Bug #1289
Infinite loop on command exiv2 -pR on a .CR2 file
100%
Description
This has been reported by Asdiel (AlienSkin Software). Test file at http://exiv2.dyndns.org:8080/userContent/testfiles/1289/IMG_1538.CR2
Associated revisions
History
Updated by Robin Mills over 4 years ago
- Category set to image format
- Assignee set to Ben Touchette
- Target version set to 0.26
- % Done changed from 0 to 100
- Estimated time set to 3.00 h
Fix submitted r4755. Thank You to Ben for providing the patch.
Ben: I shortened your fix by 4 lines, however it is your fix. Thanks for working on this.
Updated by Robin Mills over 4 years ago
A couple of additional observations about this:
1) Why is this looping when TiffVisitor reads it successfully?
I should investigate to discover why printStructure() dies on this file and TiffVisitor does not.
2) Why are we frequently running exiv2 -pR for no good reason?
Asdiel has pointed out that we shouldn't be executing Image::printStructure(kpsRecursive) on files which are not remote. This code is only required by RemoteIO objects to ensure that the metadata is in memory for TiffVisitor.
I have proposed method BasicIo::isRemote() which would return true for remote data sources. We could add the following to Image::printIFDStructure() after the first write:
if ( !out.good() && !io->isRemote() ) return;
This says skip printIFDStructure() on a stream attached to /dev/null when the IO source is remote.
It’s easy to define isRemote() for files using protocols such as http/ftp/ssh. However local files which are network shares/mounts (afp, nfs, samba) require investigation.
Updated by Robin Mills over 4 years ago
I've investigated the subject of how to determine if a file is on local storage or a network share. On Windows there is an API GetDeviceType(path) which looks like our man! https://msdn.microsoft.com/en-us/library/windows/desktop/aa364939(v=vs.85).aspx
On Mac/Linux, one solution is to call realpath (see man 3 realpath). If realpath starts with /media (Linux) or /Volumes (MacOS-X), it's a network share.
So the default implementation of BasicIo::isRemote() is to return false. For class RemoteIo (and derived classes), isRemote() returns true. FileIo::isRemote can be implemented on Windows (including MinGW and Cygwin) using GetDeviceType. Linux and MacOS-X can use the realpath /media or /Volumes strategy.
Updated by Robin Mills over 4 years ago
I've investigated why the test file loops on Image::printIFDStructure and survives TiffVisitor. I don't know the exact reason, however Ben's fix is very effective. We stop processing the subIFD in the .CR2.
778 rmills@rmillsmbp:~/Downloads $ exiv2 -pR IMG_1538.CR2 2>&1
...
STRUCTURE OF TIFF FILE (II): IMG_1538.CR2
address | tag | type | count | offset | value
998 | 0x0001 Version | SHORT | 49 | 1518 | 98 2 0 4 0 ...
1010 | 0x0002 ISOSpeed | SHORT | 4 | 1616 | 0 85 63343 21770
1022 | 0x0003 ColorMode | SHORT | 4 | 1624 | 0 0 0 0
1034 | 0x0004 Quality | SHORT | 34 | 1632 | 68 0 160 52 52 ...
1046 | 0x0006 Sharpening | ASCII | 20 | 1700 | Canon EOS REBEL SL1
1058 | 0x0007 Focus | ASCII | 24 | 1732 | Firmware Version 1.0.0.
1070 | 0x0009 GPSStatus | ASCII | 32 | 1756 | ...............................
1082 | 0x000d GPSSpeed | UNDEFINED | 1536 | 1788 | .._.`.H..../..................T ...
1094 | 0x0010 DataDump | LONG | 1 |2147484486 | 2147484486
1106 | 0x0013 GPSDestLatitudeRef | SHORT | 4 | 3324 | 0 159 7 112
1118 | 0x0019 GPSDestDistanceRef | SHORT | 1 | 1 | 1
1130 | 0x0026 | SHORT | 139 | 3332 | 278 2 31 9 5184 ...
1142 | 0x0035 | LONG | 4 | 3610 | 16 4294966816 30 0
1154 | 0x0093 | SHORT | 32 | 3626 | 64 0 0 0 0 ...
1166 | 0x0095 | ASCII | 74 | 3690 | EF85mm f/1.8 USM............... ...
1178 | 0x0096 | ASCII | 16 | 3764 | HA0751226......
1190 | 0x0097 | UNDEFINED | 1024 | 3780 | ............................... ...
1202 | 0x0098 | SHORT | 4 | 4804 | 0 0 0 0
1214 | 0x0099 | LONG | 38 | 4812 | 152 4 1 32 2 ...
1226 | 0x009a | LONG | 5 | 4964 | 0 5184 3456 0 0
1238 | 0x00a0 | SHORT | 14 | 4984 | 28 0 3 0 0 ...
1250 | 0x00aa | SHORT | 6 | 5012 | 12 835 1024 1024 407 ...
1262 | 0x00b4 | SHORT | 1 | 1 | 1
1274 | 0x00d0 | LONG | 1 | 0 | 0
1286 | 0x00e0 | SHORT | 17 | 5024 | 34 5280 3528 1 1 ...
1298 | 0x4001 | SHORT | 1313 | 5058 | 10 749 1024 1024 348 ...
1310 | 0x4002 | UNDEFINED | 43636 | 7684 | t..0""""""".................... ...
1322 | 0x4005 | UNDEFINED | 16796 | 51320 | .A..... ....................... ...
1334 | 0x4008 | SHORT | 3 | 68116 | 135 135 135
1346 | 0x4009 | SHORT | 3 | 68122 | 0 0 0
1358 | 0x4010 | ASCII | 32 | 68128 | ...............................
1370 | 0x4011 | UNDEFINED | 252 | 68160 | ............................... ...
1382 | 0x4012 | ASCII | 32 | 68412 | ...............................
1394 | 0x4013 | LONG | 11 | 68444 | 44 0 0 10 4294967295 ...
1406 | 0x4015 | UNDEFINED | 456 | 68488 | .!....................@........ ...
1418 | 0x4016 | LONG | 7 | 68944 | 28 0 1 0 1 ...
1430 | 0x4018 | LONG | 7 | 68972 | 28 0 0 0 0 ...
1442 | 0x4019 | UNDEFINED | 30 | 69000 | .............................
1454 | 0x4020 | LONG | 7 | 69030 | 28 0 0 0 2147483647 ...
invalid type value detected in Image::printIFDStructure: 0
END IMG_1538.CR2I suspect the TiffVisitor behaves in a similar manner and quietly stops processing this subIFD. TiffVisitor does detect and warn:Warning: Directory Canon, entry 0x0000 has unknown Exif (TIFF) type 0; setting type size 1.
I'm going to leave this issue closed. I'm unwilling at this time in v0.26 to introduce the new API BasicIo::isRemote() and the implementation. It's not a lot of work, however it's not essential. And with the C++ implementation being platform dependent, there will probably be build and test matters which will arise from this. And finally, there is no question that the file has illegal data.
#1289 Thanks to Asdiel (AlienSkin) for reporting this and providing a test file. Thanks to Ben for investigation and patch.