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>
35 #include "openflow-netlink.h"
37 #include "nx_act_snat.h"
47 /* Strings to describe the manufacturer, hardware, and software. This data
48 * is queriable through the switch description stats message. */
49 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks";
50 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
51 static char sw_desc[DESC_STR_LEN] = VERSION;
52 static char serial_num[SERIAL_NUM_LEN] = "None";
54 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
55 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
56 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
57 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
58 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
60 MODULE_PARM(mfr_desc, "s");
61 MODULE_PARM(hw_desc, "s");
62 MODULE_PARM(sw_desc, "s");
63 MODULE_PARM(serial_num, "s");
67 /* Number of milliseconds between runs of the maintenance thread. */
68 #define MAINT_SLEEP_MSECS 1000
70 #define UINT32_MAX 4294967295U
71 #define UINT16_MAX 65535
72 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
74 static struct genl_family dp_genl_family;
75 static struct genl_multicast_group mc_group;
77 /* It's hard to imagine wanting more than one datapath, but... */
80 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
81 * by dp_mutex. dp_mutex is almost completely redundant with genl_mutex
82 * maintained by the Generic Netlink code, but the timeout path needs mutual
85 * It is safe to access the datapath and net_bridge_port structures with just
88 static struct datapath *dps[DP_MAX];
89 DEFINE_MUTEX(dp_mutex);
90 EXPORT_SYMBOL(dp_mutex);
92 static int dp_maint_func(void *data);
93 static void init_port_status(struct net_bridge_port *p);
94 static int dp_genl_openflow_done(struct netlink_callback *);
95 static struct net_bridge_port *new_nbp(struct datapath *,
96 struct net_device *, int port_no);
98 /* nla_shrink - reduce amount of space reserved by nla_reserve
99 * @skb: socket buffer from which to recover room
100 * @nla: netlink attribute to adjust
101 * @len: new length of attribute payload
103 * Reduces amount of space reserved by a call to nla_reserve.
105 * No other attributes may be added between calling nla_reserve and this
106 * function, since it will create a hole in the message.
108 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
110 int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
114 nla->nla_len = nla_attr_size(len);
117 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
118 * If 'sender' is nonnull, then it is used as the message's destination. 'dp'
119 * must specify the datapath to use.
121 * '*max_openflow_len' receives the maximum number of bytes that are available
122 * for the embedded OpenFlow message. The caller must call
123 * resize_openflow_skb() to set the actual size of the message to this number
126 * Returns the openflow header if successful, otherwise (if 'skb' is too small)
129 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
130 const struct sender *sender, int *max_openflow_len)
132 struct ofp_header *oh;
136 /* Assemble the Generic Netlink wrapper. */
137 if (!genlmsg_put(skb,
138 sender ? sender->pid : 0,
139 sender ? sender->seq : 0,
140 &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
141 return ERR_PTR(-ENOBUFS);
142 if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
143 return ERR_PTR(-ENOBUFS);
144 openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
145 if (openflow_len < sizeof *oh)
146 return ERR_PTR(-ENOBUFS);
147 *max_openflow_len = openflow_len;
148 attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
151 /* Fill in the header. The caller is responsible for the length. */
153 oh->version = OFP_VERSION;
155 oh->xid = sender ? sender->xid : 0;
160 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
161 * length 'new_length' (in bytes), adjusting pointers and size values as
164 resize_openflow_skb(struct sk_buff *skb,
165 struct ofp_header *oh, size_t new_length)
167 struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
168 nla_shrink(skb, attr, new_length);
169 oh->length = htons(new_length);
170 nlmsg_end(skb, (struct nlmsghdr *) skb->data);
173 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
174 * length. Returns a null pointer if memory is unavailable, otherwise returns
175 * the OpenFlow header and stores a pointer to the skb in '*pskb'.
177 * 'type' is the OpenFlow message type. If 'sender' is nonnull, then it is
178 * used as the message's destination. 'dp' must specify the datapath to
181 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
182 const struct sender *sender, struct sk_buff **pskb)
184 struct ofp_header *oh;
187 int max_openflow_len;
189 if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
191 printk("alloc_openflow_skb: openflow message too large: %zu\n",
196 genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
197 genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
198 genl_len += nla_total_size(openflow_len); /* DP_GENL_A_OPENFLOW */
199 skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
202 printk("alloc_openflow_skb: genlmsg_new failed\n");
206 oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
207 BUG_ON(!oh || IS_ERR(oh));
208 resize_openflow_skb(skb, oh, openflow_len);
213 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
216 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
219 ? genlmsg_unicast(skb, sender->pid)
220 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
223 /* Generates a unique datapath id. It incorporates the datapath index
224 * and a hardware address, if available. If not, it generates a random
228 uint64_t gen_datapath_id(uint16_t dp_idx)
232 struct net_device *dev;
234 /* The top 16 bits are used to identify the datapath. The lower 48 bits
235 * use an interface address. */
236 id = (uint64_t)dp_idx << 48;
237 if ((dev = dev_get_by_name(&init_net, "ctl0"))
238 || (dev = dev_get_by_name(&init_net, "eth0"))) {
239 for (i=0; i<ETH_ALEN; i++) {
240 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
244 /* Randomly choose the lower 48 bits if we cannot find an
245 * address and mark the most significant bit to indicate that
246 * this was randomly generated. */
247 uint8_t rand[ETH_ALEN];
248 get_random_bytes(rand, ETH_ALEN);
249 id |= (uint64_t)1 << 63;
250 for (i=0; i<ETH_ALEN; i++) {
251 id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
258 /* Creates a new datapath numbered 'dp_idx'. Returns 0 for success or a
259 * negative error code. */
260 static int new_dp(int dp_idx)
265 if (dp_idx < 0 || dp_idx >= DP_MAX)
268 if (!try_module_get(THIS_MODULE))
271 /* Exit early if a datapath with that number already exists. */
278 dp = kzalloc(sizeof *dp, GFP_KERNEL);
282 /* Setup our "of" device */
283 err = dp_dev_setup(dp);
288 dp->id = gen_datapath_id(dp_idx);
289 dp->chain = chain_create(dp);
290 if (dp->chain == NULL)
291 goto err_destroy_dp_dev;
292 INIT_LIST_HEAD(&dp->port_list);
294 dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
295 if (IS_ERR(dp->local_port)) {
296 err = PTR_ERR(dp->local_port);
297 goto err_destroy_local_port;
301 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
303 dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
304 if (IS_ERR(dp->dp_task))
305 goto err_destroy_chain;
311 err_destroy_local_port:
312 dp_del_switch_port(dp->local_port);
314 chain_destroy(dp->chain);
320 module_put(THIS_MODULE);
324 /* Find and return a free port number under 'dp'. */
325 static int find_portno(struct datapath *dp)
328 for (i = 0; i < OFPP_MAX; i++)
329 if (dp->ports[i] == NULL)
334 static struct net_bridge_port *new_nbp(struct datapath *dp,
335 struct net_device *dev, int port_no)
337 struct net_bridge_port *p;
339 if (dev->br_port != NULL)
340 return ERR_PTR(-EBUSY);
342 p = kzalloc(sizeof(*p), GFP_KERNEL);
344 return ERR_PTR(-ENOMEM);
347 dev_set_promiscuity(dev, 1);
352 p->port_no = port_no;
353 spin_lock_init(&p->lock);
354 INIT_WORK(&p->port_task, NULL);
355 if (port_no != OFPP_LOCAL)
356 rcu_assign_pointer(dev->br_port, p);
357 if (port_no < OFPP_MAX)
358 rcu_assign_pointer(dp->ports[port_no], p);
359 list_add_rcu(&p->node, &dp->port_list);
364 int add_switch_port(struct datapath *dp, struct net_device *dev)
366 struct net_bridge_port *p;
369 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
373 port_no = find_portno(dp);
377 p = new_nbp(dp, dev, port_no);
383 /* Notify the ctlpath that this port has been added */
384 dp_send_port_status(p, OFPPR_ADD);
389 /* Delete 'p' from switch. */
390 int dp_del_switch_port(struct net_bridge_port *p)
396 /* First drop references to device. */
397 cancel_work_sync(&p->port_task);
399 dev_set_promiscuity(p->dev, -1);
401 list_del_rcu(&p->node);
402 if (p->port_no != OFPP_LOCAL)
403 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
404 rcu_assign_pointer(p->dev->br_port, NULL);
406 /* Then wait until no one is still using it, and destroy it. */
410 /* Free any SNAT configuration on the port. */
411 spin_lock_irqsave(&p->lock, flags);
413 spin_unlock_irqrestore(&p->lock, flags);
416 /* Notify the ctlpath that this port no longer exists */
417 dp_send_port_status(p, OFPPR_DELETE);
425 static void del_dp(struct datapath *dp)
427 struct net_bridge_port *p, *n;
429 kthread_stop(dp->dp_task);
431 /* Drop references to DP. */
432 list_for_each_entry_safe (p, n, &dp->port_list, node)
433 dp_del_switch_port(p);
434 rcu_assign_pointer(dps[dp->dp_idx], NULL);
436 /* Kill off local_port dev references from buffered packets that have
437 * associated dst entries. */
441 /* Destroy dp->netdev. (Must follow deleting switch ports since
442 * dp->local_port has a reference to it.) */
445 /* Wait until no longer in use, then destroy it. */
447 chain_destroy(dp->chain);
449 module_put(THIS_MODULE);
452 static int dp_maint_func(void *data)
454 struct datapath *dp = (struct datapath *) data;
456 while (!kthread_should_stop()) {
458 struct net_bridge_port *p;
460 /* Expire old SNAT entries */
462 list_for_each_entry_rcu (p, &dp->port_list, node)
467 /* Timeout old entries */
468 chain_timeout(dp->chain);
469 msleep_interruptible(MAINT_SLEEP_MSECS);
476 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
479 /* Check if this packet needs early SNAT processing. */
480 if (snat_pre_route(skb)) {
486 /* Push the Ethernet header back on. */
487 skb_push(skb, ETH_HLEN);
488 fwd_port_input(p->dp->chain, skb, p);
492 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
493 * different set of devices!)
495 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
496 /* Called with rcu_read_lock. */
497 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
500 do_port_input(p, skb);
503 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
504 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
506 do_port_input(p, *pskb);
510 /* NB: This has only been tested on 2.4.35 */
511 static void dp_frame_hook(struct sk_buff *skb)
513 struct net_bridge_port *p = skb->dev->br_port;
516 do_port_input(p, skb);
523 /* Forwarding output path.
524 * Based on net/bridge/br_forward.c. */
526 static inline unsigned packet_length(const struct sk_buff *skb)
528 int length = skb->len - ETH_HLEN;
529 if (skb->protocol == htons(ETH_P_8021Q))
534 /* Send packets out all the ports except the originating one. If the
535 * "flood" argument is set, only send along the minimum spanning tree.
538 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
540 u32 disable = flood ? OFPPC_NO_FLOOD : 0;
541 struct net_bridge_port *p;
544 list_for_each_entry_rcu (p, &dp->port_list, node) {
545 if (skb->dev == p->dev || p->config & disable)
547 if (prev_port != -1) {
548 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
553 dp_output_port(dp, clone, prev_port, 0);
555 prev_port = p->port_no;
558 dp_output_port(dp, skb, prev_port, 0);
565 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
566 FIXME: how are devices reference counted? */
567 void dp_set_origin(struct datapath *dp, uint16_t in_port,
570 struct net_bridge_port *p = (in_port < OFPP_MAX ? dp->ports[in_port]
571 : in_port == OFPP_LOCAL ? dp->local_port
580 dp_xmit_skb(struct sk_buff *skb)
583 if (packet_length(skb) > skb->dev->mtu) {
584 printk("dropped over-mtu packet: %d > %d\n",
585 packet_length(skb), skb->dev->mtu);
595 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
597 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
603 /* Send it out the port it came in on, which is already set in
607 printk("skb device not set forwarding to in_port\n");
611 return dp_xmit_skb(skb);
614 int retval = run_flow_through_tables(dp->chain, skb,
622 return output_all(dp, skb, 1);
625 return output_all(dp, skb, 0);
627 case OFPP_CONTROLLER:
628 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
632 struct net_device *dev = dp->netdev;
633 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
636 case 0 ... OFPP_MAX-1: {
637 struct net_bridge_port *p = dp->ports[out_port];
640 if (p->dev == skb->dev) {
641 /* To send to the input port, must use OFPP_IN_PORT */
644 printk("can't directly forward to input port\n");
647 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
652 return dp_xmit_skb(skb);
662 printk("can't forward to bad port %d\n", out_port);
666 /* Takes ownership of 'skb' and transmits it to 'dp''s control path. If
667 * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
668 * otherwise, all of 'skb' is sent. 'reason' indicates why 'skb' is being
669 * sent. 'max_len' sets the maximum number of bytes that the caller
670 * wants to be sent; a value of 0 indicates the entire packet should be
673 dp_output_control(struct datapath *dp, struct sk_buff *skb,
674 uint32_t buffer_id, size_t max_len, int reason)
676 /* FIXME? Can we avoid creating a new skbuff in the case where we
677 * forward the whole packet? */
678 struct sk_buff *f_skb;
679 struct ofp_packet_in *opi;
680 struct net_bridge_port *p;
681 size_t fwd_len, opi_len;
685 if ((buffer_id != (uint32_t) -1) && max_len)
686 fwd_len = min(fwd_len, max_len);
688 opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
689 opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
694 opi->buffer_id = htonl(buffer_id);
695 opi->total_len = htons(skb->len);
696 p = skb->dev->br_port;
697 opi->in_port = htons(p ? p->port_no : OFPP_LOCAL);
698 opi->reason = reason;
700 memcpy(opi->data, skb_mac_header(skb), fwd_len);
701 err = send_openflow_skb(f_skb, NULL);
708 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
711 desc->port_no = htons(p->port_no);
712 strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
713 desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
714 memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
717 desc->advertised = 0;
720 spin_lock_irqsave(&p->lock, flags);
721 desc->config = htonl(p->config);
722 desc->state = htonl(p->state);
723 spin_unlock_irqrestore(&p->lock, flags);
725 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
726 if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
727 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
729 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
730 /* Set the supported features */
731 if (ecmd.supported & SUPPORTED_10baseT_Half)
732 desc->supported |= OFPPF_10MB_HD;
733 if (ecmd.supported & SUPPORTED_10baseT_Full)
734 desc->supported |= OFPPF_10MB_FD;
735 if (ecmd.supported & SUPPORTED_100baseT_Half)
736 desc->supported |= OFPPF_100MB_HD;
737 if (ecmd.supported & SUPPORTED_100baseT_Full)
738 desc->supported |= OFPPF_100MB_FD;
739 if (ecmd.supported & SUPPORTED_1000baseT_Half)
740 desc->supported |= OFPPF_1GB_HD;
741 if (ecmd.supported & SUPPORTED_1000baseT_Full)
742 desc->supported |= OFPPF_1GB_FD;
743 if (ecmd.supported & SUPPORTED_10000baseT_Full)
744 desc->supported |= OFPPF_10GB_FD;
745 if (ecmd.supported & SUPPORTED_TP)
746 desc->supported |= OFPPF_COPPER;
747 if (ecmd.supported & SUPPORTED_FIBRE)
748 desc->supported |= OFPPF_FIBER;
749 if (ecmd.supported & SUPPORTED_Autoneg)
750 desc->supported |= OFPPF_AUTONEG;
751 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
752 if (ecmd.supported & SUPPORTED_Pause)
753 desc->supported |= OFPPF_PAUSE;
754 if (ecmd.supported & SUPPORTED_Asym_Pause)
755 desc->supported |= OFPPF_PAUSE_ASYM;
756 #endif /* kernel >= 2.6.14 */
758 /* Set the advertised features */
759 if (ecmd.advertising & ADVERTISED_10baseT_Half)
760 desc->advertised |= OFPPF_10MB_HD;
761 if (ecmd.advertising & ADVERTISED_10baseT_Full)
762 desc->advertised |= OFPPF_10MB_FD;
763 if (ecmd.advertising & ADVERTISED_100baseT_Half)
764 desc->advertised |= OFPPF_100MB_HD;
765 if (ecmd.advertising & ADVERTISED_100baseT_Full)
766 desc->advertised |= OFPPF_100MB_FD;
767 if (ecmd.advertising & ADVERTISED_1000baseT_Half)
768 desc->advertised |= OFPPF_1GB_HD;
769 if (ecmd.advertising & ADVERTISED_1000baseT_Full)
770 desc->advertised |= OFPPF_1GB_FD;
771 if (ecmd.advertising & ADVERTISED_10000baseT_Full)
772 desc->advertised |= OFPPF_10GB_FD;
773 if (ecmd.advertising & ADVERTISED_TP)
774 desc->advertised |= OFPPF_COPPER;
775 if (ecmd.advertising & ADVERTISED_FIBRE)
776 desc->advertised |= OFPPF_FIBER;
777 if (ecmd.advertising & ADVERTISED_Autoneg)
778 desc->advertised |= OFPPF_AUTONEG;
779 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
780 if (ecmd.advertising & ADVERTISED_Pause)
781 desc->advertised |= OFPPF_PAUSE;
782 if (ecmd.advertising & ADVERTISED_Asym_Pause)
783 desc->advertised |= OFPPF_PAUSE_ASYM;
784 #endif /* kernel >= 2.6.14 */
786 /* Set the current features */
787 if (ecmd.speed == SPEED_10)
788 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
789 else if (ecmd.speed == SPEED_100)
790 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
791 else if (ecmd.speed == SPEED_1000)
792 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
793 else if (ecmd.speed == SPEED_10000)
794 desc->curr = OFPPF_10GB_FD;
796 if (ecmd.port == PORT_TP)
797 desc->curr |= OFPPF_COPPER;
798 else if (ecmd.port == PORT_FIBRE)
799 desc->curr |= OFPPF_FIBER;
802 desc->curr |= OFPPF_AUTONEG;
806 desc->curr = htonl(desc->curr);
807 desc->supported = htonl(desc->supported);
808 desc->advertised = htonl(desc->advertised);
809 desc->peer = htonl(desc->peer);
813 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
815 struct net_bridge_port *p;
818 ofr->datapath_id = cpu_to_be64(dp->id);
820 ofr->n_buffers = htonl(N_PKT_BUFFERS);
821 ofr->n_tables = dp->chain->n_tables;
822 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
823 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
824 memset(ofr->pad, 0, sizeof ofr->pad);
826 list_for_each_entry_rcu (p, &dp->port_list, node) {
827 fill_port_desc(p, &ofr->ports[port_count]);
835 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
838 struct ofp_switch_features *ofr;
839 size_t ofr_len, port_max_len;
843 port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
844 ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
845 OFPT_FEATURES_REPLY, sender, &skb);
850 port_count = fill_features_reply(dp, ofr);
853 ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
854 resize_openflow_skb(skb, &ofr->header, ofr_len);
855 return send_openflow_skb(skb, sender);
859 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
862 struct ofp_switch_config *osc;
864 osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
869 osc->flags = htons(dp->flags);
870 osc->miss_send_len = htons(dp->miss_send_len);
872 return send_openflow_skb(skb, sender);
876 dp_send_hello(struct datapath *dp, const struct sender *sender,
877 const struct ofp_header *request)
879 if (request->version < OFP_VERSION) {
881 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
882 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
883 OFPHFC_INCOMPATIBLE, err, strlen(err));
887 struct ofp_header *reply;
889 reply = alloc_openflow_skb(dp, sizeof *reply,
890 OFPT_HELLO, sender, &skb);
894 return send_openflow_skb(skb, sender);
898 /* Callback function for a workqueue to disable an interface */
900 down_port_cb(struct work_struct *work)
902 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
906 if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
908 printk("problem bringing up port %s\n", p->dev->name);
910 p->config |= OFPPC_PORT_DOWN;
913 /* Callback function for a workqueue to enable an interface */
915 up_port_cb(struct work_struct *work)
917 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
921 if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
923 printk("problem bringing down port %s\n", p->dev->name);
925 p->config &= ~OFPPC_PORT_DOWN;
929 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
931 unsigned long int flags;
932 int port_no = ntohs(opm->port_no);
933 struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no]
934 : port_no == OFPP_LOCAL ? dp->local_port
937 /* Make sure the port id hasn't changed since this was sent */
938 if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
941 spin_lock_irqsave(&p->lock, flags);
943 uint32_t config_mask = ntohl(opm->mask);
944 p->config &= ~config_mask;
945 p->config |= ntohl(opm->config) & config_mask;
948 /* Modifying the status of an interface requires taking a lock
949 * that cannot be done from here. For this reason, we use a shared
950 * workqueue, which will cause it to be executed from a safer
952 if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
953 if ((opm->config & htonl(OFPPC_PORT_DOWN))
954 && (p->config & OFPPC_PORT_DOWN) == 0) {
955 PREPARE_WORK(&p->port_task, down_port_cb);
956 schedule_work(&p->port_task);
957 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
958 && (p->config & OFPPC_PORT_DOWN)) {
959 PREPARE_WORK(&p->port_task, up_port_cb);
960 schedule_work(&p->port_task);
963 spin_unlock_irqrestore(&p->lock, flags);
968 /* Initialize the port status field of the bridge port. */
970 init_port_status(struct net_bridge_port *p)
972 unsigned long int flags;
974 spin_lock_irqsave(&p->lock, flags);
976 if (p->dev->flags & IFF_UP)
977 p->config &= ~OFPPC_PORT_DOWN;
979 p->config |= OFPPC_PORT_DOWN;
981 if (netif_carrier_ok(p->dev))
982 p->state &= ~OFPPS_LINK_DOWN;
984 p->state |= OFPPS_LINK_DOWN;
986 spin_unlock_irqrestore(&p->lock, flags);
990 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
993 struct ofp_port_status *ops;
995 ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
999 ops->reason = status;
1000 memset(ops->pad, 0, sizeof ops->pad);
1001 fill_port_desc(p, &ops->desc);
1003 return send_openflow_skb(skb, NULL);
1007 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1008 enum ofp_flow_expired_reason reason)
1010 struct sk_buff *skb;
1011 struct ofp_flow_expired *ofe;
1013 if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1016 ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1020 flow_fill_match(&ofe->match, &flow->key);
1022 ofe->priority = htons(flow->priority);
1023 ofe->reason = reason;
1024 memset(ofe->pad, 0, sizeof ofe->pad);
1026 ofe->duration = htonl((jiffies - flow->init_time) / HZ);
1027 memset(ofe->pad2, 0, sizeof ofe->pad2);
1028 ofe->packet_count = cpu_to_be64(flow->packet_count);
1029 ofe->byte_count = cpu_to_be64(flow->byte_count);
1031 return send_openflow_skb(skb, NULL);
1033 EXPORT_SYMBOL(dp_send_flow_expired);
1036 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1037 uint16_t type, uint16_t code, const void *data, size_t len)
1039 struct sk_buff *skb;
1040 struct ofp_error_msg *oem;
1043 oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1048 oem->type = htons(type);
1049 oem->code = htons(code);
1050 memcpy(oem->data, data, len);
1052 return send_openflow_skb(skb, sender);
1056 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1057 const struct ofp_header *rq)
1059 struct sk_buff *skb;
1060 struct ofp_header *reply;
1062 reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1067 memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1068 return send_openflow_skb(skb, sender);
1071 /* Generic Netlink interface.
1073 * See netlink(7) for an introduction to netlink. See
1074 * http://linux-net.osdl.org/index.php/Netlink for more information and
1075 * pointers on how to work with netlink and Generic Netlink in the kernel and
1078 static struct genl_family dp_genl_family = {
1079 .id = GENL_ID_GENERATE,
1081 .name = DP_GENL_FAMILY_NAME,
1083 .maxattr = DP_GENL_A_MAX,
1086 /* Attribute policy: what each attribute may contain. */
1087 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1088 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1089 [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1090 [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1093 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1095 if (!info->attrs[DP_GENL_A_DP_IDX])
1098 return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1101 static struct genl_ops dp_genl_ops_add_dp = {
1102 .cmd = DP_GENL_C_ADD_DP,
1103 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1104 .policy = dp_genl_policy,
1105 .doit = dp_genl_add,
1109 struct datapath *dp_get(int dp_idx)
1111 if (dp_idx < 0 || dp_idx > DP_MAX)
1113 return rcu_dereference(dps[dp_idx]);
1116 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1118 struct datapath *dp;
1121 if (!info->attrs[DP_GENL_A_DP_IDX])
1124 dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1134 static struct genl_ops dp_genl_ops_del_dp = {
1135 .cmd = DP_GENL_C_DEL_DP,
1136 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1137 .policy = dp_genl_policy,
1138 .doit = dp_genl_del,
1142 /* Queries a datapath for related information. Currently the only relevant
1143 * information is the datapath's multicast group ID. Really we want one
1144 * multicast group per datapath, but because of locking issues[*] we can't
1145 * easily get one. Thus, every datapath will currently return the same
1146 * global multicast group ID, but in the future it would be nice to fix that.
1148 * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1149 * mutex, and genl_register_mc_group, called to acquire a new multicast
1150 * group ID, also acquires genl_lock, thus deadlock.
1152 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1154 struct datapath *dp;
1155 struct sk_buff *ans_skb = NULL;
1159 if (!info->attrs[DP_GENL_A_DP_IDX])
1163 dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1164 dp = dp_get(dp_idx);
1169 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1174 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1175 0, DP_GENL_C_QUERY_DP);
1180 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1181 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1183 genlmsg_end(ans_skb, data);
1184 err = genlmsg_reply(ans_skb, info);
1196 static struct genl_ops dp_genl_ops_query_dp = {
1197 .cmd = DP_GENL_C_QUERY_DP,
1198 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199 .policy = dp_genl_policy,
1200 .doit = dp_genl_query,
1204 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1206 struct datapath *dp;
1207 struct net_device *port;
1210 if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1214 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1220 /* Get interface to add/remove. */
1221 port = dev_get_by_name(&init_net,
1222 nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1228 /* Execute operation. */
1229 if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1230 err = add_switch_port(dp, port);
1232 if (port->br_port == NULL || port->br_port->dp != dp) {
1236 err = dp_del_switch_port(port->br_port);
1245 static struct genl_ops dp_genl_ops_add_port = {
1246 .cmd = DP_GENL_C_ADD_PORT,
1247 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1248 .policy = dp_genl_policy,
1249 .doit = dp_genl_add_del_port,
1253 static struct genl_ops dp_genl_ops_del_port = {
1254 .cmd = DP_GENL_C_DEL_PORT,
1255 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1256 .policy = dp_genl_policy,
1257 .doit = dp_genl_add_del_port,
1261 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1263 struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1264 struct datapath *dp;
1265 struct ofp_header *oh;
1266 struct sender sender;
1269 if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1272 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1276 if (nla_len(va) < sizeof(struct ofp_header))
1280 sender.xid = oh->xid;
1281 sender.pid = info->snd_pid;
1282 sender.seq = info->snd_seq;
1284 mutex_lock(&dp_mutex);
1285 err = fwd_control_input(dp->chain, &sender,
1286 nla_data(va), nla_len(va));
1287 mutex_unlock(&dp_mutex);
1291 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1292 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1295 static int desc_stats_dump(struct datapath *dp, void *state,
1296 void *body, int *body_len)
1298 struct ofp_desc_stats *ods = body;
1299 int n_bytes = sizeof *ods;
1301 if (n_bytes > *body_len) {
1304 *body_len = n_bytes;
1306 strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1307 strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1308 strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1309 strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1314 struct flow_stats_state {
1316 struct sw_table_position position;
1317 const struct ofp_flow_stats_request *rq;
1320 int bytes_used, bytes_allocated;
1323 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1326 const struct ofp_flow_stats_request *fsr = body;
1327 struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1330 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1331 memset(&s->position, 0, sizeof s->position);
1337 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1339 struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1340 struct flow_stats_state *s = private;
1341 struct ofp_flow_stats *ofs;
1344 length = sizeof *ofs + sf_acts->actions_len;
1345 if (length + s->bytes_used > s->bytes_allocated)
1348 ofs = s->body + s->bytes_used;
1349 ofs->length = htons(length);
1350 ofs->table_id = s->table_idx;
1352 ofs->match.wildcards = htonl(flow->key.wildcards);
1353 ofs->match.in_port = flow->key.in_port;
1354 memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1355 memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1356 ofs->match.dl_vlan = flow->key.dl_vlan;
1357 ofs->match.dl_type = flow->key.dl_type;
1358 ofs->match.nw_src = flow->key.nw_src;
1359 ofs->match.nw_dst = flow->key.nw_dst;
1360 ofs->match.nw_proto = flow->key.nw_proto;
1362 ofs->match.tp_src = flow->key.tp_src;
1363 ofs->match.tp_dst = flow->key.tp_dst;
1364 ofs->duration = htonl((jiffies - flow->init_time) / HZ);
1365 ofs->priority = htons(flow->priority);
1366 ofs->idle_timeout = htons(flow->idle_timeout);
1367 ofs->hard_timeout = htons(flow->hard_timeout);
1368 memset(ofs->pad2, 0, sizeof ofs->pad2);
1369 ofs->packet_count = cpu_to_be64(flow->packet_count);
1370 ofs->byte_count = cpu_to_be64(flow->byte_count);
1371 memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1373 s->bytes_used += length;
1377 static int flow_stats_dump(struct datapath *dp, void *state,
1378 void *body, int *body_len)
1380 struct flow_stats_state *s = state;
1381 struct sw_flow_key match_key;
1385 s->bytes_allocated = *body_len;
1388 flow_extract_match(&match_key, &s->rq->match);
1389 while (s->table_idx < dp->chain->n_tables
1390 && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1392 struct sw_table *table = dp->chain->tables[s->table_idx];
1394 error = table->iterate(table, &match_key, &s->position,
1395 flow_stats_dump_callback, s);
1400 memset(&s->position, 0, sizeof s->position);
1402 *body_len = s->bytes_used;
1404 /* If error is 0, we're done.
1405 * Otherwise, if some bytes were used, there are more flows to come.
1406 * Otherwise, we were not able to fit even a single flow in the body,
1407 * which indicates that we have a single flow with too many actions to
1408 * fit. We won't ever make any progress at that rate, so give up. */
1409 return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1412 static void flow_stats_done(void *state)
1417 static int aggregate_stats_init(struct datapath *dp,
1418 const void *body, int body_len,
1421 *state = (void *)body;
1425 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1427 struct ofp_aggregate_stats_reply *rpy = private;
1428 rpy->packet_count += flow->packet_count;
1429 rpy->byte_count += flow->byte_count;
1434 static int aggregate_stats_dump(struct datapath *dp, void *state,
1435 void *body, int *body_len)
1437 struct ofp_aggregate_stats_request *rq = state;
1438 struct ofp_aggregate_stats_reply *rpy;
1439 struct sw_table_position position;
1440 struct sw_flow_key match_key;
1443 if (*body_len < sizeof *rpy)
1446 *body_len = sizeof *rpy;
1448 memset(rpy, 0, sizeof *rpy);
1450 flow_extract_match(&match_key, &rq->match);
1451 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1452 memset(&position, 0, sizeof position);
1453 while (table_idx < dp->chain->n_tables
1454 && (rq->table_id == 0xff || rq->table_id == table_idx))
1456 struct sw_table *table = dp->chain->tables[table_idx];
1459 error = table->iterate(table, &match_key, &position,
1460 aggregate_stats_dump_callback, rpy);
1465 memset(&position, 0, sizeof position);
1468 rpy->packet_count = cpu_to_be64(rpy->packet_count);
1469 rpy->byte_count = cpu_to_be64(rpy->byte_count);
1470 rpy->flow_count = htonl(rpy->flow_count);
1474 static int table_stats_dump(struct datapath *dp, void *state,
1475 void *body, int *body_len)
1477 struct ofp_table_stats *ots;
1478 int n_bytes = dp->chain->n_tables * sizeof *ots;
1480 if (n_bytes > *body_len)
1482 *body_len = n_bytes;
1483 for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1484 struct sw_table_stats stats;
1485 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1486 strncpy(ots->name, stats.name, sizeof ots->name);
1488 ots->wildcards = htonl(stats.wildcards);
1489 memset(ots->pad, 0, sizeof ots->pad);
1490 ots->max_entries = htonl(stats.max_flows);
1491 ots->active_count = htonl(stats.n_flows);
1492 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1493 ots->matched_count = cpu_to_be64(stats.n_matched);
1498 struct port_stats_state {
1502 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1505 struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1513 static int port_stats_dump(struct datapath *dp, void *state,
1514 void *body, int *body_len)
1516 struct port_stats_state *s = state;
1517 struct ofp_port_stats *ops;
1518 int n_ports, max_ports;
1521 max_ports = *body_len / sizeof *ops;
1527 for (i = s->port; i < OFPP_MAX && n_ports < max_ports; i++) {
1528 struct net_bridge_port *p = dp->ports[i];
1529 struct net_device_stats *stats;
1532 stats = p->dev->get_stats(p->dev);
1533 ops->port_no = htons(p->port_no);
1534 memset(ops->pad, 0, sizeof ops->pad);
1535 ops->rx_packets = cpu_to_be64(stats->rx_packets);
1536 ops->tx_packets = cpu_to_be64(stats->tx_packets);
1537 ops->rx_bytes = cpu_to_be64(stats->rx_bytes);
1538 ops->tx_bytes = cpu_to_be64(stats->tx_bytes);
1539 ops->rx_dropped = cpu_to_be64(stats->rx_dropped);
1540 ops->tx_dropped = cpu_to_be64(stats->tx_dropped);
1541 ops->rx_errors = cpu_to_be64(stats->rx_errors);
1542 ops->tx_errors = cpu_to_be64(stats->tx_errors);
1543 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1544 ops->rx_over_err = cpu_to_be64(stats->rx_over_errors);
1545 ops->rx_crc_err = cpu_to_be64(stats->rx_crc_errors);
1546 ops->collisions = cpu_to_be64(stats->collisions);
1551 *body_len = n_ports * sizeof *ops;
1552 return n_ports >= max_ports;
1555 static void port_stats_done(void *state)
1561 /* Minimum and maximum acceptable number of bytes in body member of
1562 * struct ofp_stats_request. */
1563 size_t min_body, max_body;
1565 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1566 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1567 * Returns zero if successful, otherwise a negative error code.
1568 * May initialize '*state' to state information. May be null if no
1569 * initialization is required.*/
1570 int (*init)(struct datapath *dp, const void *body, int body_len,
1573 /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1574 * modifies '*body_len' to reflect the number of bytes actually used.
1575 * ('body' will be transmitted as the 'body' member of struct
1576 * ofp_stats_reply.) */
1577 int (*dump)(struct datapath *dp, void *state,
1578 void *body, int *body_len);
1580 /* Cleans any state created by the init or dump functions. May be null
1581 * if no cleanup is required. */
1582 void (*done)(void *state);
1585 static const struct stats_type stats[] = {
1594 sizeof(struct ofp_flow_stats_request),
1595 sizeof(struct ofp_flow_stats_request),
1600 [OFPST_AGGREGATE] = {
1601 sizeof(struct ofp_aggregate_stats_request),
1602 sizeof(struct ofp_aggregate_stats_request),
1603 aggregate_stats_init,
1604 aggregate_stats_dump,
1624 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1626 struct datapath *dp;
1627 struct sender sender;
1628 const struct stats_type *s;
1629 struct ofp_stats_reply *osr;
1631 int max_openflow_len, body_len;
1635 /* Set up the cleanup function for this dump. Linux 2.6.20 and later
1636 * support setting up cleanup functions via the .doneit member of
1637 * struct genl_ops. This kluge supports earlier versions also. */
1638 cb->done = dp_genl_openflow_done;
1640 sender.pid = NETLINK_CB(cb->skb).pid;
1641 sender.seq = cb->nlh->nlmsg_seq;
1643 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1644 struct ofp_stats_request *rq;
1646 size_t len, body_len;
1649 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1650 dp_genl_openflow_policy);
1654 if (!attrs[DP_GENL_A_DP_IDX])
1656 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1657 dp = dp_get(dp_idx);
1661 va = attrs[DP_GENL_A_OPENFLOW];
1663 if (!va || len < sizeof *rq)
1667 sender.xid = rq->header.xid;
1668 type = ntohs(rq->type);
1669 if (rq->header.version != OFP_VERSION) {
1670 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1671 OFPBRC_BAD_VERSION, rq, len);
1674 if (rq->header.type != OFPT_STATS_REQUEST
1675 || ntohs(rq->header.length) != len)
1678 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1679 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1680 OFPBRC_BAD_STAT, rq, len);
1685 body_len = len - offsetof(struct ofp_stats_request, body);
1686 if (body_len < s->min_body || body_len > s->max_body)
1690 cb->args[1] = dp_idx;
1692 cb->args[3] = rq->header.xid;
1695 err = s->init(dp, rq->body, body_len, &state);
1698 cb->args[4] = (long) state;
1700 } else if (cb->args[0] == 1) {
1701 sender.xid = cb->args[3];
1702 dp_idx = cb->args[1];
1703 s = &stats[cb->args[2]];
1705 dp = dp_get(dp_idx);
1712 osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1715 return PTR_ERR(osr);
1716 osr->type = htons(s - stats);
1718 resize_openflow_skb(skb, &osr->header, max_openflow_len);
1720 body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1722 err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1727 osr->flags = ntohs(OFPSF_REPLY_MORE);
1728 resize_openflow_skb(skb, &osr->header,
1729 (offsetof(struct ofp_stats_reply, body)
1738 dp_genl_openflow_done(struct netlink_callback *cb)
1741 const struct stats_type *s = &stats[cb->args[2]];
1743 s->done((void *) cb->args[4]);
1748 static struct genl_ops dp_genl_ops_openflow = {
1749 .cmd = DP_GENL_C_OPENFLOW,
1750 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1751 .policy = dp_genl_openflow_policy,
1752 .doit = dp_genl_openflow,
1753 .dumpit = dp_genl_openflow_dumpit,
1756 static struct genl_ops *dp_genl_all_ops[] = {
1757 /* Keep this operation first. Generic Netlink dispatching
1758 * looks up operations with linear search, so we want it at the
1760 &dp_genl_ops_openflow,
1762 &dp_genl_ops_add_dp,
1763 &dp_genl_ops_del_dp,
1764 &dp_genl_ops_query_dp,
1765 &dp_genl_ops_add_port,
1766 &dp_genl_ops_del_port,
1769 static int dp_init_netlink(void)
1774 err = genl_register_family(&dp_genl_family);
1778 for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1779 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1781 goto err_unregister;
1784 strcpy(mc_group.name, "openflow");
1785 err = genl_register_mc_group(&dp_genl_family, &mc_group);
1787 goto err_unregister;
1792 genl_unregister_family(&dp_genl_family);
1796 static void dp_uninit_netlink(void)
1798 genl_unregister_family(&dp_genl_family);
1801 static int __init dp_init(void)
1805 printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1806 "protocol 0x%02x\n", OFP_VERSION);
1812 err = register_netdevice_notifier(&dp_device_notifier);
1814 goto error_flow_exit;
1816 err = dp_init_netlink();
1818 goto error_unreg_notifier;
1820 /* Hook into callback used by the bridge to intercept packets.
1821 * Parasites we are. */
1822 if (br_handle_frame_hook)
1823 printk("openflow: hijacking bridge hook\n");
1824 br_handle_frame_hook = dp_frame_hook;
1828 error_unreg_notifier:
1829 unregister_netdevice_notifier(&dp_device_notifier);
1833 printk(KERN_EMERG "openflow: failed to install!");
1837 static void dp_cleanup(void)
1840 dp_uninit_netlink();
1841 unregister_netdevice_notifier(&dp_device_notifier);
1843 br_handle_frame_hook = NULL;
1846 module_init(dp_init);
1847 module_exit(dp_cleanup);
1849 MODULE_DESCRIPTION("OpenFlow switching datapath");
1850 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1851 MODULE_LICENSE("GPL");