Hi all,

we noticed a possible problem when using the, non-normative, file name strategy suggested in the repos-structure document section 2.2. I don't think the suggestion should be changed, but operationally the following may be of interest...


There seems to be bug in rsync when you try retrieve a single file that start with a '-' (minus) character ('-' is part of the file system safe base64 set). It seems this is related to the rsync daemon implementation, but we have not yet followed up with the rsync mailinglist (could not find anything in bugtracker):


rsync: on remote machine: -SndBLVaDchzm60h4WPrfNiVu5k.cer: invalid numeric value rsync error: requested action not supported (code 4) at clientserver.c(841) [sender=3.0.7]
rsync: read error: Connection reset by peer (54)
rsync error: error in rsync protocol data stream (code 12) at /SourceCache/rsync/rsync-40/rsync/io.c(603) [receiver=2.6.9]


Note that this probably does not affect the validation implementation for BBN and rcynic. Recursive downloads are not affected. Our validator supports bottom-up validation as well and tries to fetch all objects in the chain individually, so this functionality is affected.

From our side we are putting a work-around in place, for the time being, that will prepend any filename that would start with '-' with a '1' character. If you are implementing software that produces repositories you may want to prevent having file names starting with '-' as well.

In general one can never be sure that RPs will not try to fetch files individually and if this is a bug in rsync (as it seems to be) than even a fix to rsync may not be good enough as people out there may use old versions (impact also depends on whether the problem is server or client side..)


If anyone has any other ideas (or experience) on this, much appreciated. In the meantime we will try to follow with the rsync developers. Is there anyone on this list who could quickly verify this with them? I don't mind using the regular bug-tracker but if someone knows these people that might speed up matters. If we have more news on this we'll let the list know..



Regards
Tim Bruijnzeels
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to