Tue, 06 Jan 2015 21:39:09 +0100
Conditionally force memory storage according to privacy.thirdparty.isolate;
This solves Tor bug #9701, complying with disk avoidance documented in
https://www.torproject.org/projects/torbrowser/design/#disk-avoidance.
michael@0 | 1 | #!/usr/bin/perl |
michael@0 | 2 | # vim:sw=4:ts=4:et: |
michael@0 | 3 | # This Source Code Form is subject to the terms of the Mozilla Public |
michael@0 | 4 | # License, v. 2.0. If a copy of the MPL was not distributed with this |
michael@0 | 5 | # file, You can obtain one at http://mozilla.org/MPL/2.0/. |
michael@0 | 6 | |
michael@0 | 7 | # $Id: fix-linux-stack.pl,v 1.16 2008/05/05 21:51:11 dbaron%dbaron.org Exp $ |
michael@0 | 8 | # |
michael@0 | 9 | # This script uses addr2line (part of binutils) to process the output of |
michael@0 | 10 | # nsTraceRefcnt's Linux stack walking code. This is useful for two |
michael@0 | 11 | # things: |
michael@0 | 12 | # (1) Getting line number information out of |
michael@0 | 13 | # |nsTraceRefcnt::WalkTheStack|'s output in debug builds. |
michael@0 | 14 | # (2) Getting function names out of |nsTraceRefcnt::WalkTheStack|'s |
michael@0 | 15 | # output on optimized builds (where it mostly prints UNKNOWN |
michael@0 | 16 | # because only a handful of symbols are exported from component |
michael@0 | 17 | # libraries). |
michael@0 | 18 | # |
michael@0 | 19 | # Use the script by piping output containing stacks (such as raw stacks |
michael@0 | 20 | # or make-tree.pl balance trees) through this script. |
michael@0 | 21 | |
michael@0 | 22 | use strict; |
michael@0 | 23 | use IPC::Open2; |
michael@0 | 24 | use File::Basename; |
michael@0 | 25 | |
michael@0 | 26 | # XXX Hard-coded to gdb defaults (works on Fedora). |
michael@0 | 27 | my $global_debug_dir = '/usr/lib/debug'; |
michael@0 | 28 | |
michael@0 | 29 | # We record several things for each file encountered. |
michael@0 | 30 | # |
michael@0 | 31 | # - {pipe_read}, {pipe_write}: these constitute a bidirectional pipe to an |
michael@0 | 32 | # addr2line process that gives symbol information for a file. |
michael@0 | 33 | # |
michael@0 | 34 | # - {cache}: this table holds the results of lookups that we've done |
michael@0 | 35 | # previously for (pre-adjustment) addresses, which lets us avoid redundant |
michael@0 | 36 | # calls to addr2line. |
michael@0 | 37 | # |
michael@0 | 38 | # - {address_adjustment}: addr2line wants offsets relative to the base address |
michael@0 | 39 | # for shared libraries, but it wants addresses including the base address |
michael@0 | 40 | # offset for executables. This holds the appropriate address adjustment to |
michael@0 | 41 | # add to an offset within file. See bug 230336. |
michael@0 | 42 | # |
michael@0 | 43 | my %file_infos; |
michael@0 | 44 | |
michael@0 | 45 | sub set_address_adjustment($$) { |
michael@0 | 46 | my ($file, $file_info) = @_; |
michael@0 | 47 | |
michael@0 | 48 | # find out if it's an executable (as opposed to a shared library) |
michael@0 | 49 | my $elftype; |
michael@0 | 50 | open(ELFHDR, '-|', 'readelf', '-h', $file); |
michael@0 | 51 | while (<ELFHDR>) { |
michael@0 | 52 | if (/^\s*Type:\s+(\S+)/) { |
michael@0 | 53 | $elftype = $1; |
michael@0 | 54 | last; |
michael@0 | 55 | } |
michael@0 | 56 | } |
michael@0 | 57 | close(ELFHDR); |
michael@0 | 58 | |
michael@0 | 59 | # If it's an executable, make adjustment the base address. |
michael@0 | 60 | # Otherwise, leave it zero. |
michael@0 | 61 | my $adjustment = 0; |
michael@0 | 62 | if ($elftype eq 'EXEC') { |
michael@0 | 63 | open(ELFSECS, '-|', 'readelf', '-S', $file); |
michael@0 | 64 | while (<ELFSECS>) { |
michael@0 | 65 | if (/^\s*\[\s*\d+\]\s+\.text\s+\w+\s+(\w+)\s+(\w+)\s+/) { |
michael@0 | 66 | # Subtract the .text section's offset within the |
michael@0 | 67 | # file from its base address. |
michael@0 | 68 | $adjustment = hex($1) - hex($2); |
michael@0 | 69 | last; |
michael@0 | 70 | } |
michael@0 | 71 | } |
michael@0 | 72 | close(ELFSECS); |
michael@0 | 73 | } |
michael@0 | 74 | |
michael@0 | 75 | $file_info->{address_adjustment} = $adjustment; |
michael@0 | 76 | } |
michael@0 | 77 | |
michael@0 | 78 | # Files sometimes contain a link to a separate object file that contains |
michael@0 | 79 | # the debug sections of the binary, removed so that a smaller file can |
michael@0 | 80 | # be shipped, but kept separately so that it can be obtained by those |
michael@0 | 81 | # who want it. |
michael@0 | 82 | # See http://sources.redhat.com/gdb/current/onlinedocs/gdb_16.html#SEC154 |
michael@0 | 83 | # for documentation of debugging information in separate files. |
michael@0 | 84 | # On Fedora distributions, these files can be obtained by installing |
michael@0 | 85 | # *-debuginfo RPM packages. |
michael@0 | 86 | sub separate_debug_file_for($) { |
michael@0 | 87 | my ($file) = @_; |
michael@0 | 88 | # We can read the .gnu_debuglink section using either of: |
michael@0 | 89 | # objdump -s --section=.gnu_debuglink $file |
michael@0 | 90 | # readelf -x .gnu_debuglink $file |
michael@0 | 91 | # Since readelf prints things backwards on little-endian platforms |
michael@0 | 92 | # for some versions only (backwards on Fedora Core 6, forwards on |
michael@0 | 93 | # Fedora 7), use objdump. |
michael@0 | 94 | |
michael@0 | 95 | # See if there's a .gnu_debuglink section |
michael@0 | 96 | my $have_debuglink = 0; |
michael@0 | 97 | open(ELFSECS, '-|', 'readelf', '-S', $file); |
michael@0 | 98 | while (<ELFSECS>) { |
michael@0 | 99 | if (/^\s*\[\s*\d+\]\s+\.gnu_debuglink\s+\w+\s+(\w+)\s+(\w+)\s+/) { |
michael@0 | 100 | $have_debuglink = 1; |
michael@0 | 101 | last; |
michael@0 | 102 | } |
michael@0 | 103 | } |
michael@0 | 104 | close(ELFSECS); |
michael@0 | 105 | return '' unless ($have_debuglink); |
michael@0 | 106 | |
michael@0 | 107 | # Determine the endianness of the shared library. |
michael@0 | 108 | my $endian = ''; |
michael@0 | 109 | open(ELFHDR, '-|', 'readelf', '-h', $file); |
michael@0 | 110 | while (<ELFHDR>) { |
michael@0 | 111 | if (/^\s*Data:\s+.*(little|big) endian.*$/) { |
michael@0 | 112 | $endian = $1; |
michael@0 | 113 | last; |
michael@0 | 114 | } |
michael@0 | 115 | } |
michael@0 | 116 | close(ELFHDR); |
michael@0 | 117 | if ($endian ne 'little' && $endian ne 'big') { |
michael@0 | 118 | print STDERR "Warning: could not determine endianness of $file.\n"; |
michael@0 | 119 | return ''; |
michael@0 | 120 | } |
michael@0 | 121 | |
michael@0 | 122 | |
michael@0 | 123 | # Read the debuglink section as an array of words, in hexidecimal. |
michael@0 | 124 | open(DEBUGLINK, '-|', 'objdump', '-s', '--section=.gnu_debuglink', $file); |
michael@0 | 125 | my @words; |
michael@0 | 126 | while (<DEBUGLINK>) { |
michael@0 | 127 | if ($_ =~ /^ [0-9a-f]* ([0-9a-f ]{8}) ([0-9a-f ]{8}) ([0-9a-f ]{8}) ([0-9a-f ]{8}).*/) { |
michael@0 | 128 | push @words, $1, $2, $3, $4; |
michael@0 | 129 | } |
michael@0 | 130 | } |
michael@0 | 131 | close(DEBUGLINK); |
michael@0 | 132 | |
michael@0 | 133 | while (@words[$#words] eq ' ') { |
michael@0 | 134 | pop @words; |
michael@0 | 135 | } |
michael@0 | 136 | |
michael@0 | 137 | if ($#words < 1) { |
michael@0 | 138 | print STDERR "Warning: .gnu_debuglink section in $file too short.\n"; |
michael@0 | 139 | return ''; |
michael@0 | 140 | } |
michael@0 | 141 | |
michael@0 | 142 | my @chars; |
michael@0 | 143 | while ($#words >= 0) { |
michael@0 | 144 | my $w = shift @words; |
michael@0 | 145 | if ($w =~ /^([0-9a-f]{2})([0-9a-f]{2})([0-9a-f]{2})([0-9a-f]{2})$/) { |
michael@0 | 146 | push @chars, $1, $2, $3, $4; |
michael@0 | 147 | } else { |
michael@0 | 148 | print STDERR "Warning: malformed objdump output for $file.\n"; |
michael@0 | 149 | return ''; |
michael@0 | 150 | } |
michael@0 | 151 | } |
michael@0 | 152 | |
michael@0 | 153 | my @hash_bytes = map(hex, @chars[$#chars - 3 .. $#chars]); |
michael@0 | 154 | $#chars -= 4; |
michael@0 | 155 | |
michael@0 | 156 | my $hash; |
michael@0 | 157 | if ($endian eq 'little') { |
michael@0 | 158 | $hash = ($hash_bytes[3] << 24) | ($hash_bytes[2] << 16) | ($hash_bytes[1] << 8) | $hash_bytes[0]; |
michael@0 | 159 | } else { |
michael@0 | 160 | $hash = ($hash_bytes[0] << 24) | ($hash_bytes[1] << 16) | ($hash_bytes[2] << 8) | $hash_bytes[3]; |
michael@0 | 161 | } |
michael@0 | 162 | |
michael@0 | 163 | # The string ends with a null-terminator and then 0 to three bytes |
michael@0 | 164 | # of padding to fill the current 32-bit unit. (This padding is |
michael@0 | 165 | # usually null bytes, but I've seen null-null-H, on Ubuntu x86_64.) |
michael@0 | 166 | my $terminator = 1; |
michael@0 | 167 | while ($chars[$terminator] ne '00') { |
michael@0 | 168 | if ($terminator == $#chars) { |
michael@0 | 169 | print STDERR "Warning: missing null terminator in " . |
michael@0 | 170 | ".gnu_debuglink section of $file.\n"; |
michael@0 | 171 | return ''; |
michael@0 | 172 | } |
michael@0 | 173 | ++$terminator; |
michael@0 | 174 | } |
michael@0 | 175 | if ($#chars - $terminator > 3) { |
michael@0 | 176 | print STDERR "Warning: Excess padding in .gnu_debuglink section " . |
michael@0 | 177 | "of $file.\n"; |
michael@0 | 178 | return ''; |
michael@0 | 179 | } |
michael@0 | 180 | $#chars = $terminator - 1; |
michael@0 | 181 | |
michael@0 | 182 | my $basename = join('', map { chr(hex($_)) } @chars); |
michael@0 | 183 | |
michael@0 | 184 | # Now $basename and $hash represent the information in the |
michael@0 | 185 | # .gnu_debuglink section. |
michael@0 | 186 | #printf STDERR "%x: %s\n", $hash, $basename; |
michael@0 | 187 | |
michael@0 | 188 | my @possible_results = ( |
michael@0 | 189 | dirname($file) . $basename, |
michael@0 | 190 | dirname($file) . '.debug/' . $basename, |
michael@0 | 191 | $global_debug_dir . dirname($file) . '/' . $basename |
michael@0 | 192 | ); |
michael@0 | 193 | foreach my $result (@possible_results) { |
michael@0 | 194 | if (-f $result) { |
michael@0 | 195 | # XXX We should check the hash. |
michael@0 | 196 | return $result; |
michael@0 | 197 | } |
michael@0 | 198 | } |
michael@0 | 199 | |
michael@0 | 200 | return ''; |
michael@0 | 201 | } |
michael@0 | 202 | |
michael@0 | 203 | sub get_file_info($) { |
michael@0 | 204 | my ($file) = @_; |
michael@0 | 205 | my $file_info = $file_infos{$file}; |
michael@0 | 206 | unless (defined $file_info) { |
michael@0 | 207 | my $debug_file = separate_debug_file_for($file); |
michael@0 | 208 | $debug_file = $file if ($debug_file eq ''); |
michael@0 | 209 | |
michael@0 | 210 | my $pid = open2($file_info->{pipe_read}, $file_info->{pipe_write}, |
michael@0 | 211 | '/usr/bin/addr2line', '-C', '-f', '-e', $debug_file); |
michael@0 | 212 | |
michael@0 | 213 | set_address_adjustment($file, $file_info); |
michael@0 | 214 | |
michael@0 | 215 | $file_infos{$file} = $file_info; |
michael@0 | 216 | } |
michael@0 | 217 | return $file_info; |
michael@0 | 218 | } |
michael@0 | 219 | |
michael@0 | 220 | # Ignore SIGPIPE as a workaround for addr2line crashes in some situations. |
michael@0 | 221 | $SIG{PIPE} = 'IGNORE'; |
michael@0 | 222 | |
michael@0 | 223 | select STDOUT; $| = 1; # make STDOUT unbuffered |
michael@0 | 224 | while (<>) { |
michael@0 | 225 | my $line = $_; |
michael@0 | 226 | if ($line =~ /^([ \|0-9-]*)(.*) ?\[([^ ]*) \+(0x[0-9A-F]{1,8})\](.*)$/) { |
michael@0 | 227 | my $before = $1; # allow preservation of balance trees |
michael@0 | 228 | my $badsymbol = $2; |
michael@0 | 229 | my $file = $3; |
michael@0 | 230 | my $address = hex($4); |
michael@0 | 231 | my $after = $5; # allow preservation of counts |
michael@0 | 232 | |
michael@0 | 233 | if (-f $file) { |
michael@0 | 234 | my $file_info = get_file_info($file); |
michael@0 | 235 | my $result = $file_info->{cache}->{$address}; |
michael@0 | 236 | if (not defined $result) { |
michael@0 | 237 | my $address2 = $address + $file_info->{address_adjustment}; |
michael@0 | 238 | my $out = $file_info->{pipe_write}; |
michael@0 | 239 | my $in = $file_info->{pipe_read}; |
michael@0 | 240 | printf {$out} "0x%X\n", $address2; |
michael@0 | 241 | chomp(my $symbol = <$in>); |
michael@0 | 242 | chomp(my $fileandline = <$in>); |
michael@0 | 243 | if (!$symbol || $symbol eq '??') { $symbol = $badsymbol; } |
michael@0 | 244 | if (!$fileandline || $fileandline eq '??:0') { |
michael@0 | 245 | $fileandline = $file; |
michael@0 | 246 | } |
michael@0 | 247 | $result = "$symbol ($fileandline)"; |
michael@0 | 248 | $file_info->{cache}->{$address} = $result; |
michael@0 | 249 | } |
michael@0 | 250 | print "$before$result$after\n"; |
michael@0 | 251 | } else { |
michael@0 | 252 | print STDERR "Warning: File \"$file\" does not exist.\n"; |
michael@0 | 253 | print $line; |
michael@0 | 254 | } |
michael@0 | 255 | |
michael@0 | 256 | } else { |
michael@0 | 257 | print $line; |
michael@0 | 258 | } |
michael@0 | 259 | } |