
157	 * has already validated the 'regs' pointer.  The checks are done for
158	 * ordering reasons: if the registers are on the next stack, we don't
159	 * want to print them out yet.  Otherwise they'll be shown as part of
160	 * the wrong stack.  Later, when show_trace_log_lvl() switches to the
161	 * next stack, this function will be called again with the same regs so
162	 * they can be printed in the right context.
163	 */
164	if (!partial && on_stack(info, regs, sizeof(*regs))) {
165		__show_regs(regs, SHOW_REGS_SHORT, log_lvl);
166
167	} else if (partial && on_stack(info, (void *)regs + IRET_FRAME_OFFSET,
168				       IRET_FRAME_SIZE)) {
169		/*
170		 * When an interrupt or exception occurs in entry code, the
171		 * full pt_regs might not have been saved yet.  In that case
172		 * just print the iret frame.
173		 */
174		show_iret_regs(regs, log_lvl);
175	}
176}
177
178/*
179 * This function reads pointers from the stack and dereferences them. The
180 * pointers may not have their KMSAN shadow set up properly, which may result
181 * in false positive reports. Disable instrumentation to avoid those.
182 */
183__no_kmsan_checks
184static void __show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
185				 unsigned long *stack, const char *log_lvl)
186{
187	struct unwind_state state;
188	struct stack_info stack_info = {0};
189	unsigned long visit_mask = 0;
190	int graph_idx = 0;
191	bool partial = false;
192
193	printk("%sCall Trace:\n", log_lvl);
194
195	unwind_start(&state, task, regs, stack);
196	stack = stack ?: get_stack_pointer(task, regs);
197	regs = unwind_get_entry_regs(&state, &partial);
198
199	/*
200	 * Iterate through the stacks, starting with the current stack pointer.
201	 * Each stack has a pointer to the next one.
202	 *
203	 * x86-64 can have several stacks:
204	 * - task stack
205	 * - interrupt stack
206	 * - HW exception stacks (double fault, nmi, debug, mce)
207	 * - entry stack
208	 *
209	 * x86-32 can have up to four stacks:
210	 * - task stack
211	 * - softirq stack
212	 * - hardirq stack
213	 * - entry stack
214	 */
215	for (; stack; stack = stack_info.next_sp) {
216		const char *stack_name;
217
218		stack = PTR_ALIGN(stack, sizeof(long));
219
220		if (get_stack_info(stack, task, &stack_info, &visit_mask)) {
221			/*
222			 * We weren't on a valid stack.  It's possible that
223			 * we overflowed a valid stack into a guard page.
224			 * See if the next page up is valid so that we can
225			 * generate some kind of backtrace if this happens.
226			 */
227			stack = (unsigned long *)PAGE_ALIGN((unsigned long)stack);
228			if (get_stack_info(stack, task, &stack_info, &visit_mask))
229				break;
230		}
231
232		stack_name = stack_type_name(stack_info.type);
233		if (stack_name)
234			printk("%s <%s>\n", log_lvl, stack_name);
235
236		if (regs)
237			show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
238
239		/*
240		 * Scan the stack, printing any text addresses we find.  At the
241		 * same time, follow proper stack frames with the unwinder.
242		 *
243		 * Addresses found during the scan which are not reported by
244		 * the unwinder are considered to be additional clues which are
245		 * sometimes useful for debugging and are prefixed with '?'.
246		 * This also serves as a failsafe option in case the unwinder
247		 * goes off in the weeds.
248		 */
249		for (; stack < stack_info.end; stack++) {
250			unsigned long real_addr;
251			int reliable = 0;
252			unsigned long addr = READ_ONCE_NOCHECK(*stack);
253			unsigned long *ret_addr_p =
254				unwind_get_return_address_ptr(&state);
255
256			if (!__kernel_text_address(addr))
257				continue;
258
259			/*
260			 * Don't print regs->ip again if it was already printed
261			 * by show_regs_if_on_stack().
262			 */
263			if (regs && stack == &regs->ip)
264				goto next;
265
266			if (stack == ret_addr_p)
267				reliable = 1;
268
269			/*
270			 * When function graph tracing is enabled for a
271			 * function, its return address on the stack is
272			 * replaced with the address of an ftrace handler
273			 * (return_to_handler).  In that case, before printing
274			 * the "real" address, we want to print the handler
275			 * address as an "unreliable" hint that function graph
276			 * tracing was involved.
277			 */
278			real_addr = ftrace_graph_ret_addr(task, &graph_idx,
279							  addr, stack);
280			if (real_addr != addr)
281				printk_stack_address(addr, 0, log_lvl);
282			printk_stack_address(real_addr, reliable, log_lvl);
283
284			if (!reliable)
285				continue;
286
287next:
288			/*
289			 * Get the next frame from the unwinder.  No need to
290			 * check for an error: if anything goes wrong, the rest
291			 * of the addresses will just be printed as unreliable.
292			 */
293			unwind_next_frame(&state);
294
295			/* if the frame has entry regs, print them */
296			regs = unwind_get_entry_regs(&state, &partial);
297			if (regs)
298				show_regs_if_on_stack(&stack_info, regs, partial, log_lvl);
299		}
300
301		if (stack_name)
302			printk("%s </%s>\n", log_lvl, stack_name);
303	}
304}
305
306static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
307			       unsigned long *stack, const char *log_lvl)
308{
309	/*
310	 * Disable KASAN to avoid false positives during walking another
311	 * task's stacks, as values on these stacks may change concurrently
312	 * with task execution.
313	 */
314	bool disable_kasan = task && task != current;
315
316	if (disable_kasan)
317		kasan_disable_current();
318
319	__show_trace_log_lvl(task, regs, stack, log_lvl);
320
321	if (disable_kasan)
322		kasan_enable_current();
323}
324
325void show_stack(struct task_struct *task, unsigned long *sp,
326		       const char *loglvl)
327{
328	task = task ? : current;
329
330	/*
331	 * Stack frames below this one aren't interesting.  Don't show them
332	 * if we're printing for %current.
333	 */
334	if (!sp && task == current)
335		sp = get_stack_pointer(current, NULL);
336
337	show_trace_log_lvl(task, NULL, sp, loglvl);
338}
339
340void show_stack_regs(struct pt_regs *regs)
341{
342	show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
343}
344
345static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED;
346static int die_owner = -1;
347static unsigned int die_nest_count;
348
349unsigned long oops_begin(void)
350{
351	int cpu;
352	unsigned long flags;
353
354	oops_enter();
355
356	/* racy, but better than risking deadlock. */
357	raw_local_irq_save(flags);
358	cpu = smp_processor_id();
359	if (!arch_spin_trylock(&die_lock)) {
360		if (cpu == die_owner)
361			/* nested oops. should stop eventually */;
362		else
363			arch_spin_lock(&die_lock);
364	}
365	die_nest_count++;
366	die_owner = cpu;
367	console_verbose();
368	bust_spinlocks(1);
369	return flags;
370}
371NOKPROBE_SYMBOL(oops_begin);
372
373void __noreturn rewind_stack_and_make_dead(int signr);
374
375void oops_end(unsigned long flags, struct pt_regs *regs, int signr)
376{
377	if (regs && kexec_should_crash(current))
378		crash_kexec(regs);
379
380	bust_spinlocks(0);
381	die_owner = -1;
382	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
383	die_nest_count--;
384	if (!die_nest_count)
385		/* Nest count reaches zero, release the lock. */
386		arch_spin_unlock(&die_lock);
387	raw_local_irq_restore(flags);
388	oops_exit();
389
390	/* Executive summary in case the oops scrolled away */
391	__show_regs(&exec_summary_regs, SHOW_REGS_ALL, KERN_DEFAULT);
392
393	if (!signr)
394		return;
395	if (in_interrupt())
396		panic("Fatal exception in interrupt");
397	if (panic_on_oops)
398		panic("Fatal exception");
399
400	/*
401	 * We're not going to return, but we might be on an IST stack or
402	 * have very little stack space left.  Rewind the stack and kill
403	 * the task.
404	 * Before we rewind the stack, we have to tell KASAN that we're going to
405	 * reuse the task stack and that existing poisons are invalid.
406	 */
407	kasan_unpoison_task_stack(current);
408	rewind_stack_and_make_dead(signr);
409}
410NOKPROBE_SYMBOL(oops_end);
411
412static void __die_header(const char *str, struct pt_regs *regs, long err)
413{
414	/* Save the regs of the first oops for the executive summary later. */
415	if (!die_counter)
416		exec_summary_regs = *regs;
417
418	printk(KERN_DEFAULT
419	       "Oops: %s: %04lx [#%d]%s%s%s%s\n", str, err & 0xffff,
420	       ++die_counter,
421	       IS_ENABLED(CONFIG_SMP)     ? " SMP"             : "",
422	       debug_pagealloc_enabled()  ? " DEBUG_PAGEALLOC" : "",
423	       IS_ENABLED(CONFIG_KASAN)   ? " KASAN"           : "",
424	       IS_ENABLED(CONFIG_MITIGATION_PAGE_TABLE_ISOLATION) ?
425	       (boot_cpu_has(X86_FEATURE_PTI) ? " PTI" : " NOPTI") : "");
426}
427NOKPROBE_SYMBOL(__die_header);
428
429static int __die_body(const char *str, struct pt_regs *regs, long err)
430{
431	show_regs(regs);
432	print_modules();
433
434	if (notify_die(DIE_OOPS, str, regs, err,
435			current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
436		return 1;
437
438	return 0;
439}
440NOKPROBE_SYMBOL(__die_body);
441
442int __die(const char *str, struct pt_regs *regs, long err)
443{
444	__die_header(str, regs, err);
445	return __die_body(str, regs, err);
446}
447NOKPROBE_SYMBOL(__die);
448
449/*
450 * This is gone through when something in the kernel has done something bad
451 * and is about to be terminated:
452 */
453void die(const char *str, struct pt_regs *regs, long err)
454{
455	unsigned long flags = oops_begin();
456	int sig = SIGSEGV;
457
458	if (__die(str, regs, err))
459		sig = 0;
460	oops_end(flags, regs, sig);
461}
462
463void die_addr(const char *str, struct pt_regs *regs, long err, long gp_addr)
464{
465	unsigned long flags = oops_begin();
466	int sig = SIGSEGV;
467
468	__die_header(str, regs, err);
469	if (gp_addr)
470		kasan_non_canonical_hook(gp_addr);
471	if (__die_body(str, regs, err))
472		sig = 0;
473	oops_end(flags, regs, sig);
474}
475
476void show_regs(struct pt_regs *regs)
477{
478	enum show_regs_mode print_kernel_regs;
479
480	show_regs_print_info(KERN_DEFAULT);
481
482	print_kernel_regs = user_mode(regs) ? SHOW_REGS_USER : SHOW_REGS_ALL;
483	__show_regs(regs, print_kernel_regs, KERN_DEFAULT);
484
485	/*
486	 * When in-kernel, we also print out the stack at the time of the fault..
487	 */
488	if (!user_mode(regs))
489		show_trace_log_lvl(current, regs, NULL, KERN_DEFAULT);
490}