- Drop weak bindings, use simple #define in compat.h instead
- No need to #ifdef sources with HAVE_foo, configure handles this for us
- Move utimensat() declaration to compat.h from pidfile.c to be consistent
With these changes we can let libsyslog link with the replacement objs,
just like syslogd and logger does. Because even if the C-library does
*not* have strlcpy() & C:o *and* an application has a local copy of any
of these APIs, our versions are prefixed with __ in the symbol table.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Only reset f_fime when the filed is in normal operation, not suspended,
otherwise the INET_SUSPEND_TIME handling is broken.
Signed-off-by: Johan Askerin <johan.askerin@gmail.com>
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
When sending to a remote syslog server, configured as an IP address or
when the DNS name has already been resolved, we may get temporary error
messages like ENETUNREACH and similar from sendmsg().
Before this patch the whole filed was placed in F_FORW_SUSP, like failed
DNS resolve, which introduces a 180 sec delay before even trying again.
A better approach is to just try again with the next syslog message.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
The original idea with libcompat was to keep as few objects as
possible for linking with libsyslog. That in turn to prevent
a user of libsyslog from suddenly also getting strong binding
to symbols like strlcpy() from libsyslog, rather than their C
library of choice.
However, this caused strlcpy.c to be built as both .o and .lo
files, which in turn caused really bizarre build problems due
to bad DAG dependency.
This patch drops libcompat and instead marks all replacement APIs
as weak symbols, which a C library can override.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Internal log messages of INTERNAL_MARK time were created every 30
seconds (TIMERINTVL) instead of every MarkInterval (default 20 min).
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
The __BEGIN_DECLS and __END_DECLS are internal identifiers in glibc and
are not defined in any standard. Using them fails build on musl
libc, its better to avoid them
Signed-off-by: Khem Raj <raj.khem@gmail.com>
This patch removes one of the traditionally key pieces of the sysklogd
project, klogd. Now that syslogd performs logging of kernel messages
we no longer require a separate daemon for that.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
When building the sysklogd project --without-klogd we must disable the
kernel logging to console on Linux. This fix depends on how the sysctl
setting `kernel.printk` is configured. The patch only calls the kernel
to set console_loglevel to minimum_console_loglevel.
See the kernel docs for details:
https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch adds a alarm()/SIGALRM based generic timer API to syslogd.
The API takes care to wrap SIGALRM and serialize all timer events to
a standard UNIX pipe(2) which syslogd can poll() for like any other
incoming event.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This bug caused syslogd to fall back to logging to /dev/console for
internal log messages/errors during reconfiguration at runtime.
syslogd has the FreeBSD style of keeping already open log files ready
for logging until re:init() has completed, when new log files are rolled
in and any old ones not to be used anymore are closed.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Refactor of nslookup of unknown remote syslog servers, both when
(re)reading the .conf file and at runtime. This means we retry
DNS lookup every 30 sec, or INET_SUSPEND_TIME +/- 30 sec.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
The domark() timer handles a lot of the critical maintenance action in
syslogd, it must always be guaranteed to run.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Only block signals *after* all sanity checking of log message has been
completed, otherwise we will end up with blocked SIGHUP and SIGALRM.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
We definitely want to be able to run syslogd in debug mode for extended
periods of time and still run under finit/systemd or similar, letting
users know we run as 'PID'.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
When logging to a remote server, using @fqdn syntax in syslog.conf,
syslogd calls getaddrinfo() to resolve the IP address. Older versions
of syslogd gave up after 10 retries. We want to retry forever since we
may be running in a setup with bad network connection to the DNS server
for longer periods of time.
This patch only removes the 'give up' mechanism, which unfortunately
reused the f_prevcount value, which in turn could cause that value to
become -1 and thus trigger an assert(). With this code out of the way,
and the type change in the previous commit, the counter can never again
be negative.
Note: The configurable suspend time before trying again remains at its
default of 3 minutes.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
