1SM-NOTIFY(8) System Manager's Manual SM-NOTIFY(8)
2
6 sm-notify - send reboot notifications to NFS peers
7
9 /usr/sbin/sm-notify [-dfn] [-m minutes] [-v name] [-p notify-port] [-P
10 path]
11
13 File locks are not part of persistent file system state. Lock state is
14 thus lost when a host reboots.
15 Network file systems must also detect when lock state is lost because a
17 remote host has rebooted. After an NFS client reboots, an NFS server
18 must release all file locks held by applications that were running on
19 that client. After a server reboots, a client must remind the server
20 of file locks held by applications running on that client.
21 For NFS version 2 and version 3, the Network Status Monitor protocol
23 (or NSM for short) is used to notify NFS peers of reboots. On Linux,
24 two separate user-space components constitute the NSM service:
25 sm-notify
27 A helper program that notifies NFS peers after the local system
28 reboots
29 rpc.statd
31 A daemon that listens for reboot notifications from other hosts,
32 and manages the list of hosts to be notified when the local sys‐
33 tem reboots
34 The local NFS lock manager alerts its local rpc.statd of each remote
36 peer that should be monitored. When the local system reboots, the sm-
37 notify command notifies the NSM service on monitored peers of the
38 reboot. When a remote reboots, that peer notifies the local rpc.statd,
39 which in turn passes the reboot notification back to the local NFS lock
40 manager.
41
43 The first file locking interaction between an NFS client and server
44 causes the NFS lock managers on both peers to contact their local NSM
45 service to store information about the opposite peer. On Linux, the
46 local lock manager contacts rpc.statd.
47 rpc.statd records information about each monitored NFS peer on persis‐
49 tent storage. This information describes how to contact a remote peer
50 in case the local system reboots, how to recognize which monitored peer
51 is reporting a reboot, and how to notify the local lock manager when a
52 monitored peer indicates it has rebooted.
53 An NFS client sends a hostname, known as the client's caller_name, in
55 each file lock request. An NFS server can use this hostname to send
56 asynchronous GRANT calls to a client, or to notify the client it has
57 rebooted.
58 The Linux NFS server can provide the client's caller_name or the
60 client's network address to rpc.statd. For the purposes of the NSM
61 protocol, this name or address is known as the monitored peer's
62 mon_name. In addition, the local lock manager tells rpc.statd what it
63 thinks its own hostname is. For the purposes of the NSM protocol, this
64 hostname is known as my_name.
65 There is no equivalent interaction between an NFS server and a client
67 to inform the client of the server's caller_name. Therefore NFS
68 clients do not actually know what mon_name an NFS server might use in
69 an SM_NOTIFY request. The Linux NFS client records the server's host‐
70 name used on the mount command to identify rebooting NFS servers.
71 Reboot notification
73 When the local system reboots, the sm-notify command reads the list of
74 monitored peers from persistent storage and sends an SM_NOTIFY request
75 to the NSM service on each listed remote peer. It uses the mon_name
76 string as the destination. To identify which host has rebooted, the
77 sm-notify command normally sends my_name string recorded when that
78 remote was monitored. The remote rpc.statd matches incoming SM_NOTIFY
79 requests using this string, or the caller's network address, to one or
80 more peers on its own monitor list.
81 If rpc.statd does not find a peer on its monitor list that matches an
83 incoming SM_NOTIFY request, the notification is not forwarded to the
84 local lock manager. In addition, each peer has its own NSM state num‐
85 ber, a 32-bit integer that is bumped after each reboot by the sm-notify
86 command. rpc.statd uses this number to distinguish between actual
87 reboots and replayed notifications.
88 Part of NFS lock recovery is rediscovering which peers need to be moni‐
90 tored again. The sm-notify command clears the monitor list on persis‐
91 tent storage after each reboot.
92
94 -d Keeps sm-notify attached to its controlling terminal and running
95 in the foreground so that notification progress may be monitored
96 directly.
97 -f Send notifications even if sm-notify has already run since the
99 last system reboot.
100 -m retry-time
102 Specifies the length of time, in minutes, to continue retrying
103 notifications to unresponsive hosts. If this option is not
104 specified, sm-notify attempts to send notifications for 15 min‐
105 utes. Specifying a value of 0 causes sm-notify to continue
106 sending notifications to unresponsive peers until it is manually
107 killed.
108 Notifications are retried if sending fails, the remote does not
110 respond, the remote's NSM service is not registered, or if there
111 is a DNS failure which prevents the remote's mon_name from being
112 resolved to an address.
113 Hosts are not removed from the notification list until a valid
115 reply has been received. However, the SM_NOTIFY procedure has a
116 void result. There is no way for sm-notify to tell if the
117 remote recognized the sender and has started appropriate lock
118 recovery.
119 -n Prevents sm-notify from updating the local system's NSM state
121 number.
122 -p port
124 Specifies the source port number sm-notify should use when send‐
125 ing reboot notifications. If this option is not specified, a
126 randomly chosen ephemeral port is used.
127 This option can be used to traverse a firewall between client
129 and server.
130 -P, --state-directory-path pathname
132 Specifies the pathname of the parent directory where NSM state
133 information resides. If this option is not specified, sm-notify
134 uses /var/lib/nfs/statd by default.
135 After starting, sm-notify attempts to set its effective UID and
137 GID to the owner and group of the subdirectory sm of this direc‐
138 tory. After changing the effective ids, sm-notify only needs to
139 access files in sm and sm.bak within the state-directory-path.
140 -v ipaddr | hostname
142 Specifies the network address from which to send reboot notifi‐
143 cations, and the mon_name argument to use when sending SM_NOTIFY
144 requests. If this option is not specified, sm-notify uses a
145 wildcard address as the transport bind address, and uses the
146 my_name recorded when the remote was monitored as the mon_name
147 argument when sending SM_NOTIFY requests.
148 The ipaddr form can be expressed as either an IPv4 or an IPv6
150 presentation address. If the ipaddr form is used, the sm-notify
151 command converts this address to a hostname for use as the
152 mon_name argument when sending SM_NOTIFY requests.
153 This option can be useful in multi-homed configurations where
155 the remote requires notification from a specific network
156 address.
157
159 Many of the options that can be set on the command line can also be
160 controlled through values set in the [sm-notify] or, in one case, the
161 [statd] section of the /etc/nfs.conf configuration file.
162 Values recognized in the [sm-notify] section include: retry-time, out‐
164 going-port, and outgoing-addr. These have the same effect as the com‐
165 mand line options m, p, and v respectively.
166 An additional value recognized in the [sm-notify] section is lift-
168 grace. By default, sm-notify will lift lockd's grace period early if
169 it has no hosts to notify. Some high availability configurations will
170 run one sm-notify per floating IP address. In these configurations,
171 lifting the grace period early may prevent clients from reclaiming
172 locks. Setting lift-grace to n will prevent sm-notify from ending the
173 grace period early. lift-grace has no corresponding command line
174 option.
175 The value recognized in the [statd] section is state-directory-path.
177
180 The sm-notify command must be started as root to acquire privileges
181 needed to access the state information database. It drops root privi‐
182 leges as soon as it starts up to reduce the risk of a privilege escala‐
183 tion attack.
184 During normal operation, the effective user ID it chooses is the owner
186 of the state directory. This allows it to continue to access files in
187 that directory after it has dropped its root privileges. To control
188 which user ID rpc.statd chooses, simply use chown(1) to set the owner
189 of the state directory.
190
192 Lock recovery after a reboot is critical to maintaining data integrity
193 and preventing unnecessary application hangs.
194 To help rpc.statd match SM_NOTIFY requests to NLM requests, a number of
196 best practices should be observed, including:
197 The UTS nodename of your systems should match the DNS names that
199 NFS peers use to contact them
200 The UTS nodenames of your systems should always be fully quali‐
202 fied domain names
203 The forward and reverse DNS mapping of the UTS nodenames should
205 be consistent
206 The hostname the client uses to mount the server should match
208 the server's mon_name in SM_NOTIFY requests it sends
209 Unmounting an NFS file system does not necessarily stop either the NFS
211 client or server from monitoring each other. Both may continue moni‐
212 toring each other for a time in case subsequent NFS traffic between the
213 two results in fresh mounts and additional file locking.
214 On Linux, if the lockd kernel module is unloaded during normal opera‐
216 tion, all remote NFS peers are unmonitored. This can happen on an NFS
217 client, for example, if an automounter removes all NFS mount points due
218 to inactivity.
219 IPv6 and TI-RPC support
221 TI-RPC is a pre-requisite for supporting NFS on IPv6. If TI-RPC sup‐
222 port is built into the sm-notify command ,it will choose an appropriate
223 IPv4 or IPv6 transport based on the network address returned by DNS for
224 each remote peer. It should be fully compatible with remote systems
225 that do not support TI-RPC or IPv6.
226 Currently, the sm-notify command supports sending notification only via
228 datagram transport protocols.
229
231 /var/lib/nfs/statd/sm directory containing monitor list
232 /var/lib/nfs/statd/sm.bak
234 directory containing notify list
235 /var/lib/nfs/statd/state NSM state number for this host
237 /proc/sys/fs/nfs/nsm_local_state
239 kernel's copy of the NSM state number
240
242 rpc.statd(8), nfs(5), uname(2), hostname(7)
243 RFC 1094 - "NFS: Network File System Protocol Specification"
245 RFC 1813 - "NFS Version 3 Protocol Specification"
246 OpenGroup Protocols for Interworking: XNFS, Version 3W - Chapter 11
247
249 Olaf Kirch <okir@suse.de>
250 Chuck Lever <chuck.lever@oracle.com>
251 1 November 2009 SM-NOTIFY(8)