From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on inbox.vuxu.org
X-Spam-Level: 
X-Spam-Status: No, score=0.8 required=5.0 tests=DATE_IN_PAST_12_24
	autolearn=no autolearn_force=no version=3.4.4
Received: (qmail 31149 invoked from network); 19 Oct 2021 09:18:27 -0000
Received: from 4ess.inri.net (216.126.196.42)
  by inbox.vuxu.org with ESMTPUTF8; 19 Oct 2021 09:18:27 -0000
Received: from duke.felloff.net ([216.126.196.34]) by 4ess; Mon Oct 18 15:40:43 -0400 2021
Message-ID: <E27A20967E3BEAFF238356DEB777B2ED@felloff.net>
Date: Mon, 18 Oct 2021 21:40:33 +0200
From: cinap_lenrek@felloff.net
To: 9front@9front.org
In-Reply-To: <B693CFAA-4337-4B08-9F42-7CD9B227E3D4@quintile.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
List-ID: <9front.9front.org>
List-Help: <http://lists.9front.org>
X-Glyph: ➈
X-Bullshit: descriptor DOM standard hypervisor-oriented table 
Subject: =?utf-8?q?Re:_[9front]_Re:_[9front]_Re:_[9front]_Dealing_with_log_files=E2=80=A6?=
Reply-To: 9front@9front.org
Precedence: bulk

> i dont get this, plan9 services all (to my knowledge) open and
> close the logfile on each message so you can just rename the logfile
> and touch and chmod a new one.

interesting point.

according to the code, it keeps the fd around, but it checks 
for each message if the file name or qid changed... but apparently
only for some kind of race condition when the fd got closed and
reopand after a rfork(RFCFD)

if the dirstat fails, it will also reopen, but leak the old file
descriptors (as it assumes it is the fd realloc race).

the result is indeed that the file gets reopend when you rename
the old file and create a new one.

--
cinap

PS: /sys/src/libc/9sys/syslog.c

/*
 * Print
 *  sysname: time: mesg
 * on /sys/log/logname.
 * If cons or log file can't be opened, print on the system console, too.
 */
void
syslog(int cons, char *logname, char *fmt, ...)
{
	char buf[1024];
	char *ctim, *p;
	va_list arg;
	int n;
	Dir *d;
	char err[ERRMAX];

	err[0] = '\0';
	errstr(err, sizeof err);
	lock(&sl);

	/*
	 *  paranoia makes us stat to make sure a fork+close
	 *  hasn't broken our fd's
	 */
	d = dirfstat(sl.fd);
	if(sl.fd < 0 || sl.name == nil || strcmp(sl.name, logname) != 0 ||
	   !eqdirdev(d, sl.d)){
		free(sl.name);
		sl.name = strdup(logname);
		if(sl.name == nil)
			cons = 1;
		else{
			free(sl.d);
			sl.d = nil;
			_syslogopen();
			if(sl.fd < 0)
				cons = 1;
			else
				sl.d = dirfstat(sl.fd);
		}
	}
	free(d);
	if(cons){
		d = dirfstat(sl.consfd);
		if(sl.consfd < 0 || !eqdirdev(d, sl.consd)){
			free(sl.consd);
			sl.consd = nil;
			sl.consfd = open("#c/cons", OWRITE|OCEXEC);
			if(sl.consfd >= 0)
				sl.consd = dirfstat(sl.consfd);
		}
		free(d);
	}

	if(fmt == nil){
		unlock(&sl);
		return;
	}

	ctim = ctime(time(0));
	p = buf + snprint(buf, sizeof(buf)-1, "%s ", sysname());
	strncpy(p, ctim+4, 15);
	p += 15;
	*p++ = ' ';
	errstr(err, sizeof err);
	va_start(arg, fmt);
	p = vseprint(p, buf+sizeof(buf)-1, fmt, arg);
	va_end(arg);
	*p++ = '\n';
	n = p - buf;

	if(sl.fd >= 0){
		seek(sl.fd, 0, 2);
		write(sl.fd, buf, n);
	}

	if(cons && sl.consfd >=0)
		write(sl.consfd, buf, n);

	unlock(&sl);
}