| J. Bruce Fields | 8a4c6e1 | 2012-08-15 17:43:30 -0400 | [diff] [blame] | 1 | Administrative interfaces for nfsd |
| 2 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 3 | |
| 4 | Note that normally these interfaces are used only by the utilities in |
| 5 | nfs-utils. |
| 6 | |
| 7 | nfsd is controlled mainly by pseudofiles under the "nfsd" filesystem, |
| 8 | which is normally mounted at /proc/fs/nfsd/. |
| 9 | |
| 10 | The server is always started by the first write of a nonzero value to |
| 11 | nfsd/threads. |
| 12 | |
| 13 | Before doing that, NFSD can be told which sockets to listen on by |
| 14 | writing to nfsd/portlist; that write may be: |
| 15 | |
| 16 | - an ascii-encoded file descriptor, which should refer to a |
| 17 | bound (and listening, for tcp) socket, or |
| 18 | - "transportname port", where transportname is currently either |
| 19 | "udp", "tcp", or "rdma". |
| 20 | |
| 21 | If nfsd is started without doing any of these, then it will create one |
| 22 | udp and one tcp listener at port 2049 (see nfsd_init_socks). |
| 23 | |
| 24 | On startup, nfsd and lockd grace periods start. |
| 25 | |
| 26 | nfsd is shut down by a write of 0 to nfsd/threads. All locks and state |
| 27 | are thrown away at that point. |
| 28 | |
| 29 | Between startup and shutdown, the number of threads may be adjusted up |
| 30 | or down by additional writes to nfsd/threads or by writes to |
| 31 | nfsd/pool_threads. |
| 32 | |
| 33 | For more detail about files under nfsd/ and what they control, see |
| 34 | fs/nfsd/nfsctl.c; most of them have detailed comments. |
| 35 | |
| 36 | Implementation notes |
| 37 | ^^^^^^^^^^^^^^^^^^^^ |
| 38 | |
| 39 | Note that the rpc server requires the caller to serialize addition and |
| 40 | removal of listening sockets, and startup and shutdown of the server. |
| 41 | For nfsd this is done using nfsd_mutex. |