Alarm functions
#include <net-snmp/utilities.h>
unsigned int
snmp_alarm_register(unsigned int seconds,
unsigned int flags,
SNMPAlarmCallback *f_callback,
void *clientarg);
unsigned int
snmp_alarm_register_hr(struct timeval t,
unsigned int flags,
SNMPAlarmCallback *f_callback,
void *clientarg);
void
snmp_alarm_unregister(unsigned int reg);
These functions implement support for a generic timer handling mechanism for multiple parts of an application to register function callbacks to happen at a particular time in the future.
The usage is fairly simple and straight-forward: Simply create a function you want called back at some point in the future. The function definition should be similar to:
void my_callback(unsigned int reg, void *clientarg);
Then, call snmp_alarm_register() to register your callback to be called seconds from now. The flags field should either be SA_REPEAT or NULL. If flags is set with SA_REPEAT, then the registered callback function will be called every seconds. If flags is NULL then the function will only be called once and then removed from the alarm system registration.
The clientarg parameter in the registration function is used only by the client function and is stored and passed back directly to them on every call to the system.
The snmp_alarm_register() function returns a unique unsigned int (which is also passed as the first argument of each callback), which can then be used to remove the callback from the queue at a later point in the future using the snmp_alarm_unregister() function. If the snmp_alarm_register() call fails it returns zero. In particular, note that it is entirely permissible for an alarm function to unregister itself.
The snmp_alarm_register_hr() function is identical in operation to the snmp_alarm_register() function, but takes a struct timeval as a first parameter, and schedules the callback after the period represented by t (the letters hr stand for "high resolution"). The operation of this function is dependent on the provision of the setitimer(2) system call by the operating system. If this system call is not available, the alarm will be scheduled as if snmp_alarm_register() had been called with a first argument equal to the value of the tv_sec member of t. See, however, the notes below.
The init_snmp() function initialises the snmp_alarm subsystem by calling init_snmp_alarm() and then init_alarm_post_config() to set up the first timer to initialise the callback function. These two functions should not be used directly by applications.
The default behaviour of the snmp_alarm subsystem is to request SIGALRM signals from the operating system via the alarm(2) or setitimer(2) system calls. This has the disadvantage, however, that no other part of the application can use the SIGLARM functionality (or, if some other part of the application does use the SIGALRM functionality, the snmp_alarm subsystem will not work correctly).
If your application runs a select(2)-based event loop, however, there is no need to use SIGALRM for the snmp_alarm subsystem, leaving it available for other parts of the application. This is done by making the following call:
netsnmp_ds_set_boolean(NETSNMP_DS_LIBRARY_ID,
NETSNMP_DS_LIB_ALARM_DONT_USE_SIG, 1);
before calling init_snmp(). Then, snmp_select_info() takes alarms into account when calculating the timeout value to be used for select(2). All you need to do is call run_alarms() when select(2) times out (return value of zero). This is the approach taken in the agent; see snmpd.c. Furthermore, when using this method, high resolution alarms do not depend on the presence of the setitimer(2) system call, although overall precision is of course still determined by the underlying operating system. Recommended.