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 | ||
bff38771 | 12 | #include <linux/moduleparam.h> |
1da177e4 LT |
13 | #include <linux/netdevice.h> |
14 | #include <linux/etherdevice.h> | |
15 | #include <linux/string.h> | |
14c85021 | 16 | #include <linux/if_arp.h> |
1da177e4 LT |
17 | #include <linux/inetdevice.h> |
18 | #include <linux/inet.h> | |
19 | #include <linux/interrupt.h> | |
20 | #include <linux/netpoll.h> | |
21 | #include <linux/sched.h> | |
22 | #include <linux/delay.h> | |
23 | #include <linux/rcupdate.h> | |
24 | #include <linux/workqueue.h> | |
5a0e3ad6 | 25 | #include <linux/slab.h> |
1da177e4 LT |
26 | #include <net/tcp.h> |
27 | #include <net/udp.h> | |
28 | #include <asm/unaligned.h> | |
9cbc1cb8 | 29 | #include <trace/events/napi.h> |
1da177e4 LT |
30 | |
31 | /* | |
32 | * We maintain a small pool of fully-sized skbs, to make sure the | |
33 | * message gets out even in extreme OOM situations. | |
34 | */ | |
35 | ||
36 | #define MAX_UDP_CHUNK 1460 | |
37 | #define MAX_SKBS 32 | |
1da177e4 | 38 | |
a1bcfacd | 39 | static struct sk_buff_head skb_pool; |
1da177e4 LT |
40 | |
41 | static atomic_t trapped; | |
42 | ||
2bdfe0ba | 43 | #define USEC_PER_POLL 50 |
d9452e9f DM |
44 | #define NETPOLL_RX_ENABLED 1 |
45 | #define NETPOLL_RX_DROP 2 | |
1da177e4 LT |
46 | |
47 | #define MAX_SKB_SIZE \ | |
48 | (MAX_UDP_CHUNK + sizeof(struct udphdr) + \ | |
49 | sizeof(struct iphdr) + sizeof(struct ethhdr)) | |
50 | ||
3578b0c8 | 51 | static void zap_completion_queue(void); |
068c6e98 | 52 | static void arp_reply(struct sk_buff *skb); |
1da177e4 | 53 | |
bff38771 AV |
54 | static unsigned int carrier_timeout = 4; |
55 | module_param(carrier_timeout, uint, 0644); | |
56 | ||
c4028958 | 57 | static void queue_process(struct work_struct *work) |
1da177e4 | 58 | { |
4c1ac1b4 DH |
59 | struct netpoll_info *npinfo = |
60 | container_of(work, struct netpoll_info, tx_work.work); | |
1da177e4 | 61 | struct sk_buff *skb; |
3640543d | 62 | unsigned long flags; |
1da177e4 | 63 | |
6c43ff18 SH |
64 | while ((skb = skb_dequeue(&npinfo->txq))) { |
65 | struct net_device *dev = skb->dev; | |
00829823 | 66 | const struct net_device_ops *ops = dev->netdev_ops; |
fd2ea0a7 | 67 | struct netdev_queue *txq; |
1da177e4 | 68 | |
6c43ff18 SH |
69 | if (!netif_device_present(dev) || !netif_running(dev)) { |
70 | __kfree_skb(skb); | |
71 | continue; | |
72 | } | |
1da177e4 | 73 | |
fd2ea0a7 DM |
74 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
75 | ||
3640543d | 76 | local_irq_save(flags); |
fd2ea0a7 | 77 | __netif_tx_lock(txq, smp_processor_id()); |
5a0d2268 | 78 | if (netif_tx_queue_frozen_or_stopped(txq) || |
00829823 | 79 | ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { |
6c43ff18 | 80 | skb_queue_head(&npinfo->txq, skb); |
fd2ea0a7 | 81 | __netif_tx_unlock(txq); |
3640543d | 82 | local_irq_restore(flags); |
1da177e4 | 83 | |
25442caf | 84 | schedule_delayed_work(&npinfo->tx_work, HZ/10); |
6c43ff18 SH |
85 | return; |
86 | } | |
fd2ea0a7 | 87 | __netif_tx_unlock(txq); |
3640543d | 88 | local_irq_restore(flags); |
1da177e4 | 89 | } |
1da177e4 LT |
90 | } |
91 | ||
b51655b9 AV |
92 | static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, |
93 | unsigned short ulen, __be32 saddr, __be32 daddr) | |
1da177e4 | 94 | { |
d6f5493c | 95 | __wsum psum; |
fb286bb2 | 96 | |
60476372 | 97 | if (uh->check == 0 || skb_csum_unnecessary(skb)) |
1da177e4 LT |
98 | return 0; |
99 | ||
fb286bb2 HX |
100 | psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); |
101 | ||
84fa7933 | 102 | if (skb->ip_summed == CHECKSUM_COMPLETE && |
d3bc23e7 | 103 | !csum_fold(csum_add(psum, skb->csum))) |
fb286bb2 | 104 | return 0; |
1da177e4 | 105 | |
fb286bb2 | 106 | skb->csum = psum; |
1da177e4 | 107 | |
fb286bb2 | 108 | return __skb_checksum_complete(skb); |
1da177e4 LT |
109 | } |
110 | ||
111 | /* | |
112 | * Check whether delayed processing was scheduled for our NIC. If so, | |
113 | * we attempt to grab the poll lock and use ->poll() to pump the card. | |
114 | * If this fails, either we've recursed in ->poll() or it's already | |
115 | * running on another CPU. | |
116 | * | |
117 | * Note: we don't mask interrupts with this lock because we're using | |
118 | * trylock here and interrupts are already disabled in the softirq | |
119 | * case. Further, we test the poll_owner to avoid recursion on UP | |
120 | * systems where the lock doesn't exist. | |
121 | * | |
122 | * In cases where there is bi-directional communications, reading only | |
123 | * one message at a time can lead to packets being dropped by the | |
124 | * network adapter, forcing superfluous retries and possibly timeouts. | |
125 | * Thus, we set our budget to greater than 1. | |
126 | */ | |
0a7606c1 DM |
127 | static int poll_one_napi(struct netpoll_info *npinfo, |
128 | struct napi_struct *napi, int budget) | |
129 | { | |
130 | int work; | |
131 | ||
132 | /* net_rx_action's ->poll() invocations and our's are | |
133 | * synchronized by this test which is only made while | |
134 | * holding the napi->poll_lock. | |
135 | */ | |
136 | if (!test_bit(NAPI_STATE_SCHED, &napi->state)) | |
137 | return budget; | |
138 | ||
d9452e9f | 139 | npinfo->rx_flags |= NETPOLL_RX_DROP; |
0a7606c1 | 140 | atomic_inc(&trapped); |
7b363e44 | 141 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 DM |
142 | |
143 | work = napi->poll(napi, budget); | |
7d18f114 | 144 | trace_napi_poll(napi); |
0a7606c1 | 145 | |
7b363e44 | 146 | clear_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 | 147 | atomic_dec(&trapped); |
d9452e9f | 148 | npinfo->rx_flags &= ~NETPOLL_RX_DROP; |
0a7606c1 DM |
149 | |
150 | return budget - work; | |
151 | } | |
152 | ||
5106930b | 153 | static void poll_napi(struct net_device *dev) |
1da177e4 | 154 | { |
bea3348e | 155 | struct napi_struct *napi; |
1da177e4 LT |
156 | int budget = 16; |
157 | ||
f13d493d | 158 | list_for_each_entry(napi, &dev->napi_list, dev_list) { |
0a7606c1 | 159 | if (napi->poll_owner != smp_processor_id() && |
bea3348e | 160 | spin_trylock(&napi->poll_lock)) { |
5106930b | 161 | budget = poll_one_napi(dev->npinfo, napi, budget); |
bea3348e | 162 | spin_unlock(&napi->poll_lock); |
0a7606c1 DM |
163 | |
164 | if (!budget) | |
165 | break; | |
bea3348e | 166 | } |
1da177e4 LT |
167 | } |
168 | } | |
169 | ||
068c6e98 NH |
170 | static void service_arp_queue(struct netpoll_info *npi) |
171 | { | |
5106930b SH |
172 | if (npi) { |
173 | struct sk_buff *skb; | |
068c6e98 | 174 | |
5106930b SH |
175 | while ((skb = skb_dequeue(&npi->arp_tx))) |
176 | arp_reply(skb); | |
068c6e98 | 177 | } |
068c6e98 NH |
178 | } |
179 | ||
0e34e931 | 180 | void netpoll_poll_dev(struct net_device *dev) |
1da177e4 | 181 | { |
5e392739 | 182 | const struct net_device_ops *ops; |
5106930b | 183 | |
5e392739 PE |
184 | if (!dev || !netif_running(dev)) |
185 | return; | |
186 | ||
187 | ops = dev->netdev_ops; | |
188 | if (!ops->ndo_poll_controller) | |
1da177e4 LT |
189 | return; |
190 | ||
191 | /* Process pending work on NIC */ | |
d314774c | 192 | ops->ndo_poll_controller(dev); |
5106930b SH |
193 | |
194 | poll_napi(dev); | |
1da177e4 | 195 | |
5106930b | 196 | service_arp_queue(dev->npinfo); |
068c6e98 | 197 | |
3578b0c8 | 198 | zap_completion_queue(); |
1da177e4 | 199 | } |
9e34a5b5 | 200 | EXPORT_SYMBOL(netpoll_poll_dev); |
1da177e4 | 201 | |
0e34e931 WC |
202 | void netpoll_poll(struct netpoll *np) |
203 | { | |
204 | netpoll_poll_dev(np->dev); | |
205 | } | |
9e34a5b5 | 206 | EXPORT_SYMBOL(netpoll_poll); |
0e34e931 | 207 | |
1da177e4 LT |
208 | static void refill_skbs(void) |
209 | { | |
210 | struct sk_buff *skb; | |
211 | unsigned long flags; | |
212 | ||
a1bcfacd SH |
213 | spin_lock_irqsave(&skb_pool.lock, flags); |
214 | while (skb_pool.qlen < MAX_SKBS) { | |
1da177e4 LT |
215 | skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); |
216 | if (!skb) | |
217 | break; | |
218 | ||
a1bcfacd | 219 | __skb_queue_tail(&skb_pool, skb); |
1da177e4 | 220 | } |
a1bcfacd | 221 | spin_unlock_irqrestore(&skb_pool.lock, flags); |
1da177e4 LT |
222 | } |
223 | ||
3578b0c8 DM |
224 | static void zap_completion_queue(void) |
225 | { | |
226 | unsigned long flags; | |
227 | struct softnet_data *sd = &get_cpu_var(softnet_data); | |
228 | ||
229 | if (sd->completion_queue) { | |
230 | struct sk_buff *clist; | |
231 | ||
232 | local_irq_save(flags); | |
233 | clist = sd->completion_queue; | |
234 | sd->completion_queue = NULL; | |
235 | local_irq_restore(flags); | |
236 | ||
237 | while (clist != NULL) { | |
238 | struct sk_buff *skb = clist; | |
239 | clist = clist->next; | |
240 | if (skb->destructor) { | |
241 | atomic_inc(&skb->users); | |
242 | dev_kfree_skb_any(skb); /* put this one back */ | |
243 | } else { | |
244 | __kfree_skb(skb); | |
245 | } | |
246 | } | |
247 | } | |
248 | ||
249 | put_cpu_var(softnet_data); | |
250 | } | |
251 | ||
a1bcfacd | 252 | static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) |
1da177e4 | 253 | { |
a1bcfacd SH |
254 | int count = 0; |
255 | struct sk_buff *skb; | |
1da177e4 | 256 | |
3578b0c8 | 257 | zap_completion_queue(); |
a1bcfacd | 258 | refill_skbs(); |
1da177e4 | 259 | repeat: |
1da177e4 LT |
260 | |
261 | skb = alloc_skb(len, GFP_ATOMIC); | |
a1bcfacd SH |
262 | if (!skb) |
263 | skb = skb_dequeue(&skb_pool); | |
1da177e4 LT |
264 | |
265 | if (!skb) { | |
a1bcfacd SH |
266 | if (++count < 10) { |
267 | netpoll_poll(np); | |
268 | goto repeat; | |
1da177e4 | 269 | } |
a1bcfacd | 270 | return NULL; |
1da177e4 LT |
271 | } |
272 | ||
273 | atomic_set(&skb->users, 1); | |
274 | skb_reserve(skb, reserve); | |
275 | return skb; | |
276 | } | |
277 | ||
bea3348e SH |
278 | static int netpoll_owner_active(struct net_device *dev) |
279 | { | |
280 | struct napi_struct *napi; | |
281 | ||
282 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
283 | if (napi->poll_owner == smp_processor_id()) | |
284 | return 1; | |
285 | } | |
286 | return 0; | |
287 | } | |
288 | ||
c2355e1a NH |
289 | void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, |
290 | struct net_device *dev) | |
1da177e4 | 291 | { |
2bdfe0ba SH |
292 | int status = NETDEV_TX_BUSY; |
293 | unsigned long tries; | |
00829823 | 294 | const struct net_device_ops *ops = dev->netdev_ops; |
de85d99e | 295 | /* It is up to the caller to keep npinfo alive. */ |
4ec93edb | 296 | struct netpoll_info *npinfo = np->dev->npinfo; |
2bdfe0ba | 297 | |
4ec93edb YH |
298 | if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { |
299 | __kfree_skb(skb); | |
300 | return; | |
301 | } | |
2bdfe0ba SH |
302 | |
303 | /* don't get messages out of order, and no recursion */ | |
bea3348e | 304 | if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { |
fd2ea0a7 | 305 | struct netdev_queue *txq; |
a49f99ff AM |
306 | unsigned long flags; |
307 | ||
fd2ea0a7 DM |
308 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
309 | ||
a49f99ff | 310 | local_irq_save(flags); |
0db3dc73 SH |
311 | /* try until next clock tick */ |
312 | for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; | |
313 | tries > 0; --tries) { | |
fd2ea0a7 | 314 | if (__netif_tx_trylock(txq)) { |
08baf561 | 315 | if (!netif_tx_queue_stopped(txq)) { |
00829823 | 316 | status = ops->ndo_start_xmit(skb, dev); |
08baf561 ED |
317 | if (status == NETDEV_TX_OK) |
318 | txq_trans_update(txq); | |
319 | } | |
fd2ea0a7 | 320 | __netif_tx_unlock(txq); |
e37b8d93 AM |
321 | |
322 | if (status == NETDEV_TX_OK) | |
323 | break; | |
324 | ||
e37b8d93 | 325 | } |
0db3dc73 SH |
326 | |
327 | /* tickle device maybe there is some cleanup */ | |
328 | netpoll_poll(np); | |
329 | ||
330 | udelay(USEC_PER_POLL); | |
0db1d6fc | 331 | } |
79b1bee8 DD |
332 | |
333 | WARN_ONCE(!irqs_disabled(), | |
334 | "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", | |
335 | dev->name, ops->ndo_start_xmit); | |
336 | ||
a49f99ff | 337 | local_irq_restore(flags); |
1da177e4 | 338 | } |
1da177e4 | 339 | |
2bdfe0ba | 340 | if (status != NETDEV_TX_OK) { |
5de4a473 | 341 | skb_queue_tail(&npinfo->txq, skb); |
4c1ac1b4 | 342 | schedule_delayed_work(&npinfo->tx_work,0); |
1da177e4 | 343 | } |
1da177e4 | 344 | } |
c2355e1a | 345 | EXPORT_SYMBOL(netpoll_send_skb_on_dev); |
1da177e4 LT |
346 | |
347 | void netpoll_send_udp(struct netpoll *np, const char *msg, int len) | |
348 | { | |
349 | int total_len, eth_len, ip_len, udp_len; | |
350 | struct sk_buff *skb; | |
351 | struct udphdr *udph; | |
352 | struct iphdr *iph; | |
353 | struct ethhdr *eth; | |
354 | ||
355 | udp_len = len + sizeof(*udph); | |
356 | ip_len = eth_len = udp_len + sizeof(*iph); | |
357 | total_len = eth_len + ETH_HLEN + NET_IP_ALIGN; | |
358 | ||
359 | skb = find_skb(np, total_len, total_len - len); | |
360 | if (!skb) | |
361 | return; | |
362 | ||
27d7ff46 | 363 | skb_copy_to_linear_data(skb, msg, len); |
1da177e4 LT |
364 | skb->len += len; |
365 | ||
4bedb452 ACM |
366 | skb_push(skb, sizeof(*udph)); |
367 | skb_reset_transport_header(skb); | |
368 | udph = udp_hdr(skb); | |
1da177e4 LT |
369 | udph->source = htons(np->local_port); |
370 | udph->dest = htons(np->remote_port); | |
371 | udph->len = htons(udp_len); | |
372 | udph->check = 0; | |
e7557af5 HH |
373 | udph->check = csum_tcpudp_magic(np->local_ip, |
374 | np->remote_ip, | |
8e365eec | 375 | udp_len, IPPROTO_UDP, |
07f0757a | 376 | csum_partial(udph, udp_len, 0)); |
8e365eec | 377 | if (udph->check == 0) |
5e57dff2 | 378 | udph->check = CSUM_MANGLED_0; |
1da177e4 | 379 | |
e2d1bca7 ACM |
380 | skb_push(skb, sizeof(*iph)); |
381 | skb_reset_network_header(skb); | |
eddc9ec5 | 382 | iph = ip_hdr(skb); |
1da177e4 LT |
383 | |
384 | /* iph->version = 4; iph->ihl = 5; */ | |
385 | put_unaligned(0x45, (unsigned char *)iph); | |
386 | iph->tos = 0; | |
387 | put_unaligned(htons(ip_len), &(iph->tot_len)); | |
388 | iph->id = 0; | |
389 | iph->frag_off = 0; | |
390 | iph->ttl = 64; | |
391 | iph->protocol = IPPROTO_UDP; | |
392 | iph->check = 0; | |
e7557af5 HH |
393 | put_unaligned(np->local_ip, &(iph->saddr)); |
394 | put_unaligned(np->remote_ip, &(iph->daddr)); | |
1da177e4 LT |
395 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); |
396 | ||
397 | eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); | |
459a98ed | 398 | skb_reset_mac_header(skb); |
206daaf7 | 399 | skb->protocol = eth->h_proto = htons(ETH_P_IP); |
09538641 SH |
400 | memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); |
401 | memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); | |
1da177e4 LT |
402 | |
403 | skb->dev = np->dev; | |
404 | ||
405 | netpoll_send_skb(np, skb); | |
406 | } | |
9e34a5b5 | 407 | EXPORT_SYMBOL(netpoll_send_udp); |
1da177e4 LT |
408 | |
409 | static void arp_reply(struct sk_buff *skb) | |
410 | { | |
115c1d6e | 411 | struct netpoll_info *npinfo = skb->dev->npinfo; |
1da177e4 LT |
412 | struct arphdr *arp; |
413 | unsigned char *arp_ptr; | |
414 | int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; | |
252e3346 | 415 | __be32 sip, tip; |
47bbec02 | 416 | unsigned char *sha; |
1da177e4 | 417 | struct sk_buff *send_skb; |
508e14b4 DB |
418 | struct netpoll *np, *tmp; |
419 | unsigned long flags; | |
420 | int hits = 0; | |
421 | ||
422 | if (list_empty(&npinfo->rx_np)) | |
423 | return; | |
424 | ||
425 | /* Before checking the packet, we do some early | |
426 | inspection whether this is interesting at all */ | |
427 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
428 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
429 | if (np->dev == skb->dev) | |
430 | hits++; | |
431 | } | |
432 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 | 433 | |
508e14b4 DB |
434 | /* No netpoll struct is using this dev */ |
435 | if (!hits) | |
115c1d6e | 436 | return; |
1da177e4 LT |
437 | |
438 | /* No arp on this interface */ | |
439 | if (skb->dev->flags & IFF_NOARP) | |
440 | return; | |
441 | ||
988b7050 | 442 | if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) |
1da177e4 LT |
443 | return; |
444 | ||
c1d2bbe1 | 445 | skb_reset_network_header(skb); |
badff6d0 | 446 | skb_reset_transport_header(skb); |
d0a92be0 | 447 | arp = arp_hdr(skb); |
1da177e4 LT |
448 | |
449 | if ((arp->ar_hrd != htons(ARPHRD_ETHER) && | |
450 | arp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
451 | arp->ar_pro != htons(ETH_P_IP) || | |
452 | arp->ar_op != htons(ARPOP_REQUEST)) | |
453 | return; | |
454 | ||
47bbec02 NH |
455 | arp_ptr = (unsigned char *)(arp+1); |
456 | /* save the location of the src hw addr */ | |
457 | sha = arp_ptr; | |
458 | arp_ptr += skb->dev->addr_len; | |
1da177e4 | 459 | memcpy(&sip, arp_ptr, 4); |
47bbec02 | 460 | arp_ptr += 4; |
508e14b4 DB |
461 | /* If we actually cared about dst hw addr, |
462 | it would get copied here */ | |
47bbec02 | 463 | arp_ptr += skb->dev->addr_len; |
1da177e4 LT |
464 | memcpy(&tip, arp_ptr, 4); |
465 | ||
466 | /* Should we ignore arp? */ | |
508e14b4 | 467 | if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) |
1da177e4 LT |
468 | return; |
469 | ||
988b7050 | 470 | size = arp_hdr_len(skb->dev); |
1da177e4 | 471 | |
508e14b4 DB |
472 | spin_lock_irqsave(&npinfo->rx_lock, flags); |
473 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
474 | if (tip != np->local_ip) | |
475 | continue; | |
1da177e4 | 476 | |
508e14b4 DB |
477 | send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev), |
478 | LL_RESERVED_SPACE(np->dev)); | |
479 | if (!send_skb) | |
480 | continue; | |
1da177e4 | 481 | |
508e14b4 DB |
482 | skb_reset_network_header(send_skb); |
483 | arp = (struct arphdr *) skb_put(send_skb, size); | |
484 | send_skb->dev = skb->dev; | |
485 | send_skb->protocol = htons(ETH_P_ARP); | |
1da177e4 | 486 | |
508e14b4 DB |
487 | /* Fill the device header for the ARP frame */ |
488 | if (dev_hard_header(send_skb, skb->dev, ptype, | |
489 | sha, np->dev->dev_addr, | |
490 | send_skb->len) < 0) { | |
491 | kfree_skb(send_skb); | |
492 | continue; | |
493 | } | |
1da177e4 | 494 | |
508e14b4 DB |
495 | /* |
496 | * Fill out the arp protocol part. | |
497 | * | |
498 | * we only support ethernet device type, | |
499 | * which (according to RFC 1390) should | |
500 | * always equal 1 (Ethernet). | |
501 | */ | |
1da177e4 | 502 | |
508e14b4 DB |
503 | arp->ar_hrd = htons(np->dev->type); |
504 | arp->ar_pro = htons(ETH_P_IP); | |
505 | arp->ar_hln = np->dev->addr_len; | |
506 | arp->ar_pln = 4; | |
507 | arp->ar_op = htons(type); | |
508 | ||
509 | arp_ptr = (unsigned char *)(arp + 1); | |
510 | memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); | |
511 | arp_ptr += np->dev->addr_len; | |
512 | memcpy(arp_ptr, &tip, 4); | |
513 | arp_ptr += 4; | |
514 | memcpy(arp_ptr, sha, np->dev->addr_len); | |
515 | arp_ptr += np->dev->addr_len; | |
516 | memcpy(arp_ptr, &sip, 4); | |
517 | ||
518 | netpoll_send_skb(np, send_skb); | |
519 | ||
520 | /* If there are several rx_hooks for the same address, | |
521 | we're fine by sending a single reply */ | |
522 | break; | |
523 | } | |
524 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 LT |
525 | } |
526 | ||
527 | int __netpoll_rx(struct sk_buff *skb) | |
528 | { | |
529 | int proto, len, ulen; | |
508e14b4 | 530 | int hits = 0; |
1da177e4 LT |
531 | struct iphdr *iph; |
532 | struct udphdr *uh; | |
508e14b4 DB |
533 | struct netpoll_info *npinfo = skb->dev->npinfo; |
534 | struct netpoll *np, *tmp; | |
068c6e98 | 535 | |
508e14b4 | 536 | if (list_empty(&npinfo->rx_np)) |
1da177e4 | 537 | goto out; |
508e14b4 | 538 | |
1da177e4 LT |
539 | if (skb->dev->type != ARPHRD_ETHER) |
540 | goto out; | |
541 | ||
d9452e9f | 542 | /* check if netpoll clients need ARP */ |
724800d6 | 543 | if (skb->protocol == htons(ETH_P_ARP) && |
1da177e4 | 544 | atomic_read(&trapped)) { |
508e14b4 | 545 | skb_queue_tail(&npinfo->arp_tx, skb); |
1da177e4 LT |
546 | return 1; |
547 | } | |
548 | ||
549 | proto = ntohs(eth_hdr(skb)->h_proto); | |
550 | if (proto != ETH_P_IP) | |
551 | goto out; | |
552 | if (skb->pkt_type == PACKET_OTHERHOST) | |
553 | goto out; | |
554 | if (skb_shared(skb)) | |
555 | goto out; | |
556 | ||
557 | iph = (struct iphdr *)skb->data; | |
558 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) | |
559 | goto out; | |
560 | if (iph->ihl < 5 || iph->version != 4) | |
561 | goto out; | |
562 | if (!pskb_may_pull(skb, iph->ihl*4)) | |
563 | goto out; | |
564 | if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) | |
565 | goto out; | |
566 | ||
567 | len = ntohs(iph->tot_len); | |
568 | if (skb->len < len || len < iph->ihl*4) | |
569 | goto out; | |
570 | ||
5e7d7fa5 AL |
571 | /* |
572 | * Our transport medium may have padded the buffer out. | |
573 | * Now We trim to the true length of the frame. | |
574 | */ | |
575 | if (pskb_trim_rcsum(skb, len)) | |
576 | goto out; | |
577 | ||
1da177e4 LT |
578 | if (iph->protocol != IPPROTO_UDP) |
579 | goto out; | |
580 | ||
581 | len -= iph->ihl*4; | |
582 | uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); | |
583 | ulen = ntohs(uh->len); | |
584 | ||
585 | if (ulen != len) | |
586 | goto out; | |
fb286bb2 | 587 | if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) |
1da177e4 | 588 | goto out; |
1da177e4 | 589 | |
508e14b4 DB |
590 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { |
591 | if (np->local_ip && np->local_ip != iph->daddr) | |
592 | continue; | |
593 | if (np->remote_ip && np->remote_ip != iph->saddr) | |
594 | continue; | |
595 | if (np->local_port && np->local_port != ntohs(uh->dest)) | |
596 | continue; | |
597 | ||
598 | np->rx_hook(np, ntohs(uh->source), | |
599 | (char *)(uh+1), | |
600 | ulen - sizeof(struct udphdr)); | |
601 | hits++; | |
602 | } | |
603 | ||
604 | if (!hits) | |
605 | goto out; | |
1da177e4 LT |
606 | |
607 | kfree_skb(skb); | |
608 | return 1; | |
609 | ||
610 | out: | |
611 | if (atomic_read(&trapped)) { | |
612 | kfree_skb(skb); | |
613 | return 1; | |
614 | } | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
0bcc1816 SS |
619 | void netpoll_print_options(struct netpoll *np) |
620 | { | |
621 | printk(KERN_INFO "%s: local port %d\n", | |
622 | np->name, np->local_port); | |
e7557af5 HH |
623 | printk(KERN_INFO "%s: local IP %pI4\n", |
624 | np->name, &np->local_ip); | |
5fc05f87 | 625 | printk(KERN_INFO "%s: interface '%s'\n", |
0bcc1816 SS |
626 | np->name, np->dev_name); |
627 | printk(KERN_INFO "%s: remote port %d\n", | |
628 | np->name, np->remote_port); | |
e7557af5 HH |
629 | printk(KERN_INFO "%s: remote IP %pI4\n", |
630 | np->name, &np->remote_ip); | |
e174961c JB |
631 | printk(KERN_INFO "%s: remote ethernet address %pM\n", |
632 | np->name, np->remote_mac); | |
0bcc1816 | 633 | } |
9e34a5b5 | 634 | EXPORT_SYMBOL(netpoll_print_options); |
0bcc1816 | 635 | |
1da177e4 LT |
636 | int netpoll_parse_options(struct netpoll *np, char *opt) |
637 | { | |
638 | char *cur=opt, *delim; | |
639 | ||
c68b9070 | 640 | if (*cur != '@') { |
1da177e4 LT |
641 | if ((delim = strchr(cur, '@')) == NULL) |
642 | goto parse_failed; | |
c68b9070 DM |
643 | *delim = 0; |
644 | np->local_port = simple_strtol(cur, NULL, 10); | |
645 | cur = delim; | |
1da177e4 LT |
646 | } |
647 | cur++; | |
1da177e4 | 648 | |
c68b9070 | 649 | if (*cur != '/') { |
1da177e4 LT |
650 | if ((delim = strchr(cur, '/')) == NULL) |
651 | goto parse_failed; | |
c68b9070 | 652 | *delim = 0; |
e7557af5 | 653 | np->local_ip = in_aton(cur); |
c68b9070 | 654 | cur = delim; |
1da177e4 LT |
655 | } |
656 | cur++; | |
657 | ||
c68b9070 | 658 | if (*cur != ',') { |
1da177e4 LT |
659 | /* parse out dev name */ |
660 | if ((delim = strchr(cur, ',')) == NULL) | |
661 | goto parse_failed; | |
c68b9070 | 662 | *delim = 0; |
1da177e4 | 663 | strlcpy(np->dev_name, cur, sizeof(np->dev_name)); |
c68b9070 | 664 | cur = delim; |
1da177e4 LT |
665 | } |
666 | cur++; | |
667 | ||
c68b9070 | 668 | if (*cur != '@') { |
1da177e4 LT |
669 | /* dst port */ |
670 | if ((delim = strchr(cur, '@')) == NULL) | |
671 | goto parse_failed; | |
c68b9070 | 672 | *delim = 0; |
5fc05f87 AW |
673 | if (*cur == ' ' || *cur == '\t') |
674 | printk(KERN_INFO "%s: warning: whitespace" | |
675 | "is not allowed\n", np->name); | |
c68b9070 DM |
676 | np->remote_port = simple_strtol(cur, NULL, 10); |
677 | cur = delim; | |
1da177e4 LT |
678 | } |
679 | cur++; | |
1da177e4 LT |
680 | |
681 | /* dst ip */ | |
682 | if ((delim = strchr(cur, '/')) == NULL) | |
683 | goto parse_failed; | |
c68b9070 | 684 | *delim = 0; |
e7557af5 | 685 | np->remote_ip = in_aton(cur); |
c68b9070 | 686 | cur = delim + 1; |
1da177e4 | 687 | |
c68b9070 | 688 | if (*cur != 0) { |
1da177e4 LT |
689 | /* MAC address */ |
690 | if ((delim = strchr(cur, ':')) == NULL) | |
691 | goto parse_failed; | |
c68b9070 DM |
692 | *delim = 0; |
693 | np->remote_mac[0] = simple_strtol(cur, NULL, 16); | |
694 | cur = delim + 1; | |
1da177e4 LT |
695 | if ((delim = strchr(cur, ':')) == NULL) |
696 | goto parse_failed; | |
c68b9070 DM |
697 | *delim = 0; |
698 | np->remote_mac[1] = simple_strtol(cur, NULL, 16); | |
699 | cur = delim + 1; | |
1da177e4 LT |
700 | if ((delim = strchr(cur, ':')) == NULL) |
701 | goto parse_failed; | |
c68b9070 DM |
702 | *delim = 0; |
703 | np->remote_mac[2] = simple_strtol(cur, NULL, 16); | |
704 | cur = delim + 1; | |
1da177e4 LT |
705 | if ((delim = strchr(cur, ':')) == NULL) |
706 | goto parse_failed; | |
c68b9070 DM |
707 | *delim = 0; |
708 | np->remote_mac[3] = simple_strtol(cur, NULL, 16); | |
709 | cur = delim + 1; | |
1da177e4 LT |
710 | if ((delim = strchr(cur, ':')) == NULL) |
711 | goto parse_failed; | |
c68b9070 DM |
712 | *delim = 0; |
713 | np->remote_mac[4] = simple_strtol(cur, NULL, 16); | |
714 | cur = delim + 1; | |
715 | np->remote_mac[5] = simple_strtol(cur, NULL, 16); | |
1da177e4 LT |
716 | } |
717 | ||
0bcc1816 | 718 | netpoll_print_options(np); |
1da177e4 LT |
719 | |
720 | return 0; | |
721 | ||
722 | parse_failed: | |
5fc05f87 | 723 | printk(KERN_INFO "%s: couldn't parse config at '%s'!\n", |
1da177e4 LT |
724 | np->name, cur); |
725 | return -1; | |
726 | } | |
9e34a5b5 | 727 | EXPORT_SYMBOL(netpoll_parse_options); |
1da177e4 | 728 | |
8fdd95ec | 729 | int __netpoll_setup(struct netpoll *np) |
1da177e4 | 730 | { |
8fdd95ec | 731 | struct net_device *ndev = np->dev; |
115c1d6e | 732 | struct netpoll_info *npinfo; |
4247e161 | 733 | const struct net_device_ops *ops; |
fbeec2e1 | 734 | unsigned long flags; |
b41848b6 | 735 | int err; |
1da177e4 | 736 | |
8fdd95ec HX |
737 | if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || |
738 | !ndev->netdev_ops->ndo_poll_controller) { | |
739 | printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n", | |
740 | np->name, np->dev_name); | |
741 | err = -ENOTSUPP; | |
742 | goto out; | |
743 | } | |
744 | ||
745 | if (!ndev->npinfo) { | |
746 | npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL); | |
747 | if (!npinfo) { | |
748 | err = -ENOMEM; | |
749 | goto out; | |
750 | } | |
751 | ||
752 | npinfo->rx_flags = 0; | |
753 | INIT_LIST_HEAD(&npinfo->rx_np); | |
754 | ||
755 | spin_lock_init(&npinfo->rx_lock); | |
756 | skb_queue_head_init(&npinfo->arp_tx); | |
757 | skb_queue_head_init(&npinfo->txq); | |
758 | INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); | |
759 | ||
760 | atomic_set(&npinfo->refcnt, 1); | |
761 | ||
762 | ops = np->dev->netdev_ops; | |
763 | if (ops->ndo_netpoll_setup) { | |
764 | err = ops->ndo_netpoll_setup(ndev, npinfo); | |
765 | if (err) | |
766 | goto free_npinfo; | |
767 | } | |
768 | } else { | |
769 | npinfo = ndev->npinfo; | |
770 | atomic_inc(&npinfo->refcnt); | |
771 | } | |
772 | ||
773 | npinfo->netpoll = np; | |
774 | ||
775 | if (np->rx_hook) { | |
776 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
777 | npinfo->rx_flags |= NETPOLL_RX_ENABLED; | |
778 | list_add_tail(&np->rx, &npinfo->rx_np); | |
779 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
780 | } | |
781 | ||
782 | /* last thing to do is link it to the net device structure */ | |
783 | rcu_assign_pointer(ndev->npinfo, npinfo); | |
8fdd95ec HX |
784 | |
785 | return 0; | |
786 | ||
787 | free_npinfo: | |
788 | kfree(npinfo); | |
789 | out: | |
790 | return err; | |
791 | } | |
792 | EXPORT_SYMBOL_GPL(__netpoll_setup); | |
793 | ||
794 | int netpoll_setup(struct netpoll *np) | |
795 | { | |
796 | struct net_device *ndev = NULL; | |
797 | struct in_device *in_dev; | |
798 | int err; | |
799 | ||
1da177e4 | 800 | if (np->dev_name) |
881d966b | 801 | ndev = dev_get_by_name(&init_net, np->dev_name); |
1da177e4 LT |
802 | if (!ndev) { |
803 | printk(KERN_ERR "%s: %s doesn't exist, aborting.\n", | |
804 | np->name, np->dev_name); | |
b41848b6 | 805 | return -ENODEV; |
1da177e4 LT |
806 | } |
807 | ||
1da177e4 LT |
808 | if (!netif_running(ndev)) { |
809 | unsigned long atmost, atleast; | |
810 | ||
811 | printk(KERN_INFO "%s: device %s not up yet, forcing it\n", | |
812 | np->name, np->dev_name); | |
813 | ||
6756ae4b | 814 | rtnl_lock(); |
b41848b6 SH |
815 | err = dev_open(ndev); |
816 | rtnl_unlock(); | |
817 | ||
818 | if (err) { | |
1da177e4 | 819 | printk(KERN_ERR "%s: failed to open %s\n", |
b41848b6 | 820 | np->name, ndev->name); |
dbaa1541 | 821 | goto put; |
1da177e4 | 822 | } |
1da177e4 LT |
823 | |
824 | atleast = jiffies + HZ/10; | |
bff38771 | 825 | atmost = jiffies + carrier_timeout * HZ; |
1da177e4 LT |
826 | while (!netif_carrier_ok(ndev)) { |
827 | if (time_after(jiffies, atmost)) { | |
828 | printk(KERN_NOTICE | |
829 | "%s: timeout waiting for carrier\n", | |
830 | np->name); | |
831 | break; | |
832 | } | |
1b614fb9 | 833 | msleep(1); |
1da177e4 LT |
834 | } |
835 | ||
836 | /* If carrier appears to come up instantly, we don't | |
837 | * trust it and pause so that we don't pump all our | |
838 | * queued console messages into the bitbucket. | |
839 | */ | |
840 | ||
841 | if (time_before(jiffies, atleast)) { | |
842 | printk(KERN_NOTICE "%s: carrier detect appears" | |
843 | " untrustworthy, waiting 4 seconds\n", | |
844 | np->name); | |
845 | msleep(4000); | |
846 | } | |
847 | } | |
848 | ||
1da177e4 LT |
849 | if (!np->local_ip) { |
850 | rcu_read_lock(); | |
e5ed6399 | 851 | in_dev = __in_dev_get_rcu(ndev); |
1da177e4 LT |
852 | |
853 | if (!in_dev || !in_dev->ifa_list) { | |
854 | rcu_read_unlock(); | |
855 | printk(KERN_ERR "%s: no IP address for %s, aborting\n", | |
856 | np->name, np->dev_name); | |
b41848b6 | 857 | err = -EDESTADDRREQ; |
dbaa1541 | 858 | goto put; |
1da177e4 LT |
859 | } |
860 | ||
e7557af5 | 861 | np->local_ip = in_dev->ifa_list->ifa_local; |
1da177e4 | 862 | rcu_read_unlock(); |
e7557af5 | 863 | printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip); |
1da177e4 LT |
864 | } |
865 | ||
dbaa1541 HX |
866 | np->dev = ndev; |
867 | ||
868 | /* fill up the skb queue */ | |
869 | refill_skbs(); | |
870 | ||
871 | rtnl_lock(); | |
8fdd95ec | 872 | err = __netpoll_setup(np); |
dbaa1541 | 873 | rtnl_unlock(); |
53fb95d3 | 874 | |
8fdd95ec HX |
875 | if (err) |
876 | goto put; | |
877 | ||
1da177e4 LT |
878 | return 0; |
879 | ||
21edbb22 | 880 | put: |
1da177e4 | 881 | dev_put(ndev); |
b41848b6 | 882 | return err; |
1da177e4 | 883 | } |
9e34a5b5 | 884 | EXPORT_SYMBOL(netpoll_setup); |
1da177e4 | 885 | |
c68b9070 DM |
886 | static int __init netpoll_init(void) |
887 | { | |
a1bcfacd SH |
888 | skb_queue_head_init(&skb_pool); |
889 | return 0; | |
890 | } | |
891 | core_initcall(netpoll_init); | |
892 | ||
8fdd95ec | 893 | void __netpoll_cleanup(struct netpoll *np) |
1da177e4 | 894 | { |
fbeec2e1 JM |
895 | struct netpoll_info *npinfo; |
896 | unsigned long flags; | |
897 | ||
8fdd95ec HX |
898 | npinfo = np->dev->npinfo; |
899 | if (!npinfo) | |
dbaa1541 | 900 | return; |
93ec2c72 | 901 | |
8fdd95ec HX |
902 | if (!list_empty(&npinfo->rx_np)) { |
903 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
904 | list_del(&np->rx); | |
905 | if (list_empty(&npinfo->rx_np)) | |
906 | npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; | |
907 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
908 | } | |
de85d99e | 909 | |
8fdd95ec HX |
910 | if (atomic_dec_and_test(&npinfo->refcnt)) { |
911 | const struct net_device_ops *ops; | |
de85d99e | 912 | |
8fdd95ec HX |
913 | ops = np->dev->netdev_ops; |
914 | if (ops->ndo_netpoll_cleanup) | |
915 | ops->ndo_netpoll_cleanup(np->dev); | |
de85d99e | 916 | |
8fdd95ec | 917 | rcu_assign_pointer(np->dev->npinfo, NULL); |
de85d99e | 918 | |
dbaa1541 HX |
919 | /* avoid racing with NAPI reading npinfo */ |
920 | synchronize_rcu_bh(); | |
93ec2c72 | 921 | |
dbaa1541 HX |
922 | skb_queue_purge(&npinfo->arp_tx); |
923 | skb_queue_purge(&npinfo->txq); | |
afe2c511 | 924 | cancel_delayed_work_sync(&npinfo->tx_work); |
93ec2c72 | 925 | |
dbaa1541 HX |
926 | /* clean after last, unfinished work */ |
927 | __skb_queue_purge(&npinfo->txq); | |
928 | kfree(npinfo); | |
115c1d6e | 929 | } |
8fdd95ec HX |
930 | } |
931 | EXPORT_SYMBOL_GPL(__netpoll_cleanup); | |
fbeec2e1 | 932 | |
8fdd95ec HX |
933 | void netpoll_cleanup(struct netpoll *np) |
934 | { | |
935 | if (!np->dev) | |
936 | return; | |
dbaa1541 | 937 | |
8fdd95ec HX |
938 | rtnl_lock(); |
939 | __netpoll_cleanup(np); | |
940 | rtnl_unlock(); | |
941 | ||
942 | dev_put(np->dev); | |
1da177e4 LT |
943 | np->dev = NULL; |
944 | } | |
9e34a5b5 | 945 | EXPORT_SYMBOL(netpoll_cleanup); |
1da177e4 LT |
946 | |
947 | int netpoll_trap(void) | |
948 | { | |
949 | return atomic_read(&trapped); | |
950 | } | |
9e34a5b5 | 951 | EXPORT_SYMBOL(netpoll_trap); |
1da177e4 LT |
952 | |
953 | void netpoll_set_trap(int trap) | |
954 | { | |
955 | if (trap) | |
956 | atomic_inc(&trapped); | |
957 | else | |
958 | atomic_dec(&trapped); | |
959 | } | |
1da177e4 | 960 | EXPORT_SYMBOL(netpoll_set_trap); |