X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Factions.c;h=ed61039a0a3d6dee88d762ea62934ce95dcf2e67;hb=5bee6e2638989aff14429c5a85fa7cef196f2106;hp=5904c8312d4116464bb3c528f4d28c154e1acdf0;hpb=27bcf966b4057623f7b4d856c0348a1e0eb452e0;p=openvswitch diff --git a/datapath/actions.c b/datapath/actions.c index 5904c831..ed61039a 100644 --- a/datapath/actions.c +++ b/datapath/actions.c @@ -1,6 +1,6 @@ /* * 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. @@ -21,306 +21,168 @@ #include #include "actions.h" +#include "checksum.h" #include "datapath.h" +#include "loop_counter.h" #include "openvswitch/datapath-protocol.h" +#include "vlan.h" #include "vport.h" -static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom) +static int do_execute_actions(struct datapath *, struct sk_buff *, + struct sw_flow_actions *acts); + +static int make_writable(struct sk_buff *skb, int write_len) { - if (skb_cloned(skb)) { - struct sk_buff *nskb; - unsigned headroom = max(min_headroom, skb_headroom(skb)); + if (!skb_cloned(skb) || skb_clone_writable(skb, write_len)) + return 0; - nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), GFP_ATOMIC); - if (nskb) { - set_skb_csum_bits(skb, nskb); - kfree_skb(skb); - return nskb; - } - } else { - unsigned int hdr_len = (skb_transport_offset(skb) - + sizeof(struct tcphdr)); - if (pskb_may_pull(skb, min(hdr_len, skb->len))) - return skb; - } - kfree_skb(skb); - return NULL; + return pskb_expand_head(skb, 0, 0, GFP_ATOMIC); } -static struct sk_buff *vlan_pull_tag(struct sk_buff *skb) +static int strip_vlan(struct sk_buff *skb) { - struct vlan_ethhdr *vh = vlan_eth_hdr(skb); struct ethhdr *eh; + int err; - /* Verify we were given a vlan packet */ - if (vh->h_vlan_proto != htons(ETH_P_8021Q) || skb->len < VLAN_ETH_HLEN) - return skb; + if (vlan_tx_tag_present(skb)) { + vlan_set_tci(skb, 0); + return 0; + } + + if (unlikely(skb->protocol != htons(ETH_P_8021Q) || + skb->len < VLAN_ETH_HLEN)) + return 0; + + err = make_writable(skb, VLAN_ETH_HLEN); + if (unlikely(err)) + return err; - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) + 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 * VLAN_ETH_ALEN); + memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN); eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN); skb->protocol = eh->h_proto; skb->mac_header += VLAN_HLEN; - return skb; + return 0; } -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) +static int modify_vlan_tci(struct sk_buff *skb, __be16 tci) { - __be16 tci = a->dl_tci.tci; - - 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; - __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 = tci; - - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) { - __be16 diff[] = { ~old_tci, vh->h_vlan_TCI }; - - skb->csum = ~csum_partial((char *)diff, sizeof(diff), - ~skb->csum); - } - } else { + if (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_8021Q)) { int err; - /* 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. */ - 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. - * - * 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; - - /* GSO can change the checksum type so update.*/ - compute_ip_summed(segs, true); - - segs = __vlan_put_tag(segs, ntohs(tci)); - err = -ENOMEM; - if (segs) { - err = execute_actions(dp, segs, - key, a + 1, - n_actions - 1); - } - - 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; - compute_ip_summed(skb, true); - } + if (unlikely(skb->len < VLAN_ETH_HLEN)) + return 0; - /* 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, ntohs(tci)); - if (!skb) - return ERR_PTR(-ENOMEM); - - /* GSO doesn't fix up the hardware computed checksum so this - * will only be hit in the non-GSO case. */ - if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) - skb->csum = csum_add(skb->csum, csum_partial(skb->data - + ETH_HLEN, VLAN_HLEN, 0)); + err = strip_vlan(skb); + if (unlikely(err)) + return err; } - return skb; -} + __vlan_hwaccel_put_tag(skb, ntohs(tci)); -static struct sk_buff *strip_vlan(struct sk_buff *skb) -{ - skb = make_writable(skb, 0); - if (skb) - vlan_pull_tag(skb); - return skb; -} - -static struct sk_buff *set_dl_addr(struct sk_buff *skb, - const struct odp_action_dl_addr *a) -{ - skb = make_writable(skb, 0); - if (skb) { - struct ethhdr *eh = eth_hdr(skb); - if (a->type == ODPAT_SET_DL_SRC) - memcpy(eh->h_source, a->dl_addr, ETH_ALEN); - else - memcpy(eh->h_dest, a->dl_addr, ETH_ALEN); - } - return skb; -} - -/* 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) -{ - __be32 diff[] = { ~from, to }; - - if (OVS_CB(skb)->ip_summed != OVS_CSUM_PARTIAL) { - *sum = csum_fold(csum_partial((char *)diff, sizeof(diff), - ~csum_unfold(*sum))); - if (OVS_CB(skb)->ip_summed == OVS_CSUM_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))); + return 0; } -static bool is_ip(struct sk_buff *skb, const struct odp_flow_key *key) +static bool is_ip(struct sk_buff *skb) { - return (key->dl_type == htons(ETH_P_IP) && + return (OVS_CB(skb)->flow->key.eth.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) +static __sum16 *get_l4_checksum(struct sk_buff *skb) { + u8 nw_proto = OVS_CB(skb)->flow->key.ip.proto; int transport_len = skb->len - skb_transport_offset(skb); - if (key->nw_proto == IPPROTO_TCP) { + if (nw_proto == IPPROTO_TCP) { if (likely(transport_len >= sizeof(struct tcphdr))) return &tcp_hdr(skb)->check; - } else if (key->nw_proto == IPPROTO_UDP) { + } else if (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, - const struct odp_flow_key *key, - const struct odp_action_nw_addr *a) +static int set_nw_addr(struct sk_buff *skb, const struct nlattr *a) { + __be32 new_nwaddr = nla_get_be32(a); struct iphdr *nh; __sum16 *check; __be32 *nwaddr; + int err; - if (unlikely(!is_ip(skb, key))) - return skb; + if (unlikely(!is_ip(skb))) + return 0; - skb = make_writable(skb, 0); - if (unlikely(!skb)) - return NULL; + err = make_writable(skb, skb_network_offset(skb) + + sizeof(struct iphdr)); + if (unlikely(err)) + return err; nh = ip_hdr(skb); - nwaddr = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr; + nwaddr = nla_type(a) == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->saddr : &nh->daddr; - check = get_l4_checksum(skb, key); + check = get_l4_checksum(skb); if (likely(check)) - update_csum(check, skb, *nwaddr, a->nw_addr, 1); - update_csum(&nh->check, skb, *nwaddr, a->nw_addr, 0); + inet_proto_csum_replace4(check, skb, *nwaddr, new_nwaddr, 1); + csum_replace4(&nh->check, *nwaddr, new_nwaddr); + + skb_clear_rxhash(skb); - *nwaddr = a->nw_addr; + *nwaddr = new_nwaddr; - return skb; + return 0; } -static struct sk_buff *set_nw_tos(struct sk_buff *skb, - const struct odp_flow_key *key, - const struct odp_action_nw_tos *a) +static int set_nw_tos(struct sk_buff *skb, u8 nw_tos) { - if (unlikely(!is_ip(skb, key))) - return skb; - - skb = make_writable(skb, 0); - if (skb) { - struct iphdr *nh = ip_hdr(skb); - u8 *f = &nh->tos; - u8 old = *f; - u8 new; - - /* Set the DSCP bits and preserve the ECN bits. */ - new = a->nw_tos | (nh->tos & INET_ECN_MASK); - update_csum(&nh->check, skb, htons((u16)old), - htons((u16)new), 0); - *f = new; - } - return skb; + struct iphdr *nh = ip_hdr(skb); + u8 old, new; + int err; + + if (unlikely(!is_ip(skb))) + return 0; + + err = make_writable(skb, skb_network_offset(skb) + + sizeof(struct iphdr)); + if (unlikely(err)) + return err; + + /* Set the DSCP bits and preserve the ECN bits. */ + old = nh->tos; + new = nw_tos | (nh->tos & INET_ECN_MASK); + csum_replace4(&nh->check, (__force __be32)old, + (__force __be32)new); + nh->tos = new; + + return 0; } -static struct sk_buff *set_tp_port(struct sk_buff *skb, - const struct odp_flow_key *key, - const struct odp_action_tp_port *a) +static int set_tp_port(struct sk_buff *skb, const struct nlattr *a) { struct udphdr *th; __sum16 *check; __be16 *port; + int err; - if (unlikely(!is_ip(skb, key))) - return skb; + if (unlikely(!is_ip(skb))) + return 0; - skb = make_writable(skb, 0); - if (unlikely(!skb)) - return NULL; + err = make_writable(skb, skb_transport_offset(skb) + + sizeof(struct tcphdr)); + if (unlikely(err)) + return err; /* Must follow make_writable() since that can move the skb data. */ - check = get_l4_checksum(skb, key); + check = get_l4_checksum(skb); if (unlikely(!check)) - return skb; + return 0; /* * Update port and checksum. @@ -330,45 +192,17 @@ static struct sk_buff *set_tp_port(struct sk_buff *skb, * 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; -} + 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); -/** - * 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)); + return 0; } static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port) { - struct dp_port *p; + struct vport *p; if (!skb) goto error; @@ -377,47 +211,33 @@ static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port) if (!p) goto error; - vport_send(p->vport, skb); + vport_send(p, skb); return; error: kfree_skb(skb); } -static int output_control(struct datapath *dp, struct sk_buff *skb, u32 arg) +static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg) { + struct dp_upcall_info upcall; + skb = skb_clone(skb, GFP_ATOMIC); if (!skb) return -ENOMEM; - return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg); -} - -/* Send a copy of this packet up to the sFlow agent, along with extra - * 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, - 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_ATOMIC); - if (!nskb) - return; - memcpy(__skb_push(nskb, actlen), a, actlen); - hdr = (struct odp_sflow_sample_header*)__skb_push(nskb, hdrlen); - hdr->n_actions = n_actions; - hdr->sample_pool = atomic_read(&dp_port->sflow_pool); - dp_output_control(dp, nskb, _ODPL_SFLOW_NR, 0); + upcall.cmd = ODP_PACKET_CMD_ACTION; + upcall.key = &OVS_CB(skb)->flow->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, - const struct odp_flow_key *key, - const union odp_action *a, int n_actions) +static int do_execute_actions(struct datapath *dp, struct sk_buff *skb, + struct sw_flow_actions *acts) { /* 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 @@ -425,92 +245,147 @@ int execute_actions(struct datapath *dp, struct sk_buff *skb, * is slightly obscure just to avoid that. */ int prev_port = -1; u32 priority = skb->priority; - int err; + const struct nlattr *a; + int rem; - if (dp->sflow_probability) { - struct dp_port *p = OVS_CB(skb)->dp_port; - if (p) { - atomic_inc(&p->sflow_pool); - if (dp->sflow_probability == UINT_MAX || - net_random() < dp->sflow_probability) - sflow_sample(dp, skb, a, n_actions, p); - } - } + for (a = acts->actions, rem = acts->actions_len; rem > 0; + a = nla_next(a, &rem)) { + int err = 0; - OVS_CB(skb)->tun_id = 0; - - for (; n_actions > 0; a++, n_actions--) { if (prev_port != -1) { 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_CONTROLLER: - err = output_control(dp, skb, a->controller.arg); - if (err) { - kfree_skb(skb); - return err; - } + case ODP_ACTION_ATTR_CONTROLLER: + err = output_control(dp, skb, nla_get_u64(a)); break; - case ODPAT_SET_TUNNEL: - OVS_CB(skb)->tun_id = a->tunnel.tun_id; + case ODP_ACTION_ATTR_SET_TUNNEL: + OVS_CB(skb)->tun_id = nla_get_be64(a); break; - case ODPAT_SET_DL_TCI: - skb = modify_vlan_tci(dp, skb, key, a, n_actions); - if (IS_ERR(skb)) - return PTR_ERR(skb); + case ODP_ACTION_ATTR_SET_DL_TCI: + err = modify_vlan_tci(skb, nla_get_be16(a)); break; - case ODPAT_STRIP_VLAN: - skb = strip_vlan(skb); + case ODP_ACTION_ATTR_STRIP_VLAN: + err = strip_vlan(skb); break; - case ODPAT_SET_DL_SRC: - case ODPAT_SET_DL_DST: - skb = set_dl_addr(skb, &a->dl_addr); + case ODP_ACTION_ATTR_SET_DL_SRC: + err = make_writable(skb, ETH_HLEN); + if (likely(!err)) + memcpy(eth_hdr(skb)->h_source, nla_data(a), ETH_ALEN); break; - case ODPAT_SET_NW_SRC: - case ODPAT_SET_NW_DST: - skb = set_nw_addr(skb, key, &a->nw_addr); + case ODP_ACTION_ATTR_SET_DL_DST: + err = make_writable(skb, ETH_HLEN); + if (likely(!err)) + memcpy(eth_hdr(skb)->h_dest, nla_data(a), ETH_ALEN); break; - case ODPAT_SET_NW_TOS: - skb = set_nw_tos(skb, key, &a->nw_tos); + case ODP_ACTION_ATTR_SET_NW_SRC: + case ODP_ACTION_ATTR_SET_NW_DST: + err = set_nw_addr(skb, a); break; - case ODPAT_SET_TP_SRC: - case ODPAT_SET_TP_DST: - skb = set_tp_port(skb, key, &a->tp_port); + case ODP_ACTION_ATTR_SET_NW_TOS: + err = set_nw_tos(skb, nla_get_u8(a)); break; - case ODPAT_SET_PRIORITY: - skb->priority = a->priority.priority; + case ODP_ACTION_ATTR_SET_TP_SRC: + case ODP_ACTION_ATTR_SET_TP_DST: + err = set_tp_port(skb, a); break; - case ODPAT_POP_PRIORITY: - skb->priority = priority; + case ODP_ACTION_ATTR_SET_PRIORITY: + skb->priority = nla_get_u32(a); break; - case ODPAT_DROP_SPOOFED_ARP: - if (unlikely(is_spoofed_arp(skb, key))) - goto exit; + case ODP_ACTION_ATTR_POP_PRIORITY: + skb->priority = priority; break; } - if (!skb) - return -ENOMEM; + + if (unlikely(err)) { + kfree_skb(skb); + return err; + } } -exit: + if (prev_port != -1) do_output(dp, skb, prev_port); else - kfree_skb(skb); + consume_skb(skb); + return 0; } + +static void sflow_sample(struct datapath *dp, struct sk_buff *skb, + struct sw_flow_actions *acts) +{ + 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 = &OVS_CB(skb)->flow->key; + upcall.userdata = 0; + upcall.sample_pool = atomic_read(&p->sflow_pool); + upcall.actions = acts->actions; + upcall.actions_len = acts->actions_len; + dp_upcall(dp, nskb, &upcall); +} + +/* Execute a list of actions against 'skb'. */ +int execute_actions(struct datapath *dp, struct sk_buff *skb) +{ + struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts); + struct loop_counter *loop; + int error; + + /* Check whether we've looped too much. */ + loop = loop_get_counter(); + if (unlikely(++loop->count > MAX_LOOPS)) + loop->looping = true; + if (unlikely(loop->looping)) { + error = loop_suppress(dp, acts); + kfree_skb(skb); + goto out_loop; + } + + /* Really execute actions. */ + if (dp->sflow_probability) + sflow_sample(dp, skb, acts); + OVS_CB(skb)->tun_id = 0; + error = do_execute_actions(dp, skb, acts); + + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + error = loop_suppress(dp, acts); + +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + loop_put_counter(); + + return error; +}