#include <linux/pkt_sched.h>
#include <linux/rtnetlink.h>
#include <linux/sockios.h>
+#include <poll.h>
#include <stdlib.h>
+#include <strings.h>
+#include <sys/epoll.h>
#include <sys/stat.h>
#include <unistd.h>
enum { N_UPCALL_SOCKS = 16 };
BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
+BUILD_ASSERT_DECL(N_UPCALL_SOCKS <= 32); /* We use a 32-bit word as a mask. */
/* This ethtool flag was introduced in Linux 2.6.24, so it might be
* missing if we have old headers. */
const char *name; /* OVS_DP_ATTR_NAME. */
const uint32_t *upcall_pid; /* OVS_DP_UPCALL_PID. */
struct ovs_dp_stats stats; /* OVS_DP_ATTR_STATS. */
- enum ovs_datapath_frag ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
/* Upcall messages. */
struct nl_sock *upcall_socks[N_UPCALL_SOCKS];
- int last_read_upcall;
- unsigned int listen_mask;
+ uint32_t ready_mask; /* 1-bit for each sock with unread messages. */
+ unsigned int listen_mask; /* Mask of DPIF_UC_* bits. */
+ int epoll_fd; /* epoll fd that includes the upcall socks. */
/* Change notification. */
struct sset changed_ports; /* Ports that have changed. */
dpif = xzalloc(sizeof *dpif);
dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
dpif);
+ dpif->epoll_fd = -1;
dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
dp->dp_ifindex, dp->dp_ifindex);
{
int i;
+ if (dpif->epoll_fd >= 0) {
+ close(dpif->epoll_fd);
+ dpif->epoll_fd = -1;
+ }
for (i = 0; i < N_UPCALL_SOCKS; i++) {
nl_sock_destroy(dpif->upcall_socks[i]);
dpif->upcall_socks[i] = NULL;
error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- stats->n_frags = dp.stats.n_frags;
stats->n_hit = dp.stats.n_hit;
stats->n_missed = dp.stats.n_missed;
stats->n_lost = dp.stats.n_lost;
return error;
}
-static int
-dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
-{
- struct dpif_linux_dp dp;
- struct ofpbuf *buf;
- int error;
-
- error = dpif_linux_dp_get(dpif_, &dp, &buf);
- if (!error) {
- *drop_fragsp = dp.ipv4_frags == OVS_DP_FRAG_DROP;
- ofpbuf_delete(buf);
- }
- return error;
-}
-
-static int
-dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_dp dp;
-
- dpif_linux_dp_init(&dp);
- dp.cmd = OVS_DP_CMD_SET;
- dp.dp_ifindex = dpif->dp_ifindex;
- dp.ipv4_frags = drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
-}
-
static int
dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
uint16_t *port_nop)
return error;
}
+static void
+dpif_linux_init_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *actions, size_t actions_len,
+ struct dpif_linux_flow *request)
+{
+ static struct nlattr dummy_action;
+
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ dpif_linux_flow_init(request);
+ request->cmd = (flags & DPIF_FP_CREATE
+ ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
+ request->dp_ifindex = dpif->dp_ifindex;
+ request->key = key;
+ request->key_len = key_len;
+ /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
+ request->actions = actions ? actions : &dummy_action;
+ request->actions_len = actions_len;
+ if (flags & DPIF_FP_ZERO_STATS) {
+ request->clear = true;
+ }
+ request->nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+}
+
static int
dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags,
const struct nlattr *key, size_t key_len,
const struct nlattr *actions, size_t actions_len,
struct dpif_flow_stats *stats)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_flow request, reply;
- struct nlattr dummy_action;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = flags & DPIF_FP_CREATE ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET;
- request.dp_ifindex = dpif->dp_ifindex;
- request.key = key;
- request.key_len = key_len;
- /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
- request.actions = actions ? actions : &dummy_action;
- request.actions_len = actions_len;
- if (flags & DPIF_FP_ZERO_STATS) {
- request.clear = true;
- }
- request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+ dpif_linux_init_flow_put(dpif_, flags, key, key_len, actions, actions_len,
+ &request);
error = dpif_linux_flow_transact(&request,
stats ? &reply : NULL,
stats ? &buf : NULL);
return error;
}
-static int
-dpif_linux_execute__(int dp_ifindex, const struct nlattr *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+static struct ofpbuf *
+dpif_linux_encode_execute(int dp_ifindex,
+ const struct nlattr *key, size_t key_len,
+ const struct nlattr *actions, size_t actions_len,
+ const struct ofpbuf *packet)
{
struct ovs_header *execute;
struct ofpbuf *buf;
- int error;
buf = ofpbuf_new(128 + actions_len + packet->size);
nl_msg_put_genlmsghdr(buf, 0, ovs_packet_family, NLM_F_REQUEST,
- OVS_PACKET_CMD_EXECUTE, 1);
+ OVS_PACKET_CMD_EXECUTE, OVS_PACKET_VERSION);
execute = ofpbuf_put_uninit(buf, sizeof *execute);
execute->dp_ifindex = dp_ifindex;
nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len);
nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len);
- error = nl_sock_transact(genl_sock, buf, NULL);
- ofpbuf_delete(buf);
+ return buf;
+}
+
+static int
+dpif_linux_execute__(int dp_ifindex, const struct nlattr *key, size_t key_len,
+ const struct nlattr *actions, size_t actions_len,
+ const struct ofpbuf *packet)
+{
+ struct ofpbuf *request;
+ int error;
+
+ request = dpif_linux_encode_execute(dp_ifindex,
+ key, key_len, actions, actions_len,
+ packet);
+ error = nl_sock_transact(genl_sock, request, NULL);
+ ofpbuf_delete(request);
+
return error;
}
actions, actions_len, packet);
}
+static void
+dpif_linux_operate(struct dpif *dpif_, union dpif_op **ops, size_t n_ops)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct nl_transaction **txnsp;
+ struct nl_transaction *txns;
+ size_t i;
+
+ txns = xmalloc(n_ops * sizeof *txns);
+ for (i = 0; i < n_ops; i++) {
+ struct nl_transaction *txn = &txns[i];
+ union dpif_op *op = ops[i];
+
+ if (op->type == DPIF_OP_FLOW_PUT) {
+ struct dpif_flow_put *put = &op->flow_put;
+ struct dpif_linux_flow request;
+
+ dpif_linux_init_flow_put(dpif_, put->flags, put->key, put->key_len,
+ put->actions, put->actions_len,
+ &request);
+ if (put->stats) {
+ request.nlmsg_flags |= NLM_F_ECHO;
+ }
+ txn->request = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(&request, txn->request);
+ } else if (op->type == DPIF_OP_EXECUTE) {
+ struct dpif_execute *execute = &op->execute;
+
+ txn->request = dpif_linux_encode_execute(
+ dpif->dp_ifindex, execute->key, execute->key_len,
+ execute->actions, execute->actions_len, execute->packet);
+ } else {
+ NOT_REACHED();
+ }
+ }
+
+ txnsp = xmalloc(n_ops * sizeof *txnsp);
+ for (i = 0; i < n_ops; i++) {
+ txnsp[i] = &txns[i];
+ }
+
+ nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
+
+ free(txnsp);
+
+ for (i = 0; i < n_ops; i++) {
+ struct nl_transaction *txn = &txns[i];
+ union dpif_op *op = ops[i];
+
+ if (op->type == DPIF_OP_FLOW_PUT) {
+ struct dpif_flow_put *put = &op->flow_put;
+ int error = txn->error;
+
+ if (!error && put->stats) {
+ struct dpif_linux_flow reply;
+
+ error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply);
+ if (!error) {
+ dpif_linux_flow_get_stats(&reply, put->stats);
+ }
+ }
+ put->error = error;
+ } else if (op->type == DPIF_OP_EXECUTE) {
+ struct dpif_execute *execute = &op->execute;
+
+ execute->error = txn->error;
+ } else {
+ NOT_REACHED();
+ }
+
+ ofpbuf_delete(txn->request);
+ ofpbuf_delete(txn->reply);
+ }
+ free(txns);
+}
+
static int
dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
{
int i;
int error;
+ dpif->epoll_fd = epoll_create(N_UPCALL_SOCKS);
+ if (dpif->epoll_fd < 0) {
+ return errno;
+ }
+
for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ struct epoll_event event;
+
error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]);
if (error) {
destroy_upcall_socks(dpif);
return error;
}
+
+ event.events = EPOLLIN;
+ event.data.u32 = i;
+ if (epoll_ctl(dpif->epoll_fd, EPOLL_CTL_ADD,
+ nl_sock_fd(dpif->upcall_socks[i]), &event) < 0) {
+ error = errno;
+ destroy_upcall_socks(dpif);
+ return error;
+ }
}
+
+ dpif->ready_mask = 0;
}
dpif->listen_mask = listen_mask;
dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
int read_tries = 0;
if (!dpif->listen_mask) {
return EAGAIN;
}
- for (i = 0; i < N_UPCALL_SOCKS; i++) {
- struct nl_sock *upcall_sock;
- dpif->last_read_upcall = (dpif->last_read_upcall + 1) &
- (N_UPCALL_SOCKS - 1);
- upcall_sock = dpif->upcall_socks[dpif->last_read_upcall];
+ if (!dpif->ready_mask) {
+ struct epoll_event events[N_UPCALL_SOCKS];
+ int retval;
+ int i;
- if (nl_sock_woke(upcall_sock)) {
- int dp_ifindex;
+ do {
+ retval = epoll_wait(dpif->epoll_fd, events, N_UPCALL_SOCKS, 0);
+ } while (retval < 0 && errno == EINTR);
+ if (retval < 0) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ VLOG_WARN_RL(&rl, "epoll_wait failed (%s)", strerror(errno));
+ }
- for (;;) {
- struct ofpbuf *buf;
- int error;
+ for (i = 0; i < retval; i++) {
+ dpif->ready_mask |= 1u << events[i].data.u32;
+ }
+ }
- if (++read_tries > 50) {
- return EAGAIN;
- }
+ while (dpif->ready_mask) {
+ int indx = ffs(dpif->ready_mask) - 1;
+ struct nl_sock *upcall_sock = dpif->upcall_socks[indx];
- error = nl_sock_recv(upcall_sock, &buf, false);
- if (error == EAGAIN) {
- break;
- } else if (error) {
- return error;
- }
+ dpif->ready_mask &= ~(1u << indx);
- error = parse_odp_packet(buf, upcall, &dp_ifindex);
- if (!error
- && dp_ifindex == dpif->dp_ifindex
- && dpif->listen_mask & (1u << upcall->type)) {
- return 0;
- }
+ for (;;) {
+ struct ofpbuf *buf;
+ int dp_ifindex;
+ int error;
- ofpbuf_delete(buf);
- if (error) {
- return error;
- }
+ if (++read_tries > 50) {
+ return EAGAIN;
+ }
+
+ error = nl_sock_recv(upcall_sock, &buf, false);
+ if (error == EAGAIN) {
+ break;
+ } else if (error) {
+ return error;
+ }
+
+ error = parse_odp_packet(buf, upcall, &dp_ifindex);
+ if (!error
+ && dp_ifindex == dpif->dp_ifindex
+ && dpif->listen_mask & (1u << upcall->type)) {
+ return 0;
+ }
+
+ ofpbuf_delete(buf);
+ if (error) {
+ return error;
}
}
}
dpif_linux_recv_wait(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
if (!dpif->listen_mask) {
return;
}
- for (i = 0; i < N_UPCALL_SOCKS; i++) {
- nl_sock_wait(dpif->upcall_socks[i], POLLIN);
- }
+ poll_fd_wait(dpif->epoll_fd, POLLIN);
}
static void
dpif_linux_run,
dpif_linux_wait,
dpif_linux_get_stats,
- dpif_linux_get_drop_frags,
- dpif_linux_set_drop_frags,
dpif_linux_port_add,
dpif_linux_port_del,
dpif_linux_port_query_by_number,
dpif_linux_flow_dump_next,
dpif_linux_flow_dump_done,
dpif_linux_execute,
+ dpif_linux_operate,
dpif_linux_recv_get_mask,
dpif_linux_recv_set_mask,
dpif_linux_queue_to_priority,
uint64_t action;
ofpbuf_use_const(&packet, data, size);
- flow_extract(&packet, htonll(0), 0, &flow);
+ flow_extract(&packet, 0, htonll(0), 0, &flow);
ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
odp_flow_key_from_flow(&key, &flow);
[OVS_VPORT_ATTR_TYPE] = { .type = NL_A_U32 },
[OVS_VPORT_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NL_A_U32 },
- [OVS_VPORT_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct ovs_vport_stats),
- .max_len = sizeof(struct ovs_vport_stats),
+ [OVS_VPORT_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_vport_stats),
.optional = true },
[OVS_VPORT_ATTR_ADDRESS] = { .type = NL_A_UNSPEC,
.min_len = ETH_ADDR_LEN,
struct ovs_header *ovs_header;
nl_msg_put_genlmsghdr(buf, 0, ovs_vport_family, NLM_F_REQUEST | NLM_F_ECHO,
- vport->cmd, 1);
+ vport->cmd, OVS_VPORT_VERSION);
ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
ovs_header->dp_ifindex = vport->dp_ifindex;
{
static const struct nl_policy ovs_datapath_policy[] = {
[OVS_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
- [OVS_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct ovs_dp_stats),
- .max_len = sizeof(struct ovs_dp_stats),
+ [OVS_DP_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_dp_stats),
.optional = true },
- [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
memcpy(&dp->stats, nl_attr_get(a[OVS_DP_ATTR_STATS]),
sizeof dp->stats);
}
- if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
- dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
- }
return 0;
}
struct ovs_header *ovs_header;
nl_msg_put_genlmsghdr(buf, 0, ovs_datapath_family,
- NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd,
+ OVS_DATAPATH_VERSION);
ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
ovs_header->dp_ifindex = dp->dp_ifindex;
}
/* Skip OVS_DP_ATTR_STATS since we never have a reason to serialize it. */
-
- if (dp->ipv4_frags) {
- nl_msg_put_u32(buf, OVS_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
- }
}
/* Clears 'dp' to "empty" values. */
static const struct nl_policy ovs_flow_policy[] = {
[OVS_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
[OVS_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
- [OVS_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
- .min_len = sizeof(struct ovs_flow_stats),
- .max_len = sizeof(struct ovs_flow_stats),
+ [OVS_FLOW_ATTR_STATS] = { NL_POLICY_FOR(struct ovs_flow_stats),
.optional = true },
[OVS_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
[OVS_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
NLM_F_REQUEST | flow->nlmsg_flags,
- flow->cmd, 1);
+ flow->cmd, OVS_FLOW_VERSION);
ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
ovs_header->dp_ifindex = flow->dp_ifindex;
stats->used = flow->used ? get_32aligned_u64(flow->used) : 0;
stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
}
-
git://git.onelab.eu / sliver-openvswitch.git / blobdiff