zmq::socket_base_t *zmq::socket_base_t::create (int type_, class ctx_t *parent_,
uint32_t tid_, int sid_)
{
socket_base_t *s = NULL;
switch (type_) {
case ZMQ_PAIR:
s = new (std::nothrow) pair_t (parent_, tid_, sid_);
break;
case ZMQ_PUB:
s = new (std::nothrow) pub_t (parent_, tid_, sid_);
break;
case ZMQ_SUB:
s = new (std::nothrow) sub_t (parent_, tid_, sid_);
break;
case ZMQ_REQ:
s = new (std::nothrow) req_t (parent_, tid_, sid_);
break;
case ZMQ_REP:
s = new (std::nothrow) rep_t (parent_, tid_, sid_);
break;
case ZMQ_DEALER:
s = new (std::nothrow) dealer_t (parent_, tid_, sid_);
break;
case ZMQ_ROUTER:
s = new (std::nothrow) router_t (parent_, tid_, sid_);
break;
case ZMQ_PULL:
s = new (std::nothrow) pull_t (parent_, tid_, sid_);
break;
case ZMQ_PUSH:
s = new (std::nothrow) push_t (parent_, tid_, sid_);
break;
case ZMQ_XPUB:
s = new (std::nothrow) xpub_t (parent_, tid_, sid_);
break;
case ZMQ_XSUB:
s = new (std::nothrow) xsub_t (parent_, tid_, sid_);
break;
case ZMQ_STREAM:
s = new (std::nothrow) stream_t (parent_, tid_, sid_);
break;
case ZMQ_SERVER:
s = new (std::nothrow) server_t (parent_, tid_, sid_);
break;
case ZMQ_CLIENT:
s = new (std::nothrow) client_t (parent_, tid_, sid_);
break;
case ZMQ_RADIO:
s = new (std::nothrow) radio_t (parent_, tid_, sid_);
break;
case ZMQ_DISH:
s = new (std::nothrow) dish_t (parent_, tid_, sid_);
break;
case ZMQ_GATHER:
s = new (std::nothrow) gather_t (parent_, tid_, sid_);
break;
case ZMQ_SCATTER:
s = new (std::nothrow) scatter_t (parent_, tid_, sid_);
break;
case ZMQ_DGRAM:
s = new (std::nothrow) dgram_t (parent_, tid_, sid_);
break;
default:
errno = EINVAL;
return NULL;
}
alloc_assert (s);
if (s->mailbox == NULL) {
s->destroyed = true;
LIBZMQ_DELETE(s);
return NULL;
}
return s;
}
void zmq::session_base_t::start_connecting (bool wait_)
{
zmq_assert (connect);
// Choose I/O thread to run connecter in. Given that we are already
// running in an I/O thread, there must be at least one available.
io_thread_t *io_thread = choose_io_thread (options.affinity);
zmq_assert (io_thread);
// Create the connecter object.
if (addr->protocol == "tcp") {
tcp_connecter_t *connecter = new (std::nothrow) tcp_connecter_t (
io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
return;
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
if (addr->protocol == "ipc") {
ipc_connecter_t *connecter = new (std::nothrow) ipc_connecter_t (
io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
return;
}
#endif
#ifdef ZMQ_HAVE_OPENPGM
// Both PGM and EPGM transports are using the same infrastructure.
if (addr->protocol == "pgm" || addr->protocol == "epgm") {
zmq_assert (options.type == ZMQ_PUB || options.type == ZMQ_XPUB
|| options.type == ZMQ_SUB || options.type == ZMQ_XSUB);
// For EPGM transport with UDP encapsulation of PGM is used.
bool const udp_encapsulation = addr->protocol == "epgm";
// At this point we'll create message pipes to the session straight
// away. There's no point in delaying it as no concept of 'connect'
// exists with PGM anyway.
if (options.type == ZMQ_PUB || options.type == ZMQ_XPUB) {
// PGM sender.
pgm_sender_t *pgm_sender = new (std::nothrow) pgm_sender_t (
io_thread, options);
alloc_assert (pgm_sender);
int rc = pgm_sender->init (udp_encapsulation, addr->address.c_str ());
errno_assert (rc == 0);
send_attach (this, pgm_sender);
}
else {
// PGM receiver.
pgm_receiver_t *pgm_receiver = new (std::nothrow) pgm_receiver_t (
io_thread, options);
alloc_assert (pgm_receiver);
int rc = pgm_receiver->init (udp_encapsulation, addr->address.c_str ());
errno_assert (rc == 0);
send_attach (this, pgm_receiver);
}
return;
}
#endif
zmq_assert (false);
}
static int grid_usock_recv_raw (struct grid_usock *self, void *buf, size_t *len)
{
size_t sz;
size_t length;
ssize_t nbytes;
struct iovec iov;
struct msghdr hdr;
unsigned char ctrl [256];
#if defined GRID_HAVE_MSG_CONTROL
struct cmsghdr *cmsg;
#endif
/* If batch buffer doesn't exist, allocate it. The point of delayed
deallocation to allow non-receiving sockets, such as TCP listening
sockets, to do without the batch buffer. */
if (grid_slow (!self->in.batch)) {
self->in.batch = grid_alloc (GRID_USOCK_BATCH_SIZE, "AIO batch buffer");
alloc_assert (self->in.batch);
}
/* Try to satisfy the recv request by data from the batch buffer. */
length = *len;
sz = self->in.batch_len - self->in.batch_pos;
if (sz) {
if (sz > length)
sz = length;
memcpy (buf, self->in.batch + self->in.batch_pos, sz);
self->in.batch_pos += sz;
buf = ((char*) buf) + sz;
length -= sz;
if (!length)
return 0;
}
/* If recv request is greater than the batch buffer, get the data directly
into the place. Otherwise, read data to the batch buffer. */
if (length > GRID_USOCK_BATCH_SIZE) {
iov.iov_base = buf;
iov.iov_len = length;
}
else {
iov.iov_base = self->in.batch;
iov.iov_len = GRID_USOCK_BATCH_SIZE;
}
memset (&hdr, 0, sizeof (hdr));
hdr.msg_iov = &iov;
hdr.msg_iovlen = 1;
#if defined GRID_HAVE_MSG_CONTROL
hdr.msg_control = ctrl;
hdr.msg_controllen = sizeof (ctrl);
#else
*((int*) ctrl) = -1;
hdr.msg_accrights = ctrl;
hdr.msg_accrightslen = sizeof (int);
#endif
nbytes = recvmsg (self->s, &hdr, 0);
/* Handle any possible errors. */
if (grid_slow (nbytes <= 0)) {
if (grid_slow (nbytes == 0))
return -ECONNRESET;
/* Zero bytes received. */
if (grid_fast (errno == EAGAIN || errno == EWOULDBLOCK))
nbytes = 0;
else {
/* If the peer closes the connection, return ECONNRESET. */
return -ECONNRESET;
}
}
/* Extract the associated file descriptor, if any. */
if (nbytes > 0) {
#if defined GRID_HAVE_MSG_CONTROL
cmsg = CMSG_FIRSTHDR (&hdr);
while (cmsg) {
if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
if (self->in.pfd) {
*self->in.pfd = *((int*) CMSG_DATA (cmsg));
self->in.pfd = NULL;
}
else {
grid_closefd (*((int*) CMSG_DATA (cmsg)));
}
break;
}
cmsg = CMSG_NXTHDR (&hdr, cmsg);
}
#else
if (hdr.msg_accrightslen > 0) {
grid_assert (hdr.msg_accrightslen == sizeof (int));
if (self->in.pfd) {
*self->in.pfd = *((int*) hdr.msg_accrights);
self->in.pfd = NULL;
}
else {
grid_closefd (*((int*) hdr.msg_accrights));
}
}
#endif
}
/* If the data were received directly into the place we can return
straight away. */
if (length > GRID_USOCK_BATCH_SIZE) {
length -= nbytes;
*len -= length;
return 0;
}
/* New data were read to the batch buffer. Copy the requested amount of it
to the user-supplied buffer. */
self->in.batch_len = nbytes;
self->in.batch_pos = 0;
if (nbytes) {
sz = nbytes > (ssize_t)length ? length : (size_t)nbytes;
memcpy (buf, self->in.batch, sz);
length -= sz;
self->in.batch_pos += sz;
}
*len -= length;
return 0;
}
void zmq::mtrie_t::rm_helper (pipe_t *pipe_, unsigned char **buff_,
size_t buffsize_, size_t maxbuffsize_,
void (*func_) (unsigned char *data_, size_t size_, void *arg_),
void *arg_, bool call_on_uniq_)
{
// Remove the subscription from this node.
if (pipes && pipes->erase (pipe_)) {
if (!call_on_uniq_ || pipes->empty ()) {
func_ (*buff_, buffsize_, arg_);
}
if (pipes->empty ()) {
LIBZMQ_DELETE(pipes);
}
}
// Adjust the buffer.
if (buffsize_ >= maxbuffsize_) {
maxbuffsize_ = buffsize_ + 256;
*buff_ = (unsigned char*) realloc (*buff_, maxbuffsize_);
alloc_assert (*buff_);
}
// If there are no subnodes in the trie, return.
if (count == 0)
return;
// If there's one subnode (optimisation).
if (count == 1) {
(*buff_) [buffsize_] = min;
buffsize_++;
next.node->rm_helper (pipe_, buff_, buffsize_, maxbuffsize_,
func_, arg_, call_on_uniq_);
// Prune the node if it was made redundant by the removal
if (next.node->is_redundant ()) {
LIBZMQ_DELETE(next.node);
count = 0;
--live_nodes;
zmq_assert (live_nodes == 0);
}
return;
}
// If there are multiple subnodes.
//
// New min non-null character in the node table after the removal
unsigned char new_min = min + count - 1;
// New max non-null character in the node table after the removal
unsigned char new_max = min;
for (unsigned short c = 0; c != count; c++) {
(*buff_) [buffsize_] = min + c;
if (next.table [c]) {
next.table [c]->rm_helper (pipe_, buff_, buffsize_ + 1,
maxbuffsize_, func_, arg_, call_on_uniq_);
// Prune redundant nodes from the mtrie
if (next.table [c]->is_redundant ()) {
LIBZMQ_DELETE(next.table[c]);
zmq_assert (live_nodes > 0);
--live_nodes;
}
else {
// The node is not redundant, so it's a candidate for being
// the new min/max node.
//
// We loop through the node array from left to right, so the
// first non-null, non-redundant node encountered is the new
// minimum index. Conversely, the last non-redundant, non-null
// node encountered is the new maximum index.
if (c + min < new_min)
new_min = c + min;
if (c + min > new_max)
new_max = c + min;
}
}
}
zmq_assert (count > 1);
// Free the node table if it's no longer used.
if (live_nodes == 0) {
free (next.table);
next.table = NULL;
count = 0;
}
// Compact the node table if possible
else
if (live_nodes == 1) {
// If there's only one live node in the table we can
// switch to using the more compact single-node
// representation
zmq_assert (new_min == new_max);
zmq_assert (new_min >= min && new_min < min + count);
mtrie_t *node = next.table [new_min - min];
zmq_assert (node);
free (next.table);
next.node = node;
count = 1;
min = new_min;
}
else
if (new_min > min || new_max < min + count - 1) {
zmq_assert (new_max - new_min + 1 > 1);
mtrie_t **old_table = next.table;
zmq_assert (new_min > min || new_max < min + count - 1);
zmq_assert (new_min >= min);
zmq_assert (new_max <= min + count - 1);
zmq_assert (new_max - new_min + 1 < count);
count = new_max - new_min + 1;
next.table = (mtrie_t**) malloc (sizeof (mtrie_t*) * count);
alloc_assert (next.table);
memmove (next.table, old_table + (new_min - min),
sizeof (mtrie_t*) * count);
free (old_table);
min = new_min;
}
}
bool zmq::mtrie_t::add_helper (unsigned char *prefix_, size_t size_,
pipe_t *pipe_)
{
// We are at the node corresponding to the prefix. We are done.
if (!size_) {
bool result = !pipes;
if (!pipes) {
pipes = new (std::nothrow) pipes_t;
alloc_assert (pipes);
}
pipes->insert (pipe_);
return result;
}
unsigned char c = *prefix_;
if (c < min || c >= min + count) {
// The character is out of range of currently handled
// charcters. We have to extend the table.
if (!count) {
min = c;
count = 1;
next.node = NULL;
}
else
if (count == 1) {
unsigned char oldc = min;
mtrie_t *oldp = next.node;
count = (min < c ? c - min : min - c) + 1;
next.table = (mtrie_t**)
malloc (sizeof (mtrie_t*) * count);
alloc_assert (next.table);
for (unsigned short i = 0; i != count; ++i)
next.table [i] = 0;
min = std::min (min, c);
next.table [oldc - min] = oldp;
}
else
if (min < c) {
// The new character is above the current character range.
unsigned short old_count = count;
count = c - min + 1;
next.table = (mtrie_t**) realloc (next.table,
sizeof (mtrie_t*) * count);
alloc_assert (next.table);
for (unsigned short i = old_count; i != count; i++)
next.table [i] = NULL;
}
else {
// The new character is below the current character range.
unsigned short old_count = count;
count = (min + old_count) - c;
next.table = (mtrie_t**) realloc (next.table,
sizeof (mtrie_t*) * count);
alloc_assert (next.table);
memmove (next.table + min - c, next.table,
old_count * sizeof (mtrie_t*));
for (unsigned short i = 0; i != min - c; i++)
next.table [i] = NULL;
min = c;
}
}
// If next node does not exist, create one.
if (count == 1) {
if (!next.node) {
next.node = new (std::nothrow) mtrie_t;
alloc_assert (next.node);
++live_nodes;
}
return next.node->add_helper (prefix_ + 1, size_ - 1, pipe_);
}
else {
if (!next.table [c - min]) {
next.table [c - min] = new (std::nothrow) mtrie_t;
alloc_assert (next.table [c - min]);
++live_nodes;
}
return next.table [c - min]->add_helper (prefix_ + 1, size_ - 1, pipe_);
}
}
static bool pfx_rm (pfx_node_t *node_, const unsigned char *prefix_,
size_t size_, void *subscriber_)
{
if (!size_) {
// Remove the subscription from this node.
if (node_->subscribers) {
pfx_node_t::subscribers_t::iterator it =
node_->subscribers->find (subscriber_);
if (it != node_->subscribers->end ()) {
xs_assert (it->second);
--it->second;
if (!it->second) {
node_->subscribers->erase (it);
if (node_->subscribers->empty ()) {
delete node_->subscribers;
node_->subscribers = 0;
}
}
}
}
return !node_->subscribers;
}
unsigned char c = *prefix_;
if (!node_->count || c < node_->min || c >= node_->min + node_->count)
return false;
pfx_node_t *next_node = node_->count == 1 ? node_->next.node :
node_->next.table [c - node_->min];
if (!next_node)
return false;
bool ret = pfx_rm (next_node, prefix_ + 1, size_ - 1, subscriber_);
if (pfx_is_redundant (next_node)) {
pfx_close (next_node);
free (next_node);
xs_assert (node_->count > 0);
if (node_->count == 1) {
node_->next.node = 0;
node_->count = 0;
--node_->live_nodes;
xs_assert (node_->live_nodes == 0);
}
else {
node_->next.table [c - node_->min] = 0;
xs_assert (node_->live_nodes > 1);
--node_->live_nodes;
// Compact the table if possible.
if (node_->live_nodes == 0) {
// Free the node table if it's no longer used.
free (node_->next.table);
node_->next.table = NULL;
node_->count = 0;
}
else if (node_->live_nodes == 1) {
// If there's only one live node in the table we can
// switch to using the more compact single-node
// representation
pfx_node_t *node = 0;
for (unsigned short i = 0; i < node_->count; ++i) {
if (node_->next.table [i]) {
node = node_->next.table [i];
node_->min += i;
break;
}
}
xs_assert (node);
free (node_->next.table);
node_->next.node = node;
node_->count = 1;
}
else if (c == node_->min) {
// We can compact the table "from the left".
unsigned char new_min = node_->min;
for (unsigned short i = 1; i < node_->count; ++i) {
if (node_->next.table [i]) {
new_min = i + node_->min;
break;
}
}
xs_assert (new_min != node_->min);
pfx_node_t **old_table = node_->next.table;
xs_assert (new_min > node_->min);
xs_assert (node_->count > new_min - node_->min);
node_->count = node_->count - (new_min - node_->min);
node_->next.table = (pfx_node_t**)
malloc (sizeof (pfx_node_t*) * node_->count);
alloc_assert (node_->next.table);
memmove (node_->next.table, old_table + (new_min - node_->min),
sizeof (pfx_node_t*) * node_->count);
free (old_table);
node_->min = new_min;
}
else if (c == node_->min + node_->count - 1) {
// We can compact the table "from the right".
unsigned short new_count = node_->count;
for (unsigned short i = 1; i < node_->count; ++i) {
if (node_->next.table [node_->count - 1 - i]) {
new_count = node_->count - i;
break;
}
}
xs_assert (new_count != node_->count);
node_->count = new_count;
pfx_node_t **old_table = node_->next.table;
node_->next.table = (pfx_node_t**)
malloc (sizeof (pfx_node_t*) * node_->count);
alloc_assert (node_->next.table);
memmove (node_->next.table, old_table,
sizeof (pfx_node_t*) * node_->count);
free (old_table);
}
}
}
return ret;
}
bool zmq::trie_t::rm (unsigned char *prefix_, size_t size_)
{
// TODO: Shouldn't an error be reported if the key does not exist?
if (!size_) {
if (!refcnt)
return false;
refcnt--;
return refcnt == 0;
}
unsigned char c = *prefix_;
if (!count || c < min || c >= min + count)
return false;
trie_t *next_node =
count == 1 ? next.node : next.table [c - min];
if (!next_node)
return false;
bool ret = next_node->rm (prefix_ + 1, size_ - 1);
// Prune redundant nodes
if (next_node->is_redundant ()) {
LIBZMQ_DELETE(next_node);
zmq_assert (count > 0);
if (count == 1) {
// The just pruned node is was the only live node
next.node = 0;
count = 0;
--live_nodes;
zmq_assert (live_nodes == 0);
}
else {
next.table [c - min] = 0;
zmq_assert (live_nodes > 1);
--live_nodes;
// Compact the table if possible
if (live_nodes == 1) {
// We can switch to using the more compact single-node
// representation since the table only contains one live node
trie_t *node = 0;
// Since we always compact the table the pruned node must
// either be the left-most or right-most ptr in the node
// table
if (c == min) {
// The pruned node is the left-most node ptr in the
// node table => keep the right-most node
node = next.table [count - 1];
min += count - 1;
}
else
if (c == min + count - 1) {
// The pruned node is the right-most node ptr in the
// node table => keep the left-most node
node = next.table [0];
}
zmq_assert (node);
free (next.table);
next.node = node;
count = 1;
}
else
if (c == min) {
// We can compact the table "from the left".
// Find the left-most non-null node ptr, which we'll use as
// our new min
unsigned char new_min = min;
for (unsigned short i = 1; i < count; ++i) {
if (next.table [i]) {
new_min = i + min;
break;
}
}
zmq_assert (new_min != min);
trie_t **old_table = next.table;
zmq_assert (new_min > min);
zmq_assert (count > new_min - min);
count = count - (new_min - min);
next.table = (trie_t**) malloc (sizeof (trie_t*) * count);
alloc_assert (next.table);
memmove (next.table, old_table + (new_min - min),
sizeof (trie_t*) * count);
free (old_table);
min = new_min;
}
else
if (c == min + count - 1) {
// We can compact the table "from the right".
// Find the right-most non-null node ptr, which we'll use to
// determine the new table size
unsigned short new_count = count;
for (unsigned short i = 1; i < count; ++i) {
if (next.table [count - 1 - i]) {
new_count = count - i;
break;
}
}
zmq_assert (new_count != count);
count = new_count;
trie_t **old_table = next.table;
next.table = (trie_t**) malloc (sizeof (trie_t*) * count);
alloc_assert (next.table);
memmove (next.table, old_table, sizeof (trie_t*) * count);
free (old_table);
}
}
}
return ret;
}
void zmq::pgm_receiver_t::in_event ()
{
// Read data from the underlying pgm_socket.
const pgm_tsi_t *tsi = NULL;
if (has_rx_timer) {
cancel_timer (rx_timer_id);
has_rx_timer = false;
}
// TODO: This loop can effectively block other engines in the same I/O
// thread in the case of high load.
while (true) {
// Get new batch of data.
// Note the workaround made not to break strict-aliasing rules.
void *tmp = NULL;
ssize_t received = pgm_socket.receive (&tmp, &tsi);
inpos = (unsigned char*) tmp;
// No data to process. This may happen if the packet received is
// neither ODATA nor ODATA.
if (received == 0) {
if (errno == ENOMEM || errno == EBUSY) {
const long timeout = pgm_socket.get_rx_timeout ();
add_timer (timeout, rx_timer_id);
has_rx_timer = true;
}
break;
}
// Find the peer based on its TSI.
peers_t::iterator it = peers.find (*tsi);
// Data loss. Delete decoder and mark the peer as disjoint.
if (received == -1) {
if (it != peers.end ()) {
it->second.joined = false;
if (it->second.decoder != NULL) {
LIBZMQ_DELETE(it->second.decoder);
}
}
break;
}
// New peer. Add it to the list of know but unjoint peers.
if (it == peers.end ()) {
peer_info_t peer_info = {false, NULL};
it = peers.insert (peers_t::value_type (*tsi, peer_info)).first;
}
insize = static_cast <size_t> (received);
// Read the offset of the fist message in the current packet.
zmq_assert (insize >= sizeof (uint16_t));
uint16_t offset = get_uint16 (inpos);
inpos += sizeof (uint16_t);
insize -= sizeof (uint16_t);
// Join the stream if needed.
if (!it->second.joined) {
// There is no beginning of the message in current packet.
// Ignore the data.
if (offset == 0xffff)
continue;
zmq_assert (offset <= insize);
zmq_assert (it->second.decoder == NULL);
// We have to move data to the beginning of the first message.
inpos += offset;
insize -= offset;
// Mark the stream as joined.
it->second.joined = true;
// Create and connect decoder for the peer.
it->second.decoder = new (std::nothrow)
v1_decoder_t (0, options.maxmsgsize);
alloc_assert (it->second.decoder);
}
int rc = process_input (it->second.decoder);
if (rc == -1) {
if (errno == EAGAIN) {
active_tsi = tsi;
// Stop polling.
reset_pollin (pipe_handle);
reset_pollin (socket_handle);
break;
}
it->second.joined = false;
LIBZMQ_DELETE(it->second.decoder);
insize = 0;
}
}
// Flush any messages decoder may have produced.
session->flush ();
}
void *zmq_poller_new (void)
{
zmq::socket_poller_t *poller = new (std::nothrow) zmq::socket_poller_t;
alloc_assert (poller);
return poller;
}
void zmq::socket_base_t::copy_monitor_address (char *dest_, std::string &src_)
{
alloc_assert (dest_);
dest_[src_.size ()] = 0;
memcpy (dest_, src_.c_str (), src_.size ());
}
int zmq::socket_base_t::connect (const char *addr_)
{
if (unlikely (ctx_terminated)) {
errno = ETERM;
return -1;
}
// Process pending commands, if any.
int rc = process_commands (0, false);
if (unlikely (rc != 0))
return -1;
// Parse addr_ string.
std::string protocol;
std::string address;
rc = parse_uri (addr_, protocol, address);
if (rc != 0)
return -1;
rc = check_protocol (protocol);
if (rc != 0)
return -1;
if (protocol == "inproc") {
// TODO: inproc connect is specific with respect to creating pipes
// as there's no 'reconnect' functionality implemented. Once that
// is in place we should follow generic pipe creation algorithm.
// Find the peer endpoint.
endpoint_t peer = find_endpoint (addr_);
if (!peer.socket)
return -1;
// The total HWM for an inproc connection should be the sum of
// the binder's HWM and the connector's HWM.
int sndhwm = 0;
if (options.sndhwm != 0 && peer.options.rcvhwm != 0)
sndhwm = options.sndhwm + peer.options.rcvhwm;
int rcvhwm = 0;
if (options.rcvhwm != 0 && peer.options.sndhwm != 0)
rcvhwm = options.rcvhwm + peer.options.sndhwm;
// Create a bi-directional pipe to connect the peers.
object_t *parents [2] = {this, peer.socket};
pipe_t *pipes [2] = {NULL, NULL};
int hwms [2] = {sndhwm, rcvhwm};
bool delays [2] = {options.delay_on_disconnect, options.delay_on_close};
int rc = pipepair (parents, pipes, hwms, delays);
errno_assert (rc == 0);
// Attach local end of the pipe to this socket object.
attach_pipe (pipes [0]);
// If required, send the identity of the local socket to the peer.
if (peer.options.recv_identity) {
msg_t id;
rc = id.init_size (options.identity_size);
errno_assert (rc == 0);
memcpy (id.data (), options.identity, options.identity_size);
id.set_flags (msg_t::identity);
bool written = pipes [0]->write (&id);
zmq_assert (written);
pipes [0]->flush ();
}
// If required, send the identity of the peer to the local socket.
if (options.recv_identity) {
msg_t id;
rc = id.init_size (peer.options.identity_size);
errno_assert (rc == 0);
memcpy (id.data (), peer.options.identity, peer.options.identity_size);
id.set_flags (msg_t::identity);
bool written = pipes [1]->write (&id);
zmq_assert (written);
pipes [1]->flush ();
}
// Attach remote end of the pipe to the peer socket. Note that peer's
// seqnum was incremented in find_endpoint function. We don't need it
// increased here.
send_bind (peer.socket, pipes [1], false);
// Save last endpoint URI
options.last_endpoint.assign (addr_);
// remember inproc connections for disconnect
inprocs.insert (inprocs_t::value_type (std::string (addr_), pipes[0]));
return 0;
}
// Choose the I/O thread to run the session in.
io_thread_t *io_thread = choose_io_thread (options.affinity);
if (!io_thread) {
errno = EMTHREAD;
return -1;
}
address_t *paddr = new (std::nothrow) address_t (protocol, address);
alloc_assert (paddr);
// Resolve address (if needed by the protocol)
if (protocol == "tcp") {
paddr->resolved.tcp_addr = new (std::nothrow) tcp_address_t ();
alloc_assert (paddr->resolved.tcp_addr);
int rc = paddr->resolved.tcp_addr->resolve (
address.c_str (), false, options.ipv4only ? true : false);
if (rc != 0) {
delete paddr;
return -1;
}
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
else
if (protocol == "ipc") {
paddr->resolved.ipc_addr = new (std::nothrow) ipc_address_t ();
alloc_assert (paddr->resolved.ipc_addr);
int rc = paddr->resolved.ipc_addr->resolve (address.c_str ());
if (rc != 0) {
delete paddr;
return -1;
}
}
#endif
#ifdef ZMQ_HAVE_OPENPGM
if (protocol == "pgm" || protocol == "epgm") {
struct pgm_addrinfo_t *res = NULL;
uint16_t port_number = 0;
int rc = pgm_socket_t::init_address(address.c_str(), &res, &port_number);
if (res != NULL)
pgm_freeaddrinfo (res);
if (rc != 0 || port_number == 0)
return -1;
}
#endif
// Create session.
session_base_t *session = session_base_t::create (io_thread, true, this,
options, paddr);
errno_assert (session);
// PGM does not support subscription forwarding; ask for all data to be
// sent to this pipe.
bool icanhasall = protocol == "pgm" || protocol == "epgm";
if (options.delay_attach_on_connect != 1 || icanhasall) {
// Create a bi-directional pipe.
object_t *parents [2] = {this, session};
pipe_t *pipes [2] = {NULL, NULL};
int hwms [2] = {options.sndhwm, options.rcvhwm};
bool delays [2] = {options.delay_on_disconnect, options.delay_on_close};
rc = pipepair (parents, pipes, hwms, delays);
errno_assert (rc == 0);
// Attach local end of the pipe to the socket object.
attach_pipe (pipes [0], icanhasall);
// Attach remote end of the pipe to the session object later on.
session->attach_pipe (pipes [1]);
}
// Save last endpoint URI
paddr->to_string (options.last_endpoint);
add_endpoint (addr_, (own_t *) session);
return 0;
}
int zmq::curve_server_t::decode (msg_t *msg_)
{
zmq_assert (state == connected);
if (msg_->size () < 33) {
// Temporary support for security debugging
puts ("CURVE I: invalid CURVE client, sent malformed command");
errno = EPROTO;
return -1;
}
const uint8_t *message = static_cast <uint8_t *> (msg_->data ());
if (memcmp (message, "\x07MESSAGE", 8)) {
// Temporary support for security debugging
puts ("CURVE I: invalid CURVE client, did not send MESSAGE");
errno = EPROTO;
return -1;
}
uint8_t message_nonce [crypto_box_NONCEBYTES];
memcpy (message_nonce, "CurveZMQMESSAGEC", 16);
memcpy (message_nonce + 16, message + 8, 8);
uint64_t nonce = get_uint64(message + 8);
if (nonce <= cn_peer_nonce) {
errno = EPROTO;
return -1;
}
cn_peer_nonce = nonce;
const size_t clen = crypto_box_BOXZEROBYTES + msg_->size () - 16;
uint8_t *message_plaintext = static_cast <uint8_t *> (malloc (clen));
alloc_assert (message_plaintext);
uint8_t *message_box = static_cast <uint8_t *> (malloc (clen));
alloc_assert (message_box);
memset (message_box, 0, crypto_box_BOXZEROBYTES);
memcpy (message_box + crypto_box_BOXZEROBYTES,
message + 16, msg_->size () - 16);
int rc = crypto_box_open_afternm (message_plaintext, message_box,
clen, message_nonce, cn_precom);
if (rc == 0) {
rc = msg_->close ();
zmq_assert (rc == 0);
rc = msg_->init_size (clen - 1 - crypto_box_ZEROBYTES);
zmq_assert (rc == 0);
const uint8_t flags = message_plaintext [crypto_box_ZEROBYTES];
if (flags & 0x01)
msg_->set_flags (msg_t::more);
if (flags & 0x02)
msg_->set_flags (msg_t::command);
memcpy (msg_->data (),
message_plaintext + crypto_box_ZEROBYTES + 1,
msg_->size ());
}
else {
// Temporary support for security debugging
puts ("CURVE I: connection key used for MESSAGE is wrong");
errno = EPROTO;
}
free (message_plaintext);
free (message_box);
return rc;
}
int zmq::socket_base_t::term_endpoint (const char *addr_)
{
scoped_optional_lock_t sync_lock(thread_safe ? &sync : NULL);
// Check whether the library haven't been shut down yet.
if (unlikely (ctx_terminated)) {
errno = ETERM;
return -1;
}
// Check whether endpoint address passed to the function is valid.
if (unlikely (!addr_)) {
errno = EINVAL;
return -1;
}
// Process pending commands, if any, since there could be pending unprocessed process_own()'s
// (from launch_child() for example) we're asked to terminate now.
int rc = process_commands (0, false);
if (unlikely(rc != 0)) {
return -1;
}
// Parse addr_ string.
std::string protocol;
std::string address;
if (parse_uri(addr_, protocol, address) || check_protocol(protocol)) {
return -1;
}
// Disconnect an inproc socket
if (protocol == "inproc") {
if (unregister_endpoint (std::string(addr_), this) == 0) {
return 0;
}
std::pair <inprocs_t::iterator, inprocs_t::iterator> range = inprocs.equal_range (std::string (addr_));
if (range.first == range.second) {
errno = ENOENT;
return -1;
}
for (inprocs_t::iterator it = range.first; it != range.second; ++it)
it->second->terminate (true);
inprocs.erase (range.first, range.second);
return 0;
}
std::string resolved_addr = std::string (addr_);
std::pair <endpoints_t::iterator, endpoints_t::iterator> range;
// The resolved last_endpoint is used as a key in the endpoints map.
// The address passed by the user might not match in the TCP case due to
// IPv4-in-IPv6 mapping (EG: tcp://[::ffff:127.0.0.1]:9999), so try to
// resolve before giving up. Given at this stage we don't know whether a
// socket is connected or bound, try with both.
if (protocol == "tcp") {
range = endpoints.equal_range (resolved_addr);
if (range.first == range.second) {
tcp_address_t *tcp_addr = new (std::nothrow) tcp_address_t ();
alloc_assert (tcp_addr);
rc = tcp_addr->resolve (address.c_str (), false, options.ipv6);
if (rc == 0) {
tcp_addr->to_string (resolved_addr);
range = endpoints.equal_range (resolved_addr);
if (range.first == range.second) {
rc = tcp_addr->resolve (address.c_str (), true, options.ipv6);
if (rc == 0) {
tcp_addr->to_string (resolved_addr);
}
}
}
LIBZMQ_DELETE(tcp_addr);
}
}
// Find the endpoints range (if any) corresponding to the addr_ string.
range = endpoints.equal_range (resolved_addr);
if (range.first == range.second) {
errno = ENOENT;
return -1;
}
for (endpoints_t::iterator it = range.first; it != range.second; ++it) {
// If we have an associated pipe, terminate it.
if (it->second.second != NULL)
it->second.second->terminate (false);
term_child (it->second.first);
}
endpoints.erase (range.first, range.second);
return 0;
}
int zmq::socket_base_t::bind (const char *addr_)
{
scoped_optional_lock_t sync_lock(thread_safe ? &sync : NULL);
if (unlikely (ctx_terminated)) {
errno = ETERM;
return -1;
}
// Process pending commands, if any.
int rc = process_commands (0, false);
if (unlikely (rc != 0)) {
return -1;
}
// Parse addr_ string.
std::string protocol;
std::string address;
if (parse_uri (addr_, protocol, address) || check_protocol (protocol)) {
return -1;
}
if (protocol == "inproc") {
const endpoint_t endpoint = { this, options };
rc = register_endpoint (addr_, endpoint);
if (rc == 0) {
connect_pending (addr_, this);
last_endpoint.assign (addr_);
options.connected = true;
}
return rc;
}
if (protocol == "pgm" || protocol == "epgm" || protocol == "norm") {
// For convenience's sake, bind can be used interchangeable with
// connect for PGM, EPGM, NORM transports.
rc = connect (addr_);
if (rc != -1)
options.connected = true;
return rc;
}
if (protocol == "udp") {
if (!(options.type == ZMQ_DGRAM || options.type == ZMQ_DISH)) {
errno = ENOCOMPATPROTO;
return -1;
}
// Choose the I/O thread to run the session in.
io_thread_t *io_thread = choose_io_thread (options.affinity);
if (!io_thread) {
errno = EMTHREAD;
return -1;
}
address_t *paddr = new (std::nothrow) address_t (protocol, address, this->get_ctx ());
alloc_assert (paddr);
paddr->resolved.udp_addr = new (std::nothrow) udp_address_t ();
alloc_assert (paddr->resolved.udp_addr);
rc = paddr->resolved.udp_addr->resolve (address.c_str(), true);
if (rc != 0) {
LIBZMQ_DELETE(paddr);
return -1;
}
session_base_t *session = session_base_t::create (io_thread, true, this,
options, paddr);
errno_assert (session);
pipe_t *newpipe = NULL;
// Create a bi-directional pipe.
object_t *parents [2] = {this, session};
pipe_t *new_pipes [2] = {NULL, NULL};
int hwms [2] = {options.sndhwm, options.rcvhwm};
bool conflates [2] = {false, false};
rc = pipepair (parents, new_pipes, hwms, conflates);
errno_assert (rc == 0);
// Attach local end of the pipe to the socket object.
attach_pipe (new_pipes [0], true);
newpipe = new_pipes [0];
// Attach remote end of the pipe to the session object later on.
session->attach_pipe (new_pipes [1]);
// Save last endpoint URI
paddr->to_string (last_endpoint);
add_endpoint (addr_, (own_t *) session, newpipe);
return 0;
}
// Remaining transports require to be run in an I/O thread, so at this
// point we'll choose one.
io_thread_t *io_thread = choose_io_thread (options.affinity);
if (!io_thread) {
errno = EMTHREAD;
return -1;
}
if (protocol == "tcp") {
tcp_listener_t *listener = new (std::nothrow) tcp_listener_t (
io_thread, this, options);
alloc_assert (listener);
rc = listener->set_address (address.c_str ());
if (rc != 0) {
LIBZMQ_DELETE(listener);
event_bind_failed (address, zmq_errno());
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (last_endpoint.c_str (), (own_t *) listener, NULL);
options.connected = true;
return 0;
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
if (protocol == "ipc") {
ipc_listener_t *listener = new (std::nothrow) ipc_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
LIBZMQ_DELETE(listener);
event_bind_failed (address, zmq_errno());
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (last_endpoint.c_str (), (own_t *) listener, NULL);
options.connected = true;
return 0;
}
#endif
#if defined ZMQ_HAVE_TIPC
if (protocol == "tipc") {
tipc_listener_t *listener = new (std::nothrow) tipc_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
LIBZMQ_DELETE(listener);
event_bind_failed (address, zmq_errno());
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (addr_, (own_t *) listener, NULL);
options.connected = true;
return 0;
}
#endif
#if defined ZMQ_HAVE_VMCI
if (protocol == "vmci") {
vmci_listener_t *listener = new (std::nothrow) vmci_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
LIBZMQ_DELETE(listener);
event_bind_failed (address, zmq_errno ());
return -1;
}
listener->get_address (last_endpoint);
add_endpoint (last_endpoint.c_str(), (own_t *) listener, NULL);
options.connected = true;
return 0;
}
#endif
zmq_assert (false);
return -1;
}
static void nn_ctx_init (void)
{
int i;
#if defined NN_HAVE_WINDOWS
WSADATA data;
int rc;
#endif
/* Check whether the library was already initialised. If so, do nothing. */
if (self.socks)
return;
/* On Windows, initialise the socket library. */
#if defined NN_HAVE_WINDOWS
rc = WSAStartup (MAKEWORD (2, 2), &data);
nn_assert (rc == 0);
nn_assert (LOBYTE (data.wVersion) == 2 &&
HIBYTE (data.wVersion) == 2);
#endif
/* Initialise the memory allocation subsystem. */
nn_alloc_init ();
/* Seed the pseudo-random number generator. */
nn_random_seed ();
/* Allocate the global table of SP sockets. */
self.socks = nn_alloc ((sizeof (struct nn_sock*) * NN_MAX_SOCKETS) +
(sizeof (uint16_t) * NN_MAX_SOCKETS), "socket table");
alloc_assert (self.socks);
for (i = 0; i != NN_MAX_SOCKETS; ++i)
self.socks [i] = NULL;
self.nsocks = 0;
self.flags = 0;
/* Allocate the stack of unused file descriptors. */
self.unused = (uint16_t*) (self.socks + NN_MAX_SOCKETS);
alloc_assert (self.unused);
for (i = 0; i != NN_MAX_SOCKETS; ++i)
self.unused [i] = NN_MAX_SOCKETS - i - 1;
/* Initialise other parts of the global state. */
nn_list_init (&self.transports);
nn_list_init (&self.socktypes);
/* Plug in individual transports. */
nn_ctx_add_transport (nn_inproc);
#if !defined NN_HAVE_WINDOWS
nn_ctx_add_transport (nn_ipc);
#endif
nn_ctx_add_transport (nn_tcp);
/* Plug in individual socktypes. */
nn_ctx_add_socktype (nn_pair_socktype);
nn_ctx_add_socktype (nn_xpair_socktype);
nn_ctx_add_socktype (nn_pub_socktype);
nn_ctx_add_socktype (nn_sub_socktype);
nn_ctx_add_socktype (nn_rep_socktype);
nn_ctx_add_socktype (nn_req_socktype);
nn_ctx_add_socktype (nn_xrep_socktype);
nn_ctx_add_socktype (nn_xreq_socktype);
nn_ctx_add_socktype (nn_sink_socktype);
nn_ctx_add_socktype (nn_source_socktype);
nn_ctx_add_socktype (nn_xsink_socktype);
nn_ctx_add_socktype (nn_xsource_socktype);
nn_ctx_add_socktype (nn_push_socktype);
nn_ctx_add_socktype (nn_xpush_socktype);
nn_ctx_add_socktype (nn_pull_socktype);
nn_ctx_add_socktype (nn_xpull_socktype);
nn_ctx_add_socktype (nn_respondent_socktype);
nn_ctx_add_socktype (nn_surveyor_socktype);
nn_ctx_add_socktype (nn_xrespondent_socktype);
nn_ctx_add_socktype (nn_xsurveyor_socktype);
nn_ctx_add_socktype (nn_bus_socktype);
nn_ctx_add_socktype (nn_xbus_socktype);
#if defined NN_LATENCY_MONITOR
nn_latmon_init ();
#endif
}
void *zmq_timers_new (void)
{
zmq::timers_t *timers = new (std::nothrow) zmq::timers_t;
alloc_assert (timers);
return timers;
}
static void pfx_rm_all (pfx_node_t *node_, void *subscribers_,
unsigned char **buff_, size_t buffsize_, size_t maxbuffsize_, void *arg_)
{
// Remove the subscription from this node.
if (node_->subscribers) {
pfx_node_t::subscribers_t::iterator it =
node_->subscribers->find (subscribers_);
if (it != node_->subscribers->end ()) {
xs_assert (it->second);
--it->second;
if (!it->second) {
node_->subscribers->erase (it);
if (node_->subscribers->empty ()) {
int rc = xs_filter_unsubscribed (arg_, *buff_, buffsize_);
errno_assert (rc == 0);
delete node_->subscribers;
node_->subscribers = 0;
}
}
}
}
// Adjust the buffer.
if (buffsize_ >= maxbuffsize_) {
maxbuffsize_ = buffsize_ + 256;
*buff_ = (unsigned char*) realloc (*buff_, maxbuffsize_);
alloc_assert (*buff_);
}
// If there are no subnodes in the trie, return.
if (node_->count == 0)
return;
// If there's one subnode (optimisation).
if (node_->count == 1) {
(*buff_) [buffsize_] = node_->min;
buffsize_++;
pfx_rm_all (node_->next.node, subscribers_, buff_, buffsize_,
maxbuffsize_, arg_);
// Prune the node if it was made redundant by the removal
if (pfx_is_redundant (node_->next.node)) {
pfx_close (node_->next.node);
free (node_->next.node);
node_->next.node = 0;
node_->count = 0;
--node_->live_nodes;
xs_assert (node_->live_nodes == 0);
}
return;
}
// If there are multiple subnodes.
// New min non-null character in the node table after the removal.
unsigned char new_min = node_->min + node_->count - 1;
// New max non-null character in the node table after the removal.
unsigned char new_max = node_->min;
for (unsigned short c = 0; c != node_->count; c++) {
(*buff_) [buffsize_] = node_->min + c;
if (node_->next.table [c]) {
pfx_rm_all (node_->next.table [c], subscribers_, buff_,
buffsize_ + 1, maxbuffsize_, arg_);
// Prune redundant nodes from the the trie.
if (pfx_is_redundant (node_->next.table [c])) {
pfx_close (node_->next.table [c]);
free (node_->next.table [c]);
node_->next.table [c] = 0;
xs_assert (node_->live_nodes > 0);
--node_->live_nodes;
}
else {
// The node is not redundant, so it's a candidate for being
// the new min/max node.
//
// We loop through the node array from left to right, so the
// first non-null, non-redundant node encountered is the new
// minimum index. Conversely, the last non-redundant, non-null
// node encountered is the new maximum index.
if (c + node_->min < new_min)
new_min = c + node_->min;
if (c + node_->min > new_max)
new_max = c + node_->min;
}
}
}
xs_assert (node_->count > 1);
// Compact the node table if possible.
if (node_->live_nodes == 0) {
// Free the node table if it's no longer used.
free (node_->next.table);
node_->next.table = NULL;
node_->count = 0;
}
else if (node_->live_nodes == 1) {
// If there's only one live node in the table we can
// switch to using the more compact single-node
// representation.
xs_assert (new_min == new_max);
xs_assert (new_min >= node_->min &&
new_min < node_->min + node_->count);
pfx_node_t *node = node_->next.table [new_min - node_->min];
xs_assert (node);
free (node_->next.table);
node_->next.node = node;
node_->count = 1;
node_->min = new_min;
}
else if (new_min > node_->min ||
new_max < node_->min + node_->count - 1) {
xs_assert (new_max - new_min + 1 > 1);
pfx_node_t **old_table = node_->next.table;
xs_assert (new_min > node_->min ||
new_max < node_->min + node_->count - 1);
xs_assert (new_min >= node_->min);
xs_assert (new_max <= node_->min + node_->count - 1);
xs_assert (new_max - new_min + 1 < node_->count);
node_->count = new_max - new_min + 1;
node_->next.table =
(pfx_node_t**) malloc (sizeof (pfx_node_t*) * node_->count);
alloc_assert (node_->next.table);
memmove (node_->next.table, old_table + (new_min - node_->min),
sizeof (pfx_node_t*) * node_->count);
free (old_table);
node_->min = new_min;
}
}
int zmq_poll (zmq_pollitem_t *items_, int nitems_, long timeout_)
{
// TODO: the function implementation can just call zmq_pollfd_poll with
// pollfd as NULL, however pollfd is not yet stable.
#if defined ZMQ_POLL_BASED_ON_POLL
if (unlikely (nitems_ < 0)) {
errno = EINVAL;
return -1;
}
if (unlikely (nitems_ == 0)) {
if (timeout_ == 0)
return 0;
#if defined ZMQ_HAVE_WINDOWS
Sleep (timeout_ > 0 ? timeout_ : INFINITE);
return 0;
#elif defined ZMQ_HAVE_ANDROID
usleep (timeout_ * 1000);
return 0;
#else
return usleep (timeout_ * 1000);
#endif
}
if (!items_) {
errno = EFAULT;
return -1;
}
zmq::clock_t clock;
uint64_t now = 0;
uint64_t end = 0;
pollfd spollfds[ZMQ_POLLITEMS_DFLT];
pollfd *pollfds = spollfds;
if (nitems_ > ZMQ_POLLITEMS_DFLT) {
pollfds = (pollfd*) malloc (nitems_ * sizeof (pollfd));
alloc_assert (pollfds);
}
// Build pollset for poll () system call.
for (int i = 0; i != nitems_; i++) {
// If the poll item is a 0MQ socket, we poll on the file descriptor
// retrieved by the ZMQ_FD socket option.
if (items_ [i].socket) {
size_t zmq_fd_size = sizeof (zmq::fd_t);
if (zmq_getsockopt (items_ [i].socket, ZMQ_FD, &pollfds [i].fd,
&zmq_fd_size) == -1) {
if (pollfds != spollfds)
free (pollfds);
return -1;
}
pollfds [i].events = items_ [i].events ? POLLIN : 0;
}
// Else, the poll item is a raw file descriptor. Just convert the
// events to normal POLLIN/POLLOUT for poll ().
else {
pollfds [i].fd = items_ [i].fd;
pollfds [i].events =
(items_ [i].events & ZMQ_POLLIN ? POLLIN : 0) |
(items_ [i].events & ZMQ_POLLOUT ? POLLOUT : 0) |
(items_ [i].events & ZMQ_POLLPRI ? POLLPRI : 0);
}
}
bool first_pass = true;
int nevents = 0;
while (true) {
// Compute the timeout for the subsequent poll.
int timeout;
if (first_pass)
timeout = 0;
else
if (timeout_ < 0)
timeout = -1;
else
timeout = end - now;
// Wait for events.
while (true) {
int rc = poll (pollfds, nitems_, timeout);
if (rc == -1 && errno == EINTR) {
if (pollfds != spollfds)
free (pollfds);
return -1;
}
errno_assert (rc >= 0);
break;
}
// Check for the events.
for (int i = 0; i != nitems_; i++) {
items_ [i].revents = 0;
// The poll item is a 0MQ socket. Retrieve pending events
// using the ZMQ_EVENTS socket option.
if (items_ [i].socket) {
size_t zmq_events_size = sizeof (uint32_t);
uint32_t zmq_events;
if (zmq_getsockopt (items_ [i].socket, ZMQ_EVENTS, &zmq_events,
&zmq_events_size) == -1) {
if (pollfds != spollfds)
free (pollfds);
return -1;
}
if ((items_ [i].events & ZMQ_POLLOUT) &&
(zmq_events & ZMQ_POLLOUT))
items_ [i].revents |= ZMQ_POLLOUT;
if ((items_ [i].events & ZMQ_POLLIN) &&
(zmq_events & ZMQ_POLLIN))
items_ [i].revents |= ZMQ_POLLIN;
}
// Else, the poll item is a raw file descriptor, simply convert
// the events to zmq_pollitem_t-style format.
else {
if (pollfds [i].revents & POLLIN)
items_ [i].revents |= ZMQ_POLLIN;
if (pollfds [i].revents & POLLOUT)
items_ [i].revents |= ZMQ_POLLOUT;
if (pollfds [i].revents & POLLPRI)
items_ [i].revents |= ZMQ_POLLPRI;
if (pollfds [i].revents & ~(POLLIN | POLLOUT | POLLPRI))
items_ [i].revents |= ZMQ_POLLERR;
}
if (items_ [i].revents)
nevents++;
}
// If timeout is zero, exit immediately whether there are events or not.
if (timeout_ == 0)
break;
// If there are events to return, we can exit immediately.
if (nevents)
break;
// At this point we are meant to wait for events but there are none.
// If timeout is infinite we can just loop until we get some events.
if (timeout_ < 0) {
if (first_pass)
first_pass = false;
continue;
}
// The timeout is finite and there are no events. In the first pass
// we get a timestamp of when the polling have begun. (We assume that
// first pass have taken negligible time). We also compute the time
// when the polling should time out.
if (first_pass) {
now = clock.now_ms ();
end = now + timeout_;
if (now == end)
break;
first_pass = false;
continue;
}
// Find out whether timeout have expired.
now = clock.now_ms ();
if (now >= end)
break;
}
if (pollfds != spollfds)
free (pollfds);
return nevents;
#elif defined ZMQ_POLL_BASED_ON_SELECT
if (unlikely (nitems_ < 0)) {
errno = EINVAL;
return -1;
}
if (unlikely (nitems_ == 0)) {
if (timeout_ == 0)
return 0;
#if defined ZMQ_HAVE_WINDOWS
Sleep (timeout_ > 0 ? timeout_ : INFINITE);
return 0;
#else
return usleep (timeout_ * 1000);
#endif
}
zmq::clock_t clock;
uint64_t now = 0;
uint64_t end = 0;
// Ensure we do not attempt to select () on more than FD_SETSIZE
// file descriptors.
zmq_assert (nitems_ <= FD_SETSIZE);
fd_set pollset_in = { 0 };
fd_set pollset_out = { 0 };
fd_set pollset_err = { 0 };
zmq::fd_t maxfd = 0;
// Build the fd_sets for passing to select ().
for (int i = 0; i != nitems_; i++) {
// If the poll item is a 0MQ socket we are interested in input on the
// notification file descriptor retrieved by the ZMQ_FD socket option.
if (items_ [i].socket) {
size_t zmq_fd_size = sizeof (zmq::fd_t);
zmq::fd_t notify_fd;
if (zmq_getsockopt (items_ [i].socket, ZMQ_FD, ¬ify_fd,
&zmq_fd_size) == -1)
return -1;
if (items_ [i].events) {
FD_SET (notify_fd, &pollset_in);
if (maxfd < notify_fd)
maxfd = notify_fd;
}
}
// Else, the poll item is a raw file descriptor. Convert the poll item
// events to the appropriate fd_sets.
else {
if (items_ [i].events & ZMQ_POLLIN)
FD_SET (items_ [i].fd, &pollset_in);
if (items_ [i].events & ZMQ_POLLOUT)
FD_SET (items_ [i].fd, &pollset_out);
if (items_ [i].events & ZMQ_POLLERR)
FD_SET (items_ [i].fd, &pollset_err);
if (maxfd < items_ [i].fd)
maxfd = items_ [i].fd;
}
}
bool first_pass = true;
int nevents = 0;
fd_set inset, outset, errset;
while (true) {
// Compute the timeout for the subsequent poll.
timeval timeout;
timeval *ptimeout;
if (first_pass) {
timeout.tv_sec = 0;
timeout.tv_usec = 0;
ptimeout = &timeout;
}
else
if (timeout_ < 0)
ptimeout = NULL;
else {
timeout.tv_sec = (long) ((end - now) / 1000);
timeout.tv_usec = (long) ((end - now) % 1000 * 1000);
ptimeout = &timeout;
}
// Wait for events. Ignore interrupts if there's infinite timeout.
while (true) {
memcpy (&inset, &pollset_in, sizeof (fd_set));
memcpy (&outset, &pollset_out, sizeof (fd_set));
memcpy (&errset, &pollset_err, sizeof (fd_set));
#if defined ZMQ_HAVE_WINDOWS
int rc = select (0, &inset, &outset, &errset, ptimeout);
if (unlikely (rc == SOCKET_ERROR)) {
errno = zmq::wsa_error_to_errno (WSAGetLastError ());
wsa_assert (errno == ENOTSOCK);
return -1;
}
#else
int rc = select (maxfd + 1, &inset, &outset, &errset, ptimeout);
if (unlikely (rc == -1)) {
errno_assert (errno == EINTR || errno == EBADF);
return -1;
}
#endif
break;
}
// Check for the events.
for (int i = 0; i != nitems_; i++) {
items_ [i].revents = 0;
// The poll item is a 0MQ socket. Retrieve pending events
// using the ZMQ_EVENTS socket option.
if (items_ [i].socket) {
size_t zmq_events_size = sizeof (uint32_t);
uint32_t zmq_events;
if (zmq_getsockopt (items_ [i].socket, ZMQ_EVENTS, &zmq_events,
&zmq_events_size) == -1)
return -1;
if ((items_ [i].events & ZMQ_POLLOUT) &&
(zmq_events & ZMQ_POLLOUT))
items_ [i].revents |= ZMQ_POLLOUT;
if ((items_ [i].events & ZMQ_POLLIN) &&
(zmq_events & ZMQ_POLLIN))
items_ [i].revents |= ZMQ_POLLIN;
}
// Else, the poll item is a raw file descriptor, simply convert
// the events to zmq_pollitem_t-style format.
else {
if (FD_ISSET (items_ [i].fd, &inset))
items_ [i].revents |= ZMQ_POLLIN;
if (FD_ISSET (items_ [i].fd, &outset))
items_ [i].revents |= ZMQ_POLLOUT;
if (FD_ISSET (items_ [i].fd, &errset))
items_ [i].revents |= ZMQ_POLLERR;
}
if (items_ [i].revents)
nevents++;
}
// If timeout is zero, exit immediately whether there are events or not.
if (timeout_ == 0)
break;
// If there are events to return, we can exit immediately.
if (nevents)
break;
// At this point we are meant to wait for events but there are none.
// If timeout is infinite we can just loop until we get some events.
if (timeout_ < 0) {
if (first_pass)
first_pass = false;
continue;
}
// The timeout is finite and there are no events. In the first pass
// we get a timestamp of when the polling have begun. (We assume that
// first pass have taken negligible time). We also compute the time
// when the polling should time out.
if (first_pass) {
now = clock.now_ms ();
end = now + timeout_;
if (now == end)
break;
first_pass = false;
continue;
}
// Find out whether timeout have expired.
now = clock.now_ms ();
if (now >= end)
break;
}
return nevents;
#else
// Exotic platforms that support neither poll() nor select().
errno = ENOTSUP;
return -1;
#endif
}
static bool pfx_add (pfx_node_t *node_,
const unsigned char *prefix_, size_t size_, void *subscriber_)
{
// We are at the node corresponding to the prefix. We are done.
if (!size_) {
bool result = !node_->subscribers;
if (!node_->subscribers)
node_->subscribers = new (std::nothrow) pfx_node_t::subscribers_t;
pfx_node_t::subscribers_t::iterator it = node_->subscribers->insert (
pfx_node_t::subscribers_t::value_type (subscriber_, 0)).first;
++it->second;
return result;
}
unsigned char c = *prefix_;
if (c < node_->min || c >= node_->min + node_->count) {
// The character is out of range of currently handled
// charcters. We have to extend the table.
if (!node_->count) {
node_->min = c;
node_->count = 1;
node_->next.node = NULL;
}
else if (node_->count == 1) {
unsigned char oldc = node_->min;
pfx_node_t *oldp = node_->next.node;
node_->count =
(node_->min < c ? c - node_->min : node_->min - c) + 1;
node_->next.table = (pfx_node_t**)
malloc (sizeof (pfx_node_t*) * node_->count);
alloc_assert (node_->next.table);
for (unsigned short i = 0; i != node_->count; ++i)
node_->next.table [i] = 0;
node_->min = std::min (node_->min, c);
node_->next.table [oldc - node_->min] = oldp;
}
else if (node_->min < c) {
// The new character is above the current character range.
unsigned short old_count = node_->count;
node_->count = c - node_->min + 1;
node_->next.table =
(pfx_node_t**) realloc ((void*) node_->next.table,
sizeof (pfx_node_t*) * node_->count);
xs_assert (node_->next.table);
for (unsigned short i = old_count; i != node_->count; i++)
node_->next.table [i] = NULL;
}
else {
// The new character is below the current character range.
unsigned short old_count = node_->count;
node_->count = (node_->min + old_count) - c;
node_->next.table =
(pfx_node_t**) realloc ((void*) node_->next.table,
sizeof (pfx_node_t*) * node_->count);
xs_assert (node_->next.table);
memmove (node_->next.table + node_->min - c, node_->next.table,
old_count * sizeof (pfx_node_t*));
for (unsigned short i = 0; i != node_->min - c; i++)
node_->next.table [i] = NULL;
node_->min = c;
}
}
// If next node does not exist, create one.
if (node_->count == 1) {
if (!node_->next.node) {
node_->next.node = (pfx_node_t*) malloc (sizeof (pfx_node_t));
alloc_assert (node_->next.node);
pfx_init (node_->next.node);
++node_->live_nodes;
xs_assert (node_->next.node);
}
return pfx_add (node_->next.node, prefix_ + 1, size_ - 1, subscriber_);
}
else {
if (!node_->next.table [c - node_->min]) {
node_->next.table [c - node_->min] =
(pfx_node_t*) malloc (sizeof (pfx_node_t));
alloc_assert (node_->next.table [c - node_->min]);
pfx_init (node_->next.table [c - node_->min]);
++node_->live_nodes;
xs_assert (node_->next.table [c - node_->min]);
}
return pfx_add (node_->next.table [c - node_->min],
prefix_ + 1, size_ - 1, subscriber_);
}
}
int zmq::socket_base_t::bind (const char *addr_)
{
ENTER_MUTEX();
if (unlikely (ctx_terminated)) {
errno = ETERM;
EXIT_MUTEX();
return -1;
}
// Process pending commands, if any.
int rc = process_commands (0, false);
if (unlikely (rc != 0)) {
EXIT_MUTEX();
return -1;
}
// Parse addr_ string.
std::string protocol;
std::string address;
if (parse_uri (addr_, protocol, address) || check_protocol (protocol)) {
EXIT_MUTEX();
return -1;
}
if (protocol == "inproc") {
const endpoint_t endpoint = { this, options };
const int rc = register_endpoint (addr_, endpoint);
if (rc == 0) {
connect_pending (addr_, this);
last_endpoint.assign (addr_);
}
EXIT_MUTEX();
return rc;
}
if (protocol == "pgm" || protocol == "epgm" || protocol == "norm") {
// For convenience's sake, bind can be used interchageable with
// connect for PGM, EPGM and NORM transports.
EXIT_MUTEX();
return connect (addr_);
}
// Remaining trasnports require to be run in an I/O thread, so at this
// point we'll choose one.
io_thread_t *io_thread = choose_io_thread (options.affinity);
if (!io_thread) {
errno = EMTHREAD;
EXIT_MUTEX();
return -1;
}
if (protocol == "tcp") {
tcp_listener_t *listener = new (std::nothrow) tcp_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
delete listener;
event_bind_failed (address, zmq_errno());
EXIT_MUTEX();
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (last_endpoint.c_str (), (own_t *) listener, NULL);
EXIT_MUTEX();
return 0;
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
if (protocol == "ipc") {
ipc_listener_t *listener = new (std::nothrow) ipc_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
delete listener;
event_bind_failed (address, zmq_errno());
EXIT_MUTEX();
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (last_endpoint.c_str (), (own_t *) listener, NULL);
EXIT_MUTEX();
return 0;
}
#endif
#if defined ZMQ_HAVE_TIPC
if (protocol == "tipc") {
tipc_listener_t *listener = new (std::nothrow) tipc_listener_t (
io_thread, this, options);
alloc_assert (listener);
int rc = listener->set_address (address.c_str ());
if (rc != 0) {
delete listener;
event_bind_failed (address, zmq_errno());
EXIT_MUTEX();
return -1;
}
// Save last endpoint URI
listener->get_address (last_endpoint);
add_endpoint (addr_, (own_t *) listener, NULL);
EXIT_MUTEX();
return 0;
}
#endif
EXIT_MUTEX();
zmq_assert (false);
return -1;
}
bool zmq::stream_engine_t::handshake ()
{
zmq_assert (handshaking);
zmq_assert (greeting_bytes_read < greeting_size);
// Receive the greeting.
while (greeting_bytes_read < greeting_size) {
const int n = read (greeting_recv + greeting_bytes_read,
greeting_size - greeting_bytes_read);
if (n == 0) {
error ();
return false;
}
if (n == -1) {
if (errno != EAGAIN)
error ();
return false;
}
greeting_bytes_read += n;
// We have received at least one byte from the peer.
// If the first byte is not 0xff, we know that the
// peer is using unversioned protocol.
if (greeting_recv [0] != 0xff)
break;
if (greeting_bytes_read < signature_size)
continue;
// Inspect the right-most bit of the 10th byte (which coincides
// with the 'flags' field if a regular message was sent).
// Zero indicates this is a header of identity message
// (i.e. the peer is using the unversioned protocol).
if (!(greeting_recv [9] & 0x01))
break;
// The peer is using versioned protocol.
// Send the major version number.
if (outpos + outsize == greeting_send + signature_size) {
if (outsize == 0)
set_pollout (handle);
outpos [outsize++] = 3; // Major version number
}
if (greeting_bytes_read > signature_size) {
if (outpos + outsize == greeting_send + signature_size + 1) {
if (outsize == 0)
set_pollout (handle);
// Use ZMTP/2.0 to talk to older peers.
if (greeting_recv [10] == ZMTP_1_0
|| greeting_recv [10] == ZMTP_2_0)
outpos [outsize++] = options.type;
else {
outpos [outsize++] = 0; // Minor version number
memset (outpos + outsize, 0, 20);
zmq_assert (options.mechanism == ZMQ_NULL
|| options.mechanism == ZMQ_PLAIN
|| options.mechanism == ZMQ_CURVE);
if (options.mechanism == ZMQ_NULL)
memcpy (outpos + outsize, "NULL", 4);
else
if (options.mechanism == ZMQ_PLAIN)
memcpy (outpos + outsize, "PLAIN", 5);
else
memcpy (outpos + outsize, "CURVE", 5);
outsize += 20;
memset (outpos + outsize, 0, 32);
outsize += 32;
greeting_size = v3_greeting_size;
}
}
}
}
// Position of the revision field in the greeting.
const size_t revision_pos = 10;
// Is the peer using ZMTP/1.0 with no revision number?
// If so, we send and receive rest of identity message
if (greeting_recv [0] != 0xff || !(greeting_recv [9] & 0x01)) {
encoder = new (std::nothrow) v1_encoder_t (out_batch_size);
alloc_assert (encoder);
decoder = new (std::nothrow) v1_decoder_t (in_batch_size, options.maxmsgsize);
alloc_assert (decoder);
// We have already sent the message header.
// Since there is no way to tell the encoder to
// skip the message header, we simply throw that
// header data away.
const size_t header_size = options.identity_size + 1 >= 255 ? 10 : 2;
unsigned char tmp [10], *bufferp = tmp;
// Prepare the identity message and load it into encoder.
// Then consume bytes we have already sent to the peer.
const int rc = tx_msg.init_size (options.identity_size);
zmq_assert (rc == 0);
memcpy (tx_msg.data (), options.identity, options.identity_size);
encoder->load_msg (&tx_msg);
size_t buffer_size = encoder->encode (&bufferp, header_size);
zmq_assert (buffer_size == header_size);
// Make sure the decoder sees the data we have already received.
inpos = greeting_recv;
insize = greeting_bytes_read;
// To allow for interoperability with peers that do not forward
// their subscriptions, we inject a phantom subscription message
// message into the incoming message stream.
if (options.type == ZMQ_PUB || options.type == ZMQ_XPUB)
subscription_required = true;
// We are sending our identity now and the next message
// will come from the socket.
read_msg = &stream_engine_t::pull_msg_from_session;
// We are expecting identity message.
write_msg = &stream_engine_t::write_identity;
}
else
if (greeting_recv [revision_pos] == ZMTP_1_0) {
encoder = new (std::nothrow) v1_encoder_t (
out_batch_size);
alloc_assert (encoder);
decoder = new (std::nothrow) v1_decoder_t (
in_batch_size, options.maxmsgsize);
alloc_assert (decoder);
}
else
if (greeting_recv [revision_pos] == ZMTP_2_0) {
encoder = new (std::nothrow) v2_encoder_t (out_batch_size);
alloc_assert (encoder);
decoder = new (std::nothrow) v2_decoder_t (
in_batch_size, options.maxmsgsize);
alloc_assert (decoder);
}
else {
encoder = new (std::nothrow) v2_encoder_t (out_batch_size);
alloc_assert (encoder);
decoder = new (std::nothrow) v2_decoder_t (
in_batch_size, options.maxmsgsize);
alloc_assert (decoder);
if (memcmp (greeting_recv + 12, "NULL\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 20) == 0) {
mechanism = new (std::nothrow)
null_mechanism_t (session, peer_address, options);
alloc_assert (mechanism);
}
else
if (memcmp (greeting_recv + 12, "PLAIN\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 20) == 0) {
mechanism = new (std::nothrow)
plain_mechanism_t (session, peer_address, options);
alloc_assert (mechanism);
}
#ifdef HAVE_LIBSODIUM
else
if (memcmp (greeting_recv + 12, "CURVE\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0", 20) == 0) {
if (options.as_server)
mechanism = new (std::nothrow)
curve_server_t (session, peer_address, options);
else
mechanism = new (std::nothrow) curve_client_t (options);
alloc_assert (mechanism);
}
#endif
else {
error ();
return false;
}
read_msg = &stream_engine_t::next_handshake_command;
write_msg = &stream_engine_t::process_handshake_command;
}
// Start polling for output if necessary.
if (outsize == 0)
set_pollout (handle);
// Handshaking was successful.
// Switch into the normal message flow.
handshaking = false;
return true;
}
int zmq::socket_base_t::connect (const char *addr_)
{
ENTER_MUTEX();
if (unlikely (ctx_terminated)) {
errno = ETERM;
EXIT_MUTEX();
return -1;
}
// Process pending commands, if any.
int rc = process_commands (0, false);
if (unlikely (rc != 0)) {
EXIT_MUTEX();
return -1;
}
// Parse addr_ string.
std::string protocol;
std::string address;
if (parse_uri (addr_, protocol, address) || check_protocol (protocol)) {
EXIT_MUTEX();
return -1;
}
if (protocol == "inproc") {
// TODO: inproc connect is specific with respect to creating pipes
// as there's no 'reconnect' functionality implemented. Once that
// is in place we should follow generic pipe creation algorithm.
// Find the peer endpoint.
endpoint_t peer = find_endpoint (addr_);
// The total HWM for an inproc connection should be the sum of
// the binder's HWM and the connector's HWM.
int sndhwm = 0;
if (peer.socket == NULL)
sndhwm = options.sndhwm;
else if (options.sndhwm != 0 && peer.options.rcvhwm != 0)
sndhwm = options.sndhwm + peer.options.rcvhwm;
int rcvhwm = 0;
if (peer.socket == NULL)
rcvhwm = options.rcvhwm;
else
if (options.rcvhwm != 0 && peer.options.sndhwm != 0)
rcvhwm = options.rcvhwm + peer.options.sndhwm;
// Create a bi-directional pipe to connect the peers.
object_t *parents [2] = {this, peer.socket == NULL ? this : peer.socket};
pipe_t *new_pipes [2] = {NULL, NULL};
bool conflate = options.conflate &&
(options.type == ZMQ_DEALER ||
options.type == ZMQ_PULL ||
options.type == ZMQ_PUSH ||
options.type == ZMQ_PUB ||
options.type == ZMQ_SUB);
int hwms [2] = {conflate? -1 : sndhwm, conflate? -1 : rcvhwm};
bool conflates [2] = {conflate, conflate};
int rc = pipepair (parents, new_pipes, hwms, conflates);
errno_assert (rc == 0);
if (!peer.socket) {
// The peer doesn't exist yet so we don't know whether
// to send the identity message or not. To resolve this,
// we always send our identity and drop it later if
// the peer doesn't expect it.
msg_t id;
rc = id.init_size (options.identity_size);
errno_assert (rc == 0);
memcpy (id.data (), options.identity, options.identity_size);
id.set_flags (msg_t::identity);
bool written = new_pipes [0]->write (&id);
zmq_assert (written);
new_pipes [0]->flush ();
const endpoint_t endpoint = {this, options};
pend_connection (std::string (addr_), endpoint, new_pipes);
}
else {
// If required, send the identity of the local socket to the peer.
if (peer.options.recv_identity) {
msg_t id;
rc = id.init_size (options.identity_size);
errno_assert (rc == 0);
memcpy (id.data (), options.identity, options.identity_size);
id.set_flags (msg_t::identity);
bool written = new_pipes [0]->write (&id);
zmq_assert (written);
new_pipes [0]->flush ();
}
// If required, send the identity of the peer to the local socket.
if (options.recv_identity) {
msg_t id;
rc = id.init_size (peer.options.identity_size);
errno_assert (rc == 0);
memcpy (id.data (), peer.options.identity, peer.options.identity_size);
id.set_flags (msg_t::identity);
bool written = new_pipes [1]->write (&id);
zmq_assert (written);
new_pipes [1]->flush ();
}
// Attach remote end of the pipe to the peer socket. Note that peer's
// seqnum was incremented in find_endpoint function. We don't need it
// increased here.
send_bind (peer.socket, new_pipes [1], false);
}
// Attach local end of the pipe to this socket object.
attach_pipe (new_pipes [0]);
// Save last endpoint URI
last_endpoint.assign (addr_);
// remember inproc connections for disconnect
inprocs.insert (inprocs_t::value_type (std::string (addr_), new_pipes [0]));
EXIT_MUTEX();
return 0;
}
bool is_single_connect = (options.type == ZMQ_DEALER ||
options.type == ZMQ_SUB ||
options.type == ZMQ_REQ);
if (unlikely (is_single_connect)) {
const endpoints_t::iterator it = endpoints.find (addr_);
if (it != endpoints.end ()) {
// There is no valid use for multiple connects for SUB-PUB nor
// DEALER-ROUTER nor REQ-REP. Multiple connects produces
// nonsensical results.
EXIT_MUTEX();
return 0;
}
}
// Choose the I/O thread to run the session in.
io_thread_t *io_thread = choose_io_thread (options.affinity);
if (!io_thread) {
errno = EMTHREAD;
EXIT_MUTEX();
return -1;
}
address_t *paddr = new (std::nothrow) address_t (protocol, address);
alloc_assert (paddr);
// Resolve address (if needed by the protocol)
if (protocol == "tcp") {
// Do some basic sanity checks on tcp:// address syntax
// - hostname starts with digit or letter, with embedded '-' or '.'
// - IPv6 address may contain hex chars and colons.
// - IPv4 address may contain decimal digits and dots.
// - Address must end in ":port" where port is *, or numeric
// - Address may contain two parts separated by ':'
// Following code is quick and dirty check to catch obvious errors,
// without trying to be fully accurate.
const char *check = address.c_str ();
if (isalnum (*check) || isxdigit (*check)) {
check++;
while (isalnum (*check)
|| isxdigit (*check)
|| *check == '.' || *check == '-' || *check == ':'|| *check == ';')
check++;
}
// Assume the worst, now look for success
rc = -1;
// Did we reach the end of the address safely?
if (*check == 0) {
// Do we have a valid port string? (cannot be '*' in connect
check = strrchr (address.c_str (), ':');
if (check) {
check++;
if (*check && (isdigit (*check)))
rc = 0; // Valid
}
}
if (rc == -1) {
errno = EINVAL;
delete paddr;
EXIT_MUTEX();
return -1;
}
// Defer resolution until a socket is opened
paddr->resolved.tcp_addr = NULL;
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
else
if (protocol == "ipc") {
paddr->resolved.ipc_addr = new (std::nothrow) ipc_address_t ();
alloc_assert (paddr->resolved.ipc_addr);
int rc = paddr->resolved.ipc_addr->resolve (address.c_str ());
if (rc != 0) {
delete paddr;
EXIT_MUTEX();
return -1;
}
}
#endif
// TBD - Should we check address for ZMQ_HAVE_NORM???
#ifdef ZMQ_HAVE_OPENPGM
if (protocol == "pgm" || protocol == "epgm") {
struct pgm_addrinfo_t *res = NULL;
uint16_t port_number = 0;
int rc = pgm_socket_t::init_address(address.c_str(), &res, &port_number);
if (res != NULL)
pgm_freeaddrinfo (res);
if (rc != 0 || port_number == 0)
EXIT_MUTEX();
return -1;
}
#endif
#if defined ZMQ_HAVE_TIPC
else
if (protocol == "tipc") {
paddr->resolved.tipc_addr = new (std::nothrow) tipc_address_t ();
alloc_assert (paddr->resolved.tipc_addr);
int rc = paddr->resolved.tipc_addr->resolve (address.c_str());
if (rc != 0) {
delete paddr;
EXIT_MUTEX();
return -1;
}
}
#endif
// Create session.
session_base_t *session = session_base_t::create (io_thread, true, this,
options, paddr);
errno_assert (session);
// PGM does not support subscription forwarding; ask for all data to be
// sent to this pipe. (same for NORM, currently?)
bool subscribe_to_all = protocol == "pgm" || protocol == "epgm" || protocol == "norm";
pipe_t *newpipe = NULL;
if (options.immediate != 1 || subscribe_to_all) {
// Create a bi-directional pipe.
object_t *parents [2] = {this, session};
pipe_t *new_pipes [2] = {NULL, NULL};
bool conflate = options.conflate &&
(options.type == ZMQ_DEALER ||
options.type == ZMQ_PULL ||
options.type == ZMQ_PUSH ||
options.type == ZMQ_PUB ||
options.type == ZMQ_SUB);
int hwms [2] = {conflate? -1 : options.sndhwm,
conflate? -1 : options.rcvhwm};
bool conflates [2] = {conflate, conflate};
rc = pipepair (parents, new_pipes, hwms, conflates);
errno_assert (rc == 0);
// Attach local end of the pipe to the socket object.
attach_pipe (new_pipes [0], subscribe_to_all);
newpipe = new_pipes [0];
// Attach remote end of the pipe to the session object later on.
session->attach_pipe (new_pipes [1]);
}
// Save last endpoint URI
paddr->to_string (last_endpoint);
add_endpoint (addr_, (own_t *) session, newpipe);
EXIT_MUTEX();
return 0;
}
bool zmq::mtrie_t::rm_helper (unsigned char *prefix_, size_t size_,
pipe_t *pipe_)
{
if (!size_) {
if (pipes) {
pipes_t::size_type erased = pipes->erase (pipe_);
zmq_assert (erased == 1);
if (pipes->empty ()) {
LIBZMQ_DELETE(pipes);
}
}
return !pipes;
}
unsigned char c = *prefix_;
if (!count || c < min || c >= min + count)
return false;
mtrie_t *next_node =
count == 1 ? next.node : next.table [c - min];
if (!next_node)
return false;
bool ret = next_node->rm_helper (prefix_ + 1, size_ - 1, pipe_);
if (next_node->is_redundant ()) {
LIBZMQ_DELETE(next_node);
zmq_assert (count > 0);
if (count == 1) {
next.node = 0;
count = 0;
--live_nodes;
zmq_assert (live_nodes == 0);
}
else {
next.table [c - min] = 0;
zmq_assert (live_nodes > 1);
--live_nodes;
// Compact the table if possible
if (live_nodes == 1) {
// If there's only one live node in the table we can
// switch to using the more compact single-node
// representation
unsigned short i;
for (i = 0; i < count; ++i)
if (next.table [i])
break;
zmq_assert (i < count);
min += i;
count = 1;
mtrie_t *oldp = next.table [i];
free (next.table);
next.node = oldp;
}
else
if (c == min) {
// We can compact the table "from the left"
unsigned short i;
for (i = 1; i < count; ++i)
if (next.table [i])
break;
zmq_assert (i < count);
min += i;
count -= i;
mtrie_t **old_table = next.table;
next.table = (mtrie_t**) malloc (sizeof (mtrie_t*) * count);
alloc_assert (next.table);
memmove (next.table, old_table + i, sizeof (mtrie_t*) * count);
free (old_table);
}
else
if (c == min + count - 1) {
// We can compact the table "from the right"
unsigned short i;
for (i = 1; i < count; ++i)
if (next.table [count - 1 - i])
break;
zmq_assert (i < count);
count -= i;
mtrie_t **old_table = next.table;
next.table = (mtrie_t**) malloc (sizeof (mtrie_t*) * count);
alloc_assert (next.table);
memmove (next.table, old_table, sizeof (mtrie_t*) * count);
free (old_table);
}
}
}
return ret;
}
zmq::socket_base_t *zmq::socket_base_t::create (int type_, class ctx_t *parent_,
uint32_t tid_, int sid_)
{
socket_base_t *s = NULL;
switch (type_) {
case ZMQ_PAIR:
s = new (std::nothrow) pair_t (parent_, tid_, sid_);
break;
case ZMQ_PUB:
s = new (std::nothrow) pub_t (parent_, tid_, sid_);
break;
case ZMQ_SUB:
s = new (std::nothrow) sub_t (parent_, tid_, sid_);
break;
case ZMQ_REQ:
s = new (std::nothrow) req_t (parent_, tid_, sid_);
break;
case ZMQ_REP:
s = new (std::nothrow) rep_t (parent_, tid_, sid_);
break;
case ZMQ_DEALER:
s = new (std::nothrow) dealer_t (parent_, tid_, sid_);
break;
case ZMQ_ROUTER:
s = new (std::nothrow) router_t (parent_, tid_, sid_);
break;
case ZMQ_PULL:
s = new (std::nothrow) pull_t (parent_, tid_, sid_);
break;
case ZMQ_PUSH:
s = new (std::nothrow) push_t (parent_, tid_, sid_);
break;
case ZMQ_XPUB:
s = new (std::nothrow) xpub_t (parent_, tid_, sid_);
break;
case ZMQ_XSUB:
s = new (std::nothrow) xsub_t (parent_, tid_, sid_);
break;
case ZMQ_STREAM:
s = new (std::nothrow) stream_t (parent_, tid_, sid_);
break;
case ZMQ_SERVER:
s = new (std::nothrow) server_t (parent_, tid_, sid_);
break;
case ZMQ_CLIENT:
s = new (std::nothrow) client_t (parent_, tid_, sid_);
break;
default:
errno = EINVAL;
return NULL;
}
alloc_assert (s);
mailbox_t *mailbox = dynamic_cast<mailbox_t*> (s->mailbox);
if (mailbox != NULL && mailbox->get_fd () == retired_fd)
return NULL;
return s;
}
int zmq::tcp_connecter_t::open ()
{
zmq_assert (s == retired_fd);
// Resolve the address
if (addr->resolved.tcp_addr != NULL) {
LIBZMQ_DELETE(addr->resolved.tcp_addr);
}
addr->resolved.tcp_addr = new (std::nothrow) tcp_address_t ();
alloc_assert (addr->resolved.tcp_addr);
int rc = addr->resolved.tcp_addr->resolve (
addr->address.c_str (), false, options.ipv6);
if (rc != 0) {
LIBZMQ_DELETE(addr->resolved.tcp_addr);
return -1;
}
zmq_assert (addr->resolved.tcp_addr != NULL);
tcp_address_t * const tcp_addr = addr->resolved.tcp_addr;
// Create the socket.
s = open_socket (tcp_addr->family (), SOCK_STREAM, IPPROTO_TCP);
#ifdef ZMQ_HAVE_WINDOWS
if (s == INVALID_SOCKET) {
errno = wsa_error_to_errno (WSAGetLastError ());
return -1;
}
#else
if (s == -1)
return -1;
#endif
// On some systems, IPv4 mapping in IPv6 sockets is disabled by default.
// Switch it on in such cases.
if (tcp_addr->family () == AF_INET6)
enable_ipv4_mapping (s);
// Set the IP Type-Of-Service priority for this socket
if (options.tos != 0)
set_ip_type_of_service (s, options.tos);
// Set the socket to non-blocking mode so that we get async connect().
unblock_socket (s);
// Set the socket buffer limits for the underlying socket.
if (options.sndbuf >= 0)
set_tcp_send_buffer (s, options.sndbuf);
if (options.rcvbuf >= 0)
set_tcp_receive_buffer (s, options.rcvbuf);
// Set the IP Type-Of-Service for the underlying socket
if (options.tos != 0)
set_ip_type_of_service (s, options.tos);
// Set a source address for conversations
if (tcp_addr->has_src_addr ()) {
rc = ::bind (s, tcp_addr->src_addr (), tcp_addr->src_addrlen ());
if (rc == -1)
return -1;
}
// Connect to the remote peer.
rc = ::connect (s, tcp_addr->addr (), tcp_addr->addrlen ());
// Connect was successful immediately.
if (rc == 0)
return 0;
// Translate error codes indicating asynchronous connect has been
// launched to a uniform EINPROGRESS.
#ifdef ZMQ_HAVE_WINDOWS
const int last_error = WSAGetLastError();
if (last_error == WSAEINPROGRESS || last_error == WSAEWOULDBLOCK)
errno = EINPROGRESS;
else
errno = wsa_error_to_errno (last_error);
#else
if (errno == EINTR)
errno = EINPROGRESS;
#endif
return -1;
}
void *zmq_atomic_counter_new (void)
{
zmq::atomic_counter_t *counter = new (std::nothrow) zmq::atomic_counter_t;
alloc_assert (counter);
return counter;
}
void zmq::session_base_t::start_connecting (bool wait_)
{
zmq_assert (active);
// Choose I/O thread to run connecter in. Given that we are already
// running in an I/O thread, there must be at least one available.
io_thread_t *io_thread = choose_io_thread (options.affinity);
zmq_assert (io_thread);
// Create the connecter object.
if (addr->protocol == "tcp") {
if (!options.socks_proxy_address.empty()) {
address_t *proxy_address = new (std::nothrow)
address_t ("tcp", options.socks_proxy_address, this->get_ctx ());
alloc_assert (proxy_address);
socks_connecter_t *connecter =
new (std::nothrow) socks_connecter_t (
io_thread, this, options, addr, proxy_address, wait_);
alloc_assert (connecter);
launch_child (connecter);
}
else {
tcp_connecter_t *connecter = new (std::nothrow)
tcp_connecter_t (io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
}
return;
}
#if !defined ZMQ_HAVE_WINDOWS && !defined ZMQ_HAVE_OPENVMS
if (addr->protocol == "ipc") {
ipc_connecter_t *connecter = new (std::nothrow) ipc_connecter_t (
io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
return;
}
#endif
#if defined ZMQ_HAVE_TIPC
if (addr->protocol == "tipc") {
tipc_connecter_t *connecter = new (std::nothrow) tipc_connecter_t (
io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
return;
}
#endif
if (addr->protocol == "udp") {
zmq_assert (options.type == ZMQ_DISH || options.type == ZMQ_RADIO);
udp_engine_t* engine = new (std::nothrow) udp_engine_t ();
alloc_assert (engine);
bool recv = false;
bool send = false;
if (options.type == ZMQ_RADIO) {
send = true;
recv = false;
}
else if (options.type == ZMQ_DISH) {
send = false;
recv = true;
}
int rc = engine->init (addr, send, recv);
errno_assert (rc == 0);
send_attach (this, engine);
return;
}
#ifdef ZMQ_HAVE_OPENPGM
// Both PGM and EPGM transports are using the same infrastructure.
if (addr->protocol == "pgm" || addr->protocol == "epgm") {
zmq_assert (options.type == ZMQ_PUB || options.type == ZMQ_XPUB
|| options.type == ZMQ_SUB || options.type == ZMQ_XSUB);
// For EPGM transport with UDP encapsulation of PGM is used.
bool const udp_encapsulation = addr->protocol == "epgm";
// At this point we'll create message pipes to the session straight
// away. There's no point in delaying it as no concept of 'connect'
// exists with PGM anyway.
if (options.type == ZMQ_PUB || options.type == ZMQ_XPUB) {
// PGM sender.
pgm_sender_t *pgm_sender = new (std::nothrow) pgm_sender_t (
io_thread, options);
alloc_assert (pgm_sender);
int rc = pgm_sender->init (udp_encapsulation, addr->address.c_str ());
errno_assert (rc == 0);
send_attach (this, pgm_sender);
}
else {
// PGM receiver.
pgm_receiver_t *pgm_receiver = new (std::nothrow) pgm_receiver_t (
io_thread, options);
alloc_assert (pgm_receiver);
int rc = pgm_receiver->init (udp_encapsulation, addr->address.c_str ());
errno_assert (rc == 0);
send_attach (this, pgm_receiver);
}
return;
}
#endif
#ifdef ZMQ_HAVE_NORM
if (addr->protocol == "norm") {
// At this point we'll create message pipes to the session straight
// away. There's no point in delaying it as no concept of 'connect'
// exists with NORM anyway.
if (options.type == ZMQ_PUB || options.type == ZMQ_XPUB) {
// NORM sender.
norm_engine_t* norm_sender = new (std::nothrow) norm_engine_t(io_thread, options);
alloc_assert (norm_sender);
int rc = norm_sender->init (addr->address.c_str (), true, false);
errno_assert (rc == 0);
send_attach (this, norm_sender);
}
else { // ZMQ_SUB or ZMQ_XSUB
// NORM receiver.
norm_engine_t* norm_receiver = new (std::nothrow) norm_engine_t (io_thread, options);
alloc_assert (norm_receiver);
int rc = norm_receiver->init (addr->address.c_str (), false, true);
errno_assert (rc == 0);
send_attach (this, norm_receiver);
}
return;
}
#endif // ZMQ_HAVE_NORM
#if defined ZMQ_HAVE_VMCI
if (addr->protocol == "vmci") {
vmci_connecter_t *connecter = new (std::nothrow) vmci_connecter_t (
io_thread, this, options, addr, wait_);
alloc_assert (connecter);
launch_child (connecter);
return;
}
#endif
zmq_assert (false);
}
int testmsg()
{
int rc;
int sb;
int sc;
unsigned char *buf1, *buf2;
int i;
struct nn_iovec iov;
struct nn_msghdr hdr;
printf("test msg\n");
if ( 1 )
{
sb = test_socket (AF_SP, NN_PAIR);
test_bind (sb, SOCKET_ADDRESS);
sc = test_socket (AF_SP, NN_PAIR);
test_connect (sc, SOCKET_ADDRESS);
buf1 = nn_allocmsg (256, 0);
alloc_assert (buf1);
for (i = 0; i != 256; ++i)
buf1 [i] = (unsigned char) i;
printf("send 256\n");
rc = nn_send (sc, &buf1, NN_MSG, 0);
printf("rc.%d\n",rc);
errno_assert (rc >= 0);
nn_assert (rc == 256);
buf2 = NULL;
rc = nn_recv (sb, &buf2, NN_MSG, 0);
errno_assert (rc >= 0);
nn_assert (rc == 256);
nn_assert (buf2);
for (i = 0; i != 256; ++i)
nn_assert (buf2 [i] == (unsigned char) i);
rc = nn_freemsg (buf2);
errno_assert (rc == 0);
buf1 = nn_allocmsg (256, 0);
alloc_assert (buf1);
for (i = 0; i != 256; ++i)
buf1 [i] = (unsigned char) i;
iov.iov_base = &buf1;
iov.iov_len = NN_MSG;
memset (&hdr, 0, sizeof (hdr));
hdr.msg_iov = &iov;
hdr.msg_iovlen = 1;
rc = nn_sendmsg (sc, &hdr, 0);
errno_assert (rc >= 0);
nn_assert (rc == 256);
buf2 = NULL;
iov.iov_base = &buf2;
iov.iov_len = NN_MSG;
memset (&hdr, 0, sizeof (hdr));
hdr.msg_iov = &iov;
hdr.msg_iovlen = 1;
rc = nn_recvmsg (sb, &hdr, 0);
errno_assert (rc >= 0);
nn_assert (rc == 256);
nn_assert (buf2);
for (i = 0; i != 256; ++i)
nn_assert (buf2 [i] == (unsigned char) i);
rc = nn_freemsg (buf2);
errno_assert (rc == 0);
test_close (sc);
test_close (sb);
}
/* Test receiving of large message */
sb = test_socket(AF_SP, NN_PAIR);
//printf("test_bind.(%s)\n",SOCKET_ADDRESS_TCP);
test_bind(sb,SOCKET_ADDRESS_TCP);
sc = test_socket(AF_SP,NN_PAIR);
//printf("test_connect.(%s)\n",SOCKET_ADDRESS_TCP);
test_connect(sc,SOCKET_ADDRESS_TCP);
for (i = 0; i < (int) sizeof (longdata); ++i)
longdata[i] = '0' + (i % 10);
longdata [sizeof(longdata) - 1] = 0;
printf("send longdata.%d\n",(int32_t)sizeof(longdata));
test_send(sb,longdata);
printf("recv longdata.%d\n",(int32_t)sizeof(longdata));
rc = nn_recv (sc, &buf2, NN_MSG, 0);
errno_assert (rc >= 0);
nn_assert (rc == sizeof (longdata) - 1);
nn_assert (buf2);
for (i = 0; i < (int) sizeof (longdata) - 1; ++i)
nn_assert (buf2 [i] == longdata [i]);
rc = nn_freemsg (buf2);
errno_assert (rc == 0);
test_close (sc);
test_close (sb);
//printf("testmsg completed\n");
return 0;
}
zmq::socket_base_t *zmq::ctx_t::create_socket (int type_)
{
scoped_lock_t locker(slot_sync);
if (unlikely (starting)) {
starting = false;
// Initialise the array of mailboxes. Additional three slots are for
// zmq_ctx_term thread and reaper thread.
opt_sync.lock ();
int mazmq = max_sockets;
int ios = io_thread_count;
opt_sync.unlock ();
slot_count = mazmq + ios + 2;
slots = (i_mailbox **) malloc (sizeof (i_mailbox*) * slot_count);
alloc_assert (slots);
// Initialise the infrastructure for zmq_ctx_term thread.
slots [term_tid] = &term_mailbox;
// Create the reaper thread.
reaper = new (std::nothrow) reaper_t (this, reaper_tid);
alloc_assert (reaper);
slots [reaper_tid] = reaper->get_mailbox ();
reaper->start ();
// Create I/O thread objects and launch them.
for (int i = 2; i != ios + 2; i++) {
io_thread_t *io_thread = new (std::nothrow) io_thread_t (this, i);
alloc_assert (io_thread);
io_threads.push_back (io_thread);
slots [i] = io_thread->get_mailbox ();
io_thread->start ();
}
// In the unused part of the slot array, create a list of empty slots.
for (int32_t i = (int32_t) slot_count - 1;
i >= (int32_t) ios + 2; i--) {
empty_slots.push_back (i);
slots [i] = NULL;
}
}
// Once zmq_ctx_term() was called, we can't create new sockets.
if (terminating) {
errno = ETERM;
return NULL;
}
// If max_sockets limit was reached, return error.
if (empty_slots.empty ()) {
errno = EMFILE;
return NULL;
}
// Choose a slot for the socket.
uint32_t slot = empty_slots.back ();
empty_slots.pop_back ();
// Generate new unique socket ID.
int sid = ((int) max_socket_id.add (1)) + 1;
// Create the socket and register its mailbox.
socket_base_t *s = socket_base_t::create (type_, this, slot, sid);
if (!s) {
empty_slots.push_back (slot);
return NULL;
}
sockets.push_back (s);
slots [slot] = s->get_mailbox ();
return s;
}
inline int zmq_poller_poll (zmq_pollitem_t *items_, int nitems_, long timeout_)
{
// implement zmq_poll on top of zmq_poller
int rc;
zmq_poller_event_t *events;
events = new zmq_poller_event_t[nitems_];
alloc_assert(events);
void *poller = zmq_poller_new ();
alloc_assert(poller);
bool repeat_items = false;
// Register sockets with poller
for (int i = 0; i < nitems_; i++) {
items_[i].revents = 0;
bool modify = false;
short e = items_[i].events;
if (items_[i].socket) {
// Poll item is a 0MQ socket.
for (int j = 0; j < i; ++j) {
// Check for repeat entries
if (items_[j].socket == items_[i].socket) {
repeat_items = true;
modify = true;
e |= items_[j].events;
}
}
if (modify) {
rc = zmq_poller_modify (poller, items_[i].socket, e);
} else {
rc = zmq_poller_add (poller, items_[i].socket, NULL, e);
}
if (rc < 0) {
zmq_poller_destroy (&poller);
delete [] events;
return rc;
}
} else {
// Poll item is a raw file descriptor.
for (int j = 0; j < i; ++j) {
// Check for repeat entries
if (!items_[j].socket && items_[j].fd == items_[i].fd) {
repeat_items = true;
modify = true;
e |= items_[j].events;
}
}
if (modify) {
rc = zmq_poller_modify_fd (poller, items_[i].fd, e);
} else {
rc = zmq_poller_add_fd (poller, items_[i].fd, NULL, e);
}
if (rc < 0) {
zmq_poller_destroy (&poller);
delete [] events;
return rc;
}
}
}
// Wait for events
rc = zmq_poller_wait_all (poller, events, nitems_, timeout_);
if (rc < 0) {
zmq_poller_destroy (&poller);
delete [] events;
if (zmq_errno() == ETIMEDOUT) {
return 0;
}
return rc;
}
// Transform poller events into zmq_pollitem events.
// items_ contains all items, while events only contains fired events.
// If no sockets are repeated (likely), the two are still co-ordered, so step through the items
// checking for matches only on the first event.
// If there are repeat items, they cannot be assumed to be co-ordered,
// so each pollitem must check fired events from the beginning.
int j_start = 0, found_events = rc;
for (int i = 0; i < nitems_; i++) {
for (int j = j_start; j < found_events; ++j) {
if (
(items_[i].socket && items_[i].socket == events[j].socket) ||
(!(items_[i].socket || items_[j].socket) && items_[i].fd == events[j].fd)
) {
items_[i].revents = events[j].events & items_[i].events;
if (!repeat_items) {
// no repeats, we can ignore events we've already seen
j_start++;
}
break;
}
if (!repeat_items) {
// no repeats, never have to look at j > j_start
break;
}
}
}
// Cleanup
zmq_poller_destroy (&poller);
delete [] events;
return rc;
}
