4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * libcfs/libcfs/debug.c
34 * Author: Phil Schwan <phil@clusterfs.com>
38 # define DEBUG_SUBSYSTEM S_LNET
40 #include <linux/module.h>
41 #include <linux/ctype.h>
42 #include <libcfs/libcfs_string.h>
43 #include <linux/kthread.h>
44 #include <linux/stacktrace.h>
45 #include <linux/utsname.h>
46 #include <linux/kallsyms.h>
47 #include "tracefile.h"
49 static char debug_file_name[1024];
51 unsigned int libcfs_subsystem_debug = ~0;
52 EXPORT_SYMBOL(libcfs_subsystem_debug);
53 module_param(libcfs_subsystem_debug, int, 0644);
54 MODULE_PARM_DESC(libcfs_subsystem_debug, "Lustre kernel debug subsystem mask");
56 unsigned int libcfs_debug = (D_CANTMASK | D_NETERROR | D_HA | D_CONFIG |
57 D_IOCTL | D_LFSCK | D_TTY);
58 EXPORT_SYMBOL(libcfs_debug);
59 module_param(libcfs_debug, int, 0644);
60 MODULE_PARM_DESC(libcfs_debug, "Lustre kernel debug mask");
62 static int libcfs_param_debug_mb_set(const char *val,
63 cfs_kernel_param_arg_t *kp)
68 rc = kstrtouint(val, 0, &num);
72 num = cfs_trace_set_debug_mb(num);
74 *((unsigned int *)kp->arg) = num;
75 num = cfs_trace_get_debug_mb();
77 /* This value is more precise */
78 *((unsigned int *)kp->arg) = num;
83 /* While debug_mb setting look like unsigned int, in fact
84 * it needs quite a bunch of extra processing, so we define special
85 * debug_mb parameter type with corresponding methods to handle this case
87 static const struct kernel_param_ops param_ops_debug_mb = {
88 .set = libcfs_param_debug_mb_set,
89 .get = param_get_uint,
92 #define param_check_debug_mb(name, p) \
93 __param_check(name, p, unsigned int)
95 static unsigned int libcfs_debug_mb;
96 #ifdef HAVE_KERNEL_PARAM_OPS
97 module_param(libcfs_debug_mb, debug_mb, 0644);
99 module_param_call(libcfs_debug_mb, libcfs_param_debug_mb_set, param_get_uint,
100 ¶m_ops_debug_mb, 0644);
102 MODULE_PARM_DESC(libcfs_debug_mb, "Total debug buffer size.");
104 unsigned int libcfs_printk = D_CANTMASK;
105 module_param(libcfs_printk, uint, 0644);
106 MODULE_PARM_DESC(libcfs_printk, "Lustre kernel debug console mask");
108 unsigned int libcfs_console_ratelimit = 1;
109 module_param(libcfs_console_ratelimit, uint, 0644);
110 MODULE_PARM_DESC(libcfs_console_ratelimit, "Lustre kernel debug console ratelimit (0 to disable)");
112 static int param_set_delay_minmax(const char *val,
113 cfs_kernel_param_arg_t *kp,
120 rc = kstrtoint(val, 0, &sec);
124 /* The sysfs setting is in centiseconds */
125 d = cfs_time_seconds(sec) / 100;
126 if (d < min || d > max)
129 *((unsigned int *)kp->arg) = d;
134 static int param_get_delay(char *buffer, cfs_kernel_param_arg_t *kp)
136 unsigned int d = *(unsigned int *)kp->arg;
138 param_get_byte(buffer, kp);
139 return sprintf(buffer, "%lu%c", jiffies_to_msecs(d * 10) / MSEC_PER_SEC,
140 strnchr(buffer, PAGE_SIZE, '\n') ? '\n' : '\0');
143 unsigned int libcfs_console_max_delay;
144 unsigned int libcfs_console_min_delay;
146 static int param_set_console_max_delay(const char *val,
147 cfs_kernel_param_arg_t *kp)
149 return param_set_delay_minmax(val, kp,
150 libcfs_console_min_delay, INT_MAX);
153 static const struct kernel_param_ops param_ops_console_max_delay = {
154 .set = param_set_console_max_delay,
155 .get = param_get_delay,
158 #define param_check_console_max_delay(name, p) \
159 __param_check(name, p, unsigned int)
161 #ifdef HAVE_KERNEL_PARAM_OPS
162 module_param(libcfs_console_max_delay, console_max_delay, 0644);
164 module_param_call(libcfs_console_max_delay, param_set_console_max_delay,
165 param_get_delay, ¶m_ops_console_max_delay, 0644);
167 MODULE_PARM_DESC(libcfs_console_max_delay, "Lustre kernel debug console max delay (jiffies)");
169 static int param_set_console_min_delay(const char *val,
170 cfs_kernel_param_arg_t *kp)
172 return param_set_delay_minmax(val, kp,
173 1, libcfs_console_max_delay);
176 static const struct kernel_param_ops param_ops_console_min_delay = {
177 .set = param_set_console_min_delay,
178 .get = param_get_delay,
181 #define param_check_console_min_delay(name, p) \
182 __param_check(name, p, unsigned int)
184 #ifdef HAVE_KERNEL_PARAM_OPS
185 module_param(libcfs_console_min_delay, console_min_delay, 0644);
187 module_param_call(libcfs_console_min_delay, param_set_console_min_delay,
188 param_get_delay, ¶m_ops_console_min_delay, 0644);
190 MODULE_PARM_DESC(libcfs_console_min_delay, "Lustre kernel debug console min delay (jiffies)");
192 static int param_set_uint_minmax(const char *val,
193 cfs_kernel_param_arg_t *kp,
194 unsigned int min, unsigned int max)
202 ret = kstrtouint(val, 0, &num);
203 if (ret < 0 || num < min || num > max)
206 *((unsigned int *)kp->arg) = num;
210 static int param_set_uintpos(const char *val,
211 cfs_kernel_param_arg_t *kp)
213 return param_set_uint_minmax(val, kp, 1, -1);
216 static const struct kernel_param_ops param_ops_uintpos = {
217 .set = param_set_uintpos,
218 .get = param_get_uint,
221 #define param_check_uintpos(name, p) \
222 __param_check(name, p, unsigned int)
224 unsigned int libcfs_console_backoff = CDEBUG_DEFAULT_BACKOFF;
225 #ifdef HAVE_KERNEL_PARAM_OPS
226 module_param(libcfs_console_backoff, uintpos, 0644);
228 module_param_call(libcfs_console_backoff, param_set_uintpos, param_get_uint,
229 ¶m_ops_uintpos, 0644);
231 MODULE_PARM_DESC(libcfs_console_backoff, "Lustre kernel debug console backoff factor");
233 unsigned int libcfs_debug_binary = 1;
235 unsigned int libcfs_stack = 3 * THREAD_SIZE / 4;
236 EXPORT_SYMBOL(libcfs_stack);
238 unsigned int libcfs_catastrophe;
239 EXPORT_SYMBOL(libcfs_catastrophe);
241 unsigned int libcfs_watchdog_ratelimit = 300;
242 EXPORT_SYMBOL(libcfs_watchdog_ratelimit);
244 unsigned int libcfs_panic_on_lbug = 1;
245 module_param(libcfs_panic_on_lbug, uint, 0644);
246 MODULE_PARM_DESC(libcfs_panic_on_lbug, "Lustre kernel panic on LBUG");
248 atomic64_t libcfs_kmem = ATOMIC64_INIT(0);
249 EXPORT_SYMBOL(libcfs_kmem);
251 static DECLARE_COMPLETION(debug_complete);
253 char libcfs_debug_file_path_arr[PATH_MAX] = LIBCFS_DEBUG_FILE_PATH_DEFAULT;
254 EXPORT_SYMBOL(libcfs_debug_file_path_arr);
256 /* We need to pass a pointer here, but elsewhere this must be a const */
257 static char *libcfs_debug_file_path = LIBCFS_DEBUG_FILE_PATH_DEFAULT;
258 module_param(libcfs_debug_file_path, charp, 0644);
259 MODULE_PARM_DESC(libcfs_debug_file_path,
260 "Path for dumping debug logs, set 'NONE' to prevent log dumping");
262 int libcfs_panic_in_progress;
264 /* libcfs_debug_token2mask() expects the returned string in lower-case */
265 static const char *libcfs_debug_subsys2str(int subsys)
267 static const char * const libcfs_debug_subsystems[] =
268 LIBCFS_DEBUG_SUBSYS_NAMES;
270 if (subsys >= ARRAY_SIZE(libcfs_debug_subsystems))
273 return libcfs_debug_subsystems[subsys];
276 /* libcfs_debug_token2mask() expects the returned string in lower-case */
277 static const char *libcfs_debug_dbg2str(int debug)
279 static const char * const libcfs_debug_masks[] =
280 LIBCFS_DEBUG_MASKS_NAMES;
282 if (debug >= ARRAY_SIZE(libcfs_debug_masks))
285 return libcfs_debug_masks[debug];
289 libcfs_debug_mask2str(char *str, int size, int mask, int is_subsys)
291 const char *(*fn)(int bit) = is_subsys ? libcfs_debug_subsys2str :
292 libcfs_debug_dbg2str;
297 if (mask == 0) { /* "0" */
301 } else { /* space-separated tokens */
302 for (i = 0; i < 32; i++) {
303 if ((mask & BIT(i)) == 0)
307 if (!token) /* unused bit */
310 if (len > 0) { /* separator? */
316 while (*token != 0) {
325 /* terminate 'str' */
335 libcfs_debug_str2mask(int *mask, const char *str, int is_subsys)
337 const char *(*fn)(int bit) = is_subsys ? libcfs_debug_subsys2str :
338 libcfs_debug_dbg2str;
344 /* Allow a number for backwards compatibility */
345 for (n = strlen(str); n > 0; n--)
346 if (!isspace(str[n-1]))
349 t = sscanf(str, "%i%n", &m, &matched);
350 if (t >= 1 && matched == n) {
351 /* don't print warning for lctl set_param debug=0 or -1 */
352 if (m != 0 && m != -1)
353 CWARN("You are trying to use a numerical value for the mask - this will be deprecated in a future release.\n");
358 return cfs_str2mask(str, fn, mask, is_subsys ? 0 : D_CANTMASK,
362 char lnet_debug_log_upcall[1024] = "/usr/lib/lustre/lnet_debug_log_upcall";
364 /* Upcall function once a Lustre log has been dumped.
366 * @file path of the dumped log
368 static void libcfs_run_debug_log_upcall(char *file)
372 static const char * const envp[] = {
374 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
379 argv[0] = lnet_debug_log_upcall;
381 LASSERTF(file, "called on a null filename\n");
382 argv[1] = file; /* only need to pass the path of the file */
386 rc = call_usermodehelper(argv[0], argv, (char **)envp, 1);
387 if (rc < 0 && rc != -ENOENT) {
388 CERROR("Error %d invoking LNET debug log upcall %s %s; check /sys/kernel/debug/lnet/debug_log_upcall\n",
389 rc, argv[0], argv[1]);
391 CDEBUG(D_HA, "Invoked LNET debug log upcall %s %s\n",
397 * Dump Lustre log to ::debug_file_path by calling tracefile_dump_all_pages()
399 static void libcfs_debug_dumplog_internal(void *arg)
401 static time64_t last_dump_time;
402 time64_t current_time;
404 current_time = ktime_get_real_seconds();
406 if (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 &&
407 current_time > last_dump_time) {
408 last_dump_time = current_time;
409 snprintf(debug_file_name, sizeof(debug_file_name) - 1,
410 "%s.%lld.%ld", libcfs_debug_file_path_arr,
411 (s64)current_time, (uintptr_t)arg);
412 pr_alert("LustreError: dumping log to %s\n", debug_file_name);
413 cfs_tracefile_dump_all_pages(debug_file_name);
414 libcfs_run_debug_log_upcall(debug_file_name);
418 static int libcfs_debug_dumplog_thread(void *arg)
420 libcfs_debug_dumplog_internal(arg);
421 complete(&debug_complete);
425 static DEFINE_MUTEX(libcfs_debug_dumplog_lock);
427 void libcfs_debug_dumplog(void)
429 struct task_struct *dumper;
433 if (mutex_trylock(&libcfs_debug_dumplog_lock) == 0)
436 /* If a previous call was interrupted, debug_complete->done
437 * might be elevated, and so we won't actually wait here.
438 * So we reinit the completion to ensure we wait for
439 * one thread to complete, though it might not be the one
440 * we start if there are overlaping thread.
442 reinit_completion(&debug_complete);
443 dumper = kthread_run(libcfs_debug_dumplog_thread,
444 (void *)(long)current->pid,
445 "libcfs_debug_dumper");
447 pr_err("LustreError: cannot start log dump thread: rc = %ld\n",
450 wait_for_completion_interruptible(&debug_complete);
452 mutex_unlock(&libcfs_debug_dumplog_lock);
454 EXPORT_SYMBOL(libcfs_debug_dumplog);
456 /* coverity[+kill] */
457 void __noreturn lbug_with_loc(struct libcfs_debug_msg_data *msgdata)
459 libcfs_catastrophe = 1;
460 libcfs_debug_msg(msgdata, "LBUG\n");
462 if (in_interrupt()) {
463 panic("LBUG in interrupt.\n");
467 libcfs_debug_dumpstack(NULL);
468 if (libcfs_panic_on_lbug)
471 libcfs_debug_dumplog();
472 set_current_state(TASK_UNINTERRUPTIBLE);
476 EXPORT_SYMBOL(lbug_with_loc);
478 #ifdef CONFIG_STACKTRACE
480 #ifndef HAVE_SAVE_STACK_TRACE_TSK
481 #define save_stack_trace_tsk(tsk, trace) \
483 if (tsk == current) \
484 save_stack_trace(trace); \
486 pr_info("No stack, save_stack_trace_tsk() not exported\n"); \
490 static void cfs_print_stack_trace(unsigned long *entries, unsigned int nr)
494 /* Prefer %pB for backtraced symbolic names since it was added in:
495 * Linux v2.6.38-6557-g0f77a8d37825
496 * vsprintf: Introduce %pB format specifier
498 for (i = 0; i < nr; i++)
499 pr_info("[<0>] %pB\n", (void *)entries[i]);
502 #define MAX_ST_ENTRIES 100
503 static DEFINE_SPINLOCK(st_lock);
505 /* Linux v5.1-rc5 214d8ca6ee ("stacktrace: Provide common infrastructure")
506 * CONFIG_ARCH_STACKWALK indicates that save_stack_trace_tsk symbol is not
507 * exported. Use symbol_get() to find if save_stack_trace_tsk is available.
509 #ifdef CONFIG_ARCH_STACKWALK
510 typedef unsigned int (stack_trace_save_tsk_t)(struct task_struct *task,
511 unsigned long *store,
513 unsigned int skipnr);
514 static stack_trace_save_tsk_t *task_dump_stack;
517 void __init cfs_debug_init(void)
519 #ifdef CONFIG_ARCH_STACKWALK
520 task_dump_stack = (void *)
521 kallsyms_lookup_name("stack_trace_save_tsk");
526 static void libcfs_call_trace(struct task_struct *tsk)
528 static unsigned long entries[MAX_ST_ENTRIES];
529 #ifdef CONFIG_ARCH_STACKWALK
530 unsigned int nr_entries;
533 pr_info("Pid: %d, comm: %.20s %s %s\n", tsk->pid, tsk->comm,
534 init_utsname()->release, init_utsname()->version);
535 pr_info("Call Trace TBD:\n");
536 if (task_dump_stack) {
537 nr_entries = task_dump_stack(tsk, entries, MAX_ST_ENTRIES, 0);
538 cfs_print_stack_trace(entries, nr_entries);
540 spin_unlock(&st_lock);
542 struct stack_trace trace;
544 trace.nr_entries = 0;
545 trace.max_entries = MAX_ST_ENTRIES;
546 trace.entries = entries;
550 pr_info("Pid: %d, comm: %.20s %s %s\n", tsk->pid, tsk->comm,
551 init_utsname()->release, init_utsname()->version);
552 pr_info("Call Trace:\n");
553 save_stack_trace_tsk(tsk, &trace);
554 cfs_print_stack_trace(trace.entries, trace.nr_entries);
555 spin_unlock(&st_lock);
559 #else /* !CONFIG_STACKTRACE */
562 #include <linux/nmi.h>
563 #include <asm/stacktrace.h>
565 #ifdef HAVE_STACKTRACE_OPS
566 static int print_trace_stack(void *data, char *name)
568 printk(" <%s> ", name);
572 #ifdef STACKTRACE_OPS_ADDRESS_RETURN_INT
577 print_trace_address(void *data, unsigned long addr, int reliable)
581 touch_nmi_watchdog();
582 sprintf(fmt, " [<%016lx>] %s%%s\n", addr, reliable ? "" : "? ");
583 __print_symbol(fmt, addr);
584 #ifdef STACKTRACE_OPS_ADDRESS_RETURN_INT
589 static const struct stacktrace_ops print_trace_ops = {
590 .stack = print_trace_stack,
591 .address = print_trace_address,
592 .walk_stack = print_context_stack,
594 #endif /* HAVE_STACKTRACE_OPS */
596 static void libcfs_call_trace(struct task_struct *tsk)
598 #ifdef HAVE_STACKTRACE_OPS
599 printk("Pid: %d, comm: %.20s\n", tsk->pid, tsk->comm);
600 printk("\nCall Trace:\n");
601 dump_trace(tsk, NULL, NULL, 0, &print_trace_ops, NULL);
603 #else /* !HAVE_STACKTRACE_OPS */
607 CWARN("can't show stack: kernel doesn't export show_task\n");
608 #endif /* HAVE_STACKTRACE_OPS */
611 #else /* !CONFIG_X86 */
613 static void libcfs_call_trace(struct task_struct *tsk)
618 CWARN("can't show stack: kernel doesn't export show_task\n");
621 #endif /* CONFIG_X86 */
623 #endif /* CONFIG_STACKTRACE */
625 void libcfs_debug_dumpstack(struct task_struct *tsk)
627 libcfs_call_trace(tsk ?: current);
629 EXPORT_SYMBOL(libcfs_debug_dumpstack);
631 static int panic_notifier(struct notifier_block *self, unsigned long unused1,
634 if (libcfs_panic_in_progress)
637 libcfs_panic_in_progress = 1;
640 #ifdef LNET_DUMP_ON_PANIC
641 /* This is currently disabled because it spews far too much to the
642 * console on the rare cases it is ever triggered. */
644 if (in_interrupt()) {
645 cfs_trace_debug_print();
647 libcfs_debug_dumplog_internal((void *)(long)current->pid);
653 static struct notifier_block libcfs_panic_notifier = {
654 .notifier_call = panic_notifier,
659 static void libcfs_register_panic_notifier(void)
661 atomic_notifier_chain_register(&panic_notifier_list,
662 &libcfs_panic_notifier);
665 static void libcfs_unregister_panic_notifier(void)
667 atomic_notifier_chain_unregister(&panic_notifier_list,
668 &libcfs_panic_notifier);
671 int libcfs_debug_init(unsigned long bufsize)
673 unsigned int max = libcfs_debug_mb;
676 if (libcfs_console_max_delay <= 0 || /* not set by user or */
677 libcfs_console_min_delay <= 0 || /* set to invalid values */
678 libcfs_console_min_delay >= libcfs_console_max_delay) {
679 libcfs_console_max_delay = CDEBUG_DEFAULT_MAX_DELAY;
680 libcfs_console_min_delay = CDEBUG_DEFAULT_MIN_DELAY;
683 if (libcfs_debug_file_path) {
684 strlcpy(libcfs_debug_file_path_arr,
685 libcfs_debug_file_path,
686 sizeof(libcfs_debug_file_path_arr));
689 /* If libcfs_debug_mb is uninitialized then just make the
690 * total buffers smp_num_cpus * TCD_MAX_PAGES
692 if (max < num_possible_cpus()) {
695 max = (max / num_possible_cpus());
696 max <<= (20 - PAGE_SHIFT);
699 rc = cfs_tracefile_init(max);
703 libcfs_register_panic_notifier();
704 kernel_param_lock(THIS_MODULE);
705 if (libcfs_debug_mb == 0)
706 libcfs_debug_mb = cfs_trace_get_debug_mb();
707 kernel_param_unlock(THIS_MODULE);
711 int libcfs_debug_cleanup(void)
713 libcfs_unregister_panic_notifier();
714 kernel_param_lock(THIS_MODULE);
715 cfs_tracefile_exit();
716 kernel_param_unlock(THIS_MODULE);
720 int libcfs_debug_clear_buffer(void)
722 cfs_trace_flush_pages();
726 /* Debug markers, although printed by S_LNET should not be be marked as such. */
727 #undef DEBUG_SUBSYSTEM
728 #define DEBUG_SUBSYSTEM S_UNDEFINED
729 int libcfs_debug_mark_buffer(const char *text)
732 "**************************************************\n");
733 LCONSOLE(D_WARNING, "DEBUG MARKER: %s\n", text);
735 "**************************************************\n");
740 #undef DEBUG_SUBSYSTEM
741 #define DEBUG_SUBSYSTEM S_LNET
743 long libcfs_log_return(struct libcfs_debug_msg_data *msgdata, long rc)
745 libcfs_debug_msg(msgdata, "Process leaving (rc=%lu : %ld : %lx)\n",
749 EXPORT_SYMBOL(libcfs_log_return);
751 void libcfs_log_goto(struct libcfs_debug_msg_data *msgdata, const char *label,
754 libcfs_debug_msg(msgdata, "Process leaving via %s (rc=%lu : %ld"
755 " : %#lx)\n", label, rc, rc, rc);
757 EXPORT_SYMBOL(libcfs_log_goto);