static int
test_misc(void)
{
char memdump[] = "memdump_test";
if (rte_bsf32(129))
FAIL("rte_bsf32");
rte_memdump(stdout, "test", memdump, sizeof(memdump));
rte_hexdump(stdout, "test", memdump, sizeof(memdump));
rte_pause();
return 0;
}
static int
usock_mbuf_write(struct vr_usocket *usockp, struct rte_mbuf *mbuf)
{
unsigned int i, pkt_len;
struct msghdr mhdr;
struct rte_mbuf *m;
struct iovec *iov;
if (!mbuf)
return 0;
pkt_len = rte_pktmbuf_pkt_len(mbuf);
if (!pkt_len)
return 0;
iov = usockp->usock_iovec;
m = mbuf;
for (i = 0; (m && (i < PKT0_MAX_IOV_LEN)); i++) {
iov->iov_base = rte_pktmbuf_mtod(m, unsigned char *);
iov->iov_len = rte_pktmbuf_data_len(m);
m = m->next;
iov++;
}
if ((i == PKT0_MAX_IOV_LEN) && m)
usockp->usock_pkt_truncated++;
mhdr.msg_name = NULL;
mhdr.msg_namelen = 0;
mhdr.msg_iov = usockp->usock_iovec;
mhdr.msg_iovlen = i;
mhdr.msg_control = NULL;
mhdr.msg_controllen = 0;
mhdr.msg_flags = 0;
#ifdef VR_DPDK_USOCK_DUMP
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d sending message\n", __func__,
pthread_self(), usockp->usock_fd);
rte_hexdump(stdout, "usock message dump:", &mhdr, sizeof(mhdr));
#endif
return sendmsg(usockp->usock_fd, &mhdr, MSG_DONTWAIT);
}
static int
vr_usocket_connect(struct vr_usocket *usockp)
{
struct sockaddr_un sun;
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d\n", __func__, pthread_self(),
usockp->usock_fd);
if (usockp->usock_proto != PACKET)
return -EINVAL;
sun.sun_family = AF_UNIX;
memset(sun.sun_path, 0, sizeof(sun.sun_path));
strncpy(sun.sun_path, VR_PACKET_AGENT_UNIX_FILE, sizeof(sun.sun_path) - 1);
#ifdef VR_DPDK_USOCK_DUMP
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d retry connecting\n", __func__,
pthread_self(), usockp->usock_fd);
rte_hexdump(stdout, "usock address dump:", &sun, sizeof(sun));
#endif
return vr_dpdk_retry_connect(usockp->usock_fd, (struct sockaddr *)&sun,
sizeof(sun));
}
static int
__usock_read(struct vr_usocket *usockp)
{
int ret;
unsigned int offset = usockp->usock_read_offset;
unsigned int len = usockp->usock_read_len;
unsigned int toread = len - offset;
struct nlmsghdr *nlh;
unsigned int proto = usockp->usock_proto;
char *buf = usockp->usock_rx_buf;
if (toread > usockp->usock_buf_len) {
toread = usockp->usock_buf_len - offset;
}
retry_read:
if (usockp->usock_owner != pthread_self()) {
if (usockp->usock_owner)
RTE_LOG(WARNING, USOCK, "WARNING: thread %lx is trying to read"
" usocket FD %d owned by thread %lx\n",
pthread_self(), usockp->usock_fd, usockp->usock_owner);
usockp->usock_owner = pthread_self();
}
ret = read(usockp->usock_fd, buf + offset, toread);
#ifdef VR_DPDK_USOCK_DUMP
if (ret > 0) {
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d read %d bytes\n", __func__,
pthread_self(), usockp->usock_fd, ret);
rte_hexdump(stdout, "usock buffer dump:", buf + offset, ret);
} else if (ret < 0) {
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d read returned error %d: %s (%d)\n", __func__,
pthread_self(), usockp->usock_fd, ret, rte_strerror(errno), errno);
}
#endif
if (ret <= 0) {
if (!ret)
return -1;
if (errno == EINTR)
goto retry_read;
if ((errno == EAGAIN) ||
(errno == EWOULDBLOCK))
return 0;
RTE_LOG(ERR, USOCK, "Error reading FD %d: %s (%d)\n",
usockp->usock_fd, rte_strerror(errno), errno);
return ret;
}
offset += ret;
usockp->usock_read_offset = offset;
if (proto == NETLINK) {
if (usockp->usock_state == READING_HEADER) {
if (usockp->usock_read_offset == usockp->usock_read_len) {
usockp->usock_state = READING_DATA;
nlh = (struct nlmsghdr *)(usockp->usock_rx_buf);
usockp->usock_read_len = nlh->nlmsg_len;
}
}
if (usockp->usock_buf_len < usockp->usock_read_len) {
usockp->usock_rx_buf = vr_malloc(usockp->usock_read_len,
VR_USOCK_BUF_OBJECT);
if (!usockp->usock_rx_buf) {
/* bad, but let's recover */
usockp->usock_rx_buf = buf;
usockp->usock_read_len -= usockp->usock_read_offset;
usockp->usock_read_offset = 0;
usockp->usock_state = READING_FAULTY_DATA;
} else {
memcpy(usockp->usock_rx_buf, buf, usockp->usock_read_offset);
vr_free(buf, VR_USOCK_BUF_OBJECT);
usockp->usock_buf_len = usockp->usock_read_len;
buf = usockp->usock_rx_buf;
}
}
} else if (proto == PACKET) {
usockp->usock_read_len = ret;
}
return ret;
}
static int
__usock_write(struct vr_usocket *usockp)
{
int ret;
unsigned int len;
unsigned char *buf;
struct vr_usocket *parent = NULL;
if (usockp->usock_proto != EVENT) {
parent = usockp->usock_parent;
if (!parent)
return -1;
}
buf = usockp->usock_tx_buf;
if (!buf || !usockp->usock_write_len)
return 0;
len = usockp->usock_write_len;
buf += usockp->usock_write_offset;
len -= usockp->usock_write_offset;
retry_write:
#ifdef VR_DPDK_USOCK_DUMP
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d writing %d bytes\n",
__func__, pthread_self(), usockp->usock_fd, len);
rte_hexdump(stdout, "usock buffer dump:", buf, len);
#endif
if (usockp->usock_owner != pthread_self()) {
if (usockp->usock_owner)
RTE_LOG(WARNING, USOCK, "WARNING: thread %lx (lcore %u) is trying to write %u bytes"
" to usocket FD %d owned by thread %lx\n",
pthread_self(), rte_lcore_id(), len, usockp->usock_fd, usockp->usock_owner);
usockp->usock_owner = pthread_self();
}
ret = write(usockp->usock_fd, buf, len);
if (ret > 0) {
usockp->usock_write_offset += ret;
if (usockp->usock_write_offset == usockp->usock_write_len) {
/* remove from output poll */
if (parent)
parent->usock_pfds[usockp->usock_child_index].events = POLLIN;
usockp->usock_tx_buf = NULL;
} else {
if (parent)
parent->usock_pfds[usockp->usock_child_index].events = POLLOUT;
}
} else if (ret < 0) {
RTE_LOG(DEBUG, VROUTER, "%s[%lx]: write error FD %d\n", __func__, pthread_self(),
usockp->usock_fd);
usock_set_error(usockp, ret);
if (errno == EINTR)
goto retry_write;
if ((errno == EAGAIN) || (errno == EWOULDBLOCK)) {
if (parent) {
parent->usock_pfds[usockp->usock_child_index].events = POLLOUT;
return 0;
}
}
usockp->usock_tx_buf = NULL;
}
return ret;
}
static int
vr_usocket_bind(struct vr_usocket *usockp)
{
int error = 0;
struct sockaddr_in sin;
struct sockaddr_un sun;
struct sockaddr *addr = NULL;
socklen_t addrlen = 0;
int optval;
bool server;
optval = 1;
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d setting option\n", __func__,
pthread_self(), usockp->usock_fd);
if (setsockopt(usockp->usock_fd, SOL_SOCKET, SO_REUSEADDR, &optval,
sizeof(optval)))
return -errno;
switch (usockp->usock_type) {
case TCP:
sin.sin_family = AF_INET;
sin.sin_port = rte_cpu_to_be_16(VR_NETLINK_TCP_PORT);
sin.sin_addr.s_addr = INADDR_ANY;
addr = (struct sockaddr *)&sin;
addrlen = sizeof(sin);
server = true;
break;
case UNIX:
sun.sun_family = AF_UNIX;
memset(sun.sun_path, 0, sizeof(sun.sun_path));
strncpy(sun.sun_path, VR_NETLINK_UNIX_FILE, sizeof(sun.sun_path) - 1);
addr = (struct sockaddr *)&sun;
addrlen = sizeof(sun);
server = true;
mkdir(VR_SOCKET_DIR, VR_SOCKET_DIR_MODE);
unlink(sun.sun_path);
break;
case RAW:
sun.sun_family = AF_UNIX;
memset(sun.sun_path, 0, sizeof(sun.sun_path));
strncpy(sun.sun_path, VR_PACKET_UNIX_FILE, sizeof(sun.sun_path) - 1);
addr = (struct sockaddr *)&sun;
addrlen = sizeof(sun);
server = false;
mkdir(VR_SOCKET_DIR, VR_SOCKET_DIR_MODE);
unlink(sun.sun_path);
break;
default:
return -EINVAL;
}
#ifdef VR_DPDK_USOCK_DUMP
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d binding\n", __func__, pthread_self(),
usockp->usock_fd);
rte_hexdump(stdout, "usock address dump:", addr, addrlen);
#endif
error = bind(usockp->usock_fd, addr, addrlen);
if (error < 0)
return error;
if (server) {
RTE_LOG(DEBUG, USOCK, "%s[%lx]: FD %d listening\n", __func__,
pthread_self(), usockp->usock_fd);
error = listen(usockp->usock_fd, 1);
if (error < 0)
return error;
usockp->usock_state = LISTENING;
}
return 0;
}
/*
* vr_uvh_cl_msg_handler - handler for messages from user space vhost
* clients. Calls the appropriate handler based on the message type.
*
* Returns 0 on success, -1 on error.
*
* TODO: upon error, this function currently makes the process exit.
* Instead, it should close the socket and continue serving other clients.
*/
static int
vr_uvh_cl_msg_handler(int fd, void *arg)
{
vr_uvh_client_t *vru_cl = (vr_uvh_client_t *) arg;
struct msghdr mhdr;
struct iovec iov;
int i, err, ret = 0, read_len = 0;
struct cmsghdr *cmsg;
memset(&mhdr, 0, sizeof(mhdr));
if (vru_cl->vruc_msg_bytes_read == 0) {
mhdr.msg_control = &vru_cl->vruc_cmsg;
mhdr.msg_controllen = sizeof(vru_cl->vruc_cmsg);
iov.iov_base = (void *) &vru_cl->vruc_msg;
iov.iov_len = VHOST_USER_HSIZE;
mhdr.msg_iov = &iov;
mhdr.msg_iovlen = 1;
ret = recvmsg(fd, &mhdr, MSG_DONTWAIT);
if (ret < 0) {
if ((errno == EAGAIN) || (errno == EWOULDBLOCK)) {
ret = 0;
goto cleanup;
}
vr_uvhost_log("Receive returned %d in vhost server for client %s\n",
ret, vru_cl->vruc_path);
ret = -1;
goto cleanup;
} else if (ret > 0) {
if (mhdr.msg_flags & MSG_CTRUNC) {
vr_uvhost_log("Truncated control message from vhost client %s\n",
vru_cl->vruc_path);
ret = -1;
goto cleanup;
}
cmsg = CMSG_FIRSTHDR(&mhdr);
if (cmsg && (cmsg->cmsg_len > 0) &&
(cmsg->cmsg_level == SOL_SOCKET) &&
(cmsg->cmsg_type == SCM_RIGHTS)) {
vru_cl->vruc_num_fds_sent = (cmsg->cmsg_len - CMSG_LEN(0))/
sizeof(int);
if (vru_cl->vruc_num_fds_sent > VHOST_MEMORY_MAX_NREGIONS) {
vr_uvhost_log("Too many FDs sent for client %s: %d\n",
vru_cl->vruc_path, vru_cl->vruc_num_fds_sent);
vru_cl->vruc_num_fds_sent = VHOST_MEMORY_MAX_NREGIONS;
}
memcpy(vru_cl->vruc_fds_sent, CMSG_DATA(cmsg),
vru_cl->vruc_num_fds_sent*sizeof(int));
}
vru_cl->vruc_msg_bytes_read = ret;
if (ret < VHOST_USER_HSIZE) {
ret = 0;
goto cleanup;
}
read_len = vru_cl->vruc_msg.size;
} else {
/*
* recvmsg returned 0, so return error.
*/
vr_uvhost_log("Receive returned %d in vhost server for client %s\n",
ret, vru_cl->vruc_path);
ret = -1;
goto cleanup;
}
} else if (vru_cl->vruc_msg_bytes_read < VHOST_USER_HSIZE) {
read_len = VHOST_USER_HSIZE - vru_cl->vruc_msg_bytes_read;
} else {
read_len = vru_cl->vruc_msg.size -
(vru_cl->vruc_msg_bytes_read - VHOST_USER_HSIZE);
}
if (read_len) {
if (vru_cl->vruc_owner != pthread_self()) {
if (vru_cl->vruc_owner)
RTE_LOG(WARNING, UVHOST, "WARNING: thread %lx is trying to read"
" uvhost client FD %d owned by thread %lx\n",
pthread_self(), fd, vru_cl->vruc_owner);
vru_cl->vruc_owner = pthread_self();
}
ret = read(fd, (((char *)&vru_cl->vruc_msg) + vru_cl->vruc_msg_bytes_read),
read_len);
#ifdef VR_DPDK_RX_PKT_DUMP
if (ret > 0) {
RTE_LOG(DEBUG, UVHOST, "%s[%lx]: FD %d read %d bytes\n", __func__,
pthread_self(), fd, ret);
rte_hexdump(stdout, "uvhost full message dump:",
(((char *)&vru_cl->vruc_msg)),
ret + vru_cl->vruc_msg_bytes_read);
} else if (ret < 0) {
RTE_LOG(DEBUG, UVHOST, "%s[%lx]: FD %d read returned error %d: %s (%d)\n", __func__,
pthread_self(), fd, ret, rte_strerror(errno), errno);
}
#endif
if (ret < 0) {
if ((errno == EAGAIN) || (errno == EWOULDBLOCK)) {
ret = 0;
goto cleanup;
}
vr_uvhost_log(
"Error: read returned %d, %d %d %d in vhost server for client %s\n",
ret, errno, read_len,
vru_cl->vruc_msg_bytes_read, vru_cl->vruc_path);
ret = -1;
goto cleanup;
} else if (ret == 0) {
vr_uvhost_log("Read returned %d in vhost server for client %s\n",
ret, vru_cl->vruc_path);
ret = -1;
goto cleanup;
}
vru_cl->vruc_msg_bytes_read += ret;
if (vru_cl->vruc_msg_bytes_read < VHOST_USER_HSIZE) {
ret = 0;
goto cleanup;
}
if (vru_cl->vruc_msg_bytes_read <
(vru_cl->vruc_msg.size + VHOST_USER_HSIZE)) {
ret = 0;
goto cleanup;
}
}
ret = vr_uvh_cl_call_handler(vru_cl);
if (ret < 0) {
vr_uvhost_log("Error handling message %d client %s\n",
vru_cl->vruc_msg.request, vru_cl->vruc_path);
ret = -1;
goto cleanup;
}
ret = vr_uvh_cl_send_reply(fd, vru_cl);
if (ret < 0) {
vr_uvhost_log("Error sending reply for message %d client %s\n",
vru_cl->vruc_msg.request, vru_cl->vruc_path);
ret = -1;
goto cleanup;
}
cleanup:
err = errno;
/* close all the FDs received */
for (i = 0; i < vru_cl->vruc_num_fds_sent; i++) {
if (vru_cl->vruc_fds_sent[i] > 0)
close(vru_cl->vruc_fds_sent[i]);
}
if (ret == -1) {
/* set VQ_NOT_READY state to vif's queues. */
for (i = 0; i < VR_DPDK_VIRTIO_MAX_QUEUES; i++) {
vr_dpdk_virtio_rxqs[vru_cl->vruc_idx][i].vdv_ready_state = VQ_NOT_READY;
vr_dpdk_virtio_txqs[vru_cl->vruc_idx][i].vdv_ready_state = VQ_NOT_READY;
}
rte_wmb();
synchronize_rcu();
/*
* Unmaps qemu's FDs.
*/
vr_dpdk_virtio_uvh_vif_munmap(&vr_dpdk_virtio_uvh_vif_mmap[vru_cl->vruc_idx]);
}
/* clear state for next message from this client. */
vru_cl->vruc_msg_bytes_read = 0;
memset(&vru_cl->vruc_msg, 0, sizeof(vru_cl->vruc_msg));
memset(vru_cl->vruc_cmsg, 0, sizeof(vru_cl->vruc_cmsg));
memset(vru_cl->vruc_fds_sent, 0, sizeof(vru_cl->vruc_fds_sent));
vru_cl->vruc_num_fds_sent = 0;
errno = err;
return ret;
}