Re: [9fans] P9P venti/copy bug?

[Nathaniel W Filardo <nwf@cs.jhu.edu>]

[-- Attachment #1: Type: text/plain, Size: 1648 bytes --]

On Mon, Feb 09, 2009 at 11:55:05AM -0500, Venkatesh Srinivas wrote:
> On Mon, Feb 09, 2009 at 08:48:33AM -0800, Russ Cox wrote:
>> On Mon, Feb 9, 2009 at 4:56 AM, Venkatesh Srinivas <me@acm.jhu.edu> wrote:
>>> vcopy: reading block 0000000000000000000000000000000000000000 (type 16):
>>> read asked for 0000000000000000000000000000000000000000 got
>>> da39a3ee5e6b4b0d3255bfef95601890afd80709
>>
>> The real question is how you got an archive with an all zeros score in it.
>> There are some programs that use that internally to mean "no block"
>> but any time such a block is written to venti, the score should be the
>> zero-byte sha1 da39...0709.
>>
>
> When I use -V with Plan 9's vcopy, the zero score block isn't listed...  
> I suspect that the block with the zero score is not actually in the 
> archive?
>
> -- vs
>

I just tripped over this myself and found that apparently Plan 9 native vac
sets root.prev to all zeros when there is no previous root.  So I applied
the diff below.  I'm not sure this is a correct fix?

--nwf;

diff -r 74392a7c323c src/cmd/venti/copy.c
--- a/src/cmd/venti/copy.c      Wed Mar 11 13:37:29 2009 -0700
+++ b/src/cmd/venti/copy.c      Tue Mar 31 03:12:53 2009 -0400
@@ -10,6 +10,8 @@
 int fast;
 int verbose;
 VtConn *zsrc, *zdst;
+
+uchar allzeros[VtScoreSize];
 
 void
 usage(void)
@@ -54,7 +56,8 @@
			break;
		}
		walk(root.score, VtDirType, 0);
-   walk(root.prev, VtRootType, 0);
+   if(memcmp(root.prev, allzeros, VtScoreSize))
+     walk(root.prev, VtRootType, 0);
		if(rewrite)
			vtrootpack(&root, buf); /* walk might have changed score */
		break;


[-- Attachment #2: Type: application/pgp-signature, Size: 204 bytes --]