/* This will need only if native IP used, or (unlikely) NCP will be
* implemented on the socket level
*/
static int
ncp_soconnect(struct socket *so,struct sockaddr *target, struct thread *td) {
int error;
error = soconnect(so, (struct sockaddr*)target, td);
if (error)
return error;
/*
* Wait for the connection to complete. Cribbed from the
* connect system call but with the wait timing out so
* that interruptible mounts don't hang here for a long time.
*/
error = EIO;
crit_enter();
while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
tsleep((caddr_t)&so->so_timeo, 0, "ncpcon", 2 * hz);
if ((so->so_state & SS_ISCONNECTING) &&
so->so_error == 0 /*&& rep &&*/) {
soclrstate(so, SS_ISCONNECTING);
crit_exit();
goto bad;
}
}
if (so->so_error) {
error = so->so_error;
so->so_error = 0;
crit_exit();
goto bad;
}
crit_exit();
error=0;
bad:
return error;
}
/*
* Destroy a disconnected socket. This routine is a NOP if entities
* still have a reference on the socket:
*
* so_pcb - The protocol stack still has a reference
* SS_NOFDREF - There is no longer a file pointer reference
*/
void
sofree(struct socket *so)
{
struct socket *head;
/*
* This is a bit hackish at the moment. We need to interlock
* any accept queue we are on before we potentially lose the
* last reference to avoid races against a re-reference from
* someone operating on the queue.
*/
while ((head = so->so_head) != NULL) {
lwkt_getpooltoken(head);
if (so->so_head == head)
break;
lwkt_relpooltoken(head);
}
/*
* Arbitrage the last free.
*/
KKASSERT(so->so_refs > 0);
if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
if (head)
lwkt_relpooltoken(head);
return;
}
KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
/*
* We're done, remove ourselves from the accept queue we are
* on, if we are on one.
*/
if (head != NULL) {
if (so->so_state & SS_INCOMP) {
TAILQ_REMOVE(&head->so_incomp, so, so_list);
head->so_incqlen--;
} else if (so->so_state & SS_COMP) {
/*
* We must not decommission a socket that's
* on the accept(2) queue. If we do, then
* accept(2) may hang after select(2) indicated
* that the listening socket was ready.
*/
lwkt_relpooltoken(head);
return;
} else {
panic("sofree: not queued");
}
soclrstate(so, SS_INCOMP);
so->so_head = NULL;
lwkt_relpooltoken(head);
}
ssb_release(&so->so_snd, so);
sorflush(so);
sodealloc(so);
}
/*
* so is passed in ref'd, which becomes owned by
* the cleared SS_NOFDREF flag.
*/
int
soaccept(struct socket *so, struct sockaddr **nam)
{
int error;
if ((so->so_state & SS_NOFDREF) == 0)
panic("soaccept: !NOFDREF");
soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
error = so_pru_accept_direct(so, nam);
return (error);
}
/*
* If only one hook, allow read(2) and write(2) to work.
*/
static int
ngs_connect(hook_p hook)
{
node_p node = NG_HOOK_NODE(hook);
struct ngsock *priv = NG_NODE_PRIVATE(node);
if ((priv->datasock) && (priv->datasock->ng_socket)) {
if (NG_NODE_NUMHOOKS(node) == 1)
sosetstate(priv->datasock->ng_socket, SS_ISCONNECTED);
else
soclrstate(priv->datasock->ng_socket, SS_ISCONNECTED);
}
return (0);
}
/*
* Hook disconnection
*
* For this type, removal of the last link destroys the node
* if the NOLINGER flag is set.
*/
static int
ngs_disconnect(hook_p hook)
{
node_p node = NG_HOOK_NODE(hook);
struct ngsock *const priv = NG_NODE_PRIVATE(node);
if ((priv->datasock) && (priv->datasock->ng_socket)) {
if (NG_NODE_NUMHOOKS(node) == 1)
sosetstate(priv->datasock->ng_socket, SS_ISCONNECTED);
else
soclrstate(priv->datasock->ng_socket, SS_ISCONNECTED);
}
if ((priv->flags & NGS_FLAG_NOLINGER) &&
(NG_NODE_NUMHOOKS(node) == 0) && (NG_NODE_IS_VALID(node)))
ng_rmnode_self(node);
return (0);
}
static void
udp6_disconnect(netmsg_t msg)
{
struct socket *so = msg->disconnect.base.nm_so;
struct inpcb *inp;
int error;
inp = so->so_pcb;
if (inp == NULL) {
error = EINVAL;
goto out;
}
if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
error = ENOTCONN;
} else {
in6_pcbdisconnect(inp);
soclrstate(so, SS_ISCONNECTED); /* XXX */
error = 0;
}
out:
lwkt_replymsg(&msg->disconnect.base.lmsg, error);
}
/*
* MPSAFE
*/
int
soo_ioctl(struct file *fp, u_long cmd, caddr_t data,
struct ucred *cred, struct sysmsg *msg)
{
struct socket *so;
int error;
so = (struct socket *)fp->f_data;
switch (cmd) {
case FIOASYNC:
if (*(int *)data) {
sosetstate(so, SS_ASYNC);
atomic_set_int(&so->so_rcv.ssb_flags, SSB_ASYNC);
atomic_set_int(&so->so_snd.ssb_flags, SSB_ASYNC);
} else {
soclrstate(so, SS_ASYNC);
atomic_clear_int(&so->so_rcv.ssb_flags, SSB_ASYNC);
atomic_clear_int(&so->so_snd.ssb_flags, SSB_ASYNC);
}
error = 0;
break;
case FIONREAD:
*(int *)data = so->so_rcv.ssb_cc;
error = 0;
break;
case FIOSETOWN:
error = fsetown(*(int *)data, &so->so_sigio);
break;
case FIOGETOWN:
*(int *)data = fgetown(&so->so_sigio);
error = 0;
break;
case SIOCSPGRP:
error = fsetown(-(*(int *)data), &so->so_sigio);
break;
case SIOCGPGRP:
*(int *)data = -fgetown(&so->so_sigio);
error = 0;
break;
case SIOCATMARK:
*(int *)data = (so->so_state&SS_RCVATMARK) != 0;
error = 0;
break;
default:
/*
* Interface/routing/protocol specific ioctls:
* interface and routing ioctls should have a
* different entry since a socket's unnecessary
*/
if (IOCGROUP(cmd) == 'i') {
error = ifioctl(so, cmd, data, cred);
} else if (IOCGROUP(cmd) == 'r') {
error = rtioctl(cmd, data, cred);
} else {
error = so_pru_control_direct(so, cmd, data, NULL);
}
break;
}
return (error);
}
/*
* Receive a control message
*/
static int
ng_ksocket_rcvmsg(node_p node, item_p item, hook_p lasthook)
{
struct thread *td = curthread; /* XXX broken */
const priv_p priv = NG_NODE_PRIVATE(node);
struct socket *const so = priv->so;
struct ng_mesg *resp = NULL;
int error = 0;
struct ng_mesg *msg;
ng_ID_t raddr;
NGI_GET_MSG(item, msg);
switch (msg->header.typecookie) {
case NGM_KSOCKET_COOKIE:
switch (msg->header.cmd) {
case NGM_KSOCKET_BIND:
{
struct sockaddr *const sa
= (struct sockaddr *)msg->data;
/* Sanity check */
if (msg->header.arglen < SADATA_OFFSET
|| msg->header.arglen < sa->sa_len)
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Bind */
error = sobind(so, sa, td);
break;
}
case NGM_KSOCKET_LISTEN:
{
/* Sanity check */
if (msg->header.arglen != sizeof(int32_t))
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Listen */
error = solisten(so, *((int32_t *)msg->data), td);
break;
}
case NGM_KSOCKET_ACCEPT:
{
/* Sanity check */
if (msg->header.arglen != 0)
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Make sure the socket is capable of accepting */
if (!(so->so_options & SO_ACCEPTCONN))
ERROUT(EINVAL);
if (priv->flags & KSF_ACCEPTING)
ERROUT(EALREADY);
error = ng_ksocket_check_accept(priv);
if (error != 0 && error != EWOULDBLOCK)
ERROUT(error);
/*
* If a connection is already complete, take it.
* Otherwise let the upcall function deal with
* the connection when it comes in.
*/
priv->response_token = msg->header.token;
raddr = priv->response_addr = NGI_RETADDR(item);
if (error == 0) {
ng_ksocket_finish_accept(priv);
} else
priv->flags |= KSF_ACCEPTING;
break;
}
case NGM_KSOCKET_CONNECT:
{
struct sockaddr *const sa
= (struct sockaddr *)msg->data;
/* Sanity check */
if (msg->header.arglen < SADATA_OFFSET
|| msg->header.arglen < sa->sa_len)
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Do connect */
if ((so->so_state & SS_ISCONNECTING) != 0)
ERROUT(EALREADY);
if ((error = soconnect(so, sa, td)) != 0) {
soclrstate(so, SS_ISCONNECTING);
ERROUT(error);
}
if ((so->so_state & SS_ISCONNECTING) != 0) {
/* We will notify the sender when we connect */
priv->response_token = msg->header.token;
raddr = priv->response_addr = NGI_RETADDR(item);
priv->flags |= KSF_CONNECTING;
ERROUT(EINPROGRESS);
}
break;
}
case NGM_KSOCKET_GETNAME:
case NGM_KSOCKET_GETPEERNAME:
{
int (*func)(struct socket *so, struct sockaddr **nam);
struct sockaddr *sa = NULL;
int len;
/* Sanity check */
if (msg->header.arglen != 0)
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Get function */
if (msg->header.cmd == NGM_KSOCKET_GETPEERNAME) {
if ((so->so_state
& (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0)
ERROUT(ENOTCONN);
func = so->so_proto->pr_usrreqs->pru_peeraddr;
} else
func = so->so_proto->pr_usrreqs->pru_sockaddr;
/* Get local or peer address */
if ((error = (*func)(so, &sa)) != 0)
goto bail;
len = (sa == NULL) ? 0 : sa->sa_len;
/* Send it back in a response */
NG_MKRESPONSE(resp, msg, len, M_WAITOK | M_NULLOK);
if (resp == NULL) {
error = ENOMEM;
goto bail;
}
bcopy(sa, resp->data, len);
bail:
/* Cleanup */
if (sa != NULL)
kfree(sa, M_SONAME);
break;
}
case NGM_KSOCKET_GETOPT:
{
struct ng_ksocket_sockopt *ksopt =
(struct ng_ksocket_sockopt *)msg->data;
struct sockopt sopt;
/* Sanity check */
if (msg->header.arglen != sizeof(*ksopt))
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Get response with room for option value */
NG_MKRESPONSE(resp, msg, sizeof(*ksopt)
+ NG_KSOCKET_MAX_OPTLEN, M_WAITOK | M_NULLOK);
if (resp == NULL)
ERROUT(ENOMEM);
/* Get socket option, and put value in the response */
sopt.sopt_dir = SOPT_GET;
sopt.sopt_level = ksopt->level;
sopt.sopt_name = ksopt->name;
sopt.sopt_td = NULL;
sopt.sopt_valsize = NG_KSOCKET_MAX_OPTLEN;
ksopt = (struct ng_ksocket_sockopt *)resp->data;
sopt.sopt_val = ksopt->value;
if ((error = sogetopt(so, &sopt)) != 0) {
NG_FREE_MSG(resp);
break;
}
/* Set actual value length */
resp->header.arglen = sizeof(*ksopt)
+ sopt.sopt_valsize;
break;
}
case NGM_KSOCKET_SETOPT:
{
struct ng_ksocket_sockopt *const ksopt =
(struct ng_ksocket_sockopt *)msg->data;
const int valsize = msg->header.arglen - sizeof(*ksopt);
struct sockopt sopt;
/* Sanity check */
if (valsize < 0)
ERROUT(EINVAL);
if (so == NULL)
ERROUT(ENXIO);
/* Set socket option */
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = ksopt->level;
sopt.sopt_name = ksopt->name;
sopt.sopt_val = ksopt->value;
sopt.sopt_valsize = valsize;
sopt.sopt_td = NULL;
error = sosetopt(so, &sopt);
break;
}
default:
error = EINVAL;
break;
}
break;
default:
error = EINVAL;
break;
}
done:
NG_RESPOND_MSG(error, node, item, resp);
NG_FREE_MSG(msg);
return (error);
}
/*
* When incoming data is appended to the socket, we get notified here.
* This is also called whenever a significant event occurs for the socket.
* Our original caller may have queued this even some time ago and
* we cannot trust that he even still exists. The node however is being
* held with a reference by the queueing code and guarantied to be valid.
*/
static void
ng_ksocket_incoming2(node_p node, hook_p hook, void *arg1, int arg2)
{
struct socket *so = arg1;
const priv_p priv = NG_NODE_PRIVATE(node);
struct mbuf *m;
struct ng_mesg *response;
struct uio auio;
int s, flags, error;
s = splnet();
/* so = priv->so; *//* XXX could have derived this like so */
KASSERT(so == priv->so, ("%s: wrong socket", __func__));
/* Allow next incoming event to be queued. */
atomic_store_rel_int(&priv->fn_sent, 0);
/* Check whether a pending connect operation has completed */
if (priv->flags & KSF_CONNECTING) {
if ((error = so->so_error) != 0) {
so->so_error = 0;
soclrstate(so, SS_ISCONNECTING);
}
if (!(so->so_state & SS_ISCONNECTING)) {
NG_MKMESSAGE(response, NGM_KSOCKET_COOKIE,
NGM_KSOCKET_CONNECT, sizeof(int32_t), M_WAITOK | M_NULLOK);
if (response != NULL) {
response->header.flags |= NGF_RESP;
response->header.token = priv->response_token;
*(int32_t *)response->data = error;
/*
* send an async "response" message
* to the node that set us up
* (if it still exists)
*/
NG_SEND_MSG_ID(error, node,
response, priv->response_addr, 0);
}
priv->flags &= ~KSF_CONNECTING;
}
}
/* Check whether a pending accept operation has completed */
if (priv->flags & KSF_ACCEPTING) {
error = ng_ksocket_check_accept(priv);
if (error != EWOULDBLOCK)
priv->flags &= ~KSF_ACCEPTING;
if (error == 0)
ng_ksocket_finish_accept(priv);
}
/*
* If we don't have a hook, we must handle data events later. When
* the hook gets created and is connected, this upcall function
* will be called again.
*/
if (priv->hook == NULL) {
splx(s);
return;
}
/* Read and forward available mbuf's */
auio.uio_td = NULL;
auio.uio_resid = 1000000000;
flags = MSG_DONTWAIT;
while (1) {
struct sockaddr *sa = NULL;
struct mbuf *n;
/* Try to get next packet from socket */
if ((error = soreceive(so, (so->so_state & SS_ISCONNECTED) ?
NULL : &sa, &auio, &m, NULL, &flags)) != 0)
break;
/* See if we got anything */
if (m == NULL) {
if (sa != NULL)
kfree(sa, M_SONAME);
break;
}
/*
* Don't trust the various socket layers to get the
* packet header and length correct (e.g. kern/15175).
*
* Also, do not trust that soreceive() will clear m_nextpkt
* for us (e.g. kern/84952, kern/82413).
*/
m->m_pkthdr.csum_flags = 0;
for (n = m, m->m_pkthdr.len = 0; n != NULL; n = n->m_next) {
m->m_pkthdr.len += n->m_len;
n->m_nextpkt = NULL;
}
/* Put peer's socket address (if any) into a tag */
if (sa != NULL) {
struct sa_tag *stag;
stag = (struct sa_tag *)m_tag_alloc(NGM_KSOCKET_COOKIE,
NG_KSOCKET_TAG_SOCKADDR, sizeof(ng_ID_t) +
sa->sa_len, MB_DONTWAIT);
if (stag == NULL) {
kfree(sa, M_SONAME);
goto sendit;
}
bcopy(sa, &stag->sa, sa->sa_len);
kfree(sa, M_SONAME);
stag->id = NG_NODE_ID(node);
m_tag_prepend(m, &stag->tag);
}
sendit: /* Forward data with optional peer sockaddr as packet tag */
NG_SEND_DATA_ONLY(error, priv->hook, m);
}
/*
* If the peer has closed the connection, forward a 0-length mbuf
* to indicate end-of-file.
*/
if (so->so_rcv.sb_state & SBS_CANTRCVMORE && !(priv->flags & KSF_EOFSEEN)) {
MGETHDR(m, MB_DONTWAIT, MT_DATA);
if (m != NULL) {
m->m_len = m->m_pkthdr.len = 0;
NG_SEND_DATA_ONLY(error, priv->hook, m);
}
priv->flags |= KSF_EOFSEEN;
}
splx(s);
}
/*
* Close a socket on last file table reference removal.
* Initiate disconnect if connected.
* Free socket when disconnect complete.
*/
int
soclose(struct socket *so, int fflag)
{
int error = 0;
funsetown(&so->so_sigio);
if (so->so_pcb == NULL)
goto discard;
if (so->so_state & SS_ISCONNECTED) {
if ((so->so_state & SS_ISDISCONNECTING) == 0) {
error = sodisconnect(so);
if (error)
goto drop;
}
if (so->so_options & SO_LINGER) {
if ((so->so_state & SS_ISDISCONNECTING) &&
(fflag & FNONBLOCK))
goto drop;
while (so->so_state & SS_ISCONNECTED) {
error = tsleep(&so->so_timeo, PCATCH,
"soclos", so->so_linger * hz);
if (error)
break;
}
}
}
drop:
if (so->so_pcb) {
int error2;
error2 = so_pru_detach(so);
if (error == 0)
error = error2;
}
discard:
lwkt_getpooltoken(so);
if (so->so_options & SO_ACCEPTCONN) {
struct socket *sp;
while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
TAILQ_REMOVE(&so->so_incomp, sp, so_list);
soclrstate(sp, SS_INCOMP);
sp->so_head = NULL;
so->so_incqlen--;
soaborta(sp);
}
while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
TAILQ_REMOVE(&so->so_comp, sp, so_list);
soclrstate(sp, SS_COMP);
sp->so_head = NULL;
so->so_qlen--;
soaborta(sp);
}
}
lwkt_relpooltoken(so);
if (so->so_state & SS_NOFDREF)
panic("soclose: NOFDREF");
sosetstate(so, SS_NOFDREF); /* take ref */
sofree(so); /* dispose of ref */
return (error);
}
int
socreate(int dom, struct socket **aso, int type,
int proto, struct thread *td)
{
struct proc *p = td->td_proc;
struct protosw *prp;
struct socket *so;
struct pru_attach_info ai;
int error;
if (proto)
prp = pffindproto(dom, proto, type);
else
prp = pffindtype(dom, type);
if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
return (EPROTONOSUPPORT);
if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
prp->pr_domain->dom_family != PF_LOCAL &&
prp->pr_domain->dom_family != PF_INET &&
prp->pr_domain->dom_family != PF_INET6 &&
prp->pr_domain->dom_family != PF_ROUTE) {
return (EPROTONOSUPPORT);
}
if (prp->pr_type != type)
return (EPROTOTYPE);
so = soalloc(p != 0);
if (so == NULL)
return (ENOBUFS);
/*
* Callers of socreate() presumably will connect up a descriptor
* and call soclose() if they cannot. This represents our so_refs
* (which should be 1) from soalloc().
*/
soclrstate(so, SS_NOFDREF);
/*
* Set a default port for protocol processing. No action will occur
* on the socket on this port until an inpcb is attached to it and
* is able to match incoming packets, or until the socket becomes
* available to userland.
*
* We normally default the socket to the protocol thread on cpu 0.
* If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
* thread and all pr_*()/pru_*() calls are executed synchronously.
*/
if (prp->pr_flags & PR_SYNC_PORT)
so->so_port = &netisr_sync_port;
else
so->so_port = cpu_portfn(0);
TAILQ_INIT(&so->so_incomp);
TAILQ_INIT(&so->so_comp);
so->so_type = type;
so->so_cred = crhold(p->p_ucred);
so->so_proto = prp;
ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
ai.p_ucred = p->p_ucred;
ai.fd_rdir = p->p_fd->fd_rdir;
/*
* Auto-sizing of socket buffers is managed by the protocols and
* the appropriate flags must be set in the pru_attach function.
*/
error = so_pru_attach(so, proto, &ai);
if (error) {
sosetstate(so, SS_NOFDREF);
sofree(so); /* from soalloc */
return error;
}
/*
* NOTE: Returns referenced socket.
*/
*aso = so;
return (0);
}
/*
* Handle the first completed incoming connection, assumed to be already
* on the socket's so_comp queue.
*/
static void
ng_ksocket_finish_accept(priv_p priv)
{
struct socket *const head = priv->so;
struct socket *so;
struct sockaddr *sa = NULL;
struct ng_mesg *resp;
struct ng_ksocket_accept *resp_data;
node_p node;
priv_p priv2;
int len;
int error;
lwkt_getpooltoken(head);
so = TAILQ_FIRST(&head->so_comp);
if (so == NULL) { /* Should never happen */
lwkt_relpooltoken(head);
return;
}
TAILQ_REMOVE(&head->so_comp, so, so_list);
head->so_qlen--;
soclrstate(so, SS_COMP);
so->so_head = NULL;
soreference(so);
lwkt_relpooltoken(head);
/* XXX KNOTE(&head->so_rcv.ssb_sel.si_note, 0); */
soaccept(so, &sa);
len = OFFSETOF(struct ng_ksocket_accept, addr);
if (sa != NULL)
len += sa->sa_len;
NG_MKMESSAGE(resp, NGM_KSOCKET_COOKIE, NGM_KSOCKET_ACCEPT, len,
M_WAITOK | M_NULLOK);
if (resp == NULL) {
soclose(so, FNONBLOCK);
goto out;
}
resp->header.flags |= NGF_RESP;
resp->header.token = priv->response_token;
/* Clone a ksocket node to wrap the new socket */
error = ng_make_node_common(&ng_ksocket_typestruct, &node);
if (error) {
kfree(resp, M_NETGRAPH);
soclose(so, FNONBLOCK);
goto out;
}
if (ng_ksocket_constructor(node) != 0) {
NG_NODE_UNREF(node);
kfree(resp, M_NETGRAPH);
soclose(so, FNONBLOCK);
goto out;
}
priv2 = NG_NODE_PRIVATE(node);
priv2->so = so;
priv2->flags |= KSF_CLONED | KSF_EMBRYONIC;
/*
* Insert the cloned node into a list of embryonic children
* on the parent node. When a hook is created on the cloned
* node it will be removed from this list. When the parent
* is destroyed it will destroy any embryonic children it has.
*/
LIST_INSERT_HEAD(&priv->embryos, priv2, siblings);
so->so_upcallarg = (caddr_t)node;
so->so_upcall = ng_ksocket_incoming;
atomic_set_int(&priv->so->so_rcv.ssb_flags, SSB_UPCALL);
atomic_set_int(&priv->so->so_snd.ssb_flags, SSB_UPCALL);
/* Fill in the response data and send it or return it to the caller */
resp_data = (struct ng_ksocket_accept *)resp->data;
resp_data->nodeid = NG_NODE_ID(node);
if (sa != NULL)
bcopy(sa, &resp_data->addr, sa->sa_len);
NG_SEND_MSG_ID(error, node, resp, priv->response_addr, 0);
out:
if (sa != NULL)
kfree(sa, M_SONAME);
}