/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#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>
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
-
-enum { LRU_MAX_PORTS = 1024 };
-enum { LRU_MASK = LRU_MAX_PORTS - 1};
-BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
+enum { MAX_PORTS = USHRT_MAX };
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_frag_handling ipv4_frags; /* OVS_DP_ATTR_IPV4_FRAGS. */
};
static void dpif_linux_dp_init(struct dpif_linux_dp *);
/* Attributes.
*
- * The 'stats' and 'used' members point to 64-bit data that might only be
- * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to
- * access their values.
+ * The 'stats' member points to 64-bit data that might only be aligned on
+ * 32-bit boundaries, so get_unaligned_u64() should be used to access its
+ * values.
*
* If 'actions' is nonnull then OVS_FLOW_ATTR_ACTIONS will be included in
* the Netlink version of the command, even if actions_len is zero. */
size_t key_len;
const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
size_t actions_len;
- const uint32_t *upcall_pid; /* OVS_FLOW_ATTR_UPCALL_PID. */
const struct ovs_flow_stats *stats; /* OVS_FLOW_ATTR_STATS. */
const uint8_t *tcp_flags; /* OVS_FLOW_ATTR_TCP_FLAGS. */
- const uint64_t *used; /* OVS_FLOW_ATTR_USED. */
+ const ovs_32aligned_u64 *used; /* OVS_FLOW_ATTR_USED. */
bool clear; /* OVS_FLOW_ATTR_CLEAR. */
};
const struct ofpbuf *);
static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *,
struct ofpbuf *);
-static int dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+static int dpif_linux_flow_transact(struct dpif_linux_flow *request,
struct dpif_linux_flow *reply,
struct ofpbuf **bufp);
static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *,
/* 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. */
+ int epoll_fd; /* epoll fd that includes the upcall socks. */
/* Change notification. */
struct sset changed_ports; /* Ports that have changed. */
struct nln_notifier *port_notifier;
bool change_error;
- /* Queue of unused ports. */
- unsigned long *lru_bitmap;
- uint16_t lru_ports[LRU_MAX_PORTS];
- size_t lru_head;
- size_t lru_tail;
+ /* Port number allocation. */
+ uint16_t alloc_port_no;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
static void open_dpif(const struct dpif_linux_dp *, struct dpif **);
static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
static void dpif_linux_port_changed(const void *vport, void *dpif);
-static uint32_t get_upcall_pid_port(struct dpif_linux *, uint32_t port);
-static uint32_t get_upcall_pid_flow(struct dpif_linux *,
- const struct nlattr *key, size_t key_len);
+static uint32_t dpif_linux_port_get_pid(const struct dpif *, uint16_t port_no);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
return CONTAINER_OF(dpif, struct dpif_linux, dpif);
}
-static void
-dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
-{
- if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
- bitmap_set1(dp->lru_bitmap, port);
- dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
- }
-}
-
-static uint32_t
-dpif_linux_pop_port(struct dpif_linux *dp)
-{
- uint16_t port;
-
- if (dp->lru_head == dp->lru_tail) {
- return UINT32_MAX;
- }
-
- port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
- bitmap_set0(dp->lru_bitmap, port);
- return port;
-}
-
static int
dpif_linux_enumerate(struct sset *all_dps)
{
{
struct dpif_linux_dp dp_request, dp;
struct ofpbuf *buf;
+ uint32_t upcall_pid;
int error;
error = dpif_linux_init();
/* Create or look up datapath. */
dpif_linux_dp_init(&dp_request);
- dp_request.cmd = create ? OVS_DP_CMD_NEW : OVS_DP_CMD_GET;
+ if (create) {
+ dp_request.cmd = OVS_DP_CMD_NEW;
+ upcall_pid = 0;
+ dp_request.upcall_pid = &upcall_pid;
+ } else {
+ dp_request.cmd = OVS_DP_CMD_GET;
+ }
dp_request.name = name;
error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
if (error) {
open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
{
struct dpif_linux *dpif;
- int i;
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);
dpif->dp_ifindex = dp->dp_ifindex;
sset_init(&dpif->changed_ports);
*dpifp = &dpif->dpif;
-
- dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
- bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL);
- for (i = 1; i < LRU_MAX_PORTS; i++) {
- dpif_linux_push_port(dpif, i);
- }
}
static void
{
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;
nln_notifier_destroy(dpif->port_notifier);
destroy_upcall_socks(dpif);
sset_destroy(&dpif->changed_ports);
- free(dpif->lru_bitmap);
free(dpif);
}
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)
struct dpif_linux_vport request, reply;
const struct ofpbuf *options;
struct ofpbuf *buf;
- int error;
+ int error, i = 0, max_ports = MAX_PORTS;
dpif_linux_vport_init(&request);
request.cmd = OVS_VPORT_CMD_NEW;
do {
uint32_t upcall_pid;
- request.port_no = dpif_linux_pop_port(dpif);
- upcall_pid = get_upcall_pid_port(dpif, request.port_no);
+ request.port_no = ++dpif->alloc_port_no;
+ upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no);
request.upcall_pid = &upcall_pid;
error = dpif_linux_vport_transact(&request, &reply, &buf);
*port_nop = reply.port_no;
VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
dpif_name(dpif_), request.port_no, upcall_pid);
+ } else if (error == EFBIG) {
+ /* Older datapath has lower limit. */
+ max_ports = dpif->alloc_port_no;
+ dpif->alloc_port_no = 0;
}
+
ofpbuf_delete(buf);
- } while (request.port_no != UINT32_MAX
+ } while ((i++ < max_ports)
&& (error == EBUSY || error == EFBIG));
return error;
vport.port_no = port_no;
error = dpif_linux_vport_transact(&vport, NULL, NULL);
- if (!error) {
- dpif_linux_push_port(dpif, port_no);
- }
return error;
}
error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
- dpif_port->name = xstrdup(reply.name);
- dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
- dpif_port->port_no = reply.port_no;
+ if (reply.dp_ifindex != request.dp_ifindex) {
+ /* A query by name reported that 'port_name' is in some datapath
+ * other than 'dpif', but the caller wants to know about 'dpif'. */
+ error = ENODEV;
+ } else {
+ dpif_port->name = xstrdup(reply.name);
+ dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply));
+ dpif_port->port_no = reply.port_no;
+ }
ofpbuf_delete(buf);
}
return error;
static int
dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- /* If the datapath increases its range of supported ports, then it should
- * start reporting that. */
- return 1024;
+ return MAX_PORTS;
+}
+
+static uint32_t
+dpif_linux_port_get_pid(const struct dpif *dpif_, uint16_t port_no)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if (dpif->epoll_fd < 0) {
+ return 0;
+ } else {
+ int idx = port_no & (N_UPCALL_SOCKS - 1);
+ return nl_sock_pid(dpif->upcall_socks[idx]);
+ }
}
static int
struct dpif_linux_port_state {
struct nl_dump dump;
- unsigned long *port_bitmap; /* Ports in the datapath. */
- bool complete; /* Dump completed without error. */
};
static int
struct ofpbuf *buf;
*statep = state = xmalloc(sizeof *state);
- state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
- state->complete = false;
dpif_linux_vport_init(&request);
request.cmd = OVS_DP_CMD_GET;
int error;
if (!nl_dump_next(&state->dump, &buf)) {
- state->complete = true;
return EOF;
}
return error;
}
- if (vport.port_no < LRU_MAX_PORTS) {
- bitmap_set1(state->port_bitmap, vport.port_no);
- }
-
dpif_port->name = (char *) vport.name;
dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
dpif_port->port_no = vport.port_no;
}
static int
-dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
+dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_port_state *state = state_;
int error = nl_dump_done(&state->dump);
- if (state->complete) {
- uint16_t i;
-
- for (i = 0; i < LRU_MAX_PORTS; i++) {
- if (!bitmap_is_set(state->port_bitmap, i)) {
- dpif_linux_push_port(dpif, i);
- }
- }
- }
-
- free(state->port_bitmap);
free(state);
return error;
}
return error;
}
-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)
+static void
+dpif_linux_init_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put,
+ 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 = (put->flags & DPIF_FP_CREATE
+ ? OVS_FLOW_CMD_NEW : OVS_FLOW_CMD_SET);
+ request->dp_ifindex = dpif->dp_ifindex;
+ request->key = put->key;
+ request->key_len = put->key_len;
+ /* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
+ request->actions = put->actions ? put->actions : &dummy_action;
+ request->actions_len = put->actions_len;
+ if (put->flags & DPIF_FP_ZERO_STATS) {
+ request->clear = true;
+ }
+ request->nlmsg_flags = put->flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+}
+
+static int
+dpif_linux_flow_put(struct dpif *dpif_, const struct dpif_flow_put *put)
+{
struct dpif_linux_flow request, reply;
- struct nlattr dummy_action;
- uint32_t upcall_pid;
struct ofpbuf *buf;
int error;
- upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
-
- 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;
- request.upcall_pid = &upcall_pid;
- 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_, put, &request);
error = dpif_linux_flow_transact(&request,
- stats ? &reply : NULL,
- stats ? &buf : NULL);
- if (!error && stats) {
- dpif_linux_flow_get_stats(&reply, stats);
+ put->stats ? &reply : NULL,
+ put->stats ? &buf : NULL);
+ if (!error && put->stats) {
+ dpif_linux_flow_get_stats(&reply, put->stats);
ofpbuf_delete(buf);
}
return error;
}
-static int
-dpif_linux_flow_del(struct dpif *dpif_,
- const struct nlattr *key, size_t key_len,
- struct dpif_flow_stats *stats)
+static void
+dpif_linux_init_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del,
+ struct dpif_linux_flow *request)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ dpif_linux_flow_init(request);
+ request->cmd = OVS_FLOW_CMD_DEL;
+ request->dp_ifindex = dpif->dp_ifindex;
+ request->key = del->key;
+ request->key_len = del->key_len;
+}
+
+static int
+dpif_linux_flow_del(struct dpif *dpif_, const struct dpif_flow_del *del)
+{
struct dpif_linux_flow request, reply;
struct ofpbuf *buf;
int error;
- dpif_linux_flow_init(&request);
- request.cmd = OVS_FLOW_CMD_DEL;
- request.dp_ifindex = dpif->dp_ifindex;
- request.key = key;
- request.key_len = key_len;
+ dpif_linux_init_flow_del(dpif_, del, &request);
error = dpif_linux_flow_transact(&request,
- stats ? &reply : NULL,
- stats ? &buf : NULL);
- if (!error && stats) {
- dpif_linux_flow_get_stats(&reply, stats);
+ del->stats ? &reply : NULL,
+ del->stats ? &buf : NULL);
+ if (!error && del->stats) {
+ dpif_linux_flow_get_stats(&reply, del->stats);
ofpbuf_delete(buf);
}
return error;
return error;
}
-static int
-dpif_linux_execute__(int dp_ifindex, uint32_t upcall_pid,
- const struct nlattr *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+static void
+dpif_linux_encode_execute(int dp_ifindex, const struct dpif_execute *d_exec,
+ struct ofpbuf *buf)
{
- struct ovs_header *execute;
- struct ofpbuf *buf;
- int error;
+ struct ovs_header *k_exec;
- buf = ofpbuf_new(128 + actions_len + packet->size);
+ ofpbuf_prealloc_tailroom(buf, (64
+ + d_exec->packet->size
+ + d_exec->key_len
+ + d_exec->actions_len));
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;
+ k_exec = ofpbuf_put_uninit(buf, sizeof *k_exec);
+ k_exec->dp_ifindex = dp_ifindex;
- nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET, packet->data, packet->size);
- nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, key, key_len);
- nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS, actions, actions_len);
- nl_msg_put_u32(buf, OVS_PACKET_ATTR_UPCALL_PID, upcall_pid);
-
- error = nl_sock_transact(genl_sock, buf, NULL);
- ofpbuf_delete(buf);
- return error;
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_PACKET,
+ d_exec->packet->data, d_exec->packet->size);
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_KEY, d_exec->key, d_exec->key_len);
+ nl_msg_put_unspec(buf, OVS_PACKET_ATTR_ACTIONS,
+ d_exec->actions, d_exec->actions_len);
}
static int
-dpif_linux_execute(struct dpif *dpif_,
- const struct nlattr *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *packet)
+dpif_linux_execute__(int dp_ifindex, const struct dpif_execute *execute)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
+ uint64_t request_stub[1024 / 8];
+ struct ofpbuf request;
+ int error;
- return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len,
- actions, actions_len, packet);
+ ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
+ dpif_linux_encode_execute(dp_ifindex, execute, &request);
+ error = nl_sock_transact(genl_sock, &request, NULL);
+ ofpbuf_uninit(&request);
+
+ return error;
}
static int
-dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
+dpif_linux_execute(struct dpif *dpif_, const struct dpif_execute *execute)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- *listen_mask = dpif->listen_mask;
- return 0;
-}
-static uint32_t
-get_upcall_pid_port__(struct dpif_linux *dpif, uint32_t port)
-{
- int idx = port & (N_UPCALL_SOCKS - 1);
- return nl_sock_pid(dpif->upcall_socks[idx]);
+ return dpif_linux_execute__(dpif->dp_ifindex, execute);
}
-static uint32_t
-get_upcall_pid_port(struct dpif_linux *dpif, uint32_t port)
+#define MAX_OPS 50
+
+static void
+dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
{
- if (!(dpif->listen_mask & (1u << DPIF_UC_MISS))) {
- return 0;
- }
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return get_upcall_pid_port__(dpif, port);
-}
+ struct op_auxdata {
+ struct nl_transaction txn;
+
+ struct ofpbuf request;
+ uint64_t request_stub[1024 / 8];
+
+ struct ofpbuf reply;
+ uint64_t reply_stub[1024 / 8];
+ } auxes[MAX_OPS];
+
+ struct nl_transaction *txnsp[MAX_OPS];
+ size_t i;
+
+ assert(n_ops <= MAX_OPS);
+ for (i = 0; i < n_ops; i++) {
+ struct op_auxdata *aux = &auxes[i];
+ struct dpif_op *op = ops[i];
+ struct dpif_flow_put *put;
+ struct dpif_flow_del *del;
+ struct dpif_execute *execute;
+ struct dpif_linux_flow flow;
+
+ ofpbuf_use_stub(&aux->request,
+ aux->request_stub, sizeof aux->request_stub);
+ aux->txn.request = &aux->request;
+
+ ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
+ aux->txn.reply = NULL;
+
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ put = &op->u.flow_put;
+ dpif_linux_init_flow_put(dpif_, put, &flow);
+ if (put->stats) {
+ flow.nlmsg_flags |= NLM_F_ECHO;
+ aux->txn.reply = &aux->reply;
+ }
+ dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ del = &op->u.flow_del;
+ dpif_linux_init_flow_del(dpif_, del, &flow);
+ if (del->stats) {
+ flow.nlmsg_flags |= NLM_F_ECHO;
+ aux->txn.reply = &aux->reply;
+ }
+ dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
+ break;
-static uint32_t
-get_upcall_pid_flow(struct dpif_linux *dpif,
- const struct nlattr *key, size_t key_len)
-{
- const struct nlattr *nla;
- uint32_t port;
+ case DPIF_OP_EXECUTE:
+ execute = &op->u.execute;
+ dpif_linux_encode_execute(dpif->dp_ifindex, execute,
+ &aux->request);
+ break;
- if (!(dpif->listen_mask & (1u << DPIF_UC_ACTION))) {
- return 0;
+ default:
+ NOT_REACHED();
+ }
}
- nla = nl_attr_find__(key, key_len, OVS_KEY_ATTR_IN_PORT);
- if (nla) {
- port = nl_attr_get_u32(nla);
- } else {
- port = random_uint32();
+ for (i = 0; i < n_ops; i++) {
+ txnsp[i] = &auxes[i].txn;
}
+ nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
+
+ for (i = 0; i < n_ops; i++) {
+ struct op_auxdata *aux = &auxes[i];
+ struct nl_transaction *txn = &auxes[i].txn;
+ struct dpif_op *op = ops[i];
+ struct dpif_flow_put *put;
+ struct dpif_flow_del *del;
+
+ op->error = txn->error;
+
+ switch (op->type) {
+ case DPIF_OP_FLOW_PUT:
+ put = &op->u.flow_put;
+ if (!op->error && put->stats) {
+ struct dpif_linux_flow reply;
+
+ op->error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply);
+ if (!op->error) {
+ dpif_linux_flow_get_stats(&reply, put->stats);
+ }
+ }
+ break;
+
+ case DPIF_OP_FLOW_DEL:
+ del = &op->u.flow_del;
+ if (!op->error && del->stats) {
+ struct dpif_linux_flow reply;
+
+ op->error = dpif_linux_flow_from_ofpbuf(&reply, txn->reply);
+ if (!op->error) {
+ dpif_linux_flow_get_stats(&reply, del->stats);
+ }
+ }
+ break;
+
+ case DPIF_OP_EXECUTE:
+ break;
- return get_upcall_pid_port__(dpif, port);
+ default:
+ NOT_REACHED();
+ }
+
+ ofpbuf_uninit(&aux->request);
+ ofpbuf_uninit(&aux->reply);
+ }
}
static void
-set_upcall_pids(struct dpif_linux *dpif)
+dpif_linux_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
{
- struct dpif_port port;
+ while (n_ops > 0) {
+ size_t chunk = MIN(n_ops, MAX_OPS);
+ dpif_linux_operate__(dpif, ops, chunk);
+ ops += chunk;
+ n_ops -= chunk;
+ }
+}
+
+static void
+set_upcall_pids(struct dpif *dpif_)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_port_dump port_dump;
- struct dpif_flow_dump flow_dump;
- const struct nlattr *key;
- size_t key_len;
+ struct dpif_port port;
int error;
DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
- uint32_t upcall_pid = get_upcall_pid_port(dpif, port.port_no);
+ uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port.port_no);
struct dpif_linux_vport vport_request;
dpif_linux_vport_init(&vport_request);
dpif_name(&dpif->dpif), strerror(error));
}
}
-
- dpif_flow_dump_start(&flow_dump, &dpif->dpif);
- while (dpif_flow_dump_next(&flow_dump, &key, &key_len,
- NULL, NULL, NULL)) {
- uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
- struct dpif_linux_flow flow_request;
-
- dpif_linux_flow_init(&flow_request);
- flow_request.cmd = OVS_FLOW_CMD_SET;
- flow_request.dp_ifindex = dpif->dp_ifindex;
- flow_request.key = key;
- flow_request.key_len = key_len;
- flow_request.upcall_pid = &upcall_pid;
- error = dpif_linux_flow_transact(&flow_request, NULL, NULL);
- if (error) {
- VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on flow: %s",
- dpif_name(&dpif->dpif), strerror(error));
- }
- }
- dpif_flow_dump_done(&flow_dump);
}
static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+dpif_linux_recv_set(struct dpif *dpif_, bool enable)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- if (listen_mask == dpif->listen_mask) {
+ if ((dpif->epoll_fd >= 0) == enable) {
return 0;
}
- if (!listen_mask) {
+ if (!enable) {
destroy_upcall_socks(dpif);
- } else if (!dpif->listen_mask) {
+ } else {
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;
}
+
+ memset(&event, 0, sizeof event);
+ 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;
- set_upcall_pids(dpif);
+ set_upcall_pids(dpif_);
return 0;
}
}
static int
-dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
+dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall,
+ struct ofpbuf *buf)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
int read_tries = 0;
- if (!dpif->listen_mask) {
+ if (dpif->epoll_fd < 0) {
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 (i = 0; i < retval; i++) {
+ dpif->ready_mask |= 1u << events[i].data.u32;
+ }
+ }
- for (;;) {
- struct ofpbuf *buf;
- int error;
+ while (dpif->ready_mask) {
+ int indx = ffs(dpif->ready_mask) - 1;
+ struct nl_sock *upcall_sock = dpif->upcall_socks[indx];
- if (++read_tries > 50) {
- return EAGAIN;
- }
+ dpif->ready_mask &= ~(1u << indx);
+
+ for (;;) {
+ int dp_ifindex;
+ int error;
+
+ if (++read_tries > 50) {
+ return EAGAIN;
+ }
- error = nl_sock_recv(upcall_sock, &buf, false);
+ error = nl_sock_recv(upcall_sock, buf, false);
+ if (error) {
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;
}
+ return error;
+ }
- ofpbuf_delete(buf);
- if (error) {
- return error;
- }
+ error = parse_odp_packet(buf, upcall, &dp_ifindex);
+ if (!error && dp_ifindex == dpif->dp_ifindex) {
+ return 0;
+ }
+ 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) {
+ if (dpif->epoll_fd < 0) {
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
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
int i;
- if (!dpif->listen_mask) {
+ if (dpif->epoll_fd < 0) {
return;
}
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_port_query_by_name,
dpif_linux_get_max_ports,
+ dpif_linux_port_get_pid,
dpif_linux_port_dump_start,
dpif_linux_port_dump_next,
dpif_linux_port_dump_done,
dpif_linux_flow_dump_next,
dpif_linux_flow_dump_done,
dpif_linux_execute,
- dpif_linux_recv_get_mask,
- dpif_linux_recv_set_mask,
+ dpif_linux_operate,
+ dpif_linux_recv_set,
dpif_linux_queue_to_priority,
dpif_linux_recv,
dpif_linux_recv_wait,
{
struct ofpbuf actions, key, packet;
struct odputil_keybuf keybuf;
+ struct dpif_execute execute;
struct flow flow;
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);
ofpbuf_use_stack(&actions, &action, sizeof action);
nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
- return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size,
- actions.data, actions.size, &packet);
+ execute.key = key.data;
+ execute.key_len = key.size;
+ execute.actions = actions.data;
+ execute.actions_len = actions.size;
+ execute.packet = &packet;
+ return dpif_linux_execute__(dp_ifindex, &execute);
}
static bool
[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_UPCALL_PID] = { .type = NL_A_U32 },
- [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 },
flow->actions = nl_attr_get(a[OVS_FLOW_ATTR_ACTIONS]);
flow->actions_len = nl_attr_get_size(a[OVS_FLOW_ATTR_ACTIONS]);
}
- if (a[OVS_FLOW_ATTR_UPCALL_PID]) {
- flow->upcall_pid = nl_attr_get(a[OVS_FLOW_ATTR_UPCALL_PID]);
- }
if (a[OVS_FLOW_ATTR_STATS]) {
flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
}
struct ovs_header *ovs_header;
nl_msg_put_genlmsghdr(buf, 0, ovs_flow_family,
- NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags,
- flow->cmd, 1);
+ NLM_F_REQUEST | flow->nlmsg_flags,
+ flow->cmd, OVS_FLOW_VERSION);
ovs_header = ofpbuf_put_uninit(buf, sizeof *ovs_header);
ovs_header->dp_ifindex = flow->dp_ifindex;
flow->actions, flow->actions_len);
}
- if (flow->upcall_pid) {
- nl_msg_put_u32(buf, OVS_FLOW_ATTR_UPCALL_PID, *flow->upcall_pid);
- }
-
/* We never need to send these to the kernel. */
assert(!flow->stats);
assert(!flow->tcp_flags);
* stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply
* is no longer needed ('reply' will contain pointers into '*bufp'). */
static int
-dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+dpif_linux_flow_transact(struct dpif_linux_flow *request,
struct dpif_linux_flow *reply, struct ofpbuf **bufp)
{
struct ofpbuf *request_buf;
assert((reply != NULL) == (bufp != NULL));
+ if (reply) {
+ request->nlmsg_flags |= NLM_F_ECHO;
+ }
+
request_buf = ofpbuf_new(1024);
dpif_linux_flow_to_ofpbuf(request, request_buf);
error = nl_sock_transact(genl_sock, request_buf, bufp);
stats->n_packets = 0;
stats->n_bytes = 0;
}
- stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
+ 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