1 Date: Wed, 16 Oct 2013 04:34:01 -0400
2 From: Jeff King <peff@peff.net>
3 Subject: pack corruption post-mortem
4 Abstract: Recovering a corrupted object when no good copy is available.
5 Content-type: text/asciidoc
7 How to recover an object from scratch
8 =====================================
10 I was recently presented with a repository with a corrupted packfile,
11 and was asked if the data was recoverable. This post-mortem describes
12 the steps I took to investigate and fix the problem. I thought others
13 might find the process interesting, and it might help somebody in the
16 ********************************
17 Note: In this case, no good copy of the repository was available. For
18 the much easier case where you can get the corrupted object from
19 elsewhere, see link:recover-corrupted-blob-object.html[this howto].
20 ********************************
22 I started with an fsck, which found a problem with exactly one object
23 (I've used $pack and $obj below to keep the output readable, and also
24 because I'll refer to them later):
28 error: $pack SHA1 checksum mismatch
29 error: index CRC mismatch for object $obj from $pack at offset 51653873
30 error: inflate: data stream error (incorrect data check)
31 error: cannot unpack $obj from $pack at offset 51653873
34 The pack checksum failing means a byte is munged somewhere, and it is
35 presumably in the object mentioned (since both the index checksum and
38 Reading the zlib source code, I found that "incorrect data check" means
39 that the adler-32 checksum at the end of the zlib data did not match the
40 inflated data. So stepping the data through zlib would not help, as it
41 did not fail until the very end, when we realize the CRC does not match.
42 The problematic bytes could be anywhere in the object data.
44 The first thing I did was pull the broken data out of the packfile. I
45 needed to know how big the object was, which I found out with:
48 $ git show-index <$idx | cut -d' ' -f1 | sort -n | grep -A1 51653873
53 Show-index gives us the list of objects and their offsets. We throw away
54 everything but the offsets, and then sort them so that our interesting
55 offset (which we got from the fsck output above) is followed immediately
56 by the offset of the next object. Now we know that the object data is
57 10863 bytes long, and we can grab it with:
60 dd if=$pack of=object bs=1 skip=51653873 count=10863
63 I inspected a hexdump of the data, looking for any obvious bogosity
64 (e.g., a 4K run of zeroes would be a good sign of filesystem
65 corruption). But everything looked pretty reasonable.
67 Note that the "object" file isn't fit for feeding straight to zlib; it
68 has the git packed object header, which is variable-length. We want to
69 strip that off so we can start playing with the zlib data directly. You
70 can either work your way through it manually (the format is described in
71 link:../technical/pack-format.html[Documentation/technical/pack-format.txt]),
72 or you can walk through it in a debugger. I did the latter, creating a
76 # pack magic and version
77 printf 'PACK\0\0\0\2' >tmp.pack
79 printf '\0\0\0\1' >>tmp.pack
80 # now add our object data
82 # and then append the pack trailer
83 /path/to/git.git/test-sha1 -b <tmp.pack >trailer
84 cat trailer >>tmp.pack
87 and then running "git index-pack tmp.pack" in the debugger (stop at
88 unpack_raw_entry). Doing this, I found that there were 3 bytes of header
89 (and the header itself had a sane type and size). So I stripped those
93 dd if=object of=zlib bs=1 skip=3
96 I ran the result through zlib's inflate using a custom C program. And
97 while it did report the error, I did get the right number of output
98 bytes (i.e., it matched git's size header that we decoded above). But
99 feeding the result back to "git hash-object" didn't produce the same
100 sha1. So there were some wrong bytes, but I didn't know which. The file
101 happened to be C source code, so I hoped I could notice something
102 obviously wrong with it, but I didn't. I even got it to compile!
104 I also tried comparing it to other versions of the same path in the
105 repository, hoping that there would be some part of the diff that didn't
106 make sense. Unfortunately, this happened to be the only revision of this
107 particular file in the repository, so I had nothing to compare against.
109 So I took a different approach. Working under the guess that the
110 corruption was limited to a single byte, I wrote a program to munge each
111 byte individually, and try inflating the result. Since the object was
112 only 10K compressed, that worked out to about 2.5M attempts, which took
115 The program I used is here:
117 ----------------------------------------------
124 static int try_zlib(unsigned char *buf, int len)
126 /* make this absurdly large so we don't have to loop */
127 static unsigned char out[1024*1024];
131 memset(&z, 0, sizeof(z));
137 z.avail_out = sizeof(out);
139 ret = inflate(&z, 0);
145 static int counter = 0;
146 static void progress(int sig)
148 fprintf(stderr, "\r%d", counter);
154 /* oversized so we can read the whole buffer in */
155 unsigned char buf[1024*1024];
159 signal(SIGALRM, progress);
162 len = read(0, buf, sizeof(buf));
163 for (i = 0; i < len; i++) {
164 unsigned char c = buf[i];
165 for (j = 0; j <= 0xff; j++) {
169 if (try_zlib(buf, len))
170 printf("i=%d, j=%x\n", i, j);
176 fprintf(stderr, "\n");
179 ----------------------------------------------
181 I compiled and ran with:
184 gcc -Wall -Werror -O3 munge.c -o munge -lz
189 There were a few false positives early on (if you write "no data" in the
190 zlib header, zlib thinks it's just fine :) ). But I got a hit about
197 I let it run to completion, and got a few more hits at the end (where it
198 was munging the CRC to match our broken data). So there was a good
199 chance this middle hit was the source of the problem.
201 I confirmed by tweaking the byte in a hex editor, zlib inflating the
202 result (no errors!), and then piping the output into "git hash-object",
203 which reported the sha1 of the broken object. Success!
205 I fixed the packfile itself with:
209 printf '\xc7' | dd of=$pack bs=1 seek=51659518 conv=notrunc
213 The `\xc7` comes from the replacement byte our "munge" program found.
214 The offset 51659518 is derived by taking the original object offset
215 (51653873), adding the replacement offset found by "munge" (5642), and
216 then adding back in the 3 bytes of git header we stripped.
218 After that, "git fsck" ran clean.
220 As for the corruption itself, I was lucky that it was indeed a single
221 byte. In fact, it turned out to be a single bit. The byte 0xc7 was
222 corrupted to 0xc5. So presumably it was caused by faulty hardware, or a
225 And the aborted attempt to look at the inflated output to see what was
226 wrong? I could have looked forever and never found it. Here's the diff
227 between what the corrupted data inflates to, versus the real data:
230 - cp = strtok (arg, "+");
231 + cp = strtok (arg, ".");
234 It tweaked one byte and still ended up as valid, readable C that just
235 happened to do something totally different! One takeaway is that on a
236 less unlucky day, looking at the zlib output might have actually been
237 helpful, as most random changes would actually break the C code.
239 But more importantly, git's hashing and checksumming noticed a problem
240 that easily could have gone undetected in another system. The result
241 still compiled, but would have caused an interesting bug (that would
242 have been blamed on some random commit).