X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=datapath%2Factions.c;h=b75aecdace46d3e7c659918f0bcb9bd0acb81b54;hb=722d19c504351a3e1a6f64e5a01ff9806eb089a4;hp=402041180cf79ea310f2ded3e5895bf9f550e4a9;hpb=f2459fe7d91c4c325dfaa3ed18f56200b63ae27e;p=openvswitch diff --git a/datapath/actions.c b/datapath/actions.c index 40204118..b75aecda 100644 --- a/datapath/actions.c +++ b/datapath/actions.c @@ -14,6 +14,7 @@ #include #include #include +#include #include #include #include @@ -24,28 +25,15 @@ #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, gfp_t gfp) { - 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); if (nskb) { -#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24) - /* Before 2.6.24 these fields were not copied when - * doing an skb_copy_expand. */ - nskb->ip_summed = skb->ip_summed; - nskb->csum = skb->csum; -#endif -#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) - /* These fields are copied in skb_clone but not in - * skb_copy or related functions. We need to manually - * copy them over here. */ - nskb->proto_data_valid = skb->proto_data_valid; - nskb->proto_csum_blank = skb->proto_csum_blank; -#endif + set_skb_csum_bits(skb, nskb); kfree_skb(skb); return nskb; } @@ -59,20 +47,13 @@ make_writable(struct sk_buff *skb, unsigned min_headroom, gfp_t gfp) 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) +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) @@ -89,22 +70,19 @@ vlan_pull_tag(struct sk_buff *skb) 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 datapath *dp, struct sk_buff *skb, + const struct odp_flow_key *key, + const union odp_action *a, int n_actions, + gfp_t gfp) { u16 tci, mask; 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); @@ -113,8 +91,14 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb, 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; + + if (skb->len < VLAN_ETH_HLEN) + return skb; + + vh = vlan_eth_hdr(skb); + old_tci = vh->h_vlan_TCI; vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci); @@ -125,6 +109,8 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb, ~skb->csum); } } else { + int err; + /* Add vlan header */ /* Set up checksumming pointers for checksum-deferred packets @@ -132,7 +118,11 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb, * 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); + err = vswitch_skb_checksum_setup(skb); + 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. @@ -169,9 +159,8 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb, segs = __vlan_put_tag(segs, tci); err = -ENOMEM; if (segs) { - struct odp_flow_key segkey = *key; err = execute_actions(dp, segs, - &segkey, a + 1, + key, a + 1, n_actions - 1, gfp); } @@ -211,32 +200,26 @@ modify_vlan_tci(struct datapath *dp, struct sk_buff *skb, 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, gfp_t gfp) { skb = make_writable(skb, 0, gfp); - if (skb) { + 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) { skb = make_writable(skb, 0, gfp); 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; } @@ -261,45 +244,60 @@ static void update_csum(__sum16 *sum, struct sk_buff *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_flow_key *key, const struct odp_action_nw_addr *a, gfp_t gfp) { - 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; + 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_flow_key *key, const struct odp_action_nw_tos *a, gfp_t gfp) { - if (key->dl_type != htons(ETH_P_IP)) + if (unlikely(!is_ip(skb, key))) return skb; skb = make_writable(skb, 0, gfp); @@ -311,61 +309,80 @@ static struct sk_buff *set_nw_tos(struct sk_buff *skb, /* 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, gfp_t gfp) { - 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, gfp); + 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; } -static inline unsigned packet_length(const 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 odp_flow_key *key) { - unsigned length = skb->len - ETH_HLEN; - if (skb->protocol == htons(ETH_P_8021Q)) - length -= VLAN_HLEN; - return length; + 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) +static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port) { struct dp_port *p; - int mtu; if (!skb) goto error; @@ -374,13 +391,6 @@ do_output(struct datapath *dp, struct sk_buff *skb, int out_port) if (!p) goto error; - mtu = vport_get_mtu(p->vport); - if (packet_length(skb) > mtu && !skb_is_gso(skb)) { - printk(KERN_WARNING "%s: dropped over-mtu packet: %d > %d\n", - dp_name(dp), packet_length(skb), mtu); - goto error; - } - vport_send(p->vport, skb); return; @@ -414,8 +424,8 @@ static int output_group(struct datapath *dp, __u16 group, return prev_port; } -static int -output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp) +static int output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, + gfp_t gfp) { skb = skb_clone(skb, gfp); if (!skb) @@ -447,7 +457,7 @@ static void sflow_sample(struct datapath *dp, struct sk_buff *skb, /* Execute a list of actions against 'skb'. */ int execute_actions(struct datapath *dp, struct sk_buff *skb, - struct odp_flow_key *key, + const struct odp_flow_key *key, const union odp_action *a, int n_actions, gfp_t gfp) { @@ -456,6 +466,7 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, * then freeing the original skbuff is wasteful. So the following code * is slightly obscure just to avoid that. */ int prev_port = -1; + u32 priority = skb->priority; int err; if (dp->sflow_probability) { @@ -471,7 +482,6 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, 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); prev_port = -1; @@ -496,7 +506,7 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, 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: @@ -507,12 +517,12 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, break; case ODPAT_STRIP_VLAN: - skb = strip_vlan(skb, key, gfp); + skb = strip_vlan(skb, gfp); 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, gfp); break; case ODPAT_SET_NW_SRC: @@ -528,10 +538,24 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, case ODPAT_SET_TP_DST: skb = set_tp_port(skb, key, &a->tp_port, gfp); break; + + case ODPAT_SET_PRIORITY: + skb->priority = a->priority.priority; + break; + + 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