2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
4 * Stanford Junior University
7 /* Functions for managing the dp interface/device. */
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/if_arp.h>
12 #include <linux/if_bridge.h>
13 #include <linux/if_vlan.h>
15 #include <net/genetlink.h>
17 #include <linux/delay.h>
18 #include <linux/etherdevice.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/rtnetlink.h>
23 #include <linux/rcupdate.h>
24 #include <linux/version.h>
25 #include <linux/ethtool.h>
26 #include <linux/random.h>
27 #include <asm/system.h>
28 #include <linux/netfilter_bridge.h>
29 #include <linux/netfilter_ipv4.h>
30 #include <linux/inetdevice.h>
31 #include <linux/list.h>
32 #include <linux/rculist.h>
33 #include <linux/workqueue.h>
34 #include <linux/dmi.h>
36 #include "openflow/nicira-ext.h"
37 #include "openflow/openflow-netlink.h"
39 #include "nx_act_snat.h"
49 /* Strings to describe the manufacturer, hardware, and software. This data
50 * is queriable through the switch description stats message. */
51 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks, Inc.";
52 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
53 static char sw_desc[DESC_STR_LEN] = VERSION BUILDNR;
54 static char serial_num[SERIAL_NUM_LEN] = "None";
56 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
57 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
58 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
59 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
60 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
62 MODULE_PARM(mfr_desc, "s");
63 MODULE_PARM(hw_desc, "s");
64 MODULE_PARM(sw_desc, "s");
65 MODULE_PARM(serial_num, "s");
69 /* Number of milliseconds between runs of the maintenance thread. */
70 #define MAINT_SLEEP_MSECS 1000
72 #define UINT32_MAX 4294967295U
73 #define UINT16_MAX 65535
74 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
76 static struct genl_family dp_genl_family;
77 static struct genl_multicast_group mc_group;
79 /* It's hard to imagine wanting more than one datapath, but... */
82 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
83 * by dp_mutex. dp_mutex is almost completely redundant with genl_mutex
84 * maintained by the Generic Netlink code, but the timeout path needs mutual
87 * It is safe to access the datapath and net_bridge_port structures with just
90 static struct datapath *dps[DP_MAX];
91 DEFINE_MUTEX(dp_mutex);
92 EXPORT_SYMBOL(dp_mutex);
94 static int dp_maint_func(void *data);
95 static void init_port_status(struct net_bridge_port *p);
96 static int dp_genl_openflow_done(struct netlink_callback *);
97 static struct net_bridge_port *new_nbp(struct datapath *,
98 struct net_device *, int port_no);
100 /* nla_shrink - reduce amount of space reserved by nla_reserve
101 * @skb: socket buffer from which to recover room
102 * @nla: netlink attribute to adjust
103 * @len: new length of attribute payload
105 * Reduces amount of space reserved by a call to nla_reserve.
107 * No other attributes may be added between calling nla_reserve and this
108 * function, since it will create a hole in the message.
110 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
112 int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
116 nla->nla_len = nla_attr_size(len);
119 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
120 * If 'sender' is nonnull, then it is used as the message's destination. 'dp'
121 * must specify the datapath to use.
123 * '*max_openflow_len' receives the maximum number of bytes that are available
124 * for the embedded OpenFlow message. The caller must call
125 * resize_openflow_skb() to set the actual size of the message to this number
128 * Returns the openflow header if successful, otherwise (if 'skb' is too small)
131 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
132 const struct sender *sender, int *max_openflow_len)
134 struct ofp_header *oh;
138 /* Assemble the Generic Netlink wrapper. */
139 if (!genlmsg_put(skb,
140 sender ? sender->pid : 0,
141 sender ? sender->seq : 0,
142 &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
143 return ERR_PTR(-ENOBUFS);
144 if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
145 return ERR_PTR(-ENOBUFS);
146 openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
147 if (openflow_len < sizeof *oh)
148 return ERR_PTR(-ENOBUFS);
149 *max_openflow_len = openflow_len;
150 attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
153 /* Fill in the header. The caller is responsible for the length. */
155 oh->version = OFP_VERSION;
157 oh->xid = sender ? sender->xid : 0;
162 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
163 * length 'new_length' (in bytes), adjusting pointers and size values as
166 resize_openflow_skb(struct sk_buff *skb,
167 struct ofp_header *oh, size_t new_length)
169 struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
170 nla_shrink(skb, attr, new_length);
171 oh->length = htons(new_length);
172 nlmsg_end(skb, (struct nlmsghdr *) skb->data);
175 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
176 * length. Returns a null pointer if memory is unavailable, otherwise returns
177 * the OpenFlow header and stores a pointer to the skb in '*pskb'.
179 * 'type' is the OpenFlow message type. If 'sender' is nonnull, then it is
180 * used as the message's destination. 'dp' must specify the datapath to
183 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
184 const struct sender *sender, struct sk_buff **pskb)
186 struct ofp_header *oh;
189 int max_openflow_len;
191 if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
193 printk("alloc_openflow_skb: openflow message too large: %zu\n",
198 genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
199 genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
200 genl_len += nla_total_size(openflow_len); /* DP_GENL_A_OPENFLOW */
201 skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
204 printk("alloc_openflow_skb: genlmsg_new failed\n");
208 oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
209 BUG_ON(!oh || IS_ERR(oh));
210 resize_openflow_skb(skb, oh, openflow_len);
215 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
218 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
221 ? genlmsg_unicast(skb, sender->pid)
222 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
225 /* Retrieves the datapath id, which is the MAC address of the "of" device. */
227 uint64_t get_datapath_id(struct net_device *dev)
232 for (i=0; i<ETH_ALEN; i++)
233 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
238 /* Creates a new datapath numbered 'dp_idx'. Returns 0 for success or a
239 * negative error code. */
240 static int new_dp(int dp_idx)
245 if (dp_idx < 0 || dp_idx >= DP_MAX)
248 if (!try_module_get(THIS_MODULE))
251 /* Exit early if a datapath with that number already exists. */
258 dp = kzalloc(sizeof *dp, GFP_KERNEL);
262 /* Setup our "of" device */
263 err = dp_dev_setup(dp);
268 dp->id = get_datapath_id(dp->netdev);
269 dp->chain = chain_create(dp);
270 if (dp->chain == NULL)
271 goto err_destroy_dp_dev;
272 INIT_LIST_HEAD(&dp->port_list);
274 dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
275 if (IS_ERR(dp->local_port)) {
276 err = PTR_ERR(dp->local_port);
277 goto err_destroy_local_port;
281 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
283 dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
284 if (IS_ERR(dp->dp_task))
285 goto err_destroy_chain;
291 err_destroy_local_port:
292 dp_del_switch_port(dp->local_port);
294 chain_destroy(dp->chain);
300 module_put(THIS_MODULE);
304 /* Find and return a free port number under 'dp'. */
305 static int find_portno(struct datapath *dp)
308 for (i = 0; i < DP_MAX_PORTS; i++)
309 if (dp->ports[i] == NULL)
314 static struct net_bridge_port *new_nbp(struct datapath *dp,
315 struct net_device *dev, int port_no)
317 struct net_bridge_port *p;
319 if (dev->br_port != NULL)
320 return ERR_PTR(-EBUSY);
322 p = kzalloc(sizeof(*p), GFP_KERNEL);
324 return ERR_PTR(-ENOMEM);
327 dev_set_promiscuity(dev, 1);
332 p->port_no = port_no;
333 spin_lock_init(&p->lock);
334 INIT_WORK(&p->port_task, NULL);
335 if (port_no != OFPP_LOCAL)
336 rcu_assign_pointer(dev->br_port, p);
337 if (port_no < DP_MAX_PORTS)
338 rcu_assign_pointer(dp->ports[port_no], p);
339 list_add_rcu(&p->node, &dp->port_list);
344 int add_switch_port(struct datapath *dp, struct net_device *dev)
346 struct net_bridge_port *p;
349 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
353 port_no = find_portno(dp);
357 p = new_nbp(dp, dev, port_no);
363 /* Notify the ctlpath that this port has been added */
364 dp_send_port_status(p, OFPPR_ADD);
369 /* Delete 'p' from switch. */
370 int dp_del_switch_port(struct net_bridge_port *p)
376 /* First drop references to device. */
377 cancel_work_sync(&p->port_task);
379 dev_set_promiscuity(p->dev, -1);
381 list_del_rcu(&p->node);
382 if (p->port_no != OFPP_LOCAL)
383 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
384 rcu_assign_pointer(p->dev->br_port, NULL);
386 /* Then wait until no one is still using it, and destroy it. */
390 /* Free any SNAT configuration on the port. */
391 spin_lock_irqsave(&p->lock, flags);
393 spin_unlock_irqrestore(&p->lock, flags);
396 /* Notify the ctlpath that this port no longer exists */
397 dp_send_port_status(p, OFPPR_DELETE);
405 static void del_dp(struct datapath *dp)
407 struct net_bridge_port *p, *n;
409 kthread_stop(dp->dp_task);
411 /* Drop references to DP. */
412 list_for_each_entry_safe (p, n, &dp->port_list, node)
413 dp_del_switch_port(p);
414 rcu_assign_pointer(dps[dp->dp_idx], NULL);
416 /* Kill off local_port dev references from buffered packets that have
417 * associated dst entries. */
421 /* Destroy dp->netdev. (Must follow deleting switch ports since
422 * dp->local_port has a reference to it.) */
425 /* Wait until no longer in use, then destroy it. */
427 chain_destroy(dp->chain);
429 module_put(THIS_MODULE);
432 static int dp_maint_func(void *data)
434 struct datapath *dp = (struct datapath *) data;
436 while (!kthread_should_stop()) {
438 struct net_bridge_port *p;
440 /* Expire old SNAT entries */
442 list_for_each_entry_rcu (p, &dp->port_list, node)
447 /* Timeout old entries */
448 chain_timeout(dp->chain);
449 msleep_interruptible(MAINT_SLEEP_MSECS);
456 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
459 /* Check if this packet needs early SNAT processing. */
460 if (snat_pre_route(skb)) {
466 /* Push the Ethernet header back on. */
467 skb_push(skb, ETH_HLEN);
468 fwd_port_input(p->dp->chain, skb, p);
472 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
473 * different set of devices!)
475 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
476 /* Called with rcu_read_lock. */
477 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
480 do_port_input(p, skb);
483 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
484 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
486 do_port_input(p, *pskb);
490 /* NB: This has only been tested on 2.4.35 */
491 static void dp_frame_hook(struct sk_buff *skb)
493 struct net_bridge_port *p = skb->dev->br_port;
496 do_port_input(p, skb);
503 /* Forwarding output path.
504 * Based on net/bridge/br_forward.c. */
506 static inline unsigned packet_length(const struct sk_buff *skb)
508 int length = skb->len - ETH_HLEN;
509 if (skb->protocol == htons(ETH_P_8021Q))
514 /* Send packets out all the ports except the originating one. If the
515 * "flood" argument is set, only send along the minimum spanning tree.
518 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
520 u32 disable = flood ? OFPPC_NO_FLOOD : 0;
521 struct net_bridge_port *p;
524 list_for_each_entry_rcu (p, &dp->port_list, node) {
525 if (skb->dev == p->dev || p->config & disable)
527 if (prev_port != -1) {
528 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
533 dp_output_port(dp, clone, prev_port, 0);
535 prev_port = p->port_no;
538 dp_output_port(dp, skb, prev_port, 0);
545 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
546 FIXME: how are devices reference counted? */
547 void dp_set_origin(struct datapath *dp, uint16_t in_port,
550 struct net_bridge_port *p;
551 p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
552 : in_port == OFPP_LOCAL ? dp->local_port
561 dp_xmit_skb(struct sk_buff *skb)
564 if (packet_length(skb) > skb->dev->mtu) {
565 printk("dropped over-mtu packet: %d > %d\n",
566 packet_length(skb), skb->dev->mtu);
576 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
578 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
584 /* Send it out the port it came in on, which is already set in
588 printk("skb device not set forwarding to in_port\n");
592 return dp_xmit_skb(skb);
595 int retval = run_flow_through_tables(dp->chain, skb,
603 return output_all(dp, skb, 1);
606 return output_all(dp, skb, 0);
608 case OFPP_CONTROLLER:
609 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
613 struct net_device *dev = dp->netdev;
617 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
620 case 0 ... DP_MAX_PORTS - 1: {
621 struct net_bridge_port *p = dp->ports[out_port];
624 if (p->dev == skb->dev) {
625 /* To send to the input port, must use OFPP_IN_PORT */
628 printk("can't directly forward to input port\n");
631 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
636 return dp_xmit_skb(skb);
646 printk("can't forward to bad port %d\n", out_port);
650 /* Takes ownership of 'skb' and transmits it to 'dp''s control path. If
651 * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
652 * otherwise, all of 'skb' is sent. 'reason' indicates why 'skb' is being
653 * sent. 'max_len' sets the maximum number of bytes that the caller
654 * wants to be sent; a value of 0 indicates the entire packet should be
657 dp_output_control(struct datapath *dp, struct sk_buff *skb,
658 uint32_t buffer_id, size_t max_len, int reason)
660 /* FIXME? Can we avoid creating a new skbuff in the case where we
661 * forward the whole packet? */
662 struct sk_buff *f_skb;
663 struct ofp_packet_in *opi;
664 struct net_bridge_port *p;
665 size_t fwd_len, opi_len;
669 if ((buffer_id != (uint32_t) -1) && max_len)
670 fwd_len = min(fwd_len, max_len);
672 opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
673 opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
678 opi->buffer_id = htonl(buffer_id);
679 opi->total_len = htons(skb->len);
680 p = skb->dev->br_port;
681 opi->in_port = htons(p ? p->port_no : OFPP_LOCAL);
682 opi->reason = reason;
684 memcpy(opi->data, skb_mac_header(skb), fwd_len);
685 err = send_openflow_skb(f_skb, NULL);
692 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
695 desc->port_no = htons(p->port_no);
696 strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
697 desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
698 memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
701 desc->advertised = 0;
704 spin_lock_irqsave(&p->lock, flags);
705 desc->config = htonl(p->config);
706 desc->state = htonl(p->state);
707 spin_unlock_irqrestore(&p->lock, flags);
709 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
710 if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
711 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
713 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
714 /* Set the supported features */
715 if (ecmd.supported & SUPPORTED_10baseT_Half)
716 desc->supported |= OFPPF_10MB_HD;
717 if (ecmd.supported & SUPPORTED_10baseT_Full)
718 desc->supported |= OFPPF_10MB_FD;
719 if (ecmd.supported & SUPPORTED_100baseT_Half)
720 desc->supported |= OFPPF_100MB_HD;
721 if (ecmd.supported & SUPPORTED_100baseT_Full)
722 desc->supported |= OFPPF_100MB_FD;
723 if (ecmd.supported & SUPPORTED_1000baseT_Half)
724 desc->supported |= OFPPF_1GB_HD;
725 if (ecmd.supported & SUPPORTED_1000baseT_Full)
726 desc->supported |= OFPPF_1GB_FD;
727 if (ecmd.supported & SUPPORTED_10000baseT_Full)
728 desc->supported |= OFPPF_10GB_FD;
729 if (ecmd.supported & SUPPORTED_TP)
730 desc->supported |= OFPPF_COPPER;
731 if (ecmd.supported & SUPPORTED_FIBRE)
732 desc->supported |= OFPPF_FIBER;
733 if (ecmd.supported & SUPPORTED_Autoneg)
734 desc->supported |= OFPPF_AUTONEG;
735 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
736 if (ecmd.supported & SUPPORTED_Pause)
737 desc->supported |= OFPPF_PAUSE;
738 if (ecmd.supported & SUPPORTED_Asym_Pause)
739 desc->supported |= OFPPF_PAUSE_ASYM;
740 #endif /* kernel >= 2.6.14 */
742 /* Set the advertised features */
743 if (ecmd.advertising & ADVERTISED_10baseT_Half)
744 desc->advertised |= OFPPF_10MB_HD;
745 if (ecmd.advertising & ADVERTISED_10baseT_Full)
746 desc->advertised |= OFPPF_10MB_FD;
747 if (ecmd.advertising & ADVERTISED_100baseT_Half)
748 desc->advertised |= OFPPF_100MB_HD;
749 if (ecmd.advertising & ADVERTISED_100baseT_Full)
750 desc->advertised |= OFPPF_100MB_FD;
751 if (ecmd.advertising & ADVERTISED_1000baseT_Half)
752 desc->advertised |= OFPPF_1GB_HD;
753 if (ecmd.advertising & ADVERTISED_1000baseT_Full)
754 desc->advertised |= OFPPF_1GB_FD;
755 if (ecmd.advertising & ADVERTISED_10000baseT_Full)
756 desc->advertised |= OFPPF_10GB_FD;
757 if (ecmd.advertising & ADVERTISED_TP)
758 desc->advertised |= OFPPF_COPPER;
759 if (ecmd.advertising & ADVERTISED_FIBRE)
760 desc->advertised |= OFPPF_FIBER;
761 if (ecmd.advertising & ADVERTISED_Autoneg)
762 desc->advertised |= OFPPF_AUTONEG;
763 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
764 if (ecmd.advertising & ADVERTISED_Pause)
765 desc->advertised |= OFPPF_PAUSE;
766 if (ecmd.advertising & ADVERTISED_Asym_Pause)
767 desc->advertised |= OFPPF_PAUSE_ASYM;
768 #endif /* kernel >= 2.6.14 */
770 /* Set the current features */
771 if (ecmd.speed == SPEED_10)
772 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
773 else if (ecmd.speed == SPEED_100)
774 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
775 else if (ecmd.speed == SPEED_1000)
776 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
777 else if (ecmd.speed == SPEED_10000)
778 desc->curr = OFPPF_10GB_FD;
780 if (ecmd.port == PORT_TP)
781 desc->curr |= OFPPF_COPPER;
782 else if (ecmd.port == PORT_FIBRE)
783 desc->curr |= OFPPF_FIBER;
786 desc->curr |= OFPPF_AUTONEG;
790 desc->curr = htonl(desc->curr);
791 desc->supported = htonl(desc->supported);
792 desc->advertised = htonl(desc->advertised);
793 desc->peer = htonl(desc->peer);
797 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
799 struct net_bridge_port *p;
802 ofr->datapath_id = cpu_to_be64(dp->id);
804 ofr->n_buffers = htonl(N_PKT_BUFFERS);
805 ofr->n_tables = dp->chain->n_tables;
806 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
807 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
808 memset(ofr->pad, 0, sizeof ofr->pad);
810 list_for_each_entry_rcu (p, &dp->port_list, node) {
811 fill_port_desc(p, &ofr->ports[port_count]);
819 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
822 struct ofp_switch_features *ofr;
823 size_t ofr_len, port_max_len;
827 port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
828 ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
829 OFPT_FEATURES_REPLY, sender, &skb);
834 port_count = fill_features_reply(dp, ofr);
837 ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
838 resize_openflow_skb(skb, &ofr->header, ofr_len);
839 return send_openflow_skb(skb, sender);
843 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
846 struct ofp_switch_config *osc;
848 osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
853 osc->flags = htons(dp->flags);
854 osc->miss_send_len = htons(dp->miss_send_len);
856 return send_openflow_skb(skb, sender);
860 dp_send_hello(struct datapath *dp, const struct sender *sender,
861 const struct ofp_header *request)
863 if (request->version < OFP_VERSION) {
865 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
866 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
867 OFPHFC_INCOMPATIBLE, err, strlen(err));
871 struct ofp_header *reply;
873 reply = alloc_openflow_skb(dp, sizeof *reply,
874 OFPT_HELLO, sender, &skb);
878 return send_openflow_skb(skb, sender);
882 /* Callback function for a workqueue to disable an interface */
884 down_port_cb(struct work_struct *work)
886 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
890 if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
892 printk("problem bringing up port %s\n", p->dev->name);
894 p->config |= OFPPC_PORT_DOWN;
897 /* Callback function for a workqueue to enable an interface */
899 up_port_cb(struct work_struct *work)
901 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
905 if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
907 printk("problem bringing down port %s\n", p->dev->name);
909 p->config &= ~OFPPC_PORT_DOWN;
913 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
915 unsigned long int flags;
916 int port_no = ntohs(opm->port_no);
917 struct net_bridge_port *p;
918 p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
919 : port_no == OFPP_LOCAL ? dp->local_port
922 /* Make sure the port id hasn't changed since this was sent */
923 if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
926 spin_lock_irqsave(&p->lock, flags);
928 uint32_t config_mask = ntohl(opm->mask);
929 p->config &= ~config_mask;
930 p->config |= ntohl(opm->config) & config_mask;
933 /* Modifying the status of an interface requires taking a lock
934 * that cannot be done from here. For this reason, we use a shared
935 * workqueue, which will cause it to be executed from a safer
937 if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
938 if ((opm->config & htonl(OFPPC_PORT_DOWN))
939 && (p->config & OFPPC_PORT_DOWN) == 0) {
940 PREPARE_WORK(&p->port_task, down_port_cb);
941 schedule_work(&p->port_task);
942 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
943 && (p->config & OFPPC_PORT_DOWN)) {
944 PREPARE_WORK(&p->port_task, up_port_cb);
945 schedule_work(&p->port_task);
948 spin_unlock_irqrestore(&p->lock, flags);
953 /* Initialize the port status field of the bridge port. */
955 init_port_status(struct net_bridge_port *p)
957 unsigned long int flags;
959 spin_lock_irqsave(&p->lock, flags);
961 if (p->dev->flags & IFF_UP)
962 p->config &= ~OFPPC_PORT_DOWN;
964 p->config |= OFPPC_PORT_DOWN;
966 if (netif_carrier_ok(p->dev))
967 p->state &= ~OFPPS_LINK_DOWN;
969 p->state |= OFPPS_LINK_DOWN;
971 spin_unlock_irqrestore(&p->lock, flags);
975 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
978 struct ofp_port_status *ops;
980 ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
984 ops->reason = status;
985 memset(ops->pad, 0, sizeof ops->pad);
986 fill_port_desc(p, &ops->desc);
988 return send_openflow_skb(skb, NULL);
992 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
993 enum ofp_flow_expired_reason reason)
996 struct ofp_flow_expired *ofe;
998 if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1001 ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1005 flow_fill_match(&ofe->match, &flow->key);
1007 ofe->priority = htons(flow->priority);
1008 ofe->reason = reason;
1009 memset(ofe->pad, 0, sizeof ofe->pad);
1011 ofe->duration = htonl((jiffies - flow->init_time) / HZ);
1012 memset(ofe->pad2, 0, sizeof ofe->pad2);
1013 ofe->packet_count = cpu_to_be64(flow->packet_count);
1014 ofe->byte_count = cpu_to_be64(flow->byte_count);
1016 return send_openflow_skb(skb, NULL);
1018 EXPORT_SYMBOL(dp_send_flow_expired);
1021 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1022 uint16_t type, uint16_t code, const void *data, size_t len)
1024 struct sk_buff *skb;
1025 struct ofp_error_msg *oem;
1028 oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1033 oem->type = htons(type);
1034 oem->code = htons(code);
1035 memcpy(oem->data, data, len);
1037 return send_openflow_skb(skb, sender);
1041 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1042 const struct ofp_header *rq)
1044 struct sk_buff *skb;
1045 struct ofp_header *reply;
1047 reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1052 memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1053 return send_openflow_skb(skb, sender);
1056 /* Generic Netlink interface.
1058 * See netlink(7) for an introduction to netlink. See
1059 * http://linux-net.osdl.org/index.php/Netlink for more information and
1060 * pointers on how to work with netlink and Generic Netlink in the kernel and
1063 static struct genl_family dp_genl_family = {
1064 .id = GENL_ID_GENERATE,
1066 .name = DP_GENL_FAMILY_NAME,
1068 .maxattr = DP_GENL_A_MAX,
1071 /* Attribute policy: what each attribute may contain. */
1072 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1073 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1074 [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1075 [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1078 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1080 if (!info->attrs[DP_GENL_A_DP_IDX])
1083 return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1086 static struct genl_ops dp_genl_ops_add_dp = {
1087 .cmd = DP_GENL_C_ADD_DP,
1088 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1089 .policy = dp_genl_policy,
1090 .doit = dp_genl_add,
1094 struct datapath *dp_get(int dp_idx)
1096 if (dp_idx < 0 || dp_idx > DP_MAX)
1098 return rcu_dereference(dps[dp_idx]);
1101 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1103 struct datapath *dp;
1106 if (!info->attrs[DP_GENL_A_DP_IDX])
1109 dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1119 static struct genl_ops dp_genl_ops_del_dp = {
1120 .cmd = DP_GENL_C_DEL_DP,
1121 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1122 .policy = dp_genl_policy,
1123 .doit = dp_genl_del,
1127 /* Queries a datapath for related information. Currently the only relevant
1128 * information is the datapath's multicast group ID. Really we want one
1129 * multicast group per datapath, but because of locking issues[*] we can't
1130 * easily get one. Thus, every datapath will currently return the same
1131 * global multicast group ID, but in the future it would be nice to fix that.
1133 * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1134 * mutex, and genl_register_mc_group, called to acquire a new multicast
1135 * group ID, also acquires genl_lock, thus deadlock.
1137 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1139 struct datapath *dp;
1140 struct sk_buff *ans_skb = NULL;
1144 if (!info->attrs[DP_GENL_A_DP_IDX])
1148 dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1149 dp = dp_get(dp_idx);
1154 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1159 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1160 0, DP_GENL_C_QUERY_DP);
1165 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1166 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1168 genlmsg_end(ans_skb, data);
1169 err = genlmsg_reply(ans_skb, info);
1181 static struct genl_ops dp_genl_ops_query_dp = {
1182 .cmd = DP_GENL_C_QUERY_DP,
1183 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1184 .policy = dp_genl_policy,
1185 .doit = dp_genl_query,
1189 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1191 struct datapath *dp;
1192 struct net_device *port;
1195 if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1199 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1205 /* Get interface to add/remove. */
1206 port = dev_get_by_name(&init_net,
1207 nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1213 /* Execute operation. */
1214 if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1215 err = add_switch_port(dp, port);
1217 if (port->br_port == NULL || port->br_port->dp != dp) {
1221 err = dp_del_switch_port(port->br_port);
1230 static struct genl_ops dp_genl_ops_add_port = {
1231 .cmd = DP_GENL_C_ADD_PORT,
1232 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1233 .policy = dp_genl_policy,
1234 .doit = dp_genl_add_del_port,
1238 static struct genl_ops dp_genl_ops_del_port = {
1239 .cmd = DP_GENL_C_DEL_PORT,
1240 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1241 .policy = dp_genl_policy,
1242 .doit = dp_genl_add_del_port,
1246 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1248 struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1249 struct datapath *dp;
1250 struct ofp_header *oh;
1251 struct sender sender;
1254 if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1257 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1261 if (nla_len(va) < sizeof(struct ofp_header))
1265 sender.xid = oh->xid;
1266 sender.pid = info->snd_pid;
1267 sender.seq = info->snd_seq;
1269 mutex_lock(&dp_mutex);
1270 err = fwd_control_input(dp->chain, &sender,
1271 nla_data(va), nla_len(va));
1272 mutex_unlock(&dp_mutex);
1276 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1277 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1280 static int desc_stats_dump(struct datapath *dp, void *state,
1281 void *body, int *body_len)
1283 struct ofp_desc_stats *ods = body;
1284 int n_bytes = sizeof *ods;
1286 if (n_bytes > *body_len) {
1289 *body_len = n_bytes;
1291 strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1292 strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1293 strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1294 strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1299 struct flow_stats_state {
1301 struct sw_table_position position;
1302 const struct ofp_flow_stats_request *rq;
1305 int bytes_used, bytes_allocated;
1308 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1311 const struct ofp_flow_stats_request *fsr = body;
1312 struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1315 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1316 memset(&s->position, 0, sizeof s->position);
1322 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1324 struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1325 struct flow_stats_state *s = private;
1326 struct ofp_flow_stats *ofs;
1329 length = sizeof *ofs + sf_acts->actions_len;
1330 if (length + s->bytes_used > s->bytes_allocated)
1333 ofs = s->body + s->bytes_used;
1334 ofs->length = htons(length);
1335 ofs->table_id = s->table_idx;
1337 ofs->match.wildcards = htonl(flow->key.wildcards);
1338 ofs->match.in_port = flow->key.in_port;
1339 memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1340 memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1341 ofs->match.dl_vlan = flow->key.dl_vlan;
1342 ofs->match.dl_type = flow->key.dl_type;
1343 ofs->match.nw_src = flow->key.nw_src;
1344 ofs->match.nw_dst = flow->key.nw_dst;
1345 ofs->match.nw_proto = flow->key.nw_proto;
1347 ofs->match.tp_src = flow->key.tp_src;
1348 ofs->match.tp_dst = flow->key.tp_dst;
1349 ofs->duration = htonl((jiffies - flow->init_time) / HZ);
1350 ofs->priority = htons(flow->priority);
1351 ofs->idle_timeout = htons(flow->idle_timeout);
1352 ofs->hard_timeout = htons(flow->hard_timeout);
1353 memset(ofs->pad2, 0, sizeof ofs->pad2);
1354 ofs->packet_count = cpu_to_be64(flow->packet_count);
1355 ofs->byte_count = cpu_to_be64(flow->byte_count);
1356 memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1358 s->bytes_used += length;
1362 static int flow_stats_dump(struct datapath *dp, void *state,
1363 void *body, int *body_len)
1365 struct flow_stats_state *s = state;
1366 struct sw_flow_key match_key;
1370 s->bytes_allocated = *body_len;
1373 flow_extract_match(&match_key, &s->rq->match);
1374 while (s->table_idx < dp->chain->n_tables
1375 && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1377 struct sw_table *table = dp->chain->tables[s->table_idx];
1379 error = table->iterate(table, &match_key, &s->position,
1380 flow_stats_dump_callback, s);
1385 memset(&s->position, 0, sizeof s->position);
1387 *body_len = s->bytes_used;
1389 /* If error is 0, we're done.
1390 * Otherwise, if some bytes were used, there are more flows to come.
1391 * Otherwise, we were not able to fit even a single flow in the body,
1392 * which indicates that we have a single flow with too many actions to
1393 * fit. We won't ever make any progress at that rate, so give up. */
1394 return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1397 static void flow_stats_done(void *state)
1402 static int aggregate_stats_init(struct datapath *dp,
1403 const void *body, int body_len,
1406 *state = (void *)body;
1410 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1412 struct ofp_aggregate_stats_reply *rpy = private;
1413 rpy->packet_count += flow->packet_count;
1414 rpy->byte_count += flow->byte_count;
1419 static int aggregate_stats_dump(struct datapath *dp, void *state,
1420 void *body, int *body_len)
1422 struct ofp_aggregate_stats_request *rq = state;
1423 struct ofp_aggregate_stats_reply *rpy;
1424 struct sw_table_position position;
1425 struct sw_flow_key match_key;
1428 if (*body_len < sizeof *rpy)
1431 *body_len = sizeof *rpy;
1433 memset(rpy, 0, sizeof *rpy);
1435 flow_extract_match(&match_key, &rq->match);
1436 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1437 memset(&position, 0, sizeof position);
1438 while (table_idx < dp->chain->n_tables
1439 && (rq->table_id == 0xff || rq->table_id == table_idx))
1441 struct sw_table *table = dp->chain->tables[table_idx];
1444 error = table->iterate(table, &match_key, &position,
1445 aggregate_stats_dump_callback, rpy);
1450 memset(&position, 0, sizeof position);
1453 rpy->packet_count = cpu_to_be64(rpy->packet_count);
1454 rpy->byte_count = cpu_to_be64(rpy->byte_count);
1455 rpy->flow_count = htonl(rpy->flow_count);
1459 static int table_stats_dump(struct datapath *dp, void *state,
1460 void *body, int *body_len)
1462 struct ofp_table_stats *ots;
1463 int n_bytes = dp->chain->n_tables * sizeof *ots;
1465 if (n_bytes > *body_len)
1467 *body_len = n_bytes;
1468 for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1469 struct sw_table_stats stats;
1470 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1471 strncpy(ots->name, stats.name, sizeof ots->name);
1473 ots->wildcards = htonl(stats.wildcards);
1474 memset(ots->pad, 0, sizeof ots->pad);
1475 ots->max_entries = htonl(stats.max_flows);
1476 ots->active_count = htonl(stats.n_flows);
1477 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1478 ots->matched_count = cpu_to_be64(stats.n_matched);
1483 struct port_stats_state {
1487 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1490 struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1498 static int port_stats_dump(struct datapath *dp, void *state,
1499 void *body, int *body_len)
1501 struct port_stats_state *s = state;
1502 struct ofp_port_stats *ops;
1503 int n_ports, max_ports;
1506 max_ports = *body_len / sizeof *ops;
1512 for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1513 struct net_bridge_port *p = dp->ports[i];
1514 struct net_device_stats *stats;
1517 stats = p->dev->get_stats(p->dev);
1518 ops->port_no = htons(p->port_no);
1519 memset(ops->pad, 0, sizeof ops->pad);
1520 ops->rx_packets = cpu_to_be64(stats->rx_packets);
1521 ops->tx_packets = cpu_to_be64(stats->tx_packets);
1522 ops->rx_bytes = cpu_to_be64(stats->rx_bytes);
1523 ops->tx_bytes = cpu_to_be64(stats->tx_bytes);
1524 ops->rx_dropped = cpu_to_be64(stats->rx_dropped);
1525 ops->tx_dropped = cpu_to_be64(stats->tx_dropped);
1526 ops->rx_errors = cpu_to_be64(stats->rx_errors);
1527 ops->tx_errors = cpu_to_be64(stats->tx_errors);
1528 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1529 ops->rx_over_err = cpu_to_be64(stats->rx_over_errors);
1530 ops->rx_crc_err = cpu_to_be64(stats->rx_crc_errors);
1531 ops->collisions = cpu_to_be64(stats->collisions);
1536 *body_len = n_ports * sizeof *ops;
1537 return n_ports >= max_ports;
1540 static void port_stats_done(void *state)
1546 /* Minimum and maximum acceptable number of bytes in body member of
1547 * struct ofp_stats_request. */
1548 size_t min_body, max_body;
1550 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1551 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1552 * Returns zero if successful, otherwise a negative error code.
1553 * May initialize '*state' to state information. May be null if no
1554 * initialization is required.*/
1555 int (*init)(struct datapath *dp, const void *body, int body_len,
1558 /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1559 * modifies '*body_len' to reflect the number of bytes actually used.
1560 * ('body' will be transmitted as the 'body' member of struct
1561 * ofp_stats_reply.) */
1562 int (*dump)(struct datapath *dp, void *state,
1563 void *body, int *body_len);
1565 /* Cleans any state created by the init or dump functions. May be null
1566 * if no cleanup is required. */
1567 void (*done)(void *state);
1570 static const struct stats_type stats[] = {
1579 sizeof(struct ofp_flow_stats_request),
1580 sizeof(struct ofp_flow_stats_request),
1585 [OFPST_AGGREGATE] = {
1586 sizeof(struct ofp_aggregate_stats_request),
1587 sizeof(struct ofp_aggregate_stats_request),
1588 aggregate_stats_init,
1589 aggregate_stats_dump,
1609 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1611 struct datapath *dp;
1612 struct sender sender;
1613 const struct stats_type *s;
1614 struct ofp_stats_reply *osr;
1616 int max_openflow_len, body_len;
1620 /* Set up the cleanup function for this dump. Linux 2.6.20 and later
1621 * support setting up cleanup functions via the .doneit member of
1622 * struct genl_ops. This kluge supports earlier versions also. */
1623 cb->done = dp_genl_openflow_done;
1625 sender.pid = NETLINK_CB(cb->skb).pid;
1626 sender.seq = cb->nlh->nlmsg_seq;
1628 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1629 struct ofp_stats_request *rq;
1631 size_t len, body_len;
1634 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1635 dp_genl_openflow_policy);
1639 if (!attrs[DP_GENL_A_DP_IDX])
1641 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1642 dp = dp_get(dp_idx);
1646 va = attrs[DP_GENL_A_OPENFLOW];
1648 if (!va || len < sizeof *rq)
1652 sender.xid = rq->header.xid;
1653 type = ntohs(rq->type);
1654 if (rq->header.version != OFP_VERSION) {
1655 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1656 OFPBRC_BAD_VERSION, rq, len);
1659 if (rq->header.type != OFPT_STATS_REQUEST
1660 || ntohs(rq->header.length) != len)
1663 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1664 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1665 OFPBRC_BAD_STAT, rq, len);
1670 body_len = len - offsetof(struct ofp_stats_request, body);
1671 if (body_len < s->min_body || body_len > s->max_body)
1675 cb->args[1] = dp_idx;
1677 cb->args[3] = rq->header.xid;
1680 err = s->init(dp, rq->body, body_len, &state);
1683 cb->args[4] = (long) state;
1685 } else if (cb->args[0] == 1) {
1686 sender.xid = cb->args[3];
1687 dp_idx = cb->args[1];
1688 s = &stats[cb->args[2]];
1690 dp = dp_get(dp_idx);
1697 osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1700 return PTR_ERR(osr);
1701 osr->type = htons(s - stats);
1703 resize_openflow_skb(skb, &osr->header, max_openflow_len);
1705 body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1707 err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1712 osr->flags = ntohs(OFPSF_REPLY_MORE);
1713 resize_openflow_skb(skb, &osr->header,
1714 (offsetof(struct ofp_stats_reply, body)
1723 dp_genl_openflow_done(struct netlink_callback *cb)
1726 const struct stats_type *s = &stats[cb->args[2]];
1728 s->done((void *) cb->args[4]);
1733 static struct genl_ops dp_genl_ops_openflow = {
1734 .cmd = DP_GENL_C_OPENFLOW,
1735 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1736 .policy = dp_genl_openflow_policy,
1737 .doit = dp_genl_openflow,
1738 .dumpit = dp_genl_openflow_dumpit,
1741 static struct genl_ops *dp_genl_all_ops[] = {
1742 /* Keep this operation first. Generic Netlink dispatching
1743 * looks up operations with linear search, so we want it at the
1745 &dp_genl_ops_openflow,
1747 &dp_genl_ops_add_dp,
1748 &dp_genl_ops_del_dp,
1749 &dp_genl_ops_query_dp,
1750 &dp_genl_ops_add_port,
1751 &dp_genl_ops_del_port,
1754 static int dp_init_netlink(void)
1759 err = genl_register_family(&dp_genl_family);
1763 for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1764 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1766 goto err_unregister;
1769 strcpy(mc_group.name, "openflow");
1770 err = genl_register_mc_group(&dp_genl_family, &mc_group);
1772 goto err_unregister;
1777 genl_unregister_family(&dp_genl_family);
1781 static void dp_uninit_netlink(void)
1783 genl_unregister_family(&dp_genl_family);
1786 /* Set the description strings if appropriate values are available from
1788 static void set_desc(void)
1790 const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1791 const char *uptr = uuid + 24;
1793 if (!uuid || *uuid == '\0' || strlen(uuid) != 36)
1796 /* We are only interested version 1 UUIDs, since the last six bytes
1797 * are an IEEE 802 MAC address. */
1798 if (uuid[14] != '1')
1801 /* Only set if the UUID is from Nicira. */
1802 if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1805 strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
1806 snprintf(hw_desc, sizeof(hw_desc), "%s %s",
1807 dmi_get_system_info(DMI_PRODUCT_NAME),
1808 dmi_get_system_info(DMI_PRODUCT_VERSION));
1809 strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL),
1810 sizeof(serial_num));
1813 static int __init dp_init(void)
1817 printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1818 "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1824 err = register_netdevice_notifier(&dp_device_notifier);
1826 goto error_flow_exit;
1828 err = dp_init_netlink();
1830 goto error_unreg_notifier;
1832 /* Check if better descriptions of the switch are available than the
1836 /* Hook into callback used by the bridge to intercept packets.
1837 * Parasites we are. */
1838 if (br_handle_frame_hook)
1839 printk("openflow: hijacking bridge hook\n");
1840 br_handle_frame_hook = dp_frame_hook;
1844 error_unreg_notifier:
1845 unregister_netdevice_notifier(&dp_device_notifier);
1849 printk(KERN_EMERG "openflow: failed to install!");
1853 static void dp_cleanup(void)
1856 dp_uninit_netlink();
1857 unregister_netdevice_notifier(&dp_device_notifier);
1859 br_handle_frame_hook = NULL;
1862 module_init(dp_init);
1863 module_exit(dp_cleanup);
1865 MODULE_DESCRIPTION("OpenFlow switching datapath");
1866 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1867 MODULE_LICENSE("GPL");