#! /bin/sh # This script supersedes all previous versions of "crpt" and # "kreport" that have been emailed out. # # Some minor mods by Dave Olson 12/93 (for IRIX 405x systems). # mainly to add the time of the crash to the output, and to # be less 'noisy'. # # Mods made by Pete Sullivan 12/93 # -Made this versions work with both 4.* and 5.* core files. # -crpt will now uncompress core files as needed. # -cleaned up the arguments and allowed vmcore.N.comp files # to be passed in as args. # -workaround for dbx bug #196361 (1-14-94) # # # Frequently asked questions (FAQ's): # # Can I redistribute this script ? # # Well, it helps SGI get more detailed reports about crashing systems # quickly, so the more it's used the better. Unrestricted distribution. # # How do I interpret the results ? # # If you have a support contract or are still under warranty, # please contact your local support provider. In North America # contact the Technical Assistance Center at 1-800-800-4SGI. # If you are outside North America, contact the Silicon Graphics # subsidiary or authorized distributor in your country. # # What do I do with the files in /usr/adm/crash (/var/adm/crash) ? # # Don't delete the core files after you've generated the crash report. # Wait until your support provider tells you that this is a "known # problem". If the crash requires further investigation, a kernel expert # will need to examine the core files more closely. Even if you # desperately need the disk space, it's always wise to at least back the # core files up to tape before deleting them. # # # CRPT(1) UNIX System V CPRT(1) # # NAME # crpt - Kernel core file report generator # # SYNOPSIS # crpt [-n] [N] # crpt [-n] [unix.N] [vmcore.N] # crpt [-n] [unix.N] [vmcore.N.comp] # # DESCRIPTION # crpt generates a report from kernel core files that can aid # your support provider in determining the cause of a system # crash. The report is saved in current directory as # "Crash_N.rpt" (Where "N" is the number of the dump file as # specified by /usr/adm/crash/bounds). # # When a system crashes, it saves the contents of physical # memory and a copy of the kernel that was running. savecore # places these files in /usr/adm/crash. If your system has # crashed and you do not have unix.N and vmcore.N in # /usr/adm/crash, make sure your primary swap space on your # system disk is at least as large as your physical memory and # that you have at least that much free space in your /usr # partition (see savecore(1M)). Save unix.N and vmcore.N to # tape in case indepth debugging is required. Usually only a # kernel expert with access to the SGI source code can # accurately debug these core files. crpt drives the dbx # debugger through the basic steps a kernel expert would start # with when debugging a system crash. The dbx(1) debugger # must be installed for crpt to run. The report can be # included in a call logged through electronic services or # emailed or faxed to your local support provider. # # crpt provides the following information: # # 1) Report System Info: # # -The full path of the given core files # -Output of the utsname data structure to provide # system information: # # If the script is run on the crashing system, it # will provide additional hardware and software # information. The output from "hinv" and # "versions -b" will be incorporated into the # script. # # 2) Formatted output from the kernel's error buffer. # This buffer contains errors messages that are # flushed to the system's error log # (/usr/adm/SYSLOG), but this buffer is often not # flushed when the system crashes. # # 3) Print the name of the program that was running # at the time of the crash and the process id of # that program. # # 4) Print a kernel stack trace on the current # process id with the "print_exception_frame" # variable set. This will help trace the # functions called which lead up to the crash. # If there was an exception, the contents of the # exception frame (exception context) will be # printed out above the function which trapped # into the kernel. # # 5) Disassemble the code surrounding the exception # program counter. This will output the name of # the routine the system crashed in and the line # number in the source code. # # 6) Output for 'netstat -m' can be viewed by the -n # option. # # 7) Output for 'netstat -n' can be viewed by the -n # option. # # OPTIONS # Command line options are described below. # # -n The -n option turns on netstat -m and -s reports (see # netstat(1)). # # SEE ALSO # savecore(1M), dbx(1), netstat(1) # # #!/bin/sh # # "crpt: usage: crpt -n [unix.N] [vmcore.N] or N" # ############################################################################# # # # VARIABLES # # # ############################################################################# NETSTAT=n OSVER=`uname -r | /usr/bin/cut -d"." -f1` if [ $OSVER = "4" ] then USERSTRUCT="0xffffc000" PUTBUF="&putbuf/1000s" elif [ $OSVER = "5" ] then USERSTRUCT="0xffffd000" PUTBUF="putbuf/200s" else echo "Unsupported IRIX Operating System" exit -1 fi ############################################################################# # # # DEFINE FUNCTIONS # # # ############################################################################# usage() { echo "usage:" echo "crpt -n [N]" echo "crpt -n [unix.N] [vmcore.N]" echo "crpt -n [unix.N] [vmcore.N.comp]" exit -1 } cleanup() { if test -r /tmp/dbxktmp*; then /bin/rm -f /tmp/dbxktmp* fi } isreadable() { if [ ! -r $UNIX ] then echo "$UNIX does not exist or no read permission" exit fi if [ ! -r $VMCORE ] then echo "$VMCORE does not exist or no read permission" exit fi } notename() { if [ -w `pwd` ] then NOTES=Crash_${X}.rpt else NOTES=/tmp/Crash_${X}.rpt fi } dbxcheck() { if [ -x /usr/adm/crash/dbx ] then DBXPATH="/usr/adm/crash" elif [ -x /usr/bin/dbx ] then DBXPATH="/usr/bin" else echo "crpt requires dbx(1) for successful" echo "operation - see man on crpt(1)" exit -1 fi } netstats() { if [ ! -x /usr/etc/netstat ] then echo "-n option requires netstat(1) for successful" echo "operation - see man on crpt" exit fi if [ $NETSTAT = "y" ] then /usr/etc/netstat -s $UNIX $VMCORE > /tmp/dbxktmp.netstats /usr/etc/netstat -m $UNIX $VMCORE > /tmp/dbxktmp.netstatm fi } uncompress1() { if [ -r ${DUMPDIR}/vmcore.${X}.comp -a ! -r ${DUMPDIR}/vmcore.${X} ] then /etc/uncompvm ${DUMPDIR}/vmcore.${X}.comp fi } uncompress2() { if [ $VMCORE = "vmcore.${X}.comp" ] then VMCORE="vmcore.${X}" if [ ! -r vmcore.${X} ] then echo "Uncompressing" /etc/uncompvm vmcore.${X}.comp fi elif [ $VMCORE = "${DUMPDIR}/vmcore.${X}.comp" ] then VMCORE="vmcore.${X}" if [ ! -r ${DUMPDIR}/vmcore.${X} ] then echo "Uncompressing" /etc/uncompvm ${DUMPDIR}/vmcore.${X}.comp fi fi } ############################################################################# # # # CHECK INPUT ARG'S & FOR DBX # # # ############################################################################# trap 'cleanup; exit 1' 2 3 15 if [ $# = 1 -a $1 != "=n" ] then X=$1 DUMPDIR=`pwd` UNIX=${DUMPDIR}/unix.${X} uncompress1 VMCORE=${DUMPDIR}/vmcore.${X} elif [ $# = 2 ] then if [ $1 = "-n" ] then NETSTAT="y" X=$2 DUMPDIR=`pwd` UNIX=${DUMPDIR}/unix.${X} uncompress1 VMCORE=${DUMPDIR}/vmcore.${X} else UNIX=$1 VMCORE=$2 X=`basename $VMCORE | sed 's/^......\.\([0-9]*\).*/\1/'` DUMPDIR=`dirname $UNIX` uncompress2 fi elif [ $# = 3 -a $1 = "-n" ] then NETSTAT="y" UNIX=$2 VMCORE=$3 X=`basename $VMCORE | sed 's/^......\.\([0-9]*\).*/\1/'` DUMPDIR=`dirname $UNIX` uncompress2 else usage fi isreadable dbxcheck notename cleanup netstats # Next 90 lines create a ksh script that converts kernel 'time' variable # (seconds since 1 Jan 1970 Midnight GMT) into a date; telling us # when we crashed. Based on code written by Tom Murphy. Olson cat > /tmp/dbxktmp.hrs << \EOF #! /bin/ksh HourDate() { typeset -i i dow day year hour min leapday=0 hours=minutes/60 min=minutes%60 day=hours/24 hour=hours-24*day dow=day%7 case $dow in 0) weekday="Fri" ;; 1) weekday="Sat" ;; 2) weekday="Sun" ;; 3) weekday="Mon" ;; 4) weekday="Tue" ;; 5) weekday="Wed" ;; 6) weekday="Thu" ;; esac i=day/1461 year=1970+4*i day=day-1461*i #if ((day>1154)); then if ((day>1153)); then leapday=1 day=day-1 fi i=day/365 year=year+i day=day-365*i+1 if ((day<182)); then if ((day<91)); then if ((day<60)); then if ((day<32)); then month="Jan" else month="Feb" day=day-31+leapday fi else month="Mar" day=day-59 fi elif ((day<152)); then if ((day<121)); then month="Apr" day=day-90 else month="May" day=day-120 fi else month="Jun" day=day-151 fi elif ((day<274)); then if ((day<244)); then if ((day<213)); then month="Jul" day=day-181 else month="Aug" day=day-212 fi else month="Sep" day=day-243 fi elif ((day<335)); then if ((day<305)); then month="Oct" day=day-273 else month="Nov" day=day-304 fi else month="Dec" day=day-334 fi if ((min<10)); then tim=0$min else tim=$min fi if ((hour<10)); then tim=0$hour:$tim else tim=$hour:$tim fi echo "$weekday $month $day ${tim} GMT $year" } # arg is a file whose second line is the time converted to hours # see point where invoked minutes=`sed -n '$p' $1 ` HourDate EOF chmod 755 /tmp/dbxktmp.hrs # end of date conversion script # "Input_to_dbx" magic word starts input to dbx #---------------------------------------------------------------------------- # no regular output, but show errors $DBXPATH/dbx -k $UNIX $VMCORE << Input_to_dbx > /dev/null /############################################################################ / # / SET VARIALBES FOR DBX # / # /############################################################################ set \$pager="cat" set \$prompt="" set \$pagewindow=1000 /############################################################################ / # / PRINT THE KERNEL PUTBUF # / # /############################################################################ record output /tmp/dbxktmp.putbuf $PUTBUF unrecord all /############################################################################ / # / PRINT CORE FILE NAME & SYSTEM INFO # / # /############################################################################ record output /tmp/dbxktmp.date p time div 60 unrecord all record output /tmp/dbxktmp.utsname givenfile ;corefile print *(struct utsname *)&utsname printf "system crashed at %#x seconds from Jan 1, 1970\nor about: ", time sh /tmp/dbxktmp.hrs /tmp/dbxktmp.date >> /tmp/dbxktmp.utsname unrecord all /############################################################################ / # / IF RUNNING ON THE SYSTEM WHICH CRASHED - GET HINV, VERSIONS # / # /############################################################################ sh sh -c 'fgrep nodename /tmp/dbxktmp.utsname| tr -d \" | cut -f7 -d" " \ > /tmp/dbxktmp.name' sh sh -c "if test \`uname -n\` = \`cat /tmp/dbxktmp.name\`;then versions -b > \ /tmp/dbxktmp.versions; hinv > /tmp/dbxktmp.hinv; fi" /############################################################################ / # / NAME AND PID OF CURRENTLY RUNNING PROCESS # / # /############################################################################ record output /tmp/dbxktmp.psargs print ((struct user *) $USERSTRUCT).u_psargs unrecord all record output /tmp/dbxktmp.pid print \$pid unrecord all sh echo "PID:" > /tmp/dbxktmp.pid1 sh cat /tmp/dbxktmp.pid >> /tmp/dbxktmp.pid1 sh cat /tmp/dbxktmp.pid1 | fgrep -v record | tr -d "\012" >> /tmp/dbxktmp.pid2 sh echo "\n" >> /tmp/dbxktmp.pid2 /############################################################################ / # / PRINT STACK POINTER, FRAME POINTER AND PROGRAM COUNTER # / PRINT A KERNEL STACK TRACE # / # /############################################################################ # more or less duplicate the message that dbx prints on startup # s8 is the frame ptr record output /tmp/dbxktmp.trace printf "SP=%x fp=%x pc=%x\n", \$sp, \$s8, \$pc set \$print_exception_frame=1 t set \$print_exception_frame=0 unrecord all /############################################################################ / # / GET THE EXCEPTION FRAME POINTER FROM THE PUTBUF # / SAVE THE EXCEPTION PROGRAM COUNTER # / # /############################################################################ set \$hexints=1 sh sh -c 'fgrep ep: /tmp/dbxktmp.putbuf | sed -e "s/^.*ep:/ep:/" > \ /tmp/dbxktmp.epptr; \ if grep -s "ep:0x" /tmp/dbxktmp.epptr; \ then cat /tmp/dbxktmp.epptr | cut -c1-13 | \ sed -e "s/ep:/\(\(struct eframe_s \*\)/" | \ sed -e "s/$/\)\.ef_epc-0x10\/20i/" > /tmp/dbxktmp.trace0 ;\ else cat /tmp/dbxktmp.epptr | cut -c1-11 | \ sed -e "s/ep:/\(\(struct eframe_s \*\)0x/" | \ sed -e "s/$/\)\.ef_epc-0x10\/20i/" > /tmp/dbxktmp.trace0 ;\ fi' /############################################################################ / # / DISASSEMBLE THE CODE NEAR THE EXCEPTION PC # / # /############################################################################ record output /tmp/dbxktmp.epc_dis playback input /tmp/dbxktmp.trace0 unrecord all quit Input_to_dbx # "Input_to_dbx" magic word ends sending input to dbx #---------------------------------------------------------------------------- # cleanup and polish. ############################################################################# # # # FORMAT A REPORT ON THE KERNEL CORE DUMP # # SAVE AS "Crash_N.rpt" # # # ############################################################################# if test -s /tmp/dbxktmp.utsname then echo '____ System Info _______________________________________________________________\n' \ > /tmp/dbxktmp.out;fgrep -v "{" /tmp/dbxktmp.utsname | fgrep -v "}" \ | fgrep -v \"\" \ | sed 's/and its processid is 0//' >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.hinv then echo '\n____ Hardware Inventory ________________________________________________________\n' \ >> /tmp/dbxktmp.out;cat /tmp/dbxktmp.hinv >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.versions then echo '\n____ Versions __________________________________________________________________\n' \ >> /tmp/dbxktmp.out;cat /tmp/dbxktmp.versions >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.putbuf then echo '\n____ Putbuf Dump _______________________________________________________________\n' \ >> /tmp/dbxktmp.out; fgrep -v \"\" /tmp/dbxktmp.putbuf | sed -e '/^"\.*"$/d' -e 's/\\n/\ /g' | sed 's/\\t/\ /' | tr -d \"| uniq >> /tmp/dbxktmp.out #^ This is not a formatting error, it's just a cludge to replace literal "\n" with \012 fi if test -s /tmp/dbxktmp.psargs then echo '\n____ Program Running ___________________________________________________________\n' \ >> /tmp/dbxktmp.out;cat /tmp/dbxktmp.psargs /tmp/dbxktmp.pid2 >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.trace then echo '\n____ Trace & Exception Frame Registers _________________________________________\n' \ >> /tmp/dbxktmp.out cat /tmp/dbxktmp.trace >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.epc_dis then EPC=`cat /tmp/dbxktmp.epc_dis | wc -l | tr -d " "` if [ $EPC -gt "12" ] then echo '\n____ Exception PC Disassembled _________________________________________________\n' \ >> /tmp/dbxktmp.out;cat /tmp/dbxktmp.epc_dis >> /tmp/dbxktmp.out fi fi if test -s /tmp/dbxktmp.netstatm then echo '\n____ Netstat -m ________________________________________________________________\n' \ >> /tmp/dbxktmp.out cat /tmp/dbxktmp.netstatm >> /tmp/dbxktmp.out fi if test -s /tmp/dbxktmp.netstats then echo '\n____ Netstat -s ________________________________________________________________\n' \ >> /tmp/dbxktmp.out cat /tmp/dbxktmp.netstats >> /tmp/dbxktmp.out fi # Clean up the final report: if test -r ${NOTES} then /bin/mv ${NOTES} ${NOTES}.old fi egrep -v 'record|reference data|zero repetition count' /tmp/dbxktmp.out > $NOTES echo "Generated crash report - $NOTES" # Remove all tempory files cleanup exit 0 # ignore trailing signature stuff