add a howto document about corrupted blob recovery
Extracted from a post by Linus on the mailing list. Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Nicolas Pitre committed
Nov 9, 2007 at 12:28 UTC
2729cadca22dce2b5bb994581c45a6a71daf7998
1 file changed
+134
Documentation/howto/recover-corrupted-blob-object.txt
new
+134
@@ -0,0 +1,134 @@
1
+Date: Fri, 9 Nov 2007 08:28:38 -0800 (PST)
2
+From: Linus Torvalds <torvalds@linux-foundation.org>
3
+Subject: corrupt object on git-gc
4
+Abstract: Some tricks to reconstruct blob objects in order to fix
5
+ a corrupted repository.
6
+
7
+On Fri, 9 Nov 2007, Yossi Leybovich wrote:
8
+>
9
+> Did not help still the repository look for this object?
10
+> Any one know how can I track this object and understand which file is it
11
+
12
+So exactly *because* the SHA1 hash is cryptographically secure, the hash
13
+itself doesn't actually tell you anything, in order to fix a corrupt
14
+object you basically have to find the "original source" for it.
15
+
16
+The easiest way to do that is almost always to have backups, and find the
17
+same object somewhere else. Backups really are a good idea, and git makes
18
+it pretty easy (if nothing else, just clone the repository somewhere else,
19
+and make sure that you do *not* use a hard-linked clone, and preferably
20
+not the same disk/machine).
21
+
22
+But since you don't seem to have backups right now, the good news is that
23
+especially with a single blob being corrupt, these things *are* somewhat
24
+debuggable.
25
+
26
+First off, move the corrupt object away, and *save* it. The most common
27
+cause of corruption so far has been memory corruption, but even so, there
28
+are people who would be interested in seeing the corruption - but it's
29
+basically impossible to judge the corruption until we can also see the
30
+original object, so right now the corrupt object is useless, but it's very
31
+interesting for the future, in the hope that you can re-create a
32
+non-corrupt version.
33
+
34
+So:
35
+
36
+> ib]$ mv .git/objects/4b/9458b3786228369c63936db65827de3cc06200 ../
37
+
38
+This is the right thing to do, although it's usually best to save it under
39
+it's full SHA1 name (you just dropped the "4b" from the result ;).
40
+
41
+Let's see what that tells us:
42
+
43
+> ib]$ git-fsck --full
44
+> broken link from tree 2d9263c6d23595e7cb2a21e5ebbb53655278dff8
45
+> to blob 4b9458b3786228369c63936db65827de3cc06200
46
+> missing blob 4b9458b3786228369c63936db65827de3cc06200
47
+
48
+Ok, I removed the "dangling commit" messages, because they are just
49
+messages about the fact that you probably have rebased etc, so they're not
50
+at all interesting. But what remains is still very useful. In particular,
51
+we now know which tree points to it!
52
+
53
+Now you can do
54
+
55
+ git ls-tree 2d9263c6d23595e7cb2a21e5ebbb53655278dff8
56
+
57
+which will show something like
58
+
59
+ 100644 blob 8d14531846b95bfa3564b58ccfb7913a034323b8 .gitignore
60
+ 100644 blob ebf9bf84da0aab5ed944264a5db2a65fe3a3e883 .mailmap
61
+ 100644 blob ca442d313d86dc67e0a2e5d584b465bd382cbf5c COPYING
62
+ 100644 blob ee909f2cc49e54f0799a4739d24c4cb9151ae453 CREDITS
63
+ 040000 tree 0f5f709c17ad89e72bdbbef6ea221c69807009f6 Documentation
64
+ 100644 blob 1570d248ad9237e4fa6e4d079336b9da62d9ba32 Kbuild
65
+ 100644 blob 1c7c229a092665b11cd46a25dbd40feeb31661d9 MAINTAINERS
66
+ ...
67
+
68
+and you should now have a line that looks like
69
+
70
+ 10064 blob 4b9458b3786228369c63936db65827de3cc06200 my-magic-file
71
+
72
+in the output. This already tells you a *lot* it tells you what file the
73
+corrupt blob came from!
74
+
75
+Now, it doesn't tell you quite enough, though: it doesn't tell what
76
+*version* of the file didn't get correctly written! You might be really
77
+lucky, and it may be the version that you already have checked out in your
78
+working tree, in which case fixing this problem is really simple, just do
79
+
80
+ git hash-object -w my-magic-file
81
+
82
+again, and if it outputs the missing SHA1 (4b945..) you're now all done!
83
+
84
+But that's the really lucky case, so let's assume that it was some older
85
+version that was broken. How do you tell which version it was?
86
+
87
+The easiest way to do it is to do
88
+
89
+ git log --raw --all --full-history -- subdirectory/my-magic-file
90
+
91
+and that will show you the whole log for that file (please realize that
92
+the tree you had may not be the top-level tree, so you need to figure out
93
+which subdirectory it was in on your own), and because you're asking for
94
+raw output, you'll now get something like
95
+
96
+ commit abc
97
+ Author:
98
+ Date:
99
+ ..
100
+ :100644 100644 4b9458b... newsha... M somedirectory/my-magic-file
101
+
102
+
103
+ commit xyz
104
+ Author:
105
+ Date:
106
+
107
+ ..
108
+ :100644 100644 oldsha... 4b9458b... M somedirectory/my-magic-file
109
+
110
+and this actually tells you what the *previous* and *subsequent* versions
111
+of that file were! So now you can look at those ("oldsha" and "newsha"
112
+respectively), and hopefully you have done commits often, and can
113
+re-create the missing my-magic-file version by looking at those older and
114
+newer versions!
115
+
116
+If you can do that, you can now recreate the missing object with
117
+
118
+ git hash-object -w <recreated-file>
119
+
120
+and your repository is good again!
121
+
122
+(Btw, you could have ignored the fsck, and started with doing a
123
+
124
+ git log --raw --all
125
+
126
+and just looked for the sha of the missing object (4b9458b..) in that
127
+whole thing. It's up to you - git does *have* a lot of information, it is
128
+just missing one particular blob version.
129
+
130
+Trying to recreate trees and especially commits is *much* harder. So you
131
+were lucky that it's a blob. It's quite possible that you can recreate the
132
+thing.
133
+
134
+ Linus