typedef struct sd_event_source sd_event_source;
typedef int (*sd_event_child_handler_t)(sd_event_source *s, const siginfo_t *si, void *userdata);
int sd_event_add_child(sd_event *event, sd_event_source **source, pid_t pid, int options, sd_event_child_handler_t handler, void *userdata);
int sd_event_add_child_pidfd(sd_event *event, sd_event_source **source, int pidfd, int options, sd_event_child_handler_t handler, void *userdata);
int sd_event_source_get_child_pid(sd_event_source *source, pid_t *pid);
int sd_event_source_get_child_pidfd(sd_event_source *source);
int sd_event_source_get_child_pidfd_own(sd_event_source *source);
int sd_event_source_set_child_pidfd_own(sd_event_source *source, int own);
int sd_event_source_get_child_process_own(sd_event_source *source);
int sd_event_source_set_child_process_own(sd_event_source *source, int own);
int sd_event_source_send_child_signal(sd_event_source *source, int sig, const siginfo_t *info, unsigned flags);
Only a single handler may be installed for a specific child process. The handler is enabled for a single event (SD_EVENT_ONESHOT), but this may be changed with sd_event_source_set_enabled(3). If the handler function returns a negative error code, it will be disabled after the invocation, even if the SD_EVENT_ON mode was requested before.
To destroy an event source object use sd_event_source_unref(3), but note that the event source is only removed from the event loop when all references to the event source are dropped. To make sure an event source does not fire anymore, even when there's still a reference to it kept, consider setting the event source to SD_EVENT_OFF with sd_event_source_set_enabled(3).
If the second parameter of sd_event_add_child() is passed as NULL no reference to the event source object is returned. In this case the event source is considered "floating", and will be destroyed implicitly when the event loop itself is destroyed.
Note that the handler function is invoked at a time where the child process is not reaped yet (and thus still is exposed as a zombie process by the kernel). However, the child will be reaped automatically after the function returns. Child processes for which no child process state change event sources are installed will not be reaped by the event loop implementation.
If the handler parameter to sd_event_add_child() is NULL, and the event source fires, this will be considered a request to exit the event loop. In this case, the userdata parameter, cast to an integer, is passed as the exit code parameter to sd_event_exit(3).
If both a child process state change event source and a SIGCHLD signal event source is installed in the same event loop, the configured event source priorities decide which event source is dispatched first. If the signal handler is processed first, it should leave the child processes for which child process state change event sources are installed unreaped.
sd_event_add_child_pidfd() is similar to sd_event_add_child() but takes a file descriptor referencing the process ("pidfd") instead of the numeric PID. A suitable file descriptor may be acquired via pidfd_open(2) and related calls. The passed file descriptor is not closed when the event source is freed again, unless sd_event_source_set_child_pidfd_own() is used to turn this behaviour on. Note that regardless which of sd_event_add_child() and sd_event_add_child_pidfd() is used for allocating an event source, the watched process has to be a direct child process of the invoking process. Also in both cases SIGCHLD has to be blocked in the invoking process.
sd_event_source_get_child_pid() retrieves the configured PID of a child process state change event source created previously with sd_event_add_child(). It takes the event source object as the source parameter and a pointer to a pid_t variable to return the process ID in.
sd_event_source_get_child_pidfd() retrieves the file descriptor referencing the watched process ("pidfd") if this functionality is available. On kernels that support the concept the event loop will make use of pidfds to watch child processes, regardless if the individual event sources are allocated via sd_event_add_child() or sd_event_add_child_pidfd(). If the latter call was used to allocate the event source, this function returns the file descriptor used for allocation. On kernels that do not support the pidfd concept this function will fail with EOPNOTSUPP. This call takes the event source object as the source parameter and returns the numeric file descriptor.
sd_event_source_get_child_pidfd_own() may be used to query whether the pidfd the event source encapsulates shall be closed when the event source is freed. This function returns zero if the pidfd shall be left open, and positive if it shall be closed automatically. By default this setting defaults to on if the event source was allocated via sd_event_add_child() and off if it was allocated via sd_event_add_child_pidfd(). The sd_event_source_set_child_pidfd_own() function may be used to change the setting and takes a boolean parameter with the new setting.
sd_event_source_get_child_process_own() may be used to query whether the process the event source watches shall be killed (with SIGKILL) and reaped when the event source is freed. This function returns zero if the process shell be left running, and positive if it shall be killed and reaped automatically. By default this setting defaults to off. The sd_event_source_set_child_process_own() function may be used to change the setting and takes a boolean parameter with the new setting. Note that currently if the calling process is terminated abnormally the watched process might survive even thought the event source ceases to exist. This behaviour might change eventually.
sd_event_source_send_child_signal() may be used to send a UNIX signal to the watched process. If the pidfd concept is supported in the kernel, this is implemented via pidfd_send_signal(2) and otherwise via rt_sigqueueinfo(2) (or via kill(2) in case info is NULL). The specified parameters match those of these underlying system calls, except that the info is never modified (and is thus declared constant). Like for the underlying system calls, the flags parameter currently must be zero.