When a packet arrives via bridging, clone it if it is shared.
[sliver-openvswitch.git] / datapath / datapath.c
1 /*
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
5  */
6
7 /* Functions for managing the dp interface/device. */
8
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>
14 #include <linux/in.h>
15 #include <net/genetlink.h>
16 #include <linux/ip.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>
35
36 #include "openflow/nicira-ext.h"
37 #include "openflow/openflow-netlink.h"
38 #include "datapath.h"
39 #include "nx_act_snat.h"
40 #include "table.h"
41 #include "chain.h"
42 #include "dp_dev.h"
43 #include "forward.h"
44 #include "flow.h"
45
46 #include "compat.h"
47
48
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";
55
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);
61 #else
62 MODULE_PARM(mfr_desc, "s");
63 MODULE_PARM(hw_desc, "s");
64 MODULE_PARM(sw_desc, "s");
65 MODULE_PARM(serial_num, "s");
66 #endif
67
68
69 /* Number of milliseconds between runs of the maintenance thread. */
70 #define MAINT_SLEEP_MSECS 1000
71
72 #define UINT32_MAX                        4294967295U
73 #define UINT16_MAX                        65535
74 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
75
76 static struct genl_family dp_genl_family;
77 static struct genl_multicast_group mc_group;
78
79 /* It's hard to imagine wanting more than one datapath, but... */
80 #define DP_MAX 32
81
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
85  * exclusion too.
86  *
87  * It is safe to access the datapath and net_bridge_port structures with just
88  * dp_mutex.
89  */
90 static struct datapath *dps[DP_MAX];
91 DEFINE_MUTEX(dp_mutex);
92 EXPORT_SYMBOL(dp_mutex);
93
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);
99
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
104  *
105  * Reduces amount of space reserved by a call to nla_reserve.
106  *
107  * No other attributes may be added between calling nla_reserve and this
108  * function, since it will create a hole in the message.
109  */
110 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
111 {
112         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
113         BUG_ON(delta > 0);
114         skb->tail += delta;
115         skb->len  += delta;
116         nla->nla_len = nla_attr_size(len);
117 }
118
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.
122  *
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
126  * of bytes or less.
127  *
128  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
129  * an error code. */
130 static void *
131 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
132                      const struct sender *sender, int *max_openflow_len)
133 {
134         struct ofp_header *oh;
135         struct nlattr *attr;
136         int openflow_len;
137
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);
151         BUG_ON(!attr);
152
153         /* Fill in the header.  The caller is responsible for the length. */
154         oh = nla_data(attr);
155         oh->version = OFP_VERSION;
156         oh->type = type;
157         oh->xid = sender ? sender->xid : 0;
158
159         return oh;
160 }
161
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
164  * necessary. */
165 static void
166 resize_openflow_skb(struct sk_buff *skb,
167                     struct ofp_header *oh, size_t new_length)
168 {
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);
173 }
174
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'. 
178  *
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
181  * use.  */
182 static void *
183 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
184                    const struct sender *sender, struct sk_buff **pskb) 
185 {
186         struct ofp_header *oh;
187         size_t genl_len;
188         struct sk_buff *skb;
189         int max_openflow_len;
190
191         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
192                 if (net_ratelimit())
193                         printk("alloc_openflow_skb: openflow message too large: %zu\n", 
194                                         openflow_len);
195                 return NULL;
196         }
197
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);
202         if (!skb) {
203                 if (net_ratelimit())
204                         printk("alloc_openflow_skb: genlmsg_new failed\n");
205                 return NULL;
206         }
207
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);
211
212         return oh;
213 }
214
215 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
216  * listeners. */
217 static int
218 send_openflow_skb(struct sk_buff *skb, const struct sender *sender) 
219 {
220         return (sender
221                 ? genlmsg_unicast(skb, sender->pid)
222                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
223 }
224
225 /* Retrieves the datapath id, which is the MAC address of the "of" device. */
226 static 
227 uint64_t get_datapath_id(struct net_device *dev)
228 {
229         uint64_t id = 0;
230         int i;
231
232         for (i=0; i<ETH_ALEN; i++) 
233                 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
234
235         return id;
236 }
237
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)
241 {
242         struct datapath *dp;
243         int err;
244
245         if (dp_idx < 0 || dp_idx >= DP_MAX)
246                 return -EINVAL;
247
248         if (!try_module_get(THIS_MODULE))
249                 return -ENODEV;
250
251         /* Exit early if a datapath with that number already exists. */
252         if (dps[dp_idx]) {
253                 err = -EEXIST;
254                 goto err_unlock;
255         }
256
257         err = -ENOMEM;
258         dp = kzalloc(sizeof *dp, GFP_KERNEL);
259         if (dp == NULL)
260                 goto err_unlock;
261
262         /* Setup our "of" device */
263         err = dp_dev_setup(dp);
264         if (err)
265                 goto err_free_dp;
266
267         dp->dp_idx = dp_idx;
268         dp->id = get_datapath_id(dp->netdev);
269         dp->chain = chain_create(dp);
270         if (dp->chain == NULL)
271                 goto err_destroy_dp_dev;
272         INIT_LIST_HEAD(&dp->port_list);
273
274         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
275         if (IS_ERR(dp->local_port)) {
276                 err = PTR_ERR(dp->local_port);
277                 goto err_destroy_local_port;
278         }
279
280         dp->flags = 0;
281         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
282
283         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
284         if (IS_ERR(dp->dp_task))
285                 goto err_destroy_chain;
286
287         dps[dp_idx] = dp;
288
289         return 0;
290
291 err_destroy_local_port:
292         dp_del_switch_port(dp->local_port);
293 err_destroy_chain:
294         chain_destroy(dp->chain);
295 err_destroy_dp_dev:
296         dp_dev_destroy(dp);
297 err_free_dp:
298         kfree(dp);
299 err_unlock:
300         module_put(THIS_MODULE);
301                 return err;
302 }
303
304 /* Find and return a free port number under 'dp'. */
305 static int find_portno(struct datapath *dp)
306 {
307         int i;
308         for (i = 0; i < DP_MAX_PORTS; i++)
309                 if (dp->ports[i] == NULL)
310                         return i;
311         return -EXFULL;
312 }
313
314 static struct net_bridge_port *new_nbp(struct datapath *dp,
315                                        struct net_device *dev, int port_no)
316 {
317         struct net_bridge_port *p;
318
319         if (dev->br_port != NULL)
320                 return ERR_PTR(-EBUSY);
321
322         p = kzalloc(sizeof(*p), GFP_KERNEL);
323         if (p == NULL)
324                 return ERR_PTR(-ENOMEM);
325
326         rtnl_lock();
327         dev_set_promiscuity(dev, 1);
328         rtnl_unlock();
329         dev_hold(dev);
330         p->dp = dp;
331         p->dev = dev;
332         p->port_no = port_no;
333         spin_lock_init(&p->lock);
334         INIT_WORK(&p->port_task, NULL);
335         if (port_no != OFPP_LOCAL)
336                 rcu_assign_pointer(dev->br_port, p);
337         if (port_no < DP_MAX_PORTS)
338                 rcu_assign_pointer(dp->ports[port_no], p); 
339         list_add_rcu(&p->node, &dp->port_list);
340
341         return p;
342 }
343
344 int add_switch_port(struct datapath *dp, struct net_device *dev)
345 {
346         struct net_bridge_port *p;
347         int port_no;
348
349         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
350             || is_dp_dev(dev))
351                 return -EINVAL;
352
353         port_no = find_portno(dp);
354         if (port_no < 0)
355                 return port_no;
356
357         p = new_nbp(dp, dev, port_no);
358         if (IS_ERR(p))
359                 return PTR_ERR(p);
360
361         init_port_status(p);
362
363         /* Notify the ctlpath that this port has been added */
364         dp_send_port_status(p, OFPPR_ADD);
365
366         return 0;
367 }
368
369 /* Delete 'p' from switch. */
370 int dp_del_switch_port(struct net_bridge_port *p)
371 {
372 #ifdef SUPPORT_SNAT
373         unsigned long flags;
374 #endif
375
376         /* First drop references to device. */
377         cancel_work_sync(&p->port_task);
378         rtnl_lock();
379         dev_set_promiscuity(p->dev, -1);
380         rtnl_unlock();
381         list_del_rcu(&p->node);
382         if (p->port_no != OFPP_LOCAL)
383                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
384         rcu_assign_pointer(p->dev->br_port, NULL);
385
386         /* Then wait until no one is still using it, and destroy it. */
387         synchronize_rcu();
388
389 #ifdef SUPPORT_SNAT
390         /* Free any SNAT configuration on the port. */
391         spin_lock_irqsave(&p->lock, flags);
392         snat_free_conf(p);
393         spin_unlock_irqrestore(&p->lock, flags);
394 #endif
395
396         /* Notify the ctlpath that this port no longer exists */
397         dp_send_port_status(p, OFPPR_DELETE);
398
399         dev_put(p->dev);
400         kfree(p);
401
402         return 0;
403 }
404
405 static void del_dp(struct datapath *dp)
406 {
407         struct net_bridge_port *p, *n;
408
409         kthread_stop(dp->dp_task);
410
411         /* Drop references to DP. */
412         list_for_each_entry_safe (p, n, &dp->port_list, node)
413                 dp_del_switch_port(p);
414         rcu_assign_pointer(dps[dp->dp_idx], NULL);
415
416         /* Kill off local_port dev references from buffered packets that have
417          * associated dst entries. */
418         synchronize_rcu();
419         fwd_discard_all();
420
421         /* Destroy dp->netdev.  (Must follow deleting switch ports since
422          * dp->local_port has a reference to it.) */
423         dp_dev_destroy(dp);
424
425         /* Wait until no longer in use, then destroy it. */
426         synchronize_rcu();
427         chain_destroy(dp->chain);
428         kfree(dp);
429         module_put(THIS_MODULE);
430 }
431
432 static int dp_maint_func(void *data)
433 {
434         struct datapath *dp = (struct datapath *) data;
435
436         while (!kthread_should_stop()) {
437 #ifdef SUPPORT_SNAT
438                 struct net_bridge_port *p;
439
440                 /* Expire old SNAT entries */
441                 rcu_read_lock();
442                 list_for_each_entry_rcu (p, &dp->port_list, node) 
443                         snat_maint(p);
444                 rcu_read_unlock();
445 #endif
446
447                 /* Timeout old entries */
448                 chain_timeout(dp->chain);
449                 msleep_interruptible(MAINT_SLEEP_MSECS);
450         }
451                 
452         return 0;
453 }
454
455 static void
456 do_port_input(struct net_bridge_port *p, struct sk_buff *skb) 
457 {
458         /* Make our own copy of the packet.  Otherwise we will mangle the
459          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
460          * (No one comes after us, since we tell handle_bridge() that we took
461          * the packet.) */
462         skb = skb_share_check(skb, GFP_ATOMIC);
463         if (!skb)
464                 return;
465
466 #ifdef SUPPORT_SNAT
467         /* Check if this packet needs early SNAT processing. */
468         if (snat_pre_route(skb)) {
469                 kfree_skb(skb);
470                 return;
471         }
472 #endif
473
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);
478 }
479
480 /*
481  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
482  * different set of devices!)
483  */
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,
487                                          struct sk_buff *skb)
488 {
489         do_port_input(p, skb);
490         return NULL;
491 }
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)
494 {
495         do_port_input(p, *pskb);
496         return 1;
497 }
498 #else
499 /* NB: This has only been tested on 2.4.35 */
500 static void dp_frame_hook(struct sk_buff *skb)
501 {
502         struct net_bridge_port *p = skb->dev->br_port;
503         if (p) {
504                 rcu_read_lock();
505                 do_port_input(p, skb);
506                 rcu_read_unlock();
507         } else
508                 kfree_skb(skb);
509 }
510 #endif
511
512 /* Forwarding output path.
513  * Based on net/bridge/br_forward.c. */
514
515 static inline unsigned packet_length(const struct sk_buff *skb)
516 {
517         int length = skb->len - ETH_HLEN;
518         if (skb->protocol == htons(ETH_P_8021Q))
519                 length -= VLAN_HLEN;
520         return length;
521 }
522
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.
525  */
526 static int
527 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
528 {
529         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
530         struct net_bridge_port *p;
531         int prev_port = -1;
532
533         list_for_each_entry_rcu (p, &dp->port_list, node) {
534                 if (skb->dev == p->dev || p->config & disable)
535                         continue;
536                 if (prev_port != -1) {
537                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
538                         if (!clone) {
539                                 kfree_skb(skb);
540                                 return -ENOMEM;
541                         }
542                         dp_output_port(dp, clone, prev_port, 0); 
543                 }
544                 prev_port = p->port_no;
545         }
546         if (prev_port != -1)
547                 dp_output_port(dp, skb, prev_port, 0);
548         else
549                 kfree_skb(skb);
550
551         return 0;
552 }
553
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,
557                            struct sk_buff *skb)
558 {
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
562              : NULL);
563         if (p) 
564                 skb->dev = p->dev;
565          else 
566                 skb->dev = NULL;
567 }
568
569 int 
570 dp_xmit_skb(struct sk_buff *skb)
571 {
572         int len = skb->len;
573         if (packet_length(skb) > skb->dev->mtu) {
574                 printk("dropped over-mtu packet: %d > %d\n",
575                            packet_length(skb), skb->dev->mtu);
576                 kfree_skb(skb);
577                 return -E2BIG;
578         }
579
580         dev_queue_xmit(skb);
581
582         return len;
583 }
584
585 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
586  */
587 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
588                    int ignore_no_fwd)
589 {
590         BUG_ON(!skb);
591         switch (out_port){
592         case OFPP_IN_PORT:
593                 /* Send it out the port it came in on, which is already set in
594                  * the skb. */
595                 if (!skb->dev) {
596                         if (net_ratelimit())
597                                 printk("skb device not set forwarding to in_port\n");
598                         kfree_skb(skb);
599                         return -ESRCH;
600                 }
601                 return dp_xmit_skb(skb);
602                 
603         case OFPP_TABLE: {
604                 int retval = run_flow_through_tables(dp->chain, skb,
605                                                      skb->dev->br_port);
606                 if (retval)
607                         kfree_skb(skb);
608                 return retval;
609         }
610
611         case OFPP_FLOOD:
612                 return output_all(dp, skb, 1);
613
614         case OFPP_ALL:
615                 return output_all(dp, skb, 0);
616
617         case OFPP_CONTROLLER:
618                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
619                                                   OFPR_ACTION);
620
621         case OFPP_LOCAL: {
622                 struct net_device *dev = dp->netdev;
623 #ifdef SUPPORT_SNAT
624                 snat_local_in(skb);
625 #endif
626                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
627         }
628
629         case 0 ... DP_MAX_PORTS - 1: {
630                 struct net_bridge_port *p = dp->ports[out_port];
631                 if (p == NULL)
632                         goto bad_port;
633                 if (p->dev == skb->dev) {
634                         /* To send to the input port, must use OFPP_IN_PORT */
635                         kfree_skb(skb);
636                         if (net_ratelimit())
637                                 printk("can't directly forward to input port\n");
638                         return -EINVAL;
639                 }
640                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
641                         kfree_skb(skb);
642                         return 0;
643                 }
644                 skb->dev = p->dev; 
645                 return dp_xmit_skb(skb);
646         }
647
648         default:
649                 goto bad_port;
650         }
651
652 bad_port:
653         kfree_skb(skb);
654         if (net_ratelimit())
655                 printk("can't forward to bad port %d\n", out_port);
656         return -ENOENT;
657 }
658
659 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
660  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
661  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
662  * sent. 'max_len' sets the maximum number of bytes that the caller
663  * wants to be sent; a value of 0 indicates the entire packet should be
664  * sent. */
665 int
666 dp_output_control(struct datapath *dp, struct sk_buff *skb,
667                            uint32_t buffer_id, size_t max_len, int reason)
668 {
669         /* FIXME?  Can we avoid creating a new skbuff in the case where we
670          * forward the whole packet? */
671         struct sk_buff *f_skb;
672         struct ofp_packet_in *opi;
673         struct net_bridge_port *p;
674         size_t fwd_len, opi_len;
675         int err;
676
677         fwd_len = skb->len;
678         if ((buffer_id != (uint32_t) -1) && max_len)
679                 fwd_len = min(fwd_len, max_len);
680
681         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
682         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
683         if (!opi) {
684                 err = -ENOMEM;
685                 goto out;
686         }
687         opi->buffer_id      = htonl(buffer_id);
688         opi->total_len      = htons(skb->len);
689         p = skb->dev->br_port;
690         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
691         opi->reason         = reason;
692         opi->pad            = 0;
693         memcpy(opi->data, skb_mac_header(skb), fwd_len);
694         err = send_openflow_skb(f_skb, NULL);
695
696 out:
697         kfree_skb(skb);
698         return err;
699 }
700
701 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
702 {
703         unsigned long flags;
704         desc->port_no = htons(p->port_no);
705         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
706         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
707         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
708         desc->curr = 0;
709         desc->supported = 0;
710         desc->advertised = 0;
711         desc->peer = 0;
712
713         spin_lock_irqsave(&p->lock, flags);
714         desc->config = htonl(p->config);
715         desc->state = htonl(p->state);
716         spin_unlock_irqrestore(&p->lock, flags);
717
718 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
719         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
720                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
721
722                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
723                         /* Set the supported features */
724                         if (ecmd.supported & SUPPORTED_10baseT_Half) 
725                                 desc->supported |= OFPPF_10MB_HD;
726                         if (ecmd.supported & SUPPORTED_10baseT_Full)
727                                 desc->supported |= OFPPF_10MB_FD;
728                         if (ecmd.supported & SUPPORTED_100baseT_Half) 
729                                 desc->supported |= OFPPF_100MB_HD;
730                         if (ecmd.supported & SUPPORTED_100baseT_Full)
731                                 desc->supported |= OFPPF_100MB_FD;
732                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
733                                 desc->supported |= OFPPF_1GB_HD;
734                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
735                                 desc->supported |= OFPPF_1GB_FD;
736                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
737                                 desc->supported |= OFPPF_10GB_FD;
738                         if (ecmd.supported & SUPPORTED_TP)
739                                 desc->supported |= OFPPF_COPPER;
740                         if (ecmd.supported & SUPPORTED_FIBRE)
741                                 desc->supported |= OFPPF_FIBER;
742                         if (ecmd.supported & SUPPORTED_Autoneg)
743                                 desc->supported |= OFPPF_AUTONEG;
744 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
745                         if (ecmd.supported & SUPPORTED_Pause)
746                                 desc->supported |= OFPPF_PAUSE;
747                         if (ecmd.supported & SUPPORTED_Asym_Pause)
748                                 desc->supported |= OFPPF_PAUSE_ASYM;
749 #endif /* kernel >= 2.6.14 */
750
751                         /* Set the advertised features */
752                         if (ecmd.advertising & ADVERTISED_10baseT_Half) 
753                                 desc->advertised |= OFPPF_10MB_HD;
754                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
755                                 desc->advertised |= OFPPF_10MB_FD;
756                         if (ecmd.advertising & ADVERTISED_100baseT_Half) 
757                                 desc->advertised |= OFPPF_100MB_HD;
758                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
759                                 desc->advertised |= OFPPF_100MB_FD;
760                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
761                                 desc->advertised |= OFPPF_1GB_HD;
762                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
763                                 desc->advertised |= OFPPF_1GB_FD;
764                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
765                                 desc->advertised |= OFPPF_10GB_FD;
766                         if (ecmd.advertising & ADVERTISED_TP)
767                                 desc->advertised |= OFPPF_COPPER;
768                         if (ecmd.advertising & ADVERTISED_FIBRE)
769                                 desc->advertised |= OFPPF_FIBER;
770                         if (ecmd.advertising & ADVERTISED_Autoneg)
771                                 desc->advertised |= OFPPF_AUTONEG;
772 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
773                         if (ecmd.advertising & ADVERTISED_Pause)
774                                 desc->advertised |= OFPPF_PAUSE;
775                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
776                                 desc->advertised |= OFPPF_PAUSE_ASYM;
777 #endif /* kernel >= 2.6.14 */
778
779                         /* Set the current features */
780                         if (ecmd.speed == SPEED_10)
781                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
782                         else if (ecmd.speed == SPEED_100)
783                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
784                         else if (ecmd.speed == SPEED_1000)
785                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
786                         else if (ecmd.speed == SPEED_10000)
787                                 desc->curr = OFPPF_10GB_FD;
788
789                         if (ecmd.port == PORT_TP) 
790                                 desc->curr |= OFPPF_COPPER;
791                         else if (ecmd.port == PORT_FIBRE) 
792                                 desc->curr |= OFPPF_FIBER;
793
794                         if (ecmd.autoneg)
795                                 desc->curr |= OFPPF_AUTONEG;
796                 }
797         }
798 #endif
799         desc->curr = htonl(desc->curr);
800         desc->supported = htonl(desc->supported);
801         desc->advertised = htonl(desc->advertised);
802         desc->peer = htonl(desc->peer);
803 }
804
805 static int 
806 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
807 {
808         struct net_bridge_port *p;
809         int port_count = 0;
810
811         ofr->datapath_id  = cpu_to_be64(dp->id); 
812
813         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
814         ofr->n_tables     = dp->chain->n_tables;
815         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
816         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
817         memset(ofr->pad, 0, sizeof ofr->pad);
818
819         list_for_each_entry_rcu (p, &dp->port_list, node) {
820                 fill_port_desc(p, &ofr->ports[port_count]);
821                 port_count++;
822         }
823
824         return port_count;
825 }
826
827 int
828 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
829 {
830         struct sk_buff *skb;
831         struct ofp_switch_features *ofr;
832         size_t ofr_len, port_max_len;
833         int port_count;
834
835         /* Overallocate. */
836         port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
837         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
838                                  OFPT_FEATURES_REPLY, sender, &skb);
839         if (!ofr)
840                 return -ENOMEM;
841
842         /* Fill. */
843         port_count = fill_features_reply(dp, ofr);
844
845         /* Shrink to fit. */
846         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
847         resize_openflow_skb(skb, &ofr->header, ofr_len);
848         return send_openflow_skb(skb, sender);
849 }
850
851 int
852 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
853 {
854         struct sk_buff *skb;
855         struct ofp_switch_config *osc;
856
857         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
858                                  &skb);
859         if (!osc)
860                 return -ENOMEM;
861
862         osc->flags = htons(dp->flags);
863         osc->miss_send_len = htons(dp->miss_send_len);
864
865         return send_openflow_skb(skb, sender);
866 }
867
868 int
869 dp_send_hello(struct datapath *dp, const struct sender *sender,
870               const struct ofp_header *request)
871 {
872         if (request->version < OFP_VERSION) {
873                 char err[64];
874                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
875                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
876                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
877                 return -EINVAL;
878         } else {
879                 struct sk_buff *skb;
880                 struct ofp_header *reply;
881
882                 reply = alloc_openflow_skb(dp, sizeof *reply,
883                                            OFPT_HELLO, sender, &skb);
884                 if (!reply)
885                         return -ENOMEM;
886
887                 return send_openflow_skb(skb, sender);
888         }
889 }
890
891 /* Callback function for a workqueue to disable an interface */
892 static void
893 down_port_cb(struct work_struct *work)
894 {
895         struct net_bridge_port *p = container_of(work, struct net_bridge_port, 
896                         port_task);
897
898         rtnl_lock();
899         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
900                 if (net_ratelimit())
901                         printk("problem bringing up port %s\n", p->dev->name);
902         rtnl_unlock();
903         p->config |= OFPPC_PORT_DOWN;
904 }
905
906 /* Callback function for a workqueue to enable an interface */
907 static void
908 up_port_cb(struct work_struct *work)
909 {
910         struct net_bridge_port *p = container_of(work, struct net_bridge_port, 
911                         port_task);
912
913         rtnl_lock();
914         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
915                 if (net_ratelimit())
916                         printk("problem bringing down port %s\n", p->dev->name);
917         rtnl_unlock();
918         p->config &= ~OFPPC_PORT_DOWN;
919 }
920
921 int
922 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
923 {
924         unsigned long int flags;
925         int port_no = ntohs(opm->port_no);
926         struct net_bridge_port *p;
927         p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
928              : port_no == OFPP_LOCAL ? dp->local_port
929              : NULL);
930
931         /* Make sure the port id hasn't changed since this was sent */
932         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
933                 return -1;
934
935         spin_lock_irqsave(&p->lock, flags);
936         if (opm->mask) {
937                 uint32_t config_mask = ntohl(opm->mask);
938                 p->config &= ~config_mask;
939                 p->config |= ntohl(opm->config) & config_mask;
940         }
941
942         /* Modifying the status of an interface requires taking a lock
943          * that cannot be done from here.  For this reason, we use a shared 
944          * workqueue, which will cause it to be executed from a safer 
945          * context. */
946         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
947                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
948                     && (p->config & OFPPC_PORT_DOWN) == 0) {
949                         PREPARE_WORK(&p->port_task, down_port_cb);
950                         schedule_work(&p->port_task);
951                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
952                            && (p->config & OFPPC_PORT_DOWN)) {
953                         PREPARE_WORK(&p->port_task, up_port_cb);
954                         schedule_work(&p->port_task);
955                 }
956         }
957         spin_unlock_irqrestore(&p->lock, flags);
958
959         return 0;
960 }
961
962 /* Initialize the port status field of the bridge port. */
963 static void
964 init_port_status(struct net_bridge_port *p)
965 {
966         unsigned long int flags;
967
968         spin_lock_irqsave(&p->lock, flags);
969
970         if (p->dev->flags & IFF_UP) 
971                 p->config &= ~OFPPC_PORT_DOWN;
972         else
973                 p->config |= OFPPC_PORT_DOWN;
974
975         if (netif_carrier_ok(p->dev))
976                 p->state &= ~OFPPS_LINK_DOWN;
977         else
978                 p->state |= OFPPS_LINK_DOWN;
979
980         spin_unlock_irqrestore(&p->lock, flags);
981 }
982
983 int
984 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
985 {
986         struct sk_buff *skb;
987         struct ofp_port_status *ops;
988
989         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
990                                  &skb);
991         if (!ops)
992                 return -ENOMEM;
993         ops->reason = status;
994         memset(ops->pad, 0, sizeof ops->pad);
995         fill_port_desc(p, &ops->desc);
996
997         return send_openflow_skb(skb, NULL);
998 }
999
1000 int 
1001 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1002                      enum ofp_flow_expired_reason reason)
1003 {
1004         struct sk_buff *skb;
1005         struct ofp_flow_expired *ofe;
1006
1007         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1008                 return 0;
1009
1010         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1011         if (!ofe)
1012                 return -ENOMEM;
1013
1014         flow_fill_match(&ofe->match, &flow->key);
1015
1016         ofe->priority = htons(flow->priority);
1017         ofe->reason = reason;
1018         memset(ofe->pad, 0, sizeof ofe->pad);
1019
1020         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
1021         memset(ofe->pad2, 0, sizeof ofe->pad2);
1022         ofe->packet_count = cpu_to_be64(flow->packet_count);
1023         ofe->byte_count   = cpu_to_be64(flow->byte_count);
1024
1025         return send_openflow_skb(skb, NULL);
1026 }
1027 EXPORT_SYMBOL(dp_send_flow_expired);
1028
1029 int
1030 dp_send_error_msg(struct datapath *dp, const struct sender *sender, 
1031                 uint16_t type, uint16_t code, const void *data, size_t len)
1032 {
1033         struct sk_buff *skb;
1034         struct ofp_error_msg *oem;
1035
1036
1037         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR, 
1038                         sender, &skb);
1039         if (!oem)
1040                 return -ENOMEM;
1041
1042         oem->type = htons(type);
1043         oem->code = htons(code);
1044         memcpy(oem->data, data, len);
1045
1046         return send_openflow_skb(skb, sender);
1047 }
1048
1049 int
1050 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1051                    const struct ofp_header *rq)
1052 {
1053         struct sk_buff *skb;
1054         struct ofp_header *reply;
1055
1056         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1057                                    sender, &skb);
1058         if (!reply)
1059                 return -ENOMEM;
1060
1061         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1062         return send_openflow_skb(skb, sender);
1063 }
1064
1065 /* Generic Netlink interface.
1066  *
1067  * See netlink(7) for an introduction to netlink.  See
1068  * http://linux-net.osdl.org/index.php/Netlink for more information and
1069  * pointers on how to work with netlink and Generic Netlink in the kernel and
1070  * in userspace. */
1071
1072 static struct genl_family dp_genl_family = {
1073         .id = GENL_ID_GENERATE,
1074         .hdrsize = 0,
1075         .name = DP_GENL_FAMILY_NAME,
1076         .version = 1,
1077         .maxattr = DP_GENL_A_MAX,
1078 };
1079
1080 /* Attribute policy: what each attribute may contain.  */
1081 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1082         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1083         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1084         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1085 };
1086
1087 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1088 {
1089         if (!info->attrs[DP_GENL_A_DP_IDX])
1090                 return -EINVAL;
1091
1092         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1093 }
1094
1095 static struct genl_ops dp_genl_ops_add_dp = {
1096         .cmd = DP_GENL_C_ADD_DP,
1097         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1098         .policy = dp_genl_policy,
1099         .doit = dp_genl_add,
1100         .dumpit = NULL,
1101 };
1102
1103 struct datapath *dp_get(int dp_idx)
1104 {
1105         if (dp_idx < 0 || dp_idx > DP_MAX)
1106                 return NULL;
1107         return rcu_dereference(dps[dp_idx]);
1108 }
1109
1110 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1111 {
1112         struct datapath *dp;
1113         int err;
1114
1115         if (!info->attrs[DP_GENL_A_DP_IDX])
1116                 return -EINVAL;
1117
1118         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1119         if (!dp)
1120                 err = -ENOENT;
1121         else {
1122                 del_dp(dp);
1123                 err = 0;
1124         }
1125         return err;
1126 }
1127
1128 static struct genl_ops dp_genl_ops_del_dp = {
1129         .cmd = DP_GENL_C_DEL_DP,
1130         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1131         .policy = dp_genl_policy,
1132         .doit = dp_genl_del,
1133         .dumpit = NULL,
1134 };
1135
1136 /* Queries a datapath for related information.  Currently the only relevant
1137  * information is the datapath's multicast group ID.  Really we want one
1138  * multicast group per datapath, but because of locking issues[*] we can't
1139  * easily get one.  Thus, every datapath will currently return the same
1140  * global multicast group ID, but in the future it would be nice to fix that.
1141  *
1142  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1143  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1144  *       group ID, also acquires genl_lock, thus deadlock.
1145  */
1146 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1147 {
1148         struct datapath *dp;
1149         struct sk_buff *ans_skb = NULL;
1150         int dp_idx;
1151         int err = -ENOMEM;
1152
1153         if (!info->attrs[DP_GENL_A_DP_IDX])
1154                 return -EINVAL;
1155
1156         rcu_read_lock();
1157         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1158         dp = dp_get(dp_idx);
1159         if (!dp)
1160                 err = -ENOENT;
1161         else {
1162                 void *data;
1163                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1164                 if (!ans_skb) {
1165                         err = -ENOMEM;
1166                         goto err;
1167                 }
1168                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1169                                          0, DP_GENL_C_QUERY_DP);
1170                 if (data == NULL) {
1171                         err = -ENOMEM;
1172                         goto err;
1173                 }
1174                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1175                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1176
1177                 genlmsg_end(ans_skb, data);
1178                 err = genlmsg_reply(ans_skb, info);
1179                 if (!err)
1180                         ans_skb = NULL;
1181         }
1182 err:
1183 nla_put_failure:
1184         if (ans_skb)
1185                 kfree_skb(ans_skb);
1186         rcu_read_unlock();
1187         return err;
1188 }
1189
1190 static struct genl_ops dp_genl_ops_query_dp = {
1191         .cmd = DP_GENL_C_QUERY_DP,
1192         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1193         .policy = dp_genl_policy,
1194         .doit = dp_genl_query,
1195         .dumpit = NULL,
1196 };
1197
1198 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1199 {
1200         struct datapath *dp;
1201         struct net_device *port;
1202         int err;
1203
1204         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1205                 return -EINVAL;
1206
1207         /* Get datapath. */
1208         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1209         if (!dp) {
1210                 err = -ENOENT;
1211                 goto out;
1212         }
1213
1214         /* Get interface to add/remove. */
1215         port = dev_get_by_name(&init_net, 
1216                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1217         if (!port) {
1218                 err = -ENOENT;
1219                 goto out;
1220         }
1221
1222         /* Execute operation. */
1223         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1224                 err = add_switch_port(dp, port);
1225         else {
1226                 if (port->br_port == NULL || port->br_port->dp != dp) {
1227                         err = -ENOENT;
1228                         goto out_put;
1229                 }
1230                 err = dp_del_switch_port(port->br_port);
1231         }
1232
1233 out_put:
1234         dev_put(port);
1235 out:
1236         return err;
1237 }
1238
1239 static struct genl_ops dp_genl_ops_add_port = {
1240         .cmd = DP_GENL_C_ADD_PORT,
1241         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1242         .policy = dp_genl_policy,
1243         .doit = dp_genl_add_del_port,
1244         .dumpit = NULL,
1245 };
1246
1247 static struct genl_ops dp_genl_ops_del_port = {
1248         .cmd = DP_GENL_C_DEL_PORT,
1249         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1250         .policy = dp_genl_policy,
1251         .doit = dp_genl_add_del_port,
1252         .dumpit = NULL,
1253 };
1254
1255 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1256 {
1257         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1258         struct datapath *dp;
1259         struct ofp_header *oh;
1260         struct sender sender;
1261         int err;
1262
1263         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1264                 return -EINVAL;
1265
1266         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1267         if (!dp)
1268                 return -ENOENT;
1269
1270         if (nla_len(va) < sizeof(struct ofp_header))
1271                 return -EINVAL;
1272         oh = nla_data(va);
1273
1274         sender.xid = oh->xid;
1275         sender.pid = info->snd_pid;
1276         sender.seq = info->snd_seq;
1277
1278         mutex_lock(&dp_mutex);
1279         err = fwd_control_input(dp->chain, &sender,
1280                                 nla_data(va), nla_len(va));
1281         mutex_unlock(&dp_mutex);
1282         return err;
1283 }
1284
1285 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1286         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1287 };
1288
1289 static int desc_stats_dump(struct datapath *dp, void *state,
1290                             void *body, int *body_len)
1291 {
1292         struct ofp_desc_stats *ods = body;
1293         int n_bytes = sizeof *ods;
1294
1295         if (n_bytes > *body_len) {
1296                 return -ENOBUFS;
1297         }
1298         *body_len = n_bytes;
1299
1300         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1301         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1302         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1303         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1304
1305         return 0;
1306 }
1307
1308 struct flow_stats_state {
1309         int table_idx;
1310         struct sw_table_position position;
1311         const struct ofp_flow_stats_request *rq;
1312
1313         void *body;
1314         int bytes_used, bytes_allocated;
1315 };
1316
1317 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1318                            void **state)
1319 {
1320         const struct ofp_flow_stats_request *fsr = body;
1321         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1322         if (!s)
1323                 return -ENOMEM;
1324         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1325         memset(&s->position, 0, sizeof s->position);
1326         s->rq = fsr;
1327         *state = s;
1328         return 0;
1329 }
1330
1331 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1332 {
1333         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1334         struct flow_stats_state *s = private;
1335         struct ofp_flow_stats *ofs;
1336         int length;
1337
1338         length = sizeof *ofs + sf_acts->actions_len;
1339         if (length + s->bytes_used > s->bytes_allocated)
1340                 return 1;
1341
1342         ofs = s->body + s->bytes_used;
1343         ofs->length          = htons(length);
1344         ofs->table_id        = s->table_idx;
1345         ofs->pad             = 0;
1346         ofs->match.wildcards = htonl(flow->key.wildcards);
1347         ofs->match.in_port   = flow->key.in_port;
1348         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1349         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1350         ofs->match.dl_vlan   = flow->key.dl_vlan;
1351         ofs->match.dl_type   = flow->key.dl_type;
1352         ofs->match.nw_src    = flow->key.nw_src;
1353         ofs->match.nw_dst    = flow->key.nw_dst;
1354         ofs->match.nw_proto  = flow->key.nw_proto;
1355         ofs->match.pad       = 0;
1356         ofs->match.tp_src    = flow->key.tp_src;
1357         ofs->match.tp_dst    = flow->key.tp_dst;
1358         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1359         ofs->priority        = htons(flow->priority);
1360         ofs->idle_timeout    = htons(flow->idle_timeout);
1361         ofs->hard_timeout    = htons(flow->hard_timeout);
1362         memset(ofs->pad2, 0, sizeof ofs->pad2);
1363         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1364         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1365         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1366
1367         s->bytes_used += length;
1368         return 0;
1369 }
1370
1371 static int flow_stats_dump(struct datapath *dp, void *state,
1372                            void *body, int *body_len)
1373 {
1374         struct flow_stats_state *s = state;
1375         struct sw_flow_key match_key;
1376         int error = 0;
1377
1378         s->bytes_used = 0;
1379         s->bytes_allocated = *body_len;
1380         s->body = body;
1381
1382         flow_extract_match(&match_key, &s->rq->match);
1383         while (s->table_idx < dp->chain->n_tables
1384                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1385         {
1386                 struct sw_table *table = dp->chain->tables[s->table_idx];
1387
1388                 error = table->iterate(table, &match_key, &s->position,
1389                                        flow_stats_dump_callback, s);
1390                 if (error)
1391                         break;
1392
1393                 s->table_idx++;
1394                 memset(&s->position, 0, sizeof s->position);
1395         }
1396         *body_len = s->bytes_used;
1397
1398         /* If error is 0, we're done.
1399          * Otherwise, if some bytes were used, there are more flows to come.
1400          * Otherwise, we were not able to fit even a single flow in the body,
1401          * which indicates that we have a single flow with too many actions to
1402          * fit.  We won't ever make any progress at that rate, so give up. */
1403         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1404 }
1405
1406 static void flow_stats_done(void *state)
1407 {
1408         kfree(state);
1409 }
1410
1411 static int aggregate_stats_init(struct datapath *dp,
1412                                 const void *body, int body_len,
1413                                 void **state)
1414 {
1415         *state = (void *)body;
1416         return 0;
1417 }
1418
1419 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1420 {
1421         struct ofp_aggregate_stats_reply *rpy = private;
1422         rpy->packet_count += flow->packet_count;
1423         rpy->byte_count += flow->byte_count;
1424         rpy->flow_count++;
1425         return 0;
1426 }
1427
1428 static int aggregate_stats_dump(struct datapath *dp, void *state,
1429                                 void *body, int *body_len)
1430 {
1431         struct ofp_aggregate_stats_request *rq = state;
1432         struct ofp_aggregate_stats_reply *rpy;
1433         struct sw_table_position position;
1434         struct sw_flow_key match_key;
1435         int table_idx;
1436
1437         if (*body_len < sizeof *rpy)
1438                 return -ENOBUFS;
1439         rpy = body;
1440         *body_len = sizeof *rpy;
1441
1442         memset(rpy, 0, sizeof *rpy);
1443
1444         flow_extract_match(&match_key, &rq->match);
1445         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1446         memset(&position, 0, sizeof position);
1447         while (table_idx < dp->chain->n_tables
1448                && (rq->table_id == 0xff || rq->table_id == table_idx))
1449         {
1450                 struct sw_table *table = dp->chain->tables[table_idx];
1451                 int error;
1452
1453                 error = table->iterate(table, &match_key, &position,
1454                                        aggregate_stats_dump_callback, rpy);
1455                 if (error)
1456                         return error;
1457
1458                 table_idx++;
1459                 memset(&position, 0, sizeof position);
1460         }
1461
1462         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1463         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1464         rpy->flow_count = htonl(rpy->flow_count);
1465         return 0;
1466 }
1467
1468 static int table_stats_dump(struct datapath *dp, void *state,
1469                             void *body, int *body_len)
1470 {
1471         struct ofp_table_stats *ots;
1472         int n_bytes = dp->chain->n_tables * sizeof *ots;
1473         int i;
1474         if (n_bytes > *body_len)
1475                 return -ENOBUFS;
1476         *body_len = n_bytes;
1477         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1478                 struct sw_table_stats stats;
1479                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1480                 strncpy(ots->name, stats.name, sizeof ots->name);
1481                 ots->table_id = i;
1482                 ots->wildcards = htonl(stats.wildcards);
1483                 memset(ots->pad, 0, sizeof ots->pad);
1484                 ots->max_entries = htonl(stats.max_flows);
1485                 ots->active_count = htonl(stats.n_flows);
1486                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1487                 ots->matched_count = cpu_to_be64(stats.n_matched);
1488         }
1489         return 0;
1490 }
1491
1492 struct port_stats_state {
1493         int port;
1494 };
1495
1496 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1497                            void **state)
1498 {
1499         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1500         if (!s)
1501                 return -ENOMEM;
1502         s->port = 0;
1503         *state = s;
1504         return 0;
1505 }
1506
1507 static int port_stats_dump(struct datapath *dp, void *state,
1508                            void *body, int *body_len)
1509 {
1510         struct port_stats_state *s = state;
1511         struct ofp_port_stats *ops;
1512         int n_ports, max_ports;
1513         int i;
1514
1515         max_ports = *body_len / sizeof *ops;
1516         if (!max_ports)
1517                 return -ENOMEM;
1518         ops = body;
1519
1520         n_ports = 0;
1521         for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1522                 struct net_bridge_port *p = dp->ports[i];
1523                 struct net_device_stats *stats;
1524                 if (!p)
1525                         continue;
1526                 stats = p->dev->get_stats(p->dev);
1527                 ops->port_no = htons(p->port_no);
1528                 memset(ops->pad, 0, sizeof ops->pad);
1529                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1530                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1531                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1532                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1533                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1534                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1535                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1536                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1537                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1538                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1539                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1540                 ops->collisions   = cpu_to_be64(stats->collisions);
1541                 n_ports++;
1542                 ops++;
1543         }
1544         s->port = i;
1545         *body_len = n_ports * sizeof *ops;
1546         return n_ports >= max_ports;
1547 }
1548
1549 static void port_stats_done(void *state)
1550 {
1551         kfree(state);
1552 }
1553
1554 struct stats_type {
1555         /* Minimum and maximum acceptable number of bytes in body member of
1556          * struct ofp_stats_request. */
1557         size_t min_body, max_body;
1558
1559         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1560          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1561          * Returns zero if successful, otherwise a negative error code.
1562          * May initialize '*state' to state information.  May be null if no
1563          * initialization is required.*/
1564         int (*init)(struct datapath *dp, const void *body, int body_len,
1565                     void **state);
1566
1567         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1568          * modifies '*body_len' to reflect the number of bytes actually used.
1569          * ('body' will be transmitted as the 'body' member of struct
1570          * ofp_stats_reply.) */
1571         int (*dump)(struct datapath *dp, void *state,
1572                     void *body, int *body_len);
1573
1574         /* Cleans any state created by the init or dump functions.  May be null
1575          * if no cleanup is required. */
1576         void (*done)(void *state);
1577 };
1578
1579 static const struct stats_type stats[] = {
1580         [OFPST_DESC] = {
1581                 0,
1582                 0,
1583                 NULL,
1584                 desc_stats_dump,
1585                 NULL
1586         },
1587         [OFPST_FLOW] = {
1588                 sizeof(struct ofp_flow_stats_request),
1589                 sizeof(struct ofp_flow_stats_request),
1590                 flow_stats_init,
1591                 flow_stats_dump,
1592                 flow_stats_done
1593         },
1594         [OFPST_AGGREGATE] = {
1595                 sizeof(struct ofp_aggregate_stats_request),
1596                 sizeof(struct ofp_aggregate_stats_request),
1597                 aggregate_stats_init,
1598                 aggregate_stats_dump,
1599                 NULL
1600         },
1601         [OFPST_TABLE] = {
1602                 0,
1603                 0,
1604                 NULL,
1605                 table_stats_dump,
1606                 NULL
1607         },
1608         [OFPST_PORT] = {
1609                 0,
1610                 0,
1611                 port_stats_init,
1612                 port_stats_dump,
1613                 port_stats_done
1614         },
1615 };
1616
1617 static int
1618 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1619 {
1620         struct datapath *dp;
1621         struct sender sender;
1622         const struct stats_type *s;
1623         struct ofp_stats_reply *osr;
1624         int dp_idx;
1625         int max_openflow_len, body_len;
1626         void *body;
1627         int err;
1628
1629         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1630          * support setting up cleanup functions via the .doneit member of
1631          * struct genl_ops.  This kluge supports earlier versions also. */
1632         cb->done = dp_genl_openflow_done;
1633
1634         sender.pid = NETLINK_CB(cb->skb).pid;
1635         sender.seq = cb->nlh->nlmsg_seq;
1636         if (!cb->args[0]) {
1637                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1638                 struct ofp_stats_request *rq;
1639                 struct nlattr *va;
1640                 size_t len, body_len;
1641                 int type;
1642
1643                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1644                                   dp_genl_openflow_policy);
1645                 if (err < 0)
1646                         return err;
1647
1648                 if (!attrs[DP_GENL_A_DP_IDX])
1649                         return -EINVAL;
1650                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1651                 dp = dp_get(dp_idx);
1652                 if (!dp)
1653                         return -ENOENT;
1654
1655                 va = attrs[DP_GENL_A_OPENFLOW];
1656                 len = nla_len(va);
1657                 if (!va || len < sizeof *rq)
1658                         return -EINVAL;
1659
1660                 rq = nla_data(va);
1661                 sender.xid = rq->header.xid;
1662                 type = ntohs(rq->type);
1663                 if (rq->header.version != OFP_VERSION) {
1664                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1665                                           OFPBRC_BAD_VERSION, rq, len);
1666                         return -EINVAL;
1667                 }
1668                 if (rq->header.type != OFPT_STATS_REQUEST
1669                     || ntohs(rq->header.length) != len)
1670                         return -EINVAL;
1671
1672                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1673                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1674                                           OFPBRC_BAD_STAT, rq, len);
1675                         return -EINVAL;
1676                 }
1677
1678                 s = &stats[type];
1679                 body_len = len - offsetof(struct ofp_stats_request, body);
1680                 if (body_len < s->min_body || body_len > s->max_body)
1681                         return -EINVAL;
1682
1683                 cb->args[0] = 1;
1684                 cb->args[1] = dp_idx;
1685                 cb->args[2] = type;
1686                 cb->args[3] = rq->header.xid;
1687                 if (s->init) {
1688                         void *state;
1689                         err = s->init(dp, rq->body, body_len, &state);
1690                         if (err)
1691                                 return err;
1692                         cb->args[4] = (long) state;
1693                 }
1694         } else if (cb->args[0] == 1) {
1695                 sender.xid = cb->args[3];
1696                 dp_idx = cb->args[1];
1697                 s = &stats[cb->args[2]];
1698
1699                 dp = dp_get(dp_idx);
1700                 if (!dp)
1701                         return -ENOENT;
1702         } else {
1703                 return 0;
1704         }
1705
1706         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1707                                    &max_openflow_len);
1708         if (IS_ERR(osr))
1709                 return PTR_ERR(osr);
1710         osr->type = htons(s - stats);
1711         osr->flags = 0;
1712         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1713         body = osr->body;
1714         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1715
1716         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1717         if (err >= 0) {
1718                 if (!err)
1719                         cb->args[0] = 2;
1720                 else
1721                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1722                 resize_openflow_skb(skb, &osr->header,
1723                                     (offsetof(struct ofp_stats_reply, body)
1724                                      + body_len));
1725                 err = skb->len;
1726         }
1727
1728         return err;
1729 }
1730
1731 static int
1732 dp_genl_openflow_done(struct netlink_callback *cb)
1733 {
1734         if (cb->args[0]) {
1735                 const struct stats_type *s = &stats[cb->args[2]];
1736                 if (s->done)
1737                         s->done((void *) cb->args[4]);
1738         }
1739         return 0;
1740 }
1741
1742 static struct genl_ops dp_genl_ops_openflow = {
1743         .cmd = DP_GENL_C_OPENFLOW,
1744         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1745         .policy = dp_genl_openflow_policy,
1746         .doit = dp_genl_openflow,
1747         .dumpit = dp_genl_openflow_dumpit,
1748 };
1749
1750 static struct genl_ops *dp_genl_all_ops[] = {
1751         /* Keep this operation first.  Generic Netlink dispatching
1752          * looks up operations with linear search, so we want it at the
1753          * front. */
1754         &dp_genl_ops_openflow,
1755
1756         &dp_genl_ops_add_dp,
1757         &dp_genl_ops_del_dp,
1758         &dp_genl_ops_query_dp,
1759         &dp_genl_ops_add_port,
1760         &dp_genl_ops_del_port,
1761 };
1762
1763 static int dp_init_netlink(void)
1764 {
1765         int err;
1766         int i;
1767
1768         err = genl_register_family(&dp_genl_family);
1769         if (err)
1770                 return err;
1771
1772         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1773                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1774                 if (err)
1775                         goto err_unregister;
1776         }
1777
1778         strcpy(mc_group.name, "openflow");
1779         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1780         if (err < 0)
1781                 goto err_unregister;
1782
1783         return 0;
1784
1785 err_unregister:
1786         genl_unregister_family(&dp_genl_family);
1787                 return err;
1788 }
1789
1790 static void dp_uninit_netlink(void)
1791 {
1792         genl_unregister_family(&dp_genl_family);
1793 }
1794
1795 /* Set the description strings if appropriate values are available from
1796  * the DMI. */
1797 static void set_desc(void)
1798 {
1799         const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1800         const char *uptr = uuid + 24;
1801
1802         if (!uuid || *uuid == '\0' || strlen(uuid) != 36) 
1803                 return;
1804
1805         /* We are only interested version 1 UUIDs, since the last six bytes
1806          * are an IEEE 802 MAC address. */
1807         if (uuid[14] != '1') 
1808                 return;
1809
1810         /* Only set if the UUID is from Nicira. */
1811         if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1812                 return;
1813
1814         strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
1815         snprintf(hw_desc, sizeof(hw_desc), "%s %s", 
1816                         dmi_get_system_info(DMI_PRODUCT_NAME), 
1817                         dmi_get_system_info(DMI_PRODUCT_VERSION));
1818         strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL), 
1819                         sizeof(serial_num));
1820 }
1821
1822 static int __init dp_init(void)
1823 {
1824         int err;
1825
1826         printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1827                "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1828
1829         err = flow_init();
1830         if (err)
1831                 goto error;
1832
1833         err = register_netdevice_notifier(&dp_device_notifier);
1834         if (err)
1835                 goto error_flow_exit;
1836
1837         err = dp_init_netlink();
1838         if (err)
1839                 goto error_unreg_notifier;
1840
1841         /* Check if better descriptions of the switch are available than the
1842          * defaults. */
1843         set_desc();
1844
1845         /* Hook into callback used by the bridge to intercept packets.
1846          * Parasites we are. */
1847         if (br_handle_frame_hook)
1848                 printk("openflow: hijacking bridge hook\n");
1849         br_handle_frame_hook = dp_frame_hook;
1850
1851         return 0;
1852
1853 error_unreg_notifier:
1854         unregister_netdevice_notifier(&dp_device_notifier);
1855 error_flow_exit:
1856         flow_exit();
1857 error:
1858         printk(KERN_EMERG "openflow: failed to install!");
1859         return err;
1860 }
1861
1862 static void dp_cleanup(void)
1863 {
1864         fwd_exit();
1865         dp_uninit_netlink();
1866         unregister_netdevice_notifier(&dp_device_notifier);
1867         flow_exit();
1868         br_handle_frame_hook = NULL;
1869 }
1870
1871 module_init(dp_init);
1872 module_exit(dp_cleanup);
1873
1874 MODULE_DESCRIPTION("OpenFlow switching datapath");
1875 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1876 MODULE_LICENSE("GPL");