xref: /openbmc/linux/net/ethernet/eth.c (revision 7a010c3c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Ethernet-type device handling.
8  *
9  * Version:	@(#)eth.c	1.0.7	05/25/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  *		Mark Evans, <evansmp@uhura.aston.ac.uk>
14  *		Florian  La Roche, <rzsfl@rz.uni-sb.de>
15  *		Alan Cox, <gw4pts@gw4pts.ampr.org>
16  *
17  * Fixes:
18  *		Mr Linux	: Arp problems
19  *		Alan Cox	: Generic queue tidyup (very tiny here)
20  *		Alan Cox	: eth_header ntohs should be htons
21  *		Alan Cox	: eth_rebuild_header missing an htons and
22  *				  minor other things.
23  *		Tegge		: Arp bug fixes.
24  *		Florian		: Removed many unnecessary functions, code cleanup
25  *				  and changes for new arp and skbuff.
26  *		Alan Cox	: Redid header building to reflect new format.
27  *		Alan Cox	: ARP only when compiled with CONFIG_INET
28  *		Greg Page	: 802.2 and SNAP stuff.
29  *		Alan Cox	: MAC layer pointers/new format.
30  *		Paul Gortmaker	: eth_copy_and_sum shouldn't csum padding.
31  *		Alan Cox	: Protect against forwarding explosions with
32  *				  older network drivers and IFF_ALLMULTI.
33  *	Christer Weinigel	: Better rebuild header message.
34  *             Andrew Morton    : 26Feb01: kill ether_setup() - use netdev_boot_setup().
35  */
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/kernel.h>
39 #include <linux/string.h>
40 #include <linux/mm.h>
41 #include <linux/socket.h>
42 #include <linux/in.h>
43 #include <linux/inet.h>
44 #include <linux/ip.h>
45 #include <linux/netdevice.h>
46 #include <linux/nvmem-consumer.h>
47 #include <linux/etherdevice.h>
48 #include <linux/skbuff.h>
49 #include <linux/errno.h>
50 #include <linux/init.h>
51 #include <linux/if_ether.h>
52 #include <linux/of_net.h>
53 #include <linux/pci.h>
54 #include <net/dst.h>
55 #include <net/arp.h>
56 #include <net/sock.h>
57 #include <net/ipv6.h>
58 #include <net/ip.h>
59 #include <net/dsa.h>
60 #include <net/flow_dissector.h>
61 #include <net/gro.h>
62 #include <linux/uaccess.h>
63 #include <net/pkt_sched.h>
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 = 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 length of header for an ethernet frame
117  * @dev: pointer to network device
118  * @data: pointer to start of frame
119  * @len: total length of frame
120  *
121  * Make a best effort attempt to pull the length for all of the headers for
122  * a given frame in a linear buffer.
123  */
124 u32 eth_get_headlen(const struct net_device *dev, const void *data, u32 len)
125 {
126 	const unsigned int flags = FLOW_DISSECTOR_F_PARSE_1ST_FRAG;
127 	const struct ethhdr *eth = (const struct ethhdr *)data;
128 	struct flow_keys_basic keys;
129 
130 	/* this should never happen, but better safe than sorry */
131 	if (unlikely(len < sizeof(*eth)))
132 		return len;
133 
134 	/* parse any remaining L2/L3 headers, check for L4 */
135 	if (!skb_flow_dissect_flow_keys_basic(dev_net(dev), NULL, &keys, data,
136 					      eth->h_proto, sizeof(*eth),
137 					      len, flags))
138 		return max_t(u32, keys.control.thoff, sizeof(*eth));
139 
140 	/* parse for any L4 headers */
141 	return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len);
142 }
143 EXPORT_SYMBOL(eth_get_headlen);
144 
145 /**
146  * eth_type_trans - determine the packet's protocol ID.
147  * @skb: received socket data
148  * @dev: receiving network device
149  *
150  * The rule here is that we
151  * assume 802.3 if the type field is short enough to be a length.
152  * This is normal practice and works for any 'now in use' protocol.
153  */
154 __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev)
155 {
156 	unsigned short _service_access_point;
157 	const unsigned short *sap;
158 	const struct ethhdr *eth;
159 
160 	skb->dev = dev;
161 	skb_reset_mac_header(skb);
162 
163 	eth = (struct ethhdr *)skb->data;
164 	skb_pull_inline(skb, ETH_HLEN);
165 
166 	if (unlikely(!ether_addr_equal_64bits(eth->h_dest,
167 					      dev->dev_addr))) {
168 		if (unlikely(is_multicast_ether_addr_64bits(eth->h_dest))) {
169 			if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
170 				skb->pkt_type = PACKET_BROADCAST;
171 			else
172 				skb->pkt_type = PACKET_MULTICAST;
173 		} else {
174 			skb->pkt_type = PACKET_OTHERHOST;
175 		}
176 	}
177 
178 	/*
179 	 * Some variants of DSA tagging don't have an ethertype field
180 	 * at all, so we check here whether one of those tagging
181 	 * variants has been configured on the receiving interface,
182 	 * and if so, set skb->protocol without looking at the packet.
183 	 */
184 	if (unlikely(netdev_uses_dsa(dev)))
185 		return htons(ETH_P_XDSA);
186 
187 	if (likely(eth_proto_is_802_3(eth->h_proto)))
188 		return eth->h_proto;
189 
190 	/*
191 	 *      This is a magic hack to spot IPX packets. Older Novell breaks
192 	 *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
193 	 *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
194 	 *      won't work for fault tolerant netware but does for the rest.
195 	 */
196 	sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
197 	if (sap && *sap == 0xFFFF)
198 		return htons(ETH_P_802_3);
199 
200 	/*
201 	 *      Real 802.2 LLC
202 	 */
203 	return htons(ETH_P_802_2);
204 }
205 EXPORT_SYMBOL(eth_type_trans);
206 
207 /**
208  * eth_header_parse - extract hardware address from packet
209  * @skb: packet to extract header from
210  * @haddr: destination buffer
211  */
212 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
213 {
214 	const struct ethhdr *eth = eth_hdr(skb);
215 	memcpy(haddr, eth->h_source, ETH_ALEN);
216 	return ETH_ALEN;
217 }
218 EXPORT_SYMBOL(eth_header_parse);
219 
220 /**
221  * eth_header_cache - fill cache entry from neighbour
222  * @neigh: source neighbour
223  * @hh: destination cache entry
224  * @type: Ethernet type field
225  *
226  * Create an Ethernet header template from the neighbour.
227  */
228 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
229 {
230 	struct ethhdr *eth;
231 	const struct net_device *dev = neigh->dev;
232 
233 	eth = (struct ethhdr *)
234 	    (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
235 
236 	if (type == htons(ETH_P_802_3))
237 		return -1;
238 
239 	eth->h_proto = type;
240 	memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
241 	memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
242 
243 	/* Pairs with READ_ONCE() in neigh_resolve_output(),
244 	 * neigh_hh_output() and neigh_update_hhs().
245 	 */
246 	smp_store_release(&hh->hh_len, ETH_HLEN);
247 
248 	return 0;
249 }
250 EXPORT_SYMBOL(eth_header_cache);
251 
252 /**
253  * eth_header_cache_update - update cache entry
254  * @hh: destination cache entry
255  * @dev: network device
256  * @haddr: new hardware address
257  *
258  * Called by Address Resolution module to notify changes in address.
259  */
260 void eth_header_cache_update(struct hh_cache *hh,
261 			     const struct net_device *dev,
262 			     const unsigned char *haddr)
263 {
264 	memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
265 	       haddr, ETH_ALEN);
266 }
267 EXPORT_SYMBOL(eth_header_cache_update);
268 
269 /**
270  * eth_header_parse_protocol - extract protocol from L2 header
271  * @skb: packet to extract protocol from
272  */
273 __be16 eth_header_parse_protocol(const struct sk_buff *skb)
274 {
275 	const struct ethhdr *eth = eth_hdr(skb);
276 
277 	return eth->h_proto;
278 }
279 EXPORT_SYMBOL(eth_header_parse_protocol);
280 
281 /**
282  * eth_prepare_mac_addr_change - prepare for mac change
283  * @dev: network device
284  * @p: socket address
285  */
286 int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
287 {
288 	struct sockaddr *addr = p;
289 
290 	if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
291 		return -EBUSY;
292 	if (!is_valid_ether_addr(addr->sa_data))
293 		return -EADDRNOTAVAIL;
294 	return 0;
295 }
296 EXPORT_SYMBOL(eth_prepare_mac_addr_change);
297 
298 /**
299  * eth_commit_mac_addr_change - commit mac change
300  * @dev: network device
301  * @p: socket address
302  */
303 void eth_commit_mac_addr_change(struct net_device *dev, void *p)
304 {
305 	struct sockaddr *addr = p;
306 
307 	memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
308 }
309 EXPORT_SYMBOL(eth_commit_mac_addr_change);
310 
311 /**
312  * eth_mac_addr - set new Ethernet hardware address
313  * @dev: network device
314  * @p: socket address
315  *
316  * Change hardware address of device.
317  *
318  * This doesn't change hardware matching, so needs to be overridden
319  * for most real devices.
320  */
321 int eth_mac_addr(struct net_device *dev, void *p)
322 {
323 	int ret;
324 
325 	ret = eth_prepare_mac_addr_change(dev, p);
326 	if (ret < 0)
327 		return ret;
328 	eth_commit_mac_addr_change(dev, p);
329 	return 0;
330 }
331 EXPORT_SYMBOL(eth_mac_addr);
332 
333 int eth_validate_addr(struct net_device *dev)
334 {
335 	if (!is_valid_ether_addr(dev->dev_addr))
336 		return -EADDRNOTAVAIL;
337 
338 	return 0;
339 }
340 EXPORT_SYMBOL(eth_validate_addr);
341 
342 const struct header_ops eth_header_ops ____cacheline_aligned = {
343 	.create		= eth_header,
344 	.parse		= eth_header_parse,
345 	.cache		= eth_header_cache,
346 	.cache_update	= eth_header_cache_update,
347 	.parse_protocol	= eth_header_parse_protocol,
348 };
349 
350 /**
351  * ether_setup - setup Ethernet network device
352  * @dev: network device
353  *
354  * Fill in the fields of the device structure with Ethernet-generic values.
355  */
356 void ether_setup(struct net_device *dev)
357 {
358 	dev->header_ops		= &eth_header_ops;
359 	dev->type		= ARPHRD_ETHER;
360 	dev->hard_header_len 	= ETH_HLEN;
361 	dev->min_header_len	= ETH_HLEN;
362 	dev->mtu		= ETH_DATA_LEN;
363 	dev->min_mtu		= ETH_MIN_MTU;
364 	dev->max_mtu		= ETH_DATA_LEN;
365 	dev->addr_len		= ETH_ALEN;
366 	dev->tx_queue_len	= DEFAULT_TX_QUEUE_LEN;
367 	dev->flags		= IFF_BROADCAST|IFF_MULTICAST;
368 	dev->priv_flags		|= IFF_TX_SKB_SHARING;
369 
370 	eth_broadcast_addr(dev->broadcast);
371 
372 }
373 EXPORT_SYMBOL(ether_setup);
374 
375 /**
376  * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
377  * @sizeof_priv: Size of additional driver-private structure to be allocated
378  *	for this Ethernet device
379  * @txqs: The number of TX queues this device has.
380  * @rxqs: The number of RX queues this device has.
381  *
382  * Fill in the fields of the device structure with Ethernet-generic
383  * values. Basically does everything except registering the device.
384  *
385  * Constructs a new net device, complete with a private data area of
386  * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
387  * this private data area.
388  */
389 
390 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
391 				      unsigned int rxqs)
392 {
393 	return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN,
394 				ether_setup, txqs, rxqs);
395 }
396 EXPORT_SYMBOL(alloc_etherdev_mqs);
397 
398 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
399 {
400 	return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr);
401 }
402 EXPORT_SYMBOL(sysfs_format_mac);
403 
404 struct sk_buff *eth_gro_receive(struct list_head *head, struct sk_buff *skb)
405 {
406 	const struct packet_offload *ptype;
407 	unsigned int hlen, off_eth;
408 	struct sk_buff *pp = NULL;
409 	struct ethhdr *eh, *eh2;
410 	struct sk_buff *p;
411 	__be16 type;
412 	int flush = 1;
413 
414 	off_eth = skb_gro_offset(skb);
415 	hlen = off_eth + sizeof(*eh);
416 	eh = skb_gro_header_fast(skb, off_eth);
417 	if (skb_gro_header_hard(skb, hlen)) {
418 		eh = skb_gro_header_slow(skb, hlen, off_eth);
419 		if (unlikely(!eh))
420 			goto out;
421 	}
422 
423 	flush = 0;
424 
425 	list_for_each_entry(p, head, list) {
426 		if (!NAPI_GRO_CB(p)->same_flow)
427 			continue;
428 
429 		eh2 = (struct ethhdr *)(p->data + off_eth);
430 		if (compare_ether_header(eh, eh2)) {
431 			NAPI_GRO_CB(p)->same_flow = 0;
432 			continue;
433 		}
434 	}
435 
436 	type = eh->h_proto;
437 
438 	rcu_read_lock();
439 	ptype = gro_find_receive_by_type(type);
440 	if (ptype == NULL) {
441 		flush = 1;
442 		goto out_unlock;
443 	}
444 
445 	skb_gro_pull(skb, sizeof(*eh));
446 	skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
447 
448 	pp = indirect_call_gro_receive_inet(ptype->callbacks.gro_receive,
449 					    ipv6_gro_receive, inet_gro_receive,
450 					    head, skb);
451 
452 out_unlock:
453 	rcu_read_unlock();
454 out:
455 	skb_gro_flush_final(skb, pp, flush);
456 
457 	return pp;
458 }
459 EXPORT_SYMBOL(eth_gro_receive);
460 
461 int eth_gro_complete(struct sk_buff *skb, int nhoff)
462 {
463 	struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
464 	__be16 type = eh->h_proto;
465 	struct packet_offload *ptype;
466 	int err = -ENOSYS;
467 
468 	if (skb->encapsulation)
469 		skb_set_inner_mac_header(skb, nhoff);
470 
471 	rcu_read_lock();
472 	ptype = gro_find_complete_by_type(type);
473 	if (ptype != NULL)
474 		err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
475 					 ipv6_gro_complete, inet_gro_complete,
476 					 skb, nhoff + sizeof(*eh));
477 
478 	rcu_read_unlock();
479 	return err;
480 }
481 EXPORT_SYMBOL(eth_gro_complete);
482 
483 static struct packet_offload eth_packet_offload __read_mostly = {
484 	.type = cpu_to_be16(ETH_P_TEB),
485 	.priority = 10,
486 	.callbacks = {
487 		.gro_receive = eth_gro_receive,
488 		.gro_complete = eth_gro_complete,
489 	},
490 };
491 
492 static int __init eth_offload_init(void)
493 {
494 	dev_add_offload(&eth_packet_offload);
495 
496 	return 0;
497 }
498 
499 fs_initcall(eth_offload_init);
500 
501 unsigned char * __weak arch_get_platform_mac_address(void)
502 {
503 	return NULL;
504 }
505 
506 int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr)
507 {
508 	unsigned char *addr;
509 	int ret;
510 
511 	ret = of_get_mac_address(dev->of_node, mac_addr);
512 	if (!ret)
513 		return 0;
514 
515 	addr = arch_get_platform_mac_address();
516 	if (!addr)
517 		return -ENODEV;
518 
519 	ether_addr_copy(mac_addr, addr);
520 
521 	return 0;
522 }
523 EXPORT_SYMBOL(eth_platform_get_mac_address);
524 
525 /**
526  * nvmem_get_mac_address - Obtain the MAC address from an nvmem cell named
527  * 'mac-address' associated with given device.
528  *
529  * @dev:	Device with which the mac-address cell is associated.
530  * @addrbuf:	Buffer to which the MAC address will be copied on success.
531  *
532  * Returns 0 on success or a negative error number on failure.
533  */
534 int nvmem_get_mac_address(struct device *dev, void *addrbuf)
535 {
536 	struct nvmem_cell *cell;
537 	const void *mac;
538 	size_t len;
539 
540 	cell = nvmem_cell_get(dev, "mac-address");
541 	if (IS_ERR(cell))
542 		return PTR_ERR(cell);
543 
544 	mac = nvmem_cell_read(cell, &len);
545 	nvmem_cell_put(cell);
546 
547 	if (IS_ERR(mac))
548 		return PTR_ERR(mac);
549 
550 	if (len != ETH_ALEN || !is_valid_ether_addr(mac)) {
551 		kfree(mac);
552 		return -EINVAL;
553 	}
554 
555 	ether_addr_copy(addrbuf, mac);
556 	kfree(mac);
557 
558 	return 0;
559 }
560 EXPORT_SYMBOL(nvmem_get_mac_address);
561