qio: Track the amount of bytes read
[akaros.git] / kern / src / net / devip.c
index 1a2f7d3..410f94d 100644 (file)
@@ -1,4 +1,31 @@
-// INFERNO
+/* Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
+ * Portions Copyright © 1997-1999 Vita Nuova Limited
+ * Portions Copyright © 2000-2007 Vita Nuova Holdings Limited
+ *                                (www.vitanuova.com)
+ * Revisions Copyright © 2000-2007 Lucent Technologies Inc. and others
+ *
+ * Modified for the Akaros operating system:
+ * Copyright (c) 2013-2014 The Regents of the University of California
+ * Copyright (c) 2013-2015 Google Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE. */
+
 #include <vfs.h>
 #include <kfs.h>
 #include <slab.h>
 #include <smp.h>
 #include <ip.h>
 
+struct dev ipdevtab;
+
+static char *devname(void)
+{
+       return ipdevtab.name;
+}
+
 enum {
        Qtopdir = 1,                            /* top level directory */
        Qtopbase,
        Qarp = Qtopbase,
-       Qbootp,
        Qndb,
        Qiproute,
        Qiprouter,
@@ -50,6 +83,8 @@ enum {
        Shiftproto = Logtype + Logconv,
 
        Nfs = 32,
+       BYPASS_QMAX = 64 * MiB,
+       IPROUTE_LEN = 2 * PGSIZE,
 };
 #define TYPE(x)        ( ((uint32_t)(x).path) & Masktype )
 #define CONV(x)        ( (((uint32_t)(x).path) >> Shiftconv) & Maskconv )
@@ -63,9 +98,17 @@ struct queue *qlog;
 
 extern void nullmediumlink(void);
 extern void pktmediumlink(void);
-extern char *eve;
+extern struct username eve;
 static long ndbwrite(struct Fs *, char *unused_char_p_t, uint32_t, int);
 static void closeconv(struct conv *);
+static void setup_proto_qio_bypass(struct conv *cv);
+static void undo_proto_qio_bypass(struct conv *cv);
+
+static struct conv *chan2conv(struct chan *chan)
+{
+       /* That's a lot of pointers to get to the conv! */
+       return ipfs[chan->dev]->p[PROTO(chan->qid)]->conv[CONV(chan->qid)];
+}
 
 static inline int founddevdir(struct chan *c, struct qid q, char *n,
                                                          int64_t length, char *user, long perm,
@@ -79,7 +122,7 @@ static int topdirgen(struct chan *c, struct dir *dp)
 {
        struct qid q;
        mkqid(&q, QID(0, 0, Qtopdir), 0, QTDIR);
-       snprintf(get_cur_genbuf(), GENBUF_SZ, "#I%lu", c->dev);
+       snprintf(get_cur_genbuf(), GENBUF_SZ, "#%s%lu", devname(), c->dev);
        return founddevdir(c, q, get_cur_genbuf(), 0, network, 0555, dp);
 }
 
@@ -89,10 +132,11 @@ static int ip3gen(struct chan *c, int i, struct dir *dp)
        struct qid q;
        struct conv *cv;
        char *p;
+       int perm;
 
-       cv = ipfs[c->dev]->p[PROTO(c->qid)]->conv[CONV(c->qid)];
+       cv = chan2conv(c);
        if (cv->owner == NULL)
-               kstrdup(&cv->owner, eve);
+               kstrdup(&cv->owner, eve.name);
        mkqid(&q, QID(PROTO(c->qid), CONV(c->qid), i), 0, QTFILE);
 
        switch (i) {
@@ -102,13 +146,20 @@ static int ip3gen(struct chan *c, int i, struct dir *dp)
                        return founddevdir(c, q, "ctl", 0,
                                                   cv->owner, cv->perm, dp);
                case Qdata:
+                       perm = cv->perm;
+                       perm |= qreadable(cv->rq) ? DMREADABLE : 0;
+                       perm |= qwritable(cv->wq) ? DMWRITABLE : 0;
                        return founddevdir(c, q, "data", qlen(cv->rq),
-                                                          cv->owner, cv->perm, dp);
+                                                          cv->owner, perm, dp);
                case Qerr:
+                       perm = cv->perm;
+                       perm |= qreadable(cv->eq) ? DMREADABLE : 0;
                        return founddevdir(c, q, "err", qlen(cv->eq),
-                                                          cv->owner, cv->perm, dp);
+                                                          cv->owner, perm, dp);
                case Qlisten:
-                       return founddevdir(c, q, "listen", 0, cv->owner, cv->perm, dp);
+                       perm = cv->perm;
+                       perm |= cv->incall ? DMREADABLE : 0;
+                       return founddevdir(c, q, "listen", 0, cv->owner, perm, dp);
                case Qlocal:
                        p = "local";
                        break;
@@ -118,8 +169,10 @@ static int ip3gen(struct chan *c, int i, struct dir *dp)
                case Qsnoop:
                        if (strcmp(cv->p->name, "ipifc") != 0)
                                return -1;
+                       perm = 0400;
+                       perm |= qreadable(cv->sq) ? DMREADABLE : 0;
                        return founddevdir(c, q, "snoop", qlen(cv->sq),
-                                                          cv->owner, 0400, dp);
+                                                          cv->owner, perm, dp);
                case Qstatus:
                        p = "status";
                        break;
@@ -159,11 +212,6 @@ static int ip1gen(struct chan *c, int i, struct dir *dp)
                case Qarp:
                        p = "arp";
                        break;
-               case Qbootp:
-                       if (bootp == NULL)
-                               return 0;
-                       p = "bootp";
-                       break;
                case Qndb:
                        p = "ndb";
                        len = strlen(f->ndb);
@@ -212,7 +260,6 @@ ipgen(struct chan *c, char *unused_char_p_t, struct dirtab *d, int unused_int,
                        s -= f->np;
                        return ip1gen(c, s + Qtopbase, dp);
                case Qarp:
-               case Qbootp:
                case Qndb:
                case Qlog:
                case Qiproute:
@@ -284,7 +331,7 @@ static struct Fs *ipgetfs(int dev)
 
        qlock(&fslock);
        if (ipfs[dev] == NULL) {
-               f = kzmalloc(sizeof(struct Fs), KMALLOC_WAIT);
+               f = kzmalloc(sizeof(struct Fs), MEM_WAIT);
                rwinit(&f->rwlock);
                qlock_init(&f->iprouter.qlock);
                ip_init(f);
@@ -324,10 +371,10 @@ static struct chan *ipattach(char *spec)
 
        dev = atoi(spec);
        if (dev >= Nfs)
-               error("bad specification");
+               error(EFAIL, "bad specification");
 
        ipgetfs(dev);
-       c = devattach('I', spec);
+       c = devattach(devname(), spec);
        mkqid(&c->qid, QID(0, 0, Qtopdir), 0, QTDIR);
        c->dev = dev;
 
@@ -362,12 +409,6 @@ static int should_wake(void *arg)
        return cv->incall != NULL;
 }
 
-static int m2p[] = {
-       [OREAD] 4,
-       [OWRITE] 2,
-       [ORDWR] 6
-};
-
 static struct chan *ipopen(struct chan *c, int omode)
 {
        ERRSTACK(2);
@@ -376,7 +417,8 @@ static struct chan *ipopen(struct chan *c, int omode)
        int perm;
        struct Fs *f;
 
-       perm = m2p[omode & 3];
+       /* perm is a lone rwx, not the rwx------ from the conversion */
+       perm = omode_to_rwx(omode) >> 6;
 
        f = ipfs[c->dev];
 
@@ -384,9 +426,9 @@ static struct chan *ipopen(struct chan *c, int omode)
                default:
                        break;
                case Qndb:
-                       if (omode & (OWRITE | OTRUNC) && !iseve())
-                               error(Eperm);
-                       if ((omode & (OWRITE | OTRUNC)) == (OWRITE | OTRUNC))
+                       if (omode & (O_WRITE | O_TRUNC) && !iseve())
+                               error(EPERM, ERROR_FIXME);
+                       if ((omode & (O_WRITE | O_TRUNC)) == (O_WRITE | O_TRUNC))
                                f->ndb[0] = 0;
                        break;
                case Qlog:
@@ -396,6 +438,8 @@ static struct chan *ipopen(struct chan *c, int omode)
                        iprouteropen(f);
                        break;
                case Qiproute:
+                       c->synth_buf = kpages_zalloc(IPROUTE_LEN, MEM_WAIT);
+                       routeread(f, c->synth_buf, 0, IPROUTE_LEN);
                        break;
                case Qtopdir:
                case Qprotodir:
@@ -404,18 +448,18 @@ static struct chan *ipopen(struct chan *c, int omode)
                case Qremote:
                case Qlocal:
                case Qstats:
-               case Qbootp:
                case Qipselftab:
-                       if (!IS_RDONLY(omode))
-                               error(Eperm);
+                       if (omode & O_WRITE)
+                               error(EPERM, ERROR_FIXME);
                        break;
                case Qsnoop:
-                       if (!IS_RDONLY(omode))
-                               error(Eperm);
+                       if (omode & O_WRITE)
+                               error(EPERM, ERROR_FIXME);
+                       /* might be racy.  note the lack of a proto lock, unlike Qdata */
                        p = f->p[PROTO(c->qid)];
                        cv = p->conv[CONV(c->qid)];
                        if (strcmp(ATTACHER(c), cv->owner) != 0 && !iseve())
-                               error(Eperm);
+                               error(EPERM, ERROR_FIXME);
                        atomic_inc(&cv->snoopers);
                        break;
                case Qclone:
@@ -429,7 +473,7 @@ static struct chan *ipopen(struct chan *c, int omode)
                        qunlock(&p->qlock);
                        poperror();
                        if (cv == NULL) {
-                               error(Enodev);
+                               error(ENODEV, "Null conversation from Fsprotoclone");
                                break;
                        }
                        mkqid(&c->qid, QID(p->x, cv->x, Qctl), 0, QTFILE);
@@ -448,9 +492,9 @@ static struct chan *ipopen(struct chan *c, int omode)
                        }
                        if ((perm & (cv->perm >> 6)) != perm) {
                                if (strcmp(ATTACHER(c), cv->owner) != 0)
-                                       error(Eperm);
+                                       error(EPERM, ERROR_FIXME);
                                if ((perm & cv->perm) != perm)
-                                       error(Eperm);
+                                       error(EPERM, ERROR_FIXME);
 
                        }
                        cv->inuse++;
@@ -464,16 +508,32 @@ static struct chan *ipopen(struct chan *c, int omode)
                        break;
                case Qlisten:
                        cv = f->p[PROTO(c->qid)]->conv[CONV(c->qid)];
+                       /* No permissions or Announce checks required.  We'll see if that's
+                        * a good idea or not. (the perm check would do nothing, as is,
+                        * since an O_PATH perm is 0).
+                        *
+                        * But we probably want to incref to keep the conversation around
+                        * until this FD/chan is closed.  #ip is a little weird in that
+                        * objects never really go away (high water mark for convs, you can
+                        * always find them in the ns).  I think it is possible to
+                        * namec/ipgen a chan, then have that conv close, then have that
+                        * chan be opened.  You can probably do this with a data file. */
+                       if (omode & O_PATH) {
+                               qlock(&cv->qlock);
+                               cv->inuse++;
+                               qunlock(&cv->qlock);
+                               break;
+                       }
                        if ((perm & (cv->perm >> 6)) != perm) {
                                if (strcmp(ATTACHER(c), cv->owner) != 0)
-                                       error(Eperm);
+                                       error(EPERM, ERROR_FIXME);
                                if ((perm & cv->perm) != perm)
-                                       error(Eperm);
+                                       error(EPERM, ERROR_FIXME);
 
                        }
 
                        if (cv->state != Announced)
-                               error("not announced");
+                               error(EFAIL, "not announced");
 
                        if (waserror()) {
                                closeconv(cv);
@@ -487,14 +547,18 @@ static struct chan *ipopen(struct chan *c, int omode)
                        while (nc == NULL) {
                                /* give up if we got a hangup */
                                if (qisclosed(cv->rq))
-                                       error("listen hungup");
+                                       error(EFAIL, "listen hungup");
 
                                qlock(&cv->listenq);
                                if (waserror()) {
                                        qunlock(&cv->listenq);
                                        nexterror();
                                }
-
+                               /* we can peek at incall without grabbing the cv qlock.  if
+                                * anything is there, it'll remain there until we dequeue it.
+                                * no one else can, since we hold the listenq lock */
+                               if ((c->flag & O_NONBLOCK) && !cv->incall)
+                                       error(EAGAIN, "listen queue empty");
                                /* wait for a connect */
                                rendez_sleep(&cv->listenr, should_wake, cv);
 
@@ -533,7 +597,7 @@ static int ipwstat(struct chan *c, uint8_t * dp, int n)
        f = ipfs[c->dev];
        switch (TYPE(c->qid)) {
                default:
-                       error(Eperm);
+                       error(EPERM, ERROR_FIXME);
                        break;
                case Qctl:
                case Qdata:
@@ -547,14 +611,14 @@ static int ipwstat(struct chan *c, uint8_t * dp, int n)
        }
        n = convM2D(dp, n, d, (char *)&d[1]);
        if (n == 0)
-               error(Eshortstat);
+               error(ENODATA, ERROR_FIXME);
        p = f->p[PROTO(c->qid)];
        cv = p->conv[CONV(c->qid)];
        if (!iseve() && strcmp(ATTACHER(c), cv->owner) != 0)
-               error(Eperm);
+               error(EPERM, ERROR_FIXME);
        if (!emptystr(d->uid))
                kstrdup(&cv->owner, d->uid);
-       if (d->mode != ~0UL)
+       if (d->mode != -1)
                cv->perm = d->mode & 0777;
        poperror();
        kfree(d);
@@ -578,8 +642,11 @@ static char *ipchaninfo(struct chan *ch, char *ret, size_t ret_l)
                case Qdata:
                        proto = f->p[PROTO(ch->qid)];
                        conv = proto->conv[CONV(ch->qid)];
-                       snprintf(ret, ret_l, "Qdata, proto %s, conv idx %d", proto->name,
-                                        conv->x);
+                       snprintf(ret, ret_l,
+                                "Qdata, %s, proto %s, conv idx %d, rq len %d, wq len %d, total read %llu",
+                                SLIST_EMPTY(&conv->data_taps) ? "untapped" : "tapped",
+                                proto->name, conv->x, qlen(conv->rq), qlen(conv->wq),
+                                        q_bytes_read(conv->rq));
                        break;
                case Qarp:
                        ret = "Qarp";
@@ -587,6 +654,14 @@ static char *ipchaninfo(struct chan *ch, char *ret, size_t ret_l)
                case Qiproute:
                        ret = "Qiproute";
                        break;
+               case Qlisten:
+                       proto = f->p[PROTO(ch->qid)];
+                       conv = proto->conv[CONV(ch->qid)];
+                       snprintf(ret, ret_l,
+                                "Qlisten, %s proto %s, conv idx %d, has %sincalls",
+                                SLIST_EMPTY(&conv->listen_taps) ? "untapped" : "tapped",
+                                proto->name, conv->x, conv->incall ? "" : "no ");
+                       break;
                case Qlog:
                        ret = "Qlog";
                        break;
@@ -605,6 +680,7 @@ static char *ipchaninfo(struct chan *ch, char *ret, size_t ret_l)
 
 static void closeconv(struct conv *cv)
 {
+       ERRSTACK(1);
        struct conv *nc;
        struct Ipmulti *mp;
 
@@ -614,7 +690,10 @@ static void closeconv(struct conv *cv)
                qunlock(&cv->qlock);
                return;
        }
-
+       if (waserror()) {
+               qunlock(&cv->qlock);
+               nexterror();
+       }
        /* close all incoming calls since no listen will ever happen */
        for (nc = cv->incall; nc; nc = cv->incall) {
                cv->incall = nc->next;
@@ -630,9 +709,12 @@ static void closeconv(struct conv *cv)
 
        cv->r = NULL;
        cv->rgen = 0;
+       if (cv->state == Bypass)
+               undo_proto_qio_bypass(cv);
        cv->p->close(cv);
        cv->state = Idle;
        qunlock(&cv->qlock);
+       poperror();
 }
 
 static void ipclose(struct chan *c)
@@ -654,6 +736,7 @@ static void ipclose(struct chan *c)
                case Qdata:
                case Qctl:
                case Qerr:
+               case Qlisten:
                        if (c->flag & COPEN)
                                closeconv(f->p[PROTO(c->qid)]->conv[CONV(c->qid)]);
                        break;
@@ -661,6 +744,10 @@ static void ipclose(struct chan *c)
                        if (c->flag & COPEN)
                                atomic_dec(&f->p[PROTO(c->qid)]->conv[CONV(c->qid)]->snoopers);
                        break;
+               case Qiproute:
+                       if (c->flag & COPEN)
+                               kpages_free(c->synth_buf, IPROUTE_LEN);
+                       break;
        }
        kfree(((struct IPaux *)c->aux)->owner);
        kfree(c->aux);
@@ -678,26 +765,23 @@ static long ipread(struct chan *ch, void *a, long n, int64_t off)
        long rv;
        struct Fs *f;
        uint32_t offset = off;
-       size_t sofar;
 
        f = ipfs[ch->dev];
 
        p = a;
        switch (TYPE(ch->qid)) {
                default:
-                       error(Eperm);
+                       error(EPERM, ERROR_FIXME);
                case Qtopdir:
                case Qprotodir:
                case Qconvdir:
                        return devdirread(ch, a, n, 0, 0, ipgen);
                case Qarp:
                        return arpread(f->arp, a, offset, n);
-               case Qbootp:
-                       return bootpread(a, offset, n);
                case Qndb:
                        return readstr(offset, a, n, f->ndb);
                case Qiproute:
-                       return routeread(f, a, offset, n);
+                       return readmem(offset, a, n, ch->synth_buf, IPROUTE_LEN);
                case Qiprouter:
                        return iprouterread(f, a, n);
                case Qipselftab:
@@ -740,13 +824,19 @@ static long ipread(struct chan *ch, void *a, long n, int64_t off)
                        buf = kzmalloc(Statelen, 0);
                        x = f->p[PROTO(ch->qid)];
                        c = x->conv[CONV(ch->qid)];
-                       sofar = (*x->state) (c, buf, Statelen - 2);
+                       if (c->state == Bypass)
+                               snprintf(buf, Statelen, "Bypassed\n");
+                       else
+                               (*x->state)(c, buf, Statelen - 2);
                        rv = readstr(offset, p, n, buf);
                        kfree(buf);
                        return rv;
                case Qdata:
                        c = f->p[PROTO(ch->qid)]->conv[CONV(ch->qid)];
-                       return qread(c->rq, a, n);
+                       if (ch->flag & O_NONBLOCK)
+                               return qread_nonblock(c->rq, a, n);
+                       else
+                               return qread(c->rq, a, n);
                case Qerr:
                        c = f->p[PROTO(ch->qid)]->conv[CONV(ch->qid)];
                        return qread(c->eq, a, n);
@@ -756,7 +846,7 @@ static long ipread(struct chan *ch, void *a, long n, int64_t off)
                case Qstats:
                        x = f->p[PROTO(ch->qid)];
                        if (x->stats == NULL)
-                               error("stats not implemented");
+                               error(EFAIL, "stats not implemented");
                        buf = kzmalloc(Statelen, 0);
                        (*x->stats) (x, buf, Statelen);
                        rv = readstr(offset, p, n, buf);
@@ -768,15 +858,14 @@ static long ipread(struct chan *ch, void *a, long n, int64_t off)
 static struct block *ipbread(struct chan *ch, long n, uint32_t offset)
 {
        struct conv *c;
-       struct Proto *x;
-       struct Fs *f;
 
        switch (TYPE(ch->qid)) {
                case Qdata:
-                       f = ipfs[ch->dev];
-                       x = f->p[PROTO(ch->qid)];
-                       c = x->conv[CONV(ch->qid)];
-                       return qbread(c->rq, n);
+                       c = chan2conv(ch);
+                       if (ch->flag & O_NONBLOCK)
+                               return qbread_nonblock(c->rq, n);
+                       else
+                               return qbread(c->rq, n);
                default:
                        return devbread(ch, n, offset);
        }
@@ -793,7 +882,7 @@ static void setladdr(struct conv *c)
 /*
  *  set a local port making sure the quad of raddr,rport,laddr,lport is unique
  */
-static char *setluniqueport(struct conv *c, int lport)
+static void setluniqueport(struct conv *c, int lport)
 {
        struct Proto *p;
        struct conv *xp;
@@ -808,18 +897,18 @@ static char *setluniqueport(struct conv *c, int lport)
                        break;
                if (xp == c)
                        continue;
-               if ((xp->state == Connected || xp->state == Announced)
+               if ((xp->state == Connected || xp->state == Announced
+                                           || xp->state == Bypass)
                        && xp->lport == lport
                        && xp->rport == c->rport
                        && ipcmp(xp->raddr, c->raddr) == 0
                        && ipcmp(xp->laddr, c->laddr) == 0) {
                        qunlock(&p->qlock);
-                       return "address in use";
+                       error(EFAIL, "address in use");
                }
        }
        c->lport = lport;
        qunlock(&p->qlock);
-       return NULL;
 }
 
 /*
@@ -851,7 +940,7 @@ static void setlport(struct conv *c)
                                *pp = 600;
                } else
                        while (*pp < 5000)
-                               *pp = nrand(1 << 15);
+                               urandom_read(pp, sizeof(*pp));
 
                found = 0;
                for (x = 0; x < p->nc; x++) {
@@ -873,15 +962,12 @@ static void setlport(struct conv *c)
  *  set a local address and port from a string of the form
  *     [address!]port[!r]
  */
-static char *setladdrport(struct conv *c, char *str, int announcing)
+static void setladdrport(struct conv *c, char *str, int announcing)
 {
        char *p;
-       char *rv;
        uint16_t lport;
        uint8_t addr[IPaddrlen];
 
-       rv = NULL;
-
        /*
         *  ignore restricted part if it exists.  it's
         *  meaningless on local ports.
@@ -908,32 +994,31 @@ static char *setladdrport(struct conv *c, char *str, int announcing)
                        if (ipforme(c->p->f, addr))
                                ipmove(c->laddr, addr);
                        else
-                               return "not a local IP address";
+                               error(EFAIL, "not a local IP address");
                }
        }
 
        /* one process can get all connections */
        if (announcing && strcmp(p, "*") == 0) {
                if (!iseve())
-                       error(Eperm);
-               return setluniqueport(c, 0);
+                       error(EPERM, ERROR_FIXME);
+               setluniqueport(c, 0);
        }
 
        lport = atoi(p);
        if (lport <= 0)
                setlport(c);
        else
-               rv = setluniqueport(c, lport);
-       return rv;
+               setluniqueport(c, lport);
 }
 
-static char *setraddrport(struct conv *c, char *str)
+static void setraddrport(struct conv *c, char *str)
 {
        char *p;
 
        p = strchr(str, '!');
        if (p == NULL)
-               return "malformed address";
+               error(EFAIL, "malformed address");
        *p++ = 0;
        parseip(c->raddr, str);
        c->rport = atoi(p);
@@ -942,43 +1027,39 @@ static char *setraddrport(struct conv *c, char *str)
                if (strstr(p, "!r") != NULL)
                        c->restricted = 1;
        }
-       return NULL;
 }
 
 /*
  *  called by protocol connect routine to set addresses
  */
-char *Fsstdconnect(struct conv *c, char *argv[], int argc)
+void Fsstdconnect(struct conv *c, char *argv[], int argc)
 {
-       char *p;
-
        switch (argc) {
                default:
-                       return "bad args to connect";
+                       error(EINVAL, "bad args to %s", __func__);
                case 2:
-                       p = setraddrport(c, argv[1]);
-                       if (p != NULL)
-                               return p;
+                       setraddrport(c, argv[1]);
                        setladdr(c);
                        setlport(c);
                        break;
                case 3:
-                       p = setraddrport(c, argv[1]);
-                       if (p != NULL)
-                               return p;
-                       p = setladdrport(c, argv[2], 0);
-                       if (p != NULL)
-                               return p;
+                       setraddrport(c, argv[1]);
+                       setladdrport(c, argv[2], 0);
+                       break;
        }
 
+       /* TODO: why is an IPnoaddr (in v6 format, equivalent to v6Unspecified),
+        * a v4 format? */
        if ((memcmp(c->raddr, v4prefix, IPv4off) == 0 &&
                 memcmp(c->laddr, v4prefix, IPv4off) == 0)
                || ipcmp(c->raddr, IPnoaddr) == 0)
                c->ipversion = V4;
        else
                c->ipversion = V6;
-
-       return NULL;
+       /* Linux has taught people to use zeros for local interfaces.  TODO: We
+        * might need this for v6 in the future. */
+       if (!ipcmp(c->raddr, IPv4_zeroes))
+               ipmove(c->raddr, IPv4_loopback);
 }
 
 /*
@@ -995,14 +1076,12 @@ static void connectctlmsg(struct Proto *x, struct conv *c, struct cmdbuf *cb)
        char *p;
 
        if (c->state != 0)
-               error(Econinuse);
+               error(EBUSY, ERROR_FIXME);
        c->state = Connecting;
        c->cerr[0] = '\0';
        if (x->connect == NULL)
-               error("connect not supported");
-       p = x->connect(c, cb->f, cb->nf);
-       if (p != NULL)
-               error(p);
+               error(EFAIL, "connect not supported");
+       x->connect(c, cb->f, cb->nf);
 
        qunlock(&c->qlock);
        if (waserror()) {
@@ -1014,21 +1093,22 @@ static void connectctlmsg(struct Proto *x, struct conv *c, struct cmdbuf *cb)
        poperror();
 
        if (c->cerr[0] != '\0')
-               error(c->cerr);
+               error(EFAIL, c->cerr);
 }
 
 /*
  *  called by protocol announce routine to set addresses
  */
-char *Fsstdannounce(struct conv *c, char *argv[], int argc)
+void Fsstdannounce(struct conv *c, char *argv[], int argc)
 {
        memset(c->raddr, 0, sizeof(c->raddr));
        c->rport = 0;
        switch (argc) {
                default:
-                       return "bad args to announce";
+                       error(EINVAL, "bad args to announce");
                case 2:
-                       return setladdrport(c, argv[1], 1);
+                       setladdrport(c, argv[1], 1);
+                       break;
        }
 }
 
@@ -1046,14 +1126,12 @@ static void announcectlmsg(struct Proto *x, struct conv *c, struct cmdbuf *cb)
        char *p;
 
        if (c->state != 0)
-               error(Econinuse);
+               error(EBUSY, ERROR_FIXME);
        c->state = Announcing;
        c->cerr[0] = '\0';
        if (x->announce == NULL)
-               error("announce not supported");
-       p = x->announce(c, cb->f, cb->nf);
-       if (p != NULL)
-               error(p);
+               error(EFAIL, "announce not supported");
+       x->announce(c, cb->f, cb->nf);
 
        qunlock(&c->qlock);
        if (waserror()) {
@@ -1065,32 +1143,184 @@ static void announcectlmsg(struct Proto *x, struct conv *c, struct cmdbuf *cb)
        poperror();
 
        if (c->cerr[0] != '\0')
-               error(c->cerr);
+               error(EFAIL, c->cerr);
 }
 
 /*
  *  called by protocol bind routine to set addresses
  */
-char *Fsstdbind(struct conv *c, char *argv[], int argc)
+void Fsstdbind(struct conv *c, char *argv[], int argc)
 {
        switch (argc) {
                default:
-                       return "bad args to bind";
+                       error(EINVAL, "bad args to bind");
                case 2:
-                       return setladdrport(c, argv[1], 0);
+                       setladdrport(c, argv[1], 0);
+                       break;
        }
 }
 
 static void bindctlmsg(struct Proto *x, struct conv *c, struct cmdbuf *cb)
 {
-       char *p;
-
        if (x->bind == NULL)
-               p = Fsstdbind(c, cb->f, cb->nf);
+               Fsstdbind(c, cb->f, cb->nf);
+       else
+               x->bind(c, cb->f, cb->nf);
+}
+
+/* Helper, called by protocols to use the bypass.
+ *
+ * This is a bit nasty due to the overall nastiness of #ip.  We need to lock
+ * before checking the state and hold the qlock throughout, because a concurrent
+ * closeconv() could tear down the bypass.  Specifically, it could free the
+ * bypass queues.  The root issue is that conversation lifetimes are not managed
+ * well.
+ *
+ * If we fail, it's our responsibility to consume (free) the block(s). */
+void bypass_or_drop(struct conv *cv, struct block *bp)
+{
+       qlock(&cv->qlock);
+       if (cv->state == Bypass)
+               qpass(cv->rq, bp);
        else
-               p = x->bind(c, cb->f, cb->nf);
-       if (p != NULL)
-               error(p);
+               freeblist(bp);
+       qunlock(&cv->qlock);
+}
+
+/* Push the block directly to the approprite ipoput function.
+ *
+ * It's the protocol's responsibility (and thus ours here) to make sure there is
+ * at least the right amount of the IP header in the block (ipoput{4,6} assumes
+ * it has the right amount, and the other protocols account for the IP header in
+ * their own header).
+ *
+ * For the TTL and TOS, we just use the default ones.  If we want, we could look
+ * into the actual block and see what the user wanted, though we're bypassing
+ * the protocol layer, not the IP layer. */
+static void proto_bypass_kick(void *arg, struct block *bp)
+{
+       struct conv *cv = (struct conv*)arg;
+       uint8_t vers_nibble;
+       struct Fs *f;
+
+       f = cv->p->f;
+
+       bp = pullupblock(bp, 1);
+       if (!bp)
+               error(EINVAL, "Proto bypass unable to pullup a byte!");
+       vers_nibble = *(uint8_t*)bp->rp & 0xf0;
+       switch (vers_nibble) {
+       case IP_VER4:
+               bp = pullupblock(bp, IPV4HDR_LEN);
+               if (!bp)
+                       error(EINVAL, "Proto bypass unable to pullup v4 header");
+               ipoput4(f, bp, FALSE, MAXTTL, DFLTTOS, NULL);
+               break;
+       case IP_VER6:
+               bp = pullupblock(bp, IPV6HDR_LEN);
+               if (!bp)
+                       error(EINVAL, "Proto bypass unable to pullup v6 header");
+               ipoput6(f, bp, FALSE, MAXTTL, DFLTTOS, NULL);
+               break;
+       default:
+               error(EINVAL, "Proto bypass block had unknown IP version 0x%x",
+                     vers_nibble);
+       }
+}
+
+/* Sets up cv for the protocol bypass.  We use different queues for two reasons:
+ * 1) To be protocol independent.  For instance, TCP and UDP could use very
+ * different QIO styles.
+ * 2) To set up our own kick/bypass method.  Note how udpcreate() and here uses
+ * qbypass() (just blast it out), while TCP uses qopen() with a kick.  TCP still
+ * follows queuing discipline.
+ *
+ * It's like we are our own protocol, the bypass protocol, when it comes to how
+ * we interact with qio.  The conv still is of the real protocol type (e.g.
+ * TCP).
+ *
+ * Note that we can't free the old queues.  The way #ip works, the queues are
+ * created when the conv is created, but the conv is never freed.  It's like a
+ * slab allocator that never frees objects, but just reinitializes them a
+ * little.
+ *
+ * For the queues, we're basically like UDP:
+ * - We take packets for rq and drop on overflow.
+ * - rq is also Qmsg, but we also have Qcoalesce, to ignore out zero-len blocks
+ * - We kick for our outbound (wq) messages.
+ *
+ * Note that Qmsg can drop parts of packets.  It's up to the user to read
+ * enough.  If they didn't read enough, the extra is dropped.  This is similar
+ * to SOCK_DGRAM and recvfrom().  Minus major changes, there's no nice way to
+ * get individual messages with read().  Userspace using the bypass will need to
+ * find out the MTU of the NIC the IP stack is attached to, and make sure to
+ * read in at least that amount each time. */
+static void setup_proto_qio_bypass(struct conv *cv)
+{
+       cv->rq_save = cv->rq;
+       cv->wq_save = cv->wq;
+       cv->rq = qopen(BYPASS_QMAX, Qmsg | Qcoalesce, 0, 0);
+       cv->wq = qbypass(proto_bypass_kick, cv);
+}
+
+static void undo_proto_qio_bypass(struct conv *cv)
+{
+       qfree(cv->rq);
+       qfree(cv->wq);
+       cv->rq = cv->rq_save;
+       cv->wq = cv->wq_save;
+       cv->rq_save = NULL;
+       cv->wq_save = NULL;
+}
+
+void Fsstdbypass(struct conv *cv, char *argv[], int argc)
+{
+       memset(cv->raddr, 0, sizeof(cv->raddr));
+       cv->rport = 0;
+       switch (argc) {
+       case 2:
+               setladdrport(cv, argv[1], 1);
+               break;
+       default:
+               error(EINVAL, "Bad args (was %d, need 2) to bypass", argc);
+       }
+}
+
+static void bypassctlmsg(struct Proto *x, struct conv *cv, struct cmdbuf *cb)
+{
+       if (!x->bypass)
+               error(EFAIL, "Protocol %s does not support bypass", x->name);
+       /* The protocol needs to set the port (usually by calling Fsstdbypass) and
+        * then do whatever it needs to make sure it can find the conv again during
+        * receive (usually by adding to a hash table). */
+       x->bypass(cv, cb->f, cb->nf);
+       setup_proto_qio_bypass(cv);
+       cv->state = Bypass;
+}
+
+static void shutdownctlmsg(struct conv *cv, struct cmdbuf *cb)
+{
+       if (cb->nf < 2)
+               goto err;
+       if (!strcmp(cb->f[1], "rd")) {
+               qhangup(cv->rq, "shutdown");
+               if (cv->p->shutdown)
+                       cv->p->shutdown(cv, SHUT_RD);
+       } else if (!strcmp(cb->f[1], "wr")) {
+               qhangup(cv->wq, "shutdown");
+               if (cv->p->shutdown)
+                       cv->p->shutdown(cv, SHUT_WR);
+       } else if (!strcmp(cb->f[1], "rdwr")) {
+               qhangup(cv->rq, "shutdown");
+               qhangup(cv->wq, "shutdown");
+               if (cv->p->shutdown)
+                       cv->p->shutdown(cv, SHUT_RDWR);
+       } else {
+               goto err;
+       }
+       return;
+err:
+       error(EINVAL, "shutdown [rx|tx|rxtx]");
 }
 
 static void tosctlmsg(struct conv *c, struct cmdbuf *cb)
@@ -1109,6 +1339,22 @@ static void ttlctlmsg(struct conv *c, struct cmdbuf *cb)
                c->ttl = atoi(cb->f[1]);
 }
 
+/* Binds a conversation, as if the user wrote "bind *" into ctl. */
+static void autobind(struct conv *cv)
+{
+       ERRSTACK(1);
+       struct cmdbuf *cb;
+
+       cb = parsecmd("bind *", 7);
+       if (waserror()) {
+               kfree(cb);
+               nexterror();
+       }
+       bindctlmsg(cv->p, cv, cb);
+       poperror();
+       kfree(cb);
+}
+
 static long ipwrite(struct chan *ch, void *v, long n, int64_t off)
 {
        ERRSTACK(1);
@@ -1125,15 +1371,18 @@ static long ipwrite(struct chan *ch, void *v, long n, int64_t off)
 
        switch (TYPE(ch->qid)) {
                default:
-                       error(Eperm);
+                       error(EPERM, ERROR_FIXME);
                case Qdata:
                        x = f->p[PROTO(ch->qid)];
                        c = x->conv[CONV(ch->qid)];
-
-                       if (c->wq == NULL)
-                               error(Eperm);
-
-                       qwrite(c->wq, a, n);
+                       /* connection-less protocols (UDP) can write without manually
+                        * binding. */
+                       if (c->lport == 0)
+                               autobind(c);
+                       if (ch->flag & O_NONBLOCK)
+                               qwrite_nonblock(c->wq, a, n);
+                       else
+                               qwrite(c->wq, a, n);
                        break;
                case Qarp:
                        return arpwrite(f, a, n);
@@ -1156,13 +1405,17 @@ static long ipwrite(struct chan *ch, void *v, long n, int64_t off)
                                nexterror();
                        }
                        if (cb->nf < 1)
-                               error("short control request");
+                               error(EFAIL, "short control request");
                        if (strcmp(cb->f[0], "connect") == 0)
                                connectctlmsg(x, c, cb);
                        else if (strcmp(cb->f[0], "announce") == 0)
                                announcectlmsg(x, c, cb);
                        else if (strcmp(cb->f[0], "bind") == 0)
                                bindctlmsg(x, c, cb);
+                       else if (strcmp(cb->f[0], "bypass") == 0)
+                               bypassctlmsg(x, c, cb);
+                       else if (strcmp(cb->f[0], "shutdown") == 0)
+                               shutdownctlmsg(c, cb);
                        else if (strcmp(cb->f[0], "ttl") == 0)
                                ttlctlmsg(c, cb);
                        else if (strcmp(cb->f[0], "tos") == 0)
@@ -1171,32 +1424,30 @@ static long ipwrite(struct chan *ch, void *v, long n, int64_t off)
                                c->ignoreadvice = 1;
                        else if (strcmp(cb->f[0], "addmulti") == 0) {
                                if (cb->nf < 2)
-                                       error("addmulti needs interface address");
+                                       error(EFAIL, "addmulti needs interface address");
                                if (cb->nf == 2) {
                                        if (!ipismulticast(c->raddr))
-                                               error("addmulti for a non multicast address");
+                                               error(EFAIL, "addmulti for a non multicast address");
                                        parseip(ia, cb->f[1]);
                                        ipifcaddmulti(c, c->raddr, ia);
                                } else {
                                        parseip(ma, cb->f[2]);
                                        if (!ipismulticast(ma))
-                                               error("addmulti for a non multicast address");
+                                               error(EFAIL, "addmulti for a non multicast address");
                                        parseip(ia, cb->f[1]);
                                        ipifcaddmulti(c, ma, ia);
                                }
                        } else if (strcmp(cb->f[0], "remmulti") == 0) {
                                if (cb->nf < 2)
-                                       error("remmulti needs interface address");
+                                       error(EFAIL, "remmulti needs interface address");
                                if (!ipismulticast(c->raddr))
-                                       error("remmulti for a non multicast address");
+                                       error(EFAIL, "remmulti for a non multicast address");
                                parseip(ia, cb->f[1]);
                                ipifcremmulti(c, c->raddr, ia);
                        } else if (x->ctl != NULL) {
-                               p = x->ctl(c, cb->f, cb->nf);
-                               if (p != NULL)
-                                       error(p);
+                               x->ctl(c, cb->f, cb->nf);
                        } else
-                               error("unknown control request");
+                               error(EFAIL, "unknown control request");
                        qunlock(&c->qlock);
                        kfree(cb);
                        poperror();
@@ -1207,50 +1458,158 @@ static long ipwrite(struct chan *ch, void *v, long n, int64_t off)
 static long ipbwrite(struct chan *ch, struct block *bp, uint32_t offset)
 {
        struct conv *c;
-       struct Proto *x;
-       struct Fs *f;
        int n;
 
        switch (TYPE(ch->qid)) {
                case Qdata:
-                       f = ipfs[ch->dev];
-                       x = f->p[PROTO(ch->qid)];
-                       c = x->conv[CONV(ch->qid)];
-
-                       if (c->wq == NULL)
-                               error(Eperm);
-
+                       c = chan2conv(ch);
                        if (bp->next)
                                bp = concatblock(bp);
                        n = BLEN(bp);
-                       qbwrite(c->wq, bp);
+                       if (ch->flag & O_NONBLOCK)
+                               qbwrite_nonblock(c->wq, bp);
+                       else
+                               qbwrite(c->wq, bp);
                        return n;
                default:
                        return devbwrite(ch, bp, offset);
        }
 }
 
+static void ip_wake_cb(struct queue *q, void *data, int filter)
+{
+       struct conv *conv = (struct conv*)data;
+       struct fd_tap *tap_i;
+       /* For these two, we want to ignore events on the opposite end of the
+        * queues.  For instance, we want to know when the WQ is writable.  Our
+        * writes will actually make it readable - we don't want to trigger a tap
+        * for that.  However, qio doesn't know how/why we are using a queue, or
+        * even who the ends are (hence the callbacks) */
+       if ((filter & FDTAP_FILT_READABLE) && (q == conv->wq))
+               return;
+       if ((filter & FDTAP_FILT_WRITABLE) && (q == conv->rq))
+               return;
+       /* At this point, we have an event we want to send to our taps (if any).
+        * The lock protects list integrity and the existence of the tap.
+        *
+        * Previously, I thought of using the conv qlock.  That actually breaks, due
+        * to weird usages of the qlock (someone holds it for a long time, blocking
+        * the inbound wakeup from etherread4).
+        *
+        * I opted for a spinlock for a couple reasons:
+        * - fire_tap should not block.  ideally it'll be fast too (it's mostly a
+        * send_event).
+        * - our callers might not want to block.  A lot of network wakeups will
+        * come network processes (etherread4) or otherwise unrelated to this
+        * particular conversation.  I'd rather do something like fire off a KMSG
+        * than block those.
+        * - if fire_tap takes a while, holding the lock only slows down other
+        * events on this *same* conversation, or other tap registration.  not a
+        * huge deal. */
+       spin_lock(&conv->tap_lock);
+       SLIST_FOREACH(tap_i, &conv->data_taps, link)
+               fire_tap(tap_i, filter);
+       spin_unlock(&conv->tap_lock);
+}
+
+int iptapfd(struct chan *chan, struct fd_tap *tap, int cmd)
+{
+       struct conv *conv = chan2conv(chan);
+       int ret;
+
+       #define DEVIP_LEGAL_DATA_TAPS (FDTAP_FILT_READABLE | FDTAP_FILT_WRITABLE | \
+                                      FDTAP_FILT_HANGUP | FDTAP_FILT_PRIORITY |   \
+                                      FDTAP_FILT_ERROR)
+       #define DEVIP_LEGAL_LISTEN_TAPS (FDTAP_FILT_READABLE | FDTAP_FILT_HANGUP)
+
+       switch (TYPE(chan->qid)) {
+               case Qdata:
+                       if (tap->filter & ~DEVIP_LEGAL_DATA_TAPS) {
+                               set_errno(ENOSYS);
+                               set_errstr("Unsupported #%s data tap %p, must be %p", devname(),
+                                          tap->filter, DEVIP_LEGAL_DATA_TAPS);
+                               return -1;
+                       }
+                       spin_lock(&conv->tap_lock);
+                       switch (cmd) {
+                               case (FDTAP_CMD_ADD):
+                                       if (SLIST_EMPTY(&conv->data_taps)) {
+                                               qio_set_wake_cb(conv->rq, ip_wake_cb, conv);
+                                               qio_set_wake_cb(conv->wq, ip_wake_cb, conv);
+                                       }
+                                       SLIST_INSERT_HEAD(&conv->data_taps, tap, link);
+                                       ret = 0;
+                                       break;
+                               case (FDTAP_CMD_REM):
+                                       SLIST_REMOVE(&conv->data_taps, tap, fd_tap, link);
+                                       if (SLIST_EMPTY(&conv->data_taps)) {
+                                               qio_set_wake_cb(conv->rq, 0, conv);
+                                               qio_set_wake_cb(conv->wq, 0, conv);
+                                       }
+                                       ret = 0;
+                                       break;
+                               default:
+                                       set_errno(ENOSYS);
+                                       set_errstr("Unsupported #%s data tap command %p",
+                                                  devname(), cmd);
+                                       ret = -1;
+                       }
+                       spin_unlock(&conv->tap_lock);
+                       return ret;
+               case Qlisten:
+                       if (tap->filter & ~DEVIP_LEGAL_LISTEN_TAPS) {
+                               set_errno(ENOSYS);
+                               set_errstr("Unsupported #%s listen tap %p, must be %p",
+                                          devname(), tap->filter, DEVIP_LEGAL_LISTEN_TAPS);
+                               return -1;
+                       }
+                       spin_lock(&conv->tap_lock);
+                       switch (cmd) {
+                               case (FDTAP_CMD_ADD):
+                                       SLIST_INSERT_HEAD(&conv->listen_taps, tap, link);
+                                       ret = 0;
+                                       break;
+                               case (FDTAP_CMD_REM):
+                                       SLIST_REMOVE(&conv->listen_taps, tap, fd_tap, link);
+                                       ret = 0;
+                                       break;
+                               default:
+                                       set_errno(ENOSYS);
+                                       set_errstr("Unsupported #%s listen tap command %p",
+                                                  devname(), cmd);
+                                       ret = -1;
+                       }
+                       spin_unlock(&conv->tap_lock);
+                       return ret;
+               default:
+                       set_errno(ENOSYS);
+                       set_errstr("Can't tap #%s file type %d", devname(),
+                                  TYPE(chan->qid));
+                       return -1;
+       }
+}
+
 struct dev ipdevtab __devtab = {
-       'I',
-       "ip",
-
-       ipreset,
-       ipinit,
-       devshutdown,
-       ipattach,
-       ipwalk,
-       ipstat,
-       ipopen,
-       devcreate,
-       ipclose,
-       ipread,
-       ipbread,
-       ipwrite,
-       ipbwrite,
-       devremove,
-       ipwstat,
-       devpower,
-       ipchaninfo,
+       .name = "ip",
+
+       .reset = ipreset,
+       .init = ipinit,
+       .shutdown = devshutdown,
+       .attach = ipattach,
+       .walk = ipwalk,
+       .stat = ipstat,
+       .open = ipopen,
+       .create = devcreate,
+       .close = ipclose,
+       .read = ipread,
+       .bread = ipbread,
+       .write = ipwrite,
+       .bwrite = ipbwrite,
+       .remove = devremove,
+       .wstat = ipwstat,
+       .power = devpower,
+       .chaninfo = ipchaninfo,
+       .tapfd = iptapfd,
 };
 
 int Fsproto(struct Fs *f, struct Proto *p)
@@ -1305,11 +1664,16 @@ retry:
                if (c == NULL) {
                        c = kzmalloc(sizeof(struct conv), 0);
                        if (c == NULL)
-                               error(Enomem);
+                               error(ENOMEM,
+                                     "conv kzmalloc(%d, 0) failed in Fsprotoclone",
+                                     sizeof(struct conv));
                        qlock_init(&c->qlock);
                        qlock_init(&c->listenq);
                        rendez_init(&c->cr);
                        rendez_init(&c->listenr);
+                       SLIST_INIT(&c->data_taps);      /* already = 0; set to be futureproof */
+                       SLIST_INIT(&c->listen_taps);
+                       spinlock_init(&c->tap_lock);
                        qlock(&c->qlock);
                        c->p = p;
                        c->x = pp - p->conv;
@@ -1317,13 +1681,16 @@ retry:
                                c->ptcl = kzmalloc(p->ptclsize, 0);
                                if (c->ptcl == NULL) {
                                        kfree(c);
-                                       error(Enomem);
+                                       error(ENOMEM,
+                                             "ptcl kzmalloc(%d, 0) failed in Fsprotoclone",
+                                             p->ptclsize);
                                }
                        }
                        *pp = c;
                        p->ac++;
                        c->eq = qopen(1024, Qmsg, 0, 0);
                        (*p->create) (c);
+                       assert(c->rq && c->wq);
                        break;
                }
                if (canqlock(&c->qlock)) {
@@ -1367,10 +1734,9 @@ retry:
 int Fsconnected(struct conv *c, char *msg)
 {
        if (msg != NULL && *msg != '\0')
-               strncpy(c->cerr, msg, sizeof(c->cerr));
+               strlcpy(c->cerr, msg, sizeof(c->cerr));
 
        switch (c->state) {
-
                case Announcing:
                        c->state = Announced;
                        break;
@@ -1397,6 +1763,17 @@ struct Proto *Fsrcvpcolx(struct Fs *f, uint8_t proto)
        return f->t2p[proto];
 }
 
+static void fire_listener_taps(struct conv *conv)
+{
+       struct fd_tap *tap_i;
+       if (SLIST_EMPTY(&conv->listen_taps))
+               return;
+       spin_lock(&conv->tap_lock);
+       SLIST_FOREACH(tap_i, &conv->listen_taps, link)
+               fire_tap(tap_i, FDTAP_FILT_READABLE);
+       spin_unlock(&conv->tap_lock);
+}
+
 /*
  *  called with protocol locked
  */
@@ -1434,6 +1811,7 @@ struct conv *Fsnewcall(struct conv *c, uint8_t * raddr, uint16_t rport,
        qunlock(&c->qlock);
 
        rendez_wakeup(&c->listenr);
+       fire_listener_taps(c);
 
        return nc;
 }
@@ -1441,9 +1819,9 @@ struct conv *Fsnewcall(struct conv *c, uint8_t * raddr, uint16_t rport,
 static long ndbwrite(struct Fs *f, char *a, uint32_t off, int n)
 {
        if (off > strlen(f->ndb))
-               error(Eio);
+               error(EIO, ERROR_FIXME);
        if (off + n >= sizeof(f->ndb) - 1)
-               error(Eio);
+               error(EIO, ERROR_FIXME);
        memmove(f->ndb + off, a, n);
        f->ndb[off + n] = 0;
        f->ndbvers++;