When a network partition is combined with a server reboot, there are edge conditions that can cause the server to grant lock reclaims when other clients have taken conflicting locks in the interim. A more detailed explanation of this issue is described in RFC 3530, section 8.6.3 and in RFC 5661, section 8.4.3.
In order to prevent these problems, the server must track a small amount of per-client information on stable storage. This program provides the userspace piece of that functionality. When the kernel needs to manipulate the database that stores this info, it will execute this program to handle it.
- -d, --debug
- Enable debug level logging.
- -f, --foreground
- Log to stderr instead of syslog.
- -s storagedir, --storagedir=storage_dir
- Directory where stable storage information should be kept. The default value is /var/lib/nfs/nfsdcltrack.
- Initialize the database. This command requires no argument.
- Create a new client record (or update the timestamp on an existing one). This command requires a hex-encoded nfs_client_id4 as an argument.
- Remove a client record from the database. This command requires a hex-encoded nfs_client_id4 as an argument.
- Check to see if a nfs_client_id4 is allowed to reclaim. This command requires a hex-encoded nfs_client_id4 as an argument.
- Remove any unreclaimed client records from the database. This command requires a epoch boot time as an argument.
[nfsdcltrack] storagedir = /shared/nfs/nfsdcltrack
Debuging to syslog can also be enabled by setting "debug = 1" in this file.
On a check operation, if there is no record of the client in the database, nfsdcltrack will look to see if the NFSDCLTRACK_LEGACY_RECDIR environment variable is set. If it is, then it will fetch that value and see if a directory exists by that name. If it does, then the check operation will succeed and the directory will be removed.
On a gracedone operation, nfsdcltrack will look to see if the NFSDCLTRACK_LEGACY_TOPDIR environment variable is set. If it is, then it will attempt to clean out that directory prior to exiting.
Note that this transition is one-way. If the machine subsequently reboots back into an older kernel that does not support the nfsdcltrack upcall then the clients will not be able to recover their state.