2.6.2: fsync() over NFS hanging unless other I/O going on

[Miquel van Smoorenburg]

[Cc: to linux-kernel because this might be more general than just NFS]

I've been able to reproduce the problem I've been talking about
before - fsync() over NFS hanging. The weird thing is that it only
happens on an otherwise complete idle system.

Server and client are both  PIV 3.0 Ghz machines, 1 GB RAM running 2.6.2.
Client mounts the server with
mount -o nfsvers=3,rsize=32768,wsize=32768 jenny-eth1:/export/swr /mnt

I have a "dd" like app that writes a file over NFS then fsync()s it.
The fsync() hangs in the kernel, in fs/nfs/write.c:nfs_sync_file()
for up to half a minute, calling nfs_wait_for_requests().

I added some debug code to nfs_sync_file(), which shows the hang
(13:03:35 -> 13:04:02) :

Feb  6 13:03:35 meghan kernel: ENTER nfs_sync_file idx_start=0 npages=0 how=2
Feb  6 13:03:35 meghan kernel: Call Trace:
Feb  6 13:03:35 meghan kernel:  [<f8a06404>] nfs_sync_file+0x48/0x153 [nfs]
Feb  6 13:03:35 meghan kernel:  [<f8a047de>] nfs_writepages+0xe9/0xf0 [nfs]
Feb  6 13:03:35 meghan kernel:  [<c013e4bb>] do_writepages+0x1e/0x38
Feb  6 13:03:35 meghan kernel:  [<c0138eb3>] __filemap_fdatawrite+0xb6/0xb8
Feb  6 13:03:35 meghan kernel:  [<c0138ecc>] filemap_fdatawrite+0x17/0x1b
Feb  6 13:03:35 meghan kernel:  [<c0155cb5>] sys_fsync+0x85/0xcf
Feb  6 13:03:35 meghan kernel:  [<c010920f>] syscall_call+0x7/0xb
Feb  6 13:03:35 meghan kernel:
Feb  6 13:03:35 meghan kernel: nfs_sync_file: waiting for requests
Feb  6 13:04:02 meghan kernel: nfs_sync_file: DONE waiting for requests error 7020
Feb  6 13:04:02 meghan kernel: nfs_sync_file: waiting for requests
Feb  6 13:04:02 meghan kernel: nfs_sync_file: DONE waiting for requests error 0
Feb  6 13:04:02 meghan kernel: nfs_sync_file: flush file
Feb  6 13:04:02 meghan kernel: nfs_sync_file: DONE flushing file error 0
Feb  6 13:04:02 meghan kernel: nfs_sync_file: commit file
Feb  6 13:04:02 meghan kernel: nfs_sync_file: DONE committing file error 50000
Feb  6 13:04:02 meghan kernel: nfs_sync_file: waiting for requests
Feb  6 13:04:04 meghan kernel: nfs_sync_file: DONE waiting for requests error 11924
Feb  6 13:04:04 meghan kernel: nfs_sync_file: waiting for requests
Feb  6 13:04:04 meghan kernel: nfs_sync_file: DONE waiting for requests error 0
Feb  6 13:04:04 meghan kernel: nfs_sync_file: flush file
Feb  6 13:04:04 meghan kernel: nfs_sync_file: DONE flushing file error 0
Feb  6 13:04:04 meghan kernel: nfs_sync_file: commit file
Feb  6 13:04:04 meghan kernel: nfs_sync_file: DONE committing file error 0
Feb  6 13:04:04 meghan kernel: EXIT nfs_sync_file

Now the weird thing is, that if I start the following in another shell
on the same server:

# cd /tmp
# while :; do echo hoi > lala; done

.. then I cannot reproduce the hang. Note that that is just a write on
a local filesystem and doesn't have anything to do with NFS at all ?

Mike.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/