Make /dev/klog and kern.msgbuf* MPSAFE.

Normally msgbufp is locked using Giant. Switch it to use the
msgbuf_lock. Instead of changing the tsleep() calls to msleep(), just
convert it to condvar(9).

In my opinion the locking around msgbuf_peekbytes() still remains
questionable. It looks like locks are dropped while performing copies of
multiple blocks to userspace, which may cause the msgbuf to be reset in
the mean time. At least getting it underneath from Giant should make it
a little easier for us to figure out how to solve that.

Reminded by:	rdivacky

1 files changed, 41 insertions, 32 deletions

diff --git a/sys/kern/subr_log.c b/sys/kern/subr_log.c
index 01a3acc..972f9f9 100644
--- a/sys/kern/subr_log.c
+++ b/sys/kern/subr_log.c
@@ -53,7 +53,6 @@ __FBSDID("$FreeBSD$");
 #define LOG_RDPRI	(PZERO + 1)
 
 #define LOG_ASYNC	0x04
-#define LOG_RDWAIT	0x08
 
 static	d_open_t	logopen;
 static	d_close_t	logclose;
@@ -65,7 +64,6 @@ static	void logtimeout(void *arg);
 
 static struct cdevsw log_cdevsw = {
 	.d_version =	D_VERSION,
-	.d_flags =	D_NEEDGIANT,
 	.d_open =	logopen,
 	.d_close =	logclose,
 	.d_read =	logread,
@@ -81,7 +79,10 @@ static struct logsoftc {
 	struct	callout sc_callout;	/* callout to wakeup syslog  */
 } logsoftc;
 
-int	log_open;			/* also used in log() */
+int			log_open;	/* also used in log() */
+static struct cv	log_wakeup;
+struct mtx		msgbuf_lock;
+MTX_SYSINIT(msgbuf_lock, &msgbuf_lock, "msgbuf lock", MTX_DEF);
 
 /* Times per second to check for a pending syslog wakeup. */
 static int	log_wakeups_per_second = 5;
@@ -92,17 +93,23 @@ SYSCTL_INT(_kern, OID_AUTO, log_wakeups_per_second, CTLFLAG_RW,
 static	int
 logopen(struct cdev *dev, int flags, int mode, struct thread *td)
 {
-	if (log_open)
-		return (EBUSY);
-	log_open = 1;
-	callout_init(&logsoftc.sc_callout, 0);
-	fsetown(td->td_proc->p_pid, &logsoftc.sc_sigio);	/* signal process only */
+
 	if (log_wakeups_per_second < 1) {
 		printf("syslog wakeup is less than one.  Adjusting to 1.\n");
 		log_wakeups_per_second = 1;
 	}
+
+	mtx_lock(&msgbuf_lock);
+	if (log_open) {
+		mtx_unlock(&msgbuf_lock);
+		return (EBUSY);
+	}
+	log_open = 1;
 	callout_reset(&logsoftc.sc_callout, hz / log_wakeups_per_second,
 	    logtimeout, NULL);
+	mtx_unlock(&msgbuf_lock);
+
+	fsetown(td->td_proc->p_pid, &logsoftc.sc_sigio);	/* signal process only */
 	return (0);
 }
 
@@ -111,10 +118,14 @@ static	int
 logclose(struct cdev *dev, int flag, int mode, struct thread *td)
 {
 
-	log_open = 0;
+	funsetown(&logsoftc.sc_sigio);
+
+	mtx_lock(&msgbuf_lock);
 	callout_stop(&logsoftc.sc_callout);
 	logsoftc.sc_state = 0;
-	funsetown(&logsoftc.sc_sigio);
+	log_open = 0;
+	mtx_unlock(&msgbuf_lock);
+
 	return (0);
 }
 
@@ -124,32 +135,32 @@ logread(struct cdev *dev, struct uio *uio, int flag)
 {
 	char buf[128];
 	struct msgbuf *mbp = msgbufp;
-	int error = 0, l, s;
+	int error = 0, l;
 
-	s = splhigh();
+	mtx_lock(&msgbuf_lock);
 	while (msgbuf_getcount(mbp) == 0) {
 		if (flag & IO_NDELAY) {
-			splx(s);
+			mtx_unlock(&msgbuf_lock);
 			return (EWOULDBLOCK);
 		}
-		logsoftc.sc_state |= LOG_RDWAIT;
-		if ((error = tsleep(mbp, LOG_RDPRI | PCATCH, "klog", 0))) {
-			splx(s);
+		if ((error = cv_wait_sig(&log_wakeup, &msgbuf_lock)) != 0) {
+			mtx_unlock(&msgbuf_lock);
 			return (error);
 		}
 	}
-	splx(s);
-	logsoftc.sc_state &= ~LOG_RDWAIT;
 
 	while (uio->uio_resid > 0) {
 		l = imin(sizeof(buf), uio->uio_resid);
 		l = msgbuf_getbytes(mbp, buf, l);
 		if (l == 0)
 			break;
+		mtx_unlock(&msgbuf_lock);
 		error = uiomove(buf, l, uio);
-		if (error)
-			break;
+		if (error || uio->uio_resid == 0)
+			return (error);
+		mtx_lock(&msgbuf_lock);
 	}
+	mtx_unlock(&msgbuf_lock);
 	return (error);
 }
 
@@ -157,18 +168,16 @@ logread(struct cdev *dev, struct uio *uio, int flag)
 static	int
 logpoll(struct cdev *dev, int events, struct thread *td)
 {
-	int s;
 	int revents = 0;
 
-	s = splhigh();
-
 	if (events & (POLLIN | POLLRDNORM)) {
+		mtx_lock(&msgbuf_lock);
 		if (msgbuf_getcount(msgbufp) > 0)
 			revents |= events & (POLLIN | POLLRDNORM);
 		else
 			selrecord(td, &logsoftc.sc_selp);
+		mtx_unlock(&msgbuf_lock);
 	}
-	splx(s);
 	return (revents);
 }
 
@@ -183,20 +192,16 @@ logtimeout(void *arg)
 		log_wakeups_per_second = 1;
 	}
 	if (msgbuftrigger == 0) {
-		callout_reset(&logsoftc.sc_callout,
-		    hz / log_wakeups_per_second, logtimeout, NULL);
+		callout_schedule(&logsoftc.sc_callout,
+		    hz / log_wakeups_per_second);
 		return;
 	}
 	msgbuftrigger = 0;
 	selwakeuppri(&logsoftc.sc_selp, LOG_RDPRI);
 	if ((logsoftc.sc_state & LOG_ASYNC) && logsoftc.sc_sigio != NULL)
 		pgsigio(&logsoftc.sc_sigio, SIGIO, 0);
-	if (logsoftc.sc_state & LOG_RDWAIT) {
-		wakeup(msgbufp);
-		logsoftc.sc_state &= ~LOG_RDWAIT;
-	}
-	callout_reset(&logsoftc.sc_callout, hz / log_wakeups_per_second,
-	    logtimeout, NULL);
+	cv_broadcastpri(&log_wakeup, LOG_RDPRI);
+	callout_schedule(&logsoftc.sc_callout, hz / log_wakeups_per_second);
 }
 
 /*ARGSUSED*/
@@ -215,10 +220,12 @@ logioctl(struct cdev *dev, u_long com, caddr_t data, int flag, struct thread *td
 		break;
 
 	case FIOASYNC:
+		mtx_lock(&msgbuf_lock);
 		if (*(int *)data)
 			logsoftc.sc_state |= LOG_ASYNC;
 		else
 			logsoftc.sc_state &= ~LOG_ASYNC;
+		mtx_unlock(&msgbuf_lock);
 		break;
 
 	case FIOSETOWN:
@@ -247,6 +254,8 @@ static void
 log_drvinit(void *unused)
 {
 
+	cv_init(&log_wakeup, "klog");
+	callout_init_mtx(&logsoftc.sc_callout, &msgbuf_lock, 0);
 	make_dev(&log_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "klog");
 }