/*
* Distributed under the terms of the GNU GPL version 2.
- * Copyright (c) 2007, 2008, 2009, 2010 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 <net/inet_ecn.h>
#include <net/ip.h>
-#include "compat.h"
-
-struct kmem_cache *flow_cache;
-static unsigned int hash_seed;
-
-struct arp_eth_header
-{
- __be16 ar_hrd; /* format of hardware address */
- __be16 ar_pro; /* format of protocol address */
- unsigned char ar_hln; /* length of hardware address */
- unsigned char ar_pln; /* length of protocol address */
- __be16 ar_op; /* ARP opcode (command) */
-
- /* Ethernet+IPv4 specific members. */
- unsigned char ar_sha[ETH_ALEN]; /* sender hardware address */
- unsigned char ar_sip[4]; /* sender IP address */
- unsigned char ar_tha[ETH_ALEN]; /* target hardware address */
- unsigned char ar_tip[4]; /* target IP address */
-} __attribute__((packed));
+static struct kmem_cache *flow_cache;
+static unsigned int hash_seed __read_mostly;
static inline bool arphdr_ok(struct sk_buff *skb)
{
- int nh_ofs = skb_network_offset(skb);
- return pskb_may_pull(skb, nh_ofs + sizeof(struct arp_eth_header));
+ return skb->len >= skb_network_offset(skb) + sizeof(struct arp_eth_header);
}
-static inline bool iphdr_ok(struct sk_buff *skb)
+static inline int check_iphdr(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 false;
+ unsigned int nh_ofs = skb_network_offset(skb);
+ unsigned int ip_len;
+
+ if (skb->len < nh_ofs + sizeof(struct iphdr))
+ return -EINVAL;
+
+ ip_len = ip_hdrlen(skb);
+ if (ip_len < sizeof(struct iphdr) || skb->len < nh_ofs + ip_len)
+ return -EINVAL;
+
+ /*
+ * Pull enough header bytes to account for the IP header plus the
+ * longest transport header that we parse, currently 20 bytes for TCP.
+ */
+ if (!pskb_may_pull(skb, min(nh_ofs + ip_len + 20, skb->len)))
+ return -ENOMEM;
+
+ skb_set_transport_header(skb, nh_ofs + ip_len);
+ return 0;
}
static inline bool tcphdr_ok(struct sk_buff *skb)
{
int th_ofs = skb_transport_offset(skb);
- if (pskb_may_pull(skb, th_ofs + sizeof(struct tcphdr))) {
+ if (skb->len >= th_ofs + sizeof(struct tcphdr)) {
int tcp_len = tcp_hdrlen(skb);
return (tcp_len >= sizeof(struct tcphdr)
&& skb->len >= th_ofs + tcp_len);
static inline bool udphdr_ok(struct sk_buff *skb)
{
- int th_ofs = skb_transport_offset(skb);
- return pskb_may_pull(skb, th_ofs + sizeof(struct udphdr));
+ return skb->len >= skb_transport_offset(skb) + sizeof(struct udphdr);
}
static inline bool icmphdr_ok(struct sk_buff *skb)
{
- int th_ofs = skb_transport_offset(skb);
- return pskb_may_pull(skb, th_ofs + sizeof(struct icmphdr));
+ return skb->len >= skb_transport_offset(skb) + sizeof(struct icmphdr);
}
#define TCP_FLAGS_OFFSET 13
spin_unlock_bh(&flow->lock);
}
-struct sw_flow_actions *flow_actions_alloc(size_t n_actions)
+struct sw_flow_actions *flow_actions_alloc(u32 actions_len)
{
struct sw_flow_actions *sfa;
- if (n_actions > (PAGE_SIZE - sizeof *sfa) / sizeof(union odp_action))
+ if (actions_len % NLA_ALIGNTO)
+ return ERR_PTR(-EINVAL);
+
+ /* At least DP_MAX_PORTS actions are required to be able to flood a
+ * packet to every port. Factor of 2 allows for setting VLAN tags,
+ * etc. */
+ if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
return ERR_PTR(-EINVAL);
- sfa = kmalloc(sizeof *sfa + n_actions * sizeof(union odp_action),
- GFP_KERNEL);
+ sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
if (!sfa)
return ERR_PTR(-ENOMEM);
- sfa->n_actions = n_actions;
+ sfa->actions_len = actions_len;
return sfa;
}
-
-/* Frees 'flow' immediately. */
-static void flow_free(struct sw_flow *flow)
+struct sw_flow *flow_alloc(void)
{
- if (unlikely(!flow))
- return;
- kfree(flow->sf_acts);
- kmem_cache_free(flow_cache, flow);
+ struct sw_flow *flow;
+
+ flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
+ if (!flow)
+ return ERR_PTR(-ENOMEM);
+
+ spin_lock_init(&flow->lock);
+ atomic_set(&flow->refcnt, 1);
+ flow->dead = false;
+
+ return flow;
}
void flow_free_tbl(struct tbl_node *node)
{
struct sw_flow *flow = flow_cast(node);
- flow_free(flow);
+
+ flow->dead = true;
+ flow_put(flow);
}
/* RCU callback used by flow_deferred_free. */
static void rcu_free_flow_callback(struct rcu_head *rcu)
{
struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
- flow_free(flow);
+
+ flow->dead = true;
+ flow_put(flow);
}
/* Schedules 'flow' to be freed after the next RCU grace period.
call_rcu(&flow->rcu, rcu_free_flow_callback);
}
+void flow_hold(struct sw_flow *flow)
+{
+ atomic_inc(&flow->refcnt);
+}
+
+void flow_put(struct sw_flow *flow)
+{
+ if (unlikely(!flow))
+ return;
+
+ if (atomic_dec_and_test(&flow->refcnt)) {
+ kfree((struct sf_flow_acts __force *)flow->sf_acts);
+ kmem_cache_free(flow_cache, flow);
+ }
+}
+
/* RCU callback used by flow_deferred_free_acts. */
static void rcu_free_acts_callback(struct rcu_head *rcu)
{
- struct sw_flow_actions *sf_acts = container_of(rcu,
+ struct sw_flow_actions *sf_acts = container_of(rcu,
struct sw_flow_actions, rcu);
kfree(sf_acts);
}
return;
qp = (struct qtag_prefix *) skb->data;
- key->dl_vlan = qp->tci & htons(VLAN_VID_MASK);
- key->dl_vlan_pcp = (ntohs(qp->tci) & VLAN_PCP_MASK) >> VLAN_PCP_SHIFT;
+ key->dl_tci = qp->tci | htons(ODP_TCI_PRESENT);
__skb_pull(skb, sizeof(struct qtag_prefix));
}
u8 ssap; /* Always 0xAA */
u8 ctrl;
u8 oui[3];
- u16 ethertype;
+ __be16 ethertype;
};
struct llc_snap_hdr *llc;
__be16 proto;
* Ethernet header
* @in_port: port number on which @skb was received.
* @key: output flow key
+ * @is_frag: set to 1 if @skb contains an IPv4 fragment, or to 0 if @skb does
+ * not contain an IPv4 packet or if it is not a fragment.
*
* The caller must ensure that skb->len >= ETH_HLEN.
*
- * Sets OVS_CB(skb)->is_frag to %true if @skb is an IPv4 fragment, otherwise to
- * %false.
+ * Returns 0 if successful, otherwise a negative errno value.
+ *
+ * Initializes @skb header pointers as follows:
+ *
+ * - skb->mac_header: the Ethernet header.
+ *
+ * - skb->network_header: just past the Ethernet header, or just past the
+ * VLAN header, to the first byte of the Ethernet payload.
+ *
+ * - skb->transport_header: If key->dl_type is ETH_P_IP on output, then just
+ * past the IPv4 header, if one is present and of a correct length,
+ * otherwise the same as skb->network_header. For other key->dl_type
+ * values it is left untouched.
*/
-int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key)
+int flow_extract(struct sk_buff *skb, u16 in_port, struct odp_flow_key *key,
+ bool *is_frag)
{
struct ethhdr *eth;
- memset(key, 0, sizeof *key);
+ memset(key, 0, sizeof(*key));
key->tun_id = OVS_CB(skb)->tun_id;
key->in_port = in_port;
- key->dl_vlan = htons(ODP_VLAN_NONE);
- OVS_CB(skb)->is_frag = false;
-
+ *is_frag = false;
+
+ /*
+ * We would really like to pull as many bytes as we could possibly
+ * want to parse into the linear data area. Currently that is:
+ *
+ * 14 Ethernet header
+ * 4 VLAN header
+ * 60 max IP header with options
+ * 20 max TCP/UDP/ICMP header (don't care about options)
+ * --
+ * 98
+ *
+ * But Xen only allocates 64 or 72 bytes for the linear data area in
+ * netback, which means that we would reallocate and copy the skb's
+ * linear data on every packet if we did that. So instead just pull 64
+ * bytes, which is always sufficient without IP options, and then check
+ * whether we need to pull more later when we look at the IP header.
+ */
if (!pskb_may_pull(skb, min(skb->len, 64u)))
- return 0;
+ return -ENOMEM;
skb_reset_mac_header(skb);
__skb_push(skb, skb->data - (unsigned char *)eth);
/* Network layer. */
- if (key->dl_type == htons(ETH_P_IP) && iphdr_ok(skb)) {
- struct iphdr *nh = ip_hdr(skb);
- int th_ofs = skb_network_offset(skb) + nh->ihl * 4;
+ if (key->dl_type == htons(ETH_P_IP)) {
+ struct iphdr *nh;
+ int error;
+
+ error = check_iphdr(skb);
+ if (unlikely(error)) {
+ if (error == -EINVAL) {
+ skb->transport_header = skb->network_header;
+ return 0;
+ }
+ return error;
+ }
+
+ nh = ip_hdr(skb);
key->nw_src = nh->saddr;
key->nw_dst = nh->daddr;
key->nw_tos = nh->tos & ~INET_ECN_MASK;
key->nw_proto = nh->protocol;
- skb_set_transport_header(skb, th_ofs);
/* Transport layer. */
- if (!(nh->frag_off & htons(IP_MF | IP_OFFSET))) {
+ if (!(nh->frag_off & htons(IP_MF | IP_OFFSET)) &&
+ !(skb_shinfo(skb)->gso_type & SKB_GSO_UDP)) {
if (key->nw_proto == IPPROTO_TCP) {
if (tcphdr_ok(skb)) {
struct tcphdr *tcp = tcp_hdr(skb);
key->tp_src = tcp->source;
key->tp_dst = tcp->dest;
- } else {
- /* Avoid tricking other code into
- * thinking that this packet has an L4
- * header. */
- key->nw_proto = 0;
}
} else if (key->nw_proto == IPPROTO_UDP) {
if (udphdr_ok(skb)) {
struct udphdr *udp = udp_hdr(skb);
key->tp_src = udp->source;
key->tp_dst = udp->dest;
- } else {
- /* Avoid tricking other code into
- * thinking that this packet has an L4
- * header. */
- key->nw_proto = 0;
}
} else if (key->nw_proto == IPPROTO_ICMP) {
if (icmphdr_ok(skb)) {
* in 16-bit network byte order. */
key->tp_src = htons(icmp->type);
key->tp_dst = htons(icmp->code);
- } else {
- /* Avoid tricking other code into
- * thinking that this packet has an L4
- * header. */
- key->nw_proto = 0;
}
}
- } else {
- OVS_CB(skb)->is_frag = true;
- }
+ } else
+ *is_frag = true;
+
} else if (key->dl_type == htons(ETH_P_ARP) && arphdr_ok(skb)) {
struct arp_eth_header *arp;
&& arp->ar_pln == 4) {
/* We only match on the lower 8 bits of the opcode. */
- if (ntohs(arp->ar_op) <= 0xff) {
+ if (ntohs(arp->ar_op) <= 0xff)
key->nw_proto = ntohs(arp->ar_op);
- }
- if (key->nw_proto == ARPOP_REQUEST
+ if (key->nw_proto == ARPOP_REQUEST
|| key->nw_proto == ARPOP_REPLY) {
memcpy(&key->nw_src, arp->ar_sip, sizeof(key->nw_src));
memcpy(&key->nw_dst, arp->ar_tip, sizeof(key->nw_dst));
}
}
- } else {
- skb_reset_transport_header(skb);
}
return 0;
}
u32 flow_hash(const struct odp_flow_key *key)
{
- return jhash2((u32*)key, sizeof *key / sizeof(u32), hash_seed);
+ return jhash2((u32*)key, sizeof(*key) / sizeof(u32), hash_seed);
}
int flow_cmp(const struct tbl_node *node, void *key2_)
if (flow_cache == NULL)
return -ENOMEM;
- get_random_bytes(&hash_seed, sizeof hash_seed);
+ get_random_bytes(&hash_seed, sizeof(hash_seed));
return 0;
}