In vanilla 2.6.32, I mount nfs4 on localhost: /n/export is the exported
point; /n/lower is the client-side mount point. I tried these two commands
in rapid succession:
# cp /bin/ls /n/lower/ls ; stat /bin/ls /n/lower/ls /n/export/ls
File: `/bin/ls'
Size: 93560 Blocks: 192 IO Block: 4096 regular file
Device: 301h/769d Inode: 98160 Links: 1
Access: (0755/-rwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2009-12-05 18:14:30.000000000 -0500
Modify: 2007-10-31 09:13:32.000000000 -0400
Change: 2008-03-23 18:02:54.000000000 -0400
File: `/n/lower/ls'
Size: 93560 Blocks: 0 IO Block: 262144 regular file
Device: fh/15d Inode: 212599 Links: 1
Access: (0755/-rwxr-xr-x) Uid: ( 0/ root) Gid: ( 100/ users)
Access: 2009-12-05 22:35:53.000000000 -0500
Modify: 2009-12-05 22:35:53.000000000 -0500
Change: 2009-12-05 22:35:53.000000000 -0500
File: `/n/export/ls'
Size: 93560 Blocks: 192 IO Block: 4096 regular file
Device: 301h/769d Inode: 212599 Links: 1
Access: (0755/-rwxr-xr-x) Uid: ( 0/ root) Gid: ( 100/ users)
Access: 2009-12-05 22:35:53.000000000 -0500
Modify: 2009-12-05 22:35:53.000000000 -0500
Change: 2009-12-05 22:35:53.000000000 -0500
The NFS server sees the inode size/blocks correctly; the NFS client,
OTOH, only sees the inode size correctly. If I wait a few seconds, then the
client's inode size and blocks become correct. I can easily reproduce
this. It doesn't happen with nfs2/3, only nfs4.
So, why this discrepancy? Why does i_size become immdiately correct while
i_blocks takes a few seconds. This is messing up my regression suites which
run on top of nfs4. Plus, doesn't this violate some posix spec?
Thanks,
Erez.