blob: 7deecc8d53b77602f466c284d015710e25d1be58 [file] [log] [blame]
b.liue9582032025-04-17 19:18:16 +08001// 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 <linux/uaccess.h>
62#include <net/pkt_sched.h>
63
64__setup("ether=", netdev_boot_setup);
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 */
79int 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}
114EXPORT_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 */
125u32 eth_get_headlen(const struct net_device *dev, void *data, unsigned int 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}
144EXPORT_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
163#ifdef CONFIG_ETHERNET_PACKET_MANGLE
164 if (dev->eth_mangle_rx)
165 dev->eth_mangle_rx(dev, skb);
166#endif
167
168 skb_reset_mac_header(skb);
169
170 eth = (struct ethhdr *)skb->data;
171 skb_pull_inline(skb, ETH_HLEN);
172
173 eth_skb_pkt_type(skb, dev);
174
175 /*
176 * Some variants of DSA tagging don't have an ethertype field
177 * at all, so we check here whether one of those tagging
178 * variants has been configured on the receiving interface,
179 * and if so, set skb->protocol without looking at the packet.
180 * The DSA tagging protocol may be able to decode some but not all
181 * traffic (for example only for management). In that case give it the
182 * option to filter the packets from which it can decode source port
183 * information.
184 */
185 if (unlikely(netdev_uses_dsa(dev)) && dsa_can_decode(skb, dev))
186 return htons(ETH_P_XDSA);
187
188 if (likely(eth_proto_is_802_3(eth->h_proto)))
189 return eth->h_proto;
190
191 /*
192 * This is a magic hack to spot IPX packets. Older Novell breaks
193 * the protocol design and runs IPX over 802.3 without an 802.2 LLC
194 * layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
195 * won't work for fault tolerant netware but does for the rest.
196 */
197 sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
198 if (sap && *sap == 0xFFFF)
199 return htons(ETH_P_802_3);
200
201 /*
202 * Real 802.2 LLC
203 */
204 return htons(ETH_P_802_2);
205}
206EXPORT_SYMBOL(eth_type_trans);
207
208/**
209 * eth_header_parse - extract hardware address from packet
210 * @skb: packet to extract header from
211 * @haddr: destination buffer
212 */
213int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
214{
215 const struct ethhdr *eth = eth_hdr(skb);
216 memcpy(haddr, eth->h_source, ETH_ALEN);
217 return ETH_ALEN;
218}
219EXPORT_SYMBOL(eth_header_parse);
220
221/**
222 * eth_header_cache - fill cache entry from neighbour
223 * @neigh: source neighbour
224 * @hh: destination cache entry
225 * @type: Ethernet type field
226 *
227 * Create an Ethernet header template from the neighbour.
228 */
229int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
230{
231 struct ethhdr *eth;
232 const struct net_device *dev = neigh->dev;
233
234 eth = (struct ethhdr *)
235 (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
236
237 if (type == htons(ETH_P_802_3))
238 return -1;
239
240 eth->h_proto = type;
241 memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
242 memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
243
244 /* Pairs with READ_ONCE() in neigh_resolve_output(),
245 * neigh_hh_output() and neigh_update_hhs().
246 */
247 smp_store_release(&hh->hh_len, ETH_HLEN);
248
249 return 0;
250}
251EXPORT_SYMBOL(eth_header_cache);
252
253/**
254 * eth_header_cache_update - update cache entry
255 * @hh: destination cache entry
256 * @dev: network device
257 * @haddr: new hardware address
258 *
259 * Called by Address Resolution module to notify changes in address.
260 */
261void eth_header_cache_update(struct hh_cache *hh,
262 const struct net_device *dev,
263 const unsigned char *haddr)
264{
265 memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
266 haddr, ETH_ALEN);
267}
268EXPORT_SYMBOL(eth_header_cache_update);
269
270/**
271 * eth_header_parser_protocol - extract protocol from L2 header
272 * @skb: packet to extract protocol from
273 */
274__be16 eth_header_parse_protocol(const struct sk_buff *skb)
275{
276 const struct ethhdr *eth = eth_hdr(skb);
277
278 return eth->h_proto;
279}
280EXPORT_SYMBOL(eth_header_parse_protocol);
281
282/**
283 * eth_prepare_mac_addr_change - prepare for mac change
284 * @dev: network device
285 * @p: socket address
286 */
287int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
288{
289 struct sockaddr *addr = p;
290
291 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
292 return -EBUSY;
293 if (!is_valid_ether_addr(addr->sa_data))
294 return -EADDRNOTAVAIL;
295 return 0;
296}
297EXPORT_SYMBOL(eth_prepare_mac_addr_change);
298
299/**
300 * eth_commit_mac_addr_change - commit mac change
301 * @dev: network device
302 * @p: socket address
303 */
304void eth_commit_mac_addr_change(struct net_device *dev, void *p)
305{
306 struct sockaddr *addr = p;
307
308 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
309}
310EXPORT_SYMBOL(eth_commit_mac_addr_change);
311
312/**
313 * eth_mac_addr - set new Ethernet hardware address
314 * @dev: network device
315 * @p: socket address
316 *
317 * Change hardware address of device.
318 *
319 * This doesn't change hardware matching, so needs to be overridden
320 * for most real devices.
321 */
322int eth_mac_addr(struct net_device *dev, void *p)
323{
324 int ret;
325
326 ret = eth_prepare_mac_addr_change(dev, p);
327 if (ret < 0)
328 return ret;
329 eth_commit_mac_addr_change(dev, p);
330 return 0;
331}
332EXPORT_SYMBOL(eth_mac_addr);
333
334/**
335 * eth_change_mtu - set new MTU size
336 * @dev: network device
337 * @new_mtu: new Maximum Transfer Unit
338 *
339 * Allow changing MTU size. Needs to be overridden for devices
340 * supporting jumbo frames.
341 */
342int eth_change_mtu(struct net_device *dev, int new_mtu)
343{
344 netdev_warn(dev, "%s is deprecated\n", __func__);
345 dev->mtu = new_mtu;
346 return 0;
347}
348EXPORT_SYMBOL(eth_change_mtu);
349
350int eth_validate_addr(struct net_device *dev)
351{
352 if (!is_valid_ether_addr(dev->dev_addr))
353 return -EADDRNOTAVAIL;
354
355 return 0;
356}
357EXPORT_SYMBOL(eth_validate_addr);
358
359const struct header_ops eth_header_ops ____cacheline_aligned = {
360 .create = eth_header,
361 .parse = eth_header_parse,
362 .cache = eth_header_cache,
363 .cache_update = eth_header_cache_update,
364 .parse_protocol = eth_header_parse_protocol,
365};
366
367/**
368 * ether_setup - setup Ethernet network device
369 * @dev: network device
370 *
371 * Fill in the fields of the device structure with Ethernet-generic values.
372 */
373void ether_setup(struct net_device *dev)
374{
375 dev->header_ops = &eth_header_ops;
376 dev->type = ARPHRD_ETHER;
377 dev->hard_header_len = ETH_HLEN;
378 dev->min_header_len = ETH_HLEN;
379 dev->mtu = ETH_DATA_LEN;
380 dev->min_mtu = ETH_MIN_MTU;
381 dev->max_mtu = ETH_DATA_LEN;
382 dev->addr_len = ETH_ALEN;
383 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
384 dev->flags = IFF_BROADCAST|IFF_MULTICAST;
385 dev->priv_flags |= IFF_TX_SKB_SHARING;
386
387 eth_broadcast_addr(dev->broadcast);
388
389}
390EXPORT_SYMBOL(ether_setup);
391
392/**
393 * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
394 * @sizeof_priv: Size of additional driver-private structure to be allocated
395 * for this Ethernet device
396 * @txqs: The number of TX queues this device has.
397 * @rxqs: The number of RX queues this device has.
398 *
399 * Fill in the fields of the device structure with Ethernet-generic
400 * values. Basically does everything except registering the device.
401 *
402 * Constructs a new net device, complete with a private data area of
403 * size (sizeof_priv). A 32-byte (not bit) alignment is enforced for
404 * this private data area.
405 */
406
407struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
408 unsigned int rxqs)
409{
410 return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN,
411 ether_setup, txqs, rxqs);
412}
413EXPORT_SYMBOL(alloc_etherdev_mqs);
414
415static void devm_free_netdev(struct device *dev, void *res)
416{
417 free_netdev(*(struct net_device **)res);
418}
419
420struct net_device *devm_alloc_etherdev_mqs(struct device *dev, int sizeof_priv,
421 unsigned int txqs, unsigned int rxqs)
422{
423 struct net_device **dr;
424 struct net_device *netdev;
425
426 dr = devres_alloc(devm_free_netdev, sizeof(*dr), GFP_KERNEL);
427 if (!dr)
428 return NULL;
429
430 netdev = alloc_etherdev_mqs(sizeof_priv, txqs, rxqs);
431 if (!netdev) {
432 devres_free(dr);
433 return NULL;
434 }
435
436 *dr = netdev;
437 devres_add(dev, dr);
438
439 return netdev;
440}
441EXPORT_SYMBOL(devm_alloc_etherdev_mqs);
442
443ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
444{
445 return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr);
446}
447EXPORT_SYMBOL(sysfs_format_mac);
448
449struct sk_buff *eth_gro_receive(struct list_head *head, struct sk_buff *skb)
450{
451 const struct packet_offload *ptype;
452 unsigned int hlen, off_eth;
453 struct sk_buff *pp = NULL;
454 struct ethhdr *eh, *eh2;
455 struct sk_buff *p;
456 __be16 type;
457 int flush = 1;
458
459 off_eth = skb_gro_offset(skb);
460 hlen = off_eth + sizeof(*eh);
461 eh = skb_gro_header_fast(skb, off_eth);
462 if (skb_gro_header_hard(skb, hlen)) {
463 eh = skb_gro_header_slow(skb, hlen, off_eth);
464 if (unlikely(!eh))
465 goto out;
466 }
467
468 flush = 0;
469
470 list_for_each_entry(p, head, list) {
471 if (!NAPI_GRO_CB(p)->same_flow)
472 continue;
473
474 eh2 = (struct ethhdr *)(p->data + off_eth);
475 if (compare_ether_header(eh, eh2)) {
476 NAPI_GRO_CB(p)->same_flow = 0;
477 continue;
478 }
479 }
480
481 type = eh->h_proto;
482
483 rcu_read_lock();
484 ptype = gro_find_receive_by_type(type);
485 if (ptype == NULL) {
486 flush = 1;
487 goto out_unlock;
488 }
489
490 skb_gro_pull(skb, sizeof(*eh));
491 skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
492 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
493
494out_unlock:
495 rcu_read_unlock();
496out:
497 skb_gro_flush_final(skb, pp, flush);
498
499 return pp;
500}
501EXPORT_SYMBOL(eth_gro_receive);
502
503int eth_gro_complete(struct sk_buff *skb, int nhoff)
504{
505 struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
506 __be16 type = eh->h_proto;
507 struct packet_offload *ptype;
508 int err = -ENOSYS;
509
510 if (skb->encapsulation)
511 skb_set_inner_mac_header(skb, nhoff);
512
513 rcu_read_lock();
514 ptype = gro_find_complete_by_type(type);
515 if (ptype != NULL)
516 err = ptype->callbacks.gro_complete(skb, nhoff +
517 sizeof(struct ethhdr));
518
519 rcu_read_unlock();
520 return err;
521}
522EXPORT_SYMBOL(eth_gro_complete);
523
524static struct packet_offload eth_packet_offload __read_mostly = {
525 .type = cpu_to_be16(ETH_P_TEB),
526 .priority = 10,
527 .callbacks = {
528 .gro_receive = eth_gro_receive,
529 .gro_complete = eth_gro_complete,
530 },
531};
532
533static int __init eth_offload_init(void)
534{
535 dev_add_offload(&eth_packet_offload);
536
537 return 0;
538}
539
540fs_initcall(eth_offload_init);
541
542unsigned char * __weak arch_get_platform_mac_address(void)
543{
544 return NULL;
545}
546
547int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr)
548{
549 const unsigned char *addr = NULL;
550
551 if (dev->of_node)
552 addr = of_get_mac_address(dev->of_node);
553 if (IS_ERR_OR_NULL(addr))
554 addr = arch_get_platform_mac_address();
555
556 if (!addr)
557 return -ENODEV;
558
559 ether_addr_copy(mac_addr, addr);
560
561 return 0;
562}
563EXPORT_SYMBOL(eth_platform_get_mac_address);
564
565/**
566 * Obtain the MAC address from an nvmem cell named 'mac-address' associated
567 * with given device.
568 *
569 * @dev: Device with which the mac-address cell is associated.
570 * @addrbuf: Buffer to which the MAC address will be copied on success.
571 *
572 * Returns 0 on success or a negative error number on failure.
573 */
574int nvmem_get_mac_address(struct device *dev, void *addrbuf)
575{
576 struct nvmem_cell *cell;
577 const void *mac;
578 size_t len;
579
580 cell = nvmem_cell_get(dev, "mac-address");
581 if (IS_ERR(cell))
582 return PTR_ERR(cell);
583
584 mac = nvmem_cell_read(cell, &len);
585 nvmem_cell_put(cell);
586
587 if (IS_ERR(mac))
588 return PTR_ERR(mac);
589
590 if (len != ETH_ALEN || !is_valid_ether_addr(mac)) {
591 kfree(mac);
592 return -EINVAL;
593 }
594
595 ether_addr_copy(addrbuf, mac);
596 kfree(mac);
597
598 return 0;
599}
600EXPORT_SYMBOL(nvmem_get_mac_address);