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_QSWNAL | S_SOCKNAL |
  63                                             S_GMNAL | S_OPENIBNAL);
  64 EXPORT_SYMBOL(portal_subsystem_debug);
  65
  66 unsigned int portal_debug = (D_WARNING | D_DLMTRACE | D_ERROR | D_EMERG | D_HA |
  67                              D_RPCTRACE | D_VFSTRACE);
  68 EXPORT_SYMBOL(portal_debug);
  69
  70 unsigned int portal_printk;
  71 EXPORT_SYMBOL(portal_printk);
  72
  73 unsigned int portal_stack;
  74 EXPORT_SYMBOL(portal_stack);
  75
  76 #ifdef __KERNEL__
  77 atomic_t portal_kmemory = ATOMIC_INIT(0);
  78 EXPORT_SYMBOL(portal_kmemory);
  79 #endif
  80
  81 static DECLARE_WAIT_QUEUE_HEAD(debug_ctlwq);
  82
  83 char debug_file_path[1024] = "/tmp/lustre-log";
  84 static char debug_file_name[1024];
  85 static int handled_panic; /* to avoid recursive calls to notifiers */
  86 char portals_upcall[1024] = "/usr/lib/lustre/portals_upcall";
  87
  88 int portals_do_debug_dumplog(void *arg)
  89 {
  90         void *journal_info;
  91
  92         kportal_daemonize("");
  93
  94         reparent_to_init();
  95         journal_info = current->journal_info;
  96         current->journal_info = NULL;
  97
  98         snprintf(debug_file_name, sizeof(debug_file_path) - 1,
  99                  "%s.%ld.%ld", debug_file_path, CURRENT_SECONDS, (long)arg);
 100         tracefile_dump_all_pages(debug_file_name);
 101
 102         current->journal_info = journal_info;
 103         wake_up(&debug_ctlwq);
 104         return 0;
 105 }
 106
 107 void portals_debug_dumplog(void)
 108 {
 109         int rc;
 110         DECLARE_WAITQUEUE(wait, current);
 111         ENTRY;
 112
 113         /* we're being careful to ensure that the kernel thread is
 114          * able to set our state to running as it exits before we
 115          * get to schedule() */
 116         set_current_state(TASK_INTERRUPTIBLE);
 117         add_wait_queue(&debug_ctlwq, &wait);
 118
 119         rc = kernel_thread(portals_do_debug_dumplog, (void *)(long)current->pid,
 120                            CLONE_VM | CLONE_FS | CLONE_FILES);
 121         if (rc < 0)
 122                 printk(KERN_ERR "LustreError: cannot start log dump thread: "
 123                        "%d\n", rc);
 124         else
 125                 schedule();
 126
 127         /* be sure to teardown if kernel_thread() failed */
 128         remove_wait_queue(&debug_ctlwq, &wait);
 129         set_current_state(TASK_RUNNING);
 130 }
 131
 132 static int panic_dumplog(struct notifier_block *self, unsigned long unused1,
 133                          void *unused2)
 134 {
 135         if (handled_panic)
 136                 return 0;
 137         else
 138                 handled_panic = 1;
 139
 140         if (in_interrupt()) {
 141                 trace_debug_print();
 142                 return 0;
 143         }
 144
 145         while (current->lock_depth >= 0)
 146                 unlock_kernel();
 147         portals_debug_dumplog();
 148         return 0;
 149 }
 150
 151 static struct notifier_block lustre_panic_notifier = {
 152         notifier_call :     panic_dumplog,
 153         next :              NULL,
 154         priority :          10000
 155 };
 156
 157 int portals_debug_init(unsigned long bufsize)
 158 {
 159         notifier_chain_register(&panic_notifier_list, &lustre_panic_notifier);
 160         return tracefile_init();
 161 }
 162
 163 int portals_debug_cleanup(void)
 164 {
 165         tracefile_exit();
 166         notifier_chain_unregister(&panic_notifier_list, &lustre_panic_notifier);
 167         return 0;
 168 }
 169
 170 int portals_debug_clear_buffer(void)
 171 {
 172         trace_flush_pages();
 173         return 0;
 174 }
 175
 176 /* Debug markers, although printed by S_PORTALS
 177  * should not be be marked as such. */
 178 #undef DEBUG_SUBSYSTEM
 179 #define DEBUG_SUBSYSTEM S_UNDEFINED
 180 int portals_debug_mark_buffer(char *text)
 181 {
 182         CDEBUG(D_TRACE,"***************************************************\n");
 183         CWARN("DEBUG MARKER: %s\n", text);
 184         CDEBUG(D_TRACE,"***************************************************\n");
 185
 186         return 0;
 187 }
 188 #undef DEBUG_SUBSYSTEM
 189 #define DEBUG_SUBSYSTEM S_PORTALS
 190
 191 void portals_debug_set_level(unsigned int debug_level)
 192 {
 193         printk(KERN_WARNING "Lustre: Setting portals debug level to %08x\n",
 194                debug_level);
 195         portal_debug = debug_level;
 196 }
 197
 198 void portals_run_upcall(char **argv)
 199 {
 200         int   rc;
 201         int   argc;
 202         char *envp[] = {
 203                 "HOME=/",
 204                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
 205                 NULL};
 206         ENTRY;
 207
 208         argv[0] = portals_upcall;
 209         argc = 1;
 210         while (argv[argc] != NULL)
 211                 argc++;
 212
 213         LASSERT(argc >= 2);
 214
 215         rc = USERMODEHELPER(argv[0], argv, envp);
 216         if (rc < 0) {
 217                 CERROR("Error %d invoking portals upcall %s %s%s%s%s%s%s%s%s; "
 218                        "check /proc/sys/portals/upcall\n",
 219                        rc, argv[0], argv[1],
 220                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 221                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 222                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 223                        argc < 6 ? "" : ",...");
 224         } else {
 225                 CERROR("Invoked portals upcall %s %s%s%s%s%s%s%s%s\n",
 226                        argv[0], argv[1],
 227                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 228                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 229                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 230                        argc < 6 ? "" : ",...");
 231         }
 232 }
 233
 234 void portals_run_lbug_upcall(char *file, const char *fn, const int line)
 235 {
 236         char *argv[6];
 237         char buf[32];
 238
 239         ENTRY;
 240         snprintf (buf, sizeof buf, "%d", line);
 241
 242         argv[1] = "LBUG";
 243         argv[2] = file;
 244         argv[3] = (char *)fn;
 245         argv[4] = buf;
 246         argv[5] = NULL;
 247
 248         portals_run_upcall (argv);
 249 }
 250
 251 char *portals_nid2str(int nal, ptl_nid_t nid, char *str)
 252 {
 253         if (nid == PTL_NID_ANY) {
 254                 snprintf(str, PTL_NALFMT_SIZE - 1, "%s",
 255                          "PTL_NID_ANY");
 256                 return str;
 257         }
 258
 259         switch(nal){
 260 /* XXX this could be a nal method of some sort, 'cept it's config
 261  * dependent whether (say) socknal NIDs are actually IP addresses... */
 262 #ifndef CRAY_PORTALS
 263         case TCPNAL:
 264                 /* userspace NAL */
 265         case OPENIBNAL:
 266         case SOCKNAL:
 267                 snprintf(str, PTL_NALFMT_SIZE - 1, "%u:%u.%u.%u.%u",
 268                          (__u32)(nid >> 32), HIPQUAD(nid));
 269                 break;
 270         case QSWNAL:
 271         case GMNAL:
 272                 snprintf(str, PTL_NALFMT_SIZE - 1, "%u:%u",
 273                          (__u32)(nid >> 32), (__u32)nid);
 274                 break;
 275 #endif
 276         default:
 277                 snprintf(str, PTL_NALFMT_SIZE - 1, "?%d? %llx",
 278                          nal, (long long)nid);
 279                 break;
 280         }
 281         return str;
 282 }
 283 /*      bug #4615       */
 284 char *portals_id2str(int nal, ptl_process_id_t id, char *str)
 285 {
 286         switch(nal){
 287 #ifndef CRAY_PORTALS
 288         case TCPNAL:
 289                 /* userspace NAL */
 290         case OPENIBNAL:
 291         case SOCKNAL:
 292                 snprintf(str, PTL_NALFMT_SIZE - 1, "%u:%u.%u.%u.%u,%u",
 293                          (__u32)(id.nid >> 32), HIPQUAD((id.nid)) , id.pid);
 294                 break;
 295         case QSWNAL:
 296         case GMNAL:
 297                 snprintf(str, PTL_NALFMT_SIZE - 1, "%u:%u,%u",
 298                          (__u32)(id.nid >> 32), (__u32)id.nid, id.pid);
 299                 break;
 300 #endif
 301         default:
 302                 snprintf(str, PTL_NALFMT_SIZE - 1, "?%d? %llx,%lx",
 303                          nal, (long long)id.nid, (long)id.pid );
 304                 break;
 305         }
 306         return str;
 307 }
 308
 309
 310 #ifdef __KERNEL__
 311 char stack_backtrace[LUSTRE_TRACE_SIZE];
 312 spinlock_t stack_backtrace_lock = SPIN_LOCK_UNLOCKED;
 313
 314 #if defined(__arch_um__)
 315
 316 char *portals_debug_dumpstack(void)
 317 {
 318         asm("int $3");
 319         return "dump stack\n";
 320 }
 321
 322 #elif defined(__i386__)
 323
 324 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 325 extern int lookup_symbol(unsigned long address, char *buf, int buflen);
 326 const char *kallsyms_lookup(unsigned long addr,
 327                             unsigned long *symbolsize,
 328                             unsigned long *offset,
 329                             char **modname, char *namebuf)
 330 {
 331         int rc = lookup_symbol(addr, namebuf, 128);
 332         if (rc == -ENOSYS)
 333                 return NULL;
 334         return namebuf;
 335 }
 336 #endif
 337
 338 char *portals_debug_dumpstack(void)
 339 {
 340         unsigned long esp = current->thread.esp, addr;
 341         unsigned long *stack = (unsigned long *)&esp;
 342         char *buf = stack_backtrace, *pbuf = buf;
 343         int size;
 344
 345         /* User space on another CPU? */
 346         if ((esp ^ (unsigned long)current) & (PAGE_MASK << 1)){
 347                 buf[0] = '\0';
 348                 goto out;
 349         }
 350
 351         size = sprintf(pbuf, " Call Trace: ");
 352         pbuf += size;
 353         while (((long) stack & (THREAD_SIZE - 1)) != 0) {
 354                 addr = *stack++;
 355                 if (kernel_text_address(addr)) {
 356                         const char *sym_name;
 357                         char *modname, buffer[128];
 358                         unsigned long junk, offset;
 359
 360                         sym_name = kallsyms_lookup(addr, &junk, &offset,
 361                                                    &modname, buffer);
 362                         if (sym_name == NULL) {
 363                                 if (buf + LUSTRE_TRACE_SIZE <= pbuf + 12)
 364                                         break;
 365                                 size = sprintf(pbuf, "[<%08lx>] ", addr);
 366                         } else {
 367                                 if (buf + LUSTRE_TRACE_SIZE
 368                                             /* fix length + sizeof('\0') */
 369                                     <= pbuf + strlen(buffer) + 28 + 1)
 370                                         break;
 371                                 size = sprintf(pbuf, "([<%08lx>] %s (0x%p)) ",
 372                                                addr, buffer, stack - 1);
 373                         }
 374                         pbuf += size;
 375                 }
 376         }
 377 out:
 378         return buf;
 379 }
 380
 381 #else /* !__arch_um__ && !__i386__ */
 382
 383 char *portals_debug_dumpstack(void)
 384 {
 385         char *buf = stack_backtrace;
 386         buf[0] = '\0';
 387         return buf;
 388 }
 389
 390 #endif /* __arch_um__ */
 391 struct task_struct *portals_current(void)
 392 {
 393         CWARN("current task struct is %p\n", current);
 394         return current;
 395 }
 396
 397 EXPORT_SYMBOL(stack_backtrace_lock);
 398 EXPORT_SYMBOL(portals_debug_dumpstack);
 399 EXPORT_SYMBOL(portals_current);
 400 #endif /* __KERNEL__ */
 401
 402 EXPORT_SYMBOL(portals_debug_dumplog);
 403 EXPORT_SYMBOL(portals_debug_set_level);
 404 EXPORT_SYMBOL(portals_run_upcall);
 405 EXPORT_SYMBOL(portals_run_lbug_upcall);
 406 EXPORT_SYMBOL(portals_nid2str);
 407 EXPORT_SYMBOL(portals_id2str);