]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/tty.c
xnu-4903.241.1.tar.gz
[apple/xnu.git] / bsd / kern / tty.c
index 59205429ad6aa77cf1dd305c55113c944cf70d1f..ed7e7965820295c04f6079338897c2b896c34e96 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997-2007 Apple Inc. All rights reserved.
+ * Copyright (c) 1997-2017 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
 #include <sys/file_internal.h>
 #include <sys/conf.h>
 #include <sys/dkstat.h>
-#include <sys/uio.h>
+#include <sys/uio_internal.h>
 #include <sys/kernel.h>
 #include <sys/vnode.h>
 #include <sys/syslog.h>
 #include <dev/kmreg_com.h>
 #include <machine/cons.h>
 #include <sys/resource.h>      /* averunnable */
+#include <kern/waitq.h>
+#include <libkern/section_keywords.h>
 
-static int     ttnread(struct tty *tp);
+static lck_grp_t       *tty_lck_grp;
+static lck_grp_attr_t  *tty_lck_grp_attr;
+static lck_attr_t      *tty_lck_attr;
+
+__private_extern__ int ttnread(struct tty *tp);
 static void    ttyecho(int c, struct tty *tp);
 static int     ttyoutput(int c, struct tty *tp);
 static void    ttypend(struct tty *tp);
@@ -126,6 +132,9 @@ static void ttyunblock(struct tty *tp);
 static int     ttywflush(struct tty *tp);
 static int     proc_compare(proc_t p1, proc_t p2);
 
+void ttyhold(struct tty *tp);
+static void    ttydeallocate(struct tty *tp);
+
 static int isctty(proc_t p, struct tty  *tp);
 static int isctty_sp(proc_t p, struct tty  *tp, struct session *sessp);
 
@@ -218,7 +227,7 @@ static u_char const char_type[] = {
 #define        I_LOW_WATER     ((TTYHOG - 2 * 256) * 7 / 8)    /* XXX */
 
 static void
-termios32to64(struct termios *in, struct user_termios *out)
+termios32to64(struct termios32 *in, struct user_termios *out)
 {
        out->c_iflag = (user_tcflag_t)in->c_iflag;
        out->c_oflag = (user_tcflag_t)in->c_oflag;
@@ -233,7 +242,7 @@ termios32to64(struct termios *in, struct user_termios *out)
 }
 
 static void
-termios64to32(struct user_termios *in, struct termios *out)
+termios64to32(struct user_termios *in, struct termios32 *out)
 {
        out->c_iflag = (tcflag_t)in->c_iflag;
        out->c_oflag = (tcflag_t)in->c_oflag;
@@ -249,17 +258,79 @@ termios64to32(struct user_termios *in, struct termios *out)
 
 
 /*
+ * tty_init
+ *
+ * Initialize the tty line discipline subsystem.
+ *
+ * Parameters: void
+ *
+ * Returns:    void
+ *
+ * Locks:      No ttys can be allocated and no tty locks can be used
+ *             until after this function is called
+ *
+ * Notes:      The intent of this is to set up a log group attribute,
+ *             lock group, and loc atribute for subsequent per-tty locks.
+ *             This function is called early in bsd_init(), prior to the
+ *             console device initialization.
+ */
+void
+tty_init(void)
+{
+       tty_lck_grp_attr = lck_grp_attr_alloc_init();
+       tty_lck_grp = lck_grp_alloc_init("tty",  tty_lck_grp_attr);
+       tty_lck_attr = lck_attr_alloc_init();
+}
+
+
+/*
+ * tty_lock
+ *
+ * Lock the requested tty structure.
+ *
+ * Parameters: tp                              The tty we want to lock
+ *
+ * Returns:    void
+ *
+ * Locks:      On return, tp is locked
+ */
+void
+tty_lock(struct tty *tp)
+{
+       TTY_LOCK_NOTOWNED(tp);  /* debug assert */
+       lck_mtx_lock(&tp->t_lock);
+}
+
+
+/*
+ * tty_unlock
+ *
+ * Unlock the requested tty structure.
+ *
+ * Parameters: tp                              The tty we want to unlock
+ *
+ * Returns:    void
+ *
+ * Locks:      On return, tp is unlocked
+ */
+void
+tty_unlock(struct tty *tp)
+{
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+       lck_mtx_unlock(&tp->t_lock);
+}
+
+/*
+ * ttyopen (LDISC)
+ *
  * Initial open of tty, or (re)entry to standard tty line discipline.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttyopen(dev_t device, struct tty *tp)
 {
-       boolean_t funnel_state;
-       proc_t p = current_proc();
-       struct pgrp * pg, * oldpg;
-       struct session *sessp, *oldsess;
-
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        tp->t_dev = device;
 
@@ -270,65 +341,29 @@ ttyopen(dev_t device, struct tty *tp)
                bzero(&tp->t_winsize, sizeof(tp->t_winsize));
        }
 
-       pg = proc_pgrp(p);
-       sessp = proc_session(p);
-
-       /*
-        * First tty open affter setsid() call makes this tty its controlling
-        * tty, if the tty does not already have a session associated with it.
-        * Only do this if the process
-        */
-       if (SESS_LEADER(p, sessp) &&                    /* process is session leader */
-           sessp->s_ttyvp == NULL &&   /* but has no controlling tty */
-           tp->t_session == NULL ) {           /* and tty not controlling */
-               session_lock(sessp);
-               if ((sessp->s_flags & S_NOCTTY) == 0) { /* and no O_NOCTTY */
-                       /* Hold on to the reference */
-                       sessp->s_ttyp = tp;
-                       OSBitOrAtomic(P_CONTROLT, (UInt32 *)&p->p_flag);
-                       session_unlock(sessp);
-                       proc_list_lock();
-                       oldpg = tp->t_pgrp;
-                       oldsess = tp->t_session;
-                       if (oldsess != SESSION_NULL)
-                               oldsess->s_ttypgrpid = NO_PID;
-                       tp->t_session = sessp;
-                       tp->t_pgrp = pg;
-                       sessp->s_ttypgrpid = pg->pg_id;
-                       proc_list_unlock();
-                       if (oldpg != PGRP_NULL)
-                               pg_rele(oldpg);
-                       if (oldsess != SESSION_NULL)
-                               session_rele(oldsess);  
-                       goto out;
-               }
-               session_unlock(sessp);
-       }
-
-       if (sessp != SESSION_NULL)
-               session_rele(sessp);
-       if (pg != PGRP_NULL)
-               pg_rele(pg);
-
-out:
-       thread_funnel_set(kernel_flock, funnel_state);
-
        return (0);
 }
 
 /*
+ * ttyclose
+ *
  * Handle close() on a tty line: flush and set to initial state,
  * bumping generation number so that pending read/write calls
  * can detect recycling of the tty.
  * XXX our caller should have done `spltty(); l_close(); ttyclose();'
  * and l_close() should have flushed, but we repeat the spltty() and
  * the flush in case there are buggy callers.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttyclose(struct tty *tp)
 {
        struct pgrp * oldpg;
        struct session * oldsessp;
+       struct knote *kn;
+
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        if (constty == tp) {
                constty = NULL;
@@ -355,13 +390,23 @@ ttyclose(struct tty *tp)
                oldsessp->s_ttypgrpid = NO_PID;
        proc_list_unlock();
        /* drop the reference on prev session and pgrp */
+       /* SAFE: All callers drop the lock on return */
+       tty_unlock(tp);
        if (oldsessp != SESSION_NULL)
                session_rele(oldsessp);
        if (oldpg != PGRP_NULL)
                pg_rele(oldpg);
+       tty_lock(tp);
        tp->t_state = 0;
+       SLIST_FOREACH(kn, &tp->t_wsel.si_note, kn_selnext) {
+               KNOTE_DETACH(&tp->t_wsel.si_note, kn);
+       }
        selthreadclear(&tp->t_wsel);
+       SLIST_FOREACH(kn, &tp->t_rsel.si_note, kn_selnext) {
+               KNOTE_DETACH(&tp->t_rsel.si_note, kn);
+       }
        selthreadclear(&tp->t_rsel);
+
        return (0);
 }
 
@@ -377,17 +422,26 @@ ttyclose(struct tty *tp)
         (c) != _POSIX_VDISABLE))
 
 /*
+ * ttyinput (LDISC)
+ *
  * Process input of a single character received on a tty.
+ *
+ * Parameters: c                       The character received
+ *             tp                      The tty on which it was received
+ *
+ * Returns:    .
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttyinput(int c, struct tty *tp)
 {
        tcflag_t iflag, lflag;
        cc_t *cc;
-       int i, err, retval;
-       boolean_t funnel_state;
+       int i, err;
+       int retval = 0;                 /* default return value */
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
         
        /*
         * If input is pending take it first.
@@ -427,12 +481,14 @@ ttyinput(int c, struct tty *tp)
                CLR(c, TTY_ERRORMASK);
                if (ISSET(err, TTY_BI)) {
                        if (ISSET(iflag, IGNBRK)) {
-                               thread_funnel_set(kernel_flock, funnel_state);
-                               return (0);
+                               goto out;
                         }
                        if (ISSET(iflag, BRKINT)) {
                                ttyflush(tp, FREAD | FWRITE);
+                               /* SAFE: All callers drop the lock on return */
+                               tty_unlock(tp);
                                tty_pgsignal(tp, SIGINT, 1);
+                               tty_lock(tp);
                                goto endcase;
                        }
                        if (ISSET(iflag, PARMRK))
@@ -440,8 +496,7 @@ ttyinput(int c, struct tty *tp)
                } else if ((ISSET(err, TTY_PE) && ISSET(iflag, INPCK))
                        || ISSET(err, TTY_FE)) {
                        if (ISSET(iflag, IGNPAR)) {
-                               thread_funnel_set(kernel_flock, funnel_state);
-                               return (0);
+                               goto out;
                        }
                        else if (ISSET(iflag, PARMRK)) {
 parmrk:
@@ -512,15 +567,29 @@ parmrk:
                                if (!ISSET(lflag, NOFLSH))
                                        ttyflush(tp, FREAD | FWRITE);
                                ttyecho(c, tp);
+                               /*
+                                * SAFE: All callers drop the lock on return;
+                                * SAFE: if we lose a threaded race on change
+                                * SAFE: of the interrupt character, we could
+                                * SAFE: have lost that race anyway due to the
+                                * SAFE: scheduler executing threads in
+                                * SAFE: priority order rather than "last
+                                * SAFE: active thread" order (FEATURE).
+                                */
+                               tty_unlock(tp);
                                tty_pgsignal(tp,
                                    CCEQ(cc[VINTR], c) ? SIGINT : SIGQUIT, 1);
+                               tty_lock(tp);
                                goto endcase;
                        }
                        if (CCEQ(cc[VSUSP], c)) {
                                if (!ISSET(lflag, NOFLSH))
                                        ttyflush(tp, FREAD);
                                ttyecho(c, tp);
+                               /* SAFE: All callers drop the lock on return */
+                               tty_unlock(tp);
                                tty_pgsignal(tp, SIGTSTP, 1);
+                               tty_lock(tp);
                                goto endcase;
                        }
                }
@@ -532,12 +601,10 @@ parmrk:
                                if (!ISSET(tp->t_state, TS_TTSTOP)) {
                                        SET(tp->t_state, TS_TTSTOP);
                                         ttystop(tp, 0);
-                                       thread_funnel_set(kernel_flock, funnel_state);
-                                       return (0);
+                                       goto out;
                                }
                                if (!CCEQ(cc[VSTART], c)) {
-                                       thread_funnel_set(kernel_flock, funnel_state);
-                                       return (0);
+                                       goto out;
                                 }
                                /*
                                 * if VSTART == VSTOP then toggle
@@ -552,8 +619,7 @@ parmrk:
                 */
                if (c == '\r') {
                        if (ISSET(iflag, IGNCR)) {
-                               thread_funnel_set(kernel_flock, funnel_state);
-                               return (0);
+                               goto out;
                         }
                        else if (ISSET(iflag, ICRNL))
                                c = '\n';
@@ -650,10 +716,14 @@ parmrk:
                 * ^T - kernel info and generate SIGINFO
                 */
                if (CCEQ(cc[VSTATUS], c) && ISSET(lflag, IEXTEN)) {
-                       if (ISSET(lflag, ISIG))
+                       if (ISSET(lflag, ISIG)) {
+                               /* SAFE: All callers drop the lock on return */
+                               tty_unlock(tp);
                                tty_pgsignal(tp, SIGINFO, 1);
+                               tty_lock(tp);
+                       }
                        if (!ISSET(lflag, NOKERNINFO))
-                               ttyinfo(tp);
+                               ttyinfo_locked(tp);
                        goto endcase;
                }
        }
@@ -709,29 +779,44 @@ input_overflow:
                        }
                }
        }
+
 endcase:
        /*
         * IXANY means allow any character to restart output.
         */
        if (ISSET(tp->t_state, TS_TTSTOP) &&
            !ISSET(iflag, IXANY) && cc[VSTART] != cc[VSTOP]) {
-               thread_funnel_set(kernel_flock, funnel_state);
-               return (0);
+               goto out;
         }
+
 restartoutput:
        CLR(tp->t_lflag, FLUSHO);
        CLR(tp->t_state, TS_TTSTOP);
+
 startoutput:
-    retval = ttstart(tp);
-       thread_funnel_set(kernel_flock, funnel_state);
+       /* Start the output */
+       retval = ttstart(tp);
+
+out:
        return (retval);
 }
 
+
 /*
+ * ttyoutput
+ *
  * Output a single character on a tty, doing output processing
  * as needed (expanding tabs, newline processing, etc.).
- * Returns < 0 if succeeds, otherwise returns char to resend.
- * Must be recursive.
+ *
+ * Parameters: c                       The character to output
+ *             tp                      The tty on which to output on the tty
+ *
+ * Returns:    < 0                     Success
+ *             >= 0                    Character to resend (failure)
+ *
+ * Locks:      Assumes tp is locked on entry, remains locked on exit
+ *
+ * Notes:      Must be recursive.
  */
 static int
 ttyoutput(int c, struct tty *tp)
@@ -739,6 +824,8 @@ ttyoutput(int c, struct tty *tp)
        tcflag_t oflag;
        int col;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        oflag = tp->t_oflag;
        if (!ISSET(oflag, OPOST)) {
                if (ISSET(tp->t_lflag, FLUSHO))
@@ -815,19 +902,141 @@ ttyoutput(int c, struct tty *tp)
 }
 
 /*
- * Ioctls for all tty devices.  Called after line-discipline specific ioctl
- * has been called to do discipline-specific functions and/or reject any
- * of these ioctl commands.
+ * Sets the tty state to not allow any more changes of foreground process
+ * group. This is required to be done so that a subsequent revoke on a vnode
+ * is able to always successfully complete.
+ *
+ * Locks :   Assumes tty_lock held on entry
+ */
+void
+ttysetpgrphup(struct tty *tp)
+{
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+       SET(tp->t_state, TS_PGRPHUP);
+       /*
+        * Also wake up sleeping readers which may or may not belong to the
+        * current foreground process group.
+        *
+        * This forces any non-fg readers (which entered read when
+        * that process group was in the fg) to return with EIO (if they're
+        * catching SIGTTIN or with SIGTTIN). The ones which do belong to the fg
+        * process group will promptly go back to sleep and get a SIGHUP shortly
+        * This would normally happen as part of the close in revoke but if
+        * there is a sleeping reader from a non-fg process group we never get
+        * to the close because the sleeping reader holds an iocount on the
+        * vnode of the terminal which is going to get revoked->reclaimed.
+        */
+       wakeup(TSA_HUP_OR_INPUT(tp));
+}
+
+/*
+ * Locks : Assumes tty lock held on entry
+ */
+void
+ttyclrpgrphup(struct tty *tp)
+{
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+       CLR(tp->t_state, TS_PGRPHUP);
+}
+
+/*
+ * ttioctl
+ *
+ * Identical to ttioctl_locked, only the lock is not held
+ *
+ * Parameters: <See ttioctl_locked()>
+ *
+ * Returns:    <See ttioctl_locked()>
+ *
+ * Locks:      This function assumes the tty_lock() is not held on entry;
+ *             it takes the lock, and releases it before returning.
+ *
+ * Notes:      This is supported to ensure the line discipline interfaces
+ *             all have the same locking semantics.
+ *
+ *             This function is called from 
+ */
+int
+ttioctl(struct tty *tp, u_long cmd, caddr_t data, int flag, proc_t p)
+{
+       int     retval;
+
+       tty_lock(tp);
+       retval = ttioctl_locked(tp, cmd, data, flag, p);
+       tty_unlock(tp);
+
+       return (retval);
+}
+
+
+/*
+ * ttioctl_locked
+ *
+ * Ioctls for all tty devices.
+ *
+ * Parameters: tp                      Tty on which ioctl() is being called
+ *             cmd                     ioctl() command parameter
+ *             data                    ioctl() data argument (if any)
+ *             flag                    fileglob open modes from fcntl.h;
+ *                                     if called internally, this is usually
+ *                                     set to 0, rather than something useful
+ *             p                       Process context for the call; if the
+ *                                     call is proxied to a worker thread,
+ *                                     this will not be the current process!!!
+ *
+ * Returns:    0                       Success
+ *             EIO                     I/O error (no process group, job
+ *                                     control, etc.)
+ *             EINTR                   Interrupted by signal
+ *             EBUSY                   Attempt to become the console while
+ *                                     the console is busy
+ *             ENOTTY                  TIOCGPGRP on a non-controlling tty
+ *             EINVAL                  Invalid baud rate
+ *             ENXIO                   TIOCSETD of invalid line discipline
+ *             EPERM                   TIOCSTI, not root, not open for read
+ *             EACCES                  TIOCSTI, not root, not your controlling
+ *                                     tty
+ *             EPERM                   TIOCSCTTY failed
+ *             ENOTTY/EINVAL/EPERM     TIOCSPGRP failed
+ *             EPERM                   TIOCSDRAINWAIT as non-root user
+ *     suser:EPERM                     Console control denied
+ *     ttywait:EIO                     t_timeout too small/expired
+ *     ttywait:ERESTART                Upper layer must redrive the call;
+ *                                     this is usually done by the Libc
+ *                                     stub in user space
+ *     ttywait:EINTR                   Interrupted (usually a signal)
+ *     ttcompat:EINVAL
+ *     ttcompat:ENOTTY
+ *     ttcompat:EIOCTL
+ *     ttcompat:ENOTTY                 TIOCGSID, if no session or session
+ *                                     leader
+ *     ttcompat:ENOTTY                 All unrecognized ioctls
+ *     *tp->t_param:?                  TIOCSETA* underlying function
+ *     *linesw[t].l_open:?             TIOCSETD line discipline open failure
+ *
+ *
+ * Locks:      This function assumes that the tty_lock() is held for the
+ *             tp at the time of the call.  The lock remains held on return.
+ *
+ * Notes:      This function is called after line-discipline specific ioctl
+ *             has been called to do discipline-specific functions and/or
+ *             reject any of these ioctl() commands.
+ *
+ *             This function calls ttcompat(), which can re-call ttioctl()
+ *             to a depth of one (FORTRAN style mutual recursion); at some
+ *             point, we should just in-line ttcompat() here.
  */
 int
-ttioctl(struct tty *tp,
-       u_long cmd, caddr_t data, int flag,
-       proc_t p)
+ttioctl_locked(struct tty *tp, u_long cmd, caddr_t data, int flag, proc_t p)
 {
        int error = 0;
+       int bogusData = 1;
        struct uthread *ut;
-       struct pgrp * pg, *oldpg;
-       struct session *sessp, * oldsessp;
+       struct pgrp *pg, *oldpg;
+       struct session *sessp, *oldsessp;
+       struct tty *oldtp;
+
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        ut = (struct uthread *)get_bsdthread_info(current_thread());
        /* If the ioctl involves modification, signal if in the background. */
@@ -838,18 +1047,17 @@ ttioctl(struct tty *tp,
        case  TIOCFLUSH:
        case TIOCSTOP:
        case TIOCSTART:
-       case  TIOCSETA:
+       case  TIOCSETA_32:
        case  TIOCSETA_64:
        case  TIOCSETD:
-       case  TIOCSETAF:
+       case  TIOCSETAF_32:
        case  TIOCSETAF_64:
-       case  TIOCSETAW:
+       case  TIOCSETAW_32:
        case  TIOCSETAW_64:
        case  TIOCSPGRP:
        case  TIOCSTAT:
        case  TIOCSTI:
        case  TIOCSWINSZ:
-#if COMPAT_43_TTY || defined(COMPAT_SUNOS)
        case  TIOCLBIC:
        case  TIOCLBIS:
        case  TIOCLSET:
@@ -858,7 +1066,6 @@ ttioctl(struct tty *tp,
        case  TIOCSETN:
        case  TIOCSETP:
        case  TIOCSLTC:
-#endif
                while (isbackground(p, tp) &&
                    (p->p_lflag & P_LPPWAIT) == 0 &&
                    (p->p_sigignore & sigmask(SIGTTOU)) == 0 &&
@@ -868,13 +1075,17 @@ ttioctl(struct tty *tp,
                                error = EIO;
                                goto out;
                        }
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
                        if (pg->pg_jobc == 0) {
                                pg_rele(pg);
+                               tty_lock(tp);
                                error = EIO;
                                goto out;
                        }
                        pgsignal(pg, SIGTTOU, 1);
                        pg_rele(pg);
+                       tty_lock(tp);
 
 
                        /*
@@ -916,7 +1127,6 @@ ttioctl(struct tty *tp,
        }
        case TIOCSCONS: {
                /* Set current console device to this line */
-               int bogusData = 1;
                data = (caddr_t) &bogusData;
 
                /* No break - Fall through to BSD code */
@@ -935,8 +1145,8 @@ ttioctl(struct tty *tp,
                        constty = NULL;
                }
                if (constty) {
-                       (*cdevsw[major(cons.t_dev)].d_ioctl)
-                               (cons.t_dev, KMIOCDISABLCONS, NULL, 0, p);
+                       (*cdevsw[major(constty->t_dev)].d_ioctl)
+                               (constty->t_dev, KMIOCDISABLCONS, NULL, 0, p);
                } else {
                        (*cdevsw[major(tp->t_dev)].d_ioctl)
                                (tp->t_dev, KMIOCDISABLCONS, NULL, 0, p);
@@ -948,15 +1158,20 @@ ttioctl(struct tty *tp,
                if (error)
                        goto out;
                break;
-       case TIOCGETA:          /* get termios struct */
-       case TIOCGETA_64: {             /* get termios struct */
-               if (IS_64BIT_PROCESS(p)) {
-                       termios32to64(&tp->t_termios, (struct user_termios *)data);
-               } else {
-                       bcopy(&tp->t_termios, data, sizeof(struct termios));
-               }
+       case TIOCGETA_32:               /* get termios struct */
+#ifdef __LP64__
+               termios64to32((struct user_termios *)&tp->t_termios, (struct termios32 *)data);
+#else
+               bcopy(&tp->t_termios, data, sizeof(struct termios));
+#endif
+               break;
+       case TIOCGETA_64:               /* get termios struct */
+#ifdef __LP64__
+               bcopy(&tp->t_termios, data, sizeof(struct termios));
+#else
+               termios32to64((struct termios32 *)&tp->t_termios, (struct user_termios *)data);
+#endif
                break;
-       }
        case TIOCGETD:                  /* get line discipline */
                *(int *)data = tp->t_line;
                break;
@@ -981,19 +1196,27 @@ ttioctl(struct tty *tp,
        case TIOCOUTQ:                  /* output queue size */
                *(int *)data = tp->t_outq.c_cc;
                break;
-       case TIOCSETA:                  /* set termios struct */
+       case TIOCSETA_32:                       /* set termios struct */
        case TIOCSETA_64:
-       case TIOCSETAW:                 /* drain output, set */
+       case TIOCSETAW_32:                      /* drain output, set */
        case TIOCSETAW_64:
-       case TIOCSETAF:         /* drn out, fls in, set */
-       case TIOCSETAF_64: {            /* drn out, fls in, set */
+       case TIOCSETAF_32:              /* drn out, fls in, set */
+       case TIOCSETAF_64:
+       {               /* drn out, fls in, set */
                struct termios *t = (struct termios *)data;
                struct termios lcl_termios;
 
-               if (IS_64BIT_PROCESS(p)) {
-                       termios64to32((struct user_termios *)data, &lcl_termios);
+#ifdef __LP64__
+               if (cmd==TIOCSETA_32 || cmd==TIOCSETAW_32 || cmd==TIOCSETAF_32) {
+                       termios32to64((struct termios32 *)data, (struct user_termios *)&lcl_termios);
+                       t = &lcl_termios;
+               }
+#else
+               if (cmd==TIOCSETA_64 || cmd==TIOCSETAW_64 || cmd==TIOCSETAF_64) {
+                       termios64to32((struct user_termios *)data, (struct termios32 *)&lcl_termios);
                        t = &lcl_termios;
                }
+#endif
 #if 0
        /* XXX bogus test; always false */
                if (t->c_ispeed < 0 || t->c_ospeed < 0) {
@@ -1003,13 +1226,13 @@ ttioctl(struct tty *tp,
 #endif /* 0 - leave in; may end up being a conformance issue */
                if (t->c_ispeed == 0)
                        t->c_ispeed = t->c_ospeed;
-               if (cmd == TIOCSETAW || cmd == TIOCSETAF ||
+               if (cmd == TIOCSETAW_32 || cmd == TIOCSETAF_32 ||
                    cmd == TIOCSETAW_64 || cmd == TIOCSETAF_64) {
                        error = ttywait(tp);
                        if (error) {
                                goto out;
                        }
-                       if (cmd == TIOCSETAF || cmd == TIOCSETAF_64)
+                       if (cmd == TIOCSETAF_32 || cmd == TIOCSETAF_64)
                                ttyflush(tp, FREAD);
                }
                if (!ISSET(t->c_cflag, CIGNORE)) {
@@ -1045,7 +1268,7 @@ ttioctl(struct tty *tp,
                        ttsetwater(tp);
                }
                if (ISSET(t->c_lflag, ICANON) != ISSET(tp->t_lflag, ICANON) &&
-                   cmd != TIOCSETAF && cmd != TIOCSETAF_64) {
+                   cmd != TIOCSETAF_32 && cmd != TIOCSETAF_64) {
                        if (ISSET(t->c_lflag, ICANON))
                                SET(tp->t_lflag, PENDIN);
                        else {
@@ -1091,10 +1314,15 @@ ttioctl(struct tty *tp,
                        error = ENXIO;
                        goto out;
                }
+               /*
+                * If the new line discipline is not equal to the old one,
+                * close the old one and open the new one.
+                */
                if (t != tp->t_line) {
                        (*linesw[tp->t_line].l_close)(tp, flag);
                        error = (*linesw[t].l_open)(device, tp);
                        if (error) {
+                               /* This is racy; it's possible to lose both */
                                (void)(*linesw[tp->t_line].l_open)(device, tp);
                                goto out;
                        }
@@ -1135,18 +1363,58 @@ ttioctl(struct tty *tp,
                break;
        case TIOCSCTTY:                 /* become controlling tty */
                /* Session ctty vnode pointer set in vnode layer. */
-               pg = proc_pgrp(p);
                sessp = proc_session(p);
-               if (!SESS_LEADER(p, sessp) ||
-                   ((sessp->s_ttyvp || tp->t_session) &&
-                   (tp->t_session != sessp))) {
-                       if (sessp != SESSION_NULL)
-                               session_rele(sessp);
-                       if (pg != PGRP_NULL)
+               if (sessp == SESSION_NULL) {
+                       error = EPERM;
+                       goto out;
+               }
+
+               /*
+                * This can only be done by a session leader.
+                */
+               if (!SESS_LEADER(p, sessp)) {
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
+                       session_rele(sessp);
+                       tty_lock(tp);
+                       error = EPERM;
+                       goto out;
+               }
+               /*
+                * If this terminal is already the controlling terminal for the
+                * session, nothing to do here.
+                */
+               if (tp->t_session == sessp) {
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
+                       session_rele(sessp);
+                       tty_lock(tp);
+                       error = 0;
+                       goto out;
+               }
+               pg = proc_pgrp(p);
+               /*
+                * Deny if the terminal is already attached to another session or
+                * the session already has a terminal vnode.
+                */
+               session_lock(sessp);
+               if (sessp->s_ttyvp || tp->t_session) {
+                       session_unlock(sessp);
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
+                       if (pg != PGRP_NULL) {
                                pg_rele(pg);
+                       }
+                       session_rele(sessp);
+                       tty_lock(tp);
                        error = EPERM;
                        goto out;
                }
+               sessp->s_ttypgrpid = pg->pg_id;
+               oldtp = sessp->s_ttyp;
+               ttyhold(tp);
+               sessp->s_ttyp = tp;
+               session_unlock(sessp);
                proc_list_lock();
                oldsessp = tp->t_session;
                oldpg = tp->t_pgrp;
@@ -1154,18 +1422,19 @@ ttioctl(struct tty *tp,
                        oldsessp->s_ttypgrpid = NO_PID;
                /* do not drop refs on sessp and pg as tp holds them */
                tp->t_session = sessp;
-               sessp->s_ttypgrpid = pg->pg_id;
                tp->t_pgrp = pg;
                proc_list_unlock();
-               session_lock(sessp);
-               sessp->s_ttyp = tp;
-               session_unlock(sessp);
-               OSBitOrAtomic(P_CONTROLT, (UInt32 *)&p->p_flag);
+               OSBitOrAtomic(P_CONTROLT, &p->p_flag);
+               /* SAFE: All callers drop the lock on return */
+               tty_unlock(tp);
                /* drop the reference on prev session and pgrp */
                if (oldsessp != SESSION_NULL)
                        session_rele(oldsessp);
                if (oldpg != PGRP_NULL)
                        pg_rele(oldpg);
+               if (NULL != oldtp)
+                       ttyfree(oldtp);
+               tty_lock(tp);
                break;
 
        case TIOCSPGRP: {               /* set pgrp of tty */
@@ -1184,6 +1453,21 @@ ttioctl(struct tty *tp,
                        error = EINVAL;
                        goto out;
                 } else if (pgrp->pg_session != sessp) {
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
+                       if (sessp != SESSION_NULL)
+                               session_rele(sessp);
+                       pg_rele(pgrp);
+                       tty_lock(tp);
+                       error = EPERM;
+                       goto out;
+               }
+               /*
+                * The session leader is going away and is possibly going to revoke
+                * the terminal, we can't change the process group when that is the
+                * case.
+                */
+               if (ISSET(tp->t_state, TS_PGRPHUP)) {
                        if (sessp != SESSION_NULL)
                                session_rele(sessp);
                        pg_rele(pgrp);
@@ -1195,20 +1479,43 @@ ttioctl(struct tty *tp,
                tp->t_pgrp = pgrp;
                sessp->s_ttypgrpid = pgrp->pg_id;
                proc_list_unlock();
+
+               /*
+                * Wakeup readers to recheck if they are still the foreground
+                * process group.
+                *
+                * ttwakeup() isn't called because the readers aren't getting
+                * woken up becuse there is something to read but to force
+                * the re-evaluation of their foreground process group status.
+                *
+                * Ordinarily leaving these readers waiting wouldn't be an issue
+                * as launchd would send them a termination signal eventually
+                * (if nobody else does). But if this terminal happens to be
+                * /dev/console, launchd itself could get blocked forever behind
+                * a revoke of /dev/console and leave the system deadlocked.
+                */
+               wakeup(TSA_HUP_OR_INPUT(tp));
+
+               /* SAFE: All callers drop the lock on return */
+               tty_unlock(tp);
                if (oldpg != PGRP_NULL)
                        pg_rele(oldpg);
                if (sessp != SESSION_NULL)
                        session_rele(sessp);
+               tty_lock(tp);
                break;
        }
        case TIOCSTAT:                  /* simulate control-T */
-               ttyinfo(tp);
+               ttyinfo_locked(tp);
                break;
        case TIOCSWINSZ:                /* set window size */
                if (bcmp((caddr_t)&tp->t_winsize, data,
                    sizeof (struct winsize))) {
                        tp->t_winsize = *(struct winsize *)data;
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
                        tty_pgsignal(tp, SIGWINCH, 1);
+                       tty_lock(tp);
                }
                break;
        case TIOCSDRAINWAIT:
@@ -1224,11 +1531,7 @@ ttioctl(struct tty *tp,
                *(int *)data = tp->t_timeout / hz;
                break; 
        default:
-#if COMPAT_43_TTY || defined(COMPAT_SUNOS)
                error = ttcompat(tp, cmd, data, flag, p);
-#else
-               error = ENOTTY;
-#endif
                goto out;
        }
 
@@ -1237,48 +1540,99 @@ out:
        return(error);
 }
 
+
+/*
+ * Locks:      Assumes tp is locked on entry, remains locked on exit
+ */
 int
 ttyselect(struct tty *tp, int rw, void *wql, proc_t p)
 {
-       if (tp == NULL)
-               return (ENXIO);
+       int retval = 0;
+       /*
+        * Attaching knotes to TTYs needs to call selrecord in order to hook
+        * up the waitq to the selinfo, regardless of data being ready.  See
+        * filt_ttyattach.
+        */
+       bool needs_selrecord = rw & FMARK;
+       rw &= ~FMARK;
+
+       if (tp == NULL) {
+               return ENXIO;
+       }
+
+       TTY_LOCK_OWNED(tp);
+
+       if (tp->t_state & TS_ZOMBIE) {
+               retval = 1;
+               goto out;
+       }
 
        switch (rw) {
        case FREAD:
-               if (ttnread(tp) > 0 || ISSET(tp->t_state, TS_ZOMBIE))
-                       return(1);
+               retval = ttnread(tp);
+               if (retval > 0) {
+                       break;
+               }
+
                selrecord(p, &tp->t_rsel, wql);
                break;
        case FWRITE:
-               if ((tp->t_outq.c_cc <= tp->t_lowat &&
-                    ISSET(tp->t_state, TS_CONNECTED))
-                   || ISSET(tp->t_state, TS_ZOMBIE)) {
-                       return (1);
+               if ((tp->t_outq.c_cc <= tp->t_lowat) &&
+                   (tp->t_state & TS_CONNECTED)) {
+                       retval = tp->t_hiwat - tp->t_outq.c_cc;
+                       break;
                }
+
                selrecord(p, &tp->t_wsel, wql);
                break;
        }
-       return (0);
+
+out:
+       if (retval > 0 && needs_selrecord) {
+               switch (rw) {
+               case FREAD:
+                       selrecord(p, &tp->t_rsel, wql);
+                       break;
+               case FWRITE:
+                       selrecord(p, &tp->t_wsel, wql);
+                       break;
+               }
+       }
+
+       return retval;
 }
 
+
 /*
  * This is a wrapper for compatibility with the select vector used by
  * cdevsw.  It relies on a proper xxxdevtotty routine.
+ *
+ * Locks:      Assumes tty_lock() is not held prior to calling.
  */
 int
 ttselect(dev_t dev, int rw, void *wql, proc_t p)
 {
-       return ttyselect(cdevsw[major(dev)].d_ttys[minor(dev)], rw, wql, p);
+       int     rv;
+       struct tty *tp = cdevsw[major(dev)].d_ttys[minor(dev)];
+
+       tty_lock(tp);
+       rv =  ttyselect(tp, rw, wql, p);
+       tty_unlock(tp);
+
+       return (rv);
 }
 
+
 /*
- * Must be called at spltty().
+ * Locks:      Assumes tp is locked on entry, remains locked on exit
  */
-static int
+__private_extern__ int
 ttnread(struct tty *tp)
 {
        int nread;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        if (ISSET(tp->t_lflag, PENDIN))
                ttypend(tp);
        nread = tp->t_canq.c_cc;
@@ -1290,14 +1644,32 @@ ttnread(struct tty *tp)
        return (nread);
 }
 
+
 /*
+ * ttywait
+ *
  * Wait for output to drain.
+ *
+ * Parameters: tp                      Tty on which to wait for output to drain
+ *
+ * Returns:    0                       Success
+ *             EIO                     t_timeout too small/expired
+ *     ttysleep:ERESTART               Upper layer must redrive the call;
+ *                                     this is usually done by the Libc
+ *                                     stub in user space
+ *     ttysleep:EINTR                  Interrupted (usually a signal)
+ *
+ * Notes:      Called from proc_exit() and vproc_exit().
+ *
+ * Locks:      Assumes tp is locked on entry, remains locked on exit
  */
 int
 ttywait(struct tty *tp)
 {
        int error;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        error = 0;
        while ((tp->t_outq.c_cc || ISSET(tp->t_state, TS_BUSY)) &&
               ISSET(tp->t_state, TS_CONNECTED) && tp->t_oproc) {
@@ -1321,20 +1693,31 @@ ttywait(struct tty *tp)
        return (error);
 }
 
+/*
+ * Stop the underlying device driver.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
+ */
 static void
 ttystop(struct tty *tp, int rw)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        (*cdevsw[major(tp->t_dev)].d_stop)(tp, rw);
 }
 
 /*
  * Flush if successfully wait.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 static int
 ttywflush(struct tty *tp)
 {
        int error;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        if ((error = ttywait(tp)) == 0)
                ttyflush(tp, FREAD);
        return (error);
@@ -1342,10 +1725,14 @@ ttywflush(struct tty *tp)
 
 /*
  * Flush tty read and/or write queues, notifying anyone waiting.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 void
 ttyflush(struct tty *tp, int rw)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
 #if 0
 again:
 #endif
@@ -1407,6 +1794,10 @@ again:
 
 /*
  * Copy in the default termios characters.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
+ *
+ * Notes:      No assertion; tp is not in scope.
  */
 void
 termioschars(struct termios *t)
@@ -1414,23 +1805,19 @@ termioschars(struct termios *t)
        bcopy(ttydefchars, t->c_cc, sizeof t->c_cc);
 }
 
-/*
- * Old interface.
- */
-void
-ttychars(struct tty *tp)
-{
-       termioschars(&tp->t_termios);
-}
 
 /*
  * Handle input high water.  Send stop character for the IXOFF case.  Turn
  * on our input flow control bit and propagate the changes to the driver.
  * XXX the stop character should be put in a special high priority queue.
+ *
+ * Locks:      Assumes tty_lock() is held for the call.
  */
 void
 ttyblock(struct tty *tp)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        SET(tp->t_state, TS_TBLOCK);
        if (ISSET(tp->t_iflag, IXOFF) && tp->t_cc[VSTOP] != _POSIX_VDISABLE &&
            putc(tp->t_cc[VSTOP], &tp->t_outq) != 0)
@@ -1438,14 +1825,19 @@ ttyblock(struct tty *tp)
        ttstart(tp);
 }
 
+
 /*
  * Handle input low water.  Send start character for the IXOFF case.  Turn
  * off our input flow control bit and propagate the changes to the driver.
  * XXX the start character should be put in a special high priority queue.
+ *
+ * Locks:      Assumes tty_lock() is held for the call.
  */
 static void
 ttyunblock(struct tty *tp)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        CLR(tp->t_state, TS_TBLOCK);
        if (ISSET(tp->t_iflag, IXOFF) && tp->t_cc[VSTART] != _POSIX_VDISABLE &&
            putc(tp->t_cc[VSTART], &tp->t_outq) != 0)
@@ -1453,65 +1845,69 @@ ttyunblock(struct tty *tp)
        ttstart(tp);
 }
 
-/* FreeBSD: Not used by any current (i386) drivers. */
+
 /*
- * Restart after an inter-char delay.
+ * ttstart
+ *
+ * Start tty output
+ *
+ * Parameters: tp                      tty on which to start output
+ *
+ * Returns:    0                       Success
+ *
+ * Locks:      Assumes tty_lock() is held for the call.
+ *
+ * Notes:      This function might as well be void; it always returns success
+ *
+ *             Called from ttioctl_locked(), LDISC routines, and
+ *             ttycheckoutq(), ttyblock(), ttyunblock(), and tputchar()
  */
-void
-ttrstrt(void *tp_arg)
-{
-       struct tty *tp;
-
-#if DIAGNOSTIC
-       if (tp_arg == NULL)
-               panic("ttrstrt");
-#endif
-       tp = tp_arg;
-
-       CLR(tp->t_state, TS_TIMEOUT);
-       ttstart(tp);
-
-}
-
 int
 ttstart(struct tty *tp)
 {
-       boolean_t funnel_state;
-
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        if (tp->t_oproc != NULL)        /* XXX: Kludge for pty. */
                (*tp->t_oproc)(tp);
-       thread_funnel_set(kernel_flock, funnel_state);
+
        return (0);
 }
 
+
 /*
+ * ttylclose (LDISC)
+ *
  * "close" a line discipline
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttylclose(struct tty *tp, int flag)
 {
-       boolean_t funnel_state;
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
        if ( (flag & FNONBLOCK) || ttywflush(tp))
                ttyflush(tp, FREAD | FWRITE);
-       thread_funnel_set(kernel_flock, funnel_state);
+
        return (0);
 }
 
+
 /*
+ * ttymodem (LDISC)
+ *
  * Handle modem control transition on a tty.
  * Flag indicates new state of carrier.
  * Returns 0 if the line should be turned off, otherwise 1.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttymodem(struct tty *tp, int flag)
 {
-       boolean_t funnel_state;
+       int rval = 1;           /* default return value */
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        if (ISSET(tp->t_state, TS_CARR_ON) && ISSET(tp->t_cflag, MDMBUF)) {
                /*
@@ -1540,8 +1936,8 @@ ttymodem(struct tty *tp, int flag)
                        if (tp->t_session && tp->t_session->s_leader)
                                psignal(tp->t_session->s_leader, SIGHUP);
                        ttyflush(tp, FREAD | FWRITE);
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (0);
+                       rval = 0;
+                       goto out;
                }
        } else {
                /*
@@ -1554,13 +1950,17 @@ ttymodem(struct tty *tp, int flag)
                ttwakeup(tp);
                ttwwakeup(tp);
        }
-       thread_funnel_set(kernel_flock, funnel_state);
-       return (1);
+
+out:
+       return (rval);
 }
 
+
 /*
  * Reinput pending characters after state switch
  * call at spltty().
+ *
+ * Locks:      Assumes tty_lock() is held for the call.
  */
 static void
 ttypend(struct tty *tp)
@@ -1568,6 +1968,8 @@ ttypend(struct tty *tp)
        struct clist tq;
        int c;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        CLR(tp->t_lflag, PENDIN);
        SET(tp->t_state, TS_TYPEN);
        tq = tp->t_rawq;
@@ -1578,8 +1980,13 @@ ttypend(struct tty *tp)
        CLR(tp->t_state, TS_TYPEN);
 }
 
+
 /*
+ * ttread (LDISC)
+ *
  * Process a read call on a tty device.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttread(struct tty *tp, struct uio *uio, int flag)
@@ -1592,11 +1999,10 @@ ttread(struct tty *tp, struct uio *uio, int flag)
        int first, error = 0;
        int has_etime = 0, last_cc = 0;
        long slp = 0;           /* XXX this should be renamed `timo'. */
-       boolean_t funnel_state;
        struct uthread *ut;
        struct pgrp * pg;
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        ut = (struct uthread *)get_bsdthread_info(current_thread());
 
@@ -1617,21 +2023,27 @@ loop:
                if ((p->p_sigignore & sigmask(SIGTTIN)) ||
                   (ut->uu_sigmask & sigmask(SIGTTIN)) ||
                    p->p_lflag & P_LPPWAIT) {
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (EIO);
+                       error = EIO;
+                       goto err;
                }
                pg = proc_pgrp(p);
                if (pg == PGRP_NULL) {
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (EIO);
+                       error = EIO;
+                       goto err;
                }
                if (pg->pg_jobc == 0) {
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
                        pg_rele(pg);
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (EIO);
+                       tty_lock(tp);
+                       error = EIO;
+                       goto err;
                }
+               /* SAFE: All callers drop the lock on return */
+               tty_unlock(tp);
                pgsignal(pg, SIGTTIN, 1);
                pg_rele(pg);
+               tty_lock(tp);
 
                /*
                 * We signalled ourself, so we need to act as if we
@@ -1639,13 +2051,13 @@ loop:
                 * the signal.  If it's a signal that stops the
                 * process, that's handled in the signal sending code.
                 */
-                       thread_funnel_set(kernel_flock, funnel_state);
-               return (EINTR);
+               error = EINTR;
+               goto err;
        }
 
        if (ISSET(tp->t_state, TS_ZOMBIE)) {
-               thread_funnel_set(kernel_flock, funnel_state);
-               return (0);     /* EOF */
+               /* EOF - returning 0 */
+               goto err;
        }
 
        /*
@@ -1659,12 +2071,11 @@ loop:
        if (flag & IO_NDELAY) {
                if (qp->c_cc > 0)
                        goto read;
-               if (!ISSET(lflag, ICANON) && cc[VMIN] == 0) {
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (0);
+               if (ISSET(lflag, ICANON) || cc[VMIN] != 0) {
+                       error = EWOULDBLOCK;
                }
-               thread_funnel_set(kernel_flock, funnel_state);
-               return (EWOULDBLOCK);
+               /* else polling - returning 0 */
+               goto err;
        }
        if (!ISSET(lflag, ICANON)) {
                int m = cc[VMIN];
@@ -1687,8 +2098,7 @@ loop:
                                goto read;
 
                        /* m, t and qp->c_cc are all 0.  0 is enough input. */
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (0);
+                       goto err;
                }
                t *= 100000;            /* time in us */
 #define diff(t1, t2) (((t1).tv_sec - (t2).tv_sec) * 1000000 + \
@@ -1739,8 +2149,7 @@ loop:
                        } else {
                                if (timercmp(&etime, &timecopy, <=)) {
                                        /* Timed out, but 0 is enough input. */
-                                       thread_funnel_set(kernel_flock, funnel_state);
-                                       return (0);
+                                       goto err;
                                }
                                slp = diff(etime, timecopy);
                        }
@@ -1755,7 +2164,7 @@ loop:
                 * is large (divide by `tick' and/or arrange to
                 * use hzto() if hz is large).
                 */
-               slp = (long) (((u_long)slp * hz) + 999999) / 1000000;
+               slp = (long) (((u_int32_t)slp * hz) + 999999) / 1000000;
                goto sleep;
        }
        if (qp->c_cc <= 0) {
@@ -1769,8 +2178,7 @@ sleep:
                if (error == EWOULDBLOCK)
                        error = 0;
                else if (error) {
-                       thread_funnel_set(kernel_flock, funnel_state);
-                       return (error);
+                       goto err;
                }
                /*
                 * XXX what happens if another process eats some input
@@ -1793,7 +2201,7 @@ read:
                char ibuf[IBUFSIZ];
                int icc;
 
-               icc = min(uio_resid(uio), IBUFSIZ);
+               icc = MIN(uio_resid(uio), IBUFSIZ);
                icc = q_to_b(qp, (u_char *)ibuf, icc);
                if (icc <= 0) {
                        if (first)
@@ -1825,10 +2233,17 @@ slowcase:
                 */
                if (CCEQ(cc[VDSUSP], c) &&
                    ISSET(lflag, IEXTEN | ISIG) == (IEXTEN | ISIG)) {
+                       /*
+                        * SAFE: All callers drop the lock on return and
+                        * SAFE: current thread will not change out from
+                        * SAFE: under us in the "goto loop" case.
+                        */
+                       tty_unlock(tp);
                        tty_pgsignal(tp, SIGTSTP, 1);
+                       tty_lock(tp);
                        if (first) {
-                               error = ttysleep(tp, &lbolt, TTIPRI | PCATCH,
-                                                "ttybg3", 0);
+                               error = ttysleep(tp, &ttread, TTIPRI | PCATCH,
+                                                "ttybg3", hz);
                                if (error)
                                        break;
                                goto loop;
@@ -1867,16 +2282,21 @@ out:
            tp->t_rawq.c_cc + tp->t_canq.c_cc <= I_LOW_WATER)
                ttyunblock(tp);
 
-       thread_funnel_set(kernel_flock, funnel_state);
+err:
        return (error);
 }
 
+
 /*
  * Check the output queue on tp for space for a kernel message (from uprintf
  * or tprintf).  Allow some space over the normal hiwater mark so we don't
  * lose messages due to normal flow control, but don't let the tty run amok.
  * Sleeps here are not interruptible, but we return prematurely if new signals
  * arrive.
+ *
+ * Locks:      Assumes tty_lock() is held before calling
+ *
+ * Notes:      This function is called from tprintf() in subr_prf.c
  */
 int
 ttycheckoutq(struct tty *tp, int wait)
@@ -1885,6 +2305,8 @@ ttycheckoutq(struct tty *tp, int wait)
        sigset_t oldsig;
        struct uthread *ut;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        ut = (struct uthread *)get_bsdthread_info(current_thread());
 
        hiwat = tp->t_hiwat;
@@ -1898,13 +2320,18 @@ ttycheckoutq(struct tty *tp, int wait)
                                return (0);
                        }
                        SET(tp->t_state, TS_SO_OLOWAT);
-                       tsleep(TSA_OLOWAT(tp), PZERO - 1, "ttoutq", hz);
+                       ttysleep(tp, TSA_OLOWAT(tp), PZERO - 1, "ttoutq", hz);
                }
        return (1);
 }
 
+
 /*
+ * ttwrite (LDISC)
+ *
  * Process a write call on a tty device.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 int
 ttwrite(struct tty *tp, struct uio *uio, int flag)
@@ -1912,17 +2339,16 @@ ttwrite(struct tty *tp, struct uio *uio, int flag)
        char *cp = NULL;
        int cc, ce;
        proc_t p;
-       int i, hiwat, count, error;
+       int i, hiwat, error;
+       user_ssize_t count;
        char obuf[OBUFSIZ];
-       boolean_t funnel_state;
        struct uthread *ut;
        struct pgrp * pg;
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        ut = (struct uthread *)get_bsdthread_info(current_thread());
        hiwat = tp->t_hiwat;
-       // LP64todo - fix this!
        count = uio_resid(uio);
        error = 0;
        cc = 0;
@@ -1958,12 +2384,18 @@ loop:
                        goto out;
                }
                if (pg->pg_jobc == 0) {
+                       /* SAFE: All callers drop the lock on return */
+                       tty_unlock(tp);
                        pg_rele(pg);
+                       tty_lock(tp);
                        error = EIO;
                        goto out;
                }
+               /* SAFE: All callers drop the lock on return */
+               tty_unlock(tp);
                pgsignal(pg, SIGTTOU, 1);
                pg_rele(pg);
+               tty_lock(tp);
                /*
                 * We signalled ourself, so we need to act as if we
                 * have been "interrupted" from a "sleep" to act on
@@ -1981,7 +2413,6 @@ loop:
        while (uio_resid(uio) > 0 || cc > 0) {
                if (ISSET(tp->t_lflag, FLUSHO)) {
                        uio_setresid(uio, 0);
-                       thread_funnel_set(kernel_flock, funnel_state);
                        return (0);
                }
                if (tp->t_outq.c_cc > hiwat)
@@ -1991,7 +2422,7 @@ loop:
                 * leftover from last time.
                 */
                if (cc == 0) {
-                       cc = min(uio_resid(uio), OBUFSIZ);
+                       cc = MIN(uio_resid(uio), OBUFSIZ);
                        cp = obuf;
                        error = uiomove(cp, cc, uio);
                        if (error) {
@@ -2044,7 +2475,9 @@ loop:
                        i = b_to_q((u_char *)cp, ce, &tp->t_outq);
                        ce -= i;
                        tp->t_column += ce;
-                       cp += ce, cc -= ce, tk_nout += ce;
+                       cp += ce;
+                       cc -= ce;
+                       tk_nout += ce;
                        tp->t_outcc += ce;
                        if (i > 0) {
                                /* out of space */
@@ -2063,7 +2496,6 @@ out:
         * (the call will either return short or restart with a new uio).
         */
        uio_setresid(uio, (uio_resid(uio) + cc));
-       thread_funnel_set(kernel_flock, funnel_state);
        return (error);
 
 overfull:
@@ -2087,7 +2519,6 @@ ovhiwat:
        }
        if (flag & IO_NDELAY) {
                uio_setresid(uio, (uio_resid(uio) + cc));
-               thread_funnel_set(kernel_flock, funnel_state);
                return (uio_resid(uio) == count ? EWOULDBLOCK : 0);
        }
        SET(tp->t_state, TS_SO_OLOWAT);
@@ -2100,9 +2531,12 @@ ovhiwat:
        goto loop;
 }
 
+
 /*
  * Rubout one character from the rawq of tp
  * as cleanly as possible.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 static void
 ttyrub(int c, struct tty *tp)
@@ -2111,6 +2545,8 @@ ttyrub(int c, struct tty *tp)
        int savecol;
        int tabc;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        if (!ISSET(tp->t_lflag, ECHO) || ISSET(tp->t_lflag, EXTPROC))
                return;
        CLR(tp->t_lflag, FLUSHO);
@@ -2182,12 +2618,17 @@ ttyrub(int c, struct tty *tp)
        --tp->t_rocount;
 }
 
+
 /*
  * Back over count characters, erasing them.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 static void
 ttyrubo(struct tty *tp, int count)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        while (count-- > 0) {
                (void)ttyoutput('\b', tp);
                (void)ttyoutput(' ', tp);
@@ -2195,10 +2636,13 @@ ttyrubo(struct tty *tp, int count)
        }
 }
 
+
 /*
  * ttyretype --
  *     Reprint the rawq line.  Note, it is assumed that c_cc has already
  *     been checked.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 static void
 ttyretype(struct tty *tp)
@@ -2206,6 +2650,8 @@ ttyretype(struct tty *tp)
        u_char *cp;
        int c;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        /* Echo the reprint character. */
        if (tp->t_cc[VREPRINT] != _POSIX_VDISABLE)
                ttyecho(tp->t_cc[VREPRINT], tp);
@@ -2227,12 +2673,17 @@ ttyretype(struct tty *tp)
        tp->t_rocol = 0;
 }
 
+
 /*
  * Echo a typed character to the terminal.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 static void
 ttyecho(int c, struct tty *tp)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        if (!ISSET(tp->t_state, TS_CNTTB))
                CLR(tp->t_lflag, FLUSHO);
        if ((!ISSET(tp->t_lflag, ECHO) &&
@@ -2252,26 +2703,52 @@ ttyecho(int c, struct tty *tp)
        (void)ttyoutput(c, tp);
 }
 
+
 /*
  * Wake up any readers on a tty.
+ *
+ * Locks:      Assumes tty_lock() is held for the call.
  */
 void
 ttwakeup(struct tty *tp)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        selwakeup(&tp->t_rsel);
-       if (ISSET(tp->t_state, TS_ASYNC))
+       KNOTE(&tp->t_rsel.si_note, 1);
+       if (ISSET(tp->t_state, TS_ASYNC)) {
+               /*
+                * XXX: Callers may not revalidate it the tty is closed
+                * XXX: out from under them by another thread, but we do
+                * XXX: not support queued signals.  This should be safe,
+                * XXX: since the process we intend to wakeup is in the
+                * XXX: process group, and will wake up because of the
+                * XXX: signal anyway.
+                */
+               tty_unlock(tp);
                tty_pgsignal(tp, SIGIO, 1);
+               tty_lock(tp);
+       }
        wakeup(TSA_HUP_OR_INPUT(tp));
 }
 
+
 /*
+ * ttwwakeup (LDISC)
+ *
  * Wake up any writers on a tty.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 void
 ttwwakeup(struct tty *tp)
 {
-       if (tp->t_outq.c_cc <= tp->t_lowat)
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
+       if (tp->t_outq.c_cc <= tp->t_lowat) {
                selwakeup(&tp->t_wsel);
+               KNOTE(&tp->t_wsel.si_note, 1);
+       }
        if (ISSET(tp->t_state, TS_BUSY | TS_SO_OCOMPLETE) ==
            TS_SO_OCOMPLETE && tp->t_outq.c_cc == 0) {
                CLR(tp->t_state, TS_SO_OCOMPLETE);
@@ -2284,9 +2761,13 @@ ttwwakeup(struct tty *tp)
        }
 }
 
+
 /*
  * Look up a code for a specified speed in a conversion table;
  * used by drivers to map software speed values to hardware parameters.
+ *
+ * Notes:      No locks are assumed for this function; it does not
+ *             directly access struct tty.
  */
 int
 ttspeedtab(int speed, struct speedtab *table)
@@ -2297,12 +2778,14 @@ ttspeedtab(int speed, struct speedtab *table)
        return (-1);
 }
 
+
 /*
  * Set tty hi and low water marks.
  *
  * Try to arrange the dynamics so there's about one second
  * from hi to low water.
  *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 void
 ttsetwater(struct tty *tp)
@@ -2310,6 +2793,8 @@ ttsetwater(struct tty *tp)
        int cps;
        unsigned int x;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
 #define CLAMP(x, h, l) ((x) > h ? h : ((x) < l) ? l : (x))
 
        cps = tp->t_ospeed / 10;
@@ -2329,11 +2814,14 @@ extern kern_return_t    thread_info_internal(thread_t thread,
                                thread_info_t thread_info_out,
                                mach_msg_type_number_t *thread_info_count);
 
+
 /*
  * Report on state of foreground process group.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
  */
 void
-ttyinfo(struct tty *tp)
+ttyinfo_locked(struct tty *tp)
 {
        int             load;
        thread_t        thread;
@@ -2348,6 +2836,7 @@ ttyinfo(struct tty *tp)
        mach_msg_type_number_t          mmtn = THREAD_BASIC_INFO_COUNT;
        struct pgrp * pg;
 
+       TTY_LOCK_OWNED(tp);     /* debug assert */
 
        if (ttycheckoutq(tp,0) == 0)
                return;
@@ -2394,7 +2883,10 @@ ttyinfo(struct tty *tp)
                }
        }
        pgrp_unlock(pg);
+       /* SAFE: All callers drop the lock on return */
+       tty_unlock(tp);
        pg_rele(pg);
+       tty_lock(tp);
 
        pick = proc_find(pickpid);
        if (pick == PROC_NULL)
@@ -2434,7 +2926,7 @@ ttyinfo(struct tty *tp)
        proc_rele(pick);
 
        /* Print command, pid, state, utime, and stime */
-       ttyprintf(tp, " cmd: %s %d %s %ld.%02ldu %ld.%02lds\n",
+       ttyprintf(tp, " cmd: %s %d %s %ld.%02du %ld.%02ds\n",
                pick->p_comm,
                pick->p_pid,
                state,
@@ -2443,6 +2935,7 @@ ttyinfo(struct tty *tp)
        tp->t_rocount = 0;
 }
 
+
 /*
  * Returns 1 if p2 is "better" than p1
  *
@@ -2461,6 +2954,11 @@ ttyinfo(struct tty *tp)
 #define ONLYB   1
 #define BOTH    3
 
+/*
+ * Locks:      pgrp_lock(p2) held on call to this function
+ *             tty_lock(tp) for p2's tty, for which p2 is the foreground
+ *                     process, held on call to this function
+ */
 static int
 proc_compare(proc_t p1, proc_t p2)
 {
@@ -2513,12 +3011,19 @@ proc_compare(proc_t p1, proc_t p2)
        return (p2->p_pid > p1->p_pid);         /* tie - return highest pid */
 }
 
+
 /*
  * Output char to tty; console putchar style.
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
+ *
+ * Notes:      Only ever called from putchar() in subr_prf.c
  */
 int
 tputchar(int c, struct tty *tp)
 {
+       TTY_LOCK_OWNED(tp);     /* debug assert */
+
        if (!ISSET(tp->t_state, TS_CONNECTED)) {
                return (-1);
        }
@@ -2529,9 +3034,33 @@ tputchar(int c, struct tty *tp)
        return (0);
 }
 
+
 /*
+ * ttysleep
+ *
+ * Sleep on a wait channel waiting for an interrupt or a condition to come
+ * true so that we are woken up.
+ *
+ * Parameters: tp                      Tty going to sleep
+ *             chan                    The sleep channel (usually an address
+ *                                     of a structure member)
+ *             pri                     priority and flags
+ *             wmesg                   Wait message; shows up in debugger,
+ *                                     should show up in "ps", but doesn't
+ *             timo                    Timeout for the sleep
+ *
+ * Returns:    0                       Condition came true
+ *             ERESTART                Upper layer must redrive the call;
+ *                                     this is usually done by the Libc
+ *                                     stub in user space
+ *     msleep0:EINTR                   Interrupted (usually a signal)
+ *     msleep0:ERESTART                Interrupted (usually a masked signal)
+ *     msleep0:EWOULDBLOCK             Timeout (timo) already expired
+ *
+ * Locks:      Assumes tty_lock() is held prior to calling.
+ *
  * Sleep on chan, returning ERESTART if tty changed while we napped and
- * returning any errors (e.g. EINTR/EWOULDBLOCK) reported by tsleep.  If
+ * returning any errors (e.g. EINTR/EWOULDBLOCK) reported by msleep0.  If
  * the tty is revoked, restarting a pending call will redo validation done
  * at the start of the call.
  */
@@ -2541,13 +3070,17 @@ ttysleep(struct tty *tp, void *chan, int pri, const char *wmesg, int timo)
        int error;
        int gen;
 
+       TTY_LOCK_OWNED(tp);
+
        gen = tp->t_gen;
-       error = tsleep(chan, pri, wmesg, timo);
+       /* Use of msleep0() avoids conversion timo/timespec/timo */
+       error = msleep0(chan, &tp->t_lock, pri, wmesg, timo, (int (*)(int))0);
        if (error)
                return (error);
        return (tp->t_gen == gen ? 0 : ERESTART);
 }
 
+
 /*
  * Allocate a tty structure and its associated buffers.
  *
@@ -2555,6 +3088,9 @@ ttysleep(struct tty *tp, void *chan, int pri, const char *wmesg, int timo)
  *
  * Returns:    !NULL                           Address of new struct tty
  *             NULL                            Error ("ENOMEM")
+ *
+ * Locks:      The tty_lock() of the returned tty is not held when it
+ *             is returned.
  */
 struct tty *
 ttymalloc(void)
@@ -2568,26 +3104,77 @@ ttymalloc(void)
                clalloc(&tp->t_canq, TTYCLSIZE, 1);
                /* output queue doesn't need quoting */
                clalloc(&tp->t_outq, TTYCLSIZE, 0);
+               lck_mtx_init(&tp->t_lock, tty_lck_grp, tty_lck_attr);
+               klist_init(&tp->t_rsel.si_note);
+               klist_init(&tp->t_wsel.si_note);
+               tp->t_refcnt = 1;
        }
-       return(tp);
+       return (tp);
+}
+
+/*
+ * Increment the reference count on a tty.
+ */
+void
+ttyhold(struct tty *tp)
+{
+       TTY_LOCK_OWNED(tp);
+       tp->t_refcnt++;
 }
 
 /*
- * Free a tty structure and its buffers.
+ * Drops a reference count on a tty structure; if the reference count reaches
+ * zero, then also frees the structure and associated buffers.
  */
 void
 ttyfree(struct tty *tp)
 {
+       TTY_LOCK_NOTOWNED(tp);
+
+       tty_lock(tp);
+       if (--tp->t_refcnt == 0) {
+               tty_unlock(tp);
+               ttydeallocate(tp);
+       } else if (tp->t_refcnt < 0) {
+               panic("%s: freeing free tty %p", __func__, tp);
+       } else
+               tty_unlock(tp);
+}
+
+/*
+ * Deallocate a tty structure and its buffers.
+ *
+ * Locks:      The tty_lock() is assumed to not be held at the time of
+ *             the free; this function destroys the mutex.
+ */
+static void
+ttydeallocate(struct tty *tp)
+{
+       TTY_LOCK_NOTOWNED(tp);  /* debug assert */
+
+#if DEBUG
+       if (!(SLIST_EMPTY(&tp->t_rsel.si_note) && SLIST_EMPTY(&tp->t_wsel.si_note))) {
+               panic("knotes hooked into a tty when the tty is freed.\n");
+       }
+#endif /* DEBUG */
+
        clfree(&tp->t_rawq);
        clfree(&tp->t_canq);
        clfree(&tp->t_outq);
+       lck_mtx_destroy(&tp->t_lock, tty_lck_grp);
        FREE(tp, M_TTYS);
 }
 
+
+/*
+ * Locks:      Assumes tty_lock() is held prior to calling.
+ */
 int
 isbackground(proc_t p, struct tty  *tp)
 {
-       return (isctty(p, tp) && (p->p_pgrp != tp->t_pgrp));
+       TTY_LOCK_OWNED(tp);
+
+       return (tp->t_session != NULL && p->p_pgrp != NULL && (p->p_pgrp != tp->t_pgrp) && isctty_sp(p, tp, p->p_pgrp->pg_session));
 }
 
 static int 
@@ -2609,3 +3196,390 @@ isctty_sp(proc_t p, struct tty  *tp, struct session *sessp)
 
 }
 
+
+static int  filt_ttyattach(struct knote *kn, struct kevent_internal_s *kev);
+static void filt_ttydetach(struct knote *kn);
+static int  filt_ttyevent(struct knote *kn, long hint);
+static int  filt_ttytouch(struct knote *kn, struct kevent_internal_s *kev);
+static int  filt_ttyprocess(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev);
+
+SECURITY_READ_ONLY_EARLY(struct filterops) tty_filtops = {
+       .f_isfd    = 1,
+       .f_attach  = filt_ttyattach,
+       .f_detach  = filt_ttydetach,
+       .f_event   = filt_ttyevent,
+       .f_touch   = filt_ttytouch,
+       .f_process = filt_ttyprocess
+};
+
+/*
+ * Called with struct tty locked. Returns non-zero if there is data to be read
+ * or written.
+ */
+static int
+filt_tty_common(struct knote *kn, struct tty *tp)
+{
+       int retval = 0;
+
+       TTY_LOCK_OWNED(tp); /* debug assert */
+
+       if (tp->t_state & TS_ZOMBIE) {
+               kn->kn_flags |= EV_EOF;
+               return 1;
+       }
+
+       switch (knote_get_seltype(kn)) {
+       case FREAD:
+               retval = ttnread(tp);
+               break;
+       case FWRITE:
+               if ((tp->t_outq.c_cc <= tp->t_lowat) &&
+                   (tp->t_state & TS_CONNECTED)) {
+                       retval = tp->t_hiwat - tp->t_outq.c_cc;
+               }
+               break;
+       }
+
+       kn->kn_data = retval;
+
+       /*
+        * TODO(mwidmann, jandrus): For native knote low watermark support,
+        * check the kn_sfflags for NOTE_LOWAT and check against kn_sdata.
+        *
+        * res = ((kn->kn_sfflags & NOTE_LOWAT) != 0) ?
+        *        (kn->kn_data >= kn->kn_sdata) : kn->kn_data;
+        */
+
+       return retval;
+}
+
+/*
+ * Find the struct tty from a waitq, which is a member of one of the two struct
+ * selinfos inside the struct tty.  Use the seltype to determine which selinfo.
+ */
+static struct tty *
+tty_from_waitq(struct waitq *wq, int seltype)
+{
+       struct selinfo *si;
+       struct tty *tp = NULL;
+
+       /*
+        * The waitq is part of the selinfo structure managed by the driver. For
+        * certain drivers, we want to hook the knote into the selinfo
+        * structure's si_note field so selwakeup can call KNOTE.
+        *
+        * While 'wq' is not really a queue element, this macro only uses the
+        * pointer to calculate the offset into a structure given an element
+        * name.
+        */
+       si = qe_element(wq, struct selinfo, si_waitq);
+
+       /*
+        * For TTY drivers, the selinfo structure is somewhere in the struct
+        * tty. There are two different selinfo structures, and the one used
+        * corresponds to the type of filter requested.
+        *
+        * While 'si' is not really a queue element, this macro only uses the
+        * pointer to calculate the offset into a structure given an element
+        * name.
+        */
+       switch (seltype) {
+       case FREAD:
+               tp = qe_element(si, struct tty, t_rsel);
+               break;
+       case FWRITE:
+               tp = qe_element(si, struct tty, t_wsel);
+               break;
+       }
+
+       return tp;
+}
+
+static struct tty *
+tty_from_knote(struct knote *kn)
+{
+       return (struct tty *)kn->kn_hook;
+}
+
+/*
+ * Try to lock the TTY structure associated with a knote.
+ *
+ * On success, this function returns a locked TTY structure.  Otherwise, NULL is
+ * returned.
+ */
+__attribute__((warn_unused_result))
+static struct tty *
+tty_lock_from_knote(struct knote *kn)
+{
+       struct tty *tp = tty_from_knote(kn);
+       if (tp) {
+               tty_lock(tp);
+       }
+
+       return tp;
+}
+
+/*
+ * Set the knote's struct tty to the kn_hook field.
+ *
+ * The idea is to fake a call to select with our own waitq set.  If the driver
+ * calls selrecord, we'll get a link to their waitq and access to the tty
+ * structure.
+ *
+ * Returns -1 on failure, with the error set in the knote, or selres on success.
+ */
+static int
+tty_set_knote_hook(struct knote *kn)
+{
+       uthread_t uth;
+       vfs_context_t ctx;
+       vnode_t vp;
+       kern_return_t kr;
+       struct waitq *wq = NULL;
+       struct waitq_set *old_wqs;
+       struct waitq_set tmp_wqs;
+       uint64_t rsvd, rsvd_arg;
+       uint64_t *rlptr = NULL;
+       int selres = -1;
+       struct tty *tp;
+
+       uth = get_bsdthread_info(current_thread());
+
+       ctx = vfs_context_current();
+       vp = (vnode_t)kn->kn_fp->f_fglob->fg_data;
+
+       /*
+        * Reserve a link element to avoid potential allocation under
+        * a spinlock.
+        */
+       rsvd = rsvd_arg = waitq_link_reserve(NULL);
+       rlptr = (void *)&rsvd_arg;
+
+       /*
+        * Trick selrecord into hooking a known waitq set into the device's selinfo
+        * waitq.  Once the link is in place, we can get back into the selinfo from
+        * the waitq and subsequently the tty (see tty_from_waitq).
+        *
+        * We can't use a real waitq set (such as the kqueue's) because wakeups
+        * might happen before we can unlink it.
+        */
+       kr = waitq_set_init(&tmp_wqs, SYNC_POLICY_FIFO | SYNC_POLICY_PREPOST, NULL,
+                       NULL);
+       assert(kr == KERN_SUCCESS);
+
+       /*
+        * Lazy allocate the waitqset to avoid potential allocation under
+        * a spinlock;
+        */
+       waitq_set_lazy_init_link(&tmp_wqs);
+
+       old_wqs = uth->uu_wqset;
+       uth->uu_wqset = &tmp_wqs;
+       /*
+        * FMARK forces selects to always call selrecord, even if data is
+        * available.  See ttselect, ptsselect, ptcselect.
+        *
+        * selres also contains the data currently available in the tty.
+        */
+       selres = VNOP_SELECT(vp, knote_get_seltype(kn) | FMARK, 0, rlptr, ctx);
+       uth->uu_wqset = old_wqs;
+
+       /*
+        * Make sure to cleanup the reserved link - this guards against
+        * drivers that may not actually call selrecord().
+        */
+       waitq_link_release(rsvd);
+       if (rsvd == rsvd_arg) {
+               /*
+                * The driver didn't call selrecord -- there's no tty hooked up so we
+                * can't attach.
+                */
+               knote_set_error(kn, ENOTTY);
+               selres = -1;
+               goto out;
+       }
+
+       /* rlptr may not point to a properly aligned pointer */
+       memcpy(&wq, rlptr, sizeof(void *));
+
+       tp = tty_from_waitq(wq, knote_get_seltype(kn));
+       assert(tp != NULL);
+
+       /*
+        * Take a reference and stash the tty in the knote.
+        */
+       tty_lock(tp);
+       ttyhold(tp);
+       kn->kn_hook = tp;
+       tty_unlock(tp);
+
+out:
+       /*
+        * Cleaning up the wqset will unlink its waitq and clean up any preposts
+        * that occurred as a result of data coming in while the tty was attached.
+        */
+       waitq_set_deinit(&tmp_wqs);
+
+       return selres;
+}
+
+static int
+filt_ttyattach(struct knote *kn, __unused struct kevent_internal_s *kev)
+{
+       int selres = 0;
+       struct tty *tp;
+
+       /*
+        * This function should be called from filt_specattach (spec_vnops.c),
+        * so most of the knote data structure should already be initialized.
+        */
+
+       /* don't support offsets in ttys or drivers that don't use struct tty */
+       if (kn->kn_vnode_use_ofst || !kn->kn_vnode_kqok) {
+               knote_set_error(kn, ENOTSUP);
+               return 0;
+       }
+
+       /*
+        * Connect the struct tty to the knote through the selinfo structure
+        * referenced by the waitq within the selinfo.
+        */
+       selres = tty_set_knote_hook(kn);
+       if (selres < 0) {
+               return 0;
+       }
+
+       /*
+        * Attach the knote to selinfo's klist.
+        */
+       tp = tty_lock_from_knote(kn);
+       if (!tp) {
+               knote_set_error(kn, ENOENT);
+               return 0;
+       }
+
+       switch (knote_get_seltype(kn)) {
+       case FREAD:
+               KNOTE_ATTACH(&tp->t_rsel.si_note, kn);
+               break;
+       case FWRITE:
+               KNOTE_ATTACH(&tp->t_wsel.si_note, kn);
+               break;
+       }
+
+       tty_unlock(tp);
+
+       return selres;
+}
+
+static void
+filt_ttydetach(struct knote *kn)
+{
+       struct tty *tp;
+
+       tp = tty_lock_from_knote(kn);
+       if (!tp) {
+               knote_set_error(kn, ENOENT);
+               return;
+       }
+
+       struct selinfo *si = NULL;
+       switch (knote_get_seltype(kn)) {
+       case FREAD:
+               si = &tp->t_rsel;
+               break;
+       case FWRITE:
+               si = &tp->t_wsel;
+               break;
+       /* knote_get_seltype will panic on default */
+       }
+
+       KNOTE_DETACH(&si->si_note, kn);
+       kn->kn_hook = NULL;
+
+       tty_unlock(tp);
+       ttyfree(tp);
+}
+
+static int
+filt_ttyevent(struct knote *kn, long hint)
+{
+       int ret;
+       struct tty *tp;
+       bool revoked = hint & NOTE_REVOKE;
+       hint &= ~NOTE_REVOKE;
+
+       tp = tty_from_knote(kn);
+       if (!tp) {
+               knote_set_error(kn, ENOENT);
+               return 0;
+       }
+
+       if (!hint) {
+               tty_lock(tp);
+       }
+
+       if (revoked) {
+               kn->kn_flags |= EV_EOF | EV_ONESHOT;
+               ret = 1;
+       } else {
+               ret = filt_tty_common(kn, tp);
+       }
+
+       if (!hint) {
+               tty_unlock(tp);
+       }
+
+       return ret;
+}
+
+static int
+filt_ttytouch(struct knote *kn, struct kevent_internal_s *kev)
+{
+       struct tty *tp;
+       int res = 0;
+
+       tp = tty_lock_from_knote(kn);
+       if (!tp) {
+               knote_set_error(kn, ENOENT);
+               return 0;
+       }
+
+       kn->kn_sdata = kev->data;
+       kn->kn_sfflags = kev->fflags;
+
+       if (kn->kn_vnode_kqok) {
+               res = filt_tty_common(kn, tp);
+       }
+
+       tty_unlock(tp);
+
+       return res;
+}
+
+static int
+filt_ttyprocess(struct knote *kn, __unused struct filt_process_s *data, struct kevent_internal_s *kev)
+{
+       struct tty *tp;
+       int res;
+
+       tp = tty_lock_from_knote(kn);
+       if (!tp) {
+               knote_set_error(kn, ENOENT);
+               return 0;
+       }
+
+       res = filt_tty_common(kn, tp);
+
+       if (res) {
+               *kev = kn->kn_kevent;
+               if (kn->kn_flags & EV_CLEAR) {
+                       kn->kn_fflags = 0;
+                       kn->kn_data = 0;
+               }
+       }
+
+       tty_unlock(tp);
+
+       return res;
+}