Skip to contentSkip to conversation

Spurious NFS ESTALE errors w/NFSv3 server, non-v3 client

DISCUSSION Caleb Epstein 1 participant 2 messages Sep 6, 2001 – Sep 7, 2001

lore
Share & tools
lore
#0 lore raw

I belive this is new behavior in the latest (post-2.4.7 I believe) kernel NFS software:

I have two machines, both running kernel 2.4.9, each of which act as both an NFS client and server to the other. I am using the kernel NFS daemon and am exporting ext2fs filesystems on a local switched LAN.

One box, called tela, was configured with NFSv3 enabled for both the client and server code. The other box, hagrid, was not configured with any NFSv3 support enabled. I just neglected to enable this in the configuration, its was not for any particular reason.

When I did large file reads on hagrid (the v2 client), I would get spurious ESTALE errors on files which are totally static and haven't been
touched in months. Basically the filesystem contains a lot of audio files, and I was running md5 checksums on them from hagrid, while they were hosted on tela.

When I checked the configuration on the client, and realized that NFSv3 was not enabled, I enabled it and rebuilt the kernel. After a reboot, the errors disappeared and I can successfully read many gigabytes of data without a hiccup.

Is this one of those "if it hurts then don't do that" kind of things, or is it the expected behavior? I think I've had the two machines configured like this for several kernel revisions (2.4.0 onwards) and only noticed this behavior since I switched my server to 2.4.9 from 2.4.7. It *may* have happened before and I didn't notice it, but I think this was introduced some time in 2.4.8 or later.

Signature · 1 line
cae at bklyn dot org | Caleb Epstein | bklyn . org | Brooklyn Dust Bunny Mfg.
#1 lore raw
Quoted 2 lines · Neil Brown

NFSv2 has a limit of 2Gigabytes per file. Are the files that you are reading close to, or exceeding, this size?

Not that large, no. They're on the order of tens of megabytes, maybe 150 MBytes max.

Quoted 3 lines

However, I wouldn't expect an ESTALE for that reason. Can you run "tcpdump -s 1024", the the response that contains the error, and send the dozon or so lines around that?

See below for the end of the tcpdump log, which concludes with the ESTALE. I've got the entire log saved, which is about 1.4 MB gzipped, available as http://bklyn.org/~cae/tcpdump.log.gz

For this test, server was linux 2.4.9 w/nfsv3 enabled, client was 2.4.7 with no nfsv3. Filesystem mounted on the client as:

tela:/shn on /shn/tela type nfs (rw,rsize=8192,wsize=8192,soft,addr=192.168.1.2,addr=192.168.1.2)

Let me know if I can provide any add'l info that might help.

09:24:16.191079 tela.bklyn.org.nfs > hagrid.bklyn.org.2225077128: reply ok 96 write (DF)
09:24:16.191107 hagrid.bklyn.org > tela.bklyn.org: (frag 33381:1332@2960)
09:24:16.191116 hagrid.bklyn.org > tela.bklyn.org: (frag 33381:1480@1480+)
09:24:16.191131 hagrid.bklyn.org.2275408776 > tela.bklyn.org.nfs: 1472 write fh Unknown/1 4096 (4096) bytes @ 94208 (94208) (frag 33381:1480@0+) 09:24:16.204605 tela.bklyn.org.nfs > hagrid.bklyn.org.2241854344: reply ok 96 write (DF)
09:24:16.204632 hagrid.bklyn.org > tela.bklyn.org: (frag 33382:1332@2960)
09:24:16.204640 hagrid.bklyn.org > tela.bklyn.org: (frag 33382:1480@1480+)
09:24:16.204656 hagrid.bklyn.org.2292185992 > tela.bklyn.org.nfs: 1472 write fh Unknown/1 4096 (4096) bytes @ 102400 (102400) (frag 33382:1480@0+) 09:24:16.212418 tela.bklyn.org.nfs > hagrid.bklyn.org.2258631560: reply ok 96 write (DF)
09:24:16.212443 hagrid.bklyn.org > tela.bklyn.org: (frag 33383:988@7400)
09:24:16.212451 hagrid.bklyn.org > tela.bklyn.org: (frag 33383:1480@5920+)
09:24:16.212463 hagrid.bklyn.org > tela.bklyn.org: (frag 33383:1480@4440+)
09:24:16.212480 hagrid.bklyn.org > tela.bklyn.org: (frag 33383:1480@2960+)
09:24:16.212485 hagrid.bklyn.org > tela.bklyn.org: (frag 33383:1480@1480+)
09:24:16.212492 hagrid.bklyn.org.2308963208 > tela.bklyn.org.nfs: 1472 write fh Unknown/1 8192 (8192) bytes @ 110592 (110592) (frag 33383:1480@0+) 09:24:16.220518 tela.bklyn.org.nfs > hagrid.bklyn.org.2275408776: reply ok 96 write (DF)
09:24:16.220542 hagrid.bklyn.org > tela.bklyn.org: (frag 33384:988@7400)
09:24:16.220550 hagrid.bklyn.org > tela.bklyn.org: (frag 33384:1480@5920+)
09:24:16.220562 hagrid.bklyn.org > tela.bklyn.org: (frag 33384:1480@4440+)
09:24:16.220580 hagrid.bklyn.org > tela.bklyn.org: (frag 33384:1480@2960+)
09:24:16.220587 hagrid.bklyn.org > tela.bklyn.org: (frag 33384:1480@1480+)
09:24:16.220594 hagrid.bklyn.org.2325740424 > tela.bklyn.org.nfs: 1472 write fh Unknown/1 8192 (8192) bytes @ 118784 (118784) (frag 33384:1480@0+) 09:24:16.228087 tela.bklyn.org.nfs > hagrid.bklyn.org.2292185992: reply ok 96 write (DF)
09:24:16.235465 tela.bklyn.org.nfs > hagrid.bklyn.org.2308963208: reply ok 96 write (DF)
09:24:16.246145 tela.bklyn.org.nfs > hagrid.bklyn.org.2325740424: reply ok 96 write (DF)
09:24:16.246286 hagrid.bklyn.org.2342517640 > tela.bklyn.org.nfs: 184 read fh Unknown/1 4096 bytes @ 39305216 (DF) 09:24:16.246631 tela.bklyn.org.nfs > hagrid.bklyn.org.2342517640: reply ok 28 read ERROR: Stale NFS file handle (DF)

Signature · 1 line
cae at bklyn dot org | Caleb Epstein | bklyn . org | Brooklyn Dust Bunny Mfg.