[7177] in Athena Bugs
NFS null-block bug
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Sun Feb 24 14:52:46 1991
Date: Sun, 24 Feb 91 14:52:15 -0500
To: bug-ps2@MIT.EDU, bugs@MIT.EDU
Cc: bjaspan@MIT.EDU
In-Reply-To: ps2bugs[0224]
From: Richard Basch <probe@MIT.EDU>
[0224] daemon@ATHENA.MIT.EDU Athena_PS2_bugs 02/24/91 14:08 (7 lines)
Subject: An old discuss bug
Date: Sun, 24 Feb 91 14:08:40 -0500
From: "Barr3y Jaspan" <bjaspan@ATHENA.MIT.EDU>
To: bug-ps2@ATHENA.MIT.EDU
My .meetings file was just zeroed, presumably by discuss on the PS/2
(the last machine I used it on). Grumble.
This is at least the second time I have heard this type of complaint...
another person complained that his compilation failed, and when I said
recompile without biod's, it worked.
As expected, the NFS null block bug can strike any client that runs
biod's without our kernel changes. I expect someone will eventually get
bitten on the DECstation's as well.
There are two parts to our fix:
1) Check for duplicates
2) Check the mod-time in the create request
Sun has adopted #1, but they don't account for re-ordering of packets on
the network, so if by some chance the network allows the create request
to get to the server after writes have started, there can be a null
block. This is somewhat unlikely, but possible. It becomes even more
likely depending on the nfsd's and biod's implementations. Running
clients without biod's forces them to synchronously write out data and
wait for the NFS ack, which means that a create request will be acked
prior to data being written into the file.
-Richard
