#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 "openvswitch/datapath-protocol.h"
#include "vport.h"
-static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp)
+static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom)
{
- if (skb_shared(skb) || skb_cloned(skb)) {
+ 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);
+ nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), GFP_ATOMIC);
if (nskb) {
set_skb_csum_bits(skb, nskb);
kfree_skb(skb);
return NULL;
}
-static void set_tunnel(struct sk_buff *skb, struct odp_flow_key *key,
- __be32 tun_id)
-{
- OVS_CB(skb)->tun_id = key->tun_id = tun_id;
-}
-
static struct sk_buff *vlan_pull_tag(struct sk_buff *skb)
{
struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
struct ethhdr *eh;
/* Verify we were given a vlan packet */
- if (vh->h_vlan_proto != htons(ETH_P_8021Q))
+ if (vh->h_vlan_proto != htons(ETH_P_8021Q) || skb->len < VLAN_ETH_HLEN)
return skb;
if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
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)
+ const struct odp_flow_key *key,
+ const union odp_action *a, int n_actions)
{
- u16 tci, mask;
+ __be16 tci = a->dl_tci.tci;
- if (a->type == ODPAT_SET_VLAN_VID) {
- tci = ntohs(a->vlan_vid.vlan_vid);
- mask = VLAN_VID_MASK;
- key->dl_vlan = a->vlan_vid.vlan_vid;
- } else {
- tci = a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT;
- mask = VLAN_PCP_MASK;
- key->dl_vlan_pcp = a->vlan_pcp.vlan_pcp;
- }
-
- skb = make_writable(skb, VLAN_HLEN, gfp);
+ skb = make_writable(skb, VLAN_HLEN);
if (!skb)
return ERR_PTR(-ENOMEM);
if (skb->protocol == htons(ETH_P_8021Q)) {
/* Modify vlan id, but maintain other TCI values */
- struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
- __be16 old_tci = vh->h_vlan_TCI;
+ struct vlan_ethhdr *vh;
+ __be16 old_tci;
- vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci);
+ if (skb->len < VLAN_ETH_HLEN)
+ return skb;
+
+ vh = vlan_eth_hdr(skb);
+ old_tci = vh->h_vlan_TCI;
+
+ vh->h_vlan_TCI = tci;
if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) {
__be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
if (unlikely(err)) {
kfree_skb(skb);
return ERR_PTR(err);
- }
+ }
/* GSO is not implemented for packets with an 802.1Q header, so
* we have to do segmentation before we add that header.
/* GSO can change the checksum type so update.*/
compute_ip_summed(segs, true);
- segs = __vlan_put_tag(segs, tci);
+ segs = __vlan_put_tag(segs, ntohs(tci));
err = -ENOMEM;
if (segs) {
- struct odp_flow_key segkey = *key;
err = execute_actions(dp, segs,
- &segkey, a + 1,
- n_actions - 1,
- gfp);
+ key, a + 1,
+ n_actions - 1);
}
if (unlikely(err)) {
* e.g. vconfig(8)), so call the software-only version
* __vlan_put_tag() directly instead.
*/
- skb = __vlan_put_tag(skb, tci);
+ skb = __vlan_put_tag(skb, ntohs(tci));
if (!skb)
return ERR_PTR(-ENOMEM);
return skb;
}
-static struct sk_buff *strip_vlan(struct sk_buff *skb,
- struct odp_flow_key *key, gfp_t gfp)
+static struct sk_buff *strip_vlan(struct sk_buff *skb)
{
- skb = make_writable(skb, 0, gfp);
- if (skb) {
+ skb = make_writable(skb, 0);
+ if (skb)
vlan_pull_tag(skb);
- key->dl_vlan = htons(ODP_VLAN_NONE);
- }
return skb;
}
static struct sk_buff *set_dl_addr(struct sk_buff *skb,
- struct odp_flow_key *key,
- const struct odp_action_dl_addr *a,
- gfp_t gfp)
+ const struct odp_action_dl_addr *a)
{
- skb = make_writable(skb, 0, gfp);
+ skb = make_writable(skb, 0);
if (skb) {
struct ethhdr *eh = eth_hdr(skb);
- if (a->type == ODPAT_SET_DL_SRC) {
+ if (a->type == ODPAT_SET_DL_SRC)
memcpy(eh->h_source, a->dl_addr, ETH_ALEN);
- memcpy(key->dl_src, a->dl_addr, ETH_ALEN);
- } else {
+ else
memcpy(eh->h_dest, a->dl_addr, ETH_ALEN);
- memcpy(key->dl_dst, a->dl_addr, ETH_ALEN);
- }
}
return skb;
}
csum_unfold(*sum)));
}
+static bool is_ip(struct sk_buff *skb, const struct odp_flow_key *key)
+{
+ return (key->dl_type == htons(ETH_P_IP) &&
+ skb->transport_header > skb->network_header);
+}
+
+static __sum16 *get_l4_checksum(struct sk_buff *skb, const struct odp_flow_key *key)
+{
+ 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 NULL;
+}
+
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)
+ const struct odp_flow_key *key,
+ const struct odp_action_nw_addr *a)
{
- if (key->dl_type != htons(ETH_P_IP))
+ struct iphdr *nh;
+ __sum16 *check;
+ __be32 *nwaddr;
+
+ if (unlikely(!is_ip(skb, key)))
return skb;
- skb = make_writable(skb, 0, gfp);
- 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);
- *f = new;
+ skb = make_writable(skb, 0);
+ if (unlikely(!skb))
+ return NULL;
+
+ nh = ip_hdr(skb);
+ nwaddr = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
+
+ check = get_l4_checksum(skb, key);
+ if (likely(check))
+ update_csum(check, skb, *nwaddr, a->nw_addr, 1);
+ update_csum(&nh->check, skb, *nwaddr, a->nw_addr, 0);
+
+ *nwaddr = a->nw_addr;
- if (a->type == ODPAT_SET_NW_SRC)
- key->nw_src = a->nw_addr;
- else
- key->nw_dst = a->nw_addr;
- }
return skb;
}
static struct sk_buff *set_nw_tos(struct sk_buff *skb,
- struct odp_flow_key *key,
- const struct odp_action_nw_tos *a,
- gfp_t gfp)
+ const struct odp_flow_key *key,
+ const struct odp_action_nw_tos *a)
{
- if (key->dl_type != htons(ETH_P_IP))
+ if (unlikely(!is_ip(skb, key)))
return skb;
- skb = make_writable(skb, 0, gfp);
+ skb = make_writable(skb, 0);
if (skb) {
struct iphdr *nh = ip_hdr(skb);
u8 *f = &nh->tos;
/* Set the DSCP bits and preserve the ECN bits. */
new = a->nw_tos | (nh->tos & INET_ECN_MASK);
- update_csum(&nh->check, skb, htons((uint16_t)old),
- htons((uint16_t)new), 0);
+ update_csum(&nh->check, skb, htons((u16)old),
+ htons((u16)new), 0);
*f = new;
- key->nw_tos = a->nw_tos;
}
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 odp_flow_key *key,
+ const struct odp_action_tp_port *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, 0, 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*)(skb_transport_header(skb) + check_ofs),
- skb, old, new, 0);
- *f = new;
- if (a->type == ODPAT_SET_TP_SRC)
- key->tp_src = a->tp_port;
- else
- key->tp_dst = a->tp_port;
- }
+ /*
+ * 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 = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
+ update_csum(check, skb, *port, a->tp_port, 0);
+ *port = a->tp_port;
+
return 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 odp_flow_key *key)
+{
+ struct arp_eth_header *arp;
+
+ if (key->dl_type != htons(ETH_P_ARP))
+ return false;
+
+ if (skb_network_offset(skb) + sizeof(struct arp_eth_header) > skb->len)
+ return true;
+
+ 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)
{
struct dp_port *p;
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, u32 arg)
{
- 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 dp_port *p = rcu_dereference(dp->ports[g->ports[i]]);
- if (!p || OVS_CB(skb)->dp_port == p)
- 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;
-}
-
-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);
* information about what happened to it. */
static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
const union odp_action *a, int n_actions,
- gfp_t gfp, struct dp_port *dp_port)
+ struct dp_port *dp_port)
{
struct odp_sflow_sample_header *hdr;
unsigned int actlen = n_actions * sizeof(union odp_action);
unsigned int hdrlen = sizeof(struct odp_sflow_sample_header);
struct sk_buff *nskb;
- nskb = skb_copy_expand(skb, actlen + hdrlen, 0, gfp);
+ nskb = skb_copy_expand(skb, actlen + hdrlen, 0, GFP_ATOMIC);
if (!nskb)
return;
/* 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)
+ const struct odp_flow_key *key,
+ const union odp_action *a, int n_actions)
{
/* 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
atomic_inc(&p->sflow_pool);
if (dp->sflow_probability == UINT_MAX ||
net_random() < dp->sflow_probability)
- sflow_sample(dp, skb, a, n_actions, gfp, p);
+ sflow_sample(dp, skb, a, n_actions, p);
}
}
OVS_CB(skb)->tun_id = 0;
for (; n_actions > 0; a++, n_actions--) {
- WARN_ON_ONCE(skb_shared(skb));
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;
}
prev_port = a->output.port;
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);
+ err = output_control(dp, skb, a->controller.arg);
if (err) {
kfree_skb(skb);
return err;
break;
case ODPAT_SET_TUNNEL:
- set_tunnel(skb, key, a->tunnel.tun_id);
+ OVS_CB(skb)->tun_id = a->tunnel.tun_id;
break;
- case ODPAT_SET_VLAN_VID:
- case ODPAT_SET_VLAN_PCP:
- skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
+ case ODPAT_SET_DL_TCI:
+ skb = modify_vlan_tci(dp, skb, key, a, n_actions);
if (IS_ERR(skb))
return PTR_ERR(skb);
break;
case ODPAT_STRIP_VLAN:
- skb = strip_vlan(skb, key, gfp);
+ skb = strip_vlan(skb);
break;
case ODPAT_SET_DL_SRC:
case ODPAT_SET_DL_DST:
- skb = set_dl_addr(skb, key, &a->dl_addr, gfp);
+ skb = set_dl_addr(skb, &a->dl_addr);
break;
case ODPAT_SET_NW_SRC:
case ODPAT_SET_NW_DST:
- skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
+ skb = set_nw_addr(skb, key, &a->nw_addr);
break;
case ODPAT_SET_NW_TOS:
- skb = set_nw_tos(skb, key, &a->nw_tos, gfp);
+ skb = set_nw_tos(skb, key, &a->nw_tos);
break;
case ODPAT_SET_TP_SRC:
case ODPAT_SET_TP_DST:
- skb = set_tp_port(skb, key, &a->tp_port, gfp);
+ skb = set_tp_port(skb, key, &a->tp_port);
break;
case ODPAT_SET_PRIORITY:
case ODPAT_POP_PRIORITY:
skb->priority = priority;
break;
+
+ case ODPAT_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