Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Common framework for low-level network console, dump, and debugger code | |
3 | * | |
4 | * Sep 8 2003 Matt Mackall <mpm@selenic.com> | |
5 | * | |
6 | * based on the netconsole code from: | |
7 | * | |
8 | * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com> | |
9 | * Copyright (C) 2002 Red Hat, Inc. | |
10 | */ | |
11 | ||
e6ec2693 JP |
12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
13 | ||
bff38771 | 14 | #include <linux/moduleparam.h> |
1da177e4 LT |
15 | #include <linux/netdevice.h> |
16 | #include <linux/etherdevice.h> | |
17 | #include <linux/string.h> | |
14c85021 | 18 | #include <linux/if_arp.h> |
1da177e4 LT |
19 | #include <linux/inetdevice.h> |
20 | #include <linux/inet.h> | |
21 | #include <linux/interrupt.h> | |
22 | #include <linux/netpoll.h> | |
23 | #include <linux/sched.h> | |
24 | #include <linux/delay.h> | |
25 | #include <linux/rcupdate.h> | |
26 | #include <linux/workqueue.h> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
bc3b2d7f | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <net/tcp.h> |
30 | #include <net/udp.h> | |
31 | #include <asm/unaligned.h> | |
9cbc1cb8 | 32 | #include <trace/events/napi.h> |
1da177e4 LT |
33 | |
34 | /* | |
35 | * We maintain a small pool of fully-sized skbs, to make sure the | |
36 | * message gets out even in extreme OOM situations. | |
37 | */ | |
38 | ||
39 | #define MAX_UDP_CHUNK 1460 | |
40 | #define MAX_SKBS 32 | |
1da177e4 | 41 | |
a1bcfacd | 42 | static struct sk_buff_head skb_pool; |
1da177e4 LT |
43 | |
44 | static atomic_t trapped; | |
45 | ||
2bdfe0ba | 46 | #define USEC_PER_POLL 50 |
d9452e9f DM |
47 | #define NETPOLL_RX_ENABLED 1 |
48 | #define NETPOLL_RX_DROP 2 | |
1da177e4 | 49 | |
6f706245 JP |
50 | #define MAX_SKB_SIZE \ |
51 | (sizeof(struct ethhdr) + \ | |
52 | sizeof(struct iphdr) + \ | |
53 | sizeof(struct udphdr) + \ | |
54 | MAX_UDP_CHUNK) | |
1da177e4 | 55 | |
3578b0c8 | 56 | static void zap_completion_queue(void); |
068c6e98 | 57 | static void arp_reply(struct sk_buff *skb); |
1da177e4 | 58 | |
bff38771 AV |
59 | static unsigned int carrier_timeout = 4; |
60 | module_param(carrier_timeout, uint, 0644); | |
61 | ||
e6ec2693 JP |
62 | #define np_info(np, fmt, ...) \ |
63 | pr_info("%s: " fmt, np->name, ##__VA_ARGS__) | |
64 | #define np_err(np, fmt, ...) \ | |
65 | pr_err("%s: " fmt, np->name, ##__VA_ARGS__) | |
66 | #define np_notice(np, fmt, ...) \ | |
67 | pr_notice("%s: " fmt, np->name, ##__VA_ARGS__) | |
68 | ||
c4028958 | 69 | static void queue_process(struct work_struct *work) |
1da177e4 | 70 | { |
4c1ac1b4 DH |
71 | struct netpoll_info *npinfo = |
72 | container_of(work, struct netpoll_info, tx_work.work); | |
1da177e4 | 73 | struct sk_buff *skb; |
3640543d | 74 | unsigned long flags; |
1da177e4 | 75 | |
6c43ff18 SH |
76 | while ((skb = skb_dequeue(&npinfo->txq))) { |
77 | struct net_device *dev = skb->dev; | |
00829823 | 78 | const struct net_device_ops *ops = dev->netdev_ops; |
fd2ea0a7 | 79 | struct netdev_queue *txq; |
1da177e4 | 80 | |
6c43ff18 SH |
81 | if (!netif_device_present(dev) || !netif_running(dev)) { |
82 | __kfree_skb(skb); | |
83 | continue; | |
84 | } | |
1da177e4 | 85 | |
fd2ea0a7 DM |
86 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
87 | ||
3640543d | 88 | local_irq_save(flags); |
fd2ea0a7 | 89 | __netif_tx_lock(txq, smp_processor_id()); |
73466498 | 90 | if (netif_xmit_frozen_or_stopped(txq) || |
00829823 | 91 | ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { |
6c43ff18 | 92 | skb_queue_head(&npinfo->txq, skb); |
fd2ea0a7 | 93 | __netif_tx_unlock(txq); |
3640543d | 94 | local_irq_restore(flags); |
1da177e4 | 95 | |
25442caf | 96 | schedule_delayed_work(&npinfo->tx_work, HZ/10); |
6c43ff18 SH |
97 | return; |
98 | } | |
fd2ea0a7 | 99 | __netif_tx_unlock(txq); |
3640543d | 100 | local_irq_restore(flags); |
1da177e4 | 101 | } |
1da177e4 LT |
102 | } |
103 | ||
b51655b9 AV |
104 | static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, |
105 | unsigned short ulen, __be32 saddr, __be32 daddr) | |
1da177e4 | 106 | { |
d6f5493c | 107 | __wsum psum; |
fb286bb2 | 108 | |
60476372 | 109 | if (uh->check == 0 || skb_csum_unnecessary(skb)) |
1da177e4 LT |
110 | return 0; |
111 | ||
fb286bb2 HX |
112 | psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); |
113 | ||
84fa7933 | 114 | if (skb->ip_summed == CHECKSUM_COMPLETE && |
d3bc23e7 | 115 | !csum_fold(csum_add(psum, skb->csum))) |
fb286bb2 | 116 | return 0; |
1da177e4 | 117 | |
fb286bb2 | 118 | skb->csum = psum; |
1da177e4 | 119 | |
fb286bb2 | 120 | return __skb_checksum_complete(skb); |
1da177e4 LT |
121 | } |
122 | ||
123 | /* | |
124 | * Check whether delayed processing was scheduled for our NIC. If so, | |
125 | * we attempt to grab the poll lock and use ->poll() to pump the card. | |
126 | * If this fails, either we've recursed in ->poll() or it's already | |
127 | * running on another CPU. | |
128 | * | |
129 | * Note: we don't mask interrupts with this lock because we're using | |
130 | * trylock here and interrupts are already disabled in the softirq | |
131 | * case. Further, we test the poll_owner to avoid recursion on UP | |
132 | * systems where the lock doesn't exist. | |
133 | * | |
134 | * In cases where there is bi-directional communications, reading only | |
135 | * one message at a time can lead to packets being dropped by the | |
136 | * network adapter, forcing superfluous retries and possibly timeouts. | |
137 | * Thus, we set our budget to greater than 1. | |
138 | */ | |
0a7606c1 DM |
139 | static int poll_one_napi(struct netpoll_info *npinfo, |
140 | struct napi_struct *napi, int budget) | |
141 | { | |
142 | int work; | |
143 | ||
144 | /* net_rx_action's ->poll() invocations and our's are | |
145 | * synchronized by this test which is only made while | |
146 | * holding the napi->poll_lock. | |
147 | */ | |
148 | if (!test_bit(NAPI_STATE_SCHED, &napi->state)) | |
149 | return budget; | |
150 | ||
d9452e9f | 151 | npinfo->rx_flags |= NETPOLL_RX_DROP; |
0a7606c1 | 152 | atomic_inc(&trapped); |
7b363e44 | 153 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 DM |
154 | |
155 | work = napi->poll(napi, budget); | |
7d18f114 | 156 | trace_napi_poll(napi); |
0a7606c1 | 157 | |
7b363e44 | 158 | clear_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 | 159 | atomic_dec(&trapped); |
d9452e9f | 160 | npinfo->rx_flags &= ~NETPOLL_RX_DROP; |
0a7606c1 DM |
161 | |
162 | return budget - work; | |
163 | } | |
164 | ||
5106930b | 165 | static void poll_napi(struct net_device *dev) |
1da177e4 | 166 | { |
bea3348e | 167 | struct napi_struct *napi; |
1da177e4 LT |
168 | int budget = 16; |
169 | ||
f13d493d | 170 | list_for_each_entry(napi, &dev->napi_list, dev_list) { |
0a7606c1 | 171 | if (napi->poll_owner != smp_processor_id() && |
bea3348e | 172 | spin_trylock(&napi->poll_lock)) { |
5106930b | 173 | budget = poll_one_napi(dev->npinfo, napi, budget); |
bea3348e | 174 | spin_unlock(&napi->poll_lock); |
0a7606c1 DM |
175 | |
176 | if (!budget) | |
177 | break; | |
bea3348e | 178 | } |
1da177e4 LT |
179 | } |
180 | } | |
181 | ||
068c6e98 NH |
182 | static void service_arp_queue(struct netpoll_info *npi) |
183 | { | |
5106930b SH |
184 | if (npi) { |
185 | struct sk_buff *skb; | |
068c6e98 | 186 | |
5106930b SH |
187 | while ((skb = skb_dequeue(&npi->arp_tx))) |
188 | arp_reply(skb); | |
068c6e98 | 189 | } |
068c6e98 NH |
190 | } |
191 | ||
234b921d | 192 | static void netpoll_poll_dev(struct net_device *dev) |
1da177e4 | 193 | { |
5e392739 | 194 | const struct net_device_ops *ops; |
5106930b | 195 | |
5e392739 PE |
196 | if (!dev || !netif_running(dev)) |
197 | return; | |
198 | ||
199 | ops = dev->netdev_ops; | |
200 | if (!ops->ndo_poll_controller) | |
1da177e4 LT |
201 | return; |
202 | ||
203 | /* Process pending work on NIC */ | |
d314774c | 204 | ops->ndo_poll_controller(dev); |
5106930b SH |
205 | |
206 | poll_napi(dev); | |
1da177e4 | 207 | |
58e05f35 | 208 | if (dev->flags & IFF_SLAVE) { |
5a698af5 AW |
209 | if (dev->npinfo) { |
210 | struct net_device *bond_dev = dev->master; | |
211 | struct sk_buff *skb; | |
212 | while ((skb = skb_dequeue(&dev->npinfo->arp_tx))) { | |
213 | skb->dev = bond_dev; | |
214 | skb_queue_tail(&bond_dev->npinfo->arp_tx, skb); | |
215 | } | |
216 | } | |
217 | } | |
218 | ||
5106930b | 219 | service_arp_queue(dev->npinfo); |
068c6e98 | 220 | |
3578b0c8 | 221 | zap_completion_queue(); |
1da177e4 LT |
222 | } |
223 | ||
224 | static void refill_skbs(void) | |
225 | { | |
226 | struct sk_buff *skb; | |
227 | unsigned long flags; | |
228 | ||
a1bcfacd SH |
229 | spin_lock_irqsave(&skb_pool.lock, flags); |
230 | while (skb_pool.qlen < MAX_SKBS) { | |
1da177e4 LT |
231 | skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); |
232 | if (!skb) | |
233 | break; | |
234 | ||
a1bcfacd | 235 | __skb_queue_tail(&skb_pool, skb); |
1da177e4 | 236 | } |
a1bcfacd | 237 | spin_unlock_irqrestore(&skb_pool.lock, flags); |
1da177e4 LT |
238 | } |
239 | ||
3578b0c8 DM |
240 | static void zap_completion_queue(void) |
241 | { | |
242 | unsigned long flags; | |
243 | struct softnet_data *sd = &get_cpu_var(softnet_data); | |
244 | ||
245 | if (sd->completion_queue) { | |
246 | struct sk_buff *clist; | |
247 | ||
248 | local_irq_save(flags); | |
249 | clist = sd->completion_queue; | |
250 | sd->completion_queue = NULL; | |
251 | local_irq_restore(flags); | |
252 | ||
253 | while (clist != NULL) { | |
254 | struct sk_buff *skb = clist; | |
255 | clist = clist->next; | |
256 | if (skb->destructor) { | |
257 | atomic_inc(&skb->users); | |
258 | dev_kfree_skb_any(skb); /* put this one back */ | |
259 | } else { | |
260 | __kfree_skb(skb); | |
261 | } | |
262 | } | |
263 | } | |
264 | ||
265 | put_cpu_var(softnet_data); | |
266 | } | |
267 | ||
a1bcfacd | 268 | static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) |
1da177e4 | 269 | { |
a1bcfacd SH |
270 | int count = 0; |
271 | struct sk_buff *skb; | |
1da177e4 | 272 | |
3578b0c8 | 273 | zap_completion_queue(); |
a1bcfacd | 274 | refill_skbs(); |
1da177e4 | 275 | repeat: |
1da177e4 LT |
276 | |
277 | skb = alloc_skb(len, GFP_ATOMIC); | |
a1bcfacd SH |
278 | if (!skb) |
279 | skb = skb_dequeue(&skb_pool); | |
1da177e4 LT |
280 | |
281 | if (!skb) { | |
a1bcfacd | 282 | if (++count < 10) { |
2a49e001 | 283 | netpoll_poll_dev(np->dev); |
a1bcfacd | 284 | goto repeat; |
1da177e4 | 285 | } |
a1bcfacd | 286 | return NULL; |
1da177e4 LT |
287 | } |
288 | ||
289 | atomic_set(&skb->users, 1); | |
290 | skb_reserve(skb, reserve); | |
291 | return skb; | |
292 | } | |
293 | ||
bea3348e SH |
294 | static int netpoll_owner_active(struct net_device *dev) |
295 | { | |
296 | struct napi_struct *napi; | |
297 | ||
298 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
299 | if (napi->poll_owner == smp_processor_id()) | |
300 | return 1; | |
301 | } | |
302 | return 0; | |
303 | } | |
304 | ||
c2355e1a NH |
305 | void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, |
306 | struct net_device *dev) | |
1da177e4 | 307 | { |
2bdfe0ba SH |
308 | int status = NETDEV_TX_BUSY; |
309 | unsigned long tries; | |
00829823 | 310 | const struct net_device_ops *ops = dev->netdev_ops; |
de85d99e | 311 | /* It is up to the caller to keep npinfo alive. */ |
4ec93edb | 312 | struct netpoll_info *npinfo = np->dev->npinfo; |
2bdfe0ba | 313 | |
4ec93edb YH |
314 | if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { |
315 | __kfree_skb(skb); | |
316 | return; | |
317 | } | |
2bdfe0ba SH |
318 | |
319 | /* don't get messages out of order, and no recursion */ | |
bea3348e | 320 | if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { |
fd2ea0a7 | 321 | struct netdev_queue *txq; |
a49f99ff AM |
322 | unsigned long flags; |
323 | ||
fd2ea0a7 DM |
324 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
325 | ||
a49f99ff | 326 | local_irq_save(flags); |
0db3dc73 SH |
327 | /* try until next clock tick */ |
328 | for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; | |
329 | tries > 0; --tries) { | |
fd2ea0a7 | 330 | if (__netif_tx_trylock(txq)) { |
73466498 | 331 | if (!netif_xmit_stopped(txq)) { |
00829823 | 332 | status = ops->ndo_start_xmit(skb, dev); |
08baf561 ED |
333 | if (status == NETDEV_TX_OK) |
334 | txq_trans_update(txq); | |
335 | } | |
fd2ea0a7 | 336 | __netif_tx_unlock(txq); |
e37b8d93 AM |
337 | |
338 | if (status == NETDEV_TX_OK) | |
339 | break; | |
340 | ||
e37b8d93 | 341 | } |
0db3dc73 SH |
342 | |
343 | /* tickle device maybe there is some cleanup */ | |
2a49e001 | 344 | netpoll_poll_dev(np->dev); |
0db3dc73 SH |
345 | |
346 | udelay(USEC_PER_POLL); | |
0db1d6fc | 347 | } |
79b1bee8 DD |
348 | |
349 | WARN_ONCE(!irqs_disabled(), | |
350 | "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", | |
351 | dev->name, ops->ndo_start_xmit); | |
352 | ||
a49f99ff | 353 | local_irq_restore(flags); |
1da177e4 | 354 | } |
1da177e4 | 355 | |
2bdfe0ba | 356 | if (status != NETDEV_TX_OK) { |
5de4a473 | 357 | skb_queue_tail(&npinfo->txq, skb); |
4c1ac1b4 | 358 | schedule_delayed_work(&npinfo->tx_work,0); |
1da177e4 | 359 | } |
1da177e4 | 360 | } |
c2355e1a | 361 | EXPORT_SYMBOL(netpoll_send_skb_on_dev); |
1da177e4 LT |
362 | |
363 | void netpoll_send_udp(struct netpoll *np, const char *msg, int len) | |
364 | { | |
365 | int total_len, eth_len, ip_len, udp_len; | |
366 | struct sk_buff *skb; | |
367 | struct udphdr *udph; | |
368 | struct iphdr *iph; | |
369 | struct ethhdr *eth; | |
370 | ||
371 | udp_len = len + sizeof(*udph); | |
372 | ip_len = eth_len = udp_len + sizeof(*iph); | |
373 | total_len = eth_len + ETH_HLEN + NET_IP_ALIGN; | |
374 | ||
375 | skb = find_skb(np, total_len, total_len - len); | |
376 | if (!skb) | |
377 | return; | |
378 | ||
27d7ff46 | 379 | skb_copy_to_linear_data(skb, msg, len); |
1da177e4 LT |
380 | skb->len += len; |
381 | ||
4bedb452 ACM |
382 | skb_push(skb, sizeof(*udph)); |
383 | skb_reset_transport_header(skb); | |
384 | udph = udp_hdr(skb); | |
1da177e4 LT |
385 | udph->source = htons(np->local_port); |
386 | udph->dest = htons(np->remote_port); | |
387 | udph->len = htons(udp_len); | |
388 | udph->check = 0; | |
e7557af5 HH |
389 | udph->check = csum_tcpudp_magic(np->local_ip, |
390 | np->remote_ip, | |
8e365eec | 391 | udp_len, IPPROTO_UDP, |
07f0757a | 392 | csum_partial(udph, udp_len, 0)); |
8e365eec | 393 | if (udph->check == 0) |
5e57dff2 | 394 | udph->check = CSUM_MANGLED_0; |
1da177e4 | 395 | |
e2d1bca7 ACM |
396 | skb_push(skb, sizeof(*iph)); |
397 | skb_reset_network_header(skb); | |
eddc9ec5 | 398 | iph = ip_hdr(skb); |
1da177e4 LT |
399 | |
400 | /* iph->version = 4; iph->ihl = 5; */ | |
401 | put_unaligned(0x45, (unsigned char *)iph); | |
402 | iph->tos = 0; | |
403 | put_unaligned(htons(ip_len), &(iph->tot_len)); | |
404 | iph->id = 0; | |
405 | iph->frag_off = 0; | |
406 | iph->ttl = 64; | |
407 | iph->protocol = IPPROTO_UDP; | |
408 | iph->check = 0; | |
e7557af5 HH |
409 | put_unaligned(np->local_ip, &(iph->saddr)); |
410 | put_unaligned(np->remote_ip, &(iph->daddr)); | |
1da177e4 LT |
411 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); |
412 | ||
413 | eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); | |
459a98ed | 414 | skb_reset_mac_header(skb); |
206daaf7 | 415 | skb->protocol = eth->h_proto = htons(ETH_P_IP); |
09538641 SH |
416 | memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); |
417 | memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); | |
1da177e4 LT |
418 | |
419 | skb->dev = np->dev; | |
420 | ||
421 | netpoll_send_skb(np, skb); | |
422 | } | |
9e34a5b5 | 423 | EXPORT_SYMBOL(netpoll_send_udp); |
1da177e4 LT |
424 | |
425 | static void arp_reply(struct sk_buff *skb) | |
426 | { | |
115c1d6e | 427 | struct netpoll_info *npinfo = skb->dev->npinfo; |
1da177e4 LT |
428 | struct arphdr *arp; |
429 | unsigned char *arp_ptr; | |
430 | int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; | |
252e3346 | 431 | __be32 sip, tip; |
47bbec02 | 432 | unsigned char *sha; |
1da177e4 | 433 | struct sk_buff *send_skb; |
508e14b4 DB |
434 | struct netpoll *np, *tmp; |
435 | unsigned long flags; | |
ae641949 | 436 | int hlen, tlen; |
508e14b4 DB |
437 | int hits = 0; |
438 | ||
439 | if (list_empty(&npinfo->rx_np)) | |
440 | return; | |
441 | ||
442 | /* Before checking the packet, we do some early | |
443 | inspection whether this is interesting at all */ | |
444 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
445 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
446 | if (np->dev == skb->dev) | |
447 | hits++; | |
448 | } | |
449 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 | 450 | |
508e14b4 DB |
451 | /* No netpoll struct is using this dev */ |
452 | if (!hits) | |
115c1d6e | 453 | return; |
1da177e4 LT |
454 | |
455 | /* No arp on this interface */ | |
456 | if (skb->dev->flags & IFF_NOARP) | |
457 | return; | |
458 | ||
988b7050 | 459 | if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) |
1da177e4 LT |
460 | return; |
461 | ||
c1d2bbe1 | 462 | skb_reset_network_header(skb); |
badff6d0 | 463 | skb_reset_transport_header(skb); |
d0a92be0 | 464 | arp = arp_hdr(skb); |
1da177e4 LT |
465 | |
466 | if ((arp->ar_hrd != htons(ARPHRD_ETHER) && | |
467 | arp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
468 | arp->ar_pro != htons(ETH_P_IP) || | |
469 | arp->ar_op != htons(ARPOP_REQUEST)) | |
470 | return; | |
471 | ||
47bbec02 NH |
472 | arp_ptr = (unsigned char *)(arp+1); |
473 | /* save the location of the src hw addr */ | |
474 | sha = arp_ptr; | |
475 | arp_ptr += skb->dev->addr_len; | |
1da177e4 | 476 | memcpy(&sip, arp_ptr, 4); |
47bbec02 | 477 | arp_ptr += 4; |
508e14b4 DB |
478 | /* If we actually cared about dst hw addr, |
479 | it would get copied here */ | |
47bbec02 | 480 | arp_ptr += skb->dev->addr_len; |
1da177e4 LT |
481 | memcpy(&tip, arp_ptr, 4); |
482 | ||
483 | /* Should we ignore arp? */ | |
508e14b4 | 484 | if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) |
1da177e4 LT |
485 | return; |
486 | ||
988b7050 | 487 | size = arp_hdr_len(skb->dev); |
1da177e4 | 488 | |
508e14b4 DB |
489 | spin_lock_irqsave(&npinfo->rx_lock, flags); |
490 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
491 | if (tip != np->local_ip) | |
492 | continue; | |
1da177e4 | 493 | |
ae641949 HX |
494 | hlen = LL_RESERVED_SPACE(np->dev); |
495 | tlen = np->dev->needed_tailroom; | |
496 | send_skb = find_skb(np, size + hlen + tlen, hlen); | |
508e14b4 DB |
497 | if (!send_skb) |
498 | continue; | |
1da177e4 | 499 | |
508e14b4 DB |
500 | skb_reset_network_header(send_skb); |
501 | arp = (struct arphdr *) skb_put(send_skb, size); | |
502 | send_skb->dev = skb->dev; | |
503 | send_skb->protocol = htons(ETH_P_ARP); | |
1da177e4 | 504 | |
508e14b4 DB |
505 | /* Fill the device header for the ARP frame */ |
506 | if (dev_hard_header(send_skb, skb->dev, ptype, | |
507 | sha, np->dev->dev_addr, | |
508 | send_skb->len) < 0) { | |
509 | kfree_skb(send_skb); | |
510 | continue; | |
511 | } | |
1da177e4 | 512 | |
508e14b4 DB |
513 | /* |
514 | * Fill out the arp protocol part. | |
515 | * | |
516 | * we only support ethernet device type, | |
517 | * which (according to RFC 1390) should | |
518 | * always equal 1 (Ethernet). | |
519 | */ | |
1da177e4 | 520 | |
508e14b4 DB |
521 | arp->ar_hrd = htons(np->dev->type); |
522 | arp->ar_pro = htons(ETH_P_IP); | |
523 | arp->ar_hln = np->dev->addr_len; | |
524 | arp->ar_pln = 4; | |
525 | arp->ar_op = htons(type); | |
526 | ||
527 | arp_ptr = (unsigned char *)(arp + 1); | |
528 | memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); | |
529 | arp_ptr += np->dev->addr_len; | |
530 | memcpy(arp_ptr, &tip, 4); | |
531 | arp_ptr += 4; | |
532 | memcpy(arp_ptr, sha, np->dev->addr_len); | |
533 | arp_ptr += np->dev->addr_len; | |
534 | memcpy(arp_ptr, &sip, 4); | |
535 | ||
536 | netpoll_send_skb(np, send_skb); | |
537 | ||
538 | /* If there are several rx_hooks for the same address, | |
539 | we're fine by sending a single reply */ | |
540 | break; | |
541 | } | |
542 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 LT |
543 | } |
544 | ||
545 | int __netpoll_rx(struct sk_buff *skb) | |
546 | { | |
547 | int proto, len, ulen; | |
508e14b4 | 548 | int hits = 0; |
b71d1d42 | 549 | const struct iphdr *iph; |
1da177e4 | 550 | struct udphdr *uh; |
508e14b4 DB |
551 | struct netpoll_info *npinfo = skb->dev->npinfo; |
552 | struct netpoll *np, *tmp; | |
068c6e98 | 553 | |
508e14b4 | 554 | if (list_empty(&npinfo->rx_np)) |
1da177e4 | 555 | goto out; |
508e14b4 | 556 | |
1da177e4 LT |
557 | if (skb->dev->type != ARPHRD_ETHER) |
558 | goto out; | |
559 | ||
d9452e9f | 560 | /* check if netpoll clients need ARP */ |
724800d6 | 561 | if (skb->protocol == htons(ETH_P_ARP) && |
1da177e4 | 562 | atomic_read(&trapped)) { |
508e14b4 | 563 | skb_queue_tail(&npinfo->arp_tx, skb); |
1da177e4 LT |
564 | return 1; |
565 | } | |
566 | ||
567 | proto = ntohs(eth_hdr(skb)->h_proto); | |
568 | if (proto != ETH_P_IP) | |
569 | goto out; | |
570 | if (skb->pkt_type == PACKET_OTHERHOST) | |
571 | goto out; | |
572 | if (skb_shared(skb)) | |
573 | goto out; | |
574 | ||
1da177e4 LT |
575 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) |
576 | goto out; | |
e9278a47 | 577 | iph = (struct iphdr *)skb->data; |
1da177e4 LT |
578 | if (iph->ihl < 5 || iph->version != 4) |
579 | goto out; | |
580 | if (!pskb_may_pull(skb, iph->ihl*4)) | |
581 | goto out; | |
e9278a47 | 582 | iph = (struct iphdr *)skb->data; |
1da177e4 LT |
583 | if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) |
584 | goto out; | |
585 | ||
586 | len = ntohs(iph->tot_len); | |
587 | if (skb->len < len || len < iph->ihl*4) | |
588 | goto out; | |
589 | ||
5e7d7fa5 AL |
590 | /* |
591 | * Our transport medium may have padded the buffer out. | |
592 | * Now We trim to the true length of the frame. | |
593 | */ | |
594 | if (pskb_trim_rcsum(skb, len)) | |
595 | goto out; | |
596 | ||
e9278a47 | 597 | iph = (struct iphdr *)skb->data; |
1da177e4 LT |
598 | if (iph->protocol != IPPROTO_UDP) |
599 | goto out; | |
600 | ||
601 | len -= iph->ihl*4; | |
602 | uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); | |
603 | ulen = ntohs(uh->len); | |
604 | ||
605 | if (ulen != len) | |
606 | goto out; | |
fb286bb2 | 607 | if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) |
1da177e4 | 608 | goto out; |
1da177e4 | 609 | |
508e14b4 DB |
610 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { |
611 | if (np->local_ip && np->local_ip != iph->daddr) | |
612 | continue; | |
613 | if (np->remote_ip && np->remote_ip != iph->saddr) | |
614 | continue; | |
615 | if (np->local_port && np->local_port != ntohs(uh->dest)) | |
616 | continue; | |
617 | ||
618 | np->rx_hook(np, ntohs(uh->source), | |
619 | (char *)(uh+1), | |
620 | ulen - sizeof(struct udphdr)); | |
621 | hits++; | |
622 | } | |
623 | ||
624 | if (!hits) | |
625 | goto out; | |
1da177e4 LT |
626 | |
627 | kfree_skb(skb); | |
628 | return 1; | |
629 | ||
630 | out: | |
631 | if (atomic_read(&trapped)) { | |
632 | kfree_skb(skb); | |
633 | return 1; | |
634 | } | |
635 | ||
636 | return 0; | |
637 | } | |
638 | ||
0bcc1816 SS |
639 | void netpoll_print_options(struct netpoll *np) |
640 | { | |
e6ec2693 JP |
641 | np_info(np, "local port %d\n", np->local_port); |
642 | np_info(np, "local IP %pI4\n", &np->local_ip); | |
643 | np_info(np, "interface '%s'\n", np->dev_name); | |
644 | np_info(np, "remote port %d\n", np->remote_port); | |
645 | np_info(np, "remote IP %pI4\n", &np->remote_ip); | |
646 | np_info(np, "remote ethernet address %pM\n", np->remote_mac); | |
0bcc1816 | 647 | } |
9e34a5b5 | 648 | EXPORT_SYMBOL(netpoll_print_options); |
0bcc1816 | 649 | |
1da177e4 LT |
650 | int netpoll_parse_options(struct netpoll *np, char *opt) |
651 | { | |
652 | char *cur=opt, *delim; | |
653 | ||
c68b9070 | 654 | if (*cur != '@') { |
1da177e4 LT |
655 | if ((delim = strchr(cur, '@')) == NULL) |
656 | goto parse_failed; | |
c68b9070 DM |
657 | *delim = 0; |
658 | np->local_port = simple_strtol(cur, NULL, 10); | |
659 | cur = delim; | |
1da177e4 LT |
660 | } |
661 | cur++; | |
1da177e4 | 662 | |
c68b9070 | 663 | if (*cur != '/') { |
1da177e4 LT |
664 | if ((delim = strchr(cur, '/')) == NULL) |
665 | goto parse_failed; | |
c68b9070 | 666 | *delim = 0; |
e7557af5 | 667 | np->local_ip = in_aton(cur); |
c68b9070 | 668 | cur = delim; |
1da177e4 LT |
669 | } |
670 | cur++; | |
671 | ||
c68b9070 | 672 | if (*cur != ',') { |
1da177e4 LT |
673 | /* parse out dev name */ |
674 | if ((delim = strchr(cur, ',')) == NULL) | |
675 | goto parse_failed; | |
c68b9070 | 676 | *delim = 0; |
1da177e4 | 677 | strlcpy(np->dev_name, cur, sizeof(np->dev_name)); |
c68b9070 | 678 | cur = delim; |
1da177e4 LT |
679 | } |
680 | cur++; | |
681 | ||
c68b9070 | 682 | if (*cur != '@') { |
1da177e4 LT |
683 | /* dst port */ |
684 | if ((delim = strchr(cur, '@')) == NULL) | |
685 | goto parse_failed; | |
c68b9070 | 686 | *delim = 0; |
5fc05f87 | 687 | if (*cur == ' ' || *cur == '\t') |
e6ec2693 | 688 | np_info(np, "warning: whitespace is not allowed\n"); |
c68b9070 DM |
689 | np->remote_port = simple_strtol(cur, NULL, 10); |
690 | cur = delim; | |
1da177e4 LT |
691 | } |
692 | cur++; | |
1da177e4 LT |
693 | |
694 | /* dst ip */ | |
695 | if ((delim = strchr(cur, '/')) == NULL) | |
696 | goto parse_failed; | |
c68b9070 | 697 | *delim = 0; |
e7557af5 | 698 | np->remote_ip = in_aton(cur); |
c68b9070 | 699 | cur = delim + 1; |
1da177e4 | 700 | |
c68b9070 | 701 | if (*cur != 0) { |
1da177e4 | 702 | /* MAC address */ |
4940fc88 | 703 | if (!mac_pton(cur, np->remote_mac)) |
1da177e4 | 704 | goto parse_failed; |
1da177e4 LT |
705 | } |
706 | ||
0bcc1816 | 707 | netpoll_print_options(np); |
1da177e4 LT |
708 | |
709 | return 0; | |
710 | ||
711 | parse_failed: | |
e6ec2693 | 712 | np_info(np, "couldn't parse config at '%s'!\n", cur); |
1da177e4 LT |
713 | return -1; |
714 | } | |
9e34a5b5 | 715 | EXPORT_SYMBOL(netpoll_parse_options); |
1da177e4 | 716 | |
8fdd95ec | 717 | int __netpoll_setup(struct netpoll *np) |
1da177e4 | 718 | { |
8fdd95ec | 719 | struct net_device *ndev = np->dev; |
115c1d6e | 720 | struct netpoll_info *npinfo; |
4247e161 | 721 | const struct net_device_ops *ops; |
fbeec2e1 | 722 | unsigned long flags; |
b41848b6 | 723 | int err; |
1da177e4 | 724 | |
8fdd95ec HX |
725 | if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || |
726 | !ndev->netdev_ops->ndo_poll_controller) { | |
e6ec2693 JP |
727 | np_err(np, "%s doesn't support polling, aborting\n", |
728 | np->dev_name); | |
8fdd95ec HX |
729 | err = -ENOTSUPP; |
730 | goto out; | |
731 | } | |
732 | ||
733 | if (!ndev->npinfo) { | |
734 | npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL); | |
735 | if (!npinfo) { | |
736 | err = -ENOMEM; | |
737 | goto out; | |
738 | } | |
739 | ||
740 | npinfo->rx_flags = 0; | |
741 | INIT_LIST_HEAD(&npinfo->rx_np); | |
742 | ||
743 | spin_lock_init(&npinfo->rx_lock); | |
744 | skb_queue_head_init(&npinfo->arp_tx); | |
745 | skb_queue_head_init(&npinfo->txq); | |
746 | INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); | |
747 | ||
748 | atomic_set(&npinfo->refcnt, 1); | |
749 | ||
750 | ops = np->dev->netdev_ops; | |
751 | if (ops->ndo_netpoll_setup) { | |
752 | err = ops->ndo_netpoll_setup(ndev, npinfo); | |
753 | if (err) | |
754 | goto free_npinfo; | |
755 | } | |
756 | } else { | |
757 | npinfo = ndev->npinfo; | |
758 | atomic_inc(&npinfo->refcnt); | |
759 | } | |
760 | ||
761 | npinfo->netpoll = np; | |
762 | ||
763 | if (np->rx_hook) { | |
764 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
765 | npinfo->rx_flags |= NETPOLL_RX_ENABLED; | |
766 | list_add_tail(&np->rx, &npinfo->rx_np); | |
767 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
768 | } | |
769 | ||
770 | /* last thing to do is link it to the net device structure */ | |
cf778b00 | 771 | rcu_assign_pointer(ndev->npinfo, npinfo); |
8fdd95ec HX |
772 | |
773 | return 0; | |
774 | ||
775 | free_npinfo: | |
776 | kfree(npinfo); | |
777 | out: | |
778 | return err; | |
779 | } | |
780 | EXPORT_SYMBOL_GPL(__netpoll_setup); | |
781 | ||
782 | int netpoll_setup(struct netpoll *np) | |
783 | { | |
784 | struct net_device *ndev = NULL; | |
785 | struct in_device *in_dev; | |
786 | int err; | |
787 | ||
1da177e4 | 788 | if (np->dev_name) |
881d966b | 789 | ndev = dev_get_by_name(&init_net, np->dev_name); |
1da177e4 | 790 | if (!ndev) { |
e6ec2693 | 791 | np_err(np, "%s doesn't exist, aborting\n", np->dev_name); |
b41848b6 | 792 | return -ENODEV; |
1da177e4 LT |
793 | } |
794 | ||
0c1ad04a | 795 | if (ndev->master) { |
e6ec2693 | 796 | np_err(np, "%s is a slave device, aborting\n", np->dev_name); |
83fe32de DC |
797 | err = -EBUSY; |
798 | goto put; | |
0c1ad04a WC |
799 | } |
800 | ||
1da177e4 LT |
801 | if (!netif_running(ndev)) { |
802 | unsigned long atmost, atleast; | |
803 | ||
e6ec2693 | 804 | np_info(np, "device %s not up yet, forcing it\n", np->dev_name); |
1da177e4 | 805 | |
6756ae4b | 806 | rtnl_lock(); |
b41848b6 SH |
807 | err = dev_open(ndev); |
808 | rtnl_unlock(); | |
809 | ||
810 | if (err) { | |
e6ec2693 | 811 | np_err(np, "failed to open %s\n", ndev->name); |
dbaa1541 | 812 | goto put; |
1da177e4 | 813 | } |
1da177e4 LT |
814 | |
815 | atleast = jiffies + HZ/10; | |
bff38771 | 816 | atmost = jiffies + carrier_timeout * HZ; |
1da177e4 LT |
817 | while (!netif_carrier_ok(ndev)) { |
818 | if (time_after(jiffies, atmost)) { | |
e6ec2693 | 819 | np_notice(np, "timeout waiting for carrier\n"); |
1da177e4 LT |
820 | break; |
821 | } | |
1b614fb9 | 822 | msleep(1); |
1da177e4 LT |
823 | } |
824 | ||
825 | /* If carrier appears to come up instantly, we don't | |
826 | * trust it and pause so that we don't pump all our | |
827 | * queued console messages into the bitbucket. | |
828 | */ | |
829 | ||
830 | if (time_before(jiffies, atleast)) { | |
e6ec2693 | 831 | np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n"); |
1da177e4 LT |
832 | msleep(4000); |
833 | } | |
834 | } | |
835 | ||
1da177e4 LT |
836 | if (!np->local_ip) { |
837 | rcu_read_lock(); | |
e5ed6399 | 838 | in_dev = __in_dev_get_rcu(ndev); |
1da177e4 LT |
839 | |
840 | if (!in_dev || !in_dev->ifa_list) { | |
841 | rcu_read_unlock(); | |
e6ec2693 JP |
842 | np_err(np, "no IP address for %s, aborting\n", |
843 | np->dev_name); | |
b41848b6 | 844 | err = -EDESTADDRREQ; |
dbaa1541 | 845 | goto put; |
1da177e4 LT |
846 | } |
847 | ||
e7557af5 | 848 | np->local_ip = in_dev->ifa_list->ifa_local; |
1da177e4 | 849 | rcu_read_unlock(); |
e6ec2693 | 850 | np_info(np, "local IP %pI4\n", &np->local_ip); |
1da177e4 LT |
851 | } |
852 | ||
dbaa1541 HX |
853 | np->dev = ndev; |
854 | ||
855 | /* fill up the skb queue */ | |
856 | refill_skbs(); | |
857 | ||
858 | rtnl_lock(); | |
8fdd95ec | 859 | err = __netpoll_setup(np); |
dbaa1541 | 860 | rtnl_unlock(); |
53fb95d3 | 861 | |
8fdd95ec HX |
862 | if (err) |
863 | goto put; | |
864 | ||
1da177e4 LT |
865 | return 0; |
866 | ||
21edbb22 | 867 | put: |
1da177e4 | 868 | dev_put(ndev); |
b41848b6 | 869 | return err; |
1da177e4 | 870 | } |
9e34a5b5 | 871 | EXPORT_SYMBOL(netpoll_setup); |
1da177e4 | 872 | |
c68b9070 DM |
873 | static int __init netpoll_init(void) |
874 | { | |
a1bcfacd SH |
875 | skb_queue_head_init(&skb_pool); |
876 | return 0; | |
877 | } | |
878 | core_initcall(netpoll_init); | |
879 | ||
8fdd95ec | 880 | void __netpoll_cleanup(struct netpoll *np) |
1da177e4 | 881 | { |
fbeec2e1 JM |
882 | struct netpoll_info *npinfo; |
883 | unsigned long flags; | |
884 | ||
8fdd95ec HX |
885 | npinfo = np->dev->npinfo; |
886 | if (!npinfo) | |
dbaa1541 | 887 | return; |
93ec2c72 | 888 | |
8fdd95ec HX |
889 | if (!list_empty(&npinfo->rx_np)) { |
890 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
891 | list_del(&np->rx); | |
892 | if (list_empty(&npinfo->rx_np)) | |
893 | npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; | |
894 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
895 | } | |
de85d99e | 896 | |
8fdd95ec HX |
897 | if (atomic_dec_and_test(&npinfo->refcnt)) { |
898 | const struct net_device_ops *ops; | |
de85d99e | 899 | |
8fdd95ec HX |
900 | ops = np->dev->netdev_ops; |
901 | if (ops->ndo_netpoll_cleanup) | |
902 | ops->ndo_netpoll_cleanup(np->dev); | |
de85d99e | 903 | |
a9b3cd7f | 904 | RCU_INIT_POINTER(np->dev->npinfo, NULL); |
de85d99e | 905 | |
dbaa1541 HX |
906 | /* avoid racing with NAPI reading npinfo */ |
907 | synchronize_rcu_bh(); | |
93ec2c72 | 908 | |
dbaa1541 HX |
909 | skb_queue_purge(&npinfo->arp_tx); |
910 | skb_queue_purge(&npinfo->txq); | |
afe2c511 | 911 | cancel_delayed_work_sync(&npinfo->tx_work); |
93ec2c72 | 912 | |
dbaa1541 HX |
913 | /* clean after last, unfinished work */ |
914 | __skb_queue_purge(&npinfo->txq); | |
915 | kfree(npinfo); | |
115c1d6e | 916 | } |
8fdd95ec HX |
917 | } |
918 | EXPORT_SYMBOL_GPL(__netpoll_cleanup); | |
fbeec2e1 | 919 | |
8fdd95ec HX |
920 | void netpoll_cleanup(struct netpoll *np) |
921 | { | |
922 | if (!np->dev) | |
923 | return; | |
dbaa1541 | 924 | |
8fdd95ec HX |
925 | rtnl_lock(); |
926 | __netpoll_cleanup(np); | |
927 | rtnl_unlock(); | |
928 | ||
929 | dev_put(np->dev); | |
1da177e4 LT |
930 | np->dev = NULL; |
931 | } | |
9e34a5b5 | 932 | EXPORT_SYMBOL(netpoll_cleanup); |
1da177e4 LT |
933 | |
934 | int netpoll_trap(void) | |
935 | { | |
936 | return atomic_read(&trapped); | |
937 | } | |
9e34a5b5 | 938 | EXPORT_SYMBOL(netpoll_trap); |
1da177e4 LT |
939 | |
940 | void netpoll_set_trap(int trap) | |
941 | { | |
942 | if (trap) | |
943 | atomic_inc(&trapped); | |
944 | else | |
945 | atomic_dec(&trapped); | |
946 | } | |
1da177e4 | 947 | EXPORT_SYMBOL(netpoll_set_trap); |