xref: /linux/net/ethernet/eth.c (revision 41f7adb676f6c4aef439d78b15c7e1119216bc2b)
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 = (struct ethhdr *)skb->data;
165 	skb_pull_inline(skb, ETH_HLEN);
166 
167 	eth_skb_pkt_type(skb, dev);
168 
169 	/*
170 	 * Some variants of DSA tagging don't have an ethertype field
171 	 * at all, so we check here whether one of those tagging
172 	 * variants has been configured on the receiving interface,
173 	 * and if so, set skb->protocol without looking at the packet.
174 	 */
175 	if (unlikely(netdev_uses_dsa(dev)))
176 		return htons(ETH_P_XDSA);
177 
178 	if (likely(eth_proto_is_802_3(eth->h_proto)))
179 		return eth->h_proto;
180 
181 	/*
182 	 *      This is a magic hack to spot IPX packets. Older Novell breaks
183 	 *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
184 	 *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
185 	 *      won't work for fault tolerant netware but does for the rest.
186 	 */
187 	sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
188 	if (sap && *sap == 0xFFFF)
189 		return htons(ETH_P_802_3);
190 
191 	/*
192 	 *      Real 802.2 LLC
193 	 */
194 	return htons(ETH_P_802_2);
195 }
196 EXPORT_SYMBOL(eth_type_trans);
197 
198 /**
199  * eth_header_parse - extract hardware address from packet
200  * @skb: packet to extract header from
201  * @haddr: destination buffer
202  */
203 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
204 {
205 	const struct ethhdr *eth = eth_hdr(skb);
206 	memcpy(haddr, eth->h_source, ETH_ALEN);
207 	return ETH_ALEN;
208 }
209 EXPORT_SYMBOL(eth_header_parse);
210 
211 /**
212  * eth_header_cache - fill cache entry from neighbour
213  * @neigh: source neighbour
214  * @hh: destination cache entry
215  * @type: Ethernet type field
216  *
217  * Create an Ethernet header template from the neighbour.
218  */
219 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
220 {
221 	struct ethhdr *eth;
222 	const struct net_device *dev = neigh->dev;
223 
224 	eth = (struct ethhdr *)
225 	    (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
226 
227 	if (type == htons(ETH_P_802_3))
228 		return -1;
229 
230 	eth->h_proto = type;
231 	memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
232 	memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
233 
234 	/* Pairs with READ_ONCE() in neigh_resolve_output(),
235 	 * neigh_hh_output() and neigh_update_hhs().
236 	 */
237 	smp_store_release(&hh->hh_len, ETH_HLEN);
238 
239 	return 0;
240 }
241 EXPORT_SYMBOL(eth_header_cache);
242 
243 /**
244  * eth_header_cache_update - update cache entry
245  * @hh: destination cache entry
246  * @dev: network device
247  * @haddr: new hardware address
248  *
249  * Called by Address Resolution module to notify changes in address.
250  */
251 void eth_header_cache_update(struct hh_cache *hh,
252 			     const struct net_device *dev,
253 			     const unsigned char *haddr)
254 {
255 	memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
256 	       haddr, ETH_ALEN);
257 }
258 EXPORT_SYMBOL(eth_header_cache_update);
259 
260 /**
261  * eth_header_parse_protocol - extract protocol from L2 header
262  * @skb: packet to extract protocol from
263  */
264 __be16 eth_header_parse_protocol(const struct sk_buff *skb)
265 {
266 	const struct ethhdr *eth = eth_hdr(skb);
267 
268 	return eth->h_proto;
269 }
270 EXPORT_SYMBOL(eth_header_parse_protocol);
271 
272 /**
273  * eth_prepare_mac_addr_change - prepare for mac change
274  * @dev: network device
275  * @p: socket address
276  */
277 int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
278 {
279 	struct sockaddr *addr = p;
280 
281 	if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
282 		return -EBUSY;
283 	if (!is_valid_ether_addr(addr->sa_data))
284 		return -EADDRNOTAVAIL;
285 	return 0;
286 }
287 EXPORT_SYMBOL(eth_prepare_mac_addr_change);
288 
289 /**
290  * eth_commit_mac_addr_change - commit mac change
291  * @dev: network device
292  * @p: socket address
293  */
294 void eth_commit_mac_addr_change(struct net_device *dev, void *p)
295 {
296 	struct sockaddr *addr = p;
297 
298 	eth_hw_addr_set(dev, addr->sa_data);
299 }
300 EXPORT_SYMBOL(eth_commit_mac_addr_change);
301 
302 /**
303  * eth_mac_addr - set new Ethernet hardware address
304  * @dev: network device
305  * @p: socket address
306  *
307  * Change hardware address of device.
308  *
309  * This doesn't change hardware matching, so needs to be overridden
310  * for most real devices.
311  */
312 int eth_mac_addr(struct net_device *dev, void *p)
313 {
314 	int ret;
315 
316 	ret = eth_prepare_mac_addr_change(dev, p);
317 	if (ret < 0)
318 		return ret;
319 	eth_commit_mac_addr_change(dev, p);
320 	return 0;
321 }
322 EXPORT_SYMBOL(eth_mac_addr);
323 
324 int eth_validate_addr(struct net_device *dev)
325 {
326 	if (!is_valid_ether_addr(dev->dev_addr))
327 		return -EADDRNOTAVAIL;
328 
329 	return 0;
330 }
331 EXPORT_SYMBOL(eth_validate_addr);
332 
333 const struct header_ops eth_header_ops ____cacheline_aligned = {
334 	.create		= eth_header,
335 	.parse		= eth_header_parse,
336 	.cache		= eth_header_cache,
337 	.cache_update	= eth_header_cache_update,
338 	.parse_protocol	= eth_header_parse_protocol,
339 };
340 
341 /**
342  * ether_setup - setup Ethernet network device
343  * @dev: network device
344  *
345  * Fill in the fields of the device structure with Ethernet-generic values.
346  */
347 void ether_setup(struct net_device *dev)
348 {
349 	dev->header_ops		= &eth_header_ops;
350 	dev->type		= ARPHRD_ETHER;
351 	dev->hard_header_len 	= ETH_HLEN;
352 	dev->min_header_len	= ETH_HLEN;
353 	dev->mtu		= ETH_DATA_LEN;
354 	dev->min_mtu		= ETH_MIN_MTU;
355 	dev->max_mtu		= ETH_DATA_LEN;
356 	dev->addr_len		= ETH_ALEN;
357 	dev->tx_queue_len	= DEFAULT_TX_QUEUE_LEN;
358 	dev->flags		= IFF_BROADCAST|IFF_MULTICAST;
359 	dev->priv_flags		|= IFF_TX_SKB_SHARING;
360 
361 	eth_broadcast_addr(dev->broadcast);
362 
363 }
364 EXPORT_SYMBOL(ether_setup);
365 
366 /**
367  * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
368  * @sizeof_priv: Size of additional driver-private structure to be allocated
369  *	for this Ethernet device
370  * @txqs: The number of TX queues this device has.
371  * @rxqs: The number of RX queues this device has.
372  *
373  * Fill in the fields of the device structure with Ethernet-generic
374  * values. Basically does everything except registering the device.
375  *
376  * Constructs a new net device, complete with a private data area of
377  * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
378  * this private data area.
379  */
380 
381 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
382 				      unsigned int rxqs)
383 {
384 	return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_ENUM,
385 				ether_setup, txqs, rxqs);
386 }
387 EXPORT_SYMBOL(alloc_etherdev_mqs);
388 
389 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
390 {
391 	return sysfs_emit(buf, "%*phC\n", len, addr);
392 }
393 EXPORT_SYMBOL(sysfs_format_mac);
394 
395 struct sk_buff *eth_gro_receive(struct list_head *head, struct sk_buff *skb)
396 {
397 	const struct packet_offload *ptype;
398 	unsigned int hlen, off_eth;
399 	struct sk_buff *pp = NULL;
400 	struct ethhdr *eh, *eh2;
401 	struct sk_buff *p;
402 	__be16 type;
403 	int flush = 1;
404 
405 	off_eth = skb_gro_offset(skb);
406 	hlen = off_eth + sizeof(*eh);
407 	eh = skb_gro_header(skb, hlen, off_eth);
408 	if (unlikely(!eh))
409 		goto out;
410 
411 	flush = 0;
412 
413 	list_for_each_entry(p, head, list) {
414 		if (!NAPI_GRO_CB(p)->same_flow)
415 			continue;
416 
417 		eh2 = (struct ethhdr *)(p->data + off_eth);
418 		if (compare_ether_header(eh, eh2)) {
419 			NAPI_GRO_CB(p)->same_flow = 0;
420 			continue;
421 		}
422 	}
423 
424 	type = eh->h_proto;
425 
426 	ptype = gro_find_receive_by_type(type);
427 	if (ptype == NULL) {
428 		flush = 1;
429 		goto out;
430 	}
431 
432 	skb_gro_pull(skb, sizeof(*eh));
433 	skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
434 
435 	pp = indirect_call_gro_receive_inet(ptype->callbacks.gro_receive,
436 					    ipv6_gro_receive, inet_gro_receive,
437 					    head, skb);
438 
439 out:
440 	skb_gro_flush_final(skb, pp, flush);
441 
442 	return pp;
443 }
444 EXPORT_SYMBOL(eth_gro_receive);
445 
446 int eth_gro_complete(struct sk_buff *skb, int nhoff)
447 {
448 	struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
449 	__be16 type = eh->h_proto;
450 	struct packet_offload *ptype;
451 	int err = -ENOSYS;
452 
453 	if (skb->encapsulation)
454 		skb_set_inner_mac_header(skb, nhoff);
455 
456 	ptype = gro_find_complete_by_type(type);
457 	if (ptype != NULL)
458 		err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
459 					 ipv6_gro_complete, inet_gro_complete,
460 					 skb, nhoff + sizeof(*eh));
461 
462 	return err;
463 }
464 EXPORT_SYMBOL(eth_gro_complete);
465 
466 static struct packet_offload eth_packet_offload __read_mostly = {
467 	.type = cpu_to_be16(ETH_P_TEB),
468 	.priority = 10,
469 	.callbacks = {
470 		.gro_receive = eth_gro_receive,
471 		.gro_complete = eth_gro_complete,
472 	},
473 };
474 
475 static int __init eth_offload_init(void)
476 {
477 	dev_add_offload(&eth_packet_offload);
478 
479 	return 0;
480 }
481 
482 fs_initcall(eth_offload_init);
483 
484 unsigned char * __weak arch_get_platform_mac_address(void)
485 {
486 	return NULL;
487 }
488 
489 int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr)
490 {
491 	unsigned char *addr;
492 	int ret;
493 
494 	ret = of_get_mac_address(dev->of_node, mac_addr);
495 	if (!ret)
496 		return 0;
497 
498 	addr = arch_get_platform_mac_address();
499 	if (!addr)
500 		return -ENODEV;
501 
502 	ether_addr_copy(mac_addr, addr);
503 
504 	return 0;
505 }
506 EXPORT_SYMBOL(eth_platform_get_mac_address);
507 
508 /**
509  * platform_get_ethdev_address - Set netdev's MAC address from a given device
510  * @dev:	Pointer to the device
511  * @netdev:	Pointer to netdev to write the address to
512  *
513  * Wrapper around eth_platform_get_mac_address() which writes the address
514  * directly to netdev->dev_addr.
515  */
516 int platform_get_ethdev_address(struct device *dev, struct net_device *netdev)
517 {
518 	u8 addr[ETH_ALEN] __aligned(2);
519 	int ret;
520 
521 	ret = eth_platform_get_mac_address(dev, addr);
522 	if (!ret)
523 		eth_hw_addr_set(netdev, addr);
524 	return ret;
525 }
526 EXPORT_SYMBOL(platform_get_ethdev_address);
527 
528 /**
529  * nvmem_get_mac_address - Obtain the MAC address from an nvmem cell named
530  * 'mac-address' associated with given device.
531  *
532  * @dev:	Device with which the mac-address cell is associated.
533  * @addrbuf:	Buffer to which the MAC address will be copied on success.
534  *
535  * Returns 0 on success or a negative error number on failure.
536  */
537 int nvmem_get_mac_address(struct device *dev, void *addrbuf)
538 {
539 	struct nvmem_cell *cell;
540 	const void *mac;
541 	size_t len;
542 
543 	cell = nvmem_cell_get(dev, "mac-address");
544 	if (IS_ERR(cell))
545 		return PTR_ERR(cell);
546 
547 	mac = nvmem_cell_read(cell, &len);
548 	nvmem_cell_put(cell);
549 
550 	if (IS_ERR(mac))
551 		return PTR_ERR(mac);
552 
553 	if (len != ETH_ALEN || !is_valid_ether_addr(mac)) {
554 		kfree(mac);
555 		return -EINVAL;
556 	}
557 
558 	ether_addr_copy(addrbuf, mac);
559 	kfree(mac);
560 
561 	return 0;
562 }
563 
564 static int fwnode_get_mac_addr(struct fwnode_handle *fwnode,
565 			       const char *name, char *addr)
566 {
567 	int ret;
568 
569 	ret = fwnode_property_read_u8_array(fwnode, name, addr, ETH_ALEN);
570 	if (ret)
571 		return ret;
572 
573 	if (!is_valid_ether_addr(addr))
574 		return -EINVAL;
575 	return 0;
576 }
577 
578 /**
579  * fwnode_get_mac_address - Get the MAC from the firmware node
580  * @fwnode:	Pointer to the firmware node
581  * @addr:	Address of buffer to store the MAC in
582  *
583  * Search the firmware node for the best MAC address to use.  'mac-address' is
584  * checked first, because that is supposed to contain to "most recent" MAC
585  * address. If that isn't set, then 'local-mac-address' is checked next,
586  * because that is the default address.  If that isn't set, then the obsolete
587  * 'address' is checked, just in case we're using an old device tree.
588  *
589  * Note that the 'address' property is supposed to contain a virtual address of
590  * the register set, but some DTS files have redefined that property to be the
591  * MAC address.
592  *
593  * All-zero MAC addresses are rejected, because those could be properties that
594  * exist in the firmware tables, but were not updated by the firmware.  For
595  * example, the DTS could define 'mac-address' and 'local-mac-address', with
596  * zero MAC addresses.  Some older U-Boots only initialized 'local-mac-address'.
597  * In this case, the real MAC is in 'local-mac-address', and 'mac-address'
598  * exists but is all zeros.
599  */
600 int fwnode_get_mac_address(struct fwnode_handle *fwnode, char *addr)
601 {
602 	if (!fwnode_get_mac_addr(fwnode, "mac-address", addr) ||
603 	    !fwnode_get_mac_addr(fwnode, "local-mac-address", addr) ||
604 	    !fwnode_get_mac_addr(fwnode, "address", addr))
605 		return 0;
606 
607 	return -ENOENT;
608 }
609 EXPORT_SYMBOL(fwnode_get_mac_address);
610 
611 /**
612  * device_get_mac_address - Get the MAC for a given device
613  * @dev:	Pointer to the device
614  * @addr:	Address of buffer to store the MAC in
615  */
616 int device_get_mac_address(struct device *dev, char *addr)
617 {
618 	return fwnode_get_mac_address(dev_fwnode(dev), addr);
619 }
620 EXPORT_SYMBOL(device_get_mac_address);
621 
622 /**
623  * device_get_ethdev_address - Set netdev's MAC address from a given device
624  * @dev:	Pointer to the device
625  * @netdev:	Pointer to netdev to write the address to
626  *
627  * Wrapper around device_get_mac_address() which writes the address
628  * directly to netdev->dev_addr.
629  */
630 int device_get_ethdev_address(struct device *dev, struct net_device *netdev)
631 {
632 	u8 addr[ETH_ALEN];
633 	int ret;
634 
635 	ret = device_get_mac_address(dev, addr);
636 	if (!ret)
637 		eth_hw_addr_set(netdev, addr);
638 	return ret;
639 }
640 EXPORT_SYMBOL(device_get_ethdev_address);
641