Created attachment 2702 [details] adding timestamp to debug messages Hi, I know it is very trivial thing, but it still might be helpful when hunting problems (for example my problem occurred at boot-time on an AIX-computer (it wasn't sshd's fault, but the -d option helped to debug)). See the attached paths.
Created attachment 2740 [details] adding timestamp to debug messages Reformatting as unified diff for ease of commenting. Not sure if we want to do this, though.
Comment on attachment 2740 [details] adding timestamp to debug messages >+#include <sys/timeb.h> style(9) says this should go with the other <sys/...> includes. >+ ftime (&tb); ftime is obsolete and has been removed from POSIX.1-2008. Should use gettimeofday instead. >+ localtime_r (&tb.time, &tm); >+ p += strftime (p, 16, "%Y%m%d.%H%M%S", &tm); I'm not a fan of unnecessary pointer arithmetic. Since this is neither time nor space critical, I'd rather strftime into a static buffer then use snprintf to assemble the final string. >+ p += sprintf (p, ".%03d ", tb.millitm); also not a fan of unbounded string functions even if this particular one is safe.
Created attachment 2741 [details] Changes as suggested. Diff updated with suggested changes (also, making the timestamp format ISO8601 compliant). That said, what's the use case for this? The timestamps are bulky, and the only time I'm ever interested in the timing of them is the relative time, which could be done in a lot less space by printing the number of seconds since the previous log message.
Why not use syslog? Both ssh and sshd support it and you get timestamps for free? I definitely don't think it should be on by default.
I've had some usecases recently for some sort of timestamp or timedelta in the debug messages output. It is fairly common to request 'ssh -vvv' output when debugging connection problems. In my environment the person assisting with the debugging of the connection is almost never the one running the client (or server). While it is true that syslog would add a timestamp, it isn't always possible for the user that is running the 'ssh -vvv' to update the syslog configuration. That could be for a various reasons, eg lacking privilege to do so or the system syslog configuration is controlled via a configuration management system and isn't allowed to be altered. I agree with the prior comments that a delta rather than an absolute time is probably more generally useful but either would be sufficient.
As for it being on by default, maybe a good compromise would be that timestamps are only added when the debug output is sent to a file rather than on stderr/syslog.