/*
* Append mbuf chain m to the last record in the
* socket buffer sb. The additional space associated
* the mbuf chain is recorded in sb. Empty mbufs are
* discarded and mbufs are compacted where possible.
*/
void
sbappend(struct sockbuf *sb, struct mbuf *m)
{
struct mbuf *n;
KASSERT(solocked(sb->sb_so));
if (m == 0)
return;
#ifdef MBUFTRACE
m_claimm(m, sb->sb_mowner);
#endif
SBLASTRECORDCHK(sb, "sbappend 1");
if ((n = sb->sb_lastrecord) != NULL) {
/*
* XXX Would like to simply use sb_mbtail here, but
* XXX I need to verify that I won't miss an EOR that
* XXX way.
*/
do {
if (n->m_flags & M_EOR) {
sbappendrecord(sb, m); /* XXXXXX!!!! */
return;
}
} while (n->m_next && (n = n->m_next));
} else {
/*
* If this is the first record in the socket buffer, it's
* also the last record.
*/
sb->sb_lastrecord = m;
}
sbcompress(sb, m, n);
SBLASTRECORDCHK(sb, "sbappend 2");
}
/*
* This version of sbappend() should only be used when the caller
* absolutely knows that there will never be more than one record
* in the socket buffer, that is, a stream protocol (such as TCP).
*/
void
sbappendstream(struct sockbuf *sb, struct mbuf *m)
{
KDASSERT(m->m_nextpkt == NULL);
KASSERT(sb->sb_mb == sb->sb_lastrecord);
SBLASTMBUFCHK(sb, __func__);
sbcompress(sb, m, sb->sb_mbtail);
sb->sb_lastrecord = sb->sb_mb;
SBLASTRECORDCHK(sb, __func__);
}
/*
* This version of sbappend() should only be used when the caller absolutely
* knows that there will never be more than one record in the socket buffer,
* that is, a stream protocol (such as TCP).
*/
void
sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
{
SOCKBUF_LOCK_ASSERT(sb);
KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
SBLASTMBUFCHK(sb);
sbcompress(sb, m, sb->sb_mbtail);
sb->sb_lastrecord = sb->sb_mb;
SBLASTRECORDCHK(sb);
}
/*
* As above, except the mbuf chain
* begins a new record.
*/
void
sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
{
struct mbuf *m;
if (m0 == NULL)
return;
/*
* Put the first mbuf on the queue.
* Note this permits zero length records.
*/
sballoc(sb, m0);
SBLASTRECORDCHK(sb, "sbappendrecord 1");
SBLINKRECORD(sb, m0);
m = m0->m_next;
m0->m_next = NULL;
if (m && (m0->m_flags & M_EOR)) {
m0->m_flags &= ~M_EOR;
m->m_flags |= M_EOR;
}
sbcompress(sb, m, m0);
SBLASTRECORDCHK(sb, "sbappendrecord 2");
}
/* Helper routine that appends data, control, and address to a sockbuf. */
static int
sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
{
struct mbuf *m, *n, *nlast;
#if MSIZE <= 256
if (asa->sa_len > MLEN)
return (0);
#endif
m = m_get(M_NOWAIT, MT_SONAME);
if (m == NULL)
return (0);
m->m_len = asa->sa_len;
bcopy(asa, mtod(m, caddr_t), asa->sa_len);
if (m0) {
m_clrprotoflags(m0);
m_tag_delete_chain(m0, NULL);
/*
* Clear some persistent info from pkthdr.
* We don't use m_demote(), because some netgraph consumers
* expect M_PKTHDR presence.
*/
m0->m_pkthdr.rcvif = NULL;
m0->m_pkthdr.flowid = 0;
m0->m_pkthdr.csum_flags = 0;
m0->m_pkthdr.fibnum = 0;
m0->m_pkthdr.rsstype = 0;
}
if (ctrl_last)
ctrl_last->m_next = m0; /* concatenate data to control */
else
control = m0;
m->m_next = control;
for (n = m; n->m_next != NULL; n = n->m_next)
sballoc(sb, n);
sballoc(sb, n);
nlast = n;
SBLINKRECORD(sb, m);
sb->sb_mbtail = nlast;
SBLASTMBUFCHK(sb);
SBLASTRECORDCHK(sb);
return (1);
}
/*
* Append address and data, and optionally, control (ancillary) data to the
* receive queue of a socket. If present, m0 must include a packet header
* with total length. Returns 0 if no space in sockbuf or insufficient
* mbufs.
*/
int
sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
struct mbuf *m0, struct mbuf *control)
{
struct mbuf *m, *n, *nlast;
int space = asa->sa_len;
SOCKBUF_LOCK_ASSERT(sb);
if (m0 && (m0->m_flags & M_PKTHDR) == 0)
panic("sbappendaddr_locked");
if (m0)
space += m0->m_pkthdr.len;
space += m_length(control, &n);
if (space > sbspace(sb))
return (0);
#if MSIZE <= 256
if (asa->sa_len > MLEN)
return (0);
#endif
MGET(m, M_DONTWAIT, MT_SONAME);
if (m == 0)
return (0);
m->m_len = asa->sa_len;
bcopy(asa, mtod(m, caddr_t), asa->sa_len);
if (n){
CHECK_ADD_LINKCNT(n, m0, NULL, "sbappendaddr_locked");
n->m_next = m0; /* concatenate data to control */
}else
control = m0;
CHECK_ADD_LINKCNT(m, control, NULL, "sbappendaddr_locked");
m->m_next = control;
for (n = m; n->m_next != NULL; n = n->m_next)
sballoc(sb, n);
sballoc(sb, n);
nlast = n;
SBLINKRECORD(sb, m);
sb->sb_mbtail = nlast;
SBLASTMBUFCHK(sb);
SBLASTRECORDCHK(sb);
return (1);
}
/*
* This version of sbappend() should only be used when the caller absolutely
* knows that there will never be more than one record in the socket buffer,
* that is, a stream protocol (such as TCP).
*/
void
sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
{
SOCKBUF_LOCK_ASSERT(sb);
KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
SBLASTMBUFCHK(sb);
/* Remove all packet headers and mbuf tags to get a pure data chain. */
m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
sbcompress(sb, m, sb->sb_mbtail);
sb->sb_lastrecord = sb->sb_mb;
SBLASTRECORDCHK(sb);
}
/*
* This version of sbappend() should only be used when the caller
* absolutely knows that there will never be more than one record
* in the socket buffer, that is, a stream protocol (such as TCP).
*/
void
sbappendstream(struct sockbuf *sb, struct mbuf *m)
{
KASSERT(solocked(sb->sb_so));
KDASSERT(m->m_nextpkt == NULL);
KASSERT(sb->sb_mb == sb->sb_lastrecord);
SBLASTMBUFCHK(sb, __func__);
#ifdef MBUFTRACE
m_claimm(m, sb->sb_mowner);
#endif
service_reflect_tx_socket_buf(sb->sb_so, m->m_len);
sbcompress(sb, m, sb->sb_mbtail);
sb->sb_lastrecord = sb->sb_mb;
SBLASTRECORDCHK(sb, __func__);
}
/*
* Append address and data, and optionally, control (ancillary) data
* to the receive queue of a socket. If present,
* m0 must include a packet header with total length.
* Returns 0 if no space in sockbuf or insufficient mbufs.
*/
int
sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0,
struct mbuf *control)
{
struct mbuf *m, *n, *nlast;
int space, len;
KASSERT(solocked(sb->sb_so));
space = asa->sa_len;
if (m0 != NULL) {
if ((m0->m_flags & M_PKTHDR) == 0)
panic("sbappendaddr");
space += m0->m_pkthdr.len;
#ifdef MBUFTRACE
m_claimm(m0, sb->sb_mowner);
#endif
}
for (n = control; n; n = n->m_next) {
space += n->m_len;
MCLAIM(n, sb->sb_mowner);
if (n->m_next == 0) /* keep pointer to last control buf */
break;
}
if (space > sbspace(sb))
return (0);
MGET(m, M_DONTWAIT, MT_SONAME);
if (m == 0)
return (0);
MCLAIM(m, sb->sb_mowner);
/*
* XXX avoid 'comparison always true' warning which isn't easily
* avoided.
*/
len = asa->sa_len;
if (len > MLEN) {
MEXTMALLOC(m, asa->sa_len, M_NOWAIT);
if ((m->m_flags & M_EXT) == 0) {
m_free(m);
return (0);
}
}
m->m_len = asa->sa_len;
memcpy(mtod(m, void *), asa, asa->sa_len);
if (n)
n->m_next = m0; /* concatenate data to control */
else
control = m0;
m->m_next = control;
SBLASTRECORDCHK(sb, "sbappendaddr 1");
for (n = m; n->m_next != NULL; n = n->m_next)
sballoc(sb, n);
sballoc(sb, n);
nlast = n;
SBLINKRECORD(sb, m);
sb->sb_mbtail = nlast;
SBLASTMBUFCHK(sb, "sbappendaddr");
SBLASTRECORDCHK(sb, "sbappendaddr 2");
return (1);
}
int
sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa,
struct mbuf *m0, int sbprio)
{
int space;
struct mbuf *m, *n, *n0, *nlast;
int error;
KASSERT(solocked(sb->sb_so));
/*
* XXX sbprio reserved for encoding priority of this* request:
* SB_PRIO_NONE --> honour normal sb limits
* SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space,
* take whole chain. Intended for large requests
* that should be delivered atomically (all, or none).
* SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow
* over normal socket limits, for messages indicating
* buffer overflow in earlier normal/lower-priority messages
* SB_PRIO_BESTEFFORT --> ignore limits entirely.
* Intended for kernel-generated messages only.
* Up to generator to avoid total mbuf resource exhaustion.
*/
(void)sbprio;
if (m0 && (m0->m_flags & M_PKTHDR) == 0)
panic("sbappendaddrchain");
space = sbspace(sb);
#ifdef notyet
/*
* Enforce SB_PRIO_* limits as described above.
*/
#endif
n0 = NULL;
nlast = NULL;
for (m = m0; m; m = m->m_nextpkt) {
struct mbuf *np;
#ifdef MBUFTRACE
m_claimm(m, sb->sb_mowner);
#endif
/* Prepend sockaddr to this record (m) of input chain m0 */
n = m_prepend_sockaddr(sb, m, asa);
if (n == NULL) {
error = ENOBUFS;
goto bad;
}
/* Append record (asa+m) to end of new chain n0 */
if (n0 == NULL) {
n0 = n;
} else {
nlast->m_nextpkt = n;
}
/* Keep track of last record on new chain */
nlast = n;
for (np = n; np; np = np->m_next)
sballoc(sb, np);
}
SBLASTRECORDCHK(sb, "sbappendaddrchain 1");
/* Drop the entire chain of (asa+m) records onto the socket */
SBLINKRECORDCHAIN(sb, n0, nlast);
SBLASTRECORDCHK(sb, "sbappendaddrchain 2");
for (m = nlast; m->m_next; m = m->m_next)
;
sb->sb_mbtail = m;
SBLASTMBUFCHK(sb, "sbappendaddrchain");
return (1);
bad:
/*
* On error, free the prepended addreseses. For consistency
* with sbappendaddr(), leave it to our caller to free
* the input record chain passed to us as m0.
*/
while ((n = n0) != NULL) {
struct mbuf *np;
/* Undo the sballoc() of this record */
for (np = n; np; np = np->m_next)
sbfree(sb, np);
n0 = n->m_nextpkt; /* iterate at next prepended address */
MFREE(n, np); /* free prepended address (not data) */
}
return 0;
}
static int
kttcp_soreceive(struct socket *so, unsigned long long slen,
unsigned long long *done, struct lwp *l, int *flagsp)
{
struct mbuf *m, **mp;
int flags, len, error, offset, moff, type;
long long orig_resid, resid;
const struct protosw *pr;
struct mbuf *nextrecord;
pr = so->so_proto;
mp = NULL;
type = 0;
resid = orig_resid = slen;
if (flagsp)
flags = *flagsp &~ MSG_EOR;
else
flags = 0;
if (flags & MSG_OOB) {
m = m_get(M_WAIT, MT_DATA);
solock(so);
error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
sounlock(so);
if (error)
goto bad;
do {
resid -= min(resid, m->m_len);
m = m_free(m);
} while (resid && error == 0 && m);
bad:
if (m)
m_freem(m);
return (error);
}
if (mp)
*mp = NULL;
solock(so);
restart:
if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
return (error);
m = so->so_rcv.sb_mb;
/*
* If we have less data than requested, block awaiting more
* (subject to any timeout) if:
* 1. the current count is less than the low water mark,
* 2. MSG_WAITALL is set, and it is possible to do the entire
* receive operation at once if we block (resid <= hiwat), or
* 3. MSG_DONTWAIT is not set.
* If MSG_WAITALL is set but resid is larger than the receive buffer,
* we have to do the receive in sections, and thus risk returning
* a short count if a timeout or signal occurs after we start.
*/
if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
so->so_rcv.sb_cc < resid) &&
(so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
((flags & MSG_WAITALL) && resid <= so->so_rcv.sb_hiwat)) &&
m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
#ifdef DIAGNOSTIC
if (m == NULL && so->so_rcv.sb_cc)
panic("receive 1");
#endif
if (so->so_error) {
if (m)
goto dontblock;
error = so->so_error;
if ((flags & MSG_PEEK) == 0)
so->so_error = 0;
goto release;
}
if (so->so_state & SS_CANTRCVMORE) {
if (m)
goto dontblock;
else
goto release;
}
for (; m; m = m->m_next)
if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
m = so->so_rcv.sb_mb;
goto dontblock;
}
if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
(so->so_proto->pr_flags & PR_CONNREQUIRED)) {
error = ENOTCONN;
goto release;
}
if (resid == 0)
goto release;
if ((so->so_state & SS_NBIO) ||
(flags & (MSG_DONTWAIT|MSG_NBIO))) {
error = EWOULDBLOCK;
goto release;
}
sbunlock(&so->so_rcv);
error = sbwait(&so->so_rcv);
if (error) {
sounlock(so);
return (error);
}
goto restart;
}
dontblock:
/*
* On entry here, m points to the first record of the socket buffer.
* While we process the initial mbufs containing address and control
* info, we save a copy of m->m_nextpkt into nextrecord.
*/
#ifdef notyet /* XXXX */
if (uio->uio_lwp)
uio->uio_lwp->l_ru.ru_msgrcv++;
#endif
KASSERT(m == so->so_rcv.sb_mb);
SBLASTRECORDCHK(&so->so_rcv, "kttcp_soreceive 1");
SBLASTMBUFCHK(&so->so_rcv, "kttcp_soreceive 1");
nextrecord = m->m_nextpkt;
if (pr->pr_flags & PR_ADDR) {
#ifdef DIAGNOSTIC
if (m->m_type != MT_SONAME)
panic("receive 1a");
#endif
orig_resid = 0;
if (flags & MSG_PEEK) {
m = m->m_next;
} else {
sbfree(&so->so_rcv, m);
MFREE(m, so->so_rcv.sb_mb);
m = so->so_rcv.sb_mb;
}
}
while (m && m->m_type == MT_CONTROL && error == 0) {
if (flags & MSG_PEEK) {
m = m->m_next;
} else {
sbfree(&so->so_rcv, m);
MFREE(m, so->so_rcv.sb_mb);
m = so->so_rcv.sb_mb;
}
}
/*
* If m is non-NULL, we have some data to read. From now on,
* make sure to keep sb_lastrecord consistent when working on
* the last packet on the chain (nextrecord == NULL) and we
* change m->m_nextpkt.
*/
if (m) {
if ((flags & MSG_PEEK) == 0) {
m->m_nextpkt = nextrecord;
/*
* If nextrecord == NULL (this is a single chain),
* then sb_lastrecord may not be valid here if m
* was changed earlier.
*/
if (nextrecord == NULL) {
KASSERT(so->so_rcv.sb_mb == m);
so->so_rcv.sb_lastrecord = m;
}
}
type = m->m_type;
if (type == MT_OOBDATA)
flags |= MSG_OOB;
} else {
if ((flags & MSG_PEEK) == 0) {
KASSERT(so->so_rcv.sb_mb == m);
so->so_rcv.sb_mb = nextrecord;
SB_EMPTY_FIXUP(&so->so_rcv);
}
}
SBLASTRECORDCHK(&so->so_rcv, "kttcp_soreceive 2");
SBLASTMBUFCHK(&so->so_rcv, "kttcp_soreceive 2");
moff = 0;
offset = 0;
while (m && resid > 0 && error == 0) {
if (m->m_type == MT_OOBDATA) {
if (type != MT_OOBDATA)
break;
} else if (type == MT_OOBDATA)
break;
#ifdef DIAGNOSTIC
else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
panic("receive 3");
#endif
so->so_state &= ~SS_RCVATMARK;
len = resid;
if (so->so_oobmark && len > so->so_oobmark - offset)
len = so->so_oobmark - offset;
if (len > m->m_len - moff)
len = m->m_len - moff;
/*
* If mp is set, just pass back the mbufs.
* Otherwise copy them out via the uio, then free.
* Sockbuf must be consistent here (points to current mbuf,
* it points to next record) when we drop priority;
* we must note any additions to the sockbuf when we
* block interrupts again.
*/
resid -= len;
if (len == m->m_len - moff) {
if (m->m_flags & M_EOR)
flags |= MSG_EOR;
if (flags & MSG_PEEK) {
m = m->m_next;
moff = 0;
} else {
nextrecord = m->m_nextpkt;
sbfree(&so->so_rcv, m);
if (mp) {
*mp = m;
mp = &m->m_next;
so->so_rcv.sb_mb = m = m->m_next;
*mp = NULL;
} else {
MFREE(m, so->so_rcv.sb_mb);
m = so->so_rcv.sb_mb;
}
/*
* If m != NULL, we also know that
* so->so_rcv.sb_mb != NULL.
*/
KASSERT(so->so_rcv.sb_mb == m);
if (m) {
m->m_nextpkt = nextrecord;
if (nextrecord == NULL)
so->so_rcv.sb_lastrecord = m;
} else {
so->so_rcv.sb_mb = nextrecord;
SB_EMPTY_FIXUP(&so->so_rcv);
}
SBLASTRECORDCHK(&so->so_rcv,
"kttcp_soreceive 3");
SBLASTMBUFCHK(&so->so_rcv,
"kttcp_soreceive 3");
}
} else {
if (flags & MSG_PEEK)
moff += len;
else {
if (mp) {
sounlock(so);
*mp = m_copym(m, 0, len, M_WAIT);
solock(so);
}
m->m_data += len;
m->m_len -= len;
so->so_rcv.sb_cc -= len;
}
}
if (so->so_oobmark) {
if ((flags & MSG_PEEK) == 0) {
so->so_oobmark -= len;
if (so->so_oobmark == 0) {
so->so_state |= SS_RCVATMARK;
break;
}
} else {
offset += len;
if (offset == so->so_oobmark)
break;
}
}
if (flags & MSG_EOR)
break;
/*
* If the MSG_WAITALL flag is set (for non-atomic socket),
* we must not quit until "uio->uio_resid == 0" or an error
* termination. If a signal/timeout occurs, return
* with a short count but without error.
* Keep sockbuf locked against other readers.
*/
while (flags & MSG_WAITALL && m == NULL && resid > 0 &&
!sosendallatonce(so) && !nextrecord) {
if (so->so_error || so->so_state & SS_CANTRCVMORE)
break;
/*
* If we are peeking and the socket receive buffer is
* full, stop since we can't get more data to peek at.
*/
if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
break;
/*
* If we've drained the socket buffer, tell the
* protocol in case it needs to do something to
* get it filled again.
*/
if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb) {
(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
}
SBLASTRECORDCHK(&so->so_rcv,
"kttcp_soreceive sbwait 2");
SBLASTMBUFCHK(&so->so_rcv,
"kttcp_soreceive sbwait 2");
error = sbwait(&so->so_rcv);
if (error) {
sbunlock(&so->so_rcv);
sounlock(so);
return (0);
}
if ((m = so->so_rcv.sb_mb) != NULL)
nextrecord = m->m_nextpkt;
}
}
if (m && pr->pr_flags & PR_ATOMIC) {
flags |= MSG_TRUNC;
if ((flags & MSG_PEEK) == 0)
(void) sbdroprecord(&so->so_rcv);
}
if ((flags & MSG_PEEK) == 0) {
if (m == NULL) {
/*
* First part is an SB_EMPTY_FIXUP(). Second part
* makes sure sb_lastrecord is up-to-date if
* there is still data in the socket buffer.
*/
so->so_rcv.sb_mb = nextrecord;
if (so->so_rcv.sb_mb == NULL) {
so->so_rcv.sb_mbtail = NULL;
so->so_rcv.sb_lastrecord = NULL;
} else if (nextrecord->m_nextpkt == NULL)
so->so_rcv.sb_lastrecord = nextrecord;
}
SBLASTRECORDCHK(&so->so_rcv, "kttcp_soreceive 4");
SBLASTMBUFCHK(&so->so_rcv, "kttcp_soreceive 4");
if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) {
(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
}
}
if (orig_resid == resid && orig_resid &&
(flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
sbunlock(&so->so_rcv);
goto restart;
}
if (flagsp)
*flagsp |= flags;
release:
sbunlock(&so->so_rcv);
sounlock(so);
*done = slen - resid;
#if 0
printf("soreceive: error %d slen %llu resid %lld\n", error, slen, resid);
#endif
return (error);
}
/*
* Slightly changed version of sosend()
*/
static int
kttcp_sosend(struct socket *so, unsigned long long slen,
unsigned long long *done, struct lwp *l, int flags)
{
struct mbuf **mp, *m, *top;
long space, len, mlen;
int error, dontroute, atomic;
long long resid;
atomic = sosendallatonce(so);
resid = slen;
top = NULL;
/*
* In theory resid should be unsigned.
* However, space must be signed, as it might be less than 0
* if we over-committed, and we must use a signed comparison
* of space and resid. On the other hand, a negative resid
* causes us to loop sending 0-length segments to the protocol.
*/
if (resid < 0) {
error = EINVAL;
goto out;
}
dontroute =
(flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
(so->so_proto->pr_flags & PR_ATOMIC);
l->l_ru.ru_msgsnd++;
#define snderr(errno) { error = errno; goto release; }
solock(so);
restart:
if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
goto out;
do {
if (so->so_state & SS_CANTSENDMORE)
snderr(EPIPE);
if (so->so_error) {
error = so->so_error;
so->so_error = 0;
goto release;
}
if ((so->so_state & SS_ISCONNECTED) == 0) {
if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
snderr(ENOTCONN);
} else {
snderr(EDESTADDRREQ);
}
}
space = sbspace(&so->so_snd);
if (flags & MSG_OOB)
space += 1024;
if ((atomic && resid > so->so_snd.sb_hiwat))
snderr(EMSGSIZE);
if (space < resid && (atomic || space < so->so_snd.sb_lowat)) {
if (so->so_state & SS_NBIO)
snderr(EWOULDBLOCK);
SBLASTRECORDCHK(&so->so_rcv,
"kttcp_soreceive sbwait 1");
SBLASTMBUFCHK(&so->so_rcv,
"kttcp_soreceive sbwait 1");
sbunlock(&so->so_snd);
error = sbwait(&so->so_snd);
if (error)
goto out;
goto restart;
}
mp = ⊤
do {
sounlock(so);
do {
if (top == 0) {
m = m_gethdr(M_WAIT, MT_DATA);
mlen = MHLEN;
m->m_pkthdr.len = 0;
m->m_pkthdr.rcvif = NULL;
} else {
m = m_get(M_WAIT, MT_DATA);
mlen = MLEN;
}
if (resid >= MINCLSIZE && space >= MCLBYTES) {
m_clget(m, M_WAIT);
if ((m->m_flags & M_EXT) == 0)
goto nopages;
mlen = MCLBYTES;
#ifdef MAPPED_MBUFS
len = lmin(MCLBYTES, resid);
#else
if (atomic && top == 0) {
len = lmin(MCLBYTES - max_hdr,
resid);
m->m_data += max_hdr;
} else
len = lmin(MCLBYTES, resid);
#endif
space -= len;
} else {
nopages:
len = lmin(lmin(mlen, resid), space);
space -= len;
/*
* For datagram protocols, leave room
* for protocol headers in first mbuf.
*/
if (atomic && top == 0 && len < mlen)
MH_ALIGN(m, len);
}
resid -= len;
m->m_len = len;
*mp = m;
top->m_pkthdr.len += len;
if (error)
goto release;
mp = &m->m_next;
if (resid <= 0) {
if (flags & MSG_EOR)
top->m_flags |= M_EOR;
break;
}
} while (space > 0 && atomic);
solock(so);
if (so->so_state & SS_CANTSENDMORE)
snderr(EPIPE);
if (dontroute)
so->so_options |= SO_DONTROUTE;
if (resid > 0)
so->so_state |= SS_MORETOCOME;
if (flags & MSG_OOB)
error = (*so->so_proto->pr_usrreqs->pr_sendoob)(so,
top, NULL);
else
error = (*so->so_proto->pr_usrreqs->pr_send)(so,
top, NULL, NULL, l);
if (dontroute)
so->so_options &= ~SO_DONTROUTE;
if (resid > 0)
so->so_state &= ~SS_MORETOCOME;
top = 0;
mp = ⊤
if (error)
goto release;
} while (resid && space > 0);
} while (resid);
release:
sbunlock(&so->so_snd);
out:
sounlock(so);
if (top)
m_freem(top);
*done = slen - resid;
#if 0
printf("sosend: error %d slen %llu resid %lld\n", error, slen, resid);
#endif
return (error);
}