lnet/libcfs/debug.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002 Cluster File Systems, Inc.
   5  *   Author: Phil Schwan <phil@clusterfs.com>
   6  *
   7  *   This file is part of Lustre, http://www.lustre.org.
   8  *
   9  *   Lustre is free software; you can redistribute it and/or
  10  *   modify it under the terms of version 2 of the GNU General Public
  11  *   License as published by the Free Software Foundation.
  12  *
  13  *   Lustre is distributed in the hope that it will be useful,
  14  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *   GNU General Public License for more details.
  17  *
  18  *   You should have received a copy of the GNU General Public License
  19  *   along with Lustre; if not, write to the Free Software
  20  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  */
  22
  23 #ifndef EXPORT_SYMTAB
  24 # define EXPORT_SYMTAB
  25 #endif
  26
  27 #include <linux/config.h>
  28 #include <linux/module.h>
  29 #include <linux/kmod.h>
  30 #include <linux/notifier.h>
  31 #include <linux/kernel.h>
  32 #include <linux/mm.h>
  33 #include <linux/string.h>
  34 #include <linux/stat.h>
  35 #include <linux/errno.h>
  36 #include <linux/smp_lock.h>
  37 #include <linux/unistd.h>
  38 #include <linux/interrupt.h>
  39 #include <asm/system.h>
  40 #include <asm/uaccess.h>
  41 #include <linux/completion.h>
  42
  43 #include <linux/fs.h>
  44 #include <linux/stat.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/segment.h>
  47 #include <linux/miscdevice.h>
  48 #include <linux/version.h>
  49
  50 # define DEBUG_SUBSYSTEM S_PORTALS
  51
  52 #include <linux/kp30.h>
  53 #include <linux/portals_compat25.h>
  54 #include <linux/libcfs.h>
  55
  56 #include "tracefile.h"
  57
  58 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
  59 #include <linux/kallsyms.h>
  60 #endif
  61
  62 unsigned int portal_subsystem_debug = ~0 - (S_PORTALS | S_NAL);
  63 EXPORT_SYMBOL(portal_subsystem_debug);
  64
  65 unsigned int portal_debug = (D_WARNING | D_DLMTRACE | D_ERROR | D_EMERG | D_HA |
  66                              D_RPCTRACE | D_VFSTRACE);
  67 EXPORT_SYMBOL(portal_debug);
  68
  69 unsigned int portal_printk;
  70 EXPORT_SYMBOL(portal_printk);
  71
  72 unsigned int portal_stack;
  73 EXPORT_SYMBOL(portal_stack);
  74
  75 #ifdef __KERNEL__
  76 atomic_t portal_kmemory = ATOMIC_INIT(0);
  77 EXPORT_SYMBOL(portal_kmemory);
  78 #endif
  79
  80 static DECLARE_WAIT_QUEUE_HEAD(debug_ctlwq);
  81
  82 char debug_file_path[1024] = "/tmp/lustre-log";
  83 static char debug_file_name[1024];
  84 char portals_upcall[1024] = "/usr/lib/lustre/portals_upcall";
  85
  86 void portals_debug_dumplog_internal(void *arg)
  87 {
  88         void *journal_info = current->journal_info;
  89         current->journal_info = NULL;
  90
  91         snprintf(debug_file_name, sizeof(debug_file_path) - 1,
  92                  "%s.%ld.%ld", debug_file_path, CURRENT_SECONDS, (long)arg);
  93         printk(KERN_ALERT "LustreError: dumping log to %s\n", debug_file_name);
  94         tracefile_dump_all_pages(debug_file_name);
  95
  96         current->journal_info = journal_info;
  97 }
  98
  99 int portals_debug_dumplog_thread(void *arg)
 100 {
 101         kportal_daemonize("");
 102         reparent_to_init();
 103         portals_debug_dumplog_internal(arg);
 104         wake_up(&debug_ctlwq);
 105         return 0;
 106 }
 107
 108 void portals_debug_dumplog(void)
 109 {
 110         int rc;
 111         DECLARE_WAITQUEUE(wait, current);
 112         ENTRY;
 113
 114         /* we're being careful to ensure that the kernel thread is
 115          * able to set our state to running as it exits before we
 116          * get to schedule() */
 117         set_current_state(TASK_INTERRUPTIBLE);
 118         add_wait_queue(&debug_ctlwq, &wait);
 119
 120         rc = kernel_thread(portals_debug_dumplog_thread,
 121                            (void *)(long)current->pid,
 122                            CLONE_VM | CLONE_FS | CLONE_FILES);
 123         if (rc < 0)
 124                 printk(KERN_ERR "LustreError: cannot start log dump thread: "
 125                        "%d\n", rc);
 126         else
 127                 schedule();
 128
 129         /* be sure to teardown if kernel_thread() failed */
 130         remove_wait_queue(&debug_ctlwq, &wait);
 131         set_current_state(TASK_RUNNING);
 132 }
 133
 134 #ifdef PORTALS_DUMP_ON_PANIC
 135 static int panic_dumplog(struct notifier_block *self, unsigned long unused1,
 136                          void *unused2)
 137 {
 138         static int handled_panic; /* to avoid recursive calls to notifiers */
 139
 140         if (handled_panic)
 141                 return 0;
 142         else
 143                 handled_panic = 1;
 144
 145         if (in_interrupt()) {
 146                 trace_debug_print();
 147                 return 0;
 148         }
 149
 150         while (current->lock_depth >= 0)
 151                 unlock_kernel();
 152         portals_debug_dumplog();
 153         return 0;
 154 }
 155
 156 static struct notifier_block lustre_panic_notifier = {
 157         notifier_call :     panic_dumplog,
 158         next :              NULL,
 159         priority :          10000
 160 };
 161 #endif
 162
 163 #ifdef CRAY_PORTALS
 164 extern void *lus_portals_debug;
 165 #endif
 166
 167 int portals_debug_init(unsigned long bufsize)
 168 {
 169 #ifdef CRAY_PORTALS
 170         lus_portals_debug = &portals_debug_msg;
 171 #endif
 172 #ifdef PORTALS_DUMP_ON_PANIC
 173         /* This is currently disabled because it spews far too much to the
 174          * console on the rare cases it is ever triggered. */
 175         notifier_chain_register(&panic_notifier_list, &lustre_panic_notifier);
 176 #endif
 177         return tracefile_init();
 178 }
 179
 180 int portals_debug_cleanup(void)
 181 {
 182         tracefile_exit();
 183 #ifdef PORTALS_DUMP_ON_PANIC
 184         notifier_chain_unregister(&panic_notifier_list, &lustre_panic_notifier);
 185 #endif
 186 #ifdef CRAY_PORTALS
 187         lus_portals_debug = NULL;
 188 #endif
 189         return 0;
 190 }
 191
 192 int portals_debug_clear_buffer(void)
 193 {
 194         trace_flush_pages();
 195         return 0;
 196 }
 197
 198 /* Debug markers, although printed by S_PORTALS
 199  * should not be be marked as such. */
 200 #undef DEBUG_SUBSYSTEM
 201 #define DEBUG_SUBSYSTEM S_UNDEFINED
 202 int portals_debug_mark_buffer(char *text)
 203 {
 204         CDEBUG(D_TRACE,"***************************************************\n");
 205         CDEBUG(D_WARNING, "DEBUG MARKER: %s\n", text);
 206         CDEBUG(D_TRACE,"***************************************************\n");
 207
 208         return 0;
 209 }
 210 #undef DEBUG_SUBSYSTEM
 211 #define DEBUG_SUBSYSTEM S_PORTALS
 212
 213 void portals_debug_set_level(unsigned int debug_level)
 214 {
 215         printk(KERN_WARNING "Lustre: Setting portals debug level to %08x\n",
 216                debug_level);
 217         portal_debug = debug_level;
 218 }
 219
 220 void portals_run_upcall(char **argv)
 221 {
 222         int   rc;
 223         int   argc;
 224         char *envp[] = {
 225                 "HOME=/",
 226                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
 227                 NULL};
 228         ENTRY;
 229
 230         argv[0] = portals_upcall;
 231         argc = 1;
 232         while (argv[argc] != NULL)
 233                 argc++;
 234
 235         LASSERT(argc >= 2);
 236
 237         rc = USERMODEHELPER(argv[0], argv, envp);
 238         if (rc < 0) {
 239                 CERROR("Error %d invoking portals upcall %s %s%s%s%s%s%s%s%s; "
 240                        "check /proc/sys/portals/upcall\n",
 241                        rc, argv[0], argv[1],
 242                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 243                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 244                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 245                        argc < 6 ? "" : ",...");
 246         } else {
 247                 CERROR("Invoked portals upcall %s %s%s%s%s%s%s%s%s\n",
 248                        argv[0], argv[1],
 249                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 250                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 251                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 252                        argc < 6 ? "" : ",...");
 253         }
 254 }
 255
 256 void portals_run_lbug_upcall(char *file, const char *fn, const int line)
 257 {
 258         char *argv[6];
 259         char buf[32];
 260
 261         ENTRY;
 262         snprintf (buf, sizeof buf, "%d", line);
 263
 264         argv[1] = "LBUG";
 265         argv[2] = file;
 266         argv[3] = (char *)fn;
 267         argv[4] = buf;
 268         argv[5] = NULL;
 269
 270         portals_run_upcall (argv);
 271 }
 272
 273 char *portals_nid2str(int nal, ptl_nid_t nid, char *str)
 274 {
 275         if (nid == PTL_NID_ANY) {
 276                 snprintf(str, PTL_NALFMT_SIZE, "%s", "PTL_NID_ANY");
 277                 return str;
 278         }
 279
 280         switch(nal){
 281 /* XXX this could be a nal method of some sort, 'cept it's config
 282  * dependent whether (say) socknal NIDs are actually IP addresses... */
 283 #if !CRAY_PORTALS
 284         case TCPNAL:
 285                 /* userspace NAL */
 286         case IIBNAL:
 287         case VIBNAL:
 288         case OPENIBNAL:
 289         case RANAL:
 290         case SOCKNAL:
 291                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u.%u.%u.%u",
 292                          (__u32)(nid >> 32), HIPQUAD(nid));
 293                 break;
 294         case QSWNAL:
 295         case GMNAL:
 296         case LONAL:
 297                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u",
 298                          (__u32)(nid >> 32), (__u32)nid);
 299                 break;
 300 #endif
 301         default:
 302                 snprintf(str, PTL_NALFMT_SIZE, "?%x? %llx",
 303                          nal, (long long)nid);
 304                 break;
 305         }
 306         return str;
 307 }
 308
 309 char *portals_id2str(int nal, ptl_process_id_t id, char *str)
 310 {
 311         int   len;
 312
 313         portals_nid2str(nal, id.nid, str);
 314         len = strlen(str);
 315         snprintf(str + len, PTL_NALFMT_SIZE - len, "-%u", id.pid);
 316         return str;
 317 }
 318
 319 #ifdef __KERNEL__
 320
 321 void portals_debug_dumpstack(struct task_struct *tsk)
 322 {
 323 #if defined(__arch_um__)
 324         if (tsk != NULL)
 325                 CWARN("stack dump for pid %d (%d) requested; wake up gdb.\n",
 326                       tsk->pid, UML_PID(tsk));
 327         asm("int $3");
 328 #elif defined(HAVE_SHOW_TASK)
 329         /* this is exported by lustre kernel version 42 */
 330         extern void show_task(struct task_struct *);
 331
 332         if (tsk == NULL)
 333                 tsk = current;
 334         CWARN("showing stack for process %d\n", tsk->pid);
 335         show_task(tsk);
 336 #else
 337         CWARN("can't show stack: kernel doesn't export show_task\n");
 338 #endif
 339 }
 340
 341 struct task_struct *portals_current(void)
 342 {
 343         CWARN("current task struct is %p\n", current);
 344         return current;
 345 }
 346
 347 EXPORT_SYMBOL(portals_debug_dumpstack);
 348 EXPORT_SYMBOL(portals_current);
 349 #endif /* __KERNEL__ */
 350
 351 EXPORT_SYMBOL(portals_debug_dumplog);
 352 EXPORT_SYMBOL(portals_debug_set_level);
 353 EXPORT_SYMBOL(portals_run_upcall);
 354 EXPORT_SYMBOL(portals_run_lbug_upcall);
 355 EXPORT_SYMBOL(portals_nid2str);
 356 EXPORT_SYMBOL(portals_id2str);