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_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 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         printk(KERN_ALERT "LustreError: dumping log to %s\n", debug_file_name);
 101         tracefile_dump_all_pages(debug_file_name);
 102
 103         current->journal_info = journal_info;
 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_do_debug_dumplog, (void *)(long)current->pid,
 121                            CLONE_VM | CLONE_FS | CLONE_FILES);
 122         if (rc < 0)
 123                 printk(KERN_ERR "LustreError: cannot start log dump thread: "
 124                        "%d\n", rc);
 125         else
 126                 schedule();
 127
 128         /* be sure to teardown if kernel_thread() failed */
 129         remove_wait_queue(&debug_ctlwq, &wait);
 130         set_current_state(TASK_RUNNING);
 131 }
 132
 133 static int panic_dumplog(struct notifier_block *self, unsigned long unused1,
 134                          void *unused2)
 135 {
 136         if (handled_panic)
 137                 return 0;
 138         else
 139                 handled_panic = 1;
 140
 141         if (in_interrupt()) {
 142                 trace_debug_print();
 143                 return 0;
 144         }
 145
 146         while (current->lock_depth >= 0)
 147                 unlock_kernel();
 148         portals_debug_dumplog();
 149         return 0;
 150 }
 151
 152 static struct notifier_block lustre_panic_notifier = {
 153         notifier_call :     panic_dumplog,
 154         next :              NULL,
 155         priority :          10000
 156 };
 157
 158 int portals_debug_init(unsigned long bufsize)
 159 {
 160         notifier_chain_register(&panic_notifier_list, &lustre_panic_notifier);
 161         return tracefile_init();
 162 }
 163
 164 int portals_debug_cleanup(void)
 165 {
 166         tracefile_exit();
 167         notifier_chain_unregister(&panic_notifier_list, &lustre_panic_notifier);
 168         return 0;
 169 }
 170
 171 int portals_debug_clear_buffer(void)
 172 {
 173         trace_flush_pages();
 174         return 0;
 175 }
 176
 177 /* Debug markers, although printed by S_PORTALS
 178  * should not be be marked as such. */
 179 #undef DEBUG_SUBSYSTEM
 180 #define DEBUG_SUBSYSTEM S_UNDEFINED
 181 int portals_debug_mark_buffer(char *text)
 182 {
 183         CDEBUG(D_TRACE,"***************************************************\n");
 184         CDEBUG(D_WARNING, "DEBUG MARKER: %s\n", text);
 185         CDEBUG(D_TRACE,"***************************************************\n");
 186
 187         return 0;
 188 }
 189 #undef DEBUG_SUBSYSTEM
 190 #define DEBUG_SUBSYSTEM S_PORTALS
 191
 192 void portals_debug_set_level(unsigned int debug_level)
 193 {
 194         printk(KERN_WARNING "Lustre: Setting portals debug level to %08x\n",
 195                debug_level);
 196         portal_debug = debug_level;
 197 }
 198
 199 void portals_run_upcall(char **argv)
 200 {
 201         int   rc;
 202         int   argc;
 203         char *envp[] = {
 204                 "HOME=/",
 205                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
 206                 NULL};
 207         ENTRY;
 208
 209         argv[0] = portals_upcall;
 210         argc = 1;
 211         while (argv[argc] != NULL)
 212                 argc++;
 213
 214         LASSERT(argc >= 2);
 215
 216         rc = USERMODEHELPER(argv[0], argv, envp);
 217         if (rc < 0) {
 218                 CERROR("Error %d invoking portals upcall %s %s%s%s%s%s%s%s%s; "
 219                        "check /proc/sys/portals/upcall\n",
 220                        rc, argv[0], argv[1],
 221                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 222                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 223                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 224                        argc < 6 ? "" : ",...");
 225         } else {
 226                 CERROR("Invoked portals upcall %s %s%s%s%s%s%s%s%s\n",
 227                        argv[0], argv[1],
 228                        argc < 3 ? "" : ",", argc < 3 ? "" : argv[2],
 229                        argc < 4 ? "" : ",", argc < 4 ? "" : argv[3],
 230                        argc < 5 ? "" : ",", argc < 5 ? "" : argv[4],
 231                        argc < 6 ? "" : ",...");
 232         }
 233 }
 234
 235 void portals_run_lbug_upcall(char *file, const char *fn, const int line)
 236 {
 237         char *argv[6];
 238         char buf[32];
 239
 240         ENTRY;
 241         snprintf (buf, sizeof buf, "%d", line);
 242
 243         argv[1] = "LBUG";
 244         argv[2] = file;
 245         argv[3] = (char *)fn;
 246         argv[4] = buf;
 247         argv[5] = NULL;
 248
 249         portals_run_upcall (argv);
 250 }
 251
 252 char *portals_nid2str(int nal, ptl_nid_t nid, char *str)
 253 {
 254         if (nid == PTL_NID_ANY) {
 255                 snprintf(str, PTL_NALFMT_SIZE, "%s", "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 #if !CRAY_PORTALS
 263         case TCPNAL:
 264                 /* userspace NAL */
 265         case IIBNAL:
 266         case OPENIBNAL:
 267         case SOCKNAL:
 268                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u.%u.%u.%u",
 269                          (__u32)(nid >> 32), HIPQUAD(nid));
 270                 break;
 271         case QSWNAL:
 272         case GMNAL:
 273                 snprintf(str, PTL_NALFMT_SIZE, "%u:%u",
 274                          (__u32)(nid >> 32), (__u32)nid);
 275                 break;
 276 #endif
 277         default:
 278                 snprintf(str, PTL_NALFMT_SIZE, "?%x? %llx",
 279                          nal, (long long)nid);
 280                 break;
 281         }
 282         return str;
 283 }
 284
 285 char *portals_id2str(int nal, ptl_process_id_t id, char *str)
 286 {
 287         int   len;
 288
 289         portals_nid2str(nal, id.nid, str);
 290         len = strlen(str);
 291         snprintf(str + len, PTL_NALFMT_SIZE - len, ",%u", id.pid);
 292         return str;
 293 }
 294
 295 #ifdef __KERNEL__
 296 char stack_backtrace[LUSTRE_TRACE_SIZE];
 297 spinlock_t stack_backtrace_lock = SPIN_LOCK_UNLOCKED;
 298
 299 #if defined(__arch_um__)
 300
 301 char *portals_debug_dumpstack(void)
 302 {
 303         asm("int $3");
 304         return "dump stack\n";
 305 }
 306
 307 #elif defined(__i386__)
 308
 309 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 310 extern int lookup_symbol(unsigned long address, char *buf, int buflen);
 311 const char *kallsyms_lookup(unsigned long addr,
 312                             unsigned long *symbolsize,
 313                             unsigned long *offset,
 314                             char **modname, char *namebuf)
 315 {
 316         int rc = lookup_symbol(addr, namebuf, 128);
 317         if (rc == -ENOSYS)
 318                 return NULL;
 319         return namebuf;
 320 }
 321 #endif
 322
 323 char *portals_debug_dumpstack(void)
 324 {
 325         unsigned long esp = current->thread.esp, addr;
 326         unsigned long *stack = (unsigned long *)&esp;
 327         char *buf = stack_backtrace, *pbuf = buf;
 328         int size;
 329
 330         /* User space on another CPU? */
 331         if ((esp ^ (unsigned long)current) & (PAGE_MASK << 1)){
 332                 buf[0] = '\0';
 333                 goto out;
 334         }
 335
 336         size = sprintf(pbuf, " Call Trace: ");
 337         pbuf += size;
 338         while (((long) stack & (THREAD_SIZE - 1)) != 0) {
 339                 addr = *stack++;
 340                 if (kernel_text_address(addr)) {
 341                         const char *sym_name;
 342                         char *modname, buffer[128];
 343                         unsigned long junk, offset;
 344
 345                         sym_name = kallsyms_lookup(addr, &junk, &offset,
 346                                                    &modname, buffer);
 347                         if (sym_name == NULL) {
 348                                 if (buf + LUSTRE_TRACE_SIZE <= pbuf + 12)
 349                                         break;
 350                                 size = sprintf(pbuf, "[<%08lx>] ", addr);
 351                         } else {
 352                                 if (buf + LUSTRE_TRACE_SIZE
 353                                             /* fix length + sizeof('\0') */
 354                                     <= pbuf + strlen(buffer) + 28 + 1)
 355                                         break;
 356                                 size = sprintf(pbuf, "([<%08lx>] %s (0x%p)) ",
 357                                                addr, buffer, stack - 1);
 358                         }
 359                         pbuf += size;
 360                 }
 361         }
 362 out:
 363         return buf;
 364 }
 365
 366 #else /* !__arch_um__ && !__i386__ */
 367
 368 char *portals_debug_dumpstack(void)
 369 {
 370         char *buf = stack_backtrace;
 371         buf[0] = '\0';
 372         return buf;
 373 }
 374
 375 #endif /* __arch_um__ */
 376 struct task_struct *portals_current(void)
 377 {
 378         CWARN("current task struct is %p\n", current);
 379         return current;
 380 }
 381
 382 EXPORT_SYMBOL(stack_backtrace_lock);
 383 EXPORT_SYMBOL(portals_debug_dumpstack);
 384 EXPORT_SYMBOL(portals_current);
 385 #endif /* __KERNEL__ */
 386
 387 EXPORT_SYMBOL(portals_debug_dumplog);
 388 EXPORT_SYMBOL(portals_debug_set_level);
 389 EXPORT_SYMBOL(portals_run_upcall);
 390 EXPORT_SYMBOL(portals_run_lbug_upcall);
 391 EXPORT_SYMBOL(portals_nid2str);
 392 EXPORT_SYMBOL(portals_id2str);