lustre/portals/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 static int handled_panic; /* to avoid recursive calls to notifiers */
  85 char portals_upcall[1024] = "/usr/lib/lustre/portals_upcall";
  86
  87 void portals_debug_dumplog_internal(void *arg)
  88 {
  89         void *journal_info = current->journal_info;
  90         current->journal_info = NULL;
  91
  92         snprintf(debug_file_name, sizeof(debug_file_path) - 1,
  93                  "%s.%ld.%ld", debug_file_path, CURRENT_SECONDS, (long)arg);
  94         printk(KERN_ALERT "LustreError: dumping log to %s\n", debug_file_name);
  95         tracefile_dump_all_pages(debug_file_name);
  96
  97         current->journal_info = journal_info;
  98 }
  99
 100 int portals_debug_dumplog_thread(void *arg)
 101 {
 102         kportal_daemonize("");
 103         reparent_to_init();
 104         portals_debug_dumplog_internal(arg);
 105         wake_up(&debug_ctlwq);
 106         return 0;
 107 }
 108
 109 void portals_debug_dumplog(void)
 110 {
 111         int rc;
 112         DECLARE_WAITQUEUE(wait, current);
 113         ENTRY;
 114
 115         /* we're being careful to ensure that the kernel thread is
 116          * able to set our state to running as it exits before we
 117          * get to schedule() */
 118         set_current_state(TASK_INTERRUPTIBLE);
 119         add_wait_queue(&debug_ctlwq, &wait);
 120
 121         rc = kernel_thread(portals_debug_dumplog_thread,
 122                            (void *)(long)current->pid,
 123                            CLONE_VM | CLONE_FS | CLONE_FILES);
 124         if (rc < 0)
 125                 printk(KERN_ERR "LustreError: cannot start log dump thread: "
 126                        "%d\n", rc);
 127         else
 128                 schedule();
 129
 130         /* be sure to teardown if kernel_thread() failed */
 131         remove_wait_queue(&debug_ctlwq, &wait);
 132         set_current_state(TASK_RUNNING);
 133 }
 134
 135 static int panic_dumplog(struct notifier_block *self, unsigned long unused1,
 136                          void *unused2)
 137 {
 138         if (handled_panic)
 139                 return 0;
 140         else
 141                 handled_panic = 1;
 142
 143         if (in_interrupt()) {
 144                 trace_debug_print();
 145                 return 0;
 146         }
 147
 148         while (current->lock_depth >= 0)
 149                 unlock_kernel();
 150         portals_debug_dumplog();
 151         return 0;
 152 }
 153
 154 static struct notifier_block lustre_panic_notifier = {
 155         notifier_call :     panic_dumplog,
 156         next :              NULL,
 157         priority :          10000
 158 };
 159
 160 int portals_debug_init(unsigned long bufsize)
 161 {
 162         notifier_chain_register(&panic_notifier_list, &lustre_panic_notifier);
 163         return tracefile_init();
 164 }
 165
 166 int portals_debug_cleanup(void)
 167 {
 168         tracefile_exit();
 169         notifier_chain_unregister(&panic_notifier_list, &lustre_panic_notifier);
 170         return 0;
 171 }
 172
 173 int portals_debug_clear_buffer(void)
 174 {
 175         trace_flush_pages();
 176         return 0;
 177 }
 178
 179 /* Debug markers, although printed by S_PORTALS
 180  * should not be be marked as such. */
 181 #undef DEBUG_SUBSYSTEM
 182 #define DEBUG_SUBSYSTEM S_UNDEFINED
 183 int portals_debug_mark_buffer(char *text)
 184 {
 185         CDEBUG(D_TRACE,"***************************************************\n");
 186         CDEBUG(D_WARNING, "DEBUG MARKER: %s\n", text);
 187         CDEBUG(D_TRACE,"***************************************************\n");
 188
 189         return 0;
 190 }
 191 #undef DEBUG_SUBSYSTEM
 192 #define DEBUG_SUBSYSTEM S_PORTALS
 193
 194 void portals_debug_set_level(unsigned int debug_level)
 195 {
 196         printk(KERN_WARNING "Lustre: Setting portals debug level to %08x\n",
 197                debug_level);
 198         portal_debug = debug_level;
 199 }
 200
 201 void portals_run_upcall(char **argv)
 202 {
 203         int   rc;
 204         int   argc;
 205         char *envp[] = {
 206                 "HOME=/",
 207                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
 208                 NULL};
 209         ENTRY;
 210
 211         argv[0] = portals_upcall;
 212         argc = 1;
 213         while (argv[argc] != NULL)
 214                 argc++;
 215
 216         LASSERT(argc >= 2);
 217
 218         rc = USERMODEHELPER(argv[0], argv, envp);
 219         if (rc < 0) {
 220                 CERROR("Error %d invoking portals upcall %s %s%s%s%s%s%s%s%s; "
 221                        "check /proc/sys/portals/upcall\n",
 222                        rc, argv[0], argv[1],
 223                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 224                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 225                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 226                        argc < 6 ? "" : ",...");
 227         } else {
 228                 CERROR("Invoked portals upcall %s %s%s%s%s%s%s%s%s\n",
 229                        argv[0], argv[1],
 230                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 231                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 232                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 233                        argc < 6 ? "" : ",...");
 234         }
 235 }
 236
 237 void portals_run_lbug_upcall(char *file, const char *fn, const int line)
 238 {
 239         char *argv[6];
 240         char buf[32];
 241
 242         ENTRY;
 243         snprintf (buf, sizeof buf, "%d", line);
 244
 245         argv[1] = "LBUG";
 246         argv[2] = file;
 247         argv[3] = (char *)fn;
 248         argv[4] = buf;
 249         argv[5] = NULL;
 250
 251         portals_run_upcall (argv);
 252 }
 253
 254 char *portals_nid2str(int nal, ptl_nid_t nid, char *str)
 255 {
 256         if (nid == PTL_NID_ANY) {
 257                 snprintf(str, PTL_NALFMT_SIZE, "%s", "PTL_NID_ANY");
 258                 return str;
 259         }
 260
 261         switch(nal){
 262 /* XXX this could be a nal method of some sort, 'cept it's config
 263  * dependent whether (say) socknal NIDs are actually IP addresses... */
 264 #if !CRAY_PORTALS
 265         case TCPNAL:
 266                 /* userspace NAL */
 267         case IIBNAL:
 268         case OPENIBNAL:
 269         case RANAL:
 270         case SOCKNAL:
 271                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u.%u.%u.%u",
 272                          (__u32)(nid >> 32), HIPQUAD(nid));
 273                 break;
 274         case QSWNAL:
 275         case GMNAL:
 276         case LONAL:
 277                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u",
 278                          (__u32)(nid >> 32), (__u32)nid);
 279                 break;
 280 #endif
 281         default:
 282                 snprintf(str, PTL_NALFMT_SIZE, "?%x? %llx",
 283                          nal, (long long)nid);
 284                 break;
 285         }
 286         return str;
 287 }
 288
 289 char *portals_id2str(int nal, ptl_process_id_t id, char *str)
 290 {
 291         int   len;
 292
 293         portals_nid2str(nal, id.nid, str);
 294         len = strlen(str);
 295         snprintf(str + len, PTL_NALFMT_SIZE - len, "-%u", id.pid);
 296         return str;
 297 }
 298
 299 #ifdef __KERNEL__
 300
 301 void portals_debug_dumpstack(struct task_struct *tsk)
 302 {
 303 #if defined(__arch_um__)
 304         if (tsk != NULL)
 305                 CWARN("stack dump for pid %d (%d) requested; wake up gdb.\n",
 306                       tsk->pid, UML_PID(tsk));
 307         asm("int $3");
 308 #elif defined(HAVE_SHOW_TASK)
 309         /* this is exported by lustre kernel version 42 */
 310         extern void show_task(struct task_struct *);
 311
 312         if (tsk == NULL)
 313                 tsk = current;
 314         CWARN("showing stack for process %d\n", tsk->pid);
 315         show_task(tsk);
 316 #else
 317         CWARN("can't show stack: kernel doesn't export show_task\n");
 318 #endif
 319 }
 320
 321 struct task_struct *portals_current(void)
 322 {
 323         CWARN("current task struct is %p\n", current);
 324         return current;
 325 }
 326
 327 EXPORT_SYMBOL(portals_debug_dumpstack);
 328 EXPORT_SYMBOL(portals_current);
 329 #endif /* __KERNEL__ */
 330
 331 EXPORT_SYMBOL(portals_debug_dumplog);
 332 EXPORT_SYMBOL(portals_debug_set_level);
 333 EXPORT_SYMBOL(portals_run_upcall);
 334 EXPORT_SYMBOL(portals_run_lbug_upcall);
 335 EXPORT_SYMBOL(portals_nid2str);
 336 EXPORT_SYMBOL(portals_id2str);