DEFINE_APITEST(peeloff, goodcase)
{
int lfd, cfd, pfd;
struct sockaddr_in addr;
socklen_t addr_len;
sctp_assoc_t assoc_id;
if ((lfd = socket(AF_INET, SOCK_STREAM, IPPROTO_SCTP)) < 0) {
RETURN_FAILED_WITH_ERRNO;
}
if (sctp_bind(lfd, INADDR_LOOPBACK, 0) < 0) {
close(lfd);
RETURN_FAILED_WITH_ERRNO;
}
if (listen(lfd, 1) < 0) {
close(lfd);
RETURN_FAILED_WITH_ERRNO;
}
if ((cfd = socket(AF_INET, SOCK_SEQPACKET, IPPROTO_SCTP)) < 0) {
close(lfd);
RETURN_FAILED_WITH_ERRNO;
}
if (sctp_bind(cfd, INADDR_LOOPBACK, 0) < 0) {
close(lfd);
close(cfd);
RETURN_FAILED_WITH_ERRNO;
}
addr_len = (socklen_t)sizeof(struct sockaddr_in);
if (getsockname (lfd, (struct sockaddr *) &addr, &addr_len) < 0) {
close(lfd);
close(cfd);
RETURN_FAILED_WITH_ERRNO;
}
if (sctp_connectx(cfd, (struct sockaddr *) &addr, 1, &assoc_id) < 0) {
close(lfd);
close(cfd);
RETURN_FAILED_WITH_ERRNO;
}
sctp_delay(SCTP_SLEEP_MS);
if ((pfd = sctp_peeloff(cfd, assoc_id)) < 0) {
close(lfd);
close(cfd);
RETURN_FAILED_WITH_ERRNO;
}
if (close(pfd) < 0) {
close(cfd);
close(lfd);
RETURN_FAILED_WITH_ERRNO;
}
if (close(cfd) < 0) {
close(lfd);
RETURN_FAILED_WITH_ERRNO;
}
if (close(lfd) < 0) {
RETURN_FAILED_WITH_ERRNO;
}
RETURN_PASSED;
}
int
sctp_listen(sctp_t *sctp)
{
sctp_tf_t *tf;
RUN_SCTP(sctp);
/*
* TCP handles listen() increasing the backlog, need to check
* if it should be handled here too
*/
if (sctp->sctp_state > SCTPS_BOUND) {
WAKE_SCTP(sctp);
return (EINVAL);
}
/* Do an anonymous bind for unbound socket doing listen(). */
if (sctp->sctp_nsaddrs == 0) {
struct sockaddr_storage ss;
int ret;
bzero(&ss, sizeof (ss));
ss.ss_family = sctp->sctp_family;
WAKE_SCTP(sctp);
if ((ret = sctp_bind(sctp, (struct sockaddr *)&ss,
sizeof (ss))) != 0)
return (ret);
RUN_SCTP(sctp)
}
sctp->sctp_state = SCTPS_LISTEN;
(void) random_get_pseudo_bytes(sctp->sctp_secret, SCTP_SECRET_LEN);
sctp->sctp_last_secret_update = lbolt64;
bzero(sctp->sctp_old_secret, SCTP_SECRET_LEN);
tf = &sctp_listen_fanout[SCTP_LISTEN_HASH(ntohs(sctp->sctp_lport))];
sctp_listen_hash_insert(tf, sctp);
WAKE_SCTP(sctp);
return (0);
}
int
sctp_listen(sctp_t *sctp)
{
sctp_tf_t *tf;
sctp_stack_t *sctps = sctp->sctp_sctps;
conn_t *connp = sctp->sctp_connp;
RUN_SCTP(sctp);
/*
* TCP handles listen() increasing the backlog, need to check
* if it should be handled here too
*/
if (sctp->sctp_state > SCTPS_BOUND ||
(sctp->sctp_connp->conn_state_flags & CONN_CLOSING)) {
WAKE_SCTP(sctp);
return (EINVAL);
}
/* Do an anonymous bind for unbound socket doing listen(). */
if (sctp->sctp_nsaddrs == 0) {
struct sockaddr_storage ss;
int ret;
bzero(&ss, sizeof (ss));
ss.ss_family = connp->conn_family;
WAKE_SCTP(sctp);
if ((ret = sctp_bind(sctp, (struct sockaddr *)&ss,
sizeof (ss))) != 0)
return (ret);
RUN_SCTP(sctp)
}
/* Cache things in the ixa without any refhold */
ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
connp->conn_ixa->ixa_cred = connp->conn_cred;
connp->conn_ixa->ixa_cpid = connp->conn_cpid;
if (is_system_labeled())
connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
sctp->sctp_state = SCTPS_LISTEN;
(void) random_get_pseudo_bytes(sctp->sctp_secret, SCTP_SECRET_LEN);
sctp->sctp_last_secret_update = ddi_get_lbolt64();
bzero(sctp->sctp_old_secret, SCTP_SECRET_LEN);
/*
* If there is an association limit, allocate and initialize
* the counter struct. Note that since listen can be called
* multiple times, the struct may have been allready allocated.
*/
if (!list_is_empty(&sctps->sctps_listener_conf) &&
sctp->sctp_listen_cnt == NULL) {
sctp_listen_cnt_t *slc;
uint32_t ratio;
ratio = sctp_find_listener_conf(sctps,
ntohs(connp->conn_lport));
if (ratio != 0) {
uint32_t mem_ratio, tot_buf;
slc = kmem_alloc(sizeof (sctp_listen_cnt_t), KM_SLEEP);
/*
* Calculate the connection limit based on
* the configured ratio and maxusers. Maxusers
* are calculated based on memory size,
* ~ 1 user per MB. Note that the conn_rcvbuf
* and conn_sndbuf may change after a
* connection is accepted. So what we have
* is only an approximation.
*/
if ((tot_buf = connp->conn_rcvbuf +
connp->conn_sndbuf) < MB) {
mem_ratio = MB / tot_buf;
slc->slc_max = maxusers / ratio * mem_ratio;
} else {
mem_ratio = tot_buf / MB;
slc->slc_max = maxusers / ratio / mem_ratio;
}
/* At least we should allow some associations! */
if (slc->slc_max < sctp_min_assoc_listener)
slc->slc_max = sctp_min_assoc_listener;
slc->slc_cnt = 1;
slc->slc_drop = 0;
sctp->sctp_listen_cnt = slc;
}
}
tf = &sctps->sctps_listen_fanout[SCTP_LISTEN_HASH(
ntohs(connp->conn_lport))];
sctp_listen_hash_insert(tf, sctp);
WAKE_SCTP(sctp);
return (0);
}
/*
* Connect to a peer - this function inserts the sctp in the
* bind and conn fanouts, sends the INIT, and replies to the client
* with an OK ack.
*/
int
sctp_connect(sctp_t *sctp, const struct sockaddr *dst, uint32_t addrlen,
cred_t *cr, pid_t pid)
{
sin_t *sin;
sin6_t *sin6;
in6_addr_t dstaddr;
in_port_t dstport;
mblk_t *initmp;
sctp_tf_t *tbf;
sctp_t *lsctp;
char buf[INET6_ADDRSTRLEN];
int sleep = sctp->sctp_cansleep ? KM_SLEEP : KM_NOSLEEP;
int err;
sctp_faddr_t *cur_fp;
sctp_stack_t *sctps = sctp->sctp_sctps;
conn_t *connp = sctp->sctp_connp;
uint_t scope_id = 0;
ip_xmit_attr_t *ixa;
/*
* Determine packet type based on type of address passed in
* the request should contain an IPv4 or IPv6 address.
* Make sure that address family matches the type of
* family of the address passed down.
*/
if (addrlen < sizeof (sin_t)) {
return (EINVAL);
}
switch (dst->sa_family) {
case AF_INET:
sin = (sin_t *)dst;
/* Check for attempt to connect to non-unicast */
if (CLASSD(sin->sin_addr.s_addr) ||
(sin->sin_addr.s_addr == INADDR_BROADCAST)) {
ip0dbg(("sctp_connect: non-unicast\n"));
return (EINVAL);
}
if (connp->conn_ipv6_v6only)
return (EAFNOSUPPORT);
/* convert to v6 mapped */
/* Check for attempt to connect to INADDR_ANY */
if (sin->sin_addr.s_addr == INADDR_ANY) {
struct in_addr v4_addr;
/*
* SunOS 4.x and 4.3 BSD allow an application
* to connect a TCP socket to INADDR_ANY.
* When they do this, the kernel picks the
* address of one interface and uses it
* instead. The kernel usually ends up
* picking the address of the loopback
* interface. This is an undocumented feature.
* However, we provide the same thing here
* in case any TCP apps that use this feature
* are being ported to SCTP...
*/
v4_addr.s_addr = htonl(INADDR_LOOPBACK);
IN6_INADDR_TO_V4MAPPED(&v4_addr, &dstaddr);
} else {
IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &dstaddr);
}
dstport = sin->sin_port;
break;
case AF_INET6:
sin6 = (sin6_t *)dst;
/* Check for attempt to connect to non-unicast. */
if ((addrlen < sizeof (sin6_t)) ||
IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
ip0dbg(("sctp_connect: non-unicast\n"));
return (EINVAL);
}
if (connp->conn_ipv6_v6only &&
IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
return (EAFNOSUPPORT);
}
/* check for attempt to connect to unspec */
if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
dstaddr = ipv6_loopback;
} else {
dstaddr = sin6->sin6_addr;
if (IN6_IS_ADDR_LINKLOCAL(&dstaddr)) {
sctp->sctp_linklocal = 1;
scope_id = sin6->sin6_scope_id;
}
}
dstport = sin6->sin6_port;
connp->conn_flowinfo = sin6->sin6_flowinfo;
break;
default:
dprint(1, ("sctp_connect: unknown family %d\n",
dst->sa_family));
return (EAFNOSUPPORT);
}
(void) inet_ntop(AF_INET6, &dstaddr, buf, sizeof (buf));
dprint(1, ("sctp_connect: attempting connect to %s...\n", buf));
RUN_SCTP(sctp);
if (connp->conn_family != dst->sa_family ||
(connp->conn_state_flags & CONN_CLOSING)) {
WAKE_SCTP(sctp);
return (EINVAL);
}
/* We update our cred/cpid based on the caller of connect */
if (connp->conn_cred != cr) {
crhold(cr);
crfree(connp->conn_cred);
connp->conn_cred = cr;
}
connp->conn_cpid = pid;
/* Cache things in conn_ixa without any refhold */
ixa = connp->conn_ixa;
ixa->ixa_cred = cr;
ixa->ixa_cpid = pid;
if (is_system_labeled()) {
/* We need to restart with a label based on the cred */
ip_xmit_attr_restore_tsl(ixa, ixa->ixa_cred);
}
switch (sctp->sctp_state) {
case SCTPS_IDLE: {
struct sockaddr_storage ss;
/*
* We support a quick connect capability here, allowing
* clients to transition directly from IDLE to COOKIE_WAIT.
* sctp_bindi will pick an unused port, insert the connection
* in the bind hash and transition to BOUND state. SCTP
* picks and uses what it considers the optimal local address
* set (just like specifiying INADDR_ANY to bind()).
*/
dprint(1, ("sctp_connect: idle, attempting bind...\n"));
ASSERT(sctp->sctp_nsaddrs == 0);
bzero(&ss, sizeof (ss));
ss.ss_family = connp->conn_family;
WAKE_SCTP(sctp);
if ((err = sctp_bind(sctp, (struct sockaddr *)&ss,
sizeof (ss))) != 0) {
return (err);
}
RUN_SCTP(sctp);
/* FALLTHRU */
}
case SCTPS_BOUND:
ASSERT(sctp->sctp_nsaddrs > 0);
/* do the connect */
/* XXX check for attempt to connect to self */
connp->conn_fport = dstport;
ASSERT(sctp->sctp_iphc);
ASSERT(sctp->sctp_iphc6);
/*
* Don't allow this connection to completely duplicate
* an existing connection.
*
* Ensure that the duplicate check and insertion is atomic.
*/
sctp_conn_hash_remove(sctp);
tbf = &sctps->sctps_conn_fanout[SCTP_CONN_HASH(sctps,
connp->conn_ports)];
mutex_enter(&tbf->tf_lock);
lsctp = sctp_lookup(sctp, &dstaddr, tbf, &connp->conn_ports,
SCTPS_COOKIE_WAIT);
if (lsctp != NULL) {
/* found a duplicate connection */
mutex_exit(&tbf->tf_lock);
SCTP_REFRELE(lsctp);
WAKE_SCTP(sctp);
return (EADDRINUSE);
}
/*
* OK; set up the peer addr (this may grow after we get
* the INIT ACK from the peer with additional addresses).
*/
if ((err = sctp_add_faddr(sctp, &dstaddr, sleep,
B_FALSE)) != 0) {
mutex_exit(&tbf->tf_lock);
WAKE_SCTP(sctp);
return (err);
}
cur_fp = sctp->sctp_faddrs;
ASSERT(cur_fp->ixa != NULL);
/* No valid src addr, return. */
if (cur_fp->state == SCTP_FADDRS_UNREACH) {
mutex_exit(&tbf->tf_lock);
WAKE_SCTP(sctp);
return (EADDRNOTAVAIL);
}
sctp->sctp_primary = cur_fp;
sctp->sctp_current = cur_fp;
sctp->sctp_mss = cur_fp->sfa_pmss;
sctp_conn_hash_insert(tbf, sctp, 1);
mutex_exit(&tbf->tf_lock);
ixa = cur_fp->ixa;
ASSERT(ixa->ixa_cred != NULL);
if (scope_id != 0) {
ixa->ixa_flags |= IXAF_SCOPEID_SET;
ixa->ixa_scopeid = scope_id;
} else {
ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
}
/* initialize composite headers */
if ((err = sctp_set_hdraddrs(sctp)) != 0) {
sctp_conn_hash_remove(sctp);
WAKE_SCTP(sctp);
return (err);
}
if ((err = sctp_build_hdrs(sctp, KM_SLEEP)) != 0) {
sctp_conn_hash_remove(sctp);
WAKE_SCTP(sctp);
return (err);
}
/*
* Turn off the don't fragment bit on the (only) faddr,
* so that if one of the messages exchanged during the
* initialization sequence exceeds the path mtu, it
* at least has a chance to get there. SCTP does no
* fragmentation of initialization messages. The DF bit
* will be turned on again in sctp_send_cookie_echo()
* (but the cookie echo will still be sent with the df bit
* off).
*/
cur_fp->df = B_FALSE;
/* Mark this address as alive */
cur_fp->state = SCTP_FADDRS_ALIVE;
/* Send the INIT to the peer */
SCTP_FADDR_TIMER_RESTART(sctp, cur_fp, cur_fp->rto);
sctp->sctp_state = SCTPS_COOKIE_WAIT;
/*
* sctp_init_mp() could result in modifying the source
* address list, so take the hash lock.
*/
mutex_enter(&tbf->tf_lock);
initmp = sctp_init_mp(sctp, cur_fp);
if (initmp == NULL) {
mutex_exit(&tbf->tf_lock);
/*
* It may happen that all the source addresses
* (loopback/link local) are removed. In that case,
* faile the connect.
*/
if (sctp->sctp_nsaddrs == 0) {
sctp_conn_hash_remove(sctp);
SCTP_FADDR_TIMER_STOP(cur_fp);
WAKE_SCTP(sctp);
return (EADDRNOTAVAIL);
}
/* Otherwise, let the retransmission timer retry */
WAKE_SCTP(sctp);
goto notify_ulp;
}
mutex_exit(&tbf->tf_lock);
/*
* On a clustered note send this notification to the clustering
* subsystem.
*/
if (cl_sctp_connect != NULL) {
uchar_t *slist;
uchar_t *flist;
size_t ssize;
size_t fsize;
fsize = sizeof (in6_addr_t) * sctp->sctp_nfaddrs;
ssize = sizeof (in6_addr_t) * sctp->sctp_nsaddrs;
slist = kmem_alloc(ssize, KM_SLEEP);
flist = kmem_alloc(fsize, KM_SLEEP);
/* The clustering module frees the lists */
sctp_get_saddr_list(sctp, slist, ssize);
sctp_get_faddr_list(sctp, flist, fsize);
(*cl_sctp_connect)(connp->conn_family, slist,
sctp->sctp_nsaddrs, connp->conn_lport,
flist, sctp->sctp_nfaddrs, connp->conn_fport,
B_TRUE, (cl_sctp_handle_t)sctp);
}
ASSERT(ixa->ixa_cred != NULL);
ASSERT(ixa->ixa_ire != NULL);
(void) conn_ip_output(initmp, ixa);
BUMP_LOCAL(sctp->sctp_opkts);
WAKE_SCTP(sctp);
notify_ulp:
sctp_set_ulp_prop(sctp);
return (0);
default:
ip0dbg(("sctp_connect: invalid state. %d\n", sctp->sctp_state));
WAKE_SCTP(sctp);
return (EINVAL);
}
}
