logger: reconnect on failed send()

The libc syslog() reconnects on failed send(). We need the same thing
as logger(1) is expected as long time running tool. For example
recommended Apache configuration is:

	ErrorLog "| /usr/bin/logger -t apache_error -p local6.debug"

The issue is that connection endpoint (e.g. syslogd) maybe restarted.
The simple way how to test is:

	for i in $(seq 0 3600); do echo "This is message number $i"; sleep 1; done | logger --tcp --server 127.0.0.1 --port 514

and restart your syslog. The current implementation gets SIGPIPE or
write warning message, but it never reconnect.

Addresses: https://github.com/karelzak/util-linux/issues/363
Signed-off-by: Karel Zak <kzak@redhat.com>

diff --git a/misc-utils/logger.c b/misc-utils/logger.c
index 5e98d014447d3524b1e2f703201e523e4ff6dece..f8dbb2cef9687f9f7b933ca2fe62ffe8612ac2f3 100644 (file)
--- a/misc-utils/logger.c
+++ b/misc-utils/logger.c
@@ -139,6 +139,8 @@ struct logger_ctl {
                        octet_count:1;          /* use RFC6587 octet counting */
 };
 
+static void logger_reopen(struct logger_ctl *ctl);
+
 /*
  * For tests we want to be able to control datetime outputs
  */
@@ -425,7 +427,7 @@ static char const *rfc3164_current_time(void)
  * it is too much for the logger utility. If octet-counting is
  * selected, we use that.
  */
-static void write_output(const struct logger_ctl *ctl, const char *const msg)
+static void write_output(struct logger_ctl *ctl, const char *const msg)
 {
        struct iovec iov[4];
        int iovlen = 0;
@@ -482,9 +484,18 @@ static void write_output(const struct logger_ctl *ctl, const char *const msg)
                        cred->pid = ctl->pid;
                }
 #endif
+               /* Note that logger(1) maybe executed for long time (as pipe
+                * reader) and connection endpoint (syslogd) may be restarted.
+                *
+                * The libc syslog() function reconnects on failed send().
+                * Let's do the same to be robust.    [kzak -- Oct 2017]
+                */
 
-               if (sendmsg(ctl->fd, &message, 0) < 0)
-                       warn(_("send message failed"));
+               if (sendmsg(ctl->fd, &message, MSG_NOSIGNAL) < 0) {
+                       logger_reopen(ctl);
+                       if (sendmsg(ctl->fd, &message, MSG_NOSIGNAL) < 0)
+                               warn(_("send message failed"));
+               }
        }
 
        if (ctl->stderr_printout) {
@@ -865,26 +876,44 @@ static void generate_syslog_header(struct logger_ctl *const ctl)
        ctl->syslogfp(ctl);
 }
 
-static void logger_open(struct logger_ctl *ctl)
+/* just open, nothing else */
+static void __logger_open(struct logger_ctl *ctl)
 {
        if (ctl->server) {
                ctl->fd = inet_socket(ctl->server, ctl->port, &ctl->socket_type);
-               if (!ctl->syslogfp)
-                       ctl->syslogfp = syslog_rfc5424_header;
        } else {
                if (!ctl->unix_socket)
                        ctl->unix_socket = _PATH_DEVLOG;
 
                ctl->fd = unix_socket(ctl, ctl->unix_socket, &ctl->socket_type);
-               if (!ctl->syslogfp)
-                       ctl->syslogfp = syslog_local_header;
        }
+}
+
+/* open and initialize relevant @ctl tuff */
+static void logger_open(struct logger_ctl *ctl)
+{
+       __logger_open(ctl);
+
+       if (!ctl->syslogfp)
+               ctl->syslogfp = ctl->server ? syslog_rfc5424_header :
+                                             syslog_local_header;
        if (!ctl->tag)
                ctl->tag = xgetlogin();
+
        generate_syslog_header(ctl);
 }
 
-static void logger_command_line(const struct logger_ctl *ctl, char **argv)
+/* re-open; usually after failed connection */
+static void logger_reopen(struct logger_ctl *ctl)
+{
+       if (ctl->fd != -1)
+               close(ctl->fd);
+       ctl->fd = -1;
+
+       __logger_open(ctl);
+}
+
+static void logger_command_line(struct logger_ctl *ctl, char **argv)
 {
        /* note: we never re-generate the syslog header here, even if we
         * generate multiple messages. If so, we think it is the right thing