xref: /linux/net/ethernet/eth.c (revision 36ec807b627b4c0a0a382f0ae48eac7187d14b2b)
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 <linux/property.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 <net/flow_dissector.h>
62 #include <net/gro.h>
63 #include <linux/uaccess.h>
64 #include <net/pkt_sched.h>
65 
66 /**
67  * eth_header - create the Ethernet header
68  * @skb:	buffer to alter
69  * @dev:	source device
70  * @type:	Ethernet type field
71  * @daddr: destination address (NULL leave destination address)
72  * @saddr: source address (NULL use device source address)
73  * @len:   packet length (<= skb->len)
74  *
75  *
76  * Set the protocol type. For a packet of type ETH_P_802_3/2 we put the length
77  * in here instead.
78  */
79 int eth_header(struct sk_buff *skb, struct net_device *dev,
80 	       unsigned short type,
81 	       const void *daddr, const void *saddr, unsigned int len)
82 {
83 	struct ethhdr *eth = skb_push(skb, ETH_HLEN);
84 
85 	if (type != ETH_P_802_3 && type != ETH_P_802_2)
86 		eth->h_proto = htons(type);
87 	else
88 		eth->h_proto = htons(len);
89 
90 	/*
91 	 *      Set the source hardware address.
92 	 */
93 
94 	if (!saddr)
95 		saddr = dev->dev_addr;
96 	memcpy(eth->h_source, saddr, ETH_ALEN);
97 
98 	if (daddr) {
99 		memcpy(eth->h_dest, daddr, ETH_ALEN);
100 		return ETH_HLEN;
101 	}
102 
103 	/*
104 	 *      Anyway, the loopback-device should never use this function...
105 	 */
106 
107 	if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) {
108 		eth_zero_addr(eth->h_dest);
109 		return ETH_HLEN;
110 	}
111 
112 	return -ETH_HLEN;
113 }
114 EXPORT_SYMBOL(eth_header);
115 
116 /**
117  * eth_get_headlen - determine the length of header for an ethernet frame
118  * @dev: pointer to network device
119  * @data: pointer to start of frame
120  * @len: total length of frame
121  *
122  * Make a best effort attempt to pull the length for all of the headers for
123  * a given frame in a linear buffer.
124  */
125 u32 eth_get_headlen(const struct net_device *dev, const void *data, u32 len)
126 {
127 	const unsigned int flags = FLOW_DISSECTOR_F_PARSE_1ST_FRAG;
128 	const struct ethhdr *eth = (const struct ethhdr *)data;
129 	struct flow_keys_basic keys;
130 
131 	/* this should never happen, but better safe than sorry */
132 	if (unlikely(len < sizeof(*eth)))
133 		return len;
134 
135 	/* parse any remaining L2/L3 headers, check for L4 */
136 	if (!skb_flow_dissect_flow_keys_basic(dev_net(dev), NULL, &keys, data,
137 					      eth->h_proto, sizeof(*eth),
138 					      len, flags))
139 		return max_t(u32, keys.control.thoff, sizeof(*eth));
140 
141 	/* parse for any L4 headers */
142 	return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len);
143 }
144 EXPORT_SYMBOL(eth_get_headlen);
145 
146 /**
147  * eth_type_trans - determine the packet's protocol ID.
148  * @skb: received socket data
149  * @dev: receiving network device
150  *
151  * The rule here is that we
152  * assume 802.3 if the type field is short enough to be a length.
153  * This is normal practice and works for any 'now in use' protocol.
154  */
155 __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev)
156 {
157 	unsigned short _service_access_point;
158 	const unsigned short *sap;
159 	const struct ethhdr *eth;
160 
161 	skb->dev = dev;
162 	skb_reset_mac_header(skb);
163 
164 	eth = eth_skb_pull_mac(skb);
165 	eth_skb_pkt_type(skb, dev);
166 
167 	/*
168 	 * Some variants of DSA tagging don't have an ethertype field
169 	 * at all, so we check here whether one of those tagging
170 	 * variants has been configured on the receiving interface,
171 	 * and if so, set skb->protocol without looking at the packet.
172 	 */
173 	if (unlikely(netdev_uses_dsa(dev)))
174 		return htons(ETH_P_XDSA);
175 
176 	if (likely(eth_proto_is_802_3(eth->h_proto)))
177 		return eth->h_proto;
178 
179 	/*
180 	 *      This is a magic hack to spot IPX packets. Older Novell breaks
181 	 *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
182 	 *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
183 	 *      won't work for fault tolerant netware but does for the rest.
184 	 */
185 	sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
186 	if (sap && *sap == 0xFFFF)
187 		return htons(ETH_P_802_3);
188 
189 	/*
190 	 *      Real 802.2 LLC
191 	 */
192 	return htons(ETH_P_802_2);
193 }
194 EXPORT_SYMBOL(eth_type_trans);
195 
196 /**
197  * eth_header_parse - extract hardware address from packet
198  * @skb: packet to extract header from
199  * @haddr: destination buffer
200  */
201 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
202 {
203 	const struct ethhdr *eth = eth_hdr(skb);
204 	memcpy(haddr, eth->h_source, ETH_ALEN);
205 	return ETH_ALEN;
206 }
207 EXPORT_SYMBOL(eth_header_parse);
208 
209 /**
210  * eth_header_cache - fill cache entry from neighbour
211  * @neigh: source neighbour
212  * @hh: destination cache entry
213  * @type: Ethernet type field
214  *
215  * Create an Ethernet header template from the neighbour.
216  */
217 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
218 {
219 	struct ethhdr *eth;
220 	const struct net_device *dev = neigh->dev;
221 
222 	eth = (struct ethhdr *)
223 	    (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
224 
225 	if (type == htons(ETH_P_802_3))
226 		return -1;
227 
228 	eth->h_proto = type;
229 	memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
230 	memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
231 
232 	/* Pairs with READ_ONCE() in neigh_resolve_output(),
233 	 * neigh_hh_output() and neigh_update_hhs().
234 	 */
235 	smp_store_release(&hh->hh_len, ETH_HLEN);
236 
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_header_parse_protocol - extract protocol from L2 header
260  * @skb: packet to extract protocol from
261  */
262 __be16 eth_header_parse_protocol(const struct sk_buff *skb)
263 {
264 	const struct ethhdr *eth = eth_hdr(skb);
265 
266 	return eth->h_proto;
267 }
268 EXPORT_SYMBOL(eth_header_parse_protocol);
269 
270 /**
271  * eth_prepare_mac_addr_change - prepare for mac change
272  * @dev: network device
273  * @p: socket address
274  */
275 int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
276 {
277 	struct sockaddr *addr = p;
278 
279 	if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
280 		return -EBUSY;
281 	if (!is_valid_ether_addr(addr->sa_data))
282 		return -EADDRNOTAVAIL;
283 	return 0;
284 }
285 EXPORT_SYMBOL(eth_prepare_mac_addr_change);
286 
287 /**
288  * eth_commit_mac_addr_change - commit mac change
289  * @dev: network device
290  * @p: socket address
291  */
292 void eth_commit_mac_addr_change(struct net_device *dev, void *p)
293 {
294 	struct sockaddr *addr = p;
295 
296 	eth_hw_addr_set(dev, addr->sa_data);
297 }
298 EXPORT_SYMBOL(eth_commit_mac_addr_change);
299 
300 /**
301  * eth_mac_addr - set new Ethernet hardware address
302  * @dev: network device
303  * @p: socket address
304  *
305  * Change hardware address of device.
306  *
307  * This doesn't change hardware matching, so needs to be overridden
308  * for most real devices.
309  */
310 int eth_mac_addr(struct net_device *dev, void *p)
311 {
312 	int ret;
313 
314 	ret = eth_prepare_mac_addr_change(dev, p);
315 	if (ret < 0)
316 		return ret;
317 	eth_commit_mac_addr_change(dev, p);
318 	return 0;
319 }
320 EXPORT_SYMBOL(eth_mac_addr);
321 
322 int eth_validate_addr(struct net_device *dev)
323 {
324 	if (!is_valid_ether_addr(dev->dev_addr))
325 		return -EADDRNOTAVAIL;
326 
327 	return 0;
328 }
329 EXPORT_SYMBOL(eth_validate_addr);
330 
331 const struct header_ops eth_header_ops ____cacheline_aligned = {
332 	.create		= eth_header,
333 	.parse		= eth_header_parse,
334 	.cache		= eth_header_cache,
335 	.cache_update	= eth_header_cache_update,
336 	.parse_protocol	= eth_header_parse_protocol,
337 };
338 
339 /**
340  * ether_setup - setup Ethernet network device
341  * @dev: network device
342  *
343  * Fill in the fields of the device structure with Ethernet-generic values.
344  */
345 void ether_setup(struct net_device *dev)
346 {
347 	dev->header_ops		= &eth_header_ops;
348 	dev->type		= ARPHRD_ETHER;
349 	dev->hard_header_len 	= ETH_HLEN;
350 	dev->min_header_len	= ETH_HLEN;
351 	dev->mtu		= ETH_DATA_LEN;
352 	dev->min_mtu		= ETH_MIN_MTU;
353 	dev->max_mtu		= ETH_DATA_LEN;
354 	dev->addr_len		= ETH_ALEN;
355 	dev->tx_queue_len	= DEFAULT_TX_QUEUE_LEN;
356 	dev->flags		= IFF_BROADCAST|IFF_MULTICAST;
357 	dev->priv_flags		|= IFF_TX_SKB_SHARING;
358 
359 	eth_broadcast_addr(dev->broadcast);
360 
361 }
362 EXPORT_SYMBOL(ether_setup);
363 
364 /**
365  * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
366  * @sizeof_priv: Size of additional driver-private structure to be allocated
367  *	for this Ethernet device
368  * @txqs: The number of TX queues this device has.
369  * @rxqs: The number of RX queues this device has.
370  *
371  * Fill in the fields of the device structure with Ethernet-generic
372  * values. Basically does everything except registering the device.
373  *
374  * Constructs a new net device, complete with a private data area of
375  * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
376  * this private data area.
377  */
378 
379 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
380 				      unsigned int rxqs)
381 {
382 	return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_ENUM,
383 				ether_setup, txqs, rxqs);
384 }
385 EXPORT_SYMBOL(alloc_etherdev_mqs);
386 
387 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
388 {
389 	return sysfs_emit(buf, "%*phC\n", len, addr);
390 }
391 EXPORT_SYMBOL(sysfs_format_mac);
392 
393 struct sk_buff *eth_gro_receive(struct list_head *head, struct sk_buff *skb)
394 {
395 	const struct packet_offload *ptype;
396 	unsigned int hlen, off_eth;
397 	struct sk_buff *pp = NULL;
398 	struct ethhdr *eh, *eh2;
399 	struct sk_buff *p;
400 	__be16 type;
401 	int flush = 1;
402 
403 	off_eth = skb_gro_offset(skb);
404 	hlen = off_eth + sizeof(*eh);
405 	eh = skb_gro_header(skb, hlen, off_eth);
406 	if (unlikely(!eh))
407 		goto out;
408 
409 	flush = 0;
410 
411 	list_for_each_entry(p, head, list) {
412 		if (!NAPI_GRO_CB(p)->same_flow)
413 			continue;
414 
415 		eh2 = (struct ethhdr *)(p->data + off_eth);
416 		if (compare_ether_header(eh, eh2)) {
417 			NAPI_GRO_CB(p)->same_flow = 0;
418 			continue;
419 		}
420 	}
421 
422 	type = eh->h_proto;
423 
424 	ptype = gro_find_receive_by_type(type);
425 	if (ptype == NULL) {
426 		flush = 1;
427 		goto out;
428 	}
429 
430 	skb_gro_pull(skb, sizeof(*eh));
431 	skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
432 
433 	pp = indirect_call_gro_receive_inet(ptype->callbacks.gro_receive,
434 					    ipv6_gro_receive, inet_gro_receive,
435 					    head, skb);
436 
437 out:
438 	skb_gro_flush_final(skb, pp, flush);
439 
440 	return pp;
441 }
442 EXPORT_SYMBOL(eth_gro_receive);
443 
444 int eth_gro_complete(struct sk_buff *skb, int nhoff)
445 {
446 	struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
447 	__be16 type = eh->h_proto;
448 	struct packet_offload *ptype;
449 	int err = -ENOSYS;
450 
451 	if (skb->encapsulation)
452 		skb_set_inner_mac_header(skb, nhoff);
453 
454 	ptype = gro_find_complete_by_type(type);
455 	if (ptype != NULL)
456 		err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
457 					 ipv6_gro_complete, inet_gro_complete,
458 					 skb, nhoff + sizeof(*eh));
459 
460 	return err;
461 }
462 EXPORT_SYMBOL(eth_gro_complete);
463 
464 static struct packet_offload eth_packet_offload __read_mostly = {
465 	.type = cpu_to_be16(ETH_P_TEB),
466 	.priority = 10,
467 	.callbacks = {
468 		.gro_receive = eth_gro_receive,
469 		.gro_complete = eth_gro_complete,
470 	},
471 };
472 
473 static int __init eth_offload_init(void)
474 {
475 	dev_add_offload(&eth_packet_offload);
476 
477 	return 0;
478 }
479 
480 fs_initcall(eth_offload_init);
481 
482 unsigned char * __weak arch_get_platform_mac_address(void)
483 {
484 	return NULL;
485 }
486 
487 int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr)
488 {
489 	unsigned char *addr;
490 	int ret;
491 
492 	ret = of_get_mac_address(dev->of_node, mac_addr);
493 	if (!ret)
494 		return 0;
495 
496 	addr = arch_get_platform_mac_address();
497 	if (!addr)
498 		return -ENODEV;
499 
500 	ether_addr_copy(mac_addr, addr);
501 
502 	return 0;
503 }
504 EXPORT_SYMBOL(eth_platform_get_mac_address);
505 
506 /**
507  * platform_get_ethdev_address - Set netdev's MAC address from a given device
508  * @dev:	Pointer to the device
509  * @netdev:	Pointer to netdev to write the address to
510  *
511  * Wrapper around eth_platform_get_mac_address() which writes the address
512  * directly to netdev->dev_addr.
513  */
514 int platform_get_ethdev_address(struct device *dev, struct net_device *netdev)
515 {
516 	u8 addr[ETH_ALEN] __aligned(2);
517 	int ret;
518 
519 	ret = eth_platform_get_mac_address(dev, addr);
520 	if (!ret)
521 		eth_hw_addr_set(netdev, addr);
522 	return ret;
523 }
524 EXPORT_SYMBOL(platform_get_ethdev_address);
525 
526 /**
527  * nvmem_get_mac_address - Obtain the MAC address from an nvmem cell named
528  * 'mac-address' associated with given device.
529  *
530  * @dev:	Device with which the mac-address cell is associated.
531  * @addrbuf:	Buffer to which the MAC address will be copied on success.
532  *
533  * Returns 0 on success or a negative error number on failure.
534  */
535 int nvmem_get_mac_address(struct device *dev, void *addrbuf)
536 {
537 	struct nvmem_cell *cell;
538 	const void *mac;
539 	size_t len;
540 
541 	cell = nvmem_cell_get(dev, "mac-address");
542 	if (IS_ERR(cell))
543 		return PTR_ERR(cell);
544 
545 	mac = nvmem_cell_read(cell, &len);
546 	nvmem_cell_put(cell);
547 
548 	if (IS_ERR(mac))
549 		return PTR_ERR(mac);
550 
551 	if (len != ETH_ALEN || !is_valid_ether_addr(mac)) {
552 		kfree(mac);
553 		return -EINVAL;
554 	}
555 
556 	ether_addr_copy(addrbuf, mac);
557 	kfree(mac);
558 
559 	return 0;
560 }
561 
562 static int fwnode_get_mac_addr(struct fwnode_handle *fwnode,
563 			       const char *name, char *addr)
564 {
565 	int ret;
566 
567 	ret = fwnode_property_read_u8_array(fwnode, name, addr, ETH_ALEN);
568 	if (ret)
569 		return ret;
570 
571 	if (!is_valid_ether_addr(addr))
572 		return -EINVAL;
573 	return 0;
574 }
575 
576 /**
577  * fwnode_get_mac_address - Get the MAC from the firmware node
578  * @fwnode:	Pointer to the firmware node
579  * @addr:	Address of buffer to store the MAC in
580  *
581  * Search the firmware node for the best MAC address to use.  'mac-address' is
582  * checked first, because that is supposed to contain to "most recent" MAC
583  * address. If that isn't set, then 'local-mac-address' is checked next,
584  * because that is the default address.  If that isn't set, then the obsolete
585  * 'address' is checked, just in case we're using an old device tree.
586  *
587  * Note that the 'address' property is supposed to contain a virtual address of
588  * the register set, but some DTS files have redefined that property to be the
589  * MAC address.
590  *
591  * All-zero MAC addresses are rejected, because those could be properties that
592  * exist in the firmware tables, but were not updated by the firmware.  For
593  * example, the DTS could define 'mac-address' and 'local-mac-address', with
594  * zero MAC addresses.  Some older U-Boots only initialized 'local-mac-address'.
595  * In this case, the real MAC is in 'local-mac-address', and 'mac-address'
596  * exists but is all zeros.
597  */
598 int fwnode_get_mac_address(struct fwnode_handle *fwnode, char *addr)
599 {
600 	if (!fwnode_get_mac_addr(fwnode, "mac-address", addr) ||
601 	    !fwnode_get_mac_addr(fwnode, "local-mac-address", addr) ||
602 	    !fwnode_get_mac_addr(fwnode, "address", addr))
603 		return 0;
604 
605 	return -ENOENT;
606 }
607 EXPORT_SYMBOL(fwnode_get_mac_address);
608 
609 /**
610  * device_get_mac_address - Get the MAC for a given device
611  * @dev:	Pointer to the device
612  * @addr:	Address of buffer to store the MAC in
613  */
614 int device_get_mac_address(struct device *dev, char *addr)
615 {
616 	return fwnode_get_mac_address(dev_fwnode(dev), addr);
617 }
618 EXPORT_SYMBOL(device_get_mac_address);
619 
620 /**
621  * device_get_ethdev_address - Set netdev's MAC address from a given device
622  * @dev:	Pointer to the device
623  * @netdev:	Pointer to netdev to write the address to
624  *
625  * Wrapper around device_get_mac_address() which writes the address
626  * directly to netdev->dev_addr.
627  */
628 int device_get_ethdev_address(struct device *dev, struct net_device *netdev)
629 {
630 	u8 addr[ETH_ALEN];
631 	int ret;
632 
633 	ret = device_get_mac_address(dev, addr);
634 	if (!ret)
635 		eth_hw_addr_set(netdev, addr);
636 	return ret;
637 }
638 EXPORT_SYMBOL(device_get_ethdev_address);
639