xref: /openbmc/linux/net/ethernet/eth.c (revision 827634ad)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Ethernet-type device handling.
7  *
8  * Version:	@(#)eth.c	1.0.7	05/25/93
9  *
10  * Authors:	Ross Biro
11  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
13  *		Florian  La Roche, <rzsfl@rz.uni-sb.de>
14  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
15  *
16  * Fixes:
17  *		Mr Linux	: Arp problems
18  *		Alan Cox	: Generic queue tidyup (very tiny here)
19  *		Alan Cox	: eth_header ntohs should be htons
20  *		Alan Cox	: eth_rebuild_header missing an htons and
21  *				  minor other things.
22  *		Tegge		: Arp bug fixes.
23  *		Florian		: Removed many unnecessary functions, code cleanup
24  *				  and changes for new arp and skbuff.
25  *		Alan Cox	: Redid header building to reflect new format.
26  *		Alan Cox	: ARP only when compiled with CONFIG_INET
27  *		Greg Page	: 802.2 and SNAP stuff.
28  *		Alan Cox	: MAC layer pointers/new format.
29  *		Paul Gortmaker	: eth_copy_and_sum shouldn't csum padding.
30  *		Alan Cox	: Protect against forwarding explosions with
31  *				  older network drivers and IFF_ALLMULTI.
32  *	Christer Weinigel	: Better rebuild header message.
33  *             Andrew Morton    : 26Feb01: kill ether_setup() - use netdev_boot_setup().
34  *
35  *		This program is free software; you can redistribute it and/or
36  *		modify it under the terms of the GNU General Public License
37  *		as published by the Free Software Foundation; either version
38  *		2 of the License, or (at your option) any later version.
39  */
40 #include <linux/module.h>
41 #include <linux/types.h>
42 #include <linux/kernel.h>
43 #include <linux/string.h>
44 #include <linux/mm.h>
45 #include <linux/socket.h>
46 #include <linux/in.h>
47 #include <linux/inet.h>
48 #include <linux/ip.h>
49 #include <linux/netdevice.h>
50 #include <linux/etherdevice.h>
51 #include <linux/skbuff.h>
52 #include <linux/errno.h>
53 #include <linux/init.h>
54 #include <linux/if_ether.h>
55 #include <net/dst.h>
56 #include <net/arp.h>
57 #include <net/sock.h>
58 #include <net/ipv6.h>
59 #include <net/ip.h>
60 #include <net/dsa.h>
61 #include <linux/uaccess.h>
62 
63 __setup("ether=", netdev_boot_setup);
64 
65 /**
66  * eth_header - create the Ethernet header
67  * @skb:	buffer to alter
68  * @dev:	source device
69  * @type:	Ethernet type field
70  * @daddr: destination address (NULL leave destination address)
71  * @saddr: source address (NULL use device source address)
72  * @len:   packet length (<= skb->len)
73  *
74  *
75  * Set the protocol type. For a packet of type ETH_P_802_3/2 we put the length
76  * in here instead.
77  */
78 int eth_header(struct sk_buff *skb, struct net_device *dev,
79 	       unsigned short type,
80 	       const void *daddr, const void *saddr, unsigned int len)
81 {
82 	struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
83 
84 	if (type != ETH_P_802_3 && type != ETH_P_802_2)
85 		eth->h_proto = htons(type);
86 	else
87 		eth->h_proto = htons(len);
88 
89 	/*
90 	 *      Set the source hardware address.
91 	 */
92 
93 	if (!saddr)
94 		saddr = dev->dev_addr;
95 	memcpy(eth->h_source, saddr, ETH_ALEN);
96 
97 	if (daddr) {
98 		memcpy(eth->h_dest, daddr, ETH_ALEN);
99 		return ETH_HLEN;
100 	}
101 
102 	/*
103 	 *      Anyway, the loopback-device should never use this function...
104 	 */
105 
106 	if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) {
107 		eth_zero_addr(eth->h_dest);
108 		return ETH_HLEN;
109 	}
110 
111 	return -ETH_HLEN;
112 }
113 EXPORT_SYMBOL(eth_header);
114 
115 /**
116  * eth_get_headlen - determine the the length of header for an ethernet frame
117  * @data: pointer to start of frame
118  * @len: total length of frame
119  *
120  * Make a best effort attempt to pull the length for all of the headers for
121  * a given frame in a linear buffer.
122  */
123 u32 eth_get_headlen(void *data, unsigned int len)
124 {
125 	const struct ethhdr *eth = (const struct ethhdr *)data;
126 	struct flow_keys keys;
127 
128 	/* this should never happen, but better safe than sorry */
129 	if (len < sizeof(*eth))
130 		return len;
131 
132 	/* parse any remaining L2/L3 headers, check for L4 */
133 	if (!__skb_flow_dissect(NULL, &keys, data,
134 				eth->h_proto, sizeof(*eth), len))
135 		return max_t(u32, keys.thoff, sizeof(*eth));
136 
137 	/* parse for any L4 headers */
138 	return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len);
139 }
140 EXPORT_SYMBOL(eth_get_headlen);
141 
142 /**
143  * eth_type_trans - determine the packet's protocol ID.
144  * @skb: received socket data
145  * @dev: receiving network device
146  *
147  * The rule here is that we
148  * assume 802.3 if the type field is short enough to be a length.
149  * This is normal practice and works for any 'now in use' protocol.
150  */
151 __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev)
152 {
153 	unsigned short _service_access_point;
154 	const unsigned short *sap;
155 	const struct ethhdr *eth;
156 
157 	skb->dev = dev;
158 	skb_reset_mac_header(skb);
159 	skb_pull_inline(skb, ETH_HLEN);
160 	eth = eth_hdr(skb);
161 
162 	if (unlikely(is_multicast_ether_addr(eth->h_dest))) {
163 		if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
164 			skb->pkt_type = PACKET_BROADCAST;
165 		else
166 			skb->pkt_type = PACKET_MULTICAST;
167 	}
168 	else if (unlikely(!ether_addr_equal_64bits(eth->h_dest,
169 						   dev->dev_addr)))
170 		skb->pkt_type = PACKET_OTHERHOST;
171 
172 	/*
173 	 * Some variants of DSA tagging don't have an ethertype field
174 	 * at all, so we check here whether one of those tagging
175 	 * variants has been configured on the receiving interface,
176 	 * and if so, set skb->protocol without looking at the packet.
177 	 */
178 	if (unlikely(netdev_uses_dsa(dev)))
179 		return htons(ETH_P_XDSA);
180 
181 	if (likely(ntohs(eth->h_proto) >= ETH_P_802_3_MIN))
182 		return eth->h_proto;
183 
184 	/*
185 	 *      This is a magic hack to spot IPX packets. Older Novell breaks
186 	 *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
187 	 *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
188 	 *      won't work for fault tolerant netware but does for the rest.
189 	 */
190 	sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
191 	if (sap && *sap == 0xFFFF)
192 		return htons(ETH_P_802_3);
193 
194 	/*
195 	 *      Real 802.2 LLC
196 	 */
197 	return htons(ETH_P_802_2);
198 }
199 EXPORT_SYMBOL(eth_type_trans);
200 
201 /**
202  * eth_header_parse - extract hardware address from packet
203  * @skb: packet to extract header from
204  * @haddr: destination buffer
205  */
206 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
207 {
208 	const struct ethhdr *eth = eth_hdr(skb);
209 	memcpy(haddr, eth->h_source, ETH_ALEN);
210 	return ETH_ALEN;
211 }
212 EXPORT_SYMBOL(eth_header_parse);
213 
214 /**
215  * eth_header_cache - fill cache entry from neighbour
216  * @neigh: source neighbour
217  * @hh: destination cache entry
218  * @type: Ethernet type field
219  *
220  * Create an Ethernet header template from the neighbour.
221  */
222 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
223 {
224 	struct ethhdr *eth;
225 	const struct net_device *dev = neigh->dev;
226 
227 	eth = (struct ethhdr *)
228 	    (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
229 
230 	if (type == htons(ETH_P_802_3))
231 		return -1;
232 
233 	eth->h_proto = type;
234 	memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
235 	memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
236 	hh->hh_len = ETH_HLEN;
237 	return 0;
238 }
239 EXPORT_SYMBOL(eth_header_cache);
240 
241 /**
242  * eth_header_cache_update - update cache entry
243  * @hh: destination cache entry
244  * @dev: network device
245  * @haddr: new hardware address
246  *
247  * Called by Address Resolution module to notify changes in address.
248  */
249 void eth_header_cache_update(struct hh_cache *hh,
250 			     const struct net_device *dev,
251 			     const unsigned char *haddr)
252 {
253 	memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
254 	       haddr, ETH_ALEN);
255 }
256 EXPORT_SYMBOL(eth_header_cache_update);
257 
258 /**
259  * eth_prepare_mac_addr_change - prepare for mac change
260  * @dev: network device
261  * @p: socket address
262  */
263 int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
264 {
265 	struct sockaddr *addr = p;
266 
267 	if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
268 		return -EBUSY;
269 	if (!is_valid_ether_addr(addr->sa_data))
270 		return -EADDRNOTAVAIL;
271 	return 0;
272 }
273 EXPORT_SYMBOL(eth_prepare_mac_addr_change);
274 
275 /**
276  * eth_commit_mac_addr_change - commit mac change
277  * @dev: network device
278  * @p: socket address
279  */
280 void eth_commit_mac_addr_change(struct net_device *dev, void *p)
281 {
282 	struct sockaddr *addr = p;
283 
284 	memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
285 }
286 EXPORT_SYMBOL(eth_commit_mac_addr_change);
287 
288 /**
289  * eth_mac_addr - set new Ethernet hardware address
290  * @dev: network device
291  * @p: socket address
292  *
293  * Change hardware address of device.
294  *
295  * This doesn't change hardware matching, so needs to be overridden
296  * for most real devices.
297  */
298 int eth_mac_addr(struct net_device *dev, void *p)
299 {
300 	int ret;
301 
302 	ret = eth_prepare_mac_addr_change(dev, p);
303 	if (ret < 0)
304 		return ret;
305 	eth_commit_mac_addr_change(dev, p);
306 	return 0;
307 }
308 EXPORT_SYMBOL(eth_mac_addr);
309 
310 /**
311  * eth_change_mtu - set new MTU size
312  * @dev: network device
313  * @new_mtu: new Maximum Transfer Unit
314  *
315  * Allow changing MTU size. Needs to be overridden for devices
316  * supporting jumbo frames.
317  */
318 int eth_change_mtu(struct net_device *dev, int new_mtu)
319 {
320 	if (new_mtu < 68 || new_mtu > ETH_DATA_LEN)
321 		return -EINVAL;
322 	dev->mtu = new_mtu;
323 	return 0;
324 }
325 EXPORT_SYMBOL(eth_change_mtu);
326 
327 int eth_validate_addr(struct net_device *dev)
328 {
329 	if (!is_valid_ether_addr(dev->dev_addr))
330 		return -EADDRNOTAVAIL;
331 
332 	return 0;
333 }
334 EXPORT_SYMBOL(eth_validate_addr);
335 
336 const struct header_ops eth_header_ops ____cacheline_aligned = {
337 	.create		= eth_header,
338 	.parse		= eth_header_parse,
339 	.cache		= eth_header_cache,
340 	.cache_update	= eth_header_cache_update,
341 };
342 
343 /**
344  * ether_setup - setup Ethernet network device
345  * @dev: network device
346  *
347  * Fill in the fields of the device structure with Ethernet-generic values.
348  */
349 void ether_setup(struct net_device *dev)
350 {
351 	dev->header_ops		= &eth_header_ops;
352 	dev->type		= ARPHRD_ETHER;
353 	dev->hard_header_len 	= ETH_HLEN;
354 	dev->mtu		= ETH_DATA_LEN;
355 	dev->addr_len		= ETH_ALEN;
356 	dev->tx_queue_len	= 1000;	/* Ethernet wants good queues */
357 	dev->flags		= IFF_BROADCAST|IFF_MULTICAST;
358 	dev->priv_flags		|= IFF_TX_SKB_SHARING;
359 
360 	eth_broadcast_addr(dev->broadcast);
361 
362 }
363 EXPORT_SYMBOL(ether_setup);
364 
365 /**
366  * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
367  * @sizeof_priv: Size of additional driver-private structure to be allocated
368  *	for this Ethernet device
369  * @txqs: The number of TX queues this device has.
370  * @rxqs: The number of RX queues this device has.
371  *
372  * Fill in the fields of the device structure with Ethernet-generic
373  * values. Basically does everything except registering the device.
374  *
375  * Constructs a new net device, complete with a private data area of
376  * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
377  * this private data area.
378  */
379 
380 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
381 				      unsigned int rxqs)
382 {
383 	return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN,
384 				ether_setup, txqs, rxqs);
385 }
386 EXPORT_SYMBOL(alloc_etherdev_mqs);
387 
388 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
389 {
390 	return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr);
391 }
392 EXPORT_SYMBOL(sysfs_format_mac);
393 
394 struct sk_buff **eth_gro_receive(struct sk_buff **head,
395 				 struct sk_buff *skb)
396 {
397 	struct sk_buff *p, **pp = NULL;
398 	struct ethhdr *eh, *eh2;
399 	unsigned int hlen, off_eth;
400 	const struct packet_offload *ptype;
401 	__be16 type;
402 	int flush = 1;
403 
404 	off_eth = skb_gro_offset(skb);
405 	hlen = off_eth + sizeof(*eh);
406 	eh = skb_gro_header_fast(skb, off_eth);
407 	if (skb_gro_header_hard(skb, hlen)) {
408 		eh = skb_gro_header_slow(skb, hlen, off_eth);
409 		if (unlikely(!eh))
410 			goto out;
411 	}
412 
413 	flush = 0;
414 
415 	for (p = *head; p; p = p->next) {
416 		if (!NAPI_GRO_CB(p)->same_flow)
417 			continue;
418 
419 		eh2 = (struct ethhdr *)(p->data + off_eth);
420 		if (compare_ether_header(eh, eh2)) {
421 			NAPI_GRO_CB(p)->same_flow = 0;
422 			continue;
423 		}
424 	}
425 
426 	type = eh->h_proto;
427 
428 	rcu_read_lock();
429 	ptype = gro_find_receive_by_type(type);
430 	if (ptype == NULL) {
431 		flush = 1;
432 		goto out_unlock;
433 	}
434 
435 	skb_gro_pull(skb, sizeof(*eh));
436 	skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
437 	pp = ptype->callbacks.gro_receive(head, skb);
438 
439 out_unlock:
440 	rcu_read_unlock();
441 out:
442 	NAPI_GRO_CB(skb)->flush |= flush;
443 
444 	return pp;
445 }
446 EXPORT_SYMBOL(eth_gro_receive);
447 
448 int eth_gro_complete(struct sk_buff *skb, int nhoff)
449 {
450 	struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
451 	__be16 type = eh->h_proto;
452 	struct packet_offload *ptype;
453 	int err = -ENOSYS;
454 
455 	if (skb->encapsulation)
456 		skb_set_inner_mac_header(skb, nhoff);
457 
458 	rcu_read_lock();
459 	ptype = gro_find_complete_by_type(type);
460 	if (ptype != NULL)
461 		err = ptype->callbacks.gro_complete(skb, nhoff +
462 						    sizeof(struct ethhdr));
463 
464 	rcu_read_unlock();
465 	return err;
466 }
467 EXPORT_SYMBOL(eth_gro_complete);
468 
469 static struct packet_offload eth_packet_offload __read_mostly = {
470 	.type = cpu_to_be16(ETH_P_TEB),
471 	.callbacks = {
472 		.gro_receive = eth_gro_receive,
473 		.gro_complete = eth_gro_complete,
474 	},
475 };
476 
477 static int __init eth_offload_init(void)
478 {
479 	dev_add_offload(&eth_packet_offload);
480 
481 	return 0;
482 }
483 
484 fs_initcall(eth_offload_init);
485