1SD_NOTIFY(3) sd_notify SD_NOTIFY(3)
2
6 sd_notify, sd_notifyf, sd_pid_notify, sd_pid_notifyf,
7 sd_pid_notify_with_fds, sd_notify_barrier - Notify service manager
8 about start-up completion and other service status changes
9
11 #include <systemd/sd-daemon.h>
12 int sd_notify(int unset_environment, const char *state);
14 int sd_notifyf(int unset_environment, const char *format, ...);
16 int sd_pid_notify(pid_t pid, int unset_environment, const char *state);
18 int sd_pid_notifyf(pid_t pid, int unset_environment,
20 const char *format, ...);
21 int sd_pid_notify_with_fds(pid_t pid, int unset_environment,
23 const char *state, const int *fds,
24 unsigned n_fds);
25 int sd_notify_barrier(int unset_environment, uint64_t timeout);
27
29 sd_notify() may be called by a service to notify the service manager
30 about state changes. It can be used to send arbitrary information,
31 encoded in an environment-block-like string. Most importantly, it can
32 be used for start-up completion notification.
33 If the unset_environment parameter is non-zero, sd_notify() will unset
35 the $NOTIFY_SOCKET environment variable before returning (regardless of
36 whether the function call itself succeeded or not). Further calls to
37 sd_notify() will then fail, but the variable is no longer inherited by
38 child processes.
39 The state parameter should contain a newline-separated list of variable
41 assignments, similar in style to an environment block. A trailing
42 newline is implied if none is specified. The string may contain any
43 kind of variable assignments, but the following shall be considered
44 well-known:
45 READY=1
47 Tells the service manager that service startup is finished, or the
48 service finished re-loading its configuration. This is only used by
49 systemd if the service definition file has Type=notify or
50 Type=notify-reload set. Since there is little value in signaling
51 non-readiness, the only value services should send is "READY=1"
52 (i.e. "READY=0" is not defined).
53 RELOADING=1
55 Tells the service manager that the service is beginning to reload
56 its configuration. This is useful to allow the service manager to
57 track the service's internal state, and present it to the user.
58 Note that a service that sends this notification must also send a
59 "READY=1" notification when it completed reloading its
60 configuration. Reloads the service manager is notified about with
61 this mechanisms are propagated in the same way as they are when
62 originally initiated through the service manager. This message is
63 particularly relevant for Type=notify-reload services, to inform
64 the service manager that the request to reload the service has been
65 received and is now being processed.
66 MONOTONIC_USEC=...
68 A field carrying the monotonic timestamp (as per CLOCK_MONOTONIC)
69 formatted in decimal in µs, when the notification message was
70 generated by the client. This is typically used in combination with
71 "RELOADING=1", to allow the service manager to properly synchronize
72 reload cycles. See systemd.service(5) for details, specifically
73 "Type=notify-reload".
74 STOPPING=1
76 Tells the service manager that the service is beginning its
77 shutdown. This is useful to allow the service manager to track the
78 service's internal state, and present it to the user.
79 STATUS=...
81 Passes a single-line UTF-8 status string back to the service
82 manager that describes the service state. This is free-form and can
83 be used for various purposes: general state feedback, fsck-like
84 programs could pass completion percentages and failing programs
85 could pass a human-readable error message. Example:
86 "STATUS=Completed 66% of file system check..."
87 ERRNO=...
89 If a service fails, the errno-style error code, formatted as
90 string. Example: "ERRNO=2" for ENOENT.
91 BUSERROR=...
93 If a service fails, the D-Bus error-style error code. Example:
94 "BUSERROR=org.freedesktop.DBus.Error.TimedOut"
95 MAINPID=...
97 The main process ID (PID) of the service, in case the service
98 manager did not fork off the process itself. Example:
99 "MAINPID=4711"
100 WATCHDOG=1
102 Tells the service manager to update the watchdog timestamp. This is
103 the keep-alive ping that services need to issue in regular
104 intervals if WatchdogSec= is enabled for it. See systemd.service(5)
105 for information how to enable this functionality and
106 sd_watchdog_enabled(3) for the details of how the service can check
107 whether the watchdog is enabled.
108 WATCHDOG=trigger
110 Tells the service manager that the service detected an internal
111 error that should be handled by the configured watchdog options.
112 This will trigger the same behaviour as if WatchdogSec= is enabled
113 and the service did not send "WATCHDOG=1" in time. Note that
114 WatchdogSec= does not need to be enabled for "WATCHDOG=trigger" to
115 trigger the watchdog action. See systemd.service(5) for information
116 about the watchdog behavior.
117 WATCHDOG_USEC=...
119 Reset watchdog_usec value during runtime. Notice that this is not
120 available when using sd_event_set_watchdog() or
121 sd_watchdog_enabled(). Example : "WATCHDOG_USEC=20000000"
122 EXTEND_TIMEOUT_USEC=...
124 Tells the service manager to extend the startup, runtime or
125 shutdown service timeout corresponding the current state. The value
126 specified is a time in microseconds during which the service must
127 send a new message. A service timeout will occur if the message
128 isn't received, but only if the runtime of the current state is
129 beyond the original maximum times of TimeoutStartSec=,
130 RuntimeMaxSec=, and TimeoutStopSec=. See systemd.service(5) for
131 effects on the service timeouts.
132 FDSTORE=1
134 Stores additional file descriptors in the service manager. File
135 descriptors sent this way will be maintained per-service by the
136 service manager and will later be handed back using the usual file
137 descriptor passing logic at the next invocation of the service
138 (e.g. when it is restarted), see sd_listen_fds(3). This is useful
139 for implementing services that can restart after an explicit
140 request or a crash without losing state. Any open sockets and other
141 file descriptors which should not be closed during the restart may
142 be stored this way. Application state can either be serialized to a
143 file in /run/, or better, stored in a memfd_create(2) memory file
144 descriptor. Note that the service manager will accept messages for
145 a service only if its FileDescriptorStoreMax= setting is non-zero
146 (defaults to zero, see systemd.service(5)). If FDPOLL=0 is not set
147 and the file descriptors sent are pollable (see epoll_ctl(2)), then
148 any EPOLLHUP or EPOLLERR event seen on them will result in their
149 automatic removal from the store. Multiple arrays of file
150 descriptors may be sent in separate messages, in which case the
151 arrays are combined. Note that the service manager removes
152 duplicate (pointing to the same object) file descriptors before
153 passing them to the service. When a service is stopped, its file
154 descriptor store is discarded and all file descriptors in it are
155 closed. Use sd_pid_notify_with_fds() to send messages with
156 "FDSTORE=1", see below.
157 FDSTOREREMOVE=1
159 Removes file descriptors from the file descriptor store. This field
160 needs to be combined with FDNAME= to specify the name of the file
161 descriptors to remove.
162 FDNAME=...
164 When used in combination with FDSTORE=1, specifies a name for the
165 submitted file descriptors. When used with FDSTOREREMOVE=1,
166 specifies the name for the file descriptors to remove. This name is
167 passed to the service during activation, and may be queried using
168 sd_listen_fds_with_names(3). File descriptors submitted without
169 this field set, will implicitly get the name "stored" assigned.
170 Note that, if multiple file descriptors are submitted at once, the
171 specified name will be assigned to all of them. In order to assign
172 different names to submitted file descriptors, submit them in
173 separate invocations of sd_pid_notify_with_fds(). The name may
174 consist of arbitrary ASCII characters except control characters or
175 ":". It may not be longer than 255 characters. If a submitted name
176 does not follow these restrictions, it is ignored.
177 FDPOLL=0
179 When used in combination with FDSTORE=1, disables polling of the
180 stored file descriptors regardless of whether or not they are
181 pollable. As this option disables automatic cleanup of the stored
182 file descriptors on EPOLLERR and EPOLLHUP, care must be taken to
183 ensure proper manual cleanup. Use of this option is not generally
184 recommended except for when automatic cleanup has unwanted behavior
185 such as prematurely discarding file descriptors from the store.
186 BARRIER=1
188 Tells the service manager that the client is explicitly requesting
189 synchronization by means of closing the file descriptor sent with
190 this command. The service manager guarantees that the processing of
191 a BARRIER=1 command will only happen after all previous
192 notification messages sent before this command have been processed.
193 Hence, this command accompanied with a single file descriptor can
194 be used to synchronize against reception of all previous status
195 messages. Note that this command cannot be mixed with other
196 notifications, and has to be sent in a separate message to the
197 service manager, otherwise all assignments will be ignored. Note
198 that sending 0 or more than 1 file descriptor with this command is
199 a violation of the protocol.
200 It is recommended to prefix variable names that are not listed above
202 with X_ to avoid namespace clashes.
203 Note that systemd will accept status data sent from a service only if
205 the NotifyAccess= option is correctly set in the service definition
206 file. See systemd.service(5) for details.
207 Note that sd_notify() notifications may be attributed to units
209 correctly only if either the sending process is still around at the
210 time PID 1 processes the message, or if the sending process is
211 explicitly runtime-tracked by the service manager. The latter is the
212 case if the service manager originally forked off the process, i.e. on
213 all processes that match NotifyAccess=main or NotifyAccess=exec.
214 Conversely, if an auxiliary process of the unit sends an sd_notify()
215 message and immediately exits, the service manager might not be able to
216 properly attribute the message to the unit, and thus will ignore it,
217 even if NotifyAccess=all is set for it.
218 Hence, to eliminate all race conditions involving lookup of the
220 client's unit and attribution of notifications to units correctly,
221 sd_notify_barrier() may be used. This call acts as a synchronization
222 point and ensures all notifications sent before this call have been
223 picked up by the service manager when it returns successfully. Use of
224 sd_notify_barrier() is needed for clients which are not invoked by the
225 service manager, otherwise this synchronization mechanism is
226 unnecessary for attribution of notifications to the unit.
227 sd_notifyf() is similar to sd_notify() but takes a printf()-like format
229 string plus arguments.
230 sd_pid_notify() and sd_pid_notifyf() are similar to sd_notify() and
232 sd_notifyf() but take a process ID (PID) to use as originating PID for
233 the message as first argument. This is useful to send notification
234 messages on behalf of other processes, provided the appropriate
235 privileges are available. If the PID argument is specified as 0, the
236 process ID of the calling process is used, in which case the calls are
237 fully equivalent to sd_notify() and sd_notifyf().
238 sd_pid_notify_with_fds() is similar to sd_pid_notify() but takes an
240 additional array of file descriptors. These file descriptors are sent
241 along the notification message to the service manager. This is
242 particularly useful for sending "FDSTORE=1" messages, as described
243 above. The additional arguments are a pointer to the file descriptor
244 array plus the number of file descriptors in the array. If the number
245 of file descriptors is passed as 0, the call is fully equivalent to
246 sd_pid_notify(), i.e. no file descriptors are passed. Note that sending
247 file descriptors to the service manager on messages that do not expect
248 them (i.e. without "FDSTORE=1") they are immediately closed on
249 reception.
250 sd_notify_barrier() allows the caller to synchronize against reception
252 of previously sent notification messages and uses the BARRIER=1
253 command. It takes a relative timeout value in microseconds which is
254 passed to ppoll(2). A value of UINT64_MAX is interpreted as infinite
255 timeout.
256
258 On failure, these calls return a negative errno-style error code. If
259 $NOTIFY_SOCKET was not set and hence no status message could be sent, 0
260 is returned. If the status was sent, these functions return a positive
261 value. In order to support both service managers that implement this
262 scheme and those which do not, it is generally recommended to ignore
263 the return value of this call. Note that the return value simply
264 indicates whether the notification message was enqueued properly, it
265 does not reflect whether the message could be processed successfully.
266 Specifically, no error is returned when a file descriptor is attempted
267 to be stored using FDSTORE=1 but the service is not actually configured
268 to permit storing of file descriptors (see above).
269
271 These APIs are implemented as a shared library, which can be compiled
272 and linked to with the libsystemd pkg-config(1) file.
273 These functions send a single datagram with the state string as payload
275 to the socket referenced in the $NOTIFY_SOCKET environment variable. If
276 the first character of $NOTIFY_SOCKET is "/" or "@", the string is
277 understood as an AF_UNIX or Linux abstract namespace socket
278 (respectively), and in both cases the datagram is accompanied by the
279 process credentials of the sending service, using SCM_CREDENTIALS. If
280 the string starts with "vsock:" then the string is understood as an
281 AF_VSOCK address, which is useful for hypervisors/VMMs or other
282 processes on the host to receive a notification when a virtual machine
283 has finished booting. Note that in case the hypervisor does not support
284 SOCK_DGRAM over AF_VSOCK, SOCK_SEQPACKET will be used instead. The
285 address should be in the form: "vsock:CID:PORT". Note that unlike other
286 uses of vsock, the CID is mandatory and cannot be "VMADDR_CID_ANY".
287 Note that PID1 will send the VSOCK packets from a privileged port
288 (i.e.: lower than 1024), as an attempt to address concerns that
289 unprivileged processes in the guest might try to send malicious
290 notifications to the host, driving it to make destructive decisions
291 based on them.
292
294 $NOTIFY_SOCKET
295 Set by the service manager for supervised processes for status and
296 start-up completion notification. This environment variable
297 specifies the socket sd_notify() talks to. See above for details.
298
300 Example 1. Start-up Notification
301 When a service finished starting up, it might issue the following call
303 to notify the service manager:
304 sd_notify(0, "READY=1");
306 Example 2. Extended Start-up Notification
308 A service could send the following after completing initialization:
310 sd_notifyf(0, "READY=1\n"
312 "STATUS=Processing requests...\n"
313 "MAINPID=%lu",
314 (unsigned long) getpid());
315 Example 3. Error Cause Notification
317 A service could send the following shortly before exiting, on failure:
319 sd_notifyf(0, "STATUS=Failed to start up: %s\n"
321 "ERRNO=%i",
322 strerror_r(errnum, (char[1024]){}, 1024),
323 errnum);
324 Example 4. Store a File Descriptor in the Service Manager
326 To store an open file descriptor in the service manager, in order to
328 continue operation after a service restart without losing state, use
329 "FDSTORE=1":
330 sd_pid_notify_with_fds(0, 0, "FDSTORE=1\nFDNAME=foobar", &fd, 1);
332 Example 5. Eliminating race conditions
334 When the client sending the notifications is not spawned by the service
336 manager, it may exit too quickly and the service manager may fail to
337 attribute them correctly to the unit. To prevent such races, use
338 sd_notify_barrier() to synchronize against reception of all
339 notifications sent before this call is made.
340 sd_notify(0, "READY=1");
342 /* set timeout to 5 seconds */
343 sd_notify_barrier(0, 5 * 1000000);
344
347 systemd(1), sd-daemon(3), sd_listen_fds(3),
348 sd_listen_fds_with_names(3), sd_watchdog_enabled(3), daemon(7),
349 systemd.service(5)
350systemd 253 SD_NOTIFY(3)