#include "flow.h"
#include "hmap.h"
#include "list.h"
+#include "meta-flow.h"
#include "netdev.h"
#include "netdev-vport.h"
#include "netlink.h"
static int do_del_port(struct dp_netdev *, odp_port_t port_no);
static int dpif_netdev_open(const struct dpif_class *, const char *name,
bool create, struct dpif **);
-static int dp_netdev_output_userspace(struct dp_netdev *, const struct ofpbuf *,
+static int dp_netdev_output_userspace(struct dp_netdev *, struct ofpbuf *,
int queue_no, const struct flow *,
const struct nlattr *userdata);
-static void dp_netdev_execute_actions(struct dp_netdev *,
- struct ofpbuf *, struct flow *,
+static void dp_netdev_execute_actions(struct dp_netdev *, const struct flow *,
+ struct ofpbuf *,
const struct nlattr *actions,
size_t actions_len);
static void dp_netdev_port_input(struct dp_netdev *dp,
struct dp_netdev_port *port,
- struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t pkt_mark, const struct flow_tnl *tnl);
+ struct ofpbuf *packet);
static struct dpif_netdev *
dpif_netdev_cast(const struct dpif *dpif)
dp->queues[i].head = dp->queues[i].tail = 0;
}
dp->queue_seq = seq_create();
- classifier_init(&dp->cls);
+ classifier_init(&dp->cls, NULL);
hmap_init(&dp->flow_table);
list_init(&dp->port_list);
dp->port_seq = seq_create();
while (q->tail != q->head) {
struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
+ ofpbuf_uninit(&u->upcall.packet);
ofpbuf_uninit(&u->buf);
}
}
stats->n_hit = dp->n_hit;
stats->n_missed = dp->n_missed;
stats->n_lost = dp->n_lost;
- stats->n_masks = UINT64_MAX;
+ stats->n_masks = UINT32_MAX;
stats->n_mask_hit = UINT64_MAX;
ovs_mutex_unlock(&dp_netdev_mutex);
static void
dp_netdev_free_flow(struct dp_netdev *dp, struct dp_netdev_flow *netdev_flow)
{
- ovs_rwlock_wrlock(&dp->cls.rwlock);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
classifier_remove(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ fat_rwlock_unlock(&dp->cls.rwlock);
cls_rule_destroy(&netdev_flow->cr);
hmap_remove(&dp->flow_table, &netdev_flow->node);
{
struct cls_rule *cr;
- ovs_rwlock_wrlock(&dp->cls.rwlock);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
cr = classifier_lookup(&dp->cls, flow, NULL);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ fat_rwlock_unlock(&dp->cls.rwlock);
return (cr
? CONTAINER_OF(cr, struct dp_netdev_flow, cr)
}
static int
-dpif_netdev_flow_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
- const struct nlattr *mask_key,
- uint32_t mask_key_len, struct flow *flow,
- struct flow *mask)
+dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ const struct nlattr *mask_key,
+ uint32_t mask_key_len, const struct flow *flow,
+ struct flow *mask)
+{
+ if (mask_key_len) {
+ if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
+ /* This should not happen: it indicates that
+ * odp_flow_key_from_mask() and odp_flow_key_to_mask()
+ * disagree on the acceptable form of a mask. Log the problem
+ * as an error, with enough details to enable debugging. */
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+
+ if (!VLOG_DROP_ERR(&rl)) {
+ struct ds s;
+
+ ds_init(&s);
+ odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
+ true);
+ VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
+ ds_destroy(&s);
+ }
+
+ return EINVAL;
+ }
+ /* Force unwildcard the in_port. */
+ mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
+ } else {
+ enum mf_field_id id;
+ /* No mask key, unwildcard everything except fields whose
+ * prerequisities are not met. */
+ memset(mask, 0x0, sizeof *mask);
+
+ for (id = 0; id < MFF_N_IDS; ++id) {
+ /* Skip registers and metadata. */
+ if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
+ && id != MFF_METADATA) {
+ const struct mf_field *mf = mf_from_id(id);
+ if (mf_are_prereqs_ok(mf, flow)) {
+ mf_mask_field(mf, mask);
+ }
+ }
+ }
+ }
+
+ return 0;
+}
+
+static int
+dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
+ struct flow *flow)
{
odp_port_t in_port;
- if (odp_flow_key_to_flow(key, key_len, flow)
- || (mask_key
- && odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow))) {
+ if (odp_flow_key_to_flow(key, key_len, flow)) {
/* This should not happen: it indicates that odp_flow_key_from_flow()
- * and odp_flow_key_to_flow() disagree on the acceptable form of a flow
- * or odp_flow_key_from_mask() and odp_flow_key_to_mask() disagree on
- * the acceptable form of a mask. Log the problem as an error, with
- * enough details to enable debugging. */
+ * and odp_flow_key_to_flow() disagree on the acceptable form of a
+ * flow. Log the problem as an error, with enough details to enable
+ * debugging. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
if (!VLOG_DROP_ERR(&rl)) {
struct ds s;
ds_init(&s);
- odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
- true);
+ odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
ds_destroy(&s);
}
return EINVAL;
}
- if (mask_key) {
- /* Force unwildcard the in_port. */
- mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
- }
-
in_port = flow->in_port.odp_port;
if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
return EINVAL;
return 0;
}
-static int
-dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
- struct flow *flow)
-{
- return dpif_netdev_flow_mask_from_nlattrs(key, key_len, NULL, 0, flow,
- NULL);
-}
-
static int
dpif_netdev_flow_get(const struct dpif *dpif,
const struct nlattr *nl_key, size_t nl_key_len,
match_init(&match, flow, wc);
cls_rule_init(&netdev_flow->cr, &match, NETDEV_RULE_PRIORITY);
- ovs_rwlock_wrlock(&dp->cls.rwlock);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
classifier_insert(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ fat_rwlock_unlock(&dp->cls.rwlock);
error = set_flow_actions(netdev_flow, actions, actions_len);
if (error) {
- ovs_rwlock_wrlock(&dp->cls.rwlock);
+ fat_rwlock_wrlock(&dp->cls.rwlock);
classifier_remove(&dp->cls, &netdev_flow->cr);
- ovs_rwlock_unlock(&dp->cls.rwlock);
+ fat_rwlock_unlock(&dp->cls.rwlock);
cls_rule_destroy(&netdev_flow->cr);
free(netdev_flow);
struct flow_wildcards wc;
int error;
- error = dpif_netdev_flow_mask_from_nlattrs(put->key, put->key_len,
- put->mask, put->mask_len, &flow, &wc.masks);
+ error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
+ if (error) {
+ return error;
+ }
+ error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
+ put->mask, put->mask_len,
+ &flow, &wc.masks);
if (error) {
return error;
}
dpif_netdev_execute(struct dpif *dpif, const struct dpif_execute *execute)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- struct ofpbuf copy;
- struct flow key;
+ struct flow md;
int error;
if (execute->packet->size < ETH_HEADER_LEN ||
return EINVAL;
}
- /* Make a deep copy of 'packet', because we might modify its data. */
- ofpbuf_init(©, DP_NETDEV_HEADROOM + execute->packet->size);
- ofpbuf_reserve(©, DP_NETDEV_HEADROOM);
- ofpbuf_put(©, execute->packet->data, execute->packet->size);
-
- flow_extract(©, 0, 0, NULL, NULL, &key);
- error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
- &key);
+ /* Get packet metadata. */
+ error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len, &md);
if (!error) {
+ struct flow key;
+
+ /* Extract flow key. */
+ flow_extract(execute->packet, md.skb_priority, md.pkt_mark, &md.tunnel,
+ &md.in_port, &key);
ovs_mutex_lock(&dp_netdev_mutex);
- dp_netdev_execute_actions(dp, ©, &key,
+ dp_netdev_execute_actions(dp, &key, execute->packet,
execute->actions, execute->actions_len);
ovs_mutex_unlock(&dp_netdev_mutex);
}
-
- ofpbuf_uninit(©);
return error;
}
struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
*upcall = u->upcall;
- upcall->packet = buf;
ofpbuf_uninit(buf);
*buf = u->buf;
\f
static void
dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
- const struct ofpbuf *packet)
+ const struct ofpbuf *packet,
+ const struct flow *key)
{
netdev_flow->used = time_msec();
netdev_flow->packet_count++;
netdev_flow->byte_count += packet->size;
- netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
+ netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, key);
}
static void
dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
- struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t pkt_mark, const struct flow_tnl *tnl)
+ struct ofpbuf *packet)
{
struct dp_netdev_flow *netdev_flow;
struct flow key;
return;
}
in_port_.odp_port = port->port_no;
- flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
+ flow_extract(packet, 0, 0, NULL, &in_port_, &key);
netdev_flow = dp_netdev_lookup_flow(dp, &key);
if (netdev_flow) {
- dp_netdev_flow_used(netdev_flow, packet);
- dp_netdev_execute_actions(dp, packet, &key,
+ dp_netdev_flow_used(netdev_flow, packet, &key);
+ dp_netdev_execute_actions(dp, &key, packet,
netdev_flow->actions,
netdev_flow->actions_len);
dp->n_hit++;
struct dp_netdev_port *port;
struct dp_netdev *dp;
struct ofpbuf packet;
+ size_t buf_size;
ovs_mutex_lock(&dp_netdev_mutex);
dp = get_dp_netdev(dpif);
- ofpbuf_init(&packet,
- DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
+ ofpbuf_init(&packet, 0);
+
+ buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu;
LIST_FOR_EACH (port, node, &dp->port_list) {
int error;
- /* Reset packet contents. */
+ /* Reset packet contents. Packet data may have been stolen. */
ofpbuf_clear(&packet);
- ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
+ ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
if (!error) {
- dp_netdev_port_input(dp, port, &packet, 0, 0, NULL);
+ dp_netdev_port_input(dp, port, &packet);
} else if (error != EAGAIN && error != EOPNOTSUPP) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
ovs_mutex_unlock(&dp_netdev_mutex);
}
-static void
-dp_netdev_output_port(void *dp_, struct ofpbuf *packet,
- const struct flow *flow OVS_UNUSED, odp_port_t out_port)
-{
- struct dp_netdev *dp = dp_;
- struct dp_netdev_port *p = dp->ports[odp_to_u32(out_port)];
- if (p) {
- netdev_send(p->netdev, packet);
- }
-}
-
static int
-dp_netdev_output_userspace(struct dp_netdev *dp, const struct ofpbuf *packet,
+dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
int queue_no, const struct flow *flow,
const struct nlattr *userdata)
{
upcall->type = queue_no;
/* Allocate buffer big enough for everything. */
- buf_size = ODPUTIL_FLOW_KEY_BYTES + 2 + packet->size;
+ buf_size = ODPUTIL_FLOW_KEY_BYTES;
if (userdata) {
buf_size += NLA_ALIGN(userdata->nla_len);
}
NLA_ALIGN(userdata->nla_len));
}
- /* Put packet.
- *
- * We adjust 'data' and 'size' in 'buf' so that only the packet itself
- * is visible in 'upcall->packet'. The ODP flow and (if present)
- * userdata become part of the headroom. */
- ofpbuf_put_zeros(buf, 2);
- buf->data = ofpbuf_put(buf, packet->data, packet->size);
- buf->size = packet->size;
- upcall->packet = buf;
+ /* Steal packet data. */
+ ovs_assert(packet->source == OFPBUF_MALLOC);
+ upcall->packet = *packet;
+ ofpbuf_use(packet, NULL, 0);
seq_change(dp->queue_seq);
}
}
+struct dp_netdev_execute_aux {
+ struct dp_netdev *dp;
+ const struct flow *key;
+};
+
+static void
+dp_netdev_action_output(void *aux_, struct ofpbuf *packet,
+ const struct flow *flow OVS_UNUSED,
+ odp_port_t out_port)
+{
+ struct dp_netdev_execute_aux *aux = aux_;
+ struct dp_netdev_port *p = aux->dp->ports[odp_to_u32(out_port)];
+ if (p) {
+ netdev_send(p->netdev, packet);
+ }
+}
+
static void
-dp_netdev_action_userspace(void *dp, struct ofpbuf *packet,
- const struct flow *key,
- const struct nlattr *a)
+dp_netdev_action_userspace(void *aux_, struct ofpbuf *packet,
+ const struct flow *flow OVS_UNUSED,
+ const struct nlattr *a, bool may_steal)
{
+ struct dp_netdev_execute_aux *aux = aux_;
const struct nlattr *userdata;
userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
- dp_netdev_output_userspace(dp, packet, DPIF_UC_ACTION, key, userdata);
+
+ /* Make a copy if we are not allowed to steal the packet's data. */
+ if (!may_steal) {
+ packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
+ }
+ dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
+ userdata);
+ if (!may_steal) {
+ ofpbuf_uninit(packet);
+ }
}
static void
-dp_netdev_execute_actions(struct dp_netdev *dp,
- struct ofpbuf *packet, struct flow *key,
- const struct nlattr *actions,
- size_t actions_len)
+dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
+ struct ofpbuf *packet,
+ const struct nlattr *actions, size_t actions_len)
{
- odp_execute_actions(dp, packet, key, actions, actions_len,
- dp_netdev_output_port, dp_netdev_action_userspace);
+ struct dp_netdev_execute_aux aux = {dp, key};
+ struct flow md = *key; /* Packet metadata, may be modified by actions. */
+
+ odp_execute_actions(&aux, packet, &md, actions, actions_len,
+ dp_netdev_action_output, dp_netdev_action_userspace);
}
const struct dpif_class dpif_netdev_class = {
projects / cascardo / ovs.git / blobdiff