X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=d405fc1525370a9a7e1a4ad4dffca34b57283c93;hb=8f56e1fd10878ebad7fcbe6b8659862d9be3e6d7;hp=193a4526da5957ccc5af519750f7289a264d4bd7;hpb=5caaf64f202b5daaa6eda08aa279202b16524062;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index 193a4526d..d405fc152 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -29,6 +29,7 @@ #include #include #include +#include #include "openflow-netlink.h" #include "datapath.h" @@ -41,10 +42,37 @@ #include "compat.h" +/* Strings to describe the manufacturer, hardware, and software. This data + * is queriable through the switch description stats message. */ +static char mfr_desc[DESC_STR_LEN] = "Nicira Networks"; +static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module"; +static char sw_desc[DESC_STR_LEN] = VERSION; +static char serial_num[SERIAL_NUM_LEN] = "None"; + +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0) +module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444); +module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444); +module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444); +module_param_string(serial_num, serial_num, sizeof serial_num, 0444); +#else +MODULE_PARM(mfr_desc, "s"); +MODULE_PARM(hw_desc, "s"); +MODULE_PARM(sw_desc, "s"); +MODULE_PARM(serial_num, "s"); +#endif + + /* Number of milliseconds between runs of the maintenance thread. */ #define MAINT_SLEEP_MSECS 1000 -#define BRIDGE_PORT_NO_FLOOD 0x00000001 +enum br_port_flags { + BRPF_NO_FLOOD = 1 << 0, +}; + +enum br_port_status { + BRPS_PORT_DOWN = 1 << 0, + BRPS_LINK_DOWN = 1 << 1, +}; #define UINT32_MAX 4294967295U #define UINT16_MAX 65535 @@ -52,7 +80,10 @@ struct net_bridge_port { u16 port_no; - u32 flags; + u32 flags; /* BRPF_* flags */ + u32 status; /* BRPS_* flags */ + spinlock_t lock; + struct work_struct port_task; struct datapath *dp; struct net_device *dev; struct list_head node; /* Element in datapath.ports. */ @@ -77,6 +108,7 @@ DEFINE_MUTEX(dp_mutex); EXPORT_SYMBOL(dp_mutex); static int dp_maint_func(void *data); +static int update_port_status(struct net_bridge_port *p); static int send_port_status(struct net_bridge_port *p, uint8_t status); static int dp_genl_openflow_done(struct netlink_callback *); static struct net_bridge_port *new_nbp(struct datapath *, @@ -338,6 +370,8 @@ static struct net_bridge_port *new_nbp(struct datapath *dp, p->dp = dp; p->dev = dev; p->port_no = port_no; + spin_lock_init(&p->lock); + INIT_WORK(&p->port_task, NULL); if (port_no != OFPP_LOCAL) rcu_assign_pointer(dev->br_port, p); if (port_no < OFPP_MAX) @@ -364,6 +398,8 @@ int add_switch_port(struct datapath *dp, struct net_device *dev) if (IS_ERR(p)) return PTR_ERR(p); + update_port_status(p); + /* Notify the ctlpath that this port has been added */ send_port_status(p, OFPPR_ADD); @@ -374,6 +410,7 @@ int add_switch_port(struct datapath *dp, struct net_device *dev) static int del_switch_port(struct net_bridge_port *p) { /* First drop references to device. */ + cancel_work_sync(&p->port_task); rtnl_lock(); dev_set_promiscuity(p->dev, -1); rtnl_unlock(); @@ -426,6 +463,16 @@ static int dp_maint_func(void *data) struct datapath *dp = (struct datapath *) data; while (!kthread_should_stop()) { + struct net_bridge_port *p; + + /* Check if port status has changed */ + rcu_read_lock(); + list_for_each_entry_rcu (p, &dp->port_list, node) + if (update_port_status(p)) + send_port_status(p, OFPPR_MOD); + rcu_read_unlock(); + + /* Timeout old entries */ chain_timeout(dp->chain); msleep_interruptible(MAINT_SLEEP_MSECS); } @@ -490,7 +537,7 @@ static inline unsigned packet_length(const struct sk_buff *skb) static int output_all(struct datapath *dp, struct sk_buff *skb, int flood) { - u32 disable = flood ? BRIDGE_PORT_NO_FLOOD : 0; + u32 disable = flood ? BRPF_NO_FLOOD : 0; struct net_bridge_port *p; int prev_port = -1; @@ -530,50 +577,80 @@ int dp_set_origin(struct datapath *dp, uint16_t in_port, return -ENOENT; } +static int xmit_skb(struct sk_buff *skb) +{ + int len = skb->len; + if (packet_length(skb) > skb->dev->mtu) { + printk("dropped over-mtu packet: %d > %d\n", + packet_length(skb), skb->dev->mtu); + kfree_skb(skb); + return -E2BIG; + } + + dev_queue_xmit(skb); + + return len; +} + /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'. */ int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port) { BUG_ON(!skb); - if (out_port == OFPP_FLOOD) + switch (out_port){ + case OFPP_IN_PORT: + /* Send it out the port it came in on, which is already set in + * the skb. */ + if (!skb->dev) { + if (net_ratelimit()) + printk("skb device not set forwarding to in_port\n"); + kfree(skb); + return -ESRCH; + } + return xmit_skb(skb); + + case OFPP_TABLE: { + struct net_bridge_port *p = skb->dev->br_port; + int retval; + retval = run_flow_through_tables(dp->chain, skb, + p ? p->port_no : OFPP_LOCAL); + if (retval) + kfree_skb(skb); + return retval; + } + + case OFPP_FLOOD: return output_all(dp, skb, 1); - else if (out_port == OFPP_ALL) + + case OFPP_ALL: return output_all(dp, skb, 0); - else if (out_port == OFPP_CONTROLLER) + + case OFPP_CONTROLLER: return dp_output_control(dp, skb, fwd_save_skb(skb), 0, OFPR_ACTION); - else if (out_port == OFPP_TABLE) { - struct net_bridge_port *p = skb->dev->br_port; - struct sw_flow_key key; - struct sw_flow *flow; - - flow_extract(skb, p ? p->port_no : OFPP_LOCAL, &key); - flow = chain_lookup(dp->chain, &key); - if (likely(flow != NULL)) { - flow_used(flow, skb); - execute_actions(dp, skb, &key, flow->actions, flow->n_actions); - return 0; - } - return -ESRCH; - } else if (out_port == OFPP_LOCAL) { + + case OFPP_LOCAL: { struct net_device *dev = dp->netdev; return dev ? dp_dev_recv(dev, skb) : -ESRCH; - } else if (out_port >= 0 && out_port < OFPP_MAX) { + } + + case 0 ... OFPP_MAX-1: { struct net_bridge_port *p = dp->ports[out_port]; - int len = skb->len; if (p == NULL) goto bad_port; - skb->dev = p->dev; - if (packet_length(skb) > skb->dev->mtu) { - printk("dropped over-mtu packet: %d > %d\n", - packet_length(skb), skb->dev->mtu); + if (p->dev == skb->dev) { + /* To send to the input port, must use OFPP_IN_PORT */ kfree_skb(skb); - return -E2BIG; + if (net_ratelimit()) + printk("can't directly forward to input port\n"); + return -EINVAL; } + skb->dev = p->dev; + return xmit_skb(skb); + } - dev_queue_xmit(skb); - - return len; + default: + goto bad_port; } bad_port: @@ -627,14 +704,24 @@ out: static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc) { + unsigned long flags; desc->port_no = htons(p->port_no); strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN); desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0'; memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN); - desc->flags = htonl(p->flags); + desc->flags = 0; desc->features = 0; desc->speed = 0; + spin_lock_irqsave(&p->lock, flags); + if (p->flags & BRPF_NO_FLOOD) + desc->flags |= htonl(OFPPFL_NO_FLOOD); + else if (p->status & BRPS_PORT_DOWN) + desc->flags |= htonl(OFPPFL_PORT_DOWN); + else if (p->status & BRPS_LINK_DOWN) + desc->flags |= htonl(OFPPFL_LINK_DOWN); + spin_unlock_irqrestore(&p->lock, flags); + #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24) if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) { struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET }; @@ -728,9 +815,41 @@ dp_send_config_reply(struct datapath *dp, const struct sender *sender) return send_openflow_skb(skb, sender); } +/* Callback function for a workqueue to disable an interface */ +static void +down_port_cb(struct work_struct *work) +{ + struct net_bridge_port *p = container_of(work, struct net_bridge_port, + port_task); + + rtnl_lock(); + if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0) + if (net_ratelimit()) + printk("problem bringing up port %s\n", p->dev->name); + rtnl_unlock(); + p->status |= BRPS_PORT_DOWN; +} + +/* Callback function for a workqueue to enable an interface */ +static void +up_port_cb(struct work_struct *work) +{ + struct net_bridge_port *p = container_of(work, struct net_bridge_port, + port_task); + + rtnl_lock(); + if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0) + if (net_ratelimit()) + printk("problem bringing down port %s\n", p->dev->name); + rtnl_unlock(); + p->status &= ~BRPS_PORT_DOWN; +} + int -dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp) +dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm) { + unsigned long int flags; + const struct ofp_phy_port *opp = &opm->desc; int port_no = ntohs(opp->port_no); struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no] : port_no == OFPP_LOCAL ? dp->local_port @@ -738,10 +857,63 @@ dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp) /* Make sure the port id hasn't changed since this was sent */ if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN)) return -1; - p->flags = htonl(opp->flags); + + spin_lock_irqsave(&p->lock, flags); + if (opm->mask & htonl(OFPPFL_NO_FLOOD)) { + if (opp->flags & htonl(OFPPFL_NO_FLOOD)) + p->flags |= BRPF_NO_FLOOD; + else + p->flags &= ~BRPF_NO_FLOOD; + } + + /* Modifying the status of an interface requires taking a lock + * that cannot be done from here. For this reason, we use a shared + * workqueue, which will cause it to be executed from a safer + * context. */ + if (opm->mask & htonl(OFPPFL_PORT_DOWN)) { + if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) + && (p->status & BRPS_PORT_DOWN) == 0) { + PREPARE_WORK(&p->port_task, down_port_cb); + schedule_work(&p->port_task); + } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0 + && (p->status & BRPS_PORT_DOWN)) { + PREPARE_WORK(&p->port_task, up_port_cb); + schedule_work(&p->port_task); + } + } + spin_unlock_irqrestore(&p->lock, flags); + return 0; } +/* Update the port status field of the bridge port. A non-zero return + * value indicates some field has changed. + * + * NB: Callers of this function may hold the RCU read lock, so any + * additional checks must not sleep. + */ +static int +update_port_status(struct net_bridge_port *p) +{ + unsigned long int flags; + uint32_t orig_status; + + spin_lock_irqsave(&p->lock, flags); + orig_status = p->status; + + if (p->dev->flags & IFF_UP) + p->status &= ~BRPS_PORT_DOWN; + else + p->status |= BRPS_PORT_DOWN; + + if (netif_carrier_ok(p->dev)) + p->status &= ~BRPS_LINK_DOWN; + else + p->status |= BRPS_LINK_DOWN; + + spin_unlock_irqrestore(&p->lock, flags); + return (orig_status != p->status); +} static int send_port_status(struct net_bridge_port *p, uint8_t status) @@ -761,11 +933,14 @@ send_port_status(struct net_bridge_port *p, uint8_t status) } int -dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow) +dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow, + enum ofp_flow_expired_reason reason) { struct sk_buff *skb; struct ofp_flow_expired *ofe; - unsigned long duration_j; + + if (!(dp->flags & OFPC_SEND_FLOW_EXP)) + return 0; ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb); if (!ofe) @@ -773,11 +948,12 @@ dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow) flow_fill_match(&ofe->match, &flow->key); - memset(ofe->pad, 0, sizeof ofe->pad); ofe->priority = htons(flow->priority); + ofe->reason = reason; + memset(ofe->pad, 0, sizeof ofe->pad); - duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time; - ofe->duration = htonl(duration_j / HZ); + ofe->duration = htonl((jiffies - flow->init_time) / HZ); + memset(ofe->pad2, 0, sizeof ofe->pad2); ofe->packet_count = cpu_to_be64(flow->packet_count); ofe->byte_count = cpu_to_be64(flow->byte_count); @@ -1045,6 +1221,25 @@ static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = { [DP_GENL_A_DP_IDX] = { .type = NLA_U32 }, }; +static int desc_stats_dump(struct datapath *dp, void *state, + void *body, int *body_len) +{ + struct ofp_desc_stats *ods = body; + int n_bytes = sizeof *ods; + + if (n_bytes > *body_len) { + return -ENOBUFS; + } + *body_len = n_bytes; + + strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc); + strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc); + strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc); + strncpy(ods->serial_num, serial_num, sizeof ods->serial_num); + + return 0; +} + struct flow_stats_state { int table_idx; struct sw_table_position position; @@ -1084,7 +1279,7 @@ static int flow_stats_dump_callback(struct sw_flow *flow, void *private) ofs->length = htons(length); ofs->table_id = s->table_idx; ofs->pad = 0; - ofs->match.wildcards = htons(flow->key.wildcards); + ofs->match.wildcards = htonl(flow->key.wildcards); ofs->match.in_port = flow->key.in_port; memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN); memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN); @@ -1093,14 +1288,16 @@ static int flow_stats_dump_callback(struct sw_flow *flow, void *private) ofs->match.nw_src = flow->key.nw_src; ofs->match.nw_dst = flow->key.nw_dst; ofs->match.nw_proto = flow->key.nw_proto; - memset(ofs->match.pad, 0, sizeof ofs->match.pad); + ofs->match.pad = 0; ofs->match.tp_src = flow->key.tp_src; ofs->match.tp_dst = flow->key.tp_dst; ofs->duration = htonl((jiffies - flow->init_time) / HZ); + ofs->priority = htons(flow->priority); + ofs->idle_timeout = htons(flow->idle_timeout); + ofs->hard_timeout = htons(flow->hard_timeout); + memset(ofs->pad2, 0, sizeof ofs->pad2); ofs->packet_count = cpu_to_be64(flow->packet_count); ofs->byte_count = cpu_to_be64(flow->byte_count); - ofs->priority = htons(flow->priority); - ofs->max_idle = htons(flow->max_idle); memcpy(ofs->actions, flow->actions, actions_length); s->bytes_used += length; @@ -1208,11 +1405,11 @@ static int table_stats_dump(struct datapath *dp, void *state, void *body, int *body_len) { struct ofp_table_stats *ots; - int nbytes = dp->chain->n_tables * sizeof *ots; + int n_bytes = dp->chain->n_tables * sizeof *ots; int i; - if (nbytes > *body_len) + if (n_bytes > *body_len) return -ENOBUFS; - *body_len = nbytes; + *body_len = n_bytes; for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) { struct sw_table_stats stats; dp->chain->tables[i]->stats(dp->chain->tables[i], &stats); @@ -1263,10 +1460,18 @@ static int port_stats_dump(struct datapath *dp, void *state, stats = p->dev->get_stats(p->dev); ops->port_no = htons(p->port_no); memset(ops->pad, 0, sizeof ops->pad); - ops->rx_count = cpu_to_be64(stats->rx_packets); - ops->tx_count = cpu_to_be64(stats->tx_packets); - ops->drop_count = cpu_to_be64(stats->rx_dropped - + stats->tx_dropped); + ops->rx_packets = cpu_to_be64(stats->rx_packets); + ops->tx_packets = cpu_to_be64(stats->tx_packets); + ops->rx_bytes = cpu_to_be64(stats->rx_bytes); + ops->tx_bytes = cpu_to_be64(stats->tx_bytes); + ops->rx_dropped = cpu_to_be64(stats->rx_dropped); + ops->tx_dropped = cpu_to_be64(stats->tx_dropped); + ops->rx_errors = cpu_to_be64(stats->rx_errors); + ops->tx_errors = cpu_to_be64(stats->tx_errors); + ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors); + ops->rx_over_err = cpu_to_be64(stats->rx_over_errors); + ops->rx_crc_err = cpu_to_be64(stats->rx_crc_errors); + ops->collisions = cpu_to_be64(stats->collisions); n_ports++; ops++; } @@ -1306,6 +1511,13 @@ struct stats_type { }; static const struct stats_type stats[] = { + [OFPST_DESC] = { + 0, + 0, + NULL, + desc_stats_dump, + NULL + }, [OFPST_FLOW] = { sizeof(struct ofp_flow_stats_request), sizeof(struct ofp_flow_stats_request), @@ -1506,19 +1718,12 @@ static void dp_uninit_netlink(void) genl_unregister_family(&dp_genl_family); } -#define DRV_NAME "openflow" -#define DRV_VERSION VERSION -#define DRV_DESCRIPTION "OpenFlow switching datapath implementation" -#define DRV_COPYRIGHT "Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University" - - static int __init dp_init(void) { int err; - printk(KERN_INFO DRV_NAME ": " DRV_DESCRIPTION "\n"); - printk(KERN_INFO DRV_NAME ": " VERSION" built on "__DATE__" "__TIME__"\n"); - printk(KERN_INFO DRV_NAME ": " DRV_COPYRIGHT "\n"); + printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", " + "protocol 0x%02x\n", OFP_VERSION); err = flow_init(); if (err) @@ -1554,6 +1759,6 @@ static void dp_cleanup(void) module_init(dp_init); module_exit(dp_cleanup); -MODULE_DESCRIPTION(DRV_DESCRIPTION); -MODULE_AUTHOR(DRV_COPYRIGHT); +MODULE_DESCRIPTION("OpenFlow switching datapath"); +MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University"); MODULE_LICENSE("GPL");