#include "openvswitch/tunnel.h"
#include "packets.h"
#include "poll-loop.h"
+#include "random.h"
#include "shash.h"
#include "sset.h"
#include "unaligned.h"
enum { LRU_MASK = LRU_MAX_PORTS - 1};
BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
+enum { N_UPCALL_SOCKS = 16 };
+BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
+
/* This ethtool flag was introduced in Linux 2.6.24, so it might be
* missing if we have old headers. */
#define ETH_FLAG_LRO (1 << 15) /* LRO is enabled */
/* Attributes. */
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. */
- const uint32_t *sampling; /* OVS_DP_ATTR_SAMPLING. */
- uint32_t mcgroups[DPIF_N_UC_TYPES]; /* OVS_DP_ATTR_MCGROUPS. */
};
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 actions_len;
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. */
};
struct dpif dpif;
int dp_ifindex;
- /* Multicast group messages. */
- struct nl_sock *mc_sock;
- uint32_t mcgroups[DPIF_N_UC_TYPES];
+ /* Upcall messages. */
+ struct nl_sock *upcall_socks[N_UPCALL_SOCKS];
+ int last_read_upcall;
unsigned int listen_mask;
/* Change notification. */
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 dpif_linux_port_get_pid__(const struct dpif *,
+ uint16_t port_no,
+ enum dpif_upcall_type);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
struct dpif_linux *dpif;
int i;
- dpif = xmalloc(sizeof *dpif);
+ dpif = xzalloc(sizeof *dpif);
dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
dpif);
dpif_init(&dpif->dpif, &dpif_linux_class, dp->name,
dp->dp_ifindex, dp->dp_ifindex);
- dpif->mc_sock = NULL;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- dpif->mcgroups[i] = dp->mcgroups[i];
- }
- dpif->listen_mask = 0;
dpif->dp_ifindex = dp->dp_ifindex;
sset_init(&dpif->changed_ports);
- dpif->change_error = false;
*dpifp = &dpif->dpif;
- dpif->lru_head = dpif->lru_tail = 0;
dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL);
for (i = 1; i < LRU_MAX_PORTS; i++) {
}
}
+static void
+destroy_upcall_socks(struct dpif_linux *dpif)
+{
+ int i;
+
+ for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ nl_sock_destroy(dpif->upcall_socks[i]);
+ dpif->upcall_socks[i] = NULL;
+ }
+}
+
static void
dpif_linux_close(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
nln_notifier_destroy(dpif->port_notifier);
- nl_sock_destroy(dpif->mc_sock);
+ destroy_upcall_socks(dpif);
sset_destroy(&dpif->changed_ports);
free(dpif->lru_bitmap);
free(dpif);
}
static int
-dpif_linux_get_stats(const struct dpif *dpif_, struct ovs_dp_stats *stats)
+dpif_linux_get_stats(const struct dpif *dpif_, struct dpif_dp_stats *stats)
{
struct dpif_linux_dp dp;
struct ofpbuf *buf;
error = dpif_linux_dp_get(dpif_, &dp, &buf);
if (!error) {
- *stats = dp.stats;
+ 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;
+ stats->n_flows = dp.stats.n_flows;
ofpbuf_delete(buf);
}
return error;
/* Loop until we find a port that isn't used. */
do {
+ uint32_t upcall_pid;
+
request.port_no = dpif_linux_pop_port(dpif);
+ upcall_pid = dpif_linux_port_get_pid__(dpif_, request.port_no,
+ DPIF_UC_MISS);
+ request.upcall_pid = &upcall_pid;
error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
*port_nop = reply.port_no;
+ VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
+ dpif_name(dpif_), request.port_no, upcall_pid);
}
ofpbuf_delete(buf);
} while (request.port_no != UINT32_MAX
return 1024;
}
+static uint32_t
+dpif_linux_port_get_pid__(const struct dpif *dpif_, uint16_t port_no,
+ enum dpif_upcall_type upcall_type)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+
+ if (!(dpif->listen_mask & (1u << upcall_type))) {
+ return 0;
+ } else {
+ int idx = port_no & (N_UPCALL_SOCKS - 1);
+ return nl_sock_pid(dpif->upcall_socks[idx]);
+ }
+}
+
+static uint32_t
+dpif_linux_port_get_pid(const struct dpif *dpif, uint16_t port_no)
+{
+ return dpif_linux_port_get_pid__(dpif, port_no, DPIF_UC_ACTION);
+}
+
static int
dpif_linux_flow_flush(struct dpif *dpif_)
{
}
static int
-dpif_linux_execute__(int dp_ifindex,
- const struct nlattr *key, size_t key_len,
+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)
{
return 0;
}
-static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+static void
+set_upcall_pids(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_port_dump port_dump;
+ struct dpif_port port;
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;
+ DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
+ uint32_t upcall_pid = dpif_linux_port_get_pid__(dpif_, port.port_no,
+ DPIF_UC_MISS);
+ struct dpif_linux_vport vport_request;
+
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = OVS_VPORT_CMD_SET;
+ vport_request.dp_ifindex = dpif->dp_ifindex;
+ vport_request.port_no = port.port_no;
+ vport_request.upcall_pid = &upcall_pid;
+ error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
+ if (!error) {
+ VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
+ dpif_name(&dpif->dpif), vport_request.port_no,
+ upcall_pid);
+ } else {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
+ dpif_name(&dpif->dpif), strerror(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]);
- }
- }
+static int
+dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- /* Update listen_mask. */
- dpif->listen_mask = listen_mask;
+ if (listen_mask == dpif->listen_mask) {
+ return 0;
+ }
- /* Subscribe to new groups. */
- error = 0;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- if (dpif->listen_mask & (1u << i)) {
- int retval;
+ if (!listen_mask) {
+ destroy_upcall_socks(dpif);
+ } else if (!dpif->listen_mask) {
+ int i;
+ int error;
- retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
- if (retval) {
- error = retval;
+ for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ error = nl_sock_create(NETLINK_GENERIC, &dpif->upcall_socks[i]);
+ if (error) {
+ destroy_upcall_socks(dpif);
+ return error;
}
}
}
- return error;
-}
-
-static int
-dpif_linux_get_sflow_probability(const struct dpif *dpif_,
- uint32_t *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)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_dp dp;
+ dpif->listen_mask = listen_mask;
+ set_upcall_pids(dpif_);
- dpif_linux_dp_init(&dp);
- dp.cmd = OVS_DP_CMD_SET;
- dp.dp_ifindex = dpif->dp_ifindex;
- dp.sampling = &probability;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
+ return 0;
}
static int
/* OVS_PACKET_CMD_ACTION only. */
[OVS_PACKET_ATTR_USERDATA] = { .type = NL_A_U64, .optional = true },
-
- /* OVS_PACKET_CMD_SAMPLE only. */
- [OVS_PACKET_ATTR_SAMPLE_POOL] = { .type = NL_A_U32, .optional = true },
- [OVS_PACKET_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true },
};
struct ovs_header *ovs_header;
type = (genl->cmd == OVS_PACKET_CMD_MISS ? DPIF_UC_MISS
: genl->cmd == OVS_PACKET_CMD_ACTION ? DPIF_UC_ACTION
- : genl->cmd == OVS_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE
: -1);
if (type < 0) {
return EINVAL;
upcall->userdata = (a[OVS_PACKET_ATTR_USERDATA]
? nl_attr_get_u64(a[OVS_PACKET_ATTR_USERDATA])
: 0);
- upcall->sample_pool = (a[OVS_PACKET_ATTR_SAMPLE_POOL]
- ? nl_attr_get_u32(a[OVS_PACKET_ATTR_SAMPLE_POOL])
- : 0);
- if (a[OVS_PACKET_ATTR_ACTIONS]) {
- upcall->actions = (void *) nl_attr_get(a[OVS_PACKET_ATTR_ACTIONS]);
- upcall->actions_len = nl_attr_get_size(a[OVS_PACKET_ATTR_ACTIONS]);
- }
-
*dp_ifindex = ovs_header->dp_ifindex;
return 0;
dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct ofpbuf *buf;
- int error;
int i;
-
- if (!dpif->mc_sock) {
- return EAGAIN;
- }
-
- for (i = 0; i < 50; i++) {
- int dp_ifindex;
-
- error = nl_sock_recv(dpif->mc_sock, &buf, false);
- 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;
+ 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 (nl_sock_woke(upcall_sock)) {
+ int dp_ifindex;
+
+ for (;;) {
+ struct ofpbuf *buf;
+ int 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_);
- if (dpif->mc_sock) {
- nl_sock_wait(dpif->mc_sock, POLLIN);
+ int i;
+
+ if (!dpif->listen_mask) {
+ return;
+ }
+
+ for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ nl_sock_wait(dpif->upcall_socks[i], POLLIN);
}
}
dpif_linux_recv_purge(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ int i;
- if (dpif->mc_sock) {
- nl_sock_drain(dpif->mc_sock);
+ if (!dpif->listen_mask) {
+ return;
+ }
+
+ for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ nl_sock_drain(dpif->upcall_socks[i]);
}
}
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_execute,
dpif_linux_recv_get_mask,
dpif_linux_recv_set_mask,
- dpif_linux_get_sflow_probability,
- dpif_linux_set_sflow_probability,
dpif_linux_queue_to_priority,
dpif_linux_recv,
dpif_linux_recv_wait,
[OVS_VPORT_ATTR_PORT_NO] = { .type = NL_A_U32 },
[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),
.max_len = ETH_ADDR_LEN,
.optional = true },
[OVS_VPORT_ATTR_OPTIONS] = { .type = NL_A_NESTED, .optional = true },
- [OVS_VPORT_ATTR_IFINDEX] = { .type = NL_A_U32, .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_vport_policy)];
vport->port_no = nl_attr_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
vport->type = nl_attr_get_u32(a[OVS_VPORT_ATTR_TYPE]);
vport->name = nl_attr_get_string(a[OVS_VPORT_ATTR_NAME]);
+ if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
+ vport->upcall_pid = nl_attr_get(a[OVS_VPORT_ATTR_UPCALL_PID]);
+ }
if (a[OVS_VPORT_ATTR_STATS]) {
vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
}
vport->options = nl_attr_get(a[OVS_VPORT_ATTR_OPTIONS]);
vport->options_len = nl_attr_get_size(a[OVS_VPORT_ATTR_OPTIONS]);
}
- if (a[OVS_VPORT_ATTR_IFINDEX]) {
- vport->ifindex = nl_attr_get_u32(a[OVS_VPORT_ATTR_IFINDEX]);
- }
return 0;
}
nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
}
+ if (vport->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_VPORT_ATTR_UPCALL_PID, *vport->upcall_pid);
+ }
+
if (vport->stats) {
nl_msg_put_unspec(buf, OVS_VPORT_ATTR_STATS,
vport->stats, sizeof *vport->stats);
nl_msg_put_nested(buf, OVS_VPORT_ATTR_OPTIONS,
vport->options, vport->options_len);
}
-
- if (vport->ifindex) {
- nl_msg_put_u32(buf, OVS_VPORT_ATTR_IFINDEX, vport->ifindex);
- }
}
/* Clears 'vport' to "empty" values. */
.max_len = sizeof(struct ovs_dp_stats),
.optional = true },
[OVS_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true },
- [OVS_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true },
- [OVS_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true },
};
struct nlattr *a[ARRAY_SIZE(ovs_datapath_policy)];
if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
dp->ipv4_frags = nl_attr_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
}
- if (a[OVS_DP_ATTR_SAMPLING]) {
- dp->sampling = nl_attr_get(a[OVS_DP_ATTR_SAMPLING]);
- }
-
- if (a[OVS_DP_ATTR_MCGROUPS]) {
- static const struct nl_policy ovs_mcgroup_policy[] = {
- [OVS_PACKET_CMD_MISS] = { .type = NL_A_U32, .optional = true },
- [OVS_PACKET_CMD_ACTION] = { .type = NL_A_U32, .optional = true },
- [OVS_PACKET_CMD_SAMPLE] = { .type = NL_A_U32, .optional = true },
- };
-
- struct nlattr *mcgroups[ARRAY_SIZE(ovs_mcgroup_policy)];
-
- if (!nl_parse_nested(a[OVS_DP_ATTR_MCGROUPS], ovs_mcgroup_policy,
- mcgroups, ARRAY_SIZE(ovs_mcgroup_policy))) {
- return EINVAL;
- }
-
- if (mcgroups[OVS_PACKET_CMD_MISS]) {
- dp->mcgroups[DPIF_UC_MISS]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_MISS]);
- }
- if (mcgroups[OVS_PACKET_CMD_ACTION]) {
- dp->mcgroups[DPIF_UC_ACTION]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_ACTION]);
- }
- if (mcgroups[OVS_PACKET_CMD_SAMPLE]) {
- dp->mcgroups[DPIF_UC_SAMPLE]
- = nl_attr_get_u32(mcgroups[OVS_PACKET_CMD_SAMPLE]);
- }
- }
return 0;
}
nl_msg_put_string(buf, OVS_DP_ATTR_NAME, dp->name);
}
+ if (dp->upcall_pid) {
+ nl_msg_put_u32(buf, OVS_DP_ATTR_UPCALL_PID, *dp->upcall_pid);
+ }
+
/* 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);
}
-
- if (dp->sampling) {
- nl_msg_put_u32(buf, OVS_DP_ATTR_SAMPLING, *dp->sampling);
- }
}
/* Clears 'dp' to "empty" values. */
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