#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. */
+ 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 *);
size_t key_len;
const struct nlattr *actions; /* OVS_FLOW_ATTR_ACTIONS. */
size_t actions_len;
+ 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. */
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 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 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);
/* Loop until we find a port that isn't used. */
do {
request.port_no = dpif_linux_pop_port(dpif);
+ request.upcall_pid = get_upcall_pid_port(dpif, request.port_no);
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,
+ request.upcall_pid);
}
ofpbuf_delete(buf);
} while (request.port_no != UINT32_MAX
/* Ensure that OVS_FLOW_ATTR_ACTIONS will always be included. */
request.actions = actions ? actions : &dummy_action;
request.actions_len = actions_len;
+ request.upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
if (flags & DPIF_FP_ZERO_STATS) {
request.clear = true;
}
}
static int
-dpif_linux_execute__(int dp_ifindex,
+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)
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);
const struct ofpbuf *packet)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ uint32_t upcall_pid = get_upcall_pid_flow(dpif, key, key_len);
- return dpif_linux_execute__(dpif->dp_ifindex, key, key_len,
+ return dpif_linux_execute__(dpif->dp_ifindex, upcall_pid, key, key_len,
actions, actions_len, packet);
}
return 0;
}
-static int
-dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
+static uint32_t
+get_upcall_pid_port__(struct dpif_linux *dpif, uint32_t port)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- int error;
- int i;
+ int idx = port & (N_UPCALL_SOCKS - 1);
+ return nl_sock_pid(dpif->upcall_socks[idx]);
+}
- 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;
+static uint32_t
+get_upcall_pid_port(struct dpif_linux *dpif, uint32_t port)
+{
+ if (!(dpif->listen_mask & (1u << DPIF_UC_MISS))) {
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]);
- }
- }
+ return get_upcall_pid_port__(dpif, port);
+}
- /* Update listen_mask. */
- dpif->listen_mask = listen_mask;
+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;
- /* Subscribe to new groups. */
- error = 0;
- for (i = 0; i < DPIF_N_UC_TYPES; i++) {
- if (dpif->listen_mask & (1u << i)) {
- int retval;
+ if (!(dpif->listen_mask & (1u << DPIF_UC_ACTION))) {
+ return 0;
+ }
- retval = nl_sock_join_mcgroup(dpif->mc_sock, dpif->mcgroups[i]);
- if (retval) {
- error = retval;
- }
- }
+ nla = nl_attr_find__(key, key_len, OVS_KEY_ATTR_IN_PORT);
+ if (nla) {
+ port = nl_attr_get_u32(nla);
+ } else {
+ port = random_uint32();
}
- return error;
+
+ return get_upcall_pid_port__(dpif, port);
}
-static int
-dpif_linux_get_sflow_probability(const struct dpif *dpif_,
- uint32_t *probability)
+static void
+set_upcall_pids(struct dpif_linux *dpif)
{
- struct dpif_linux_dp dp;
- struct ofpbuf *buf;
+ struct dpif_port port;
+ struct dpif_port_dump port_dump;
+ struct dpif_flow_dump flow_dump;
+ const struct nlattr *key;
+ size_t key_len;
int error;
- error = dpif_linux_dp_get(dpif_, &dp, &buf);
- if (!error) {
- *probability = dp.sampling ? *dp.sampling : 0;
- ofpbuf_delete(buf);
+ DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
+ 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 = get_upcall_pid_port(dpif,
+ vport_request.port_no);
+ 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,
+ vport_request.upcall_pid);
+ } else {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
+ dpif_name(&dpif->dpif), strerror(error));
+ }
}
- return error;
+
+ dpif_flow_dump_start(&flow_dump, &dpif->dpif);
+ while (dpif_flow_dump_next(&flow_dump, &key, &key_len,
+ NULL, NULL, NULL)) {
+ 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 = get_upcall_pid_flow(dpif, key, key_len);
+ 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_set_sflow_probability(struct dpif *dpif_, uint32_t probability)
+dpif_linux_recv_set_mask(struct dpif *dpif_, int listen_mask)
{
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.sampling = &probability;
- return dpif_linux_dp_transact(&dp, NULL, NULL);
+ if (listen_mask == dpif->listen_mask) {
+ return 0;
+ }
+
+ if (!listen_mask) {
+ destroy_upcall_socks(dpif);
+ } else if (!dpif->listen_mask) {
+ int i;
+ int error;
+
+ 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;
+ }
+ }
+ }
+
+ dpif->listen_mask = listen_mask;
+ set_upcall_pids(dpif);
+
+ 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->listen_mask) {
+ return;
+ }
- if (dpif->mc_sock) {
- nl_sock_drain(dpif->mc_sock);
+ for (i = 0; i < N_UPCALL_SOCKS; i++) {
+ nl_sock_drain(dpif->upcall_socks[i]);
}
}
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,
ofpbuf_use_stack(&actions, &action, sizeof action);
nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, port_no);
- return dpif_linux_execute__(dp_ifindex, key.data, key.size,
+ return dpif_linux_execute__(dp_ifindex, 0, key.data, key.size,
actions.data, actions.size, &packet);
}
[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),
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_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
+ }
if (a[OVS_VPORT_ATTR_STATS]) {
vport->stats = nl_attr_get(a[OVS_VPORT_ATTR_STATS]);
}
nl_msg_put_string(buf, OVS_VPORT_ATTR_NAME, vport->name);
}
+ 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);
.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);
}
+ 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. */
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),
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_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
+ }
if (a[OVS_FLOW_ATTR_STATS]) {
flow->stats = nl_attr_get(a[OVS_FLOW_ATTR_STATS]);
}
flow->actions, flow->actions_len);
}
+ 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);
git://git.onelab.eu / sliver-openvswitch.git / blobdiff