| 1 |
.\"@(#)sm-notify.8" |
| 2 |
.\" |
| 3 |
.\" Copyright (C) 2004 Olaf Kirch <[email protected]> |
| 4 |
.\" |
| 5 |
.\" Rewritten by Chuck Lever <[email protected]>, 2009. |
| 6 |
.\" Copyright 2009 Oracle. All rights reserved. |
| 7 |
.\" |
| 8 |
.TH SM-NOTIFY 8 "1 November 2009 |
| 9 |
.SH NAME |
| 10 |
sm-notify \- send reboot notifications to NFS peers |
| 11 |
.SH SYNOPSIS |
| 12 |
.BI "/usr/sbin/sm-notify [-dfn] [-m " minutes "] [-v " name "] [-p " notify-port "] [-P " path "] |
| 13 |
.SH DESCRIPTION |
| 14 |
File locks are not part of persistent file system state. |
| 15 |
Lock state is thus lost when a host reboots. |
| 16 |
.PP |
| 17 |
Network file systems must also detect when lock state is lost |
| 18 |
because a remote host has rebooted. |
| 19 |
After an NFS client reboots, an NFS server must release all file locks |
| 20 |
held by applications that were running on that client. |
| 21 |
After a server reboots, a client must remind the |
| 22 |
server of file locks held by applications running on that client. |
| 23 |
.PP |
| 24 |
For NFS version 2 and version 3, the |
| 25 |
.I Network Status Monitor |
| 26 |
protocol (or NSM for short) |
| 27 |
is used to notify NFS peers of reboots. |
| 28 |
On Linux, two separate user-space components constitute the NSM service: |
| 29 |
.TP |
| 30 |
.B sm-notify |
| 31 |
A helper program that notifies NFS peers after the local system reboots |
| 32 |
.TP |
| 33 |
.B rpc.statd |
| 34 |
A daemon that listens for reboot notifications from other hosts, and |
| 35 |
manages the list of hosts to be notified when the local system reboots |
| 36 |
.PP |
| 37 |
The local NFS lock manager alerts its local |
| 38 |
.B rpc.statd |
| 39 |
of each remote peer that should be monitored. |
| 40 |
When the local system reboots, the |
| 41 |
.B sm-notify |
| 42 |
command notifies the NSM service on monitored peers of the reboot. |
| 43 |
When a remote reboots, that peer notifies the local |
| 44 |
.BR rpc.statd , |
| 45 |
which in turn passes the reboot notification |
| 46 |
back to the local NFS lock manager. |
| 47 |
.SH NSM OPERATION IN DETAIL |
| 48 |
The first file locking interaction between an NFS client and server causes |
| 49 |
the NFS lock managers on both peers to contact their local NSM service to |
| 50 |
store information about the opposite peer. |
| 51 |
On Linux, the local lock manager contacts |
| 52 |
.BR rpc.statd . |
| 53 |
.PP |
| 54 |
.B rpc.statd |
| 55 |
records information about each monitored NFS peer on persistent storage. |
| 56 |
This information describes how to contact a remote peer |
| 57 |
in case the local system reboots, |
| 58 |
how to recognize which monitored peer is reporting a reboot, |
| 59 |
and how to notify the local lock manager when a monitored peer |
| 60 |
indicates it has rebooted. |
| 61 |
.PP |
| 62 |
An NFS client sends a hostname, known as the client's |
| 63 |
.IR caller_name , |
| 64 |
in each file lock request. |
| 65 |
An NFS server can use this hostname to send asynchronous GRANT |
| 66 |
calls to a client, or to notify the client it has rebooted. |
| 67 |
.PP |
| 68 |
The Linux NFS server can provide the client's |
| 69 |
.I caller_name |
| 70 |
or the client's network address to |
| 71 |
.BR rpc.statd . |
| 72 |
For the purposes of the NSM protocol, |
| 73 |
this name or address is known as the monitored peer's |
| 74 |
.IR mon_name . |
| 75 |
In addition, the local lock manager tells |
| 76 |
.B rpc.statd |
| 77 |
what it thinks its own hostname is. |
| 78 |
For the purposes of the NSM protocol, |
| 79 |
this hostname is known as |
| 80 |
.IR my_name . |
| 81 |
.PP |
| 82 |
There is no equivalent interaction between an NFS server and a client |
| 83 |
to inform the client of the server's |
| 84 |
.IR caller_name . |
| 85 |
Therefore NFS clients do not actually know what |
| 86 |
.I mon_name |
| 87 |
an NFS server might use in an SM_NOTIFY request. |
| 88 |
The Linux NFS client records the server's hostname used on the mount command |
| 89 |
to identify rebooting NFS servers. |
| 90 |
.SS Reboot notification |
| 91 |
When the local system reboots, the |
| 92 |
.B sm-notify |
| 93 |
command reads the list of monitored peers from persistent storage and |
| 94 |
sends an SM_NOTIFY request to the NSM service on each listed remote peer. |
| 95 |
It uses the |
| 96 |
.I mon_name |
| 97 |
string as the destination. |
| 98 |
To identify which host has rebooted, the |
| 99 |
.B sm-notify |
| 100 |
command normally sends |
| 101 |
.I my_name |
| 102 |
string recorded when that remote was monitored. |
| 103 |
The remote |
| 104 |
.B rpc.statd |
| 105 |
matches incoming SM_NOTIFY requests using this string, |
| 106 |
or the caller's network address, |
| 107 |
to one or more peers on its own monitor list. |
| 108 |
.PP |
| 109 |
If |
| 110 |
.B rpc.statd |
| 111 |
does not find a peer on its monitor list that matches |
| 112 |
an incoming SM_NOTIFY request, |
| 113 |
the notification is not forwarded to the local lock manager. |
| 114 |
In addition, each peer has its own |
| 115 |
.IR "NSM state number" , |
| 116 |
a 32-bit integer that is bumped after each reboot by the |
| 117 |
.B sm-notify |
| 118 |
command. |
| 119 |
.B rpc.statd |
| 120 |
uses this number to distinguish between actual reboots |
| 121 |
and replayed notifications. |
| 122 |
.PP |
| 123 |
Part of NFS lock recovery is rediscovering |
| 124 |
which peers need to be monitored again. |
| 125 |
The |
| 126 |
.B sm-notify |
| 127 |
command clears the monitor list on persistent storage after each reboot. |
| 128 |
.SH OPTIONS |
| 129 |
.TP |
| 130 |
.B -d |
| 131 |
Keeps |
| 132 |
.B sm-notify |
| 133 |
attached to its controlling terminal and running in the foreground |
| 134 |
so that notification progress may be monitored directly. |
| 135 |
.TP |
| 136 |
.B -f |
| 137 |
Send notifications even if |
| 138 |
.B sm-notify |
| 139 |
has already run since the last system reboot. |
| 140 |
.TP |
| 141 |
.BI -m " retry-time |
| 142 |
Specifies the length of time, in minutes, to continue retrying |
| 143 |
notifications to unresponsive hosts. |
| 144 |
If this option is not specified, |
| 145 |
.B sm-notify |
| 146 |
attempts to send notifications for 15 minutes. |
| 147 |
Specifying a value of 0 causes |
| 148 |
.B sm-notify |
| 149 |
to continue sending notifications to unresponsive peers |
| 150 |
until it is manually killed. |
| 151 |
.IP |
| 152 |
Notifications are retried if sending fails, |
| 153 |
the remote does not respond, |
| 154 |
the remote's NSM service is not registered, |
| 155 |
or if there is a DNS failure |
| 156 |
which prevents the remote's |
| 157 |
.I mon_name |
| 158 |
from being resolved to an address. |
| 159 |
.IP |
| 160 |
Hosts are not removed from the notification list until a valid |
| 161 |
reply has been received. |
| 162 |
However, the SM_NOTIFY procedure has a void result. |
| 163 |
There is no way for |
| 164 |
.B sm-notify |
| 165 |
to tell if the remote recognized the sender and has started |
| 166 |
appropriate lock recovery. |
| 167 |
.TP |
| 168 |
.B -n |
| 169 |
Prevents |
| 170 |
.B sm-notify |
| 171 |
from updating the local system's NSM state number. |
| 172 |
.TP |
| 173 |
.BI -p " port |
| 174 |
Specifies the source port number |
| 175 |
.B sm-notify |
| 176 |
should use when sending reboot notifications. |
| 177 |
If this option is not specified, a randomly chosen ephemeral port is used. |
| 178 |
.IP |
| 179 |
This option can be used to traverse a firewall between client and server. |
| 180 |
.TP |
| 181 |
.BI "\-P, " "" \-\-state\-directory\-path " pathname |
| 182 |
Specifies the pathname of the parent directory |
| 183 |
where NSM state information resides. |
| 184 |
If this option is not specified, |
| 185 |
.B sm-notify |
| 186 |
uses |
| 187 |
.I /var/lib/nfs |
| 188 |
by default. |
| 189 |
.IP |
| 190 |
After starting, |
| 191 |
.B sm-notify |
| 192 |
attempts to set its effective UID and GID to the owner |
| 193 |
and group of the subdirectory |
| 194 |
.B sm |
| 195 |
of this directory. After changing the effective ids, |
| 196 |
.B sm-notify |
| 197 |
only needs to access files in |
| 198 |
.B sm |
| 199 |
and |
| 200 |
.B sm.bak |
| 201 |
within the state-directory-path. |
| 202 |
.TP |
| 203 |
.BI -v " ipaddr " | " hostname |
| 204 |
Specifies the network address from which to send reboot notifications, |
| 205 |
and the |
| 206 |
.I mon_name |
| 207 |
argument to use when sending SM_NOTIFY requests. |
| 208 |
If this option is not specified, |
| 209 |
.B sm-notify |
| 210 |
uses a wildcard address as the transport bind address, |
| 211 |
and uses the |
| 212 |
.I my_name |
| 213 |
recorded when the remote was monitored as the |
| 214 |
.I mon_name |
| 215 |
argument when sending SM_NOTIFY requests. |
| 216 |
.IP |
| 217 |
The |
| 218 |
.I ipaddr |
| 219 |
form can be expressed as either an IPv4 or an IPv6 presentation address. |
| 220 |
If the |
| 221 |
.I ipaddr |
| 222 |
form is used, the |
| 223 |
.B sm-notify |
| 224 |
command converts this address to a hostname for use as the |
| 225 |
.I mon_name |
| 226 |
argument when sending SM_NOTIFY requests. |
| 227 |
.IP |
| 228 |
This option can be useful in multi-homed configurations where |
| 229 |
the remote requires notification from a specific network address. |
| 230 |
.SH CONFIGURATION FILE |
| 231 |
Many of the options that can be set on the command line can also be |
| 232 |
controlled through values set in the |
| 233 |
.B [sm-notify] |
| 234 |
or, in one case, the |
| 235 |
.B [statd] |
| 236 |
section of the |
| 237 |
.I /etc/nfs.conf |
| 238 |
configuration file. |
| 239 |
|
| 240 |
Values recognized in the |
| 241 |
.B [sm-notify] |
| 242 |
section include: |
| 243 |
.BR retry-time , |
| 244 |
.BR outgoing-port ", and" |
| 245 |
.BR outgoing-addr . |
| 246 |
These have the same effect as the command line options |
| 247 |
.BR m , |
| 248 |
.BR p ", and" |
| 249 |
.B v |
| 250 |
respectively. |
| 251 |
|
| 252 |
An additional value recognized in the |
| 253 |
.B [sm-notify] |
| 254 |
section is |
| 255 |
.BR lift-grace . |
| 256 |
By default, |
| 257 |
.B sm-notify |
| 258 |
will lift lockd's grace period early if it has no hosts to notify. |
| 259 |
Some high availability configurations will run one |
| 260 |
.B sm-notify |
| 261 |
per floating IP address. In these configurations, lifting the |
| 262 |
grace period early may prevent clients from reclaiming locks. |
| 263 |
.RB "Setting " lift-grace " to " n |
| 264 |
will prevent |
| 265 |
.B sm-notify |
| 266 |
from ending the grace period early. |
| 267 |
.B lift-grace |
| 268 |
has no corresponding command line option. |
| 269 |
|
| 270 |
The value recognized in the |
| 271 |
.B [statd] |
| 272 |
section is |
| 273 |
.BR state-directory-path . |
| 274 |
|
| 275 |
.SH SECURITY |
| 276 |
The |
| 277 |
.B sm-notify |
| 278 |
command is, generally, started as root to acquire privileges needed |
| 279 |
to access the state information database. |
| 280 |
It drops root privileges |
| 281 |
as soon as it starts up to reduce the risk of a privilege escalation attack. |
| 282 |
.PP |
| 283 |
During normal operation, |
| 284 |
the effective user ID it chooses is the owner of the state directory. |
| 285 |
This allows it to continue to access files in that directory after it |
| 286 |
has dropped its root privileges. |
| 287 |
To control which user ID |
| 288 |
.B rpc.statd |
| 289 |
chooses, simply use |
| 290 |
.BR chown (1) |
| 291 |
to set the owner of |
| 292 |
the state directory. |
| 293 |
.PP |
| 294 |
The |
| 295 |
.B sm-notify |
| 296 |
command can also be started by the user ID that owns the state directory. |
| 297 |
That is useful for the situation when it is called from an rpc-stat |
| 298 |
ha-callout program that is already running as the |
| 299 |
non-privileged user. The non-root user will not have the capabilities |
| 300 |
to drop the root privileges as described above. |
| 301 |
.SH ADDITIONAL NOTES |
| 302 |
Lock recovery after a reboot is critical to maintaining data integrity |
| 303 |
and preventing unnecessary application hangs. |
| 304 |
.PP |
| 305 |
To help |
| 306 |
.B rpc.statd |
| 307 |
match SM_NOTIFY requests to NLM requests, a number of best practices |
| 308 |
should be observed, including: |
| 309 |
.IP |
| 310 |
The UTS nodename of your systems should match the DNS names that NFS |
| 311 |
peers use to contact them |
| 312 |
.IP |
| 313 |
The UTS nodenames of your systems should always be fully qualified domain names |
| 314 |
.IP |
| 315 |
The forward and reverse DNS mapping of the UTS nodenames should be |
| 316 |
consistent |
| 317 |
.IP |
| 318 |
The hostname the client uses to mount the server should match the server's |
| 319 |
.I mon_name |
| 320 |
in SM_NOTIFY requests it sends |
| 321 |
.PP |
| 322 |
Unmounting an NFS file system does not necessarily stop |
| 323 |
either the NFS client or server from monitoring each other. |
| 324 |
Both may continue monitoring each other for a time in case subsequent |
| 325 |
NFS traffic between the two results in fresh mounts and additional |
| 326 |
file locking. |
| 327 |
.PP |
| 328 |
On Linux, if the |
| 329 |
.B lockd |
| 330 |
kernel module is unloaded during normal operation, |
| 331 |
all remote NFS peers are unmonitored. |
| 332 |
This can happen on an NFS client, for example, |
| 333 |
if an automounter removes all NFS mount |
| 334 |
points due to inactivity. |
| 335 |
.SS IPv6 and TI-RPC support |
| 336 |
TI-RPC is a pre-requisite for supporting NFS on IPv6. |
| 337 |
If TI-RPC support is built into the |
| 338 |
.B sm-notify |
| 339 |
command ,it will choose an appropriate IPv4 or IPv6 transport |
| 340 |
based on the network address returned by DNS for each remote peer. |
| 341 |
It should be fully compatible with remote systems |
| 342 |
that do not support TI-RPC or IPv6. |
| 343 |
.PP |
| 344 |
Currently, the |
| 345 |
.B sm-notify |
| 346 |
command supports sending notification only via datagram transport protocols. |
| 347 |
.SH FILES |
| 348 |
.TP 2.5i |
| 349 |
.I /var/lib/nfs/sm |
| 350 |
directory containing monitor list |
| 351 |
.TP 2.5i |
| 352 |
.I /var/lib/nfs/sm.bak |
| 353 |
directory containing notify list |
| 354 |
.TP 2.5i |
| 355 |
.I /var/lib/nfs/state |
| 356 |
NSM state number for this host |
| 357 |
.TP 2.5i |
| 358 |
.I /proc/sys/fs/nfs/nsm_local_state |
| 359 |
kernel's copy of the NSM state number |
| 360 |
.SH SEE ALSO |
| 361 |
.BR rpc.statd (8), |
| 362 |
.BR nfs (5), |
| 363 |
.BR uname (2), |
| 364 |
.BR hostname (7) |
| 365 |
.PP |
| 366 |
RFC 1094 - "NFS: Network File System Protocol Specification" |
| 367 |
.br |
| 368 |
RFC 1813 - "NFS Version 3 Protocol Specification" |
| 369 |
.br |
| 370 |
OpenGroup Protocols for Interworking: XNFS, Version 3W - Chapter 11 |
| 371 |
.SH AUTHORS |
| 372 |
Olaf Kirch <[email protected]> |
| 373 |
.br |
| 374 |
Chuck Lever <[email protected]> |