Created attachment 26141 [details] Syslog containing call traces. I am attempting to run a diskless client using nfs4. I succeed in getting all the filesystems mounted, service started etc. But it always locks up after a relatively short time. See attached syslog output. I've included the whole thing. It is not that long, but search for "Call Trace:" There are two modalities observed. Several processes block with a call trace starting at [<c06fe6c4>] __mutex_lock_common+0x104/0x153 but there is also one process (sendmail in case it is significant) which blocks with a call trace starting at [<f81c40cb>] nfs_idmap_id+0x17c/0x1de [nfs]
This is as expected. NFSv4 on diskless systems is an unsupported configuration that is known to deadlock due to the needs of the idmapper process (as you see above). The point is that rpc.idmapd needs to be able to open and read files such as nsswitch.conf, passwd, and group in order to do its job. However, in doing so, it risks invoking itself. It will take some time before we are ready to support diskless nfsv4.
Ah, of course! I am actually using an initramfs, so it might be possible (while rpc.idmapd is running off the ramdisk) to copy the important files to a tmpfs, then bind mount them over the original or run rpc.idmapd chroot, but still there might be path searches which would deadlock on directory lookups so perhaps its best to give up on nfs4 root for now.