Integrate the new MPSAFE TTY layer to the FreeBSD operating system.

The last half year I've been working on a replacement TTY layer for the
FreeBSD kernel. The new TTY layer was designed to improve the following:

- Improved driver model:

  The old TTY layer has a driver model that is not abstract enough to
  make it friendly to use. A good example is the output path, where the
  device drivers directly access the output buffers. This means that an
  in-kernel PPP implementation must always convert network buffers into
  TTY buffers.

  If a PPP implementation would be built on top of the new TTY layer
  (still needs a hooks layer, though), it would allow the PPP
  implementation to directly hand the data to the TTY driver.

- Improved hotplugging:

  With the old TTY layer, it isn't entirely safe to destroy TTY's from
  the system. This implementation has a two-step destructing design,
  where the driver first abandons the TTY. After all threads have left
  the TTY, the TTY layer calls a routine in the driver, which can be
  used to free resources (unit numbers, etc).

  The pts(4) driver also implements this feature, which means
  posix_openpt() will now return PTY's that are created on the fly.

- Improved performance:

  One of the major improvements is the per-TTY mutex, which is expected
  to improve scalability when compared to the old Giant locking.
  Another change is the unbuffered copying to userspace, which is both
  used on TTY device nodes and PTY masters.

Upgrading should be quite straightforward. Unlike previous versions,
existing kernel configuration files do not need to be changed, except
when they reference device drivers that are listed in UPDATING.

Obtained from:		//depot/projects/mpsafetty/...
Approved by:		philip (ex-mentor)
Discussed:		on the lists, at BSDCan, at the DevSummit
Sponsored by:		Snow B.V., the Netherlands
dcons(4) fixed by:	kan

1 files changed, 159 insertions, 253 deletions

diff --git a/sys/dev/nmdm/nmdm.c b/sys/dev/nmdm/nmdm.c
index 8536c18..706d416 100644
--- a/sys/dev/nmdm/nmdm.c
+++ b/sys/dev/nmdm/nmdm.c
@@ -36,9 +36,6 @@ __FBSDID("$FreeBSD$");
  * Mighty handy for use with serial console in Vmware
  */
 
-#include "opt_compat.h"
-#include "opt_tty.h"
-
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/priv.h>
@@ -48,252 +45,176 @@ __FBSDID("$FreeBSD$");
 #include <sys/fcntl.h>
 #include <sys/poll.h>
 #include <sys/kernel.h>
+#include <sys/limits.h>
 #include <sys/module.h>
 #include <sys/serial.h>
 #include <sys/signalvar.h>
 #include <sys/malloc.h>
 #include <sys/taskqueue.h>
 
-MALLOC_DEFINE(M_NLMDM, "nullmodem", "nullmodem data structures");
-
-static d_close_t	nmdmclose;
-static t_modem_t	nmdmmodem;
-static d_open_t		nmdmopen;
-static t_oproc_t	nmdmoproc;
-static t_param_t	nmdmparam;
-static t_stop_t		nmdmstop;
-
-static struct cdevsw nmdm_cdevsw = {
-	.d_version =	D_VERSION,
-	.d_open =	nmdmopen,
-	.d_close =	nmdmclose,
-	.d_name =	"nmdn",
-	.d_flags =	D_TTY | D_PSEUDO | D_NEEDGIANT | D_NEEDMINOR,
+MALLOC_DEFINE(M_NMDM, "nullmodem", "nullmodem data structures");
+
+static tsw_inwakeup_t	nmdm_outwakeup;
+static tsw_outwakeup_t	nmdm_inwakeup;
+static tsw_param_t	nmdm_param;
+static tsw_modem_t	nmdm_modem;
+
+static struct ttydevsw nmdm_class = {
+	.tsw_flags	= TF_NOPREFIX,
+	.tsw_inwakeup	= nmdm_inwakeup,
+	.tsw_outwakeup	= nmdm_outwakeup,
+	.tsw_param	= nmdm_param,
+	.tsw_modem	= nmdm_modem,
 };
 
-#define BUFSIZ 		100		/* Chunk size iomoved to/from user */
-#define NMDM_MAX_NUM	128		/* Artificially limit # devices. */
-#define	PF_STOPPED	0x10		/* user told stopped */
-#define BFLAG		CLONE_FLAG0
-
-struct softpart {
-	struct tty		*nm_tty;
-	struct cdev 		*dev;
-	int			nm_dcd;
-	struct task		pt_task;
-	struct softpart		*other;
-	struct callout		co;
-	u_long			quota;
-	u_long			accumulator;
-	int			rate;
-	int			credits;
+static void nmdm_task_tty(void *, int);
+
+struct nmdmsoftc;
+
+struct nmdmpart {
+	struct tty		*np_tty;
+	int			 np_dcd;
+	struct task		 np_task;
+	struct nmdmpart		*np_other;
+	struct nmdmsoftc	*np_pair;
+	struct callout		 np_callout;
+	u_long			 np_quota;
+	u_long			 np_accumulator;
+	int			 np_rate;
+	int			 np_credits;
 
 #define QS 8	/* Quota shift */
 };
 
-struct	nm_softc {
-	TAILQ_ENTRY(nm_softc)	pt_list;
-	int			pt_flags;
-	struct softpart 	part1, part2;
-	struct	prison 		*pt_prison;
+struct nmdmsoftc {
+	struct nmdmpart	ns_part1;
+	struct nmdmpart	ns_part2;
+	struct mtx	ns_mtx;
 };
 
-static struct clonedevs *nmdmclones;
-static TAILQ_HEAD(,nm_softc) nmdmhead = TAILQ_HEAD_INITIALIZER(nmdmhead);
+static int nmdm_count = 0;
+
+static struct nmdmsoftc *
+nmdm_alloc(unsigned long unit)
+{
+	struct nmdmsoftc *ns;
+	struct tty *tp;
+
+	atomic_add_acq_int(&nmdm_count, 1);
+
+	ns = malloc(sizeof(*ns), M_NMDM, M_WAITOK|M_ZERO);
+	mtx_init(&ns->ns_mtx, "nmdm", NULL, MTX_DEF);
+
+	/* Hook the pairs together. */
+	ns->ns_part1.np_pair = ns;
+	ns->ns_part1.np_other = &ns->ns_part2;
+	TASK_INIT(&ns->ns_part1.np_task, 0, nmdm_task_tty, &ns->ns_part1);
+	callout_init(&ns->ns_part1.np_callout, 0);
+
+	ns->ns_part2.np_pair = ns;
+	ns->ns_part2.np_other = &ns->ns_part1;
+	TASK_INIT(&ns->ns_part2.np_task, 0, nmdm_task_tty, &ns->ns_part2);
+	callout_init(&ns->ns_part2.np_callout, 0);
+
+	/* Create device nodes. */
+	tp = ns->ns_part1.np_tty = tty_alloc(&nmdm_class, &ns->ns_part1,
+	    &ns->ns_mtx);
+	tty_makedev(tp, NULL, "nmdm%luA", unit);
+
+	tp = ns->ns_part2.np_tty = tty_alloc(&nmdm_class, &ns->ns_part2,
+	    &ns->ns_mtx);
+	tty_makedev(tp, NULL, "nmdm%luB", unit);
+
+	return (ns);
+}
 
 static void
 nmdm_clone(void *arg, struct ucred *cred, char *name, int nameen,
     struct cdev **dev)
 {
-	int i, unit;
-	char *p;
-	struct cdev *d1, *d2;
+	unsigned long unit;
+	char *end;
+	struct nmdmsoftc *ns;
 
 	if (*dev != NULL)
 		return;
-	if (strcmp(name, "nmdm") == 0) {
-		p = NULL;
-		unit = -1;
-	} else {
-		i = dev_stdclone(name, &p, "nmdm", &unit);
-		if (i == 0)
-			return;
-		if (p[0] != '\0' && p[0] != 'A' && p[0] != 'B')
-			return;
-		else if (p[0] != '\0' && p[1] != '\0')
-			return;
-	}
-	i = clone_create(&nmdmclones, &nmdm_cdevsw, &unit, &d1, 0);
-	if (i) {
-		d1 = make_dev(&nmdm_cdevsw, unit2minor(unit),
-		     0, 0, 0666, "nmdm%dA", unit);
-		if (d1 == NULL)
-			return;
-		d2 = make_dev(&nmdm_cdevsw, unit2minor(unit) | BFLAG,
-		     0, 0, 0666, "nmdm%dB", unit);
-		if (d2 == NULL) {
-			destroy_dev(d1);
-			return;
-		}
-		d2->si_drv2 = d1;
-		d1->si_drv2 = d2;
-		dev_depends(d1, d2);
-		dev_depends(d2, d1);
-		d1->si_flags |= SI_CHEAPCLONE;
-		d2->si_flags |= SI_CHEAPCLONE;
-	}
-	if (p != NULL && p[0] == 'B')
-		*dev = d1->si_drv2;
+	if (strncmp(name, "nmdm", 4) != 0)
+		return;
+
+	/* Device name must be "nmdm%lu%c", where %c is 'A' or 'B'. */
+	name += 4;
+	unit = strtoul(name, &end, 10);
+	if (unit == ULONG_MAX || name == end)
+		return;
+	if ((end[0] != 'A' && end[0] != 'B') || end[1] != '\0')
+		return;
+
+	/* XXX: pass privileges? */
+	ns = nmdm_alloc(unit);
+
+	if (end[0] == 'A')
+		*dev = ns->ns_part1.np_tty->t_dev;
 	else
-		*dev = d1;
-	dev_ref(*dev);
+		*dev = ns->ns_part2.np_tty->t_dev;
 }
 
 static void
 nmdm_timeout(void *arg)
 {
-	struct softpart *sp;
-
-	sp = arg;
+	struct nmdmpart *np = arg;
 
-	if (sp->rate == 0)
+	if (np->np_rate == 0)
 		return;
 
 	/*
 	 * Do a simple Floyd-Steinberg dither here to avoid FP math.
 	 * Wipe out unused quota from last tick.
 	 */
-	sp->accumulator += sp->credits;
-	sp->quota = sp->accumulator >> QS;
-	sp->accumulator &= ((1 << QS) - 1);
+	np->np_accumulator += np->np_credits;
+	np->np_quota = np->np_accumulator >> QS;
+	np->np_accumulator &= ((1 << QS) - 1);
 
-	taskqueue_enqueue(taskqueue_swi_giant, &sp->pt_task);
-	callout_reset(&sp->co, sp->rate, nmdm_timeout, arg);
+	taskqueue_enqueue(taskqueue_swi, &np->np_task);
+	callout_reset(&np->np_callout, np->np_rate, nmdm_timeout, np);
 }
 
 static void
 nmdm_task_tty(void *arg, int pending __unused)
 {
 	struct tty *tp, *otp;
-	struct softpart *sp;
-	int c;
+	struct nmdmpart *np = arg;
+	char c;
 
-	tp = arg;
-	sp = tp->t_sc;
-	otp = sp->other->nm_tty;
+	tp = np->np_tty;
+	tty_lock(tp);
+
+	otp = np->np_other->np_tty;
 	KASSERT(otp != NULL, ("NULL otp in nmdmstart"));
 	KASSERT(otp != tp, ("NULL otp == tp nmdmstart"));
-	if (sp->other->nm_dcd) {
-		if (!(tp->t_state & TS_ISOPEN)) {
-			sp->other->nm_dcd = 0;
-			(void)ttyld_modem(otp, 0);
+	if (np->np_other->np_dcd) {
+		if (!tty_opened(tp)) {
+			np->np_other->np_dcd = 0;
+			ttydisc_modem(otp, 0);
 		}
 	} else {
-		if (tp->t_state & TS_ISOPEN) {
-			sp->other->nm_dcd = 1;
-			(void)ttyld_modem(otp, 1);
+		if (tty_opened(tp)) {
+			np->np_other->np_dcd = 1;
+			ttydisc_modem(otp, 1);
 		}
 	}
-	if (tp->t_state & TS_TTSTOP)
-		return;
-	while (tp->t_outq.c_cc != 0) {
-		if (sp->rate && !sp->quota)
-			return;
-		if (otp->t_state & TS_TBLOCK)
-			return;
-		sp->quota--;
-		c = getc(&tp->t_outq);
-		if (otp->t_state & TS_ISOPEN)
-			ttyld_rint(otp, c);
-	}
-	if (tp->t_outq.c_cc == 0)
-		ttwwakeup(tp);
 
-}
-
-/*
- * This function creates and initializes a pair of ttys.
- */
-static void
-nmdminit(struct cdev *dev1)
-{
-	struct cdev *dev2;
-	struct nm_softc *pt;
-
-	dev2 = dev1->si_drv2;
-
-	dev1->si_flags &= ~SI_CHEAPCLONE;
-	dev2->si_flags &= ~SI_CHEAPCLONE;
-
-	pt = malloc(sizeof(*pt), M_NLMDM, M_WAITOK | M_ZERO);
-	TAILQ_INSERT_TAIL(&nmdmhead, pt, pt_list);
-
-	dev1->si_drv1 = dev2->si_drv1 = pt;
-
-	pt->part1.dev = dev1;
-	pt->part2.dev = dev2;
-
-	pt->part1.nm_tty = ttyalloc();
-	pt->part1.nm_tty->t_oproc = nmdmoproc;
-	pt->part1.nm_tty->t_stop = nmdmstop;
-	pt->part1.nm_tty->t_modem = nmdmmodem;
-	pt->part1.nm_tty->t_param = nmdmparam;
-	pt->part1.nm_tty->t_dev = dev1;
-	pt->part1.nm_tty->t_sc = &pt->part1;
-	TASK_INIT(&pt->part1.pt_task, 0, nmdm_task_tty, pt->part1.nm_tty);
-	callout_init(&pt->part1.co, 0);
-
-	pt->part2.nm_tty = ttyalloc();
-	pt->part2.nm_tty->t_oproc = nmdmoproc;
-	pt->part2.nm_tty->t_stop = nmdmstop;
-	pt->part2.nm_tty->t_modem = nmdmmodem;
-	pt->part2.nm_tty->t_param = nmdmparam;
-	pt->part2.nm_tty->t_dev = dev2;
-	pt->part2.nm_tty->t_sc = &pt->part2;
-	TASK_INIT(&pt->part2.pt_task, 0, nmdm_task_tty, pt->part2.nm_tty);
-	callout_init(&pt->part2.co, 0);
-
-	pt->part1.other = &pt->part2;
-	pt->part2.other = &pt->part1;
-
-	dev1->si_tty = pt->part1.nm_tty;
-	dev1->si_drv1 = pt;
-
-	dev2->si_tty = pt->part2.nm_tty;
-	dev2->si_drv1 = pt;
-}
-
-/*
- * Device opened from userland
- */
-static	int
-nmdmopen(struct cdev *dev, int flag, int devtype, struct thread *td)
-{
-	struct tty *tp, *tp2;
-	int error;
-	struct nm_softc *pti;
-	struct softpart *sp;
-
-	if (dev->si_drv1 == NULL)
-		nmdminit(dev);
-	pti = dev->si_drv1;
-	if (pti->pt_prison != td->td_ucred->cr_prison)
-		return (EBUSY);
-
-	tp = dev->si_tty;
-	sp = tp->t_sc;
-	tp2 = sp->other->nm_tty;
-
-	if ((tp->t_state & TS_ISOPEN) == 0) {
-		ttyinitmode(tp, 0, 0);
-		ttsetwater(tp); /* XXX ? */
-	} else if (tp->t_state & TS_XCLUDE &&
-	    priv_check(td, PRIV_TTY_EXCLUSIVE)) {
-		return (EBUSY);
+	while (ttydisc_rint_poll(otp) > 0) {
+		if (np->np_rate && !np->np_quota)
+			break;
+		if (ttydisc_getc(tp, &c, 1) != 1)
+			break;
+		np->np_quota--;
+		ttydisc_rint(otp, c, 0);
 	}
 
-	error = ttyld_open(tp, dev);
-	return (error);
+	ttydisc_rint_done(otp);
+
+	tty_unlock(tp);
 }
 
 static int
@@ -317,18 +238,17 @@ bits_per_char(struct termios *t)
 }
 
 static int
-nmdmparam(struct tty *tp, struct termios *t)
+nmdm_param(struct tty *tp, struct termios *t)
 {
-	struct softpart *sp;
+	struct nmdmpart *np = tty_softc(tp);
 	struct tty *tp2;
 	int bpc, rate, speed, i;
 
-	sp = tp->t_sc;
-	tp2 = sp->other->nm_tty;
+	tp2 = np->np_other->np_tty;
 
-	if (!((t->c_cflag | tp2->t_cflag) & CDSR_OFLOW)) {
-		sp->rate = 0;
-		sp->other->rate = 0;
+	if (!((t->c_cflag | tp2->t_termios.c_cflag) & CDSR_OFLOW)) {
+		np->np_rate = 0;
+		np->np_other->np_rate = 0;
 		return (0);
 	}
 
@@ -343,10 +263,10 @@ nmdmparam(struct tty *tp, struct termios *t)
 
 	for (i = 0; i < 2; i++) {
 		/* Use the slower of our receive and their transmit rate */
-		speed = imin(tp2->t_ospeed, t->c_ispeed);
+		speed = imin(tp2->t_termios.c_ospeed, t->c_ispeed);
 		if (speed == 0) {
-			sp->rate = 0;
-			sp->other->rate = 0;
+			np->np_rate = 0;
+			np->np_other->np_rate = 0;
 			return (0);
 		}
 
@@ -359,73 +279,63 @@ nmdmparam(struct tty *tp, struct termios *t)
 		speed *= rate;
 		speed /= hz;			/* [(char/sec)/tick, scaled */
 
-		sp->credits = speed;
-		sp->rate = rate;
-		callout_reset(&sp->co, rate, nmdm_timeout, sp);
+		np->np_credits = speed;
+		np->np_rate = rate;
+		callout_reset(&np->np_callout, rate, nmdm_timeout, np);
 
 		/*
 		 * swap pointers for second pass so the other end gets
 		 * updated as well.
 		 */
-		sp = sp->other;
+		np = np->np_other;
 		t = &tp2->t_termios;
 		tp2 = tp;
 	}
+
 	return (0);
 }
 
 static int
-nmdmmodem(struct tty *tp, int sigon, int sigoff)
+nmdm_modem(struct tty *tp, int sigon, int sigoff)
 {
-	struct softpart *sp;
-	int i;
+	struct nmdmpart *np = tty_softc(tp);
+	int i = 0;
 
-	sp = tp->t_sc;
 	if (sigon || sigoff) {
 		if (sigon & SER_DTR)
-			sp->other->nm_dcd = 1;
+			np->np_other->np_dcd = 1;
 		if (sigoff & SER_DTR)
-			sp->other->nm_dcd = 0;
-		ttyld_modem(sp->other->nm_tty, sp->other->nm_dcd);
+			np->np_other->np_dcd = 0;
+
+		ttydisc_modem(np->np_other->np_tty, np->np_other->np_dcd);
+
 		return (0);
 	} else {
-		i = 0;
-		if (sp->nm_dcd)
+		if (np->np_dcd)
 			i |= SER_DCD;
-		if (sp->other->nm_dcd)
+		if (np->np_other->np_dcd)
 			i |= SER_DTR;
+
 		return (i);
 	}
 }
 
-static int
-nmdmclose(struct cdev *dev, int flag, int mode, struct thread *td)
-{
-	struct tty *tp = dev->si_tty;
-	int error;
-
-	error = ttyld_close(tp, flag);
-	(void) tty_close(dev->si_tty);
-
-	return (error);
-}
-
 static void
-nmdmoproc(struct tty *tp)
+nmdm_inwakeup(struct tty *tp)
 {
-	struct softpart *pt;
+	struct nmdmpart *np = tty_softc(tp);
 
-	pt = tp->t_sc;
-	taskqueue_enqueue(taskqueue_swi_giant, &pt->pt_task);
+	/* We can receive again, so wake up the other side. */
+	taskqueue_enqueue(taskqueue_swi, &np->np_other->np_task);
 }
 
 static void
-nmdmstop(struct tty *tp, int flush)
+nmdm_outwakeup(struct tty *tp)
 {
-	struct softpart *pt;
+	struct nmdmpart *np = tty_softc(tp);
 
-	pt = tp->t_sc;
-	taskqueue_enqueue(taskqueue_swi_giant, &pt->pt_task);
+	/* We can transmit again, so wake up our side. */
+	taskqueue_enqueue(taskqueue_swi, &np->np_task);
 }
 
 /*
@@ -435,32 +345,28 @@ static int
 nmdm_modevent(module_t mod, int type, void *data)
 {
 	static eventhandler_tag tag;
-	struct nm_softc *pt, *tpt;
-        int error = 0;
 
         switch(type) {
         case MOD_LOAD: 
-		clone_setup(&nmdmclones);
 		tag = EVENTHANDLER_REGISTER(dev_clone, nmdm_clone, 0, 1000);
 		if (tag == NULL)
 			return (ENOMEM);
 		break;
 
 	case MOD_SHUTDOWN:
-		/* FALLTHROUGH */
+		break;
+
 	case MOD_UNLOAD:
+		if (nmdm_count != 0)
+			return (EBUSY);
 		EVENTHANDLER_DEREGISTER(dev_clone, tag);
-		TAILQ_FOREACH_SAFE(pt, &nmdmhead, pt_list, tpt) {
-			destroy_dev(pt->part1.dev);
-			TAILQ_REMOVE(&nmdmhead, pt, pt_list);
-			free(pt, M_NLMDM);
-		}
-		clone_cleanup(&nmdmclones);
 		break;
+
 	default:
-		error = EOPNOTSUPP;
+		return (EOPNOTSUPP);
 	}
-	return (error);
+
+	return (0);
 }
 
 DEV_MODULE(nmdm, nmdm_modevent, NULL);