/* FIXME: do we need to use GFP_ATOMIC everywhere here? */
-static void execute_actions(struct datapath *, struct sk_buff *,
- const struct sw_flow_key *,
- const struct ofp_action *, int n_actions);
static int make_writable(struct sk_buff **);
static struct sk_buff *retrieve_skb(uint32_t id);
static void discard_skb(uint32_t id);
/* 'skb' was received on 'in_port', a physical switch port between 0 and
- * OFPP_MAX. Process it according to 'chain'. */
-void fwd_port_input(struct sw_chain *chain, struct sk_buff *skb, int in_port)
+ * OFPP_MAX. Process it according to 'chain'. Returns 0 if successful, in
+ * which case 'skb' is destroyed, or -ESRCH if there is no matching flow, in
+ * which case 'skb' still belongs to the caller. */
+int run_flow_through_tables(struct sw_chain *chain, struct sk_buff *skb,
+ int in_port)
{
struct sw_flow_key key;
struct sw_flow *flow;
- flow_extract(skb, in_port, &key);
+ if (flow_extract(skb, in_port, &key)
+ && (chain->dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
+ /* Drop fragment. */
+ kfree_skb(skb);
+ return 0;
+ }
+
flow = chain_lookup(chain, &key);
if (likely(flow != NULL)) {
flow_used(flow, skb);
execute_actions(chain->dp, skb, &key,
flow->actions, flow->n_actions);
+ return 0;
} else {
+ return -ESRCH;
+ }
+}
+
+/* 'skb' was received on 'in_port', a physical switch port between 0 and
+ * OFPP_MAX. Process it according to 'chain', sending it up to the controller
+ * if no flow matches. Takes ownership of 'skb'. */
+void fwd_port_input(struct sw_chain *chain, struct sk_buff *skb, int in_port)
+{
+ if (run_flow_through_tables(chain, skb, in_port))
dp_output_control(chain->dp, skb, fwd_save_skb(skb),
- chain->dp->config.miss_send_len,
+ chain->dp->miss_send_len,
OFPR_NO_MATCH);
- }
}
static int do_output(struct datapath *dp, struct sk_buff *skb, size_t max_len,
max_len, OFPR_ACTION));
}
-static void execute_actions(struct datapath *dp, struct sk_buff *skb,
+void execute_actions(struct datapath *dp, struct sk_buff *skb,
const struct sw_flow_key *key,
const struct ofp_action *actions, int n_actions)
{
max_len = ntohs(a->arg.output.max_len);
} else {
if (!make_writable(&skb)) {
- printk("make_writable failed\n");
+ if (net_ratelimit())
+ printk("make_writable failed\n");
break;
}
skb = execute_setter(skb, eth_proto, key, a);
+ if (!skb) {
+ if (net_ratelimit())
+ printk("execute_setter lost skb\n");
+ return;
+ }
}
}
if (prev_port != -1)
static struct sk_buff *modify_vlan(struct sk_buff *skb,
const struct sw_flow_key *key, const struct ofp_action *a)
{
- uint16_t new_id = a->arg.vlan_id;
+ uint16_t new_id = ntohs(a->arg.vlan_id);
if (new_id != OFP_VLAN_NONE) {
if (key->dl_vlan != htons(OFP_VLAN_NONE)) {
/* Modify vlan id, but maintain other TCI values */
struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
vh->h_vlan_TCI = (vh->h_vlan_TCI
- & ~(htons(VLAN_VID_MASK))) | htons(new_id);
+ & ~(htons(VLAN_VID_MASK))) | a->arg.vlan_id;
} else {
/* Add vlan header */
- skb = vlan_put_tag(skb, new_id);
+
+ /* xxx The vlan_put_tag function, doesn't seem to work
+ * xxx reliably when it attempts to use the hardware-accelerated
+ * xxx version. We'll directly use the software version
+ * xxx until the problem can be diagnosed.
+ */
+ skb = __vlan_put_tag(skb, new_id);
}
} else {
/* Remove an existing vlan header if it exists */
}
static int
-recv_features_request(struct sw_chain *chain, const void *msg)
+recv_features_request(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
{
- const struct ofp_header *ofr = msg;
- return dp_send_features_reply(chain->dp, ofr->xid);
+ return dp_send_features_reply(chain->dp, sender);
}
static int
-recv_get_config_request(struct sw_chain *chain, const void *msg)
+recv_get_config_request(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
{
- const struct ofp_header *ofr = msg;
- return dp_send_config_reply(chain->dp, ofr->xid);
+ return dp_send_config_reply(chain->dp, sender);
}
static int
-recv_set_config(struct sw_chain *chain, const void *msg)
+recv_set_config(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
{
const struct ofp_switch_config *osc = msg;
- chain->dp->config = *osc;
+ int flags;
+
+ flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
+ if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
+ && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
+ flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
+ }
+ chain->dp->flags = flags;
+
+ chain->dp->miss_send_len = ntohs(osc->miss_send_len);
+
return 0;
}
static int
-recv_packet_out(struct sw_chain *chain, const void *msg)
+recv_packet_out(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
{
const struct ofp_packet_out *opo = msg;
struct sk_buff *skb;
}
static int
-recv_port_mod(struct sw_chain *chain, const void *msg)
+recv_port_mod(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
{
const struct ofp_port_mod *opm = msg;
return 0;
}
+static int
+recv_echo_request(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
+{
+ return dp_send_echo_reply(chain->dp, sender, msg);
+}
+
+static int
+recv_echo_reply(struct sw_chain *chain, const struct sender *sender,
+ const void *msg)
+{
+ return 0;
+}
+
static int
add_flow(struct sw_chain *chain, const struct ofp_flow_mod *ofm)
{
int error = -ENOMEM;
+ int i;
int n_acts;
struct sw_flow *flow;
- /* Check number of actions. */
+ /* To prevent loops, make sure there's no action to send to the
+ * OFP_TABLE virtual port.
+ */
n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
- if (n_acts > MAX_ACTIONS) {
- error = -E2BIG;
- goto error;
+ for (i=0; i<n_acts; i++) {
+ const struct ofp_action *a = &ofm->actions[i];
+
+ if (a->type == htons(OFPAT_OUTPUT)
+ && (a->arg.output.port == htons(OFPP_TABLE)
+ || a->arg.output.port == htons(OFPP_NONE))) {
+ /* xxx Send fancy new error message? */
+ goto error;
+ }
}
/* Allocate memory. */
/* Fill out flow. */
flow_extract_match(&flow->key, &ofm->match);
- flow->group_id = ntohl(ofm->group_id);
- flow->max_idle = ntohs(ofm->max_idle);
- flow->timeout = jiffies + flow->max_idle * HZ;
+ flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
+ flow->idle_timeout = ntohs(ofm->idle_timeout);
+ flow->hard_timeout = ntohs(ofm->hard_timeout);
+ flow->used = jiffies;
flow->n_actions = n_acts;
flow->init_time = jiffies;
flow->byte_count = 0;
flow->packet_count = 0;
- atomic_set(&flow->deleted, 0);
spin_lock_init(&flow->lock);
memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
}
static int
-recv_flow(struct sw_chain *chain, const void *msg)
+recv_flow(struct sw_chain *chain, const struct sender *sender, const void *msg)
{
const struct ofp_flow_mod *ofm = msg;
uint16_t command = ntohs(ofm->command);
} else if (command == OFPFC_DELETE) {
struct sw_flow_key key;
flow_extract_match(&key, &ofm->match);
- return chain_delete(chain, &key, 0) ? 0 : -ESRCH;
+ return chain_delete(chain, &key, 0, 0) ? 0 : -ESRCH;
} else if (command == OFPFC_DELETE_STRICT) {
struct sw_flow_key key;
+ uint16_t priority;
flow_extract_match(&key, &ofm->match);
- return chain_delete(chain, &key, 1) ? 0 : -ESRCH;
+ priority = key.wildcards ? ntohs(ofm->priority) : -1;
+ return chain_delete(chain, &key, priority, 1) ? 0 : -ESRCH;
} else {
return -ENOTSUPP;
}
}
-/* 'msg', which is 'length' bytes long, was received from the control path.
- * Apply it to 'chain'. */
+/* 'msg', which is 'length' bytes long, was received across Netlink from
+ * 'sender'. Apply it to 'chain'. */
int
-fwd_control_input(struct sw_chain *chain, const void *msg, size_t length)
+fwd_control_input(struct sw_chain *chain, const struct sender *sender,
+ const void *msg, size_t length)
{
struct openflow_packet {
size_t min_size;
- int (*handler)(struct sw_chain *, const void *);
+ int (*handler)(struct sw_chain *, const struct sender *,
+ const void *);
};
static const struct openflow_packet packets[] = {
sizeof (struct ofp_port_mod),
recv_port_mod,
},
+ [OFPT_ECHO_REQUEST] = {
+ sizeof (struct ofp_header),
+ recv_echo_request,
+ },
+ [OFPT_ECHO_REPLY] = {
+ sizeof (struct ofp_header),
+ recv_echo_reply,
+ },
};
const struct openflow_packet *pkt;
struct ofp_header *oh;
- if (length < sizeof(struct ofp_header))
- return -EINVAL;
-
oh = (struct ofp_header *) msg;
- if (oh->version != 1 || oh->type >= ARRAY_SIZE(packets)
+ if (oh->version != OFP_VERSION || oh->type >= ARRAY_SIZE(packets)
|| ntohs(oh->length) > length)
return -EINVAL;
if (length < pkt->min_size)
return -EFAULT;
- return pkt->handler(chain, msg);
+ return pkt->handler(chain, sender, msg);
}
/* Packet buffering. */
uint32_t fwd_save_skb(struct sk_buff *skb)
{
+ struct sk_buff *old_skb = NULL;
struct packet_buffer *p;
unsigned long int flags;
uint32_t id;
if (time_before(jiffies, p->exp_jiffies)) {
spin_unlock_irqrestore(&buffer_lock, flags);
return -1;
- } else
- kfree_skb(p->skb);
+ } else {
+ /* Defer kfree_skb() until interrupts re-enabled. */
+ old_skb = p->skb;
+ }
}
/* Don't use maximum cookie value since the all-bits-1 id is
* special. */
id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
spin_unlock_irqrestore(&buffer_lock, flags);
+ if (old_skb)
+ kfree_skb(old_skb);
+
return id;
}
return skb;
}
+void fwd_discard_all(void)
+{
+ int i;
+
+ for (i = 0; i < N_PKT_BUFFERS; i++) {
+ struct sk_buff *skb;
+ unsigned long int flags;
+
+ /* Defer kfree_skb() until interrupts re-enabled. */
+ spin_lock_irqsave(&buffer_lock, flags);
+ skb = buffers[i].skb;
+ buffers[i].skb = NULL;
+ spin_unlock_irqrestore(&buffer_lock, flags);
+
+ kfree_skb(skb);
+ }
+}
+
static void discard_skb(uint32_t id)
{
+ struct sk_buff *old_skb = NULL;
unsigned long int flags;
struct packet_buffer *p;
spin_lock_irqsave(&buffer_lock, flags);
p = &buffers[id & PKT_BUFFER_MASK];
if (p->cookie == id >> PKT_BUFFER_BITS) {
- kfree_skb(p->skb);
+ /* Defer kfree_skb() until interrupts re-enabled. */
+ old_skb = p->skb;
p->skb = NULL;
}
spin_unlock_irqrestore(&buffer_lock, flags);
+
+ if (old_skb)
+ kfree_skb(old_skb);
}
void fwd_exit(void)
{
- int i;
-
- for (i = 0; i < N_PKT_BUFFERS; i++)
- kfree_skb(buffers[i].skb);
+ fwd_discard_all();
}
/* Utility functions. */
if (skb_shared(*pskb) || skb_cloned(*pskb))
goto copy_skb;
- return pskb_may_pull(*pskb, 64); /* FIXME? */
+ return pskb_may_pull(*pskb, 40); /* FIXME? */
copy_skb:
nskb = skb_copy(*pskb, GFP_ATOMIC);