pcap_t *
pcap_open_live(const char *device, int snaplen,
int promisc, int to_ms, char *errbuf)
{
struct nm_desc_t *d;
int l;
if (!device) {
D("missing device name");
return NULL;
}
l = strlen(device) + 1;
D("request to open %s snaplen %d promisc %d timeout %dms",
device, snaplen, promisc, to_ms);
d = nm_open(device, NULL, 0, 0);
if (d == NULL) {
D("error opening %s", device);
return NULL;
}
d->to_ms = to_ms;
d->snaplen = snaplen;
d->errbuf = errbuf;
d->promisc = promisc;
return d;
}
/* Netmap.__init__(), may be called many times, or not called at all. */
static int
NetmapDesc_init(NetmapDesc *self, PyObject *args, PyObject *kwds)
{
PyObject *dev_name = NULL;
static char *kwlist[] = {"ifname", "flags", NULL};
const char *ifname;
unsigned long flags = 0;
int ret;
if (!PyArg_ParseTupleAndKeywords(args, kwds, "s|k", kwlist,
&ifname, &flags)) {
return -1;
}
/* Open the netmap device and register an interface. */
self->nmd = nm_open(ifname, NULL, flags, NULL);
if (self->nmd == NULL) {
PyErr_SetString(NetmapError, "nm_open() failed");
return -1;
}
/* Setup the netmap memory layout. The +1 are here to take into account
the host rings. */
ret = NetmapMemory_setup(&self->memory, self->nmd->nifp,
self->nmd->req.nr_tx_rings + 1,
self->nmd->req.nr_rx_rings + 1);
return ret;
}
struct pico_device *
pico_netmap_create(char *interface, char *name, uint8_t *mac) {
struct pico_device_netmap *netmap;
char ifname[IFNAMSIZ + 7];
netmap = PICO_ZALLOC(sizeof(struct pico_device_netmap));
if (!netmap) {
return NULL;
}
if (pico_device_init((struct pico_device *)netmap, name, mac)) {
pico_netmap_destroy((struct pico_device *)netmap);
return NULL;
}
sprintf(ifname, "netmap:%s", interface);
netmap->dev.overhead = 0;
netmap->conn = nm_open(ifname, NULL, 0, 0);
if (! netmap->conn) {
pico_netmap_destroy((struct pico_device *)netmap);
return NULL;
}
netmap->dev.send = pico_netmap_send;
netmap->dev.poll = pico_netmap_poll;
netmap->dev.destroy = pico_netmap_destroy;
return (struct pico_device *) netmap;
}
int main(int argc, char *argv[]) {
if (argc < 3) {
fprintf(stderr, "Usage: %s [interface] [RX ring number]\n", argv[0]);
exit(1);
}
struct sigaction sa;
// Setup the sighub handler
sa.sa_handler = &handle_signal;
// Restart the system call, if at all possible
sa.sa_flags = SA_RESTART;
// Block every signal during the handler
sigfillset(&sa.sa_mask);
// Intercept SIGHUP and SIGINT
if (sigaction(SIGHUP, &sa, NULL) == -1) {
perror("Error: cannot handle SIGHUP"); // Should not happen
}
L = luaL_newstate();
luaL_openlibs(L); /* Load Lua libraries */
// Try loading the file containing the script and run it
if ( luaL_loadfile(L, "script.lua") || lua_pcall(L, 0, 0, 0) ) {
fprintf(stderr, "Couldn't load file: %s\n", lua_tostring(L, -1));
exit(1);
}
lua_getglobal(L, "callback");
int cb_ref = luaL_ref(L, LUA_REGISTRYINDEX);
char netmap_ifname[IFNAMSIZ + 21];
const char *interface;
unsigned int ring_id;
struct nm_desc *d;
interface = argv[1];
ring_id = atoi(argv[2]);
snprintf(netmap_ifname, sizeof netmap_ifname, "netmap:%s-%d/R", interface, ring_id);
d = nm_open(netmap_ifname, NULL, 0, 0);
if (!d) {
perror("nm_open()");
exit(2);
}
printf("[+] Receiving packets on interface %s, RX ring %d\n", interface, ring_id);
receiver(L, cb_ref, d, ring_id);
lua_close(L);
return 0;
}
struct nm_desc*
usnet_init( struct nm_desc *gg_nmd, const char *dev_name, u_int flags)
{
struct nmreq nmr;
struct nm_desc *nmd = NULL;
struct netmap_if *nifp = NULL;
struct netmap_ring *txr, *rxr;
signal(SIGINT, sigint_h);
bzero(&nmr, sizeof(nmr));
strcpy(nmr.nr_name, dev_name);
// XXX: which netmap flags?
//nmr.nr_flags = NR_REG_ALL_NIC; //| flags;
printf("nm_open: %s\n", nmr.nr_name);
nmd = nm_open(nmr.nr_name, &nmr, 0, NULL);
if ( nmd == NULL ) {
DEBUG("Cannot open interface %s", nmr.nr_name);
exit(1);
}
nifp = nmd->nifp;
txr = NETMAP_TXRING(nifp, 0);
rxr = NETMAP_RXRING(nifp, 0);
printf("nmreq info, name=%s, version=%d,"
" flags=%d, memsize=%d,"
" ni_tx_rings=%d, ni_rx_rings=%d, num_tx_slots=%d, num_rx_slot=%d \n",
nifp->ni_name,
nifp->ni_version,
nifp->ni_flags,
nmd->memsize,
nifp->ni_tx_rings,
nifp->ni_rx_rings,
txr->num_slots,
rxr->num_slots);
memset(&g_config, 0, sizeof(g_config));
g_config.burst = 1000;
g_config.tx_rate = 0;
memset(&g_ipq, 0, sizeof(g_ipq));
usnet_init_internal();
usnet_route_init();
usnet_network_init();
usnet_udp_init();
usnet_ipv4_init();
usnet_socket_init();
return nmd;
}
/*----------------------------------------------------------------------------*/
void
netmap_init_handle(struct mtcp_thread_context *ctxt)
{
struct netmap_private_context *npc;
char ifname[MAX_IFNAMELEN];
char nifname[MAX_IFNAMELEN];
struct netdev_entry **ent;
int j;
ent = g_config.mos->netdev_table->ent;
/* create and initialize private I/O module context */
ctxt->io_private_context = calloc(1, sizeof(struct netmap_private_context));
if (ctxt->io_private_context == NULL) {
TRACE_ERROR("Failed to initialize ctxt->io_private_context: "
"Can't allocate memory\n");
exit(EXIT_FAILURE);
}
npc = (struct netmap_private_context *)ctxt->io_private_context;
/* initialize per-thread netmap interfaces */
for (j = 0; j < g_config.mos->netdev_table->num; j++) {
#if 0
if (if_indextoname(devices_attached[j], ifname) == NULL) {
TRACE_ERROR("Failed to initialize interface %s with ifidx: %d - "
"error string: %s\n",
ifname, devices_attached[j], strerror(errno));
exit(EXIT_FAILURE);
}
#else
strcpy(ifname, ent[j]->dev_name);
#endif
if (unlikely(g_config.mos->num_cores == 1))
sprintf(nifname, "netmap:%s", ifname);
else
sprintf(nifname, "netmap:%s-%d", ifname, ctxt->cpu);
TRACE_INFO("Opening %s with j: %d (cpu: %d)\n", nifname, j, ctxt->cpu);
struct nmreq base_nmd;
memset(&base_nmd, 0, sizeof(base_nmd));
base_nmd.nr_arg3 = EXTRA_BUFS;
npc->local_nmd[j] = nm_open(nifname, &base_nmd, 0, NULL);
if (npc->local_nmd[j] == NULL) {
TRACE_ERROR("Unable to open %s: %s\n",
nifname, strerror(errno));
exit(EXIT_FAILURE);
}
}
}
static int
thr_ctx_setup(struct thr_ctx *th, char *ifa, char *ifb,
int zerocopy, int burst, struct thr_ctx *parent, int cpuid)
{
int nm_flags = 0;
struct nm_desc *p_pa = NULL;
bzero(th, sizeof(struct thr_ctx));
th->ifa = strdup(ifa);
th->ifb = strdup(ifb);
th->zerocopy = zerocopy;
th->burst = burst;
th->cpuid = cpuid;
/* Setup netmap rings */
if (parent) {
nm_flags |= NM_OPEN_NO_MMAP;
p_pa = parent->pa;
}
th->pa = nm_open(ifa, NULL, nm_flags, p_pa);
if (th->pa == NULL) {
D("cannot open %s", ifa);
exit(1);
}
th->pb = nm_open(ifb, NULL, NM_OPEN_NO_MMAP, th->pa);
if (th->pb == NULL) {
D("cannot open %s", ifb);
nm_close(th->pa);
exit(1);
}
th->zerocopy = th->zerocopy && (th->pa->mem == th->pb->mem);
/* Done */
return (0);
}
/* Attempt to open port in netmap mode */
struct fp_device*
fp_netmap_open(const char* name)
{
/* Build the device. */
struct fp_netmap_device* dev = fp_allocate(struct fp_netmap_device);
dev->base.vtbl = &netmap_vtbl;
dev->handle = nm_open(name, NULL, 0, NULL);
if (dev->handle == NULL) {
fprintf(stderr, "error: unable to open netmap device %s\n", name);
fp_deallocate(dev);
return NULL;
}
return (struct fp_device*)dev;
}
/*
* Open a netmap device. We assume there is only one queue
* (which is the case for the VALE bridge).
*/
static struct nm_desc *netmap_open(const NetdevNetmapOptions *nm_opts,
Error **errp)
{
struct nm_desc *nmd;
struct nmreq req;
memset(&req, 0, sizeof(req));
nmd = nm_open(nm_opts->ifname, &req, NETMAP_NO_TX_POLL,
NULL);
if (nmd == NULL) {
error_setg_errno(errp, errno, "Failed to nm_open() %s",
nm_opts->ifname);
return NULL;
}
return nmd;
}
/*
* add a netmap port. We add them in pairs, so forwarding occurs
* between two of them.
*/
void
netmap_add_port(const char *dev)
{
static struct sess *s1 = NULL; // XXX stateful; bad!
struct my_netmap_port *port;
int l;
struct sess *s2;
D("opening netmap device %s", dev);
l = strlen(dev) + 1;
if (l >= IFNAMSIZ) {
D("name %s too long, max %d", dev, IFNAMSIZ - 1);
sleep(2);
return;
}
port = calloc(1, sizeof(*port));
port->d = nm_open(dev, NULL, 0, NULL);
if (port->d == NULL) {
D("error opening %s", dev);
kern_free(port); // XXX compat
return;
}
strncpy(port->ifp.if_xname, dev, IFNAMSIZ-1);
port->allocator_id = port->d->req.nr_arg2;
D("--- mem_id %d", port->allocator_id);
s2 = new_session(port->d->fd, netmap_read, port, WANT_READ);
port->sess = s2;
D("create sess %p my_netmap_port %p", s2, port);
if (s1 == NULL) { /* first of a pair */
s1 = s2;
} else { /* second of a pair, cross link */
struct my_netmap_port *peer = s1->arg;
port->peer = peer;
peer->peer = port;
port->can_swap_bufs = peer->can_swap_bufs =
(port->allocator_id == peer->allocator_id);
D("%p %s %d <-> %p %s %d %s",
port, port->d->req.nr_name, port->allocator_id,
peer, peer->d->req.nr_name, peer->allocator_id,
port->can_swap_bufs ? "SWAP" : "COPY");
s1 = NULL;
}
}
int
main(int argc, char *argv[])
{
struct nm_desc *d;
struct pollfd pfd;
char buf[2048];
int count = 0;
if (argc < 2) {
usage(argv[0]);
}
bzero(&pfd, sizeof(pfd));
d = nm_open(argv[1], NULL, 0, 0);
if (d == NULL) {
fprintf(stderr, "no netmap\n");
exit(0);
}
pfd.fd = d->fd;
pfd.events = argv[2] && argv[2][0] == 'w' ? POLLOUT : POLLIN;
fprintf(stderr, "working on %s in %s mode\n", argv[1], pfd.events == POLLIN ? "read" : "write");
for (;;) {
if (pfd.events == POLLIN) {
nm_dispatch(d, -1, my_cb, (void *)&count);
} else {
if (nm_inject(d, buf, 60) > 0) {
count++;
continue;
}
fprintf(stderr, "polling after sending %d\n", count);
count = 0;
}
poll(&pfd, 1, 1000);
}
nm_close(d);
return 0;
}
static void
pci_vtnet_netmap_setup(struct pci_vtnet_softc *sc, char *ifname)
{
sc->pci_vtnet_rx = pci_vtnet_netmap_rx;
sc->pci_vtnet_tx = pci_vtnet_netmap_tx;
sc->vsc_nmd = nm_open(ifname, NULL, 0, 0);
if (sc->vsc_nmd == NULL) {
WPRINTF(("open of netmap device %s failed\n", ifname));
return;
}
sc->vsc_mevp = mevent_add(sc->vsc_nmd->fd,
EVF_READ,
pci_vtnet_rx_callback,
sc);
if (sc->vsc_mevp == NULL) {
WPRINTF(("Could not register event\n"));
nm_close(sc->vsc_nmd);
sc->vsc_nmd = NULL;
}
}
void test_netmap(usn_mbuf_t *m)
{
int tosend = 0;
int n, i;
int rate_limit = 0;
int sent = 0;
struct pollfd pfd = { .fd = g_nmd->fd, .events = POLLOUT };
struct netmap_if *nifp = g_nmd->nifp;
struct timeval stime, etime;
struct nm_desc nmd = *g_nmd;
struct nm_desc *t_nmd;
uint64_t nmd_flags = 0;
// re-open netmap device.
nmd.req.nr_flags = NR_REG_ONE_NIC;
nmd.req.nr_ringid = 0;
printf("interface name:%s,len=%d\n",g_interface, m->mlen);
t_nmd = nm_open(g_interface, NULL, nmd_flags
| NM_OPEN_IFNAME | NM_OPEN_NO_MMAP, &nmd);
if (t_nmd == NULL) {
printf("Unable to open %s: %s", g_interface, strerror(errno));
return;
}
nifp =t_nmd->nifp;
pfd.fd =t_nmd->fd;
pfd.events = POLLOUT;
n = 10000;
sent = 0;
g_config.burst = 512;
printf("g_config.burst=%d\n", g_config.burst);
gettimeofday(&stime, 0);
while ( sent < n ) {
/*
* wait for available room in the send queue(s)
*/
if (poll(&pfd, 1, 1000) <= 0) {
D("poll error/timeout on queue: %s", strerror(errno));
// goto quit;
}
if (pfd.revents & POLLERR) {
D("poll error");
goto quit;
}
for (i = g_nmd->first_tx_ring; i <= g_nmd->last_tx_ring; i++) {
int limit = rate_limit ? tosend : g_config.burst;
int cnt = 0;
if (n > 0 && n - sent < limit)
limit = n - sent;
struct netmap_ring *txring = NETMAP_TXRING(nifp, i);
if (nm_ring_empty(txring))
continue;
cnt = test_send(txring, m, limit);
DEBUG("limit %d tail %d cnt %d",
limit, txring->tail, cnt);
sent += cnt;
}
}
// print info stats
gettimeofday(&etime, 0);
timersub(&etime,&stime,&etime);
printf("num of sent pkts: %d\n", n);
printf("total time: %lu (seconds) %lu (microseconds) \n",
etime.tv_sec, etime.tv_usec);
/* flush any remaining packets */
ioctl(pfd.fd, NIOCTXSYNC, NULL);
/* final part: wait all the TX queues to be empty. */
for (i = g_nmd->first_tx_ring; i <= g_nmd->last_tx_ring; i++) {
struct netmap_ring *txring = NETMAP_TXRING(nifp, i);
while (nm_tx_pending(txring)) {
ioctl(pfd.fd, NIOCTXSYNC, NULL);
usleep(1); /* wait 1 tick */
}
}
quit:
return;
}
/* set the thread affinity. */
int
setaffinity( int i)
{
cpuset_t cpumask;
if (i == -1)
return 0;
/* Set thread affinity affinity.*/
CPU_ZERO(&cpumask);
CPU_SET(i, &cpumask);
if (cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_CPUSET,
-1, sizeof(cpuset_t), &cpumask) != 0) {
DEBUG("Unable to set affinity: %s", strerror(errno));
return 1;
}
return 0;
}
void receiver(void) {
struct nm_desc *netmap_descriptor;
u_int num_cpus = sysconf( _SC_NPROCESSORS_ONLN );
printf("We have %d cpus\n", num_cpus);
struct nmreq base_nmd;
bzero(&base_nmd, sizeof(base_nmd));
// Magic from pkt-gen.c
base_nmd.nr_tx_rings = base_nmd.nr_rx_rings = 0;
base_nmd.nr_tx_slots = base_nmd.nr_rx_slots = 0;
std::string interface = "netmap:eth4";
netmap_descriptor = nm_open(interface.c_str(), &base_nmd, 0, NULL);
if (netmap_descriptor == NULL) {
printf("Can't open netmap device %s\n", interface.c_str());
exit(1);
return;
}
printf("Mapped %dKB memory at %p\n", netmap_descriptor->req.nr_memsize>>10, netmap_descriptor->mem);
printf("We have %d tx and %d rx rings\n", netmap_descriptor->req.nr_tx_rings, netmap_descriptor->req.nr_rx_rings);
/*
protocol stack and may cause a reset of the card,
which in turn may take some time for the PHY to
reconfigure. We do the open here to have time to reset.
*/
int wait_link = 2;
printf("Wait %d seconds for NIC reset\n", wait_link);
sleep(wait_link);
boost::thread* boost_threads_array[num_cpus];
for (int i = 0; i < num_cpus; i++) {
struct nm_desc nmd = *netmap_descriptor;
// This operation is VERY important!
nmd.self = &nmd;
uint64_t nmd_flags = 0;
if (nmd.req.nr_flags != NR_REG_ALL_NIC) {
printf("SHIT SHIT SHIT HAPPINED\n");
}
nmd.req.nr_flags = NR_REG_ONE_NIC;
nmd.req.nr_ringid = i;
/* Only touch one of the rings (rx is already ok) */
nmd_flags |= NETMAP_NO_TX_POLL;
struct nm_desc* new_nmd = nm_open(interface.c_str(), NULL, nmd_flags | NM_OPEN_IFNAME | NM_OPEN_NO_MMAP, &nmd);
if (new_nmd == NULL) {
printf("Can't open netmap descripto for netmap\n");
exit(1);
}
printf("My first ring is %d and last ring id is %d I'm thread %d\n", new_nmd->first_rx_ring, new_nmd->last_rx_ring, i);
printf("Start new thread %d\n", i);
// Start thread and pass netmap descriptor to it
boost_threads_array[i] = new boost::thread(netmap_thread, new_nmd, i);
}
printf("Wait for thread finish\n");
// Wait all threads for completion
for (int i = 0; i < num_cpus; i++) {
boost_threads_array[i]->join();
}
}
int main(int argc, char **argv)
{
int index = 0;
int channel_nums = 4;
int member = 2;
const char *ifname = NULL;
struct thread_args_t targs[MAX_RINGS];
memset(targs, 0x0, sizeof(targs));
if (argc < 2 || argc > 3) {
printf("Usage:%s [interface] <channel nums>\n", argv[1]);
return 1;
}
ifname = argv[1];
if (argc == 3)
{
channel_nums = atoi(argv[2]);
}
if (channel_nums < 1 || channel_nums > MAX_RINGS
|| (MAX_RINGS % channel_nums) != 0 )
{
printf("channel nums error.\n");
return 1;
}
member = MAX_RINGS / channel_nums;
for (index = 0; index < channel_nums; index++)
{
char buff[64];
memset(buff,0x0, 64);
struct thread_args_t *thread_arg = &targs[index];
unsigned short start = index * member;
unsigned short end = start + member - 1;
snprintf(buff, 63, "netmap:%s+%d.%d", ifname, start, end);
thread_arg->desc = nm_open(buff, NULL, NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL, NULL);
if ( thread_arg->desc == NULL) {
D("cannot open eth0");
return 1;
}
printf("%d first_rx_ring:%d last_rx_rings:%d ", index,
thread_arg->desc->first_rx_ring,
thread_arg->desc->last_rx_ring);
printf(" first_tx_ring:%d last_tx_rings:%d\n",
thread_arg->desc->first_tx_ring,
thread_arg->desc->last_tx_ring);
thread_arg->affinity = index;
pthread_create(&(thread_arg->thread), NULL, run, (void*)thread_arg);
pthread_detach(thread_arg->thread);
}
main_loop_statistics(targs, channel_nums);
for (index = 0; index < channel_nums; index ++)
{
nm_close(targs[index].desc);
}
D("exiting");
return (0);
}
static int
pcap_netmap_dispatch(pcap_t *p, int cnt, pcap_handler cb, u_char *user)
{
int ret;
struct pcap_netmap *pn = p->priv;
struct nm_desc *d = pn->d;
struct pollfd pfd = { .fd = p->fd, .events = POLLIN, .revents = 0 };
pn->cb = cb;
pn->cb_arg = user;
for (;;) {
if (p->break_loop) {
p->break_loop = 0;
return PCAP_ERROR_BREAK;
}
/* nm_dispatch won't run forever */
ret = nm_dispatch((void *)d, cnt, (void *)pcap_netmap_filter, (void *)p);
if (ret != 0)
break;
errno = 0;
ret = poll(&pfd, 1, p->opt.timeout);
}
return ret;
}
/* XXX need to check the NIOCTXSYNC/poll */
static int
pcap_netmap_inject(pcap_t *p, const void *buf, size_t size)
{
struct pcap_netmap *pn = p->priv;
struct nm_desc *d = pn->d;
return nm_inject(d, buf, size);
}
static int
pcap_netmap_ioctl(pcap_t *p, u_long what, uint32_t *if_flags)
{
struct pcap_netmap *pn = p->priv;
struct nm_desc *d = pn->d;
struct ifreq ifr;
int error, fd = d->fd;
#ifdef linux
fd = socket(AF_INET, SOCK_DGRAM, 0);
if (fd < 0) {
fprintf(stderr, "Error: cannot get device control socket.\n");
return -1;
}
#endif /* linux */
bzero(&ifr, sizeof(ifr));
strncpy(ifr.ifr_name, d->req.nr_name, sizeof(ifr.ifr_name));
switch (what) {
case SIOCSIFFLAGS:
/*
* The flags we pass in are 32-bit and unsigned.
*
* On most if not all UN*Xes, ifr_flags is 16-bit and
* signed, and the result of assigning a longer
* unsigned value to a shorter signed value is
* implementation-defined (even if, in practice, it'll
* do what's intended on all platforms we support
* result of assigning a 32-bit unsigned value).
* So we mask out the upper 16 bits.
*/
ifr.ifr_flags = *if_flags & 0xffff;
#ifdef __FreeBSD__
/*
* In FreeBSD, we need to set the high-order flags,
* as we're using IFF_PPROMISC, which is in those bits.
*
* XXX - DragonFly BSD?
*/
ifr.ifr_flagshigh = *if_flags >> 16;
#endif /* __FreeBSD__ */
break;
}
error = ioctl(fd, what, &ifr);
if (!error) {
switch (what) {
case SIOCGIFFLAGS:
/*
* The flags we return are 32-bit.
*
* On most if not all UN*Xes, ifr_flags is
* 16-bit and signed, and will get sign-
* extended, so that the upper 16 bits of
* those flags will be forced on. So we
* mask out the upper 16 bits of the
* sign-extended value.
*/
*if_flags = ifr.ifr_flags & 0xffff;
#ifdef __FreeBSD__
/*
* In FreeBSD, we need to return the
* high-order flags, as we're using
* IFF_PPROMISC, which is in those bits.
*
* XXX - DragonFly BSD?
*/
*if_flags |= (ifr.ifr_flagshigh << 16);
#endif /* __FreeBSD__ */
}
}
#ifdef linux
close(fd);
#endif /* linux */
return error ? -1 : 0;
}
static void
pcap_netmap_close(pcap_t *p)
{
struct pcap_netmap *pn = p->priv;
struct nm_desc *d = pn->d;
uint32_t if_flags = 0;
if (pn->must_clear_promisc) {
pcap_netmap_ioctl(p, SIOCGIFFLAGS, &if_flags); /* fetch flags */
if (if_flags & IFF_PPROMISC) {
if_flags &= ~IFF_PPROMISC;
pcap_netmap_ioctl(p, SIOCSIFFLAGS, &if_flags);
}
}
nm_close(d);
pcap_cleanup_live_common(p);
}
static int
pcap_netmap_activate(pcap_t *p)
{
struct pcap_netmap *pn = p->priv;
struct nm_desc *d;
uint32_t if_flags = 0;
d = nm_open(p->opt.device, NULL, 0, NULL);
if (d == NULL) {
pcap_fmt_errmsg_for_errno(p->errbuf, PCAP_ERRBUF_SIZE,
errno, "netmap open: cannot access %s",
p->opt.device);
pcap_cleanup_live_common(p);
return (PCAP_ERROR);
}
if (0)
fprintf(stderr, "%s device %s priv %p fd %d ports %d..%d\n",
__FUNCTION__, p->opt.device, d, d->fd,
d->first_rx_ring, d->last_rx_ring);
pn->d = d;
p->fd = d->fd;
/*
* Turn a negative snapshot value (invalid), a snapshot value of
* 0 (unspecified), or a value bigger than the normal maximum
* value, into the maximum allowed value.
*
* If some application really *needs* a bigger snapshot
* length, we should just increase MAXIMUM_SNAPLEN.
*/
if (p->snapshot <= 0 || p->snapshot > MAXIMUM_SNAPLEN)
p->snapshot = MAXIMUM_SNAPLEN;
if (p->opt.promisc && !(d->req.nr_ringid & NETMAP_SW_RING)) {
pcap_netmap_ioctl(p, SIOCGIFFLAGS, &if_flags); /* fetch flags */
if (!(if_flags & IFF_PPROMISC)) {
pn->must_clear_promisc = 1;
if_flags |= IFF_PPROMISC;
pcap_netmap_ioctl(p, SIOCSIFFLAGS, &if_flags);
}
}
p->linktype = DLT_EN10MB;
p->selectable_fd = p->fd;
p->read_op = pcap_netmap_dispatch;
p->inject_op = pcap_netmap_inject,
p->setfilter_op = install_bpf_program;
p->setdirection_op = NULL;
p->set_datalink_op = NULL;
p->getnonblock_op = pcap_getnonblock_fd;
p->setnonblock_op = pcap_setnonblock_fd;
p->stats_op = pcap_netmap_stats;
p->cleanup_op = pcap_netmap_close;
return (0);
}
pcap_t *
pcap_netmap_create(const char *device, char *ebuf, int *is_ours)
{
pcap_t *p;
*is_ours = (!strncmp(device, "netmap:", 7) || !strncmp(device, "vale", 4));
if (! *is_ours)
return NULL;
p = pcap_create_common(ebuf, sizeof (struct pcap_netmap));
if (p == NULL)
return (NULL);
p->activate_op = pcap_netmap_activate;
return (p);
}
static int netmap_open(odp_pktio_t id ODP_UNUSED, pktio_entry_t *pktio_entry,
const char *netdev, odp_pool_t pool)
{
int i;
int err;
int sockfd;
int mtu;
uint32_t buf_size;
pkt_netmap_t *pkt_nm = &pktio_entry->s.pkt_nm;
struct nm_desc *desc;
struct netmap_ring *ring;
odp_pktin_hash_proto_t hash_proto;
odp_pktio_stats_t cur_stats;
if (getenv("ODP_PKTIO_DISABLE_NETMAP"))
return -1;
if (pool == ODP_POOL_INVALID)
return -1;
/* Init pktio entry */
memset(pkt_nm, 0, sizeof(*pkt_nm));
pkt_nm->sockfd = -1;
pkt_nm->pool = pool;
/* max frame len taking into account the l2-offset */
pkt_nm->max_frame_len = ODP_CONFIG_PACKET_BUF_LEN_MAX -
odp_buffer_pool_headroom(pool) -
odp_buffer_pool_tailroom(pool);
snprintf(pktio_entry->s.name, sizeof(pktio_entry->s.name), "%s",
netdev);
snprintf(pkt_nm->nm_name, sizeof(pkt_nm->nm_name), "netmap:%s",
netdev);
/* Dummy open here to check if netmap module is available and to read
* capability info. */
desc = nm_open(pkt_nm->nm_name, NULL, 0, NULL);
if (desc == NULL) {
ODP_ERR("nm_open(%s) failed\n", pkt_nm->nm_name);
goto error;
}
if (desc->nifp->ni_rx_rings > NM_MAX_DESC) {
ODP_ERR("Unable to store all rx rings\n");
nm_close(desc);
goto error;
}
pkt_nm->num_rx_rings = desc->nifp->ni_rx_rings;
pkt_nm->capa.max_input_queues = PKTIO_MAX_QUEUES;
if (desc->nifp->ni_rx_rings < PKTIO_MAX_QUEUES)
pkt_nm->capa.max_input_queues = desc->nifp->ni_rx_rings;
if (desc->nifp->ni_tx_rings > NM_MAX_DESC) {
ODP_ERR("Unable to store all tx rings\n");
nm_close(desc);
goto error;
}
pkt_nm->num_tx_rings = desc->nifp->ni_tx_rings;
pkt_nm->capa.max_output_queues = PKTIO_MAX_QUEUES;
if (desc->nifp->ni_tx_rings < PKTIO_MAX_QUEUES)
pkt_nm->capa.max_output_queues = desc->nifp->ni_tx_rings;
ring = NETMAP_RXRING(desc->nifp, desc->cur_rx_ring);
buf_size = ring->nr_buf_size;
nm_close(desc);
sockfd = socket(AF_INET, SOCK_DGRAM, 0);
if (sockfd == -1) {
ODP_ERR("Cannot get device control socket\n");
goto error;
}
pkt_nm->sockfd = sockfd;
/* Use either interface MTU (+ ethernet header length) or netmap buffer
* size as MTU, whichever is smaller. */
mtu = mtu_get_fd(pktio_entry->s.pkt_nm.sockfd, pktio_entry->s.name) +
ODPH_ETHHDR_LEN;
if (mtu < 0) {
ODP_ERR("Unable to read interface MTU\n");
goto error;
}
pkt_nm->mtu = ((uint32_t)mtu < buf_size) ? (uint32_t)mtu : buf_size;
/* Check if RSS is supported. If not, set 'max_input_queues' to 1. */
if (rss_conf_get_supported_fd(sockfd, netdev, &hash_proto) == 0) {
ODP_DBG("RSS not supported\n");
pkt_nm->capa.max_input_queues = 1;
}
err = netmap_do_ioctl(pktio_entry, SIOCGIFFLAGS, 0);
if (err)
goto error;
if ((pkt_nm->if_flags & IFF_UP) == 0)
ODP_DBG("%s is down\n", pktio_entry->s.name);
err = mac_addr_get_fd(sockfd, netdev, pkt_nm->if_mac);
if (err)
goto error;
for (i = 0; i < PKTIO_MAX_QUEUES; i++) {
odp_ticketlock_init(&pkt_nm->rx_desc_ring[i].s.lock);
odp_ticketlock_init(&pkt_nm->tx_desc_ring[i].s.lock);
}
/* netmap uses only ethtool to get statistics counters */
err = ethtool_stats_get_fd(pktio_entry->s.pkt_nm.sockfd,
pktio_entry->s.name,
&cur_stats);
if (err) {
ODP_ERR(
"netmap pktio %s does not support statistics counters\n",
pktio_entry->s.name);
pktio_entry->s.stats_type = STATS_UNSUPPORTED;
} else {
pktio_entry->s.stats_type = STATS_ETHTOOL;
}
(void)netmap_stats_reset(pktio_entry);
return 0;
error:
netmap_close(pktio_entry);
return -1;
}
static int netmap_start(pktio_entry_t *pktio_entry)
{
pkt_netmap_t *pkt_nm = &pktio_entry->s.pkt_nm;
netmap_ring_t *desc_ring;
struct nm_desc base_desc;
unsigned i;
unsigned j;
unsigned num_rx_desc = 0;
uint64_t flags;
odp_pktin_mode_t in_mode = pktio_entry->s.param.in_mode;
odp_pktout_mode_t out_mode = pktio_entry->s.param.out_mode;
/* If no pktin/pktout queues have been configured. Configure one
* for each direction. */
if (!pktio_entry->s.num_in_queue &&
in_mode != ODP_PKTIN_MODE_DISABLED) {
odp_pktin_queue_param_t param;
odp_pktin_queue_param_init(¶m);
param.num_queues = 1;
if (odp_pktin_queue_config(pktio_entry->s.handle, ¶m))
return -1;
}
if (!pktio_entry->s.num_out_queue &&
out_mode == ODP_PKTOUT_MODE_DIRECT) {
odp_pktout_queue_param_t param;
odp_pktout_queue_param_init(¶m);
param.num_queues = 1;
if (odp_pktout_queue_config(pktio_entry->s.handle, ¶m))
return -1;
}
if (pkt_nm->num_rx_desc_rings == pktio_entry->s.num_in_queue &&
pkt_nm->num_tx_desc_rings == pktio_entry->s.num_out_queue)
return (netmap_wait_for_link(pktio_entry) == 1) ? 0 : -1;
netmap_close_descriptors(pktio_entry);
/* Map pktin/pktout queues to netmap rings */
if (pktio_entry->s.num_in_queue) {
/* In single queue case only one netmap descriptor is
* required. */
num_rx_desc = (pktio_entry->s.num_in_queue == 1) ? 1 :
pkt_nm->num_rx_rings;
map_netmap_rings(pkt_nm->rx_desc_ring,
pktio_entry->s.num_in_queue, num_rx_desc);
}
if (pktio_entry->s.num_out_queue)
/* Enough to map only one netmap tx ring per pktout queue */
map_netmap_rings(pkt_nm->tx_desc_ring,
pktio_entry->s.num_out_queue,
pktio_entry->s.num_out_queue);
base_desc.self = &base_desc;
base_desc.mem = NULL;
memcpy(base_desc.req.nr_name, pktio_entry->s.name,
sizeof(pktio_entry->s.name));
base_desc.req.nr_flags &= ~NR_REG_MASK;
if (num_rx_desc == 1)
base_desc.req.nr_flags |= NR_REG_ALL_NIC;
else
base_desc.req.nr_flags |= NR_REG_ONE_NIC;
base_desc.req.nr_ringid = 0;
/* Only the first rx descriptor does mmap */
desc_ring = pkt_nm->rx_desc_ring;
flags = NM_OPEN_IFNAME | NETMAP_NO_TX_POLL;
desc_ring[0].s.desc[0] = nm_open(pkt_nm->nm_name, NULL, flags,
&base_desc);
if (desc_ring[0].s.desc[0] == NULL) {
ODP_ERR("nm_start(%s) failed\n", pkt_nm->nm_name);
goto error;
}
/* Open rest of the rx descriptors (one per netmap ring) */
flags = NM_OPEN_IFNAME | NETMAP_NO_TX_POLL | NM_OPEN_NO_MMAP;
for (i = 0; i < pktio_entry->s.num_in_queue; i++)
for (j = desc_ring[i].s.first; j <= desc_ring[i].s.last; j++) {
if (i == 0 && j == 0) { /* First already opened */
if (num_rx_desc > 1)
continue;
else
break;
}
base_desc.req.nr_ringid = j;
desc_ring[i].s.desc[j] = nm_open(pkt_nm->nm_name, NULL,
flags, &base_desc);
if (desc_ring[i].s.desc[j] == NULL) {
ODP_ERR("nm_start(%s) failed\n",
pkt_nm->nm_name);
goto error;
}
}
/* Open tx descriptors */
desc_ring = pkt_nm->tx_desc_ring;
flags = NM_OPEN_IFNAME | NM_OPEN_NO_MMAP;
base_desc.req.nr_flags &= !NR_REG_ALL_NIC;
base_desc.req.nr_flags |= NR_REG_ONE_NIC;
for (i = 0; i < pktio_entry->s.num_out_queue; i++)
for (j = desc_ring[i].s.first; j <= desc_ring[i].s.last; j++) {
base_desc.req.nr_ringid = j;
desc_ring[i].s.desc[j] = nm_open(pkt_nm->nm_name, NULL,
flags, &base_desc);
if (desc_ring[i].s.desc[j] == NULL) {
ODP_ERR("nm_start(%s) failed\n",
pkt_nm->nm_name);
goto error;
}
}
pkt_nm->num_rx_desc_rings = pktio_entry->s.num_in_queue;
pkt_nm->num_tx_desc_rings = pktio_entry->s.num_out_queue;
/* Wait for the link to come up */
return (netmap_wait_for_link(pktio_entry) == 1) ? 0 : -1;
error:
netmap_close_descriptors(pktio_entry);
return -1;
}
int
main(int argc, char **argv)
{
struct pollfd pollfd[2];
int ch;
char *ifa = NULL, *ifb = NULL;
int wait_link = 2;
int win_size_usec = 50;
unsigned int fifo_size = 10;
int n;
int hold = 0;
struct nmreq base_req;
uint32_t buf_head = 0;
#ifdef DEDUP_HASH_STAT
time_t last_hash_output = 0;
#endif
fprintf(stderr, "%s built %s %s\n\n", argv[0], __DATE__, __TIME__);
while ((ch = getopt(argc, argv, "hci:vw:W:F:H")) != -1) {
switch (ch) {
default:
D("bad option %c %s", ch, optarg);
/* fallthrough */
case 'h':
usage();
break;
case 'i': /* interface */
if (ifa == NULL)
ifa = optarg;
else if (ifb == NULL)
ifb = optarg;
else
D("%s ignored, already have 2 interfaces",
optarg);
break;
case 'c':
zerocopy = 0; /* do not zerocopy */
break;
case 'v':
verbose++;
break;
case 'w':
wait_link = atoi(optarg);
break;
case 'W':
win_size_usec = atoi(optarg);
break;
case 'F':
fifo_size = atoi(optarg);
break;
case 'H':
hold = 1;
break;
}
}
if (!ifa || !ifb) {
D("missing interface");
usage();
}
memset(&base_req, 0, sizeof(base_req));
if (!hold) {
base_req.nr_arg3 = fifo_size;
}
pa = nm_open(ifa, &base_req, 0, NULL);
if (pa == NULL) {
D("cannot open %s", ifa);
return (1);
}
if (!hold) {
if (base_req.nr_arg3 != fifo_size) {
D("failed to allocate %u extra buffers", fifo_size);
return (1); // XXX failover to copy?
} else {
buf_head = pa->nifp->ni_bufs_head;
}
}
if (pa->first_rx_ring != pa->last_rx_ring) {
D("%s: too many RX rings (%d)", pa->req.nr_name,
pa->last_rx_ring - pa->first_rx_ring + 1);
return (1);
}
/* try to reuse the mmap() of the first interface, if possible */
pb = nm_open(ifb, NULL, NM_OPEN_NO_MMAP, pa);
if (pb == NULL) {
D("cannot open %s", ifb);
nm_close(pa);
return (1);
}
if (pb->first_tx_ring != pb->last_tx_ring) {
D("%s: too many TX rings (%d)", pb->req.nr_name,
pb->last_rx_ring - pb->first_rx_ring + 1);
nm_close(pa);
return (1);
}
memset(&dedup, 0, sizeof(dedup));
dedup.out_slot = dedup.out_ring->slot;
if (dedup_init(&dedup, fifo_size,
NETMAP_RXRING(pa->nifp, pa->first_rx_ring),
NETMAP_TXRING(pb->nifp, pb->first_tx_ring)) < 0) {
D("failed to initialize dedup with fifo_size %u", fifo_size);
return (1);
}
if (fifo_size >= dedup.out_ring->num_slots - 1) {
D("fifo_size %u too large (max %u)", fifo_size, dedup.out_ring->num_slots - 1);
return (1);
}
if (dedup_set_fifo_buffers(&dedup, NULL, buf_head) != 0) {
D("failed to set 'hold packets' option");
return (1);
}
pa->nifp->ni_bufs_head = 0;
atexit(free_buffers);
/* enable/disable zerocopy */
dedup.in_memid = pa->req.nr_arg2;
dedup.out_memid = (zerocopy ? pb->req.nr_arg2 : -1 );
dedup.fifo_memid = hold ? dedup.out_memid : dedup.in_memid;
D("memids: in %d out %d fifo %d", dedup.in_memid, dedup.out_memid,
dedup.fifo_memid);
dedup.win_size.tv_sec = win_size_usec / 1000000;
dedup.win_size.tv_usec = win_size_usec % 1000000;
D("win_size %lld+%lld", (long long) dedup.win_size.tv_sec,
(long long) dedup.win_size.tv_usec);
/* setup poll(2) array */
memset(pollfd, 0, sizeof(pollfd));
pollfd[0].fd = pa->fd;
pollfd[1].fd = pb->fd;
D("Wait %d secs for link to come up...", wait_link);
sleep(wait_link);
D("Ready to go, %s -> %s", pa->req.nr_name, pb->req.nr_name);
/* main loop */
signal(SIGINT, sigint_h);
n = 0;
while (!do_abort) {
int ret;
struct timeval now;
pollfd[0].events = pollfd[1].events = 0;
pollfd[0].revents = pollfd[1].revents = 0;
if (!n)
pollfd[0].events = POLLIN;
else
pollfd[1].events = POLLOUT;
/* poll() also cause kernel to txsync/rxsync the NICs */
ret = poll(pollfd, 2, 1000);
gettimeofday(&now, NULL);
if (ret <= 0 || verbose)
D("poll %s [0] ev %x %x"
" [1] ev %x %x",
ret <= 0 ? "timeout" : "ok",
pollfd[0].events,
pollfd[0].revents,
pollfd[1].events,
pollfd[1].revents
);
n = dedup_push_in(&dedup, &now);
#ifdef DEDUP_HASH_STAT
if (now.tv_sec != last_hash_output) {
unsigned int i;
last_hash_output = now.tv_sec;
printf("buckets: ");
for (i = 0; i <= dedup.hashmap_mask; i++) {
if (dedup.hashmap[i].bucket_size)
printf("%u: %u, ", i, dedup.hashmap[i].bucket_size);
}
printf("\n");
}
#endif
}
return (0);
}
/*---------------------------------------------------------------------*/
int32_t
netmap_link_iface(void *ctxt, const unsigned char *iface,
const uint16_t batchsize, int8_t qid)
{
TRACE_NETMAP_FUNC_START();
char nifname[MAX_IFNAMELEN];
netmap_module_context *nmc = (netmap_module_context *)ctxt;
netmap_iface_context *nic = NULL;
/* setting nm-ifname*/
sprintf(nifname, "netmap:%s", iface);
/* check if the interface has been registered with some other engine */
netiface *nif = interface_find((char *)iface);
if (nif == NULL) {
nic = calloc(1, sizeof(netmap_iface_context));
if (nic == NULL) {
TRACE_ERR("Can't allocate memory for "
"netmap_iface_context (for %s)\n", iface);
TRACE_NETMAP_FUNC_END();
return -1;
}
/* resetting base_nmd */
memset(&nic->base_nmd, 0, sizeof(struct nm_desc));
/* resetting fd to -1 */
nic->global_fd = nmc->local_fd = -1;
/* use some extra rings */
nic->base_nmd.req.nr_arg3 = NM_EXTRA_BUFS;
nic->nmd_flags |= NM_OPEN_ARG3;
nic->global_nmd = nm_open((char *)nifname, NULL,
nic->nmd_flags,
&nic->base_nmd);
if (nic->global_nmd == NULL) {
TRACE_LOG("Unable to open %s: %s\n",
iface, strerror(errno));
free(nic);
TRACE_NETMAP_FUNC_END();
return -1;
}
nic->global_fd = nic->global_nmd->fd;
TRACE_DEBUG_LOG("mapped %dKB at %p\n",
nic->global_nmd->req.nr_memsize>>10,
nic->global_nmd->mem);
TRACE_DEBUG_LOG("zerocopy %s",
(nic->global_nmd->mem == nic->base_nmd.mem) ?
"enabled\n" : "disabled\n");
if (qid != -1) {
nic->global_nmd->req.nr_flags = NR_REG_ONE_NIC;
nic->global_nmd->req.nr_ringid = qid;
}
/* create interface entry */
create_interface_entry(iface, (qid == -1) ? NO_QUEUES : HW_QUEUES,
IO_NETMAP, nic, nmc->eng);
} else { /* otherwise check if that interface can be registered */
int
main(int arc, char **argv)
{
int i;
struct glob_arg g;
int ch;
int wait_link = 2;
int devqueues = 1; /* how many device queues */
bzero(&g, sizeof(g));
g.main_fd = -1;
g.td_body = receiver_body;
g.report_interval = 1000; /* report interval */
g.affinity = -1;
/* ip addresses can also be a range x.x.x.x-x.x.x.y */
g.src_ip.name = "10.0.0.1";
g.dst_ip.name = "10.1.0.1";
g.dst_mac.name = "ff:ff:ff:ff:ff:ff";
g.src_mac.name = NULL;
g.pkt_size = 60;
g.burst = 512; // default
g.nthreads = 1;
g.cpus = 1;
g.forever = 1;
g.tx_rate = 0;
g.frags = 1;
g.nmr_config = "";
g.virt_header = 0;
while ( (ch = getopt(arc, argv,
"a:f:F:n:i:Il:d:s:D:S:b:c:o:p:T:w:WvR:XC:H:e:m:P:zZ")) != -1) {
struct sf *fn;
switch(ch) {
default:
D("bad option %c %s", ch, optarg);
usage();
break;
case 'n':
g.npackets = atoi(optarg);
break;
case 'F':
i = atoi(optarg);
if (i < 1 || i > 63) {
D("invalid frags %d [1..63], ignore", i);
break;
}
g.frags = i;
break;
case 'f':
for (fn = func; fn->key; fn++) {
if (!strcmp(fn->key, optarg))
break;
}
if (fn->key)
g.td_body = fn->f;
else
D("unrecognised function %s", optarg);
break;
case 'o': /* data generation options */
g.options = atoi(optarg);
break;
case 'a': /* force affinity */
g.affinity = atoi(optarg);
break;
case 'i': /* interface */
/* a prefix of tap: netmap: or pcap: forces the mode.
* otherwise we guess
*/
D("interface is %s", optarg);
if (strlen(optarg) > MAX_IFNAMELEN - 8) {
D("ifname too long %s", optarg);
break;
}
strcpy(g.ifname, optarg);
if (!strcmp(optarg, "null")) {
g.dev_type = DEV_NETMAP;
g.dummy_send = 1;
} else if (!strncmp(optarg, "tap:", 4)) {
g.dev_type = DEV_TAP;
strcpy(g.ifname, optarg + 4);
} else if (!strncmp(optarg, "pcap:", 5)) {
g.dev_type = DEV_PCAP;
strcpy(g.ifname, optarg + 5);
} else if (!strncmp(optarg, "netmap:", 7) ||
!strncmp(optarg, "vale", 4)) {
g.dev_type = DEV_NETMAP;
} else if (!strncmp(optarg, "tap", 3)) {
g.dev_type = DEV_TAP;
} else { /* prepend netmap: */
g.dev_type = DEV_NETMAP;
sprintf(g.ifname, "netmap:%s", optarg);
}
break;
case 'I':
g.options |= OPT_INDIRECT; /* XXX use indirect buffer */
break;
case 'l': /* pkt_size */
g.pkt_size = atoi(optarg);
break;
case 'd':
g.dst_ip.name = optarg;
break;
case 's':
g.src_ip.name = optarg;
break;
case 'T': /* report interval */
g.report_interval = atoi(optarg);
break;
case 'w':
wait_link = atoi(optarg);
break;
case 'W': /* XXX changed default */
g.forever = 0; /* do not exit rx even with no traffic */
break;
case 'b': /* burst */
g.burst = atoi(optarg);
break;
case 'c':
g.cpus = atoi(optarg);
break;
case 'p':
g.nthreads = atoi(optarg);
break;
case 'D': /* destination mac */
g.dst_mac.name = optarg;
break;
case 'S': /* source mac */
g.src_mac.name = optarg;
break;
case 'v':
verbose++;
break;
case 'R':
g.tx_rate = atoi(optarg);
break;
case 'X':
g.options |= OPT_DUMP;
break;
case 'C':
g.nmr_config = strdup(optarg);
break;
case 'H':
g.virt_header = atoi(optarg);
break;
case 'e': /* extra bufs */
g.extra_bufs = atoi(optarg);
break;
case 'm':
if (strcmp(optarg, "tx") == 0) {
g.options |= OPT_MONITOR_TX;
} else if (strcmp(optarg, "rx") == 0) {
g.options |= OPT_MONITOR_RX;
} else {
D("unrecognized monitor mode %s", optarg);
}
break;
case 'P':
g.packet_file = strdup(optarg);
break;
case 'z':
g.options |= OPT_RANDOM_SRC;
break;
case 'Z':
g.options |= OPT_RANDOM_DST;
break;
}
}
if (strlen(g.ifname) <=0 ) {
D("missing ifname");
usage();
}
i = system_ncpus();
if (g.cpus < 0 || g.cpus > i) {
D("%d cpus is too high, have only %d cpus", g.cpus, i);
usage();
}
if (g.cpus == 0)
g.cpus = i;
if (g.pkt_size < 16 || g.pkt_size > MAX_PKTSIZE) {
D("bad pktsize %d [16..%d]\n", g.pkt_size, MAX_PKTSIZE);
usage();
}
if (g.src_mac.name == NULL) {
static char mybuf[20] = "00:00:00:00:00:00";
/* retrieve source mac address. */
if (source_hwaddr(g.ifname, mybuf) == -1) {
D("Unable to retrieve source mac");
// continue, fail later
}
g.src_mac.name = mybuf;
}
/* extract address ranges */
extract_ip_range(&g.src_ip);
extract_ip_range(&g.dst_ip);
extract_mac_range(&g.src_mac);
extract_mac_range(&g.dst_mac);
if (g.src_ip.start != g.src_ip.end ||
g.src_ip.port0 != g.src_ip.port1 ||
g.dst_ip.start != g.dst_ip.end ||
g.dst_ip.port0 != g.dst_ip.port1)
g.options |= OPT_COPY;
if (g.virt_header != 0 && g.virt_header != VIRT_HDR_1
&& g.virt_header != VIRT_HDR_2) {
D("bad virtio-net-header length");
usage();
}
if (g.dev_type == DEV_TAP) {
D("want to use tap %s", g.ifname);
g.main_fd = tap_alloc(g.ifname);
if (g.main_fd < 0) {
D("cannot open tap %s", g.ifname);
usage();
}
#ifndef NO_PCAP
} else if (g.dev_type == DEV_PCAP) {
char pcap_errbuf[PCAP_ERRBUF_SIZE];
pcap_errbuf[0] = '\0'; // init the buffer
g.p = pcap_open_live(g.ifname, 256 /* XXX */, 1, 100, pcap_errbuf);
if (g.p == NULL) {
D("cannot open pcap on %s", g.ifname);
usage();
}
g.main_fd = pcap_fileno(g.p);
D("using pcap on %s fileno %d", g.ifname, g.main_fd);
#endif /* !NO_PCAP */
} else if (g.dummy_send) { /* but DEV_NETMAP */
D("using a dummy send routine");
} else {
struct nmreq base_nmd;
bzero(&base_nmd, sizeof(base_nmd));
parse_nmr_config(g.nmr_config, &base_nmd);
if (g.extra_bufs) {
base_nmd.nr_arg3 = g.extra_bufs;
}
/*
* Open the netmap device using nm_open().
*
* protocol stack and may cause a reset of the card,
* which in turn may take some time for the PHY to
* reconfigure. We do the open here to have time to reset.
*/
g.nmd = nm_open(g.ifname, &base_nmd, 0, NULL);
if (g.nmd == NULL) {
D("Unable to open %s: %s", g.ifname, strerror(errno));
goto out;
}
g.main_fd = g.nmd->fd;
D("mapped %dKB at %p", g.nmd->req.nr_memsize>>10, g.nmd->mem);
/* get num of queues in tx or rx */
if (g.td_body == sender_body)
devqueues = g.nmd->req.nr_tx_rings;
else
devqueues = g.nmd->req.nr_rx_rings;
/* validate provided nthreads. */
if (g.nthreads < 1 || g.nthreads > devqueues) {
D("bad nthreads %d, have %d queues", g.nthreads, devqueues);
// continue, fail later
}
if (verbose) {
struct netmap_if *nifp = g.nmd->nifp;
struct nmreq *req = &g.nmd->req;
D("nifp at offset %d, %d tx %d rx region %d",
req->nr_offset, req->nr_tx_rings, req->nr_rx_rings,
req->nr_arg2);
for (i = 0; i <= req->nr_tx_rings; i++) {
struct netmap_ring *ring = NETMAP_TXRING(nifp, i);
D(" TX%d at 0x%lx slots %d", i,
(char *)ring - (char *)nifp, ring->num_slots);
}
for (i = 0; i <= req->nr_rx_rings; i++) {
struct netmap_ring *ring = NETMAP_RXRING(nifp, i);
D(" RX%d at 0x%lx slots %d", i,
(char *)ring - (char *)nifp, ring->num_slots);
}
}
/* Print some debug information. */
fprintf(stdout,
"%s %s: %d queues, %d threads and %d cpus.\n",
(g.td_body == sender_body) ? "Sending on" : "Receiving from",
g.ifname,
devqueues,
g.nthreads,
g.cpus);
if (g.td_body == sender_body) {
fprintf(stdout, "%s -> %s (%s -> %s)\n",
g.src_ip.name, g.dst_ip.name,
g.src_mac.name, g.dst_mac.name);
}
out:
/* Exit if something went wrong. */
if (g.main_fd < 0) {
D("aborting");
usage();
}
}
if (g.options) {
D("--- SPECIAL OPTIONS:%s%s%s%s%s\n",
g.options & OPT_PREFETCH ? " prefetch" : "",
g.options & OPT_ACCESS ? " access" : "",
g.options & OPT_MEMCPY ? " memcpy" : "",
g.options & OPT_INDIRECT ? " indirect" : "",
g.options & OPT_COPY ? " copy" : "");
}
g.tx_period.tv_sec = g.tx_period.tv_nsec = 0;
if (g.tx_rate > 0) {
/* try to have at least something every second,
* reducing the burst size to some 0.01s worth of data
* (but no less than one full set of fragments)
*/
uint64_t x;
int lim = (g.tx_rate)/300;
if (g.burst > lim)
g.burst = lim;
if (g.burst < g.frags)
g.burst = g.frags;
x = ((uint64_t)1000000000 * (uint64_t)g.burst) / (uint64_t) g.tx_rate;
g.tx_period.tv_nsec = x;
g.tx_period.tv_sec = g.tx_period.tv_nsec / 1000000000;
g.tx_period.tv_nsec = g.tx_period.tv_nsec % 1000000000;
}
if (g.td_body == sender_body)
D("Sending %d packets every %ld.%09ld s",
g.burst, g.tx_period.tv_sec, g.tx_period.tv_nsec);
/* Wait for PHY reset. */
D("Wait %d secs for phy reset", wait_link);
sleep(wait_link);
D("Ready...");
/* Install ^C handler. */
global_nthreads = g.nthreads;
signal(SIGINT, sigint_h);
start_threads(&g);
main_thread(&g);
return 0;
}
/*
* bridge [-v] if1 [if2]
*
* If only one name, or the two interfaces are the same,
* bridges userland and the adapter. Otherwise bridge
* two intefaces.
*/
int
main(int argc, char **argv)
{
struct pollfd pollfd[2];
int ch;
u_int burst = 1024, wait_link = 4;
struct nm_desc *pa = NULL, *pb = NULL;
char *ifa = NULL, *ifb = NULL;
char ifabuf[64] = { 0 };
fprintf(stderr, "%s built %s %s\n",
argv[0], __DATE__, __TIME__);
while ( (ch = getopt(argc, argv, "b:ci:vw:")) != -1) {
switch (ch) {
default:
D("bad option %c %s", ch, optarg);
usage();
break;
case 'b': /* burst */
burst = atoi(optarg);
break;
case 'i': /* interface */
if (ifa == NULL)
ifa = optarg;
else if (ifb == NULL)
ifb = optarg;
else
D("%s ignored, already have 2 interfaces",
optarg);
break;
case 'c':
zerocopy = 0; /* do not zerocopy */
break;
case 'v':
verbose++;
break;
case 'w':
wait_link = atoi(optarg);
break;
}
}
argc -= optind;
argv += optind;
if (argc > 1)
ifa = argv[1];
if (argc > 2)
ifb = argv[2];
if (argc > 3)
burst = atoi(argv[3]);
if (!ifb)
ifb = ifa;
if (!ifa) {
D("missing interface");
usage();
}
if (burst < 1 || burst > 8192) {
D("invalid burst %d, set to 1024", burst);
burst = 1024;
}
if (wait_link > 100) {
D("invalid wait_link %d, set to 4", wait_link);
wait_link = 4;
}
if (!strcmp(ifa, ifb)) {
D("same interface, endpoint 0 goes to host");
snprintf(ifabuf, sizeof(ifabuf) - 1, "%s^", ifa);
ifa = ifabuf;
} else {
/* two different interfaces. Take all rings on if1 */
}
pa = nm_open(ifa, NULL, 0, NULL);
if (pa == NULL) {
D("cannot open %s", ifa);
return (1);
}
// XXX use a single mmap ?
pb = nm_open(ifb, NULL, NM_OPEN_NO_MMAP, pa);
if (pb == NULL) {
D("cannot open %s", ifb);
nm_close(pa);
return (1);
}
zerocopy = zerocopy && (pa->mem == pb->mem);
D("------- zerocopy %ssupported", zerocopy ? "" : "NOT ");
/* setup poll(2) variables. */
memset(pollfd, 0, sizeof(pollfd));
pollfd[0].fd = pa->fd;
pollfd[1].fd = pb->fd;
D("Wait %d secs for link to come up...", wait_link);
sleep(wait_link);
D("Ready to go, %s 0x%x/%d <-> %s 0x%x/%d.",
pa->req.nr_name, pa->first_rx_ring, pa->req.nr_rx_rings,
pb->req.nr_name, pb->first_rx_ring, pb->req.nr_rx_rings);
/* main loop */
signal(SIGINT, sigint_h);
while (!do_abort) {
int n0, n1, ret;
pollfd[0].events = pollfd[1].events = 0;
pollfd[0].revents = pollfd[1].revents = 0;
n0 = pkt_queued(pa, 0);
n1 = pkt_queued(pb, 0);
if (n0)
pollfd[1].events |= POLLOUT;
else
pollfd[0].events |= POLLIN;
if (n1)
pollfd[0].events |= POLLOUT;
else
pollfd[1].events |= POLLIN;
ret = poll(pollfd, 2, 2500);
if (ret <= 0 || verbose)
D("poll %s [0] ev %x %x rx %d@%d tx %d,"
" [1] ev %x %x rx %d@%d tx %d",
ret <= 0 ? "timeout" : "ok",
pollfd[0].events,
pollfd[0].revents,
pkt_queued(pa, 0),
NETMAP_RXRING(pa->nifp, pa->cur_rx_ring)->cur,
pkt_queued(pa, 1),
pollfd[1].events,
pollfd[1].revents,
pkt_queued(pb, 0),
NETMAP_RXRING(pb->nifp, pb->cur_rx_ring)->cur,
pkt_queued(pb, 1)
);
if (ret < 0)
continue;
if (pollfd[0].revents & POLLERR) {
struct netmap_ring *rx = NETMAP_RXRING(pa->nifp, pa->cur_rx_ring);
D("error on fd0, rx [%d,%d,%d)",
rx->head, rx->cur, rx->tail);
}
if (pollfd[1].revents & POLLERR) {
struct netmap_ring *rx = NETMAP_RXRING(pb->nifp, pb->cur_rx_ring);
D("error on fd1, rx [%d,%d,%d)",
rx->head, rx->cur, rx->tail);
}
if (pollfd[0].revents & POLLOUT) {
move(pb, pa, burst);
// XXX we don't need the ioctl */
// ioctl(me[0].fd, NIOCTXSYNC, NULL);
}
if (pollfd[1].revents & POLLOUT) {
move(pa, pb, burst);
// XXX we don't need the ioctl */
// ioctl(me[1].fd, NIOCTXSYNC, NULL);
}
}
D("exiting");
nm_close(pb);
nm_close(pa);
return (0);
}
static void
start_threads(struct glob_arg *g)
{
int i;
targs = calloc(g->nthreads, sizeof(*targs));
/*
* Now create the desired number of threads, each one
* using a single descriptor.
*/
for (i = 0; i < g->nthreads; i++) {
struct targ *t = &targs[i];
bzero(t, sizeof(*t));
t->fd = -1; /* default, with pcap */
t->g = g;
if (g->dev_type == DEV_NETMAP) {
struct nm_desc nmd = *g->nmd; /* copy, we overwrite ringid */
uint64_t nmd_flags = 0;
nmd.self = &nmd;
if (g->nthreads > 1) {
if (nmd.req.nr_flags != NR_REG_ALL_NIC) {
D("invalid nthreads mode %d", nmd.req.nr_flags);
continue;
}
nmd.req.nr_flags = NR_REG_ONE_NIC;
nmd.req.nr_ringid = i;
}
/* Only touch one of the rings (rx is already ok) */
if (g->td_body == receiver_body)
nmd_flags |= NETMAP_NO_TX_POLL;
/* register interface. Override ifname and ringid etc. */
if (g->options & OPT_MONITOR_TX)
nmd.req.nr_flags |= NR_MONITOR_TX;
if (g->options & OPT_MONITOR_RX)
nmd.req.nr_flags |= NR_MONITOR_RX;
t->nmd = nm_open(t->g->ifname, NULL, nmd_flags |
NM_OPEN_IFNAME | NM_OPEN_NO_MMAP, &nmd);
if (t->nmd == NULL) {
D("Unable to open %s: %s",
t->g->ifname, strerror(errno));
continue;
}
t->fd = t->nmd->fd;
set_vnet_hdr_len(t);
} else {
targs[i].fd = g->main_fd;
}
t->used = 1;
t->me = i;
if (g->affinity >= 0) {
if (g->affinity < g->cpus)
t->affinity = g->affinity;
else
t->affinity = i % g->cpus;
} else {
t->affinity = -1;
}
/* default, init packets */
initialize_packet(t);
if (pthread_create(&t->thread, NULL, g->td_body, t) == -1) {
D("Unable to create thread %d: %s", i, strerror(errno));
t->used = 0;
}
}
}
/*
* bridge [-v] if1 [if2]
*
* If only one name, or the two interfaces are the same,
* bridges userland and the adapter. Otherwise bridge
* two intefaces.
*/
int
main(int argc, char **argv)
{
struct pollfd pollfd[2];
int ch;
u_int burst = 1024, wait_link = 4;
struct nm_desc *pa = NULL, *pb = NULL;
char *ifa = NULL, *ifb = NULL;
char ifabuf[64] = { 0 };
int loopback = 0;
fprintf(stderr, "%s built %s %s\n\n", argv[0], __DATE__, __TIME__);
while ((ch = getopt(argc, argv, "hb:ci:vw:L")) != -1) {
switch (ch) {
default:
D("bad option %c %s", ch, optarg);
/* fallthrough */
case 'h':
usage();
break;
case 'b': /* burst */
burst = atoi(optarg);
break;
case 'i': /* interface */
if (ifa == NULL)
ifa = optarg;
else if (ifb == NULL)
ifb = optarg;
else
D("%s ignored, already have 2 interfaces",
optarg);
break;
case 'c':
zerocopy = 0; /* do not zerocopy */
break;
case 'v':
verbose++;
break;
case 'w':
wait_link = atoi(optarg);
break;
case 'L':
loopback = 1;
break;
}
}
argc -= optind;
argv += optind;
if (argc > 0)
ifa = argv[0];
if (argc > 1)
ifb = argv[1];
if (argc > 2)
burst = atoi(argv[2]);
if (!ifb)
ifb = ifa;
if (!ifa) {
D("missing interface");
usage();
}
if (burst < 1 || burst > 8192) {
D("invalid burst %d, set to 1024", burst);
burst = 1024;
}
if (wait_link > 100) {
D("invalid wait_link %d, set to 4", wait_link);
wait_link = 4;
}
if (!strcmp(ifa, ifb)) {
if (!loopback) {
D("same interface, endpoint 0 goes to host");
snprintf(ifabuf, sizeof(ifabuf) - 1, "%s^", ifa);
ifa = ifabuf;
} else {
D("same interface, loopbacking traffic");
}
} else {
/* two different interfaces. Take all rings on if1 */
}
pa = nm_open(ifa, NULL, 0, NULL);
if (pa == NULL) {
D("cannot open %s", ifa);
return (1);
}
/* try to reuse the mmap() of the first interface, if possible */
pb = nm_open(ifb, NULL, NM_OPEN_NO_MMAP, pa);
if (pb == NULL) {
D("cannot open %s", ifb);
nm_close(pa);
return (1);
}
zerocopy = zerocopy && (pa->mem == pb->mem);
D("------- zerocopy %ssupported", zerocopy ? "" : "NOT ");
/* setup poll(2) array */
memset(pollfd, 0, sizeof(pollfd));
pollfd[0].fd = pa->fd;
pollfd[1].fd = pb->fd;
D("Wait %d secs for link to come up...", wait_link);
sleep(wait_link);
D("Ready to go, %s 0x%x/%d <-> %s 0x%x/%d.",
pa->req.nr_name, pa->first_rx_ring, pa->req.nr_rx_rings,
pb->req.nr_name, pb->first_rx_ring, pb->req.nr_rx_rings);
/* main loop */
signal(SIGINT, sigint_h);
while (!do_abort) {
int n0, n1, ret;
pollfd[0].events = pollfd[1].events = 0;
pollfd[0].revents = pollfd[1].revents = 0;
n0 = pkt_queued(pa, 0);
n1 = pkt_queued(pb, 0);
#if defined(_WIN32) || defined(BUSYWAIT)
if (n0) {
ioctl(pollfd[1].fd, NIOCTXSYNC, NULL);
pollfd[1].revents = POLLOUT;
} else {
ioctl(pollfd[0].fd, NIOCRXSYNC, NULL);
}
if (n1) {
ioctl(pollfd[0].fd, NIOCTXSYNC, NULL);
pollfd[0].revents = POLLOUT;
} else {
ioctl(pollfd[1].fd, NIOCRXSYNC, NULL);
}
ret = 1;
#else
if (n0)
pollfd[1].events |= POLLOUT;
else
pollfd[0].events |= POLLIN;
if (n1)
pollfd[0].events |= POLLOUT;
else
pollfd[1].events |= POLLIN;
/* poll() also cause kernel to txsync/rxsync the NICs */
ret = poll(pollfd, 2, 2500);
#endif /* defined(_WIN32) || defined(BUSYWAIT) */
if (ret <= 0 || verbose)
D("poll %s [0] ev %x %x rx %d@%d tx %d,"
" [1] ev %x %x rx %d@%d tx %d",
ret <= 0 ? "timeout" : "ok",
pollfd[0].events,
pollfd[0].revents,
pkt_queued(pa, 0),
NETMAP_RXRING(pa->nifp, pa->cur_rx_ring)->cur,
pkt_queued(pa, 1),
pollfd[1].events,
pollfd[1].revents,
pkt_queued(pb, 0),
NETMAP_RXRING(pb->nifp, pb->cur_rx_ring)->cur,
pkt_queued(pb, 1)
);
if (ret < 0)
continue;
if (pollfd[0].revents & POLLERR) {
struct netmap_ring *rx = NETMAP_RXRING(pa->nifp, pa->cur_rx_ring);
D("error on fd0, rx [%d,%d,%d)",
rx->head, rx->cur, rx->tail);
}
if (pollfd[1].revents & POLLERR) {
struct netmap_ring *rx = NETMAP_RXRING(pb->nifp, pb->cur_rx_ring);
D("error on fd1, rx [%d,%d,%d)",
rx->head, rx->cur, rx->tail);
}
if (pollfd[0].revents & POLLOUT)
move(pb, pa, burst);
if (pollfd[1].revents & POLLOUT)
move(pa, pb, burst);
/* We don't need ioctl(NIOCTXSYNC) on the two file descriptors here,
* kernel will txsync on next poll(). */
}
nm_close(pb);
nm_close(pa);
return (0);
}
int main(int argc, char **argv)
{
int ch;
uint32_t i;
int rv;
unsigned int iter = 0;
glob_arg.ifname[0] = '\0';
glob_arg.output_rings = DEF_OUT_PIPES;
glob_arg.batch = DEF_BATCH;
glob_arg.syslog_interval = DEF_SYSLOG_INT;
while ( (ch = getopt(argc, argv, "i:p:b:B:s:")) != -1) {
switch (ch) {
case 'i':
D("interface is %s", optarg);
if (strlen(optarg) > MAX_IFNAMELEN - 8) {
D("ifname too long %s", optarg);
return 1;
}
if (strncmp(optarg, "netmap:", 7) && strncmp(optarg, "vale", 4)) {
sprintf(glob_arg.ifname, "netmap:%s", optarg);
} else {
strcpy(glob_arg.ifname, optarg);
}
break;
case 'p':
glob_arg.output_rings = atoi(optarg);
if (glob_arg.output_rings < 1) {
D("you must output to at least one pipe");
usage();
return 1;
}
break;
case 'B':
glob_arg.extra_bufs = atoi(optarg);
D("requested %d extra buffers", glob_arg.extra_bufs);
break;
case 'b':
glob_arg.batch = atoi(optarg);
D("batch is %d", glob_arg.batch);
break;
case 's':
glob_arg.syslog_interval = atoi(optarg);
D("syslog interval is %d", glob_arg.syslog_interval);
break;
default:
D("bad option %c %s", ch, optarg);
usage();
return 1;
}
}
if (glob_arg.ifname[0] == '\0') {
D("missing interface name");
usage();
return 1;
}
setlogmask(LOG_UPTO(LOG_INFO));
openlog("lb", LOG_CONS | LOG_PID | LOG_NDELAY, LOG_LOCAL1);
uint32_t npipes = glob_arg.output_rings;
struct overflow_queue *freeq = NULL;
pthread_t stat_thread;
ports = calloc(npipes + 1, sizeof(struct port_des));
if (!ports) {
D("failed to allocate the stats array");
return 1;
}
struct port_des *rxport = &ports[npipes];
if (pthread_create(&stat_thread, NULL, print_stats, NULL) == -1) {
D("unable to create the stats thread: %s", strerror(errno));
return 1;
}
/* we need base_req to specify pipes and extra bufs */
struct nmreq base_req;
memset(&base_req, 0, sizeof(base_req));
base_req.nr_arg1 = npipes;
base_req.nr_arg3 = glob_arg.extra_bufs;
rxport->nmd = nm_open(glob_arg.ifname, &base_req, 0, NULL);
if (rxport->nmd == NULL) {
D("cannot open %s", glob_arg.ifname);
return (1);
} else {
D("successfully opened %s (tx rings: %u)", glob_arg.ifname,
rxport->nmd->req.nr_tx_slots);
}
uint32_t extra_bufs = rxport->nmd->req.nr_arg3;
struct overflow_queue *oq = NULL;
/* reference ring to access the buffers */
rxport->ring = NETMAP_RXRING(rxport->nmd->nifp, 0);
if (!glob_arg.extra_bufs)
goto run;
D("obtained %d extra buffers", extra_bufs);
if (!extra_bufs)
goto run;
/* one overflow queue for each output pipe, plus one for the
* free extra buffers
*/
oq = calloc(npipes + 1, sizeof(struct overflow_queue));
if (!oq) {
D("failed to allocated overflow queues descriptors");
goto run;
}
freeq = &oq[npipes];
rxport->oq = freeq;
freeq->slots = calloc(extra_bufs, sizeof(struct netmap_slot));
if (!freeq->slots) {
D("failed to allocate the free list");
}
freeq->size = extra_bufs;
snprintf(freeq->name, MAX_IFNAMELEN, "free queue");
/*
* the list of buffers uses the first uint32_t in each buffer
* as the index of the next buffer.
*/
uint32_t scan;
for (scan = rxport->nmd->nifp->ni_bufs_head;
scan;
scan = *(uint32_t *)NETMAP_BUF(rxport->ring, scan))
{
struct netmap_slot s;
s.buf_idx = scan;
ND("freeq <- %d", s.buf_idx);
oq_enq(freeq, &s);
}
atexit(free_buffers);
if (freeq->n != extra_bufs) {
D("something went wrong: netmap reported %d extra_bufs, but the free list contained %d",
extra_bufs, freeq->n);
return 1;
}
rxport->nmd->nifp->ni_bufs_head = 0;
run:
for (i = 0; i < npipes; ++i) {
char interface[25];
sprintf(interface, "%s{%d", glob_arg.ifname, i);
D("opening pipe named %s", interface);
//ports[i].nmd = nm_open(interface, NULL, NM_OPEN_NO_MMAP | NM_OPEN_ARG3 | NM_OPEN_RING_CFG, rxport->nmd);
ports[i].nmd = nm_open(interface, NULL, 0, rxport->nmd);
if (ports[i].nmd == NULL) {
D("cannot open %s", interface);
return (1);
} else {
D("successfully opened pipe #%d %s (tx slots: %d)",
i + 1, interface, ports[i].nmd->req.nr_tx_slots);
ports[i].ring = NETMAP_TXRING(ports[i].nmd->nifp, 0);
}
D("zerocopy %s",
(rxport->nmd->mem == ports[i].nmd->mem) ? "enabled" : "disabled");
if (extra_bufs) {
struct overflow_queue *q = &oq[i];
q->slots = calloc(extra_bufs, sizeof(struct netmap_slot));
if (!q->slots) {
D("failed to allocate overflow queue for pipe %d", i);
/* make all overflow queue management fail */
extra_bufs = 0;
}
q->size = extra_bufs;
snprintf(q->name, MAX_IFNAMELEN, "oq %d", i);
ports[i].oq = q;
}
}
if (glob_arg.extra_bufs && !extra_bufs) {
if (oq) {
for (i = 0; i < npipes + 1; i++) {
free(oq[i].slots);
oq[i].slots = NULL;
}
free(oq);
oq = NULL;
}
D("*** overflow queues disabled ***");
}
sleep(2);
struct pollfd pollfd[npipes + 1];
memset(&pollfd, 0, sizeof(pollfd));
signal(SIGINT, sigint_h);
while (!do_abort) {
u_int polli = 0;
iter++;
for (i = 0; i < npipes; ++i) {
struct netmap_ring *ring = ports[i].ring;
if (nm_ring_next(ring, ring->tail) == ring->cur) {
/* no need to poll, there are no packets pending */
continue;
}
pollfd[polli].fd = ports[i].nmd->fd;
pollfd[polli].events = POLLOUT;
pollfd[polli].revents = 0;
++polli;
}
pollfd[polli].fd = rxport->nmd->fd;
pollfd[polli].events = POLLIN;
pollfd[polli].revents = 0;
++polli;
//RD(5, "polling %d file descriptors", polli+1);
rv = poll(pollfd, polli, 10);
if (rv <= 0) {
if (rv < 0 && errno != EAGAIN && errno != EINTR)
RD(1, "poll error %s", strerror(errno));
continue;
}
if (oq) {
/* try to push packets from the overflow queues
* to the corresponding pipes
*/
for (i = 0; i < npipes; i++) {
struct port_des *p = &ports[i];
struct overflow_queue *q = p->oq;
uint32_t j, lim;
struct netmap_ring *ring;
struct netmap_slot *slot;
if (!q->n)
continue;
ring = p->ring;
lim = nm_ring_space(ring);
if (!lim)
continue;
if (q->n < lim)
lim = q->n;
for (j = 0; j < lim; j++) {
struct netmap_slot s = oq_deq(q);
slot = &ring->slot[ring->cur];
oq_enq(freeq, slot);
*slot = s;
slot->flags |= NS_BUF_CHANGED;
ring->cur = nm_ring_next(ring, ring->cur);
}
ring->head = ring->cur;
forwarded += lim;
p->ctr.pkts += lim;
}
}
int batch = 0;
for (i = rxport->nmd->first_rx_ring; i <= rxport->nmd->last_rx_ring; i++) {
struct netmap_ring *rxring = NETMAP_RXRING(rxport->nmd->nifp, i);
//D("prepare to scan rings");
int next_cur = rxring->cur;
struct netmap_slot *next_slot = &rxring->slot[next_cur];
const char *next_buf = NETMAP_BUF(rxring, next_slot->buf_idx);
while (!nm_ring_empty(rxring)) {
struct overflow_queue *q;
struct netmap_slot *rs = next_slot;
// CHOOSE THE CORRECT OUTPUT PIPE
uint32_t hash = pkt_hdr_hash((const unsigned char *)next_buf, 4, 'B');
if (hash == 0)
non_ip++; // XXX ??
// prefetch the buffer for the next round
next_cur = nm_ring_next(rxring, next_cur);
next_slot = &rxring->slot[next_cur];
next_buf = NETMAP_BUF(rxring, next_slot->buf_idx);
__builtin_prefetch(next_buf);
// 'B' is just a hashing seed
uint32_t output_port = hash % glob_arg.output_rings;
struct port_des *port = &ports[output_port];
struct netmap_ring *ring = port->ring;
uint32_t free_buf;
// Move the packet to the output pipe.
if (nm_ring_space(ring)) {
struct netmap_slot *ts = &ring->slot[ring->cur];
free_buf = ts->buf_idx;
ts->buf_idx = rs->buf_idx;
ts->len = rs->len;
ts->flags |= NS_BUF_CHANGED;
ring->head = ring->cur = nm_ring_next(ring, ring->cur);
port->ctr.pkts++;
forwarded++;
goto forward;
}
/* use the overflow queue, if available */
if (!oq) {
dropped++;
port->ctr.drop++;
goto next;
}
q = &oq[output_port];
if (!freeq->n) {
/* revoke some buffers from the longest overflow queue */
uint32_t j;
struct port_des *lp = &ports[0];
uint32_t max = lp->oq->n;
for (j = 1; j < npipes; j++) {
struct port_des *cp = &ports[j];
if (cp->oq->n > max) {
lp = cp;
max = cp->oq->n;
}
}
// XXX optimize this cycle
for (j = 0; lp->oq->n && j < BUF_REVOKE; j++) {
struct netmap_slot tmp = oq_deq(lp->oq);
oq_enq(freeq, &tmp);
}
ND(1, "revoked %d buffers from %s", j, lq->name);
lp->ctr.drop += j;
dropped += j;
}
free_buf = oq_deq(freeq).buf_idx;
oq_enq(q, rs);
forward:
rs->buf_idx = free_buf;
rs->flags |= NS_BUF_CHANGED;
next:
rxring->head = rxring->cur = next_cur;
batch++;
if (unlikely(batch >= glob_arg.batch)) {
ioctl(rxport->nmd->fd, NIOCRXSYNC, NULL);
batch = 0;
}
ND(1,
"Forwarded Packets: %"PRIu64" Dropped packets: %"PRIu64" Percent: %.2f",
forwarded, dropped,
((float)dropped / (float)forwarded * 100));
}
}
}
pthread_join(stat_thread, NULL);
printf("%"PRIu64" packets forwarded. %"PRIu64" packets dropped. Total %"PRIu64"\n", forwarded,
dropped, forwarded + dropped);
return 0;
}
int main(int argc, char **argv)
{
int ch;
uint32_t i;
int rv;
unsigned int iter = 0;
glob_arg.ifname[0] = '\0';
glob_arg.output_rings = 0;
glob_arg.batch = DEF_BATCH;
glob_arg.syslog_interval = DEF_SYSLOG_INT;
while ( (ch = getopt(argc, argv, "i:p:b:B:s:")) != -1) {
switch (ch) {
case 'i':
D("interface is %s", optarg);
if (strlen(optarg) > MAX_IFNAMELEN - 8) {
D("ifname too long %s", optarg);
return 1;
}
if (strncmp(optarg, "netmap:", 7) && strncmp(optarg, "vale", 4)) {
sprintf(glob_arg.ifname, "netmap:%s", optarg);
} else {
strcpy(glob_arg.ifname, optarg);
}
break;
case 'p':
if (parse_pipes(optarg)) {
usage();
return 1;
}
break;
case 'B':
glob_arg.extra_bufs = atoi(optarg);
D("requested %d extra buffers", glob_arg.extra_bufs);
break;
case 'b':
glob_arg.batch = atoi(optarg);
D("batch is %d", glob_arg.batch);
break;
case 's':
glob_arg.syslog_interval = atoi(optarg);
D("syslog interval is %d", glob_arg.syslog_interval);
break;
default:
D("bad option %c %s", ch, optarg);
usage();
return 1;
}
}
if (glob_arg.ifname[0] == '\0') {
D("missing interface name");
usage();
return 1;
}
/* extract the base name */
char *nscan = strncmp(glob_arg.ifname, "netmap:", 7) ?
glob_arg.ifname : glob_arg.ifname + 7;
strncpy(glob_arg.base_name, nscan, MAX_IFNAMELEN);
for (nscan = glob_arg.base_name; *nscan && !index("-*^{}/@", *nscan); nscan++)
;
*nscan = '\0';
if (glob_arg.num_groups == 0)
parse_pipes("");
setlogmask(LOG_UPTO(LOG_INFO));
openlog("lb", LOG_CONS | LOG_PID | LOG_NDELAY, LOG_LOCAL1);
uint32_t npipes = glob_arg.output_rings;
pthread_t stat_thread;
ports = calloc(npipes + 1, sizeof(struct port_des));
if (!ports) {
D("failed to allocate the stats array");
return 1;
}
struct port_des *rxport = &ports[npipes];
init_groups();
if (pthread_create(&stat_thread, NULL, print_stats, NULL) == -1) {
D("unable to create the stats thread: %s", strerror(errno));
return 1;
}
/* we need base_req to specify pipes and extra bufs */
struct nmreq base_req;
memset(&base_req, 0, sizeof(base_req));
base_req.nr_arg1 = npipes;
base_req.nr_arg3 = glob_arg.extra_bufs;
rxport->nmd = nm_open(glob_arg.ifname, &base_req, 0, NULL);
if (rxport->nmd == NULL) {
D("cannot open %s", glob_arg.ifname);
return (1);
} else {
D("successfully opened %s (tx rings: %u)", glob_arg.ifname,
rxport->nmd->req.nr_tx_slots);
}
uint32_t extra_bufs = rxport->nmd->req.nr_arg3;
struct overflow_queue *oq = NULL;
/* reference ring to access the buffers */
rxport->ring = NETMAP_RXRING(rxport->nmd->nifp, 0);
if (!glob_arg.extra_bufs)
goto run;
D("obtained %d extra buffers", extra_bufs);
if (!extra_bufs)
goto run;
/* one overflow queue for each output pipe, plus one for the
* free extra buffers
*/
oq = calloc(npipes + 1, sizeof(struct overflow_queue));
if (!oq) {
D("failed to allocated overflow queues descriptors");
goto run;
}
freeq = &oq[npipes];
rxport->oq = freeq;
freeq->slots = calloc(extra_bufs, sizeof(struct netmap_slot));
if (!freeq->slots) {
D("failed to allocate the free list");
}
freeq->size = extra_bufs;
snprintf(freeq->name, MAX_IFNAMELEN, "free queue");
/*
* the list of buffers uses the first uint32_t in each buffer
* as the index of the next buffer.
*/
uint32_t scan;
for (scan = rxport->nmd->nifp->ni_bufs_head;
scan;
scan = *(uint32_t *)NETMAP_BUF(rxport->ring, scan))
{
struct netmap_slot s;
s.buf_idx = scan;
ND("freeq <- %d", s.buf_idx);
oq_enq(freeq, &s);
}
if (freeq->n != extra_bufs) {
D("something went wrong: netmap reported %d extra_bufs, but the free list contained %d",
extra_bufs, freeq->n);
return 1;
}
rxport->nmd->nifp->ni_bufs_head = 0;
run:
/* we need to create the persistent vale ports */
if (create_custom_ports(rxport->nmd->req.nr_arg2)) {
free_buffers();
return 1;
}
atexit(delete_custom_ports);
atexit(free_buffers);
int j, t = 0;
for (j = 0; j < glob_arg.num_groups; j++) {
struct group_des *g = &groups[j];
int k;
for (k = 0; k < g->nports; ++k) {
struct port_des *p = &g->ports[k];
char interface[25];
sprintf(interface, "netmap:%s{%d/xT", g->pipename, g->first_id + k);
D("opening pipe named %s", interface);
p->nmd = nm_open(interface, NULL, 0, rxport->nmd);
if (p->nmd == NULL) {
D("cannot open %s", interface);
return (1);
} else {
D("successfully opened pipe #%d %s (tx slots: %d)",
k + 1, interface, p->nmd->req.nr_tx_slots);
p->ring = NETMAP_TXRING(p->nmd->nifp, 0);
}
D("zerocopy %s",
(rxport->nmd->mem == p->nmd->mem) ? "enabled" : "disabled");
if (extra_bufs) {
struct overflow_queue *q = &oq[t + k];
q->slots = calloc(extra_bufs, sizeof(struct netmap_slot));
if (!q->slots) {
D("failed to allocate overflow queue for pipe %d", k);
/* make all overflow queue management fail */
extra_bufs = 0;
}
q->size = extra_bufs;
snprintf(q->name, MAX_IFNAMELEN, "oq %s{%d", g->pipename, k);
p->oq = q;
}
}
t += g->nports;
}
if (glob_arg.extra_bufs && !extra_bufs) {
if (oq) {
for (i = 0; i < npipes + 1; i++) {
free(oq[i].slots);
oq[i].slots = NULL;
}
free(oq);
oq = NULL;
}
D("*** overflow queues disabled ***");
}
sleep(2);
struct pollfd pollfd[npipes + 1];
memset(&pollfd, 0, sizeof(pollfd));
signal(SIGINT, sigint_h);
while (!do_abort) {
u_int polli = 0;
iter++;
for (i = 0; i < npipes; ++i) {
struct netmap_ring *ring = ports[i].ring;
if (nm_ring_next(ring, ring->tail) == ring->cur) {
/* no need to poll, there are no packets pending */
continue;
}
pollfd[polli].fd = ports[i].nmd->fd;
pollfd[polli].events = POLLOUT;
pollfd[polli].revents = 0;
++polli;
}
pollfd[polli].fd = rxport->nmd->fd;
pollfd[polli].events = POLLIN;
pollfd[polli].revents = 0;
++polli;
//RD(5, "polling %d file descriptors", polli+1);
rv = poll(pollfd, polli, 10);
if (rv <= 0) {
if (rv < 0 && errno != EAGAIN && errno != EINTR)
RD(1, "poll error %s", strerror(errno));
continue;
}
if (oq) {
/* try to push packets from the overflow queues
* to the corresponding pipes
*/
for (i = 0; i < npipes; i++) {
struct port_des *p = &ports[i];
struct overflow_queue *q = p->oq;
struct group_des *g = p->group;
uint32_t j, lim;
struct netmap_ring *ring;
struct netmap_slot *slot;
if (oq_empty(q))
continue;
ring = p->ring;
lim = nm_ring_space(ring);
if (!lim)
continue;
if (q->n < lim)
lim = q->n;
for (j = 0; j < lim; j++) {
struct netmap_slot s = oq_deq(q), tmp;
tmp.ptr = 0;
slot = &ring->slot[ring->cur];
if (slot->ptr && !g->last) {
tmp.buf_idx = forward_packet(g + 1, slot);
/* the forwarding may have removed packets
* from the current queue
*/
if (q->n < lim)
lim = q->n;
} else {
tmp.buf_idx = slot->buf_idx;
}
oq_enq(freeq, &tmp);
*slot = s;
slot->flags |= NS_BUF_CHANGED;
ring->cur = nm_ring_next(ring, ring->cur);
}
ring->head = ring->cur;
forwarded += lim;
p->ctr.pkts += lim;
}
}
int batch = 0;
for (i = rxport->nmd->first_rx_ring; i <= rxport->nmd->last_rx_ring; i++) {
struct netmap_ring *rxring = NETMAP_RXRING(rxport->nmd->nifp, i);
//D("prepare to scan rings");
int next_cur = rxring->cur;
struct netmap_slot *next_slot = &rxring->slot[next_cur];
const char *next_buf = NETMAP_BUF(rxring, next_slot->buf_idx);
while (!nm_ring_empty(rxring)) {
struct netmap_slot *rs = next_slot;
struct group_des *g = &groups[0];
// CHOOSE THE CORRECT OUTPUT PIPE
uint32_t hash = pkt_hdr_hash((const unsigned char *)next_buf, 4, 'B');
if (hash == 0) {
non_ip++; // XXX ??
}
rs->ptr = hash | (1UL << 32);
// prefetch the buffer for the next round
next_cur = nm_ring_next(rxring, next_cur);
next_slot = &rxring->slot[next_cur];
next_buf = NETMAP_BUF(rxring, next_slot->buf_idx);
__builtin_prefetch(next_buf);
// 'B' is just a hashing seed
rs->buf_idx = forward_packet(g, rs);
rs->flags |= NS_BUF_CHANGED;
rxring->head = rxring->cur = next_cur;
batch++;
if (unlikely(batch >= glob_arg.batch)) {
ioctl(rxport->nmd->fd, NIOCRXSYNC, NULL);
batch = 0;
}
ND(1,
"Forwarded Packets: %"PRIu64" Dropped packets: %"PRIu64" Percent: %.2f",
forwarded, dropped,
((float)dropped / (float)forwarded * 100));
}
}
}
pthread_join(stat_thread, NULL);
printf("%"PRIu64" packets forwarded. %"PRIu64" packets dropped. Total %"PRIu64"\n", forwarded,
dropped, forwarded + dropped);
return 0;
}
static int
pcap_netmap_dispatch(pcap_t *p, int cnt, pcap_handler cb, u_char *user)
{
int ret;
struct pcap_netmap *pn = NM_PRIV(p);
struct nm_desc *d = pn->d;
struct pollfd pfd = { .fd = p->fd, .events = POLLIN, .revents = 0 };
pn->cb = cb;
pn->cb_arg = user;
for (;;) {
if (p->break_loop) {
p->break_loop = 0;
return PCAP_ERROR_BREAK;
}
/* nm_dispatch won't run forever */
ret = nm_dispatch((void *)d, cnt, (void *)pcap_netmap_filter, (void *)p);
if (ret != 0)
break;
errno = 0;
ret = poll(&pfd, 1, p->the_timeout);
}
return ret;
}
/* XXX need to check the NIOCTXSYNC/poll */
static int
pcap_netmap_inject(pcap_t *p, const void *buf, size_t size)
{
struct nm_desc *d = NM_PRIV(p)->d;
return nm_inject(d, buf, size);
}
static int
pcap_netmap_ioctl(pcap_t *p, u_long what, uint32_t *if_flags)
{
struct pcap_netmap *pn = NM_PRIV(p);
struct nm_desc *d = pn->d;
struct ifreq ifr;
int error, fd = d->fd;
#ifdef linux
fd = socket(AF_INET, SOCK_DGRAM, 0);
if (fd < 0) {
fprintf(stderr, "Error: cannot get device control socket.\n");
return -1;
}
#endif /* linux */
bzero(&ifr, sizeof(ifr));
strncpy(ifr.ifr_name, d->req.nr_name, sizeof(ifr.ifr_name));
switch (what) {
case SIOCSIFFLAGS:
ifr.ifr_flags = *if_flags;
#ifdef __FreeBSD__
ifr.ifr_flagshigh = *if_flags >> 16;
#endif /* __FreeBSD__ */
break;
}
error = ioctl(fd, what, &ifr);
if (!error) {
switch (what) {
case SIOCGIFFLAGS:
*if_flags = ifr.ifr_flags;
#ifdef __FreeBSD__
*if_flags |= (ifr.ifr_flagshigh << 16);
#endif /* __FreeBSD__ */
}
}
#ifdef linux
close(fd);
#endif /* linux */
return error ? -1 : 0;
}
static void
pcap_netmap_close(pcap_t *p)
{
struct pcap_netmap *pn = NM_PRIV(p);
struct nm_desc *d = pn->d;
uint32_t if_flags = 0;
if (pn->must_clear_promisc) {
pcap_netmap_ioctl(p, SIOCGIFFLAGS, &if_flags); /* fetch flags */
if (if_flags & IFF_PPROMISC) {
if_flags &= ~IFF_PPROMISC;
pcap_netmap_ioctl(p, SIOCSIFFLAGS, &if_flags);
}
}
nm_close(d);
#ifdef HAVE_NO_PRIV
free(pn);
SET_PRIV(p, NULL); // unnecessary
#endif
pcap_cleanup_live_common(p);
}
static int
pcap_netmap_activate(pcap_t *p)
{
struct pcap_netmap *pn = NM_PRIV(p);
struct nm_desc *d = nm_open(p->opt.source, NULL, 0, NULL);
uint32_t if_flags = 0;
if (d == NULL) {
snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
"netmap open: cannot access %s: %s\n",
p->opt.source, pcap_strerror(errno));
#ifdef HAVE_NO_PRIV
free(pn);
SET_PRIV(p, NULL); // unnecessary
#endif
pcap_cleanup_live_common(p);
return (PCAP_ERROR);
}
if (0)
fprintf(stderr, "%s device %s priv %p fd %d ports %d..%d\n",
__FUNCTION__, p->opt.source, d, d->fd,
d->first_rx_ring, d->last_rx_ring);
pn->d = d;
p->fd = d->fd;
if (p->opt.promisc && !(d->req.nr_ringid & NETMAP_SW_RING)) {
pcap_netmap_ioctl(p, SIOCGIFFLAGS, &if_flags); /* fetch flags */
if (!(if_flags & IFF_PPROMISC)) {
pn->must_clear_promisc = 1;
if_flags |= IFF_PPROMISC;
pcap_netmap_ioctl(p, SIOCSIFFLAGS, &if_flags);
}
}
p->linktype = DLT_EN10MB;
p->selectable_fd = p->fd;
p->read_op = pcap_netmap_dispatch;
p->inject_op = pcap_netmap_inject,
p->setfilter_op = install_bpf_program;
p->setdirection_op = NULL;
p->set_datalink_op = NULL;
p->getnonblock_op = pcap_getnonblock_fd;
p->setnonblock_op = pcap_setnonblock_fd;
p->stats_op = pcap_netmap_stats;
p->cleanup_op = pcap_netmap_close;
return (0);
}
pcap_t *
pcap_netmap_create(const char *device, char *ebuf, int *is_ours)
{
pcap_t *p;
*is_ours = (!strncmp(device, "netmap:", 7) || !strncmp(device, "vale", 4));
if (! *is_ours)
return NULL;
#ifdef HAVE_NO_PRIV
{
void *pn = calloc(1, sizeof(struct pcap_netmap));
if (pn == NULL)
return NULL;
p = pcap_create_common(device, ebuf);
if (p == NULL) {
free(pn);
return NULL;
}
SET_PRIV(p, pn);
}
#else
p = pcap_create_common(device, ebuf, sizeof (struct pcap_netmap));
if (p == NULL)
return (NULL);
#endif
p->activate_op = pcap_netmap_activate;
return (p);
}