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/time.h>
19 #include <linux/etherdevice.h>
20 #include <linux/kernel.h>
21 #include <linux/kthread.h>
22 #include <linux/mutex.h>
23 #include <linux/rtnetlink.h>
24 #include <linux/rcupdate.h>
25 #include <linux/version.h>
26 #include <linux/ethtool.h>
27 #include <linux/random.h>
28 #include <asm/system.h>
29 #include <asm/div64.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/rculist.h>
35 #include <linux/workqueue.h>
36 #include <linux/dmi.h>
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow-netlink.h"
41 #include "nx_act_snat.h"
51 /* Strings to describe the manufacturer, hardware, and software. This data
52 * is queriable through the switch description stats message. */
53 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks, Inc.";
54 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
55 static char sw_desc[DESC_STR_LEN] = VERSION BUILDNR;
56 static char serial_num[SERIAL_NUM_LEN] = "None";
58 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
59 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
60 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
61 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
62 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
64 MODULE_PARM(mfr_desc, "s");
65 MODULE_PARM(hw_desc, "s");
66 MODULE_PARM(sw_desc, "s");
67 MODULE_PARM(serial_num, "s");
71 /* Number of milliseconds between runs of the maintenance thread. */
72 #define MAINT_SLEEP_MSECS 1000
74 #define UINT32_MAX 4294967295U
75 #define UINT16_MAX 65535
76 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
78 static struct genl_family dp_genl_family;
79 static struct genl_multicast_group mc_group;
81 /* It's hard to imagine wanting more than one datapath, but... */
84 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
85 * by dp_mutex. dp_mutex is almost completely redundant with genl_mutex
86 * maintained by the Generic Netlink code, but the timeout path needs mutual
89 * It is safe to access the datapath and net_bridge_port structures with just
92 static struct datapath *dps[DP_MAX];
93 DEFINE_MUTEX(dp_mutex);
94 EXPORT_SYMBOL(dp_mutex);
96 static int dp_maint_func(void *data);
97 static void init_port_status(struct net_bridge_port *p);
98 static int dp_genl_openflow_done(struct netlink_callback *);
99 static struct net_bridge_port *new_nbp(struct datapath *,
100 struct net_device *, int port_no);
102 /* nla_shrink - reduce amount of space reserved by nla_reserve
103 * @skb: socket buffer from which to recover room
104 * @nla: netlink attribute to adjust
105 * @len: new length of attribute payload
107 * Reduces amount of space reserved by a call to nla_reserve.
109 * No other attributes may be added between calling nla_reserve and this
110 * function, since it will create a hole in the message.
112 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
114 int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
118 nla->nla_len = nla_attr_size(len);
121 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
122 * If 'sender' is nonnull, then it is used as the message's destination. 'dp'
123 * must specify the datapath to use.
125 * '*max_openflow_len' receives the maximum number of bytes that are available
126 * for the embedded OpenFlow message. The caller must call
127 * resize_openflow_skb() to set the actual size of the message to this number
130 * Returns the openflow header if successful, otherwise (if 'skb' is too small)
133 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
134 const struct sender *sender, int *max_openflow_len)
136 struct ofp_header *oh;
140 /* Assemble the Generic Netlink wrapper. */
141 if (!genlmsg_put(skb,
142 sender ? sender->pid : 0,
143 sender ? sender->seq : 0,
144 &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
145 return ERR_PTR(-ENOBUFS);
146 if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
147 return ERR_PTR(-ENOBUFS);
148 openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
149 if (openflow_len < sizeof *oh)
150 return ERR_PTR(-ENOBUFS);
151 *max_openflow_len = openflow_len;
152 attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
155 /* Fill in the header. The caller is responsible for the length. */
157 oh->version = OFP_VERSION;
159 oh->xid = sender ? sender->xid : 0;
164 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
165 * length 'new_length' (in bytes), adjusting pointers and size values as
168 resize_openflow_skb(struct sk_buff *skb,
169 struct ofp_header *oh, size_t new_length)
171 struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
172 nla_shrink(skb, attr, new_length);
173 oh->length = htons(new_length);
174 nlmsg_end(skb, (struct nlmsghdr *) skb->data);
177 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
178 * length. Returns a null pointer if memory is unavailable, otherwise returns
179 * the OpenFlow header and stores a pointer to the skb in '*pskb'.
181 * 'type' is the OpenFlow message type. If 'sender' is nonnull, then it is
182 * used as the message's destination. 'dp' must specify the datapath to
185 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
186 const struct sender *sender, struct sk_buff **pskb)
188 struct ofp_header *oh;
191 int max_openflow_len;
193 if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
195 printk("alloc_openflow_skb: openflow message too large: %zu\n",
200 genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
201 genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
202 genl_len += nla_total_size(openflow_len); /* DP_GENL_A_OPENFLOW */
203 skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
206 printk("alloc_openflow_skb: genlmsg_new failed\n");
210 oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
211 BUG_ON(!oh || IS_ERR(oh));
212 resize_openflow_skb(skb, oh, openflow_len);
217 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
220 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
223 ? genlmsg_unicast(skb, sender->pid)
224 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
227 /* Retrieves the datapath id, which is the MAC address of the "of" device. */
229 uint64_t get_datapath_id(struct net_device *dev)
234 for (i=0; i<ETH_ALEN; i++)
235 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
240 /* Creates a new datapath numbered 'dp_idx'. Returns 0 for success or a
241 * negative error code. */
242 static int new_dp(int dp_idx)
247 if (dp_idx < 0 || dp_idx >= DP_MAX)
250 if (!try_module_get(THIS_MODULE))
253 /* Exit early if a datapath with that number already exists. */
260 dp = kzalloc(sizeof *dp, GFP_KERNEL);
266 /* Setup our "of" device */
267 err = dp_dev_setup(dp);
271 dp->chain = chain_create(dp);
272 if (dp->chain == NULL)
273 goto err_destroy_dp_dev;
274 INIT_LIST_HEAD(&dp->port_list);
276 dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
277 if (IS_ERR(dp->local_port)) {
278 err = PTR_ERR(dp->local_port);
279 goto err_destroy_local_port;
283 dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
285 dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
286 if (IS_ERR(dp->dp_task))
287 goto err_destroy_chain;
293 err_destroy_local_port:
294 dp_del_switch_port(dp->local_port);
296 chain_destroy(dp->chain);
302 module_put(THIS_MODULE);
306 /* Find and return a free port number under 'dp'. */
307 static int find_portno(struct datapath *dp)
310 for (i = 0; i < DP_MAX_PORTS; i++)
311 if (dp->ports[i] == NULL)
316 static struct net_bridge_port *new_nbp(struct datapath *dp,
317 struct net_device *dev, int port_no)
319 struct net_bridge_port *p;
321 if (dev->br_port != NULL)
322 return ERR_PTR(-EBUSY);
324 p = kzalloc(sizeof(*p), GFP_KERNEL);
326 return ERR_PTR(-ENOMEM);
329 dev_set_promiscuity(dev, 1);
334 p->port_no = port_no;
335 spin_lock_init(&p->lock);
336 INIT_WORK(&p->port_task, NULL);
337 if (port_no != OFPP_LOCAL)
338 rcu_assign_pointer(dev->br_port, p);
339 if (port_no < DP_MAX_PORTS)
340 rcu_assign_pointer(dp->ports[port_no], p);
341 list_add_rcu(&p->node, &dp->port_list);
346 int add_switch_port(struct datapath *dp, struct net_device *dev)
348 struct net_bridge_port *p;
351 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
355 port_no = find_portno(dp);
359 p = new_nbp(dp, dev, port_no);
365 /* Notify the ctlpath that this port has been added */
366 dp_send_port_status(p, OFPPR_ADD);
371 /* Delete 'p' from switch. */
372 int dp_del_switch_port(struct net_bridge_port *p)
378 /* First drop references to device. */
379 cancel_work_sync(&p->port_task);
381 dev_set_promiscuity(p->dev, -1);
383 list_del_rcu(&p->node);
384 if (p->port_no != OFPP_LOCAL)
385 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
386 rcu_assign_pointer(p->dev->br_port, NULL);
388 /* Then wait until no one is still using it, and destroy it. */
392 /* Free any SNAT configuration on the port. */
393 spin_lock_irqsave(&p->lock, flags);
395 spin_unlock_irqrestore(&p->lock, flags);
398 /* Notify the ctlpath that this port no longer exists */
399 dp_send_port_status(p, OFPPR_DELETE);
407 static void del_dp(struct datapath *dp)
409 struct net_bridge_port *p, *n;
411 kthread_stop(dp->dp_task);
413 /* Drop references to DP. */
414 list_for_each_entry_safe (p, n, &dp->port_list, node)
415 dp_del_switch_port(p);
416 rcu_assign_pointer(dps[dp->dp_idx], NULL);
418 /* Kill off local_port dev references from buffered packets that have
419 * associated dst entries. */
423 /* Destroy dp->netdev. (Must follow deleting switch ports since
424 * dp->local_port has a reference to it.) */
427 /* Wait until no longer in use, then destroy it. */
429 chain_destroy(dp->chain);
431 module_put(THIS_MODULE);
434 static int dp_maint_func(void *data)
436 struct datapath *dp = (struct datapath *) data;
438 while (!kthread_should_stop()) {
440 struct net_bridge_port *p;
442 /* Expire old SNAT entries */
444 list_for_each_entry_rcu (p, &dp->port_list, node)
449 /* Timeout old entries */
450 chain_timeout(dp->chain);
451 msleep_interruptible(MAINT_SLEEP_MSECS);
458 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
460 /* Make our own copy of the packet. Otherwise we will mangle the
461 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
462 * (No one comes after us, since we tell handle_bridge() that we took
464 skb = skb_share_check(skb, GFP_ATOMIC);
469 /* Check if this packet needs early SNAT processing. */
470 if (snat_pre_route(skb))
474 /* Push the Ethernet header back on. */
475 skb_push(skb, ETH_HLEN);
476 skb_reset_mac_header(skb);
477 fwd_port_input(p->dp->chain, skb, p);
481 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
482 * different set of devices!)
484 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
485 /* Called with rcu_read_lock. */
486 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
489 do_port_input(p, skb);
492 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
493 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
495 do_port_input(p, *pskb);
499 /* NB: This has only been tested on 2.4.35 */
500 static void dp_frame_hook(struct sk_buff *skb)
502 struct net_bridge_port *p = skb->dev->br_port;
505 do_port_input(p, skb);
512 /* Forwarding output path.
513 * Based on net/bridge/br_forward.c. */
515 static inline unsigned packet_length(const struct sk_buff *skb)
517 int length = skb->len - ETH_HLEN;
518 if (skb->protocol == htons(ETH_P_8021Q))
523 /* Send packets out all the ports except the originating one. If the
524 * "flood" argument is set, only send along the minimum spanning tree.
527 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
529 u32 disable = flood ? OFPPC_NO_FLOOD : 0;
530 struct net_bridge_port *p;
533 list_for_each_entry_rcu (p, &dp->port_list, node) {
534 if (skb->dev == p->dev || p->config & disable)
536 if (prev_port != -1) {
537 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
542 dp_output_port(dp, clone, prev_port, 0);
544 prev_port = p->port_no;
547 dp_output_port(dp, skb, prev_port, 0);
554 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
555 FIXME: how are devices reference counted? */
556 void dp_set_origin(struct datapath *dp, uint16_t in_port,
559 struct net_bridge_port *p;
560 p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
561 : in_port == OFPP_LOCAL ? dp->local_port
571 dp_xmit_skb_finish(struct sk_buff *skb)
573 /* Copy back the Ethernet header that was stowed earlier. */
574 if (skb->protocol == htons(ETH_P_IP) && snat_copy_header(skb)) {
578 skb_reset_mac_header(skb);
580 if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb)) {
581 printk("dropped over-mtu packet: %d > %d\n",
582 packet_length(skb), skb->dev->mtu);
587 skb_push(skb, ETH_HLEN);
594 dp_xmit_skb(struct sk_buff *skb)
599 skb_pull(skb, ETH_HLEN);
601 /* The ip_fragment function does not copy the Ethernet header into
602 * the newly generated frames, so stow the original. */
603 if (skb->protocol == htons(ETH_P_IP))
604 snat_save_header(skb);
606 if (skb->protocol == htons(ETH_P_IP) &&
607 skb->len > skb->dev->mtu &&
609 err = ip_fragment(skb, dp_xmit_skb_finish);
611 err = dp_xmit_skb_finish(skb);
620 dp_xmit_skb(struct sk_buff *skb)
623 if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb)) {
624 printk("dropped over-mtu packet: %d > %d\n",
625 packet_length(skb), skb->dev->mtu);
636 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
638 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
644 /* Send it out the port it came in on, which is already set in
648 printk("skb device not set forwarding to in_port\n");
652 return dp_xmit_skb(skb);
655 int retval = run_flow_through_tables(dp->chain, skb,
663 return output_all(dp, skb, 1);
666 return output_all(dp, skb, 0);
668 case OFPP_CONTROLLER:
669 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
673 struct net_device *dev = dp->netdev;
677 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
680 case 0 ... DP_MAX_PORTS - 1: {
681 struct net_bridge_port *p = dp->ports[out_port];
684 if (p->dev == skb->dev) {
685 /* To send to the input port, must use OFPP_IN_PORT */
688 printk("can't directly forward to input port\n");
691 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
696 return dp_xmit_skb(skb);
706 printk("can't forward to bad port %d\n", out_port);
710 /* Takes ownership of 'skb' and transmits it to 'dp''s control path. If
711 * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
712 * otherwise, all of 'skb' is sent. 'reason' indicates why 'skb' is being
713 * sent. 'max_len' sets the maximum number of bytes that the caller
714 * wants to be sent; a value of 0 indicates the entire packet should be
717 dp_output_control(struct datapath *dp, struct sk_buff *skb,
718 uint32_t buffer_id, size_t max_len, int reason)
720 /* FIXME? Can we avoid creating a new skbuff in the case where we
721 * forward the whole packet? */
722 struct sk_buff *f_skb;
723 struct ofp_packet_in *opi;
724 size_t fwd_len, opi_len;
728 if ((buffer_id != (uint32_t) -1) && max_len)
729 fwd_len = min(fwd_len, max_len);
731 opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
732 opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
737 opi->buffer_id = htonl(buffer_id);
738 opi->total_len = htons(skb->len);
739 opi->in_port = htons(skb->dev && skb->dev->br_port
740 ? skb->dev->br_port->port_no
742 opi->reason = reason;
744 skb_copy_bits(skb, 0, opi->data, fwd_len);
745 err = send_openflow_skb(f_skb, NULL);
752 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
755 desc->port_no = htons(p->port_no);
756 strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
757 desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
758 memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
761 desc->advertised = 0;
764 spin_lock_irqsave(&p->lock, flags);
765 desc->config = htonl(p->config);
766 desc->state = htonl(p->state);
767 spin_unlock_irqrestore(&p->lock, flags);
769 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
770 if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
771 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
773 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
774 /* Set the supported features */
775 if (ecmd.supported & SUPPORTED_10baseT_Half)
776 desc->supported |= OFPPF_10MB_HD;
777 if (ecmd.supported & SUPPORTED_10baseT_Full)
778 desc->supported |= OFPPF_10MB_FD;
779 if (ecmd.supported & SUPPORTED_100baseT_Half)
780 desc->supported |= OFPPF_100MB_HD;
781 if (ecmd.supported & SUPPORTED_100baseT_Full)
782 desc->supported |= OFPPF_100MB_FD;
783 if (ecmd.supported & SUPPORTED_1000baseT_Half)
784 desc->supported |= OFPPF_1GB_HD;
785 if (ecmd.supported & SUPPORTED_1000baseT_Full)
786 desc->supported |= OFPPF_1GB_FD;
787 if (ecmd.supported & SUPPORTED_10000baseT_Full)
788 desc->supported |= OFPPF_10GB_FD;
789 if (ecmd.supported & SUPPORTED_TP)
790 desc->supported |= OFPPF_COPPER;
791 if (ecmd.supported & SUPPORTED_FIBRE)
792 desc->supported |= OFPPF_FIBER;
793 if (ecmd.supported & SUPPORTED_Autoneg)
794 desc->supported |= OFPPF_AUTONEG;
795 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
796 if (ecmd.supported & SUPPORTED_Pause)
797 desc->supported |= OFPPF_PAUSE;
798 if (ecmd.supported & SUPPORTED_Asym_Pause)
799 desc->supported |= OFPPF_PAUSE_ASYM;
800 #endif /* kernel >= 2.6.14 */
802 /* Set the advertised features */
803 if (ecmd.advertising & ADVERTISED_10baseT_Half)
804 desc->advertised |= OFPPF_10MB_HD;
805 if (ecmd.advertising & ADVERTISED_10baseT_Full)
806 desc->advertised |= OFPPF_10MB_FD;
807 if (ecmd.advertising & ADVERTISED_100baseT_Half)
808 desc->advertised |= OFPPF_100MB_HD;
809 if (ecmd.advertising & ADVERTISED_100baseT_Full)
810 desc->advertised |= OFPPF_100MB_FD;
811 if (ecmd.advertising & ADVERTISED_1000baseT_Half)
812 desc->advertised |= OFPPF_1GB_HD;
813 if (ecmd.advertising & ADVERTISED_1000baseT_Full)
814 desc->advertised |= OFPPF_1GB_FD;
815 if (ecmd.advertising & ADVERTISED_10000baseT_Full)
816 desc->advertised |= OFPPF_10GB_FD;
817 if (ecmd.advertising & ADVERTISED_TP)
818 desc->advertised |= OFPPF_COPPER;
819 if (ecmd.advertising & ADVERTISED_FIBRE)
820 desc->advertised |= OFPPF_FIBER;
821 if (ecmd.advertising & ADVERTISED_Autoneg)
822 desc->advertised |= OFPPF_AUTONEG;
823 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
824 if (ecmd.advertising & ADVERTISED_Pause)
825 desc->advertised |= OFPPF_PAUSE;
826 if (ecmd.advertising & ADVERTISED_Asym_Pause)
827 desc->advertised |= OFPPF_PAUSE_ASYM;
828 #endif /* kernel >= 2.6.14 */
830 /* Set the current features */
831 if (ecmd.speed == SPEED_10)
832 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
833 else if (ecmd.speed == SPEED_100)
834 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
835 else if (ecmd.speed == SPEED_1000)
836 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
837 else if (ecmd.speed == SPEED_10000)
838 desc->curr = OFPPF_10GB_FD;
840 if (ecmd.port == PORT_TP)
841 desc->curr |= OFPPF_COPPER;
842 else if (ecmd.port == PORT_FIBRE)
843 desc->curr |= OFPPF_FIBER;
846 desc->curr |= OFPPF_AUTONEG;
850 desc->curr = htonl(desc->curr);
851 desc->supported = htonl(desc->supported);
852 desc->advertised = htonl(desc->advertised);
853 desc->peer = htonl(desc->peer);
857 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
859 struct net_bridge_port *p;
860 uint64_t dpid = get_datapath_id(dp->netdev);
863 ofr->datapath_id = cpu_to_be64(dpid);
865 ofr->n_buffers = htonl(N_PKT_BUFFERS);
866 ofr->n_tables = dp->chain->n_tables;
867 ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
868 ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
869 memset(ofr->pad, 0, sizeof ofr->pad);
871 list_for_each_entry_rcu (p, &dp->port_list, node) {
872 fill_port_desc(p, &ofr->ports[port_count]);
880 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
883 struct ofp_switch_features *ofr;
884 size_t ofr_len, port_max_len;
888 port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
889 ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
890 OFPT_FEATURES_REPLY, sender, &skb);
895 port_count = fill_features_reply(dp, ofr);
898 ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
899 resize_openflow_skb(skb, &ofr->header, ofr_len);
900 return send_openflow_skb(skb, sender);
904 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
907 struct ofp_switch_config *osc;
909 osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
914 osc->flags = htons(dp->flags);
915 osc->miss_send_len = htons(dp->miss_send_len);
917 return send_openflow_skb(skb, sender);
921 dp_send_hello(struct datapath *dp, const struct sender *sender,
922 const struct ofp_header *request)
924 if (request->version < OFP_VERSION) {
926 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
927 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
928 OFPHFC_INCOMPATIBLE, err, strlen(err));
932 struct ofp_header *reply;
934 reply = alloc_openflow_skb(dp, sizeof *reply,
935 OFPT_HELLO, sender, &skb);
939 return send_openflow_skb(skb, sender);
943 /* Callback function for a workqueue to disable an interface */
945 down_port_cb(struct work_struct *work)
947 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
951 if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
953 printk("problem bringing up port %s\n", p->dev->name);
955 p->config |= OFPPC_PORT_DOWN;
958 /* Callback function for a workqueue to enable an interface */
960 up_port_cb(struct work_struct *work)
962 struct net_bridge_port *p = container_of(work, struct net_bridge_port,
966 if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
968 printk("problem bringing down port %s\n", p->dev->name);
970 p->config &= ~OFPPC_PORT_DOWN;
974 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
976 unsigned long int flags;
977 int port_no = ntohs(opm->port_no);
978 struct net_bridge_port *p;
979 p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
980 : port_no == OFPP_LOCAL ? dp->local_port
983 /* Make sure the port id hasn't changed since this was sent */
984 if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
987 spin_lock_irqsave(&p->lock, flags);
989 uint32_t config_mask = ntohl(opm->mask);
990 p->config &= ~config_mask;
991 p->config |= ntohl(opm->config) & config_mask;
994 /* Modifying the status of an interface requires taking a lock
995 * that cannot be done from here. For this reason, we use a shared
996 * workqueue, which will cause it to be executed from a safer
998 if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
999 if ((opm->config & htonl(OFPPC_PORT_DOWN))
1000 && (p->config & OFPPC_PORT_DOWN) == 0) {
1001 PREPARE_WORK(&p->port_task, down_port_cb);
1002 schedule_work(&p->port_task);
1003 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
1004 && (p->config & OFPPC_PORT_DOWN)) {
1005 PREPARE_WORK(&p->port_task, up_port_cb);
1006 schedule_work(&p->port_task);
1009 spin_unlock_irqrestore(&p->lock, flags);
1014 /* Initialize the port status field of the bridge port. */
1016 init_port_status(struct net_bridge_port *p)
1018 unsigned long int flags;
1020 spin_lock_irqsave(&p->lock, flags);
1022 if (p->dev->flags & IFF_UP)
1023 p->config &= ~OFPPC_PORT_DOWN;
1025 p->config |= OFPPC_PORT_DOWN;
1027 if (netif_carrier_ok(p->dev))
1028 p->state &= ~OFPPS_LINK_DOWN;
1030 p->state |= OFPPS_LINK_DOWN;
1032 spin_unlock_irqrestore(&p->lock, flags);
1036 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
1038 struct sk_buff *skb;
1039 struct ofp_port_status *ops;
1041 ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
1045 ops->reason = status;
1046 memset(ops->pad, 0, sizeof ops->pad);
1047 fill_port_desc(p, &ops->desc);
1049 return send_openflow_skb(skb, NULL);
1052 /* Convert jiffies_64 to milliseconds. */
1053 static u64 inline jiffies_64_to_msecs(const u64 j)
1055 #if HZ <= MSEC_PER_SEC && !(MSEC_PER_SEC % HZ)
1056 return (MSEC_PER_SEC / HZ) * j;
1057 #elif HZ > MSEC_PER_SEC && !(HZ % MSEC_PER_SEC)
1058 return (j + (HZ / MSEC_PER_SEC) - 1)/(HZ / MSEC_PER_SEC);
1060 return (j * MSEC_PER_SEC) / HZ;
1065 dp_send_flow_end(struct datapath *dp, struct sw_flow *flow,
1066 enum nx_flow_end_reason reason)
1068 struct sk_buff *skb;
1069 struct nx_flow_end *nfe;
1071 if (!dp->send_flow_end)
1074 nfe = alloc_openflow_skb(dp, sizeof *nfe, OFPT_VENDOR, 0, &skb);
1078 nfe->header.vendor = htonl(NX_VENDOR_ID);
1079 nfe->header.subtype = htonl(NXT_FLOW_END);
1081 flow_fill_match(&nfe->match, &flow->key);
1083 nfe->priority = htons(flow->priority);
1084 nfe->reason = reason;
1086 nfe->tcp_flags = flow->tcp_flags;
1087 nfe->ip_tos = flow->ip_tos;
1089 memset(nfe->pad, 0, sizeof nfe->pad);
1091 nfe->init_time = cpu_to_be64(jiffies_64_to_msecs(flow->created));
1092 nfe->used_time = cpu_to_be64(jiffies_64_to_msecs(flow->used));
1093 nfe->end_time = cpu_to_be64(jiffies_64_to_msecs(get_jiffies_64()));
1095 nfe->packet_count = cpu_to_be64(flow->packet_count);
1096 nfe->byte_count = cpu_to_be64(flow->byte_count);
1098 return send_openflow_skb(skb, NULL);
1100 EXPORT_SYMBOL(dp_send_flow_end);
1103 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1104 uint16_t type, uint16_t code, const void *data, size_t len)
1106 struct sk_buff *skb;
1107 struct ofp_error_msg *oem;
1110 oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1115 oem->type = htons(type);
1116 oem->code = htons(code);
1117 memcpy(oem->data, data, len);
1119 return send_openflow_skb(skb, sender);
1123 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1124 const struct ofp_header *rq)
1126 struct sk_buff *skb;
1127 struct ofp_header *reply;
1129 reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1134 memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1135 return send_openflow_skb(skb, sender);
1138 /* Generic Netlink interface.
1140 * See netlink(7) for an introduction to netlink. See
1141 * http://linux-net.osdl.org/index.php/Netlink for more information and
1142 * pointers on how to work with netlink and Generic Netlink in the kernel and
1145 static struct genl_family dp_genl_family = {
1146 .id = GENL_ID_GENERATE,
1148 .name = DP_GENL_FAMILY_NAME,
1150 .maxattr = DP_GENL_A_MAX,
1153 /* Attribute policy: what each attribute may contain. */
1154 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1155 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1156 [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1157 [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1160 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1162 if (!info->attrs[DP_GENL_A_DP_IDX])
1165 return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1168 static struct genl_ops dp_genl_ops_add_dp = {
1169 .cmd = DP_GENL_C_ADD_DP,
1170 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1171 .policy = dp_genl_policy,
1172 .doit = dp_genl_add,
1176 struct datapath *dp_get(int dp_idx)
1178 if (dp_idx < 0 || dp_idx >= DP_MAX)
1180 return rcu_dereference(dps[dp_idx]);
1183 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1185 struct datapath *dp;
1188 if (!info->attrs[DP_GENL_A_DP_IDX])
1191 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1201 static struct genl_ops dp_genl_ops_del_dp = {
1202 .cmd = DP_GENL_C_DEL_DP,
1203 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1204 .policy = dp_genl_policy,
1205 .doit = dp_genl_del,
1209 /* Queries a datapath for related information. Currently the only relevant
1210 * information is the datapath's multicast group ID. Really we want one
1211 * multicast group per datapath, but because of locking issues[*] we can't
1212 * easily get one. Thus, every datapath will currently return the same
1213 * global multicast group ID, but in the future it would be nice to fix that.
1215 * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1216 * mutex, and genl_register_mc_group, called to acquire a new multicast
1217 * group ID, also acquires genl_lock, thus deadlock.
1219 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1221 struct datapath *dp;
1222 struct sk_buff *ans_skb = NULL;
1226 if (!info->attrs[DP_GENL_A_DP_IDX])
1230 dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1231 dp = dp_get(dp_idx);
1236 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1241 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1242 0, DP_GENL_C_QUERY_DP);
1247 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1248 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1250 genlmsg_end(ans_skb, data);
1251 err = genlmsg_reply(ans_skb, info);
1261 static struct genl_ops dp_genl_ops_query_dp = {
1262 .cmd = DP_GENL_C_QUERY_DP,
1263 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1264 .policy = dp_genl_policy,
1265 .doit = dp_genl_query,
1269 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1271 struct datapath *dp;
1272 struct net_device *port;
1275 if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1279 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1285 /* Get interface to add/remove. */
1286 port = dev_get_by_name(&init_net,
1287 nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1293 /* Execute operation. */
1294 if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1295 err = add_switch_port(dp, port);
1297 if (port->br_port == NULL || port->br_port->dp != dp) {
1301 err = dp_del_switch_port(port->br_port);
1310 static struct genl_ops dp_genl_ops_add_port = {
1311 .cmd = DP_GENL_C_ADD_PORT,
1312 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1313 .policy = dp_genl_policy,
1314 .doit = dp_genl_add_del_port,
1318 static struct genl_ops dp_genl_ops_del_port = {
1319 .cmd = DP_GENL_C_DEL_PORT,
1320 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1321 .policy = dp_genl_policy,
1322 .doit = dp_genl_add_del_port,
1326 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1328 struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1329 struct datapath *dp;
1330 struct ofp_header *oh;
1331 struct sender sender;
1334 if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1337 dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1341 if (nla_len(va) < sizeof(struct ofp_header))
1345 sender.xid = oh->xid;
1346 sender.pid = info->snd_pid;
1347 sender.seq = info->snd_seq;
1349 mutex_lock(&dp_mutex);
1350 err = fwd_control_input(dp->chain, &sender,
1351 nla_data(va), nla_len(va));
1352 mutex_unlock(&dp_mutex);
1356 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1357 [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1360 static int desc_stats_dump(struct datapath *dp, void *state,
1361 void *body, int *body_len)
1363 struct ofp_desc_stats *ods = body;
1364 int n_bytes = sizeof *ods;
1366 if (n_bytes > *body_len) {
1369 *body_len = n_bytes;
1371 strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1372 strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1373 strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1374 strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1379 struct flow_stats_state {
1381 struct sw_table_position position;
1382 const struct ofp_flow_stats_request *rq;
1385 int bytes_used, bytes_allocated;
1388 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1391 const struct ofp_flow_stats_request *fsr = body;
1392 struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1395 s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1396 memset(&s->position, 0, sizeof s->position);
1402 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1404 struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1405 struct flow_stats_state *s = private;
1406 struct ofp_flow_stats *ofs;
1410 length = sizeof *ofs + sf_acts->actions_len;
1411 if (length + s->bytes_used > s->bytes_allocated)
1414 ofs = s->body + s->bytes_used;
1415 ofs->length = htons(length);
1416 ofs->table_id = s->table_idx;
1418 ofs->match.wildcards = htonl(flow->key.wildcards);
1419 ofs->match.in_port = flow->key.in_port;
1420 memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1421 memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1422 ofs->match.dl_vlan = flow->key.dl_vlan;
1423 ofs->match.dl_type = flow->key.dl_type;
1424 ofs->match.nw_src = flow->key.nw_src;
1425 ofs->match.nw_dst = flow->key.nw_dst;
1426 ofs->match.nw_proto = flow->key.nw_proto;
1428 ofs->match.tp_src = flow->key.tp_src;
1429 ofs->match.tp_dst = flow->key.tp_dst;
1431 /* The kernel doesn't support 64-bit division, so use the 'do_div'
1432 * macro instead. The first argument is replaced with the quotient,
1433 * while the remainder is the return value. */
1434 duration = get_jiffies_64() - flow->created;
1435 do_div(duration, HZ);
1436 ofs->duration = htonl(duration);
1438 ofs->priority = htons(flow->priority);
1439 ofs->idle_timeout = htons(flow->idle_timeout);
1440 ofs->hard_timeout = htons(flow->hard_timeout);
1441 memset(ofs->pad2, 0, sizeof ofs->pad2);
1442 ofs->packet_count = cpu_to_be64(flow->packet_count);
1443 ofs->byte_count = cpu_to_be64(flow->byte_count);
1444 memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1446 s->bytes_used += length;
1450 static int flow_stats_dump(struct datapath *dp, void *state,
1451 void *body, int *body_len)
1453 struct flow_stats_state *s = state;
1454 struct sw_flow_key match_key;
1458 s->bytes_allocated = *body_len;
1461 flow_extract_match(&match_key, &s->rq->match);
1462 while (s->table_idx < dp->chain->n_tables
1463 && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1465 struct sw_table *table = dp->chain->tables[s->table_idx];
1467 error = table->iterate(table, &match_key, s->rq->out_port,
1468 &s->position, flow_stats_dump_callback, s);
1473 memset(&s->position, 0, sizeof s->position);
1475 *body_len = s->bytes_used;
1477 /* If error is 0, we're done.
1478 * Otherwise, if some bytes were used, there are more flows to come.
1479 * Otherwise, we were not able to fit even a single flow in the body,
1480 * which indicates that we have a single flow with too many actions to
1481 * fit. We won't ever make any progress at that rate, so give up. */
1482 return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1485 static void flow_stats_done(void *state)
1490 static int aggregate_stats_init(struct datapath *dp,
1491 const void *body, int body_len,
1494 *state = (void *)body;
1498 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1500 struct ofp_aggregate_stats_reply *rpy = private;
1501 rpy->packet_count += flow->packet_count;
1502 rpy->byte_count += flow->byte_count;
1507 static int aggregate_stats_dump(struct datapath *dp, void *state,
1508 void *body, int *body_len)
1510 struct ofp_aggregate_stats_request *rq = state;
1511 struct ofp_aggregate_stats_reply *rpy;
1512 struct sw_table_position position;
1513 struct sw_flow_key match_key;
1516 if (*body_len < sizeof *rpy)
1519 *body_len = sizeof *rpy;
1521 memset(rpy, 0, sizeof *rpy);
1523 flow_extract_match(&match_key, &rq->match);
1524 table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1525 memset(&position, 0, sizeof position);
1526 while (table_idx < dp->chain->n_tables
1527 && (rq->table_id == 0xff || rq->table_id == table_idx))
1529 struct sw_table *table = dp->chain->tables[table_idx];
1532 error = table->iterate(table, &match_key, rq->out_port, &position,
1533 aggregate_stats_dump_callback, rpy);
1538 memset(&position, 0, sizeof position);
1541 rpy->packet_count = cpu_to_be64(rpy->packet_count);
1542 rpy->byte_count = cpu_to_be64(rpy->byte_count);
1543 rpy->flow_count = htonl(rpy->flow_count);
1547 static int table_stats_dump(struct datapath *dp, void *state,
1548 void *body, int *body_len)
1550 struct ofp_table_stats *ots;
1551 int n_bytes = dp->chain->n_tables * sizeof *ots;
1553 if (n_bytes > *body_len)
1555 *body_len = n_bytes;
1556 for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1557 struct sw_table_stats stats;
1558 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1559 strncpy(ots->name, stats.name, sizeof ots->name);
1561 ots->wildcards = htonl(stats.wildcards);
1562 memset(ots->pad, 0, sizeof ots->pad);
1563 ots->max_entries = htonl(stats.max_flows);
1564 ots->active_count = htonl(stats.n_flows);
1565 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1566 ots->matched_count = cpu_to_be64(stats.n_matched);
1571 struct port_stats_state {
1575 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1578 struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1586 static int port_stats_dump(struct datapath *dp, void *state,
1587 void *body, int *body_len)
1589 struct port_stats_state *s = state;
1590 struct ofp_port_stats *ops;
1591 int n_ports, max_ports;
1594 max_ports = *body_len / sizeof *ops;
1600 for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1601 struct net_bridge_port *p = dp->ports[i];
1602 struct net_device_stats *stats;
1605 stats = p->dev->get_stats(p->dev);
1606 ops->port_no = htons(p->port_no);
1607 memset(ops->pad, 0, sizeof ops->pad);
1608 ops->rx_packets = cpu_to_be64(stats->rx_packets);
1609 ops->tx_packets = cpu_to_be64(stats->tx_packets);
1610 ops->rx_bytes = cpu_to_be64(stats->rx_bytes);
1611 ops->tx_bytes = cpu_to_be64(stats->tx_bytes);
1612 ops->rx_dropped = cpu_to_be64(stats->rx_dropped);
1613 ops->tx_dropped = cpu_to_be64(stats->tx_dropped);
1614 ops->rx_errors = cpu_to_be64(stats->rx_errors);
1615 ops->tx_errors = cpu_to_be64(stats->tx_errors);
1616 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1617 ops->rx_over_err = cpu_to_be64(stats->rx_over_errors);
1618 ops->rx_crc_err = cpu_to_be64(stats->rx_crc_errors);
1619 ops->collisions = cpu_to_be64(stats->collisions);
1624 *body_len = n_ports * sizeof *ops;
1625 return n_ports >= max_ports;
1628 static void port_stats_done(void *state)
1634 /* Minimum and maximum acceptable number of bytes in body member of
1635 * struct ofp_stats_request. */
1636 size_t min_body, max_body;
1638 /* Prepares to dump some kind of statistics on 'dp'. 'body' and
1639 * 'body_len' are the 'body' member of the struct ofp_stats_request.
1640 * Returns zero if successful, otherwise a negative error code.
1641 * May initialize '*state' to state information. May be null if no
1642 * initialization is required.*/
1643 int (*init)(struct datapath *dp, const void *body, int body_len,
1646 /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1647 * modifies '*body_len' to reflect the number of bytes actually used.
1648 * ('body' will be transmitted as the 'body' member of struct
1649 * ofp_stats_reply.) */
1650 int (*dump)(struct datapath *dp, void *state,
1651 void *body, int *body_len);
1653 /* Cleans any state created by the init or dump functions. May be null
1654 * if no cleanup is required. */
1655 void (*done)(void *state);
1658 static const struct stats_type stats[] = {
1667 sizeof(struct ofp_flow_stats_request),
1668 sizeof(struct ofp_flow_stats_request),
1673 [OFPST_AGGREGATE] = {
1674 sizeof(struct ofp_aggregate_stats_request),
1675 sizeof(struct ofp_aggregate_stats_request),
1676 aggregate_stats_init,
1677 aggregate_stats_dump,
1697 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1699 struct datapath *dp;
1700 struct sender sender;
1701 const struct stats_type *s;
1702 struct ofp_stats_reply *osr;
1704 int max_openflow_len, body_len;
1708 /* Set up the cleanup function for this dump. Linux 2.6.20 and later
1709 * support setting up cleanup functions via the .doneit member of
1710 * struct genl_ops. This kluge supports earlier versions also. */
1711 cb->done = dp_genl_openflow_done;
1713 sender.pid = NETLINK_CB(cb->skb).pid;
1714 sender.seq = cb->nlh->nlmsg_seq;
1716 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1717 struct ofp_stats_request *rq;
1719 size_t len, body_len;
1722 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1723 dp_genl_openflow_policy);
1727 if (!attrs[DP_GENL_A_DP_IDX])
1729 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1730 dp = dp_get(dp_idx);
1734 va = attrs[DP_GENL_A_OPENFLOW];
1736 if (!va || len < sizeof *rq)
1740 sender.xid = rq->header.xid;
1741 type = ntohs(rq->type);
1742 if (rq->header.version != OFP_VERSION) {
1743 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1744 OFPBRC_BAD_VERSION, rq, len);
1747 if (rq->header.type != OFPT_STATS_REQUEST
1748 || ntohs(rq->header.length) != len)
1751 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1752 dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1753 OFPBRC_BAD_STAT, rq, len);
1758 body_len = len - offsetof(struct ofp_stats_request, body);
1759 if (body_len < s->min_body || body_len > s->max_body)
1763 cb->args[1] = dp_idx;
1765 cb->args[3] = rq->header.xid;
1768 err = s->init(dp, rq->body, body_len, &state);
1771 cb->args[4] = (long) state;
1773 } else if (cb->args[0] == 1) {
1774 sender.xid = cb->args[3];
1775 dp_idx = cb->args[1];
1776 s = &stats[cb->args[2]];
1778 dp = dp_get(dp_idx);
1785 osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1788 return PTR_ERR(osr);
1789 osr->type = htons(s - stats);
1791 resize_openflow_skb(skb, &osr->header, max_openflow_len);
1793 body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1795 err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1800 osr->flags = ntohs(OFPSF_REPLY_MORE);
1801 resize_openflow_skb(skb, &osr->header,
1802 (offsetof(struct ofp_stats_reply, body)
1811 dp_genl_openflow_done(struct netlink_callback *cb)
1814 const struct stats_type *s = &stats[cb->args[2]];
1816 s->done((void *) cb->args[4]);
1821 static struct genl_ops dp_genl_ops_openflow = {
1822 .cmd = DP_GENL_C_OPENFLOW,
1823 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1824 .policy = dp_genl_openflow_policy,
1825 .doit = dp_genl_openflow,
1826 .dumpit = dp_genl_openflow_dumpit,
1829 static struct genl_ops *dp_genl_all_ops[] = {
1830 /* Keep this operation first. Generic Netlink dispatching
1831 * looks up operations with linear search, so we want it at the
1833 &dp_genl_ops_openflow,
1835 &dp_genl_ops_add_dp,
1836 &dp_genl_ops_del_dp,
1837 &dp_genl_ops_query_dp,
1838 &dp_genl_ops_add_port,
1839 &dp_genl_ops_del_port,
1842 static int dp_init_netlink(void)
1847 err = genl_register_family(&dp_genl_family);
1851 for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1852 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1854 goto err_unregister;
1857 strcpy(mc_group.name, "openflow");
1858 err = genl_register_mc_group(&dp_genl_family, &mc_group);
1860 goto err_unregister;
1865 genl_unregister_family(&dp_genl_family);
1869 static void dp_uninit_netlink(void)
1871 genl_unregister_family(&dp_genl_family);
1874 /* Set the description strings if appropriate values are available from
1876 static void set_desc(void)
1878 const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1879 const char *vendor = dmi_get_system_info(DMI_SYS_VENDOR);
1880 const char *name = dmi_get_system_info(DMI_PRODUCT_NAME);
1881 const char *version = dmi_get_system_info(DMI_PRODUCT_VERSION);
1882 const char *serial = dmi_get_system_info(DMI_PRODUCT_SERIAL);
1885 if (!uuid || *uuid == '\0' || strlen(uuid) != 36)
1888 /* We are only interested version 1 UUIDs, since the last six bytes
1889 * are an IEEE 802 MAC address. */
1890 if (uuid[14] != '1')
1893 /* Only set if the UUID is from Nicira. */
1895 if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1899 strlcpy(mfr_desc, vendor, sizeof(mfr_desc));
1900 if (name || version)
1901 snprintf(hw_desc, sizeof(hw_desc), "%s %s",
1903 version ? version : "");
1905 strlcpy(serial_num, serial, sizeof(serial_num));
1908 static int __init dp_init(void)
1912 printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1913 "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1919 err = register_netdevice_notifier(&dp_device_notifier);
1921 goto error_flow_exit;
1923 err = dp_init_netlink();
1925 goto error_unreg_notifier;
1927 /* Check if better descriptions of the switch are available than the
1931 /* Hook into callback used by the bridge to intercept packets.
1932 * Parasites we are. */
1933 if (br_handle_frame_hook)
1934 printk("openflow: hijacking bridge hook\n");
1935 br_handle_frame_hook = dp_frame_hook;
1939 error_unreg_notifier:
1940 unregister_netdevice_notifier(&dp_device_notifier);
1944 printk(KERN_EMERG "openflow: failed to install!");
1948 static void dp_cleanup(void)
1951 dp_uninit_netlink();
1952 unregister_netdevice_notifier(&dp_device_notifier);
1954 br_handle_frame_hook = NULL;
1957 module_init(dp_init);
1958 module_exit(dp_cleanup);
1960 MODULE_DESCRIPTION("OpenFlow switching datapath");
1961 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1962 MODULE_LICENSE("GPL");