#include "dpif-provider.h"
#include "netdev.h"
#include "netdev-vport.h"
+#include "netlink-socket.h"
#include "netlink.h"
#include "odp-util.h"
#include "ofpbuf.h"
#include "rtnetlink-link.h"
#include "shash.h"
#include "svec.h"
+#include "unaligned.h"
#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(dpif_linux);
+struct dpif_linux_dp {
+ /* Generic Netlink header. */
+ uint8_t cmd;
+
+ /* struct odp_header. */
+ uint32_t dp_idx;
+
+ /* Attributes. */
+ const char *name; /* ODP_DP_ATTR_NAME. */
+ struct odp_stats stats; /* ODP_DP_ATTR_STATS. */
+ enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */
+ const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */
+ uint32_t mcgroups[DPIF_N_UC_TYPES]; /* ODP_DP_ATTR_MCGROUPS. */
+};
+
+static void dpif_linux_dp_init(struct dpif_linux_dp *);
+static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *,
+ const struct ofpbuf *);
+static void dpif_linux_dp_dump_start(struct nl_dump *);
+static int dpif_linux_dp_transact(const struct dpif_linux_dp *request,
+ struct dpif_linux_dp *reply,
+ struct ofpbuf **bufp);
+static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply,
+ struct ofpbuf **bufp);
+
+struct dpif_linux_flow {
+ /* ioctl command argument. */
+ int cmd;
+
+ /* struct odp_flow header. */
+ unsigned int nlmsg_flags;
+ uint32_t dp_idx;
+
+ /* Attributes.
+ *
+ * The 'stats', 'used', and 'state' 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. */
+ const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */
+ size_t key_len;
+ const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */
+ size_t actions_len;
+ const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */
+ const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */
+ const uint64_t *used; /* ODP_FLOW_ATTR_USED. */
+ bool clear; /* ODP_FLOW_ATTR_CLEAR. */
+ const uint64_t *state; /* ODP_FLOW_ATTR_STATE. */
+};
+
+static void dpif_linux_flow_init(struct dpif_linux_flow *);
+static int dpif_linux_flow_transact(const 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 *,
+ struct dpif_flow_stats *);
+
/* Datapath interface for the openvswitch Linux kernel module. */
struct dpif_linux {
struct dpif dpif;
int fd;
+ /* Multicast group messages. */
+ struct nl_sock *mc_sock;
+ uint32_t mcgroups[DPIF_N_UC_TYPES];
+ unsigned int listen_mask;
+
/* Used by dpif_linux_get_all_names(). */
char *local_ifname;
int minor;
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
-static int do_ioctl(const struct dpif *, int cmd, const void *arg);
-static int open_dpif(const struct dpif_linux_vport *local_vport,
+/* Generic Netlink family numbers for ODP. */
+static int odp_datapath_family;
+static int odp_packet_family;
+
+/* Generic Netlink socket. */
+static struct nl_sock *genl_sock;
+
+static int dpif_linux_init(void);
+static int open_dpif(const struct dpif_linux_dp *,
+ const struct dpif_linux_vport *local_vport,
struct dpif **);
static int get_openvswitch_major(void);
-static int create_minor(const char *name, int minor);
static int open_minor(int minor, int *fdp);
static int make_openvswitch_device(int minor, char **fnp);
static void dpif_linux_port_changed(const struct rtnetlink_link_change *,
static int
dpif_linux_enumerate(struct svec *all_dps)
{
+ struct nl_dump dump;
+ struct ofpbuf msg;
int major;
int error;
- int i;
+
+ error = dpif_linux_init();
+ if (error) {
+ return error;
+ }
/* Check that the Open vSwitch module is loaded. */
major = get_openvswitch_major();
return -major;
}
- error = 0;
- for (i = 0; i < ODP_MAX; i++) {
- struct dpif *dpif;
- char devname[16];
- int retval;
-
- sprintf(devname, "dp%d", i);
- retval = dpif_open(devname, "system", &dpif);
- if (!retval) {
- svec_add(all_dps, devname);
- dpif_uninit(dpif, true);
- } else if (retval != ENODEV && !error) {
- error = retval;
+ dpif_linux_dp_dump_start(&dump);
+ while (nl_dump_next(&dump, &msg)) {
+ struct dpif_linux_dp dp;
+
+ if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) {
+ svec_add(all_dps, dp.name);
}
}
- return error;
+ return nl_dump_done(&dump);
}
static int
dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name,
bool create, struct dpif **dpifp)
{
- struct dpif_linux_vport request, reply;
+ struct dpif_linux_vport vport_request, vport;
+ struct dpif_linux_dp dp_request, dp;
struct ofpbuf *buf;
int minor;
int error;
+ error = dpif_linux_init();
+ if (error) {
+ return error;
+ }
+
minor = !strncmp(name, "dp", 2)
&& isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1;
- if (create) {
- if (minor >= 0) {
- error = create_minor(name, minor);
- if (error) {
- return error;
- }
- } else {
- /* Scan for unused minor number. */
- for (minor = 0; ; minor++) {
- if (minor >= ODP_MAX) {
- /* All datapath numbers in use. */
- return ENOBUFS;
- }
- error = create_minor(name, minor);
- if (!error) {
- break;
- } else if (error != EBUSY) {
- return error;
- }
- }
- }
- }
-
- dpif_linux_vport_init(&request);
- request.cmd = ODP_VPORT_GET;
- request.port_no = ODPP_LOCAL;
- if (minor >= 0) {
- request.dp_idx = minor;
- } else {
- request.name = name;
+ /* Create or look up datapath. */
+ dpif_linux_dp_init(&dp_request);
+ dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET;
+ dp_request.dp_idx = minor;
+ dp_request.name = minor < 0 ? name : NULL;
+ error = dpif_linux_dp_transact(&dp_request, &dp, &buf);
+ if (error) {
+ return error;
}
-
- error = dpif_linux_vport_transact(&request, &reply, &buf);
+ ofpbuf_delete(buf); /* Pointers inside 'dp' are now invalid! */
+
+ /* Look up local port. */
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = ODP_VPORT_GET;
+ vport_request.dp_idx = dp.dp_idx;
+ vport_request.port_no = ODPP_LOCAL;
+ vport_request.name = minor < 0 ? name : NULL;
+ error = dpif_linux_vport_transact(&vport_request, &vport, &buf);
if (error) {
return error;
- } else if (reply.port_no != ODPP_LOCAL) {
+ } else if (vport.port_no != ODPP_LOCAL) {
/* This is an Open vSwitch device but not the local port. We
* intentionally support only using the name of the local port as the
* name of a datapath; otherwise, it would be too difficult to
* enumerate all the names of a datapath. */
error = EOPNOTSUPP;
} else {
- error = open_dpif(&reply, dpifp);
+ error = open_dpif(&dp, &vport, dpifp);
}
-
ofpbuf_delete(buf);
return error;
}
static int
-open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **dpifp)
+open_dpif(const struct dpif_linux_dp *dp,
+ const struct dpif_linux_vport *local_vport, struct dpif **dpifp)
{
int dp_idx = local_vport->dp_idx;
struct dpif_linux *dpif;
char *name;
int error;
int fd;
+ int i;
error = open_minor(dp_idx, &fd);
if (error) {
free(name);
dpif->fd = fd;
+ dpif->mc_sock = NULL;
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ dpif->mcgroups[i] = dp->mcgroups[i];
+ }
+ dpif->listen_mask = 0;
dpif->local_ifname = xstrdup(local_vport->name);
dpif->local_ifindex = local_vport->ifindex;
dpif->minor = dp_idx;
static int
dpif_linux_destroy(struct dpif *dpif_)
{
- return do_ioctl(dpif_, ODP_DP_DESTROY, NULL);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_DEL;
+ dp.dp_idx = dpif->minor;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats)
{
- memset(stats, 0, sizeof *stats);
- return do_ioctl(dpif_, ODP_DP_STATS, stats);
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
+ if (!error) {
+ *stats = dp.stats;
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp)
{
- int drop_frags;
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
int error;
- error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags);
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- *drop_fragsp = drop_frags & 1;
+ *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP;
+ ofpbuf_delete(buf);
}
return error;
}
static int
dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags)
{
- int drop_frags_int = drop_frags;
- return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_SET;
+ dp.dp_idx = dpif->minor;
+ dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
static int
dpif_linux_flow_flush(struct dpif *dpif_)
{
- return do_ioctl(dpif_, ODP_FLOW_FLUSH, NULL);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ return ioctl(dpif->fd, ODP_FLOW_FLUSH, dpif->minor) ? errno : 0;
}
struct dpif_linux_port_state {
}
}
-static void
-odp_flow_stats_to_dpif_flow_stats(const struct odp_flow_stats *ofs,
- struct dpif_flow_stats *dfs)
-{
- dfs->n_packets = ofs->n_packets;
- dfs->n_bytes = ofs->n_bytes;
- dfs->used = ofs->used_sec * 1000 + ofs->used_nsec / 1000000;
- dfs->tcp_flags = ofs->tcp_flags;
-}
-
static int
-dpif_linux_flow_get(const struct dpif *dpif_, int flags,
+dpif_linux_flow_get(const struct dpif *dpif_,
const struct nlattr *key, size_t key_len,
struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
{
- struct ofpbuf *actions = NULL;
- struct odp_flow odp_flow;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
int error;
- memset(&odp_flow, 0, sizeof odp_flow);
- odp_flow.key = (struct nlattr *) key;
- odp_flow.key_len = key_len;
- if (actionsp) {
- actions = *actionsp = ofpbuf_new(65536);
- odp_flow.actions = actions->base;
- odp_flow.actions_len = actions->allocated;
- }
- odp_flow.flags = flags;
-
- error = do_ioctl(dpif_, ODP_FLOW_GET, &odp_flow);
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_GET;
+ request.dp_idx = dpif->minor;
+ request.key = key;
+ request.key_len = key_len;
+ error = dpif_linux_flow_transact(&request, &reply, &buf);
if (!error) {
if (stats) {
- odp_flow_stats_to_dpif_flow_stats(&odp_flow.stats, stats);
- }
- if (actions) {
- actions->size = odp_flow.actions_len;
- ofpbuf_trim(actions);
+ dpif_linux_flow_get_stats(&reply, stats);
}
- } else {
- if (actions) {
- ofpbuf_delete(actions);
+ if (actionsp) {
+ buf->data = (void *) reply.actions;
+ buf->size = reply.actions_len;
+ *actionsp = buf;
+ } else {
+ ofpbuf_delete(buf);
}
}
return error;
const struct nlattr *actions, size_t actions_len,
struct dpif_flow_stats *stats)
{
- struct odp_flow_put put;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
int error;
- memset(&put, 0, sizeof put);
- put.flow.key = (struct nlattr *) key;
- put.flow.key_len = key_len;
- put.flow.actions = (struct nlattr *) actions;
- put.flow.actions_len = actions_len;
- put.flow.flags = 0;
- put.flags = 0;
- if (flags & DPIF_FP_CREATE) {
- put.flags |= ODPPF_CREATE;
- }
- if (flags & DPIF_FP_MODIFY) {
- put.flags |= ODPPF_MODIFY;
- }
+ dpif_linux_flow_init(&request);
+ request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_NEW : ODP_FLOW_SET;
+ request.dp_idx = dpif->minor;
+ request.key = key;
+ request.key_len = key_len;
+ request.actions = actions;
+ request.actions_len = actions_len;
if (flags & DPIF_FP_ZERO_STATS) {
- put.flags |= ODPPF_ZERO_STATS;
+ request.clear = true;
}
- error = do_ioctl(dpif_, ODP_FLOW_PUT, &put);
+ request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE;
+ error = dpif_linux_flow_transact(&request,
+ stats ? &reply : NULL,
+ stats ? &buf : NULL);
if (!error && stats) {
- odp_flow_stats_to_dpif_flow_stats(&put.flow.stats, stats);
+ dpif_linux_flow_get_stats(&reply, stats);
+ ofpbuf_delete(buf);
}
return error;
}
const struct nlattr *key, size_t key_len,
struct dpif_flow_stats *stats)
{
- struct odp_flow odp_flow;
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_flow request, reply;
+ struct ofpbuf *buf;
int error;
- memset(&odp_flow, 0, sizeof odp_flow);
- odp_flow.key = (struct nlattr *) key;
- odp_flow.key_len = key_len;
- error = do_ioctl(dpif_, ODP_FLOW_DEL, &odp_flow);
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_DEL;
+ request.dp_idx = dpif->minor;
+ request.key = key;
+ request.key_len = key_len;
+ error = dpif_linux_flow_transact(&request,
+ stats ? &reply : NULL,
+ stats ? &buf : NULL);
if (!error && stats) {
- odp_flow_stats_to_dpif_flow_stats(&odp_flow.stats, stats);
+ dpif_linux_flow_get_stats(&reply, stats);
+ ofpbuf_delete(buf);
}
return error;
}
struct dpif_linux_flow_state {
- struct odp_flow_dump dump;
- struct odp_flow flow;
- uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S];
- uint32_t actionsbuf[65536 / sizeof(uint32_t)];
+ struct dpif_linux_flow flow;
+ struct ofpbuf *buf;
struct dpif_flow_stats stats;
};
static int
dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
{
- struct dpif_linux_flow_state *state;
-
- *statep = state = xmalloc(sizeof *state);
- state->dump.state[0] = 0;
- state->dump.state[1] = 0;
- state->dump.flow = &state->flow;
+ *statep = xzalloc(sizeof(struct dpif_linux_flow_state));
return 0;
}
static int
-dpif_linux_flow_dump_next(const struct dpif *dpif, void *state_,
+dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_,
const struct nlattr **key, size_t *key_len,
const struct nlattr **actions, size_t *actions_len,
const struct dpif_flow_stats **stats)
{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_flow_state *state = state_;
+ struct ofpbuf *old_buf = state->buf;
+ struct dpif_linux_flow request;
int error;
- memset(&state->flow, 0, sizeof state->flow);
- state->flow.key = (struct nlattr *) state->keybuf;
- state->flow.key_len = sizeof state->keybuf;
- if (actions) {
- state->flow.actions = (struct nlattr *) state->actionsbuf;
- state->flow.actions_len = sizeof state->actionsbuf;
- }
+ dpif_linux_flow_init(&request);
+ request.cmd = ODP_FLOW_DUMP;
+ request.dp_idx = dpif->minor;
+ request.state = state->flow.state;
+ error = dpif_linux_flow_transact(&request, &state->flow, &state->buf);
+ ofpbuf_delete(old_buf);
- error = do_ioctl(dpif, ODP_FLOW_DUMP, &state->dump);
if (!error) {
- if (state->flow.flags & ODPFF_EOF) {
- return EOF;
- }
if (key) {
- *key = (const struct nlattr *) state->keybuf;
+ *key = state->flow.key;
*key_len = state->flow.key_len;
}
if (actions) {
- *actions = (const struct nlattr *) state->actionsbuf;
+ *actions = state->flow.actions;
*actions_len = state->flow.actions_len;
}
if (stats) {
- odp_flow_stats_to_dpif_flow_stats(&state->flow.stats,
- &state->stats);
+ dpif_linux_flow_get_stats(&state->flow, &state->stats);
*stats = &state->stats;
}
}
- return error;
+ return error == ENODEV ? EOF : error;
}
static int
-dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state)
+dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
{
+ struct dpif_linux_flow_state *state = state_;
+
+ ofpbuf_delete(state->buf);
free(state);
return 0;
}
static int
dpif_linux_execute(struct dpif *dpif_,
const struct nlattr *actions, size_t actions_len,
- const struct ofpbuf *buf)
+ const struct ofpbuf *packet)
{
- struct odp_execute execute;
- memset(&execute, 0, sizeof execute);
- execute.actions = (struct nlattr *) actions;
- execute.actions_len = actions_len;
- execute.data = buf->data;
- execute.length = buf->size;
- return do_ioctl(dpif_, ODP_EXECUTE, &execute);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct odp_header *execute;
+ struct ofpbuf *buf;
+ int error;
+
+ buf = ofpbuf_new(128 + actions_len + packet->size);
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST,
+ ODP_PACKET_CMD_EXECUTE, 1);
+
+ execute = ofpbuf_put_uninit(buf, sizeof *execute);
+ execute->dp_idx = dpif->minor;
+
+ nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size);
+ nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len);
+
+ error = nl_sock_transact(genl_sock, buf, NULL);
+ ofpbuf_delete(buf);
+ return error;
}
static int
dpif_linux_recv_get_mask(const struct dpif *dpif_, int *listen_mask)
{
- return do_ioctl(dpif_, ODP_GET_LISTEN_MASK, listen_mask);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ *listen_mask = dpif->listen_mask;
+ return 0;
}
static int
dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
{
- return do_ioctl(dpif_, ODP_SET_LISTEN_MASK, &listen_mask);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int error;
+ int i;
+
+ if (listen_mask == dpif->listen_mask) {
+ return 0;
+ } else if (!listen_mask) {
+ nl_sock_destroy(dpif->mc_sock);
+ dpif->mc_sock = NULL;
+ dpif->listen_mask = 0;
+ return 0;
+ } else if (!dpif->mc_sock) {
+ error = nl_sock_create(NETLINK_GENERIC, &dpif->mc_sock);
+ if (error) {
+ return error;
+ }
+ }
+
+ /* Unsubscribe from old groups. */
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ if (dpif->listen_mask & (1u << i)) {
+ nl_sock_leave_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
+ }
+ }
+
+ /* Update listen_mask. */
+ dpif->listen_mask = listen_mask;
+
+ /* Subscribe to new groups. */
+ error = 0;
+ for (i = 0; i < DPIF_N_UC_TYPES; i++) {
+ if (dpif->listen_mask & (1u << i)) {
+ int retval;
+
+ retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
+ if (retval) {
+ error = retval;
+ }
+ }
+ }
+ return error;
}
static int
dpif_linux_get_sflow_probability(const struct dpif *dpif_,
uint32_t *probability)
{
- return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability);
+ struct dpif_linux_dp dp;
+ struct ofpbuf *buf;
+ int error;
+
+ error = dpif_linux_dp_get(dpif_, &dp, &buf);
+ if (!error) {
+ *probability = dp.sampling ? *dp.sampling : 0;
+ ofpbuf_delete(buf);
+ }
+ return error;
}
static int
dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
{
- return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability);
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp dp;
+
+ dpif_linux_dp_init(&dp);
+ dp.cmd = ODP_DP_CMD_SET;
+ dp.dp_idx = dpif->minor;
+ dp.sampling = &probability;
+ return dpif_linux_dp_transact(&dp, NULL, NULL);
}
static int
}
static int
-parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall)
+parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall,
+ uint32_t *dp_idx)
{
static const struct nl_policy odp_packet_policy[] = {
/* Always present. */
- [ODP_PACKET_ATTR_TYPE] = { .type = NL_A_U32 },
[ODP_PACKET_ATTR_PACKET] = { .type = NL_A_UNSPEC,
.min_len = ETH_HEADER_LEN },
[ODP_PACKET_ATTR_KEY] = { .type = NL_A_NESTED },
- /* _ODPL_ACTION_NR only. */
+ /* ODP_PACKET_CMD_ACTION only. */
[ODP_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
- /* _ODPL_SFLOW_NR only. */
+ /* ODP_PACKET_CMD_SAMPLE only. */
[ODP_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
[ODP_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
};
- struct odp_packet *odp_packet = buf->data;
+ struct odp_header *odp_header;
struct nlattr *a[ARRAY_SIZE(odp_packet_policy)];
+ struct nlmsghdr *nlmsg;
+ struct genlmsghdr *genl;
+ struct ofpbuf b;
+ int type;
+
+ ofpbuf_use_const(&b, buf->data, buf->size);
+
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_packet_family
+ || !nl_policy_parse(&b, 0, odp_packet_policy, a,
+ ARRAY_SIZE(odp_packet_policy))) {
+ return EINVAL;
+ }
- if (!nl_policy_parse(buf, sizeof *odp_packet, odp_packet_policy,
- a, ARRAY_SIZE(odp_packet_policy))) {
+ type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS
+ : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION
+ : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
+ : -1);
+ if (type < 0) {
return EINVAL;
}
memset(upcall, 0, sizeof *upcall);
- upcall->type = nl_attr_get_u32(a[ODP_PACKET_ATTR_TYPE]);
+ upcall->type = type;
upcall->packet = buf;
upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]);
upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]);
upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]);
}
+ *dp_idx = odp_header->dp_idx;
+
return 0;
}
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct ofpbuf *buf;
- int retval;
int error;
+ int i;
- buf = ofpbuf_new(65536);
- retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf));
- if (retval < 0) {
- error = errno;
- if (error != EAGAIN) {
- VLOG_WARN_RL(&error_rl, "%s: read failed: %s",
- dpif_name(dpif_), strerror(error));
+ if (!dpif->mc_sock) {
+ return EAGAIN;
+ }
+
+ for (i = 0; i < 50; i++) {
+ uint32_t dp_idx;
+
+ error = nl_sock_recv(dpif->mc_sock, &buf, false);
+ if (error) {
+ return error;
}
- } else if (retval >= sizeof(struct odp_packet)) {
- struct odp_packet *odp_packet = buf->data;
- buf->size += retval;
- if (odp_packet->len <= retval) {
- error = parse_odp_packet(buf, upcall);
- } else {
- VLOG_WARN_RL(&error_rl, "%s: discarding message truncated "
- "from %"PRIu32" bytes to %d",
- dpif_name(dpif_), odp_packet->len, retval);
- error = ERANGE;
+ error = parse_odp_packet(buf, upcall, &dp_idx);
+ if (!error
+ && dp_idx == dpif->minor
+ && dpif->listen_mask & (1u << upcall->type)) {
+ return 0;
}
- } else if (!retval) {
- VLOG_WARN_RL(&error_rl, "%s: unexpected end of file", dpif_name(dpif_));
- error = EPROTO;
- } else {
- VLOG_WARN_RL(&error_rl, "%s: discarding too-short message (%d bytes)",
- dpif_name(dpif_), retval);
- error = ERANGE;
- }
- if (error) {
ofpbuf_delete(buf);
+ if (error) {
+ return error;
+ }
}
- return error;
+
+ return EAGAIN;
}
static void
dpif_linux_recv_wait(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- poll_fd_wait(dpif->fd, POLLIN);
+ if (dpif->mc_sock) {
+ nl_sock_wait(dpif->mc_sock, POLLIN);
+ }
}
static void
dpif_linux_recv_purge(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int i;
- /* This is somewhat bogus because it assumes that the following macros have
- * fixed values, but it's going to go away later. */
-#define DP_N_QUEUES 3
-#define DP_MAX_QUEUE_LEN 100
- for (i = 0; i < DP_N_QUEUES * DP_MAX_QUEUE_LEN; i++) {
- /* Reading even 1 byte discards a whole datagram and saves time. */
- char buffer;
-
- if (read(dpif->fd, &buffer, 1) != 1) {
- break;
- }
+ if (dpif->mc_sock) {
+ nl_sock_drain(dpif->mc_sock);
}
}
const struct dpif_class dpif_linux_class = {
"system",
- NULL,
- NULL,
+ NULL, /* run */
+ NULL, /* wait */
dpif_linux_enumerate,
dpif_linux_open,
dpif_linux_close,
dpif_linux_recv_purge,
};
\f
-static int get_openvswitch_major(void);
static int get_major(const char *target);
static int
-do_ioctl(const struct dpif *dpif_, int cmd, const void *arg)
+dpif_linux_init(void)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- return ioctl(dpif->fd, cmd, arg) ? errno : 0;
+ static int error = -1;
+
+ if (error < 0) {
+ error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY,
+ &odp_datapath_family);
+ if (!error) {
+ error = nl_lookup_genl_family(ODP_PACKET_FAMILY,
+ &odp_packet_family);
+ }
+ if (!error) {
+ error = nl_sock_create(NETLINK_GENERIC, &genl_sock);
+ }
+ }
+
+ return error;
}
bool
return -ENODEV;
}
-static int
-create_minor(const char *name, int minor)
-{
- int error;
- int fd;
-
- error = open_minor(minor, &fd);
- if (error) {
- return error;
- }
-
- error = ioctl(fd, ODP_DP_CREATE, name) ? errno : 0;
- close(fd);
- return error;
-}
-
static int
open_minor(int minor, int *fdp)
{
dpif->change_error = true;
}
}
+
+static int
+get_dp0_fd(int *dp0_fdp)
+{
+ static int dp0_fd = -1;
+ if (dp0_fd < 0) {
+ int error;
+ int fd;
+
+ error = open_minor(0, &fd);
+ if (error) {
+ return error;
+ }
+ dp0_fd = fd;
+ }
+ *dp0_fdp = dp0_fd;
+ return 0;
+}
\f
/* Parses the contents of 'buf', which contains a "struct odp_vport" followed
* by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a
struct dpif_linux_vport *reply,
struct ofpbuf **bufp)
{
- static int dp0_fd = -1;
struct ofpbuf *buf = NULL;
int error;
+ int fd;
assert((reply != NULL) == (bufp != NULL));
- if (dp0_fd < 0) {
- int fd;
- error = open_minor(0, &fd);
- if (error) {
- goto error;
- }
- dp0_fd = fd;
+ error = get_dp0_fd(&fd);
+ if (error) {
+ goto error;
}
buf = ofpbuf_new(1024);
dpif_linux_vport_to_ofpbuf(request, buf);
- error = ioctl(dp0_fd, request->cmd, buf->data) ? errno : 0;
+ error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
if (error) {
goto error;
}
return dpif_linux_vport_transact(&request, reply, bufp);
}
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_header" followed
+ * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a
+ * positive errno value.
+ *
+ * 'dp' will contain pointers into 'buf', so the caller should not free 'buf'
+ * while 'dp' is still in use. */
+static int
+dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf)
+{
+ static const struct nl_policy odp_datapath_policy[] = {
+ [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ },
+ [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC,
+ .min_len = sizeof(struct odp_stats),
+ .max_len = sizeof(struct odp_stats),
+ .optional = true },
+ [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
+ [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
+ [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
+ };
+
+ struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)];
+ struct odp_header *odp_header;
+ struct nlmsghdr *nlmsg;
+ struct genlmsghdr *genl;
+ struct ofpbuf b;
+
+ dpif_linux_dp_init(dp);
+
+ ofpbuf_use_const(&b, buf->data, buf->size);
+ nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+ genl = ofpbuf_try_pull(&b, sizeof *genl);
+ odp_header = ofpbuf_try_pull(&b, sizeof *odp_header);
+ if (!nlmsg || !genl || !odp_header
+ || nlmsg->nlmsg_type != odp_datapath_family
+ || !nl_policy_parse(&b, 0, odp_datapath_policy, a,
+ ARRAY_SIZE(odp_datapath_policy))) {
+ return EINVAL;
+ }
+
+ dp->cmd = genl->cmd;
+ dp->dp_idx = odp_header->dp_idx;
+ dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]);
+ if (a[ODP_DP_ATTR_STATS]) {
+ /* Can't use structure assignment because Netlink doesn't ensure
+ * sufficient alignment for 64-bit members. */
+ memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]),
+ sizeof dp->stats);
+ }
+ if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
+ dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
+ }
+ if (a[ODP_DP_ATTR_SAMPLING]) {
+ dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]);
+ }
+
+ if (a[ODP_DP_ATTR_MCGROUPS]) {
+ static const struct nl_policy odp_mcgroup_policy[] = {
+ [ODP_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
+ [ODP_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
+ };
+
+ struct nlattr *mcgroups[ARRAY_SIZE(odp_mcgroup_policy)];
+
+ if (!nl_parse_nested(a[ODP_DP_ATTR_MCGROUPS], odp_mcgroup_policy,
+ mcgroups, ARRAY_SIZE(odp_mcgroup_policy))) {
+ return EINVAL;
+ }
+
+ if (mcgroups[ODP_PACKET_CMD_MISS]) {
+ dp->mcgroups[DPIF_UC_MISS]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_MISS]);
+ }
+ if (mcgroups[ODP_PACKET_CMD_ACTION]) {
+ dp->mcgroups[DPIF_UC_ACTION]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_ACTION]);
+ }
+ if (mcgroups[ODP_PACKET_CMD_SAMPLE]) {
+ dp->mcgroups[DPIF_UC_SAMPLE]
+ = nl_attr_get_u32(mcgroups[ODP_PACKET_CMD_SAMPLE]);
+ }
+ }
+
+ return 0;
+}
+
+/* Appends to 'buf' the Generic Netlink message described by 'dp'. */
+static void
+dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf)
+{
+ struct odp_header *odp_header;
+
+ nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family,
+ NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1);
+
+ odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header);
+ odp_header->dp_idx = dp->dp_idx;
+
+ if (dp->name) {
+ nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name);
+ }
+
+ /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */
+
+ if (dp->ipv4_frags) {
+ nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags);
+ }
+
+ if (dp->sampling) {
+ nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling);
+ }
+}
+
+/* Clears 'dp' to "empty" values. */
+void
+dpif_linux_dp_init(struct dpif_linux_dp *dp)
+{
+ memset(dp, 0, sizeof *dp);
+ dp->dp_idx = -1;
+}
+
+static void
+dpif_linux_dp_dump_start(struct nl_dump *dump)
+{
+ struct dpif_linux_dp request;
+ struct ofpbuf *buf;
+
+ dpif_linux_dp_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(&request, buf);
+ nl_dump_start(dump, genl_sock, buf);
+ ofpbuf_delete(buf);
+}
+
+/* Executes 'request' in the kernel datapath. If the command fails, returns a
+ * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
+ * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
+ * result of the command is expected to be of the same form, which is decoded
+ * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
+ * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
+int
+dpif_linux_dp_transact(const struct dpif_linux_dp *request,
+ struct dpif_linux_dp *reply, struct ofpbuf **bufp)
+{
+ struct ofpbuf *request_buf;
+ int error;
+
+ assert((reply != NULL) == (bufp != NULL));
+
+ request_buf = ofpbuf_new(1024);
+ dpif_linux_dp_to_ofpbuf(request, request_buf);
+ error = nl_sock_transact(genl_sock, request_buf, bufp);
+ ofpbuf_delete(request_buf);
+
+ if (reply) {
+ if (!error) {
+ error = dpif_linux_dp_from_ofpbuf(reply, *bufp);
+ }
+ if (error) {
+ dpif_linux_dp_init(reply);
+ ofpbuf_delete(*bufp);
+ *bufp = NULL;
+ }
+ }
+ return error;
+}
+
+/* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'.
+ * The caller must free '*bufp' when the reply is no longer needed ('reply'
+ * will contain pointers into '*bufp'). */
+int
+dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply,
+ struct ofpbuf **bufp)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_dp request;
+
+ dpif_linux_dp_init(&request);
+ request.cmd = ODP_DP_CMD_GET;
+ request.dp_idx = dpif->minor;
+
+ return dpif_linux_dp_transact(&request, reply, bufp);
+}
+\f
+/* Parses the contents of 'buf', which contains a "struct odp_flow" followed by
+ * Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a
+ * positive errno value.
+ *
+ * 'flow' will contain pointers into 'buf', so the caller should not free 'buf'
+ * while 'flow' is still in use. */
+static int
+dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow,
+ const struct ofpbuf *buf)
+{
+ static const struct nl_policy odp_flow_policy[] = {
+ [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED },
+ [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
+ [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC,
+ .min_len = sizeof(struct odp_flow_stats),
+ .max_len = sizeof(struct odp_flow_stats),
+ .optional = true },
+ [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true },
+ [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true },
+ /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */
+ [ODP_FLOW_ATTR_STATE] = { .type = NL_A_U64, .optional = true },
+ };
+
+ struct odp_flow *odp_flow;
+ struct nlattr *a[ARRAY_SIZE(odp_flow_policy)];
+
+ dpif_linux_flow_init(flow);
+
+ if (!nl_policy_parse(buf, sizeof *odp_flow, odp_flow_policy,
+ a, ARRAY_SIZE(odp_flow_policy))) {
+ return EINVAL;
+ }
+ odp_flow = buf->data;
+
+ flow->nlmsg_flags = odp_flow->nlmsg_flags;
+ flow->dp_idx = odp_flow->dp_idx;
+ flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]);
+ flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]);
+ if (a[ODP_FLOW_ATTR_ACTIONS]) {
+ flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]);
+ flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]);
+ }
+ if (a[ODP_FLOW_ATTR_STATS]) {
+ flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]);
+ }
+ if (a[ODP_FLOW_ATTR_TCP_FLAGS]) {
+ flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]);
+ }
+ if (a[ODP_FLOW_ATTR_STATE]) {
+ flow->state = nl_attr_get(a[ODP_FLOW_ATTR_STATE]);
+ }
+ return 0;
+}
+
+/* Appends to 'buf' (which must initially be empty) a "struct odp_flow"
+ * followed by Netlink attributes corresponding to 'flow'. */
+static void
+dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow,
+ struct ofpbuf *buf)
+{
+ struct odp_flow *odp_flow;
+
+ ofpbuf_reserve(buf, sizeof odp_flow);
+
+ if (flow->key_len) {
+ nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len);
+ }
+
+ if (flow->actions_len) {
+ nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS,
+ flow->actions, flow->actions_len);
+ }
+
+ /* We never need to send these to the kernel. */
+ assert(!flow->stats);
+ assert(!flow->tcp_flags);
+ assert(!flow->used);
+
+ if (flow->clear) {
+ nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR);
+ }
+
+ if (flow->state) {
+ nl_msg_put_u64(buf, ODP_FLOW_ATTR_STATE,
+ get_unaligned_u64(flow->state));
+ }
+
+ odp_flow = ofpbuf_push_uninit(buf, sizeof *odp_flow);
+ odp_flow->nlmsg_flags = flow->nlmsg_flags;
+ odp_flow->dp_idx = flow->dp_idx;
+ odp_flow->len = buf->size;
+ odp_flow->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data;
+}
+
+/* Clears 'flow' to "empty" values. */
+void
+dpif_linux_flow_init(struct dpif_linux_flow *flow)
+{
+ memset(flow, 0, sizeof *flow);
+}
+
+/* Executes 'request' in the kernel datapath. If the command fails, returns a
+ * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0
+ * without doing anything else. If 'reply' and 'bufp' are nonnull, then the
+ * result of the command is expected to be an odp_flow also, which is decoded
+ * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the
+ * reply is no longer needed ('reply' will contain pointers into '*bufp'). */
+int
+dpif_linux_flow_transact(const struct dpif_linux_flow *request,
+ struct dpif_linux_flow *reply, struct ofpbuf **bufp)
+{
+ struct ofpbuf *buf = NULL;
+ int error;
+ int fd;
+
+ assert((reply != NULL) == (bufp != NULL));
+
+ error = get_dp0_fd(&fd);
+ if (error) {
+ goto error;
+ }
+
+ buf = ofpbuf_new(1024);
+ dpif_linux_flow_to_ofpbuf(request, buf);
+
+ error = ioctl(fd, request->cmd, buf->data) ? errno : 0;
+ if (error) {
+ goto error;
+ }
+
+ if (bufp) {
+ buf->size = ((struct odp_flow *) buf->data)->len;
+ error = dpif_linux_flow_from_ofpbuf(reply, buf);
+ if (error) {
+ goto error;
+ }
+ *bufp = buf;
+ } else {
+ ofpbuf_delete(buf);
+ }
+ return 0;
+
+error:
+ ofpbuf_delete(buf);
+ if (bufp) {
+ memset(reply, 0, sizeof *reply);
+ *bufp = NULL;
+ }
+ return error;
+}
+
+static void
+dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow,
+ struct dpif_flow_stats *stats)
+{
+ if (flow->stats) {
+ stats->n_packets = get_unaligned_u64(&flow->stats->n_packets);
+ stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes);
+ } else {
+ stats->n_packets = 0;
+ stats->n_bytes = 0;
+ }
+ stats->used = flow->used ? get_unaligned_u64(flow->used) : 0;
+ stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0;
+}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff