mariadb/mysql-test/lib/My/CoreDump.pm
Vladislav Vaintroub f812f8e1ab MDEV-30475 Windows, mtr - Remove outdated instructions on how to install post-mortem debugger
Also, use standard C:\symbols location for OS debugging symbols cache,
rather than own invention C:\cdb_symbols.
2023-01-26 12:32:15 +01:00

516 lines
14 KiB
Perl

# -*- cperl -*-
# Copyright (c) 2008, 2013, Oracle and/or its affiliates. All rights reserved.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; version 2 of the License.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335 USA
package My::CoreDump;
use strict;
use Carp;
use My::Platform;
use Text::Wrap;
use Data::Dumper;
use File::Temp qw/ tempfile tempdir /;
use File::Find;
use File::Basename;
use mtr_results;
use mtr_report;
my %opts;
my %config;
my $help = "\n\nOptions for printing core dumps\n\n";
sub register_opt($$$) {
my ($name, $format, $msg)= @_;
my @names= split(/\|/, $name);
my $option_name= $names[0];
$option_name=~ s/-/_/;
$opts{$name. $format}= \$config{$option_name};
$help.= wrap(sprintf(" %-23s", join(', ', @names)), ' 'x25, "$msg\n");
}
# To preserve order we use array instead of hash
my @print_formats= (
short => {
description => "Failing stack trace",
codes => {}
},
medium => {
description => "All stack traces",
codes => {}
},
detailed => {
description => "All stack traces with debug context",
codes => {}
},
custom => {
description => "Custom debugger script for printing stack"
},
# 'no' must be last (check generated help)
no => {
description => "Skip stack trace printing"
}
);
# TODO: make class for each {method, get_code}
my @print_methods= (IS_WINDOWS) ? (cdb => { method => \&_cdb }) : (
gdb => {
method => \&_gdb,
get_code => \&_gdb_format,
},
dbx => {
method => \&_dbx
},
lldb => {
method => \&_lldb
},
# 'auto' must be last (check generated help)
auto => {
method => \&_auto
}
);
# But we also use hash
my %print_formats= @print_formats;
my %print_methods= @print_methods;
# and scalar
my $x= 0;
my $print_formats= join(', ', grep { ++$x % 2 } @print_formats);
$x= 0;
my $print_methods= join(', ', grep { ++$x % 2 } @print_methods);
# Fill 'short' and 'detailed' formats per each print_method
# that has interface for that
for my $f (keys %print_formats)
{
next unless exists $print_formats{$f}->{codes};
for my $m (keys %print_methods)
{
next unless exists $print_methods{$m}->{get_code};
# That calls f.ex. _gdb_format('short')
# and assigns { gdb => value-of-_gdb_format } into $print_formats{short}->{format}:
$print_formats{$f}->{codes}->{$m}= $print_methods{$m}->{get_code}->($f);
}
}
register_opt('print-core|C', ':s',
"Print core dump format: ". $print_formats. " (for not printing cores). ".
"Defaults to value of MTR_PRINT_CORE or 'medium'");
if (!IS_WINDOWS)
{
register_opt('print-method', '=s',
"Print core method: ". join(', ', $print_methods). " (try each method until success). ".
"Defaults to 'auto'");
}
sub options() { %opts }
sub help() { $help }
sub env_or_default($$) {
my ($default, $env)= @_;
if (exists $ENV{$env}) {
my $f= $ENV{$env};
$f= 'custom'
if $f =~ m/^custom:/;
return $ENV{$env}
if exists $print_formats{$f};
mtr_verbose("$env value ignored: $ENV{$env}");
}
return $default;
}
sub pre_setup() {
$config{print_core}= env_or_default('medium', 'MTR_PRINT_CORE')
if not defined $config{print_core};
$config{print_method}= (IS_WINDOWS) ? 'cdb' : 'auto'
if not defined $config{print_method};
# If the user has specified 'custom' we fill appropriate print_format
# and that will be used automatically
# Note: this can assign 'custom' to method 'auto'.
if ($config{print_core} =~ m/^custom:(.+)$/) {
$config{print_core}= 'custom';
$print_formats{'custom'}= {
$config{print_method} => $1
}
}
mtr_error "Wrong value for --print-core: $config{print_core}"
if not exists $print_formats{$config{print_core}};
mtr_error "Wrong value for --print-method: $config{print_method}"
if not exists $print_methods{$config{print_method}};
mtr_debug(Data::Dumper->Dump(
[\%config, \%print_formats, \%print_methods],
[qw(config print_formats print_methods)]));
}
my $hint_mysqld; # Last resort guess for executable path
# If path in core file is 79 chars we assume it's been truncated
# Looks like we can still find the full path using 'strings'
# If that doesn't work, use the hint (mysqld path) as last resort.
sub _verify_binpath {
my ($binary, $core_name)= @_;
my $binpath;
if (length $binary != 79) {
$binpath= $binary;
print "Core generated by '$binpath'\n";
} else {
# Last occurrence of path ending in /mysql*, cut from first /
if (`strings '$core_name' | grep "/mysql[^/. ]*\$" | tail -1` =~ /(\/.*)/) {
$binpath= $1;
print "Guessing that core was generated by '$binpath'\n";
} else {
return unless $hint_mysqld;
$binpath= $hint_mysqld;
print "Wild guess that core was generated by '$binpath'\n";
}
}
return $binpath;
}
# Returns GDB code according to specified format
# Note: this is like simple hash, separate interface was made
# in advance for implementing below TODO
# TODO: _gdb_format() and _gdb() should be separate class
# (like the other printing methods)
sub _gdb_format($) {
my ($format)= @_;
my %formats= (
short => "bt\n",
medium => "thread apply all bt\n",
detailed =>
"bt\n".
"set print sevenbit on\n".
"set print static-members off\n".
"set print frame-arguments all\n".
"thread apply all bt full\n".
"quit\n"
);
confess "Unknown format: ". $format
unless exists $formats{$format};
return $formats{$format};
}
sub _gdb {
my ($core_name, $code)= @_;
confess "Undefined format"
unless defined $code;
# Check that gdb exists
`gdb --version`;
if ($?) {
print "gdb not found, cannot get the stack trace\n";
return;
}
if (-f $core_name) {
mtr_verbose("Trying 'gdb' to get a backtrace from coredump $core_name");
} else {
print "\nCoredump $core_name does not exist, cannot run 'gdb'\n";
return;
}
# Find out name of binary that generated core
`gdb -c '$core_name' --batch 2>&1` =~
/Core was generated by `([^\s\'\`]+)/;
my $binary= $1 or return;
$binary= _verify_binpath ($binary, $core_name) or return;
# Create tempfile containing gdb commands
my ($tmp, $tmp_name) = tempfile();
print $tmp $code;
close $tmp or die "Error closing $tmp_name: $!";
# Run gdb
my $gdb_output=
`gdb '$binary' -c '$core_name' -x '$tmp_name' --batch 2>&1`;
unlink $tmp_name or die "Error removing $tmp_name: $!";
return if $? >> 8;
return unless $gdb_output;
resfile_print <<EOF . $gdb_output . "\n";
Output from gdb follows. The first stack trace is from the failing thread.
The following stack traces are from all threads (so the failing one is
duplicated).
--------------------------
EOF
return 1;
}
sub _dbx {
my ($core_name, $format)= @_;
print "\nTrying 'dbx' to get a backtrace\n";
return unless -f $core_name;
# Find out name of binary that generated core
`echo | dbx - '$core_name' 2>&1` =~
/Corefile specified executable: "([^"]+)"/;
my $binary= $1 or return;
$binary= _verify_binpath ($binary, $core_name) or return;
# Find all threads
my @thr_ids = `echo threads | dbx '$binary' '$core_name' 2>&1` =~ /t@\d+/g;
# Create tempfile containing dbx commands
my ($tmp, $tmp_name) = tempfile();
foreach my $thread (@thr_ids) {
print $tmp "where $thread\n";
}
print $tmp "exit\n";
close $tmp or die "Error closing $tmp_name: $!";
# Run dbx
my $dbx_output=
`cat '$tmp_name' | dbx '$binary' '$core_name' 2>&1`;
unlink $tmp_name or die "Error removing $tmp_name: $!";
return if $? >> 8;
return unless $dbx_output;
resfile_print <<EOF . $dbx_output . "\n";
Output from dbx follows. Stack trace is printed for all threads in order,
above this you should see info about which thread was the failing one.
----------------------------
EOF
return 1;
}
# Check that Debugging tools for Windows are installed
sub cdb_check {
`cdb -? 2>&1`;
if ($? >> 8)
{
print "Cannot find the cdb debugger. Please install Debugging tools for Windows\n";
print "and set PATH environment variable to include location of cdb.exe";
}
}
sub _cdb {
my ($core_name, $format)= @_;
print "\nTrying 'cdb' to get a backtrace\n";
return unless -f $core_name;
# Read module list, find out the name of executable and
# build symbol path (required by cdb if executable was built on
# different machine)
my $tmp_name= $core_name.".cdb_lmv";
`cdb -z $core_name -c \"lmv;q\" > $tmp_name 2>&1`;
if ($? >> 8)
{
unlink($tmp_name);
# check if cdb is installed and complain if not
cdb_check();
return;
}
open(temp,"< $tmp_name");
my %dirhash=();
while(<temp>)
{
if($_ =~ /Image path\: (.*)/)
{
if (rindex($1,'\\') != -1)
{
my $dir= substr($1, 0, rindex($1,'\\'));
$dirhash{$dir}++;
}
}
}
close(temp);
unlink($tmp_name);
my $image_path= join(";", (keys %dirhash),".");
# For better callstacks, setup _NT_SYMBOL_PATH to include
# OS symbols. Note : Dowloading symbols for the first time
# can take some minutes
if (!$ENV{'_NT_SYMBOL_PATH'})
{
my $windir= $ENV{'windir'};
my $symbol_cache= substr($windir ,0, index($windir,'\\'))."\\symbols";
print "OS debug symbols will be downloaded and stored in $symbol_cache.\n";
print "You can control the location of symbol cache with _NT_SYMBOL_PATH\n";
print "environment variable. Please refer to Microsoft KB article\n";
print "http://support.microsoft.com/kb/311503 for details about _NT_SYMBOL_PATH\n";
print "-------------------------------------------------------------------------\n";
$ENV{'_NT_SYMBOL_PATH'}.=
"srv*".$symbol_cache."*http://msdl.microsoft.com/download/symbols";
}
my $symbol_path= $image_path.";".$ENV{'_NT_SYMBOL_PATH'};
# Run cdb. Use "analyze" extension to print crashing thread stacktrace
# and "uniqstack" to print other threads
my $cdb_cmd = "!sym prompts off; !analyze -v; .ecxr; !for_each_frame dv /t;!uniqstack -p;q";
my $cdb_output=
`cdb -c "$cdb_cmd" -z $core_name -i "$image_path" -y "$symbol_path" -t 0 -lines 2>&1`;
return if $? >> 8;
return unless $cdb_output;
# Remove comments (lines starting with *), stack pointer and frame
# pointer adresses and offsets to function to make output better readable
$cdb_output=~ s/^\*.*\n//gm;
$cdb_output=~ s/^([\:0-9a-fA-F\`]+ )+//gm;
$cdb_output=~ s/^ChildEBP RetAddr//gm;
$cdb_output=~ s/^Child\-SP RetAddr Call Site//gm;
$cdb_output=~ s/\+0x([0-9a-fA-F]+)//gm;
resfile_print <<EOF . $cdb_output . "\n";
Output from cdb follows. Faulting thread is printed twice,with and without function parameters
Search for STACK_TEXT to see the stack trace of
the faulting thread. Callstacks of other threads are printed after it.
EOF
return 1;
}
sub _lldb
{
my ($core_name)= @_;
print "\nTrying 'lldb' to get a backtrace from coredump $core_name\n";
# Create tempfile containing lldb commands
my ($tmp, $tmp_name)= tempfile();
print $tmp
"bt\n",
"thread backtrace all\n",
"quit\n";
close $tmp or die "Error closing $tmp_name: $!";
my $lldb_output= `lldb -c '$core_name' -s '$tmp_name' 2>&1`;
unlink $tmp_name or die "Error removing $tmp_name: $!";
if ($? == 127)
{
print "lldb not found, cannot get the stack trace\n";
return;
}
return if $?;
return unless $lldb_output;
resfile_print <<EOF . $lldb_output . "\n";
Output from lldb follows. The first stack trace is from the failing thread.
The following stack traces are from all threads (so the failing one is
duplicated).
--------------------------
EOF
return 1;
}
sub _auto
{
my ($core_name, $code, $rest)= @_;
# We use ordered array @print_methods and omit auto itself
my @valid_methods= @print_methods[0 .. $#print_methods - 2];
my $x= 0;
my @methods= grep { ++$x % 2} @valid_methods;
my $f= $config{print_core};
foreach my $m (@methods)
{
my $debugger= $print_methods{$m};
confess "Broken @print_methods"
if $debugger->{method} == \&_auto;
# If we didn't find format for 'auto' (that is only possible for 'custom')
# we get format for specific debugger
if (not defined $code && defined $print_formats{$f} and
exists $print_formats{$f}->{codes}->{$m})
{
$code= $print_formats{$f}->{codes}->{$m};
}
mtr_verbose2("Trying to print with method ${m}:${f}");
if ($debugger->{method}->($core_name, $code)) {
return;
}
}
}
sub show {
my ($core_name, $exe_mysqld, $parallel)= @_;
if ($config{print_core} ne 'no') {
my $f= $config{print_core};
my $m= $config{print_method};
my $code= undef;
if (exists $print_formats{$f}->{codes} and
exists $print_formats{$f}->{codes}->{$m}) {
$code= $print_formats{$f}->{codes}->{$m};
}
mtr_verbose2("Printing core with method ${m}:${f}");
mtr_debug("code: ${code}");
$print_methods{$m}->{method}->($core_name, $code);
}
return;
}
sub core_wanted($$$$$) {
my ($num_saved_cores, $opt_max_save_core, $compress,
$exe_mysqld, $opt_parallel)= @_;
my $core_file= $File::Find::name;
my $core_name= basename($core_file);
# Name beginning with core, not ending in .gz
if (($core_name =~ /^core/ and $core_name !~ /\.gz$/)
or (IS_WINDOWS and $core_name =~ /\.dmp$/))
{
# Ending with .dmp
mtr_report(" - found '$core_name'",
"($$num_saved_cores/$opt_max_save_core)");
show($core_file, $exe_mysqld, $opt_parallel);
# Limit number of core files saved
if ($$num_saved_cores >= $opt_max_save_core)
{
mtr_report(" - deleting it, already saved",
"$opt_max_save_core");
unlink("$core_file");
}
else
{
main::mtr_compress_file($core_file) if $compress;
++$$num_saved_cores;
}
}
}
1;