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->chain = chain_create(dp);
269 if (dp->chain == NULL)
270 goto err_destroy_dp_dev;
271 INIT_LIST_HEAD(&dp->port_list);
273 dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
274 if (IS_ERR(dp->local_port)) {
275 err = PTR_ERR(dp->local_port);
276 goto err_destroy_local_port;
280 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
282 dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
283 if (IS_ERR(dp->dp_task))
284 goto err_destroy_chain;
290 err_destroy_local_port:
291 dp_del_switch_port(dp->local_port);
293 chain_destroy(dp->chain);
299 module_put(THIS_MODULE);
303 /* Find and return a free port number under 'dp'. */
304 static int find_portno(struct datapath *dp)
307 for (i = 0; i < DP_MAX_PORTS; i++)
308 if (dp->ports[i] == NULL)
313 static struct net_bridge_port *new_nbp(struct datapath *dp,
314 struct net_device *dev, int port_no)
316 struct net_bridge_port *p;
318 if (dev->br_port != NULL)
319 return ERR_PTR(-EBUSY);
321 p = kzalloc(sizeof(*p), GFP_KERNEL);
323 return ERR_PTR(-ENOMEM);
326 dev_set_promiscuity(dev, 1);
331 p->port_no = port_no;
332 spin_lock_init(&p->lock);
333 INIT_WORK(&p->port_task, NULL);
334 if (port_no != OFPP_LOCAL)
335 rcu_assign_pointer(dev->br_port, p);
336 if (port_no < DP_MAX_PORTS)
337 rcu_assign_pointer(dp->ports[port_no], p);
338 list_add_rcu(&p->node, &dp->port_list);
343 int add_switch_port(struct datapath *dp, struct net_device *dev)
345 struct net_bridge_port *p;
348 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
352 port_no = find_portno(dp);
356 p = new_nbp(dp, dev, port_no);
362 /* Notify the ctlpath that this port has been added */
363 dp_send_port_status(p, OFPPR_ADD);
368 /* Delete 'p' from switch. */
369 int dp_del_switch_port(struct net_bridge_port *p)
375 /* First drop references to device. */
376 cancel_work_sync(&p->port_task);
378 dev_set_promiscuity(p->dev, -1);
380 list_del_rcu(&p->node);
381 if (p->port_no != OFPP_LOCAL)
382 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
383 rcu_assign_pointer(p->dev->br_port, NULL);
385 /* Then wait until no one is still using it, and destroy it. */
389 /* Free any SNAT configuration on the port. */
390 spin_lock_irqsave(&p->lock, flags);
392 spin_unlock_irqrestore(&p->lock, flags);
395 /* Notify the ctlpath that this port no longer exists */
396 dp_send_port_status(p, OFPPR_DELETE);
404 static void del_dp(struct datapath *dp)
406 struct net_bridge_port *p, *n;
408 kthread_stop(dp->dp_task);
410 /* Drop references to DP. */
411 list_for_each_entry_safe (p, n, &dp->port_list, node)
412 dp_del_switch_port(p);
413 rcu_assign_pointer(dps[dp->dp_idx], NULL);
415 /* Kill off local_port dev references from buffered packets that have
416 * associated dst entries. */
420 /* Destroy dp->netdev. (Must follow deleting switch ports since
421 * dp->local_port has a reference to it.) */
424 /* Wait until no longer in use, then destroy it. */
426 chain_destroy(dp->chain);
428 module_put(THIS_MODULE);
431 static int dp_maint_func(void *data)
433 struct datapath *dp = (struct datapath *) data;
435 while (!kthread_should_stop()) {
437 struct net_bridge_port *p;
439 /* Expire old SNAT entries */
441 list_for_each_entry_rcu (p, &dp->port_list, node)
446 /* Timeout old entries */
447 chain_timeout(dp->chain);
448 msleep_interruptible(MAINT_SLEEP_MSECS);
455 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
457 /* Make our own copy of the packet. Otherwise we will mangle the
458 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
459 * (No one comes after us, since we tell handle_bridge() that we took
461 skb = skb_share_check(skb, GFP_ATOMIC);
466 /* Check if this packet needs early SNAT processing. */
467 if (snat_pre_route(skb)) {
472 /* Push the Ethernet header back on. */
473 skb_push(skb, ETH_HLEN);
474 skb_reset_mac_header(skb);
475 fwd_port_input(p->dp->chain, skb, p);
479 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
480 * different set of devices!)
482 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
483 /* Called with rcu_read_lock. */
484 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
487 do_port_input(p, skb);
490 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
491 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
493 do_port_input(p, *pskb);
497 /* NB: This has only been tested on 2.4.35 */
498 static void dp_frame_hook(struct sk_buff *skb)
500 struct net_bridge_port *p = skb->dev->br_port;
503 do_port_input(p, skb);
510 /* Forwarding output path.
511 * Based on net/bridge/br_forward.c. */
513 static inline unsigned packet_length(const struct sk_buff *skb)
515 int length = skb->len - ETH_HLEN;
516 if (skb->protocol == htons(ETH_P_8021Q))
521 /* Send packets out all the ports except the originating one. If the
522 * "flood" argument is set, only send along the minimum spanning tree.
525 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
527 u32 disable = flood ? OFPPC_NO_FLOOD : 0;
528 struct net_bridge_port *p;
531 list_for_each_entry_rcu (p, &dp->port_list, node) {
532 if (skb->dev == p->dev || p->config & disable)
534 if (prev_port != -1) {
535 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
540 dp_output_port(dp, clone, prev_port, 0);
542 prev_port = p->port_no;
545 dp_output_port(dp, skb, prev_port, 0);
552 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
553 FIXME: how are devices reference counted? */
554 void dp_set_origin(struct datapath *dp, uint16_t in_port,
557 struct net_bridge_port *p;
558 p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
559 : in_port == OFPP_LOCAL ? dp->local_port
568 dp_xmit_skb(struct sk_buff *skb)
571 if (packet_length(skb) > skb->dev->mtu) {
572 printk("dropped over-mtu packet: %d > %d\n",
573 packet_length(skb), skb->dev->mtu);
583 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
585 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
591 /* Send it out the port it came in on, which is already set in
595 printk("skb device not set forwarding to in_port\n");
599 return dp_xmit_skb(skb);
602 int retval = run_flow_through_tables(dp->chain, skb,
610 return output_all(dp, skb, 1);
613 return output_all(dp, skb, 0);
615 case OFPP_CONTROLLER:
616 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
620 struct net_device *dev = dp->netdev;
624 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
627 case 0 ... DP_MAX_PORTS - 1: {
628 struct net_bridge_port *p = dp->ports[out_port];
631 if (p->dev == skb->dev) {
632 /* To send to the input port, must use OFPP_IN_PORT */
635 printk("can't directly forward to input port\n");
638 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
643 return dp_xmit_skb(skb);
653 printk("can't forward to bad port %d\n", out_port);
657 /* Takes ownership of 'skb' and transmits it to 'dp''s control path. If
658 * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
659 * otherwise, all of 'skb' is sent. 'reason' indicates why 'skb' is being
660 * sent. 'max_len' sets the maximum number of bytes that the caller
661 * wants to be sent; a value of 0 indicates the entire packet should be
664 dp_output_control(struct datapath *dp, struct sk_buff *skb,
665 uint32_t buffer_id, size_t max_len, int reason)
667 /* FIXME? Can we avoid creating a new skbuff in the case where we
668 * forward the whole packet? */
669 struct sk_buff *f_skb;
670 struct ofp_packet_in *opi;
671 size_t fwd_len, opi_len;
675 if ((buffer_id != (uint32_t) -1) && max_len)
676 fwd_len = min(fwd_len, max_len);
678 opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
679 opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
684 opi->buffer_id = htonl(buffer_id);
685 opi->total_len = htons(skb->len);
686 opi->in_port = htons(skb->dev && skb->dev->br_port
687 ? skb->dev->br_port->port_no
689 opi->reason = reason;
691 skb_copy_bits(skb, 0, opi->data, fwd_len);
692 err = send_openflow_skb(f_skb, NULL);
699 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
702 desc->port_no = htons(p->port_no);
703 strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
704 desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
705 memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
708 desc->advertised = 0;
711 spin_lock_irqsave(&p->lock, flags);
712 desc->config = htonl(p->config);
713 desc->state = htonl(p->state);
714 spin_unlock_irqrestore(&p->lock, flags);
716 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
717 if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
718 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
720 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
721 /* Set the supported features */
722 if (ecmd.supported & SUPPORTED_10baseT_Half)
723 desc->supported |= OFPPF_10MB_HD;
724 if (ecmd.supported & SUPPORTED_10baseT_Full)
725 desc->supported |= OFPPF_10MB_FD;
726 if (ecmd.supported & SUPPORTED_100baseT_Half)
727 desc->supported |= OFPPF_100MB_HD;
728 if (ecmd.supported & SUPPORTED_100baseT_Full)
729 desc->supported |= OFPPF_100MB_FD;
730 if (ecmd.supported & SUPPORTED_1000baseT_Half)
731 desc->supported |= OFPPF_1GB_HD;
732 if (ecmd.supported & SUPPORTED_1000baseT_Full)
733 desc->supported |= OFPPF_1GB_FD;
734 if (ecmd.supported & SUPPORTED_10000baseT_Full)
735 desc->supported |= OFPPF_10GB_FD;
736 if (ecmd.supported & SUPPORTED_TP)
737 desc->supported |= OFPPF_COPPER;
738 if (ecmd.supported & SUPPORTED_FIBRE)
739 desc->supported |= OFPPF_FIBER;
740 if (ecmd.supported & SUPPORTED_Autoneg)
741 desc->supported |= OFPPF_AUTONEG;
742 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
743 if (ecmd.supported & SUPPORTED_Pause)
744 desc->supported |= OFPPF_PAUSE;
745 if (ecmd.supported & SUPPORTED_Asym_Pause)
746 desc->supported |= OFPPF_PAUSE_ASYM;
747 #endif /* kernel >= 2.6.14 */
749 /* Set the advertised features */
750 if (ecmd.advertising & ADVERTISED_10baseT_Half)
751 desc->advertised |= OFPPF_10MB_HD;
752 if (ecmd.advertising & ADVERTISED_10baseT_Full)
753 desc->advertised |= OFPPF_10MB_FD;
754 if (ecmd.advertising & ADVERTISED_100baseT_Half)
755 desc->advertised |= OFPPF_100MB_HD;
756 if (ecmd.advertising & ADVERTISED_100baseT_Full)
757 desc->advertised |= OFPPF_100MB_FD;
758 if (ecmd.advertising & ADVERTISED_1000baseT_Half)
759 desc->advertised |= OFPPF_1GB_HD;
760 if (ecmd.advertising & ADVERTISED_1000baseT_Full)
761 desc->advertised |= OFPPF_1GB_FD;
762 if (ecmd.advertising & ADVERTISED_10000baseT_Full)
763 desc->advertised |= OFPPF_10GB_FD;
764 if (ecmd.advertising & ADVERTISED_TP)
765 desc->advertised |= OFPPF_COPPER;
766 if (ecmd.advertising & ADVERTISED_FIBRE)
767 desc->advertised |= OFPPF_FIBER;
768 if (ecmd.advertising & ADVERTISED_Autoneg)
769 desc->advertised |= OFPPF_AUTONEG;
770 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
771 if (ecmd.advertising & ADVERTISED_Pause)
772 desc->advertised |= OFPPF_PAUSE;
773 if (ecmd.advertising & ADVERTISED_Asym_Pause)
774 desc->advertised |= OFPPF_PAUSE_ASYM;
775 #endif /* kernel >= 2.6.14 */
777 /* Set the current features */
778 if (ecmd.speed == SPEED_10)
779 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
780 else if (ecmd.speed == SPEED_100)
781 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
782 else if (ecmd.speed == SPEED_1000)
783 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
784 else if (ecmd.speed == SPEED_10000)
785 desc->curr = OFPPF_10GB_FD;
787 if (ecmd.port == PORT_TP)
788 desc->curr |= OFPPF_COPPER;
789 else if (ecmd.port == PORT_FIBRE)
790 desc->curr |= OFPPF_FIBER;
793 desc->curr |= OFPPF_AUTONEG;
797 desc->curr = htonl(desc->curr);
798 desc->supported = htonl(desc->supported);
799 desc->advertised = htonl(desc->advertised);
800 desc->peer = htonl(desc->peer);
804 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
806 struct net_bridge_port *p;
807 uint64_t dpid = get_datapath_id(dp->netdev);
810 ofr->datapath_id = cpu_to_be64(dpid);
812 ofr->n_buffers = htonl(N_PKT_BUFFERS);
813 ofr->n_tables = dp->chain->n_tables;
814 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
815 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
816 memset(ofr->pad, 0, sizeof ofr->pad);
818 list_for_each_entry_rcu (p, &dp->port_list, node) {
819 fill_port_desc(p, &ofr->ports[port_count]);
827 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
830 struct ofp_switch_features *ofr;
831 size_t ofr_len, port_max_len;
835 port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
836 ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
837 OFPT_FEATURES_REPLY, sender, &skb);
842 port_count = fill_features_reply(dp, ofr);
845 ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
846 resize_openflow_skb(skb, &ofr->header, ofr_len);
847 return send_openflow_skb(skb, sender);
851 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
854 struct ofp_switch_config *osc;
856 osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
861 osc->flags = htons(dp->flags);
862 osc->miss_send_len = htons(dp->miss_send_len);
864 return send_openflow_skb(skb, sender);
868 dp_send_hello(struct datapath *dp, const struct sender *sender,
869 const struct ofp_header *request)
871 if (request->version < OFP_VERSION) {
873 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
874 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
875 OFPHFC_INCOMPATIBLE, err, strlen(err));
879 struct ofp_header *reply;
881 reply = alloc_openflow_skb(dp, sizeof *reply,
882 OFPT_HELLO, sender, &skb);
886 return send_openflow_skb(skb, sender);
890 /* Callback function for a workqueue to disable an interface */
892 down_port_cb(struct work_struct *work)
894 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
898 if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
900 printk("problem bringing up port %s\n", p->dev->name);
902 p->config |= OFPPC_PORT_DOWN;
905 /* Callback function for a workqueue to enable an interface */
907 up_port_cb(struct work_struct *work)
909 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
913 if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
915 printk("problem bringing down port %s\n", p->dev->name);
917 p->config &= ~OFPPC_PORT_DOWN;
921 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
923 unsigned long int flags;
924 int port_no = ntohs(opm->port_no);
925 struct net_bridge_port *p;
926 p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
927 : port_no == OFPP_LOCAL ? dp->local_port
930 /* Make sure the port id hasn't changed since this was sent */
931 if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
934 spin_lock_irqsave(&p->lock, flags);
936 uint32_t config_mask = ntohl(opm->mask);
937 p->config &= ~config_mask;
938 p->config |= ntohl(opm->config) & config_mask;
941 /* Modifying the status of an interface requires taking a lock
942 * that cannot be done from here. For this reason, we use a shared
943 * workqueue, which will cause it to be executed from a safer
945 if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
946 if ((opm->config & htonl(OFPPC_PORT_DOWN))
947 && (p->config & OFPPC_PORT_DOWN) == 0) {
948 PREPARE_WORK(&p->port_task, down_port_cb);
949 schedule_work(&p->port_task);
950 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
951 && (p->config & OFPPC_PORT_DOWN)) {
952 PREPARE_WORK(&p->port_task, up_port_cb);
953 schedule_work(&p->port_task);
956 spin_unlock_irqrestore(&p->lock, flags);
961 /* Initialize the port status field of the bridge port. */
963 init_port_status(struct net_bridge_port *p)
965 unsigned long int flags;
967 spin_lock_irqsave(&p->lock, flags);
969 if (p->dev->flags & IFF_UP)
970 p->config &= ~OFPPC_PORT_DOWN;
972 p->config |= OFPPC_PORT_DOWN;
974 if (netif_carrier_ok(p->dev))
975 p->state &= ~OFPPS_LINK_DOWN;
977 p->state |= OFPPS_LINK_DOWN;
979 spin_unlock_irqrestore(&p->lock, flags);
983 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
986 struct ofp_port_status *ops;
988 ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
992 ops->reason = status;
993 memset(ops->pad, 0, sizeof ops->pad);
994 fill_port_desc(p, &ops->desc);
996 return send_openflow_skb(skb, NULL);
1000 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1001 enum ofp_flow_expired_reason reason)
1003 struct sk_buff *skb;
1004 struct ofp_flow_expired *ofe;
1006 if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1009 ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1013 flow_fill_match(&ofe->match, &flow->key);
1015 ofe->priority = htons(flow->priority);
1016 ofe->reason = reason;
1017 memset(ofe->pad, 0, sizeof ofe->pad);
1019 ofe->duration = htonl((jiffies - flow->init_time) / HZ);
1020 memset(ofe->pad2, 0, sizeof ofe->pad2);
1021 ofe->packet_count = cpu_to_be64(flow->packet_count);
1022 ofe->byte_count = cpu_to_be64(flow->byte_count);
1024 return send_openflow_skb(skb, NULL);
1026 EXPORT_SYMBOL(dp_send_flow_expired);
1029 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1030 uint16_t type, uint16_t code, const void *data, size_t len)
1032 struct sk_buff *skb;
1033 struct ofp_error_msg *oem;
1036 oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1041 oem->type = htons(type);
1042 oem->code = htons(code);
1043 memcpy(oem->data, data, len);
1045 return send_openflow_skb(skb, sender);
1049 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1050 const struct ofp_header *rq)
1052 struct sk_buff *skb;
1053 struct ofp_header *reply;
1055 reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1060 memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1061 return send_openflow_skb(skb, sender);
1064 /* Generic Netlink interface.
1066 * See netlink(7) for an introduction to netlink. See
1067 * http://linux-net.osdl.org/index.php/Netlink for more information and
1068 * pointers on how to work with netlink and Generic Netlink in the kernel and
1071 static struct genl_family dp_genl_family = {
1072 .id = GENL_ID_GENERATE,
1074 .name = DP_GENL_FAMILY_NAME,
1076 .maxattr = DP_GENL_A_MAX,
1079 /* Attribute policy: what each attribute may contain. */
1080 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1081 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1082 [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1083 [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1086 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1088 if (!info->attrs[DP_GENL_A_DP_IDX])
1091 return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1094 static struct genl_ops dp_genl_ops_add_dp = {
1095 .cmd = DP_GENL_C_ADD_DP,
1096 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1097 .policy = dp_genl_policy,
1098 .doit = dp_genl_add,
1102 struct datapath *dp_get(int dp_idx)
1104 if (dp_idx < 0 || dp_idx > DP_MAX)
1106 return rcu_dereference(dps[dp_idx]);
1109 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1111 struct datapath *dp;
1114 if (!info->attrs[DP_GENL_A_DP_IDX])
1117 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1127 static struct genl_ops dp_genl_ops_del_dp = {
1128 .cmd = DP_GENL_C_DEL_DP,
1129 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1130 .policy = dp_genl_policy,
1131 .doit = dp_genl_del,
1135 /* Queries a datapath for related information. Currently the only relevant
1136 * information is the datapath's multicast group ID. Really we want one
1137 * multicast group per datapath, but because of locking issues[*] we can't
1138 * easily get one. Thus, every datapath will currently return the same
1139 * global multicast group ID, but in the future it would be nice to fix that.
1141 * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1142 * mutex, and genl_register_mc_group, called to acquire a new multicast
1143 * group ID, also acquires genl_lock, thus deadlock.
1145 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1147 struct datapath *dp;
1148 struct sk_buff *ans_skb = NULL;
1152 if (!info->attrs[DP_GENL_A_DP_IDX])
1156 dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1157 dp = dp_get(dp_idx);
1162 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1167 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1168 0, DP_GENL_C_QUERY_DP);
1173 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1174 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1176 genlmsg_end(ans_skb, data);
1177 err = genlmsg_reply(ans_skb, info);
1187 static struct genl_ops dp_genl_ops_query_dp = {
1188 .cmd = DP_GENL_C_QUERY_DP,
1189 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1190 .policy = dp_genl_policy,
1191 .doit = dp_genl_query,
1195 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1197 struct datapath *dp;
1198 struct net_device *port;
1201 if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1205 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1211 /* Get interface to add/remove. */
1212 port = dev_get_by_name(&init_net,
1213 nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1219 /* Execute operation. */
1220 if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1221 err = add_switch_port(dp, port);
1223 if (port->br_port == NULL || port->br_port->dp != dp) {
1227 err = dp_del_switch_port(port->br_port);
1236 static struct genl_ops dp_genl_ops_add_port = {
1237 .cmd = DP_GENL_C_ADD_PORT,
1238 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1239 .policy = dp_genl_policy,
1240 .doit = dp_genl_add_del_port,
1244 static struct genl_ops dp_genl_ops_del_port = {
1245 .cmd = DP_GENL_C_DEL_PORT,
1246 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1247 .policy = dp_genl_policy,
1248 .doit = dp_genl_add_del_port,
1252 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1254 struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1255 struct datapath *dp;
1256 struct ofp_header *oh;
1257 struct sender sender;
1260 if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1263 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1267 if (nla_len(va) < sizeof(struct ofp_header))
1271 sender.xid = oh->xid;
1272 sender.pid = info->snd_pid;
1273 sender.seq = info->snd_seq;
1275 mutex_lock(&dp_mutex);
1276 err = fwd_control_input(dp->chain, &sender,
1277 nla_data(va), nla_len(va));
1278 mutex_unlock(&dp_mutex);
1282 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1283 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1286 static int desc_stats_dump(struct datapath *dp, void *state,
1287 void *body, int *body_len)
1289 struct ofp_desc_stats *ods = body;
1290 int n_bytes = sizeof *ods;
1292 if (n_bytes > *body_len) {
1295 *body_len = n_bytes;
1297 strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1298 strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1299 strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1300 strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1305 struct flow_stats_state {
1307 struct sw_table_position position;
1308 const struct ofp_flow_stats_request *rq;
1311 int bytes_used, bytes_allocated;
1314 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1317 const struct ofp_flow_stats_request *fsr = body;
1318 struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1321 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1322 memset(&s->position, 0, sizeof s->position);
1328 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1330 struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1331 struct flow_stats_state *s = private;
1332 struct ofp_flow_stats *ofs;
1335 length = sizeof *ofs + sf_acts->actions_len;
1336 if (length + s->bytes_used > s->bytes_allocated)
1339 ofs = s->body + s->bytes_used;
1340 ofs->length = htons(length);
1341 ofs->table_id = s->table_idx;
1343 ofs->match.wildcards = htonl(flow->key.wildcards);
1344 ofs->match.in_port = flow->key.in_port;
1345 memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1346 memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1347 ofs->match.dl_vlan = flow->key.dl_vlan;
1348 ofs->match.dl_type = flow->key.dl_type;
1349 ofs->match.nw_src = flow->key.nw_src;
1350 ofs->match.nw_dst = flow->key.nw_dst;
1351 ofs->match.nw_proto = flow->key.nw_proto;
1353 ofs->match.tp_src = flow->key.tp_src;
1354 ofs->match.tp_dst = flow->key.tp_dst;
1355 ofs->duration = htonl((jiffies - flow->init_time) / HZ);
1356 ofs->priority = htons(flow->priority);
1357 ofs->idle_timeout = htons(flow->idle_timeout);
1358 ofs->hard_timeout = htons(flow->hard_timeout);
1359 memset(ofs->pad2, 0, sizeof ofs->pad2);
1360 ofs->packet_count = cpu_to_be64(flow->packet_count);
1361 ofs->byte_count = cpu_to_be64(flow->byte_count);
1362 memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1364 s->bytes_used += length;
1368 static int flow_stats_dump(struct datapath *dp, void *state,
1369 void *body, int *body_len)
1371 struct flow_stats_state *s = state;
1372 struct sw_flow_key match_key;
1376 s->bytes_allocated = *body_len;
1379 flow_extract_match(&match_key, &s->rq->match);
1380 while (s->table_idx < dp->chain->n_tables
1381 && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1383 struct sw_table *table = dp->chain->tables[s->table_idx];
1385 error = table->iterate(table, &match_key, s->rq->out_port,
1386 &s->position, flow_stats_dump_callback, s);
1391 memset(&s->position, 0, sizeof s->position);
1393 *body_len = s->bytes_used;
1395 /* If error is 0, we're done.
1396 * Otherwise, if some bytes were used, there are more flows to come.
1397 * Otherwise, we were not able to fit even a single flow in the body,
1398 * which indicates that we have a single flow with too many actions to
1399 * fit. We won't ever make any progress at that rate, so give up. */
1400 return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1403 static void flow_stats_done(void *state)
1408 static int aggregate_stats_init(struct datapath *dp,
1409 const void *body, int body_len,
1412 *state = (void *)body;
1416 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1418 struct ofp_aggregate_stats_reply *rpy = private;
1419 rpy->packet_count += flow->packet_count;
1420 rpy->byte_count += flow->byte_count;
1425 static int aggregate_stats_dump(struct datapath *dp, void *state,
1426 void *body, int *body_len)
1428 struct ofp_aggregate_stats_request *rq = state;
1429 struct ofp_aggregate_stats_reply *rpy;
1430 struct sw_table_position position;
1431 struct sw_flow_key match_key;
1434 if (*body_len < sizeof *rpy)
1437 *body_len = sizeof *rpy;
1439 memset(rpy, 0, sizeof *rpy);
1441 flow_extract_match(&match_key, &rq->match);
1442 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1443 memset(&position, 0, sizeof position);
1444 while (table_idx < dp->chain->n_tables
1445 && (rq->table_id == 0xff || rq->table_id == table_idx))
1447 struct sw_table *table = dp->chain->tables[table_idx];
1450 error = table->iterate(table, &match_key, rq->out_port, &position,
1451 aggregate_stats_dump_callback, rpy);
1456 memset(&position, 0, sizeof position);
1459 rpy->packet_count = cpu_to_be64(rpy->packet_count);
1460 rpy->byte_count = cpu_to_be64(rpy->byte_count);
1461 rpy->flow_count = htonl(rpy->flow_count);
1465 static int table_stats_dump(struct datapath *dp, void *state,
1466 void *body, int *body_len)
1468 struct ofp_table_stats *ots;
1469 int n_bytes = dp->chain->n_tables * sizeof *ots;
1471 if (n_bytes > *body_len)
1473 *body_len = n_bytes;
1474 for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1475 struct sw_table_stats stats;
1476 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1477 strncpy(ots->name, stats.name, sizeof ots->name);
1479 ots->wildcards = htonl(stats.wildcards);
1480 memset(ots->pad, 0, sizeof ots->pad);
1481 ots->max_entries = htonl(stats.max_flows);
1482 ots->active_count = htonl(stats.n_flows);
1483 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1484 ots->matched_count = cpu_to_be64(stats.n_matched);
1489 struct port_stats_state {
1493 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1496 struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1504 static int port_stats_dump(struct datapath *dp, void *state,
1505 void *body, int *body_len)
1507 struct port_stats_state *s = state;
1508 struct ofp_port_stats *ops;
1509 int n_ports, max_ports;
1512 max_ports = *body_len / sizeof *ops;
1518 for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1519 struct net_bridge_port *p = dp->ports[i];
1520 struct net_device_stats *stats;
1523 stats = p->dev->get_stats(p->dev);
1524 ops->port_no = htons(p->port_no);
1525 memset(ops->pad, 0, sizeof ops->pad);
1526 ops->rx_packets = cpu_to_be64(stats->rx_packets);
1527 ops->tx_packets = cpu_to_be64(stats->tx_packets);
1528 ops->rx_bytes = cpu_to_be64(stats->rx_bytes);
1529 ops->tx_bytes = cpu_to_be64(stats->tx_bytes);
1530 ops->rx_dropped = cpu_to_be64(stats->rx_dropped);
1531 ops->tx_dropped = cpu_to_be64(stats->tx_dropped);
1532 ops->rx_errors = cpu_to_be64(stats->rx_errors);
1533 ops->tx_errors = cpu_to_be64(stats->tx_errors);
1534 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1535 ops->rx_over_err = cpu_to_be64(stats->rx_over_errors);
1536 ops->rx_crc_err = cpu_to_be64(stats->rx_crc_errors);
1537 ops->collisions = cpu_to_be64(stats->collisions);
1542 *body_len = n_ports * sizeof *ops;
1543 return n_ports >= max_ports;
1546 static void port_stats_done(void *state)
1552 /* Minimum and maximum acceptable number of bytes in body member of
1553 * struct ofp_stats_request. */
1554 size_t min_body, max_body;
1556 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1557 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1558 * Returns zero if successful, otherwise a negative error code.
1559 * May initialize '*state' to state information. May be null if no
1560 * initialization is required.*/
1561 int (*init)(struct datapath *dp, const void *body, int body_len,
1564 /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1565 * modifies '*body_len' to reflect the number of bytes actually used.
1566 * ('body' will be transmitted as the 'body' member of struct
1567 * ofp_stats_reply.) */
1568 int (*dump)(struct datapath *dp, void *state,
1569 void *body, int *body_len);
1571 /* Cleans any state created by the init or dump functions. May be null
1572 * if no cleanup is required. */
1573 void (*done)(void *state);
1576 static const struct stats_type stats[] = {
1585 sizeof(struct ofp_flow_stats_request),
1586 sizeof(struct ofp_flow_stats_request),
1591 [OFPST_AGGREGATE] = {
1592 sizeof(struct ofp_aggregate_stats_request),
1593 sizeof(struct ofp_aggregate_stats_request),
1594 aggregate_stats_init,
1595 aggregate_stats_dump,
1615 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1617 struct datapath *dp;
1618 struct sender sender;
1619 const struct stats_type *s;
1620 struct ofp_stats_reply *osr;
1622 int max_openflow_len, body_len;
1626 /* Set up the cleanup function for this dump. Linux 2.6.20 and later
1627 * support setting up cleanup functions via the .doneit member of
1628 * struct genl_ops. This kluge supports earlier versions also. */
1629 cb->done = dp_genl_openflow_done;
1631 sender.pid = NETLINK_CB(cb->skb).pid;
1632 sender.seq = cb->nlh->nlmsg_seq;
1634 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1635 struct ofp_stats_request *rq;
1637 size_t len, body_len;
1640 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1641 dp_genl_openflow_policy);
1645 if (!attrs[DP_GENL_A_DP_IDX])
1647 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1648 dp = dp_get(dp_idx);
1652 va = attrs[DP_GENL_A_OPENFLOW];
1654 if (!va || len < sizeof *rq)
1658 sender.xid = rq->header.xid;
1659 type = ntohs(rq->type);
1660 if (rq->header.version != OFP_VERSION) {
1661 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1662 OFPBRC_BAD_VERSION, rq, len);
1665 if (rq->header.type != OFPT_STATS_REQUEST
1666 || ntohs(rq->header.length) != len)
1669 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1670 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1671 OFPBRC_BAD_STAT, rq, len);
1676 body_len = len - offsetof(struct ofp_stats_request, body);
1677 if (body_len < s->min_body || body_len > s->max_body)
1681 cb->args[1] = dp_idx;
1683 cb->args[3] = rq->header.xid;
1686 err = s->init(dp, rq->body, body_len, &state);
1689 cb->args[4] = (long) state;
1691 } else if (cb->args[0] == 1) {
1692 sender.xid = cb->args[3];
1693 dp_idx = cb->args[1];
1694 s = &stats[cb->args[2]];
1696 dp = dp_get(dp_idx);
1703 osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1706 return PTR_ERR(osr);
1707 osr->type = htons(s - stats);
1709 resize_openflow_skb(skb, &osr->header, max_openflow_len);
1711 body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1713 err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1718 osr->flags = ntohs(OFPSF_REPLY_MORE);
1719 resize_openflow_skb(skb, &osr->header,
1720 (offsetof(struct ofp_stats_reply, body)
1729 dp_genl_openflow_done(struct netlink_callback *cb)
1732 const struct stats_type *s = &stats[cb->args[2]];
1734 s->done((void *) cb->args[4]);
1739 static struct genl_ops dp_genl_ops_openflow = {
1740 .cmd = DP_GENL_C_OPENFLOW,
1741 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1742 .policy = dp_genl_openflow_policy,
1743 .doit = dp_genl_openflow,
1744 .dumpit = dp_genl_openflow_dumpit,
1747 static struct genl_ops *dp_genl_all_ops[] = {
1748 /* Keep this operation first. Generic Netlink dispatching
1749 * looks up operations with linear search, so we want it at the
1751 &dp_genl_ops_openflow,
1753 &dp_genl_ops_add_dp,
1754 &dp_genl_ops_del_dp,
1755 &dp_genl_ops_query_dp,
1756 &dp_genl_ops_add_port,
1757 &dp_genl_ops_del_port,
1760 static int dp_init_netlink(void)
1765 err = genl_register_family(&dp_genl_family);
1769 for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1770 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1772 goto err_unregister;
1775 strcpy(mc_group.name, "openflow");
1776 err = genl_register_mc_group(&dp_genl_family, &mc_group);
1778 goto err_unregister;
1783 genl_unregister_family(&dp_genl_family);
1787 static void dp_uninit_netlink(void)
1789 genl_unregister_family(&dp_genl_family);
1792 /* Set the description strings if appropriate values are available from
1794 static void set_desc(void)
1796 const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1797 const char *uptr = uuid + 24;
1799 if (!uuid || *uuid == '\0' || strlen(uuid) != 36)
1802 /* We are only interested version 1 UUIDs, since the last six bytes
1803 * are an IEEE 802 MAC address. */
1804 if (uuid[14] != '1')
1807 /* Only set if the UUID is from Nicira. */
1808 if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1811 strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
1812 snprintf(hw_desc, sizeof(hw_desc), "%s %s",
1813 dmi_get_system_info(DMI_PRODUCT_NAME),
1814 dmi_get_system_info(DMI_PRODUCT_VERSION));
1815 strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL),
1816 sizeof(serial_num));
1819 static int __init dp_init(void)
1823 printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1824 "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1830 err = register_netdevice_notifier(&dp_device_notifier);
1832 goto error_flow_exit;
1834 err = dp_init_netlink();
1836 goto error_unreg_notifier;
1838 /* Check if better descriptions of the switch are available than the
1842 /* Hook into callback used by the bridge to intercept packets.
1843 * Parasites we are. */
1844 if (br_handle_frame_hook)
1845 printk("openflow: hijacking bridge hook\n");
1846 br_handle_frame_hook = dp_frame_hook;
1850 error_unreg_notifier:
1851 unregister_netdevice_notifier(&dp_device_notifier);
1855 printk(KERN_EMERG "openflow: failed to install!");
1859 static void dp_cleanup(void)
1862 dp_uninit_netlink();
1863 unregister_netdevice_notifier(&dp_device_notifier);
1865 br_handle_frame_hook = NULL;
1868 module_init(dp_init);
1869 module_exit(dp_cleanup);
1871 MODULE_DESCRIPTION("OpenFlow switching datapath");
1872 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1873 MODULE_LICENSE("GPL");