/*
* Distributed under the terms of the GNU GPL version 2.
- * Copyright (c) 2007, 2008, 2009 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
*
* Significant portions of this file may be copied from parts of the Linux
* kernel, by Linus Torvalds and others.
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/in6.h>
+#include <linux/if_arp.h>
#include <linux/if_vlan.h>
+#include <net/inet_ecn.h>
#include <net/ip.h>
#include <net/checksum.h>
-#include "datapath.h"
-#include "dp_dev.h"
+
#include "actions.h"
+#include "checksum.h"
+#include "datapath.h"
#include "openvswitch/datapath-protocol.h"
+#include "vlan.h"
+#include "vport.h"
-struct sk_buff *
-make_writable(struct sk_buff *skb, gfp_t gfp)
+static int do_execute_actions(struct datapath *, struct sk_buff *,
+ const struct sw_flow_key *,
+ const struct nlattr *actions, u32 actions_len);
+
+static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom)
{
- if (skb_shared(skb) || skb_cloned(skb)) {
- struct sk_buff *nskb = skb_copy(skb, gfp);
+ if (skb_cloned(skb)) {
+ struct sk_buff *nskb;
+ unsigned headroom = max(min_headroom, skb_headroom(skb));
+
+ nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), GFP_ATOMIC);
if (nskb) {
+ set_skb_csum_bits(skb, nskb);
kfree_skb(skb);
return nskb;
}
return NULL;
}
-
-static struct sk_buff *
-vlan_pull_tag(struct sk_buff *skb)
+static struct sk_buff *strip_vlan(struct sk_buff *skb)
{
- struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
struct ethhdr *eh;
+ if (vlan_tx_tag_present(skb)) {
+ vlan_set_tci(skb, 0);
+ return skb;
+ }
- /* Verify we were given a vlan packet */
- if (vh->h_vlan_proto != htons(ETH_P_8021Q))
+ if (unlikely(vlan_eth_hdr(skb)->h_vlan_proto != htons(ETH_P_8021Q) ||
+ skb->len < VLAN_ETH_HLEN))
return skb;
- memmove(skb->data + VLAN_HLEN, skb->data, 2 * VLAN_ETH_ALEN);
+ skb = make_writable(skb, 0);
+ if (unlikely(!skb))
+ return NULL;
+
+ if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
+ skb->csum = csum_sub(skb->csum, csum_partial(skb->data
+ + ETH_HLEN, VLAN_HLEN, 0));
+
+ memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
return skb;
}
-
-static struct sk_buff *
-modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
- struct odp_flow_key *key, const union odp_action *a,
- int n_actions, gfp_t gfp)
+static struct sk_buff *modify_vlan_tci(struct sk_buff *skb, __be16 tci)
{
- u16 tci, mask;
+ struct vlan_ethhdr *vh;
+ __be16 old_tci;
- if (a->type == ODPAT_SET_VLAN_VID) {
- tci = ntohs(a->vlan_vid.vlan_vid);
- mask = VLAN_VID_MASK;
- key->dl_vlan = htons(tci & mask);
- } else {
- tci = a->vlan_pcp.vlan_pcp << 13;
- mask = VLAN_PCP_MASK;
- }
+ if (vlan_tx_tag_present(skb) || skb->protocol != htons(ETH_P_8021Q))
+ return __vlan_hwaccel_put_tag(skb, ntohs(tci));
- skb = make_writable(skb, gfp);
- if (!skb)
- return ERR_PTR(-ENOMEM);
+ skb = make_writable(skb, 0);
+ if (unlikely(!skb))
+ return NULL;
- if (skb->protocol == htons(ETH_P_8021Q)) {
- /* Modify vlan id, but maintain other TCI values */
- struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
- vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci);
- } else {
- /* Add vlan header */
-
- /* Set up checksumming pointers for checksum-deferred packets
- * on Xen. Otherwise, dev_queue_xmit() will try to do this
- * when we send the packet out on the wire, and it will fail at
- * that point because skb_checksum_setup() will not look inside
- * an 802.1Q header. */
- vswitch_skb_checksum_setup(skb);
-
- /* GSO is not implemented for packets with an 802.1Q header, so
- * we have to do segmentation before we add that header.
- *
- * GSO does work with hardware-accelerated VLAN tagging, but we
- * can't use hardware-accelerated VLAN tagging since it
- * requires the device to have a VLAN group configured (with
- * e.g. vconfig(8)) and we don't do that.
- *
- * Having to do this here may be a performance loss, since we
- * can't take advantage of TSO hardware support, although it
- * does not make a measurable network performance difference
- * for 1G Ethernet. Fixing that would require patching the
- * kernel (either to add GSO support to the VLAN protocol or to
- * support hardware-accelerated VLAN tagging without VLAN
- * groups configured). */
- if (skb_is_gso(skb)) {
- struct sk_buff *segs;
-
- segs = skb_gso_segment(skb, 0);
- kfree_skb(skb);
- if (unlikely(IS_ERR(segs)))
- return ERR_CAST(segs);
-
- do {
- struct sk_buff *nskb = segs->next;
- int err;
-
- segs->next = NULL;
-
- segs = __vlan_put_tag(segs, tci);
- err = -ENOMEM;
- if (segs) {
- struct odp_flow_key segkey = *key;
- err = execute_actions(dp, segs,
- &segkey, a + 1,
- n_actions - 1,
- gfp);
- }
-
- if (unlikely(err)) {
- while ((segs = nskb)) {
- nskb = segs->next;
- segs->next = NULL;
- kfree_skb(segs);
- }
- return ERR_PTR(err);
- }
-
- segs = nskb;
- } while (segs->next);
-
- skb = segs;
- }
+ if (unlikely(skb->len < VLAN_ETH_HLEN))
+ return skb;
- /* The hardware-accelerated version of vlan_put_tag() works
- * only for a device that has a VLAN group configured (with
- * e.g. vconfig(8)), so call the software-only version
- * __vlan_put_tag() directly instead.
- */
- skb = __vlan_put_tag(skb, tci);
- if (!skb)
- return ERR_PTR(-ENOMEM);
+ vh = vlan_eth_hdr(skb);
+
+ old_tci = vh->h_vlan_TCI;
+ vh->h_vlan_TCI = tci;
+
+ if (get_ip_summed(skb) == OVS_CSUM_COMPLETE) {
+ __be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
+ skb->csum = ~csum_partial((char *)diff, sizeof(diff), ~skb->csum);
}
return skb;
}
-static struct sk_buff *strip_vlan(struct sk_buff *skb,
- struct odp_flow_key *key, gfp_t gfp)
+static bool is_ip(struct sk_buff *skb, const struct sw_flow_key *key)
{
- skb = make_writable(skb, gfp);
- if (skb) {
- vlan_pull_tag(skb);
- key->dl_vlan = htons(ODP_VLAN_NONE);
- }
- return skb;
+ return (key->dl_type == htons(ETH_P_IP) &&
+ skb->transport_header > skb->network_header);
}
-static struct sk_buff *set_dl_addr(struct sk_buff *skb,
- const struct odp_action_dl_addr *a,
- gfp_t gfp)
+static __sum16 *get_l4_checksum(struct sk_buff *skb, const struct sw_flow_key *key)
{
- skb = make_writable(skb, gfp);
- if (skb) {
- struct ethhdr *eh = eth_hdr(skb);
- memcpy(a->type == ODPAT_SET_DL_SRC ? eh->h_source : eh->h_dest,
- a->dl_addr, ETH_ALEN);
+ int transport_len = skb->len - skb_transport_offset(skb);
+ if (key->nw_proto == IPPROTO_TCP) {
+ if (likely(transport_len >= sizeof(struct tcphdr)))
+ return &tcp_hdr(skb)->check;
+ } else if (key->nw_proto == IPPROTO_UDP) {
+ if (likely(transport_len >= sizeof(struct udphdr)))
+ return &udp_hdr(skb)->check;
}
- return skb;
+ return NULL;
}
-/* Updates 'sum', which is a field in 'skb''s data, given that a 4-byte field
- * covered by the sum has been changed from 'from' to 'to'. If set,
- * 'pseudohdr' indicates that the field is in the TCP or UDP pseudo-header.
- * Based on nf_proto_csum_replace4. */
-static void update_csum(__sum16 *sum, struct sk_buff *skb,
- __be32 from, __be32 to, int pseudohdr)
+static struct sk_buff *set_nw_addr(struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ const struct nlattr *a)
{
- __be32 diff[] = { ~from, to };
- if (skb->ip_summed != CHECKSUM_PARTIAL) {
- *sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
- ~csum_unfold(*sum)));
- if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr)
- skb->csum = ~csum_partial((char *)diff, sizeof(diff),
- ~skb->csum);
- } else if (pseudohdr)
- *sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
- csum_unfold(*sum)));
+ __be32 new_nwaddr = nla_get_be32(a);
+ struct iphdr *nh;
+ __sum16 *check;
+ __be32 *nwaddr;
+
+ if (unlikely(!is_ip(skb, key)))
+ return skb;
+
+ skb = make_writable(skb, 0);
+ if (unlikely(!skb))
+ return NULL;
+
+ nh = ip_hdr(skb);
+ nwaddr = nla_type(a) == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->saddr : &nh->daddr;
+
+ check = get_l4_checksum(skb, key);
+ if (likely(check))
+ inet_proto_csum_replace4(check, skb, *nwaddr, new_nwaddr, 1);
+ csum_replace4(&nh->check, *nwaddr, new_nwaddr);
+
+ skb_clear_rxhash(skb);
+
+ *nwaddr = new_nwaddr;
+
+ return skb;
}
-static struct sk_buff *set_nw_addr(struct sk_buff *skb,
- struct odp_flow_key *key,
- const struct odp_action_nw_addr *a,
- gfp_t gfp)
+static struct sk_buff *set_nw_tos(struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ u8 nw_tos)
{
- if (key->dl_type != htons(ETH_P_IP))
+ if (unlikely(!is_ip(skb, key)))
return skb;
- skb = make_writable(skb, gfp);
+ skb = make_writable(skb, 0);
if (skb) {
struct iphdr *nh = ip_hdr(skb);
- u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
- u32 old = *f;
- u32 new = a->nw_addr;
-
- if (key->nw_proto == IPPROTO_TCP) {
- struct tcphdr *th = tcp_hdr(skb);
- update_csum(&th->check, skb, old, new, 1);
- } else if (key->nw_proto == IPPROTO_UDP) {
- struct udphdr *th = udp_hdr(skb);
- update_csum(&th->check, skb, old, new, 1);
- }
- update_csum(&nh->check, skb, old, new, 0);
+ u8 *f = &nh->tos;
+ u8 old = *f;
+ u8 new;
+
+ /* Set the DSCP bits and preserve the ECN bits. */
+ new = nw_tos | (nh->tos & INET_ECN_MASK);
+ csum_replace4(&nh->check, (__force __be32)old,
+ (__force __be32)new);
*f = new;
}
return skb;
}
-static struct sk_buff *
-set_tp_port(struct sk_buff *skb, struct odp_flow_key *key,
- const struct odp_action_tp_port *a,
- gfp_t gfp)
+static struct sk_buff *set_tp_port(struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ const struct nlattr *a)
{
- int check_ofs;
+ struct udphdr *th;
+ __sum16 *check;
+ __be16 *port;
- if (key->dl_type != htons(ETH_P_IP))
+ if (unlikely(!is_ip(skb, key)))
return skb;
- if (key->nw_proto == IPPROTO_TCP)
- check_ofs = offsetof(struct tcphdr, check);
- else if (key->nw_proto == IPPROTO_UDP)
- check_ofs = offsetof(struct udphdr, check);
- else
+ skb = make_writable(skb, 0);
+ if (unlikely(!skb))
+ return NULL;
+
+ /* Must follow make_writable() since that can move the skb data. */
+ check = get_l4_checksum(skb, key);
+ if (unlikely(!check))
return skb;
- skb = make_writable(skb, gfp);
- if (skb) {
- struct udphdr *th = udp_hdr(skb);
- u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
- u16 old = *f;
- u16 new = a->tp_port;
- update_csum((u16*)((u8*)skb->data + check_ofs),
- skb, old, new, 1);
- *f = new;
- }
- return skb;
-}
+ /*
+ * Update port and checksum.
+ *
+ * This is OK because source and destination port numbers are at the
+ * same offsets in both UDP and TCP headers, and get_l4_checksum() only
+ * supports those protocols.
+ */
+ th = udp_hdr(skb);
+ port = nla_type(a) == ODP_ACTION_ATTR_SET_TP_SRC ? &th->source : &th->dest;
+ inet_proto_csum_replace2(check, skb, *port, nla_get_be16(a), 0);
+ *port = nla_get_be16(a);
+ skb_clear_rxhash(skb);
-static inline unsigned packet_length(const struct sk_buff *skb)
-{
- unsigned length = skb->len - ETH_HLEN;
- if (skb->protocol == htons(ETH_P_8021Q))
- length -= VLAN_HLEN;
- return length;
+ return skb;
}
-int dp_xmit_skb(struct sk_buff *skb)
+/**
+ * is_spoofed_arp - check for invalid ARP packet
+ *
+ * @skb: skbuff containing an Ethernet packet, with network header pointing
+ * just past the Ethernet and optional 802.1Q header.
+ * @key: flow key extracted from @skb by flow_extract()
+ *
+ * Returns true if @skb is an invalid Ethernet+IPv4 ARP packet: one with screwy
+ * or truncated header fields or one whose inner and outer Ethernet address
+ * differ.
+ */
+static bool is_spoofed_arp(struct sk_buff *skb, const struct sw_flow_key *key)
{
- struct datapath *dp = skb->dev->br_port->dp;
- int len = skb->len;
+ struct arp_eth_header *arp;
- if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb)) {
- printk(KERN_WARNING "%s: dropped over-mtu packet: %d > %d\n",
- dp_name(dp), packet_length(skb), skb->dev->mtu);
- kfree_skb(skb);
- return -E2BIG;
- }
+ if (key->dl_type != htons(ETH_P_ARP))
+ return false;
- dev_queue_xmit(skb);
+ if (skb_network_offset(skb) + sizeof(struct arp_eth_header) > skb->len)
+ return true;
- return len;
+ arp = (struct arp_eth_header *)skb_network_header(skb);
+ return (arp->ar_hrd != htons(ARPHRD_ETHER) ||
+ arp->ar_pro != htons(ETH_P_IP) ||
+ arp->ar_hln != ETH_ALEN ||
+ arp->ar_pln != 4 ||
+ compare_ether_addr(arp->ar_sha, eth_hdr(skb)->h_source));
}
-static void
-do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
+static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
{
- struct net_bridge_port *p;
- struct net_device *dev;
+ struct vport *p;
if (!skb)
goto error;
- p = dp->ports[out_port];
+ p = rcu_dereference(dp->ports[out_port]);
if (!p)
goto error;
- dev = skb->dev = p->dev;
- if (is_dp_dev(dev))
- dp_dev_recv(dev, skb);
- else
- dp_xmit_skb(skb);
+ vport_send(p, skb);
return;
error:
kfree_skb(skb);
}
-/* Never consumes 'skb'. Returns a port that 'skb' should be sent to, -1 if
- * none. */
-static int output_group(struct datapath *dp, __u16 group,
- struct sk_buff *skb, gfp_t gfp)
+static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg,
+ const struct sw_flow_key *key)
{
- struct dp_port_group *g = rcu_dereference(dp->groups[group]);
- int prev_port = -1;
- int i;
-
- if (!g)
- return -1;
- for (i = 0; i < g->n_ports; i++) {
- struct net_bridge_port *p = dp->ports[g->ports[i]];
- if (!p || skb->dev == p->dev)
- continue;
- if (prev_port != -1) {
- struct sk_buff *clone = skb_clone(skb, gfp);
- if (!clone)
- return -1;
- do_output(dp, clone, prev_port);
- }
- prev_port = p->port_no;
- }
- return prev_port;
-}
+ struct dp_upcall_info upcall;
-static int
-output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp)
-{
- skb = skb_clone(skb, gfp);
+ skb = skb_clone(skb, GFP_ATOMIC);
if (!skb)
return -ENOMEM;
- return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
+
+ upcall.cmd = ODP_PACKET_CMD_ACTION;
+ upcall.key = key;
+ upcall.userdata = arg;
+ upcall.sample_pool = 0;
+ upcall.actions = NULL;
+ upcall.actions_len = 0;
+ return dp_upcall(dp, skb, &upcall);
}
/* Execute a list of actions against 'skb'. */
-int execute_actions(struct datapath *dp, struct sk_buff *skb,
- struct odp_flow_key *key,
- const union odp_action *a, int n_actions,
- gfp_t gfp)
+static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ const struct nlattr *actions, u32 actions_len)
{
/* Every output action needs a separate clone of 'skb', but the common
* case is just a single output action, so that doing a clone and
* then freeing the original skbuff is wasteful. So the following code
* is slightly obscure just to avoid that. */
int prev_port = -1;
- int err;
- for (; n_actions > 0; a++, n_actions--) {
- WARN_ON_ONCE(skb_shared(skb));
+ u32 priority = skb->priority;
+ const struct nlattr *a;
+ int rem, err;
+
+ for (a = actions, rem = actions_len; rem > 0; a = nla_next(a, &rem)) {
if (prev_port != -1) {
- do_output(dp, skb_clone(skb, gfp), prev_port);
+ do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
prev_port = -1;
}
- switch (a->type) {
- case ODPAT_OUTPUT:
- prev_port = a->output.port;
+ switch (nla_type(a)) {
+ case ODP_ACTION_ATTR_OUTPUT:
+ prev_port = nla_get_u32(a);
break;
- case ODPAT_OUTPUT_GROUP:
- prev_port = output_group(dp, a->output_group.group,
- skb, gfp);
- break;
-
- case ODPAT_CONTROLLER:
- err = output_control(dp, skb, a->controller.arg, gfp);
+ case ODP_ACTION_ATTR_CONTROLLER:
+ err = output_control(dp, skb, nla_get_u64(a), key);
if (err) {
kfree_skb(skb);
return err;
}
break;
- case ODPAT_SET_VLAN_VID:
- case ODPAT_SET_VLAN_PCP:
- skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
- if (IS_ERR(skb))
- return PTR_ERR(skb);
+ case ODP_ACTION_ATTR_SET_TUNNEL:
+ OVS_CB(skb)->tun_id = nla_get_be64(a);
break;
- case ODPAT_STRIP_VLAN:
- skb = strip_vlan(skb, key, gfp);
+ case ODP_ACTION_ATTR_SET_DL_TCI:
+ skb = modify_vlan_tci(skb, nla_get_be16(a));
break;
- case ODPAT_SET_DL_SRC:
- case ODPAT_SET_DL_DST:
- skb = set_dl_addr(skb, &a->dl_addr, gfp);
+ case ODP_ACTION_ATTR_STRIP_VLAN:
+ skb = strip_vlan(skb);
break;
- case ODPAT_SET_NW_SRC:
- case ODPAT_SET_NW_DST:
- skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
+ case ODP_ACTION_ATTR_SET_DL_SRC:
+ skb = make_writable(skb, 0);
+ if (!skb)
+ return -ENOMEM;
+ memcpy(eth_hdr(skb)->h_source, nla_data(a), ETH_ALEN);
break;
- case ODPAT_SET_TP_SRC:
- case ODPAT_SET_TP_DST:
- skb = set_tp_port(skb, key, &a->tp_port, gfp);
+ case ODP_ACTION_ATTR_SET_DL_DST:
+ skb = make_writable(skb, 0);
+ if (!skb)
+ return -ENOMEM;
+ memcpy(eth_hdr(skb)->h_dest, nla_data(a), ETH_ALEN);
+ break;
+
+ case ODP_ACTION_ATTR_SET_NW_SRC:
+ case ODP_ACTION_ATTR_SET_NW_DST:
+ skb = set_nw_addr(skb, key, a);
+ break;
+
+ case ODP_ACTION_ATTR_SET_NW_TOS:
+ skb = set_nw_tos(skb, key, nla_get_u8(a));
+ break;
+
+ case ODP_ACTION_ATTR_SET_TP_SRC:
+ case ODP_ACTION_ATTR_SET_TP_DST:
+ skb = set_tp_port(skb, key, a);
+ break;
+
+ case ODP_ACTION_ATTR_SET_PRIORITY:
+ skb->priority = nla_get_u32(a);
+ break;
+
+ case ODP_ACTION_ATTR_POP_PRIORITY:
+ skb->priority = priority;
+ break;
+
+ case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
+ if (unlikely(is_spoofed_arp(skb, key)))
+ goto exit;
break;
}
if (!skb)
return -ENOMEM;
}
+exit:
if (prev_port != -1)
do_output(dp, skb, prev_port);
else
kfree_skb(skb);
return 0;
}
+
+static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ const struct nlattr *a, u32 actions_len)
+{
+ struct sk_buff *nskb;
+ struct vport *p = OVS_CB(skb)->vport;
+ struct dp_upcall_info upcall;
+
+ if (unlikely(!p))
+ return;
+
+ atomic_inc(&p->sflow_pool);
+ if (net_random() >= dp->sflow_probability)
+ return;
+
+ nskb = skb_clone(skb, GFP_ATOMIC);
+ if (unlikely(!nskb))
+ return;
+
+ upcall.cmd = ODP_PACKET_CMD_SAMPLE;
+ upcall.key = key;
+ upcall.userdata = 0;
+ upcall.sample_pool = atomic_read(&p->sflow_pool);
+ upcall.actions = a;
+ upcall.actions_len = actions_len;
+ dp_upcall(dp, nskb, &upcall);
+}
+
+/* Execute a list of actions against 'skb'. */
+int execute_actions(struct datapath *dp, struct sk_buff *skb,
+ const struct sw_flow_key *key,
+ const struct nlattr *actions, u32 actions_len)
+{
+ if (dp->sflow_probability)
+ sflow_sample(dp, skb, key, actions, actions_len);
+
+ OVS_CB(skb)->tun_id = 0;
+
+ return do_execute_actions(dp, skb, key, actions, actions_len);
+}