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