Add support for listing and deleting entries based on an output port.
[sliver-openvswitch.git] / datapath / flow.c
index ae2f787..a8c3368 100644 (file)
 #include <linux/ip.h>
 #include <linux/jiffies.h>
 #include <linux/kernel.h>
+#include <linux/llc.h>
 #include <linux/module.h>
 #include <linux/tcp.h>
 #include <linux/udp.h>
+#include <linux/icmp.h>
 #include <linux/in.h>
 #include <linux/rcupdate.h>
 #include <net/ip.h>
 
-#include "openflow.h"
+#include "openflow/openflow.h"
 #include "compat.h"
-#include "snap.h"
 
 struct kmem_cache *flow_cache;
 
@@ -114,7 +115,8 @@ void flow_extract_match(struct sw_flow_key* to, const struct ofp_match* from)
                         * network protocol is unknown. */
                        to->wildcards |= OFPFW_TP;
                } else if (from->nw_proto == IPPROTO_TCP
-                          || from->nw_proto == IPPROTO_UDP) {
+                               || from->nw_proto == IPPROTO_UDP
+                               || from->nw_proto == IPPROTO_ICMP) {
                        to->tp_src = from->tp_src;
                        to->tp_dst = from->tp_dst;
                } else {
@@ -165,12 +167,47 @@ int flow_timeout(struct sw_flow *flow)
 }
 EXPORT_SYMBOL(flow_timeout);
 
-/* Allocates and returns a new flow with 'n_actions' action, using allocation
- * flags 'flags'.  Returns the new flow or a null pointer on failure. */
-struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
+/* Returns nonzero if 'flow' contains an output action to 'out_port' or
+ * has the value OFPP_NONE. 'out_port' is in network-byte order. */
+int flow_has_out_port(struct sw_flow *flow, uint16_t out_port)
+{
+       struct sw_flow_actions *sf_acts;
+       size_t actions_len;
+       uint8_t *p;
+
+       if (out_port == htons(OFPP_NONE))
+               return 1;
+
+       sf_acts = rcu_dereference(flow->sf_acts);
+
+       actions_len = sf_acts->actions_len;
+       p = (uint8_t *)sf_acts->actions;
+
+       while (actions_len > 0) {
+               struct ofp_action_header *ah = (struct ofp_action_header *)p;
+               size_t len = ntohs(ah->len);
+
+               if (ah->type == htons(OFPAT_OUTPUT)) {
+                       struct ofp_action_output *oa = (struct ofp_action_output *)p;
+                       if (oa->port == out_port)
+                               return 1;
+               }
+
+               p += len;
+               actions_len -= len;
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(flow_has_out_port);
+
+/* Allocates and returns a new flow with room for 'actions_len' actions, 
+ * using allocation flags 'flags'.  Returns the new flow or a null pointer 
+ * on failure. */
+struct sw_flow *flow_alloc(size_t actions_len, gfp_t flags)
 {
        struct sw_flow_actions *sfa;
-       int size = sizeof *sfa + (n_actions * sizeof sfa->actions[0]);
+       size_t size = sizeof *sfa + actions_len;
        struct sw_flow *flow = kmem_cache_alloc(flow_cache, flags);
        if (unlikely(!flow))
                return NULL;
@@ -180,7 +217,7 @@ struct sw_flow *flow_alloc(int n_actions, gfp_t flags)
                kmem_cache_free(flow_cache, flow);
                return NULL;
        }
-       sfa->n_actions = n_actions;
+       sfa->actions_len = actions_len;
        flow->sf_acts = sfa;
 
        return flow;
@@ -229,19 +266,19 @@ EXPORT_SYMBOL(flow_deferred_free_acts);
 
 /* Copies 'actions' into a newly allocated structure for use by 'flow'
  * and safely frees the structure that defined the previous actions. */
-void flow_replace_acts(struct sw_flow *flow, const struct ofp_action *actions,
-                       int n_actions)
+void flow_replace_acts(struct sw_flow *flow, 
+               const struct ofp_action_header *actions, size_t actions_len)
 {
        struct sw_flow_actions *sfa;
        struct sw_flow_actions *orig_sfa = flow->sf_acts;
-       int size = sizeof *sfa + (n_actions * sizeof sfa->actions[0]);
+       size_t size = sizeof *sfa + actions_len;
 
        sfa = kmalloc(size, GFP_ATOMIC);
        if (unlikely(!sfa))
                return;
 
-       sfa->n_actions = n_actions;
-       memcpy(sfa->actions, actions, n_actions * sizeof sfa->actions[0]);
+       sfa->actions_len = actions_len;
+       memcpy(sfa->actions, actions, actions_len);
 
        rcu_assign_pointer(flow->sf_acts, sfa);
        flow_deferred_free_acts(orig_sfa);
@@ -274,10 +311,40 @@ void print_flow(const struct sw_flow_key *key)
 }
 EXPORT_SYMBOL(print_flow);
 
+#define SNAP_OUI_LEN 3
+
+struct eth_snap_hdr
+{
+       struct ethhdr eth;
+       uint8_t  dsap;  /* Always 0xAA */
+       uint8_t  ssap;  /* Always 0xAA */
+       uint8_t  ctrl;
+       uint8_t  oui[SNAP_OUI_LEN];
+       uint16_t ethertype;
+} __attribute__ ((packed));
+
+static int is_snap(const struct eth_snap_hdr *esh)
+{
+       return (esh->dsap == LLC_SAP_SNAP
+               && esh->ssap == LLC_SAP_SNAP
+               && !memcmp(esh->oui, "\0\0\0", 3));
+}
+
+static int iphdr_ok(struct sk_buff *skb)
+{
+       int nh_ofs = skb_network_offset(skb);
+       if (skb->len >= nh_ofs + sizeof(struct iphdr)) {
+               int ip_len = ip_hdrlen(skb);
+               return (ip_len >= sizeof(struct iphdr)
+                       && pskb_may_pull(skb, nh_ofs + ip_len));
+       }
+       return 0;
+}
+
 static int tcphdr_ok(struct sk_buff *skb)
 {
        int th_ofs = skb_transport_offset(skb);
-       if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
+       if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
                int tcp_len = tcp_hdrlen(skb);
                return (tcp_len >= sizeof(struct tcphdr)
                        && skb->len >= th_ofs + tcp_len);
@@ -288,7 +355,13 @@ static int tcphdr_ok(struct sk_buff *skb)
 static int udphdr_ok(struct sk_buff *skb)
 {
        int th_ofs = skb_transport_offset(skb);
-       return skb->len >= th_ofs + sizeof(struct udphdr);
+       return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
+}
+
+static int icmphdr_ok(struct sk_buff *skb)
+{
+       int th_ofs = skb_transport_offset(skb);
+       return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
 }
 
 /* Parses the Ethernet frame in 'skb', which was received on 'in_port',
@@ -297,58 +370,56 @@ static int udphdr_ok(struct sk_buff *skb)
 int flow_extract(struct sk_buff *skb, uint16_t in_port,
                 struct sw_flow_key *key)
 {
-       struct ethhdr *mac;
-       int nh_ofs, th_ofs;
+       struct ethhdr *eth;
+       struct eth_snap_hdr *esh;
        int retval = 0;
+       int nh_ofs;
 
+       memset(key, 0, sizeof *key);
+       key->dl_vlan = htons(OFP_VLAN_NONE);
        key->in_port = htons(in_port);
-       key->pad = 0;
-       key->wildcards = 0;
-       key->nw_src_mask = 0;
-       key->nw_dst_mask = 0;
-
-       /* This code doesn't check that skb->len is long enough to contain the
-        * MAC or network header.  With a 46-byte minimum length frame this
-        * assumption is always correct. */
-
-       /* Doesn't verify checksums.  Should it? */
-
-       /* Data link layer.  We only support Ethernet. */
-       mac = eth_hdr(skb);
-       nh_ofs = sizeof(struct ethhdr);
-       if (likely(ntohs(mac->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF)) {
-               /* This is an Ethernet II frame */
-               key->dl_type = mac->h_proto;
+
+       if (skb->len < sizeof *eth)
+               return 0;
+       if (!pskb_may_pull(skb, skb->len >= 64 ? 64 : skb->len)) {
+               return 0;
+       }
+
+       skb_reset_mac_header(skb);
+       eth = eth_hdr(skb);
+       esh = (struct eth_snap_hdr *) eth;
+       nh_ofs = sizeof *eth;
+       if (likely(ntohs(eth->h_proto) >= OFP_DL_TYPE_ETH2_CUTOFF))
+               key->dl_type = eth->h_proto;
+       else if (skb->len >= sizeof *esh && is_snap(esh)) {
+               key->dl_type = esh->ethertype;
+               nh_ofs = sizeof *esh;
        } else {
-               /* This is an 802.2 frame */
-               if (snap_get_ethertype(skb, &key->dl_type) != -EINVAL) {
-                       nh_ofs += sizeof(struct snap_hdr);
-               } else {
-                       key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
-                       nh_ofs += sizeof(struct llc_pdu_un);
+               key->dl_type = htons(OFP_DL_TYPE_NOT_ETH_TYPE);
+               if (skb->len >= nh_ofs + sizeof(struct llc_pdu_un)) {
+                       nh_ofs += sizeof(struct llc_pdu_un); 
                }
        }
 
        /* Check for a VLAN tag */
-       if (likely(key->dl_type != htons(ETH_P_8021Q))) {
-               key->dl_vlan = htons(OFP_VLAN_NONE);
-       } else {
-               struct vlan_hdr *vh = (struct vlan_hdr *)(skb_mac_header(skb) + nh_ofs);
+       if (key->dl_type == htons(ETH_P_8021Q) &&
+           skb->len >= nh_ofs + sizeof(struct vlan_hdr)) {
+               struct vlan_hdr *vh = (struct vlan_hdr*)(skb->data + nh_ofs);
                key->dl_type = vh->h_vlan_encapsulated_proto;
                key->dl_vlan = vh->h_vlan_TCI & htons(VLAN_VID_MASK);
-               nh_ofs += sizeof(*vh);
+               nh_ofs += sizeof(struct vlan_hdr);
        }
-       memcpy(key->dl_src, mac->h_source, ETH_ALEN);
-       memcpy(key->dl_dst, mac->h_dest, ETH_ALEN);
+       memcpy(key->dl_src, eth->h_source, ETH_ALEN);
+       memcpy(key->dl_dst, eth->h_dest, ETH_ALEN);
        skb_set_network_header(skb, nh_ofs);
 
        /* Network layer. */
-       if (likely(key->dl_type == htons(ETH_P_IP))) {
+       if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
                struct iphdr *nh = ip_hdr(skb);
+               int th_ofs = nh_ofs + nh->ihl * 4;
                key->nw_src = nh->saddr;
                key->nw_dst = nh->daddr;
                key->nw_proto = nh->protocol;
-               th_ofs = nh_ofs + nh->ihl * 4;
                skb_set_transport_header(skb, th_ofs);
 
                /* Transport layer. */
@@ -362,7 +433,7 @@ int flow_extract(struct sk_buff *skb, uint16_t in_port,
                                        /* Avoid tricking other code into
                                         * thinking that this packet has an L4
                                         * header. */
-                                       goto no_proto;
+                                       key->nw_proto = 0;
                                }
                        } else if (key->nw_proto == IPPROTO_UDP) {
                                if (udphdr_ok(skb)) {
@@ -373,28 +444,29 @@ int flow_extract(struct sk_buff *skb, uint16_t in_port,
                                        /* Avoid tricking other code into
                                         * thinking that this packet has an L4
                                         * header. */
-                                       goto no_proto;
+                                       key->nw_proto = 0;
+                               }
+                       } else if (key->nw_proto == IPPROTO_ICMP) {
+                               if (icmphdr_ok(skb)) {
+                                       struct icmphdr *icmp = icmp_hdr(skb);
+                                       /* The ICMP type and code fields use the 16-bit
+                                        * transport port fields, so we need to store them
+                                        * in 16-bit network byte order. */
+                                       key->icmp_type = htons(icmp->type);
+                                       key->icmp_code = htons(icmp->code);
+                               } else {
+                                       /* Avoid tricking other code into
+                                        * thinking that this packet has an L4
+                                        * header. */
+                                       key->nw_proto = 0;
                                }
-                       } else {
-                               goto no_th;
                        }
                } else {
                        retval = 1;
-                       goto no_th;
                }
-
-               return 0;
+       } else {
+               skb_reset_transport_header(skb);
        }
-
-       key->nw_src = 0;
-       key->nw_dst = 0;
-
-no_proto:
-       key->nw_proto = 0;
-
-no_th:
-       key->tp_src = 0;
-       key->tp_dst = 0;
        return retval;
 }