]>
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> | |
25 | #include <net/tcp.h> | |
26 | #include <net/udp.h> | |
27 | #include <asm/unaligned.h> | |
9cbc1cb8 | 28 | #include <trace/events/napi.h> |
1da177e4 LT |
29 | |
30 | /* | |
31 | * We maintain a small pool of fully-sized skbs, to make sure the | |
32 | * message gets out even in extreme OOM situations. | |
33 | */ | |
34 | ||
35 | #define MAX_UDP_CHUNK 1460 | |
36 | #define MAX_SKBS 32 | |
37 | #define MAX_QUEUE_DEPTH (MAX_SKBS / 2) | |
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 | ||
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 DM |
77 | __netif_tx_lock(txq, smp_processor_id()); |
78 | if (netif_tx_queue_stopped(txq) || | |
c3f26a26 | 79 | netif_tx_queue_frozen(txq) || |
00829823 | 80 | ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { |
6c43ff18 | 81 | skb_queue_head(&npinfo->txq, skb); |
fd2ea0a7 | 82 | __netif_tx_unlock(txq); |
3640543d | 83 | local_irq_restore(flags); |
1da177e4 | 84 | |
25442caf | 85 | schedule_delayed_work(&npinfo->tx_work, HZ/10); |
6c43ff18 SH |
86 | return; |
87 | } | |
fd2ea0a7 | 88 | __netif_tx_unlock(txq); |
3640543d | 89 | local_irq_restore(flags); |
1da177e4 | 90 | } |
1da177e4 LT |
91 | } |
92 | ||
b51655b9 AV |
93 | static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, |
94 | unsigned short ulen, __be32 saddr, __be32 daddr) | |
1da177e4 | 95 | { |
d6f5493c | 96 | __wsum psum; |
fb286bb2 | 97 | |
60476372 | 98 | if (uh->check == 0 || skb_csum_unnecessary(skb)) |
1da177e4 LT |
99 | return 0; |
100 | ||
fb286bb2 HX |
101 | psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); |
102 | ||
84fa7933 | 103 | if (skb->ip_summed == CHECKSUM_COMPLETE && |
d3bc23e7 | 104 | !csum_fold(csum_add(psum, skb->csum))) |
fb286bb2 | 105 | return 0; |
1da177e4 | 106 | |
fb286bb2 | 107 | skb->csum = psum; |
1da177e4 | 108 | |
fb286bb2 | 109 | return __skb_checksum_complete(skb); |
1da177e4 LT |
110 | } |
111 | ||
112 | /* | |
113 | * Check whether delayed processing was scheduled for our NIC. If so, | |
114 | * we attempt to grab the poll lock and use ->poll() to pump the card. | |
115 | * If this fails, either we've recursed in ->poll() or it's already | |
116 | * running on another CPU. | |
117 | * | |
118 | * Note: we don't mask interrupts with this lock because we're using | |
119 | * trylock here and interrupts are already disabled in the softirq | |
120 | * case. Further, we test the poll_owner to avoid recursion on UP | |
121 | * systems where the lock doesn't exist. | |
122 | * | |
123 | * In cases where there is bi-directional communications, reading only | |
124 | * one message at a time can lead to packets being dropped by the | |
125 | * network adapter, forcing superfluous retries and possibly timeouts. | |
126 | * Thus, we set our budget to greater than 1. | |
127 | */ | |
0a7606c1 DM |
128 | static int poll_one_napi(struct netpoll_info *npinfo, |
129 | struct napi_struct *napi, int budget) | |
130 | { | |
131 | int work; | |
132 | ||
133 | /* net_rx_action's ->poll() invocations and our's are | |
134 | * synchronized by this test which is only made while | |
135 | * holding the napi->poll_lock. | |
136 | */ | |
137 | if (!test_bit(NAPI_STATE_SCHED, &napi->state)) | |
138 | return budget; | |
139 | ||
d9452e9f | 140 | npinfo->rx_flags |= NETPOLL_RX_DROP; |
0a7606c1 | 141 | atomic_inc(&trapped); |
7b363e44 | 142 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 DM |
143 | |
144 | work = napi->poll(napi, budget); | |
7d18f114 | 145 | trace_napi_poll(napi); |
0a7606c1 | 146 | |
7b363e44 | 147 | clear_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 | 148 | atomic_dec(&trapped); |
d9452e9f | 149 | npinfo->rx_flags &= ~NETPOLL_RX_DROP; |
0a7606c1 DM |
150 | |
151 | return budget - work; | |
152 | } | |
153 | ||
5106930b | 154 | static void poll_napi(struct net_device *dev) |
1da177e4 | 155 | { |
bea3348e | 156 | struct napi_struct *napi; |
1da177e4 LT |
157 | int budget = 16; |
158 | ||
5106930b | 159 | list_for_each_entry(napi, &dev->napi_list, dev_list) { |
0a7606c1 | 160 | if (napi->poll_owner != smp_processor_id() && |
bea3348e | 161 | spin_trylock(&napi->poll_lock)) { |
5106930b | 162 | budget = poll_one_napi(dev->npinfo, napi, budget); |
bea3348e | 163 | spin_unlock(&napi->poll_lock); |
0a7606c1 DM |
164 | |
165 | if (!budget) | |
166 | break; | |
bea3348e | 167 | } |
1da177e4 LT |
168 | } |
169 | } | |
170 | ||
068c6e98 NH |
171 | static void service_arp_queue(struct netpoll_info *npi) |
172 | { | |
5106930b SH |
173 | if (npi) { |
174 | struct sk_buff *skb; | |
068c6e98 | 175 | |
5106930b SH |
176 | while ((skb = skb_dequeue(&npi->arp_tx))) |
177 | arp_reply(skb); | |
068c6e98 | 178 | } |
068c6e98 NH |
179 | } |
180 | ||
1da177e4 LT |
181 | void netpoll_poll(struct netpoll *np) |
182 | { | |
5106930b | 183 | struct net_device *dev = np->dev; |
5e392739 | 184 | const struct net_device_ops *ops; |
5106930b | 185 | |
5e392739 PE |
186 | if (!dev || !netif_running(dev)) |
187 | return; | |
188 | ||
189 | ops = dev->netdev_ops; | |
190 | if (!ops->ndo_poll_controller) | |
1da177e4 LT |
191 | return; |
192 | ||
193 | /* Process pending work on NIC */ | |
d314774c | 194 | ops->ndo_poll_controller(dev); |
5106930b SH |
195 | |
196 | poll_napi(dev); | |
1da177e4 | 197 | |
5106930b | 198 | service_arp_queue(dev->npinfo); |
068c6e98 | 199 | |
1da177e4 LT |
200 | zap_completion_queue(); |
201 | } | |
202 | ||
203 | static void refill_skbs(void) | |
204 | { | |
205 | struct sk_buff *skb; | |
206 | unsigned long flags; | |
207 | ||
a1bcfacd SH |
208 | spin_lock_irqsave(&skb_pool.lock, flags); |
209 | while (skb_pool.qlen < MAX_SKBS) { | |
1da177e4 LT |
210 | skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); |
211 | if (!skb) | |
212 | break; | |
213 | ||
a1bcfacd | 214 | __skb_queue_tail(&skb_pool, skb); |
1da177e4 | 215 | } |
a1bcfacd | 216 | spin_unlock_irqrestore(&skb_pool.lock, flags); |
1da177e4 LT |
217 | } |
218 | ||
219 | static void zap_completion_queue(void) | |
220 | { | |
221 | unsigned long flags; | |
222 | struct softnet_data *sd = &get_cpu_var(softnet_data); | |
223 | ||
224 | if (sd->completion_queue) { | |
225 | struct sk_buff *clist; | |
226 | ||
227 | local_irq_save(flags); | |
228 | clist = sd->completion_queue; | |
229 | sd->completion_queue = NULL; | |
230 | local_irq_restore(flags); | |
231 | ||
232 | while (clist != NULL) { | |
233 | struct sk_buff *skb = clist; | |
234 | clist = clist->next; | |
8a455b08 JP |
235 | if (skb->destructor) { |
236 | atomic_inc(&skb->users); | |
1da177e4 | 237 | dev_kfree_skb_any(skb); /* put this one back */ |
8a455b08 | 238 | } else { |
1da177e4 | 239 | __kfree_skb(skb); |
8a455b08 | 240 | } |
1da177e4 LT |
241 | } |
242 | } | |
243 | ||
244 | put_cpu_var(softnet_data); | |
245 | } | |
246 | ||
a1bcfacd | 247 | static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) |
1da177e4 | 248 | { |
a1bcfacd SH |
249 | int count = 0; |
250 | struct sk_buff *skb; | |
1da177e4 LT |
251 | |
252 | zap_completion_queue(); | |
a1bcfacd | 253 | refill_skbs(); |
1da177e4 | 254 | repeat: |
1da177e4 LT |
255 | |
256 | skb = alloc_skb(len, GFP_ATOMIC); | |
a1bcfacd SH |
257 | if (!skb) |
258 | skb = skb_dequeue(&skb_pool); | |
1da177e4 LT |
259 | |
260 | if (!skb) { | |
a1bcfacd SH |
261 | if (++count < 10) { |
262 | netpoll_poll(np); | |
263 | goto repeat; | |
1da177e4 | 264 | } |
a1bcfacd | 265 | return NULL; |
1da177e4 LT |
266 | } |
267 | ||
268 | atomic_set(&skb->users, 1); | |
269 | skb_reserve(skb, reserve); | |
270 | return skb; | |
271 | } | |
272 | ||
bea3348e SH |
273 | static int netpoll_owner_active(struct net_device *dev) |
274 | { | |
275 | struct napi_struct *napi; | |
276 | ||
277 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
278 | if (napi->poll_owner == smp_processor_id()) | |
279 | return 1; | |
280 | } | |
281 | return 0; | |
282 | } | |
283 | ||
1da177e4 LT |
284 | static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) |
285 | { | |
2bdfe0ba SH |
286 | int status = NETDEV_TX_BUSY; |
287 | unsigned long tries; | |
4ec93edb | 288 | struct net_device *dev = np->dev; |
00829823 | 289 | const struct net_device_ops *ops = dev->netdev_ops; |
4ec93edb | 290 | struct netpoll_info *npinfo = np->dev->npinfo; |
2bdfe0ba | 291 | |
4ec93edb YH |
292 | if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { |
293 | __kfree_skb(skb); | |
294 | return; | |
295 | } | |
2bdfe0ba SH |
296 | |
297 | /* don't get messages out of order, and no recursion */ | |
bea3348e | 298 | if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { |
fd2ea0a7 | 299 | struct netdev_queue *txq; |
a49f99ff AM |
300 | unsigned long flags; |
301 | ||
fd2ea0a7 DM |
302 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
303 | ||
a49f99ff | 304 | local_irq_save(flags); |
0db3dc73 SH |
305 | /* try until next clock tick */ |
306 | for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; | |
307 | tries > 0; --tries) { | |
fd2ea0a7 | 308 | if (__netif_tx_trylock(txq)) { |
08baf561 | 309 | if (!netif_tx_queue_stopped(txq)) { |
00829823 | 310 | status = ops->ndo_start_xmit(skb, dev); |
08baf561 ED |
311 | if (status == NETDEV_TX_OK) |
312 | txq_trans_update(txq); | |
313 | } | |
fd2ea0a7 | 314 | __netif_tx_unlock(txq); |
e37b8d93 AM |
315 | |
316 | if (status == NETDEV_TX_OK) | |
317 | break; | |
318 | ||
e37b8d93 | 319 | } |
0db3dc73 SH |
320 | |
321 | /* tickle device maybe there is some cleanup */ | |
322 | netpoll_poll(np); | |
323 | ||
324 | udelay(USEC_PER_POLL); | |
0db1d6fc | 325 | } |
79b1bee8 DD |
326 | |
327 | WARN_ONCE(!irqs_disabled(), | |
328 | "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", | |
329 | dev->name, ops->ndo_start_xmit); | |
330 | ||
a49f99ff | 331 | local_irq_restore(flags); |
1da177e4 | 332 | } |
1da177e4 | 333 | |
2bdfe0ba | 334 | if (status != NETDEV_TX_OK) { |
5de4a473 | 335 | skb_queue_tail(&npinfo->txq, skb); |
4c1ac1b4 | 336 | schedule_delayed_work(&npinfo->tx_work,0); |
1da177e4 | 337 | } |
1da177e4 LT |
338 | } |
339 | ||
340 | void netpoll_send_udp(struct netpoll *np, const char *msg, int len) | |
341 | { | |
342 | int total_len, eth_len, ip_len, udp_len; | |
343 | struct sk_buff *skb; | |
344 | struct udphdr *udph; | |
345 | struct iphdr *iph; | |
346 | struct ethhdr *eth; | |
347 | ||
348 | udp_len = len + sizeof(*udph); | |
349 | ip_len = eth_len = udp_len + sizeof(*iph); | |
350 | total_len = eth_len + ETH_HLEN + NET_IP_ALIGN; | |
351 | ||
352 | skb = find_skb(np, total_len, total_len - len); | |
353 | if (!skb) | |
354 | return; | |
355 | ||
27d7ff46 | 356 | skb_copy_to_linear_data(skb, msg, len); |
1da177e4 LT |
357 | skb->len += len; |
358 | ||
4bedb452 ACM |
359 | skb_push(skb, sizeof(*udph)); |
360 | skb_reset_transport_header(skb); | |
361 | udph = udp_hdr(skb); | |
1da177e4 LT |
362 | udph->source = htons(np->local_port); |
363 | udph->dest = htons(np->remote_port); | |
364 | udph->len = htons(udp_len); | |
365 | udph->check = 0; | |
e7557af5 HH |
366 | udph->check = csum_tcpudp_magic(np->local_ip, |
367 | np->remote_ip, | |
8e365eec | 368 | udp_len, IPPROTO_UDP, |
07f0757a | 369 | csum_partial(udph, udp_len, 0)); |
8e365eec | 370 | if (udph->check == 0) |
5e57dff2 | 371 | udph->check = CSUM_MANGLED_0; |
1da177e4 | 372 | |
e2d1bca7 ACM |
373 | skb_push(skb, sizeof(*iph)); |
374 | skb_reset_network_header(skb); | |
eddc9ec5 | 375 | iph = ip_hdr(skb); |
1da177e4 LT |
376 | |
377 | /* iph->version = 4; iph->ihl = 5; */ | |
378 | put_unaligned(0x45, (unsigned char *)iph); | |
379 | iph->tos = 0; | |
380 | put_unaligned(htons(ip_len), &(iph->tot_len)); | |
381 | iph->id = 0; | |
382 | iph->frag_off = 0; | |
383 | iph->ttl = 64; | |
384 | iph->protocol = IPPROTO_UDP; | |
385 | iph->check = 0; | |
e7557af5 HH |
386 | put_unaligned(np->local_ip, &(iph->saddr)); |
387 | put_unaligned(np->remote_ip, &(iph->daddr)); | |
1da177e4 LT |
388 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); |
389 | ||
390 | eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); | |
459a98ed | 391 | skb_reset_mac_header(skb); |
206daaf7 | 392 | skb->protocol = eth->h_proto = htons(ETH_P_IP); |
09538641 SH |
393 | memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); |
394 | memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); | |
1da177e4 LT |
395 | |
396 | skb->dev = np->dev; | |
397 | ||
398 | netpoll_send_skb(np, skb); | |
399 | } | |
400 | ||
401 | static void arp_reply(struct sk_buff *skb) | |
402 | { | |
115c1d6e | 403 | struct netpoll_info *npinfo = skb->dev->npinfo; |
1da177e4 LT |
404 | struct arphdr *arp; |
405 | unsigned char *arp_ptr; | |
406 | int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; | |
252e3346 | 407 | __be32 sip, tip; |
47bbec02 | 408 | unsigned char *sha; |
1da177e4 | 409 | struct sk_buff *send_skb; |
508e14b4 DB |
410 | struct netpoll *np, *tmp; |
411 | unsigned long flags; | |
412 | int hits = 0; | |
413 | ||
414 | if (list_empty(&npinfo->rx_np)) | |
415 | return; | |
416 | ||
417 | /* Before checking the packet, we do some early | |
418 | inspection whether this is interesting at all */ | |
419 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
420 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
421 | if (np->dev == skb->dev) | |
422 | hits++; | |
423 | } | |
424 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 | 425 | |
508e14b4 DB |
426 | /* No netpoll struct is using this dev */ |
427 | if (!hits) | |
115c1d6e | 428 | return; |
1da177e4 LT |
429 | |
430 | /* No arp on this interface */ | |
431 | if (skb->dev->flags & IFF_NOARP) | |
432 | return; | |
433 | ||
988b7050 | 434 | if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) |
1da177e4 LT |
435 | return; |
436 | ||
c1d2bbe1 | 437 | skb_reset_network_header(skb); |
badff6d0 | 438 | skb_reset_transport_header(skb); |
d0a92be0 | 439 | arp = arp_hdr(skb); |
1da177e4 LT |
440 | |
441 | if ((arp->ar_hrd != htons(ARPHRD_ETHER) && | |
442 | arp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
443 | arp->ar_pro != htons(ETH_P_IP) || | |
444 | arp->ar_op != htons(ARPOP_REQUEST)) | |
445 | return; | |
446 | ||
47bbec02 NH |
447 | arp_ptr = (unsigned char *)(arp+1); |
448 | /* save the location of the src hw addr */ | |
449 | sha = arp_ptr; | |
450 | arp_ptr += skb->dev->addr_len; | |
1da177e4 | 451 | memcpy(&sip, arp_ptr, 4); |
47bbec02 | 452 | arp_ptr += 4; |
508e14b4 DB |
453 | /* If we actually cared about dst hw addr, |
454 | it would get copied here */ | |
47bbec02 | 455 | arp_ptr += skb->dev->addr_len; |
1da177e4 LT |
456 | memcpy(&tip, arp_ptr, 4); |
457 | ||
458 | /* Should we ignore arp? */ | |
508e14b4 | 459 | if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) |
1da177e4 LT |
460 | return; |
461 | ||
988b7050 | 462 | size = arp_hdr_len(skb->dev); |
1da177e4 | 463 | |
508e14b4 DB |
464 | spin_lock_irqsave(&npinfo->rx_lock, flags); |
465 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
466 | if (tip != np->local_ip) | |
467 | continue; | |
1da177e4 | 468 | |
508e14b4 DB |
469 | send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev), |
470 | LL_RESERVED_SPACE(np->dev)); | |
471 | if (!send_skb) | |
472 | continue; | |
1da177e4 | 473 | |
508e14b4 DB |
474 | skb_reset_network_header(send_skb); |
475 | arp = (struct arphdr *) skb_put(send_skb, size); | |
476 | send_skb->dev = skb->dev; | |
477 | send_skb->protocol = htons(ETH_P_ARP); | |
1da177e4 | 478 | |
508e14b4 DB |
479 | /* Fill the device header for the ARP frame */ |
480 | if (dev_hard_header(send_skb, skb->dev, ptype, | |
481 | sha, np->dev->dev_addr, | |
482 | send_skb->len) < 0) { | |
483 | kfree_skb(send_skb); | |
484 | continue; | |
485 | } | |
1da177e4 | 486 | |
508e14b4 DB |
487 | /* |
488 | * Fill out the arp protocol part. | |
489 | * | |
490 | * we only support ethernet device type, | |
491 | * which (according to RFC 1390) should | |
492 | * always equal 1 (Ethernet). | |
493 | */ | |
1da177e4 | 494 | |
508e14b4 DB |
495 | arp->ar_hrd = htons(np->dev->type); |
496 | arp->ar_pro = htons(ETH_P_IP); | |
497 | arp->ar_hln = np->dev->addr_len; | |
498 | arp->ar_pln = 4; | |
499 | arp->ar_op = htons(type); | |
500 | ||
501 | arp_ptr = (unsigned char *)(arp + 1); | |
502 | memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); | |
503 | arp_ptr += np->dev->addr_len; | |
504 | memcpy(arp_ptr, &tip, 4); | |
505 | arp_ptr += 4; | |
506 | memcpy(arp_ptr, sha, np->dev->addr_len); | |
507 | arp_ptr += np->dev->addr_len; | |
508 | memcpy(arp_ptr, &sip, 4); | |
509 | ||
510 | netpoll_send_skb(np, send_skb); | |
511 | ||
512 | /* If there are several rx_hooks for the same address, | |
513 | we're fine by sending a single reply */ | |
514 | break; | |
515 | } | |
516 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 LT |
517 | } |
518 | ||
519 | int __netpoll_rx(struct sk_buff *skb) | |
520 | { | |
521 | int proto, len, ulen; | |
508e14b4 | 522 | int hits = 0; |
1da177e4 LT |
523 | struct iphdr *iph; |
524 | struct udphdr *uh; | |
508e14b4 DB |
525 | struct netpoll_info *npinfo = skb->dev->npinfo; |
526 | struct netpoll *np, *tmp; | |
068c6e98 | 527 | |
508e14b4 | 528 | if (list_empty(&npinfo->rx_np)) |
1da177e4 | 529 | goto out; |
508e14b4 | 530 | |
1da177e4 LT |
531 | if (skb->dev->type != ARPHRD_ETHER) |
532 | goto out; | |
533 | ||
d9452e9f | 534 | /* check if netpoll clients need ARP */ |
724800d6 | 535 | if (skb->protocol == htons(ETH_P_ARP) && |
1da177e4 | 536 | atomic_read(&trapped)) { |
508e14b4 | 537 | skb_queue_tail(&npinfo->arp_tx, skb); |
1da177e4 LT |
538 | return 1; |
539 | } | |
540 | ||
541 | proto = ntohs(eth_hdr(skb)->h_proto); | |
542 | if (proto != ETH_P_IP) | |
543 | goto out; | |
544 | if (skb->pkt_type == PACKET_OTHERHOST) | |
545 | goto out; | |
546 | if (skb_shared(skb)) | |
547 | goto out; | |
548 | ||
549 | iph = (struct iphdr *)skb->data; | |
550 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) | |
551 | goto out; | |
552 | if (iph->ihl < 5 || iph->version != 4) | |
553 | goto out; | |
554 | if (!pskb_may_pull(skb, iph->ihl*4)) | |
555 | goto out; | |
556 | if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) | |
557 | goto out; | |
558 | ||
559 | len = ntohs(iph->tot_len); | |
560 | if (skb->len < len || len < iph->ihl*4) | |
561 | goto out; | |
562 | ||
5e7d7fa5 AL |
563 | /* |
564 | * Our transport medium may have padded the buffer out. | |
565 | * Now We trim to the true length of the frame. | |
566 | */ | |
567 | if (pskb_trim_rcsum(skb, len)) | |
568 | goto out; | |
569 | ||
1da177e4 LT |
570 | if (iph->protocol != IPPROTO_UDP) |
571 | goto out; | |
572 | ||
573 | len -= iph->ihl*4; | |
574 | uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); | |
575 | ulen = ntohs(uh->len); | |
576 | ||
577 | if (ulen != len) | |
578 | goto out; | |
fb286bb2 | 579 | if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) |
1da177e4 | 580 | goto out; |
1da177e4 | 581 | |
508e14b4 DB |
582 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { |
583 | if (np->local_ip && np->local_ip != iph->daddr) | |
584 | continue; | |
585 | if (np->remote_ip && np->remote_ip != iph->saddr) | |
586 | continue; | |
587 | if (np->local_port && np->local_port != ntohs(uh->dest)) | |
588 | continue; | |
589 | ||
590 | np->rx_hook(np, ntohs(uh->source), | |
591 | (char *)(uh+1), | |
592 | ulen - sizeof(struct udphdr)); | |
593 | hits++; | |
594 | } | |
595 | ||
596 | if (!hits) | |
597 | goto out; | |
1da177e4 LT |
598 | |
599 | kfree_skb(skb); | |
600 | return 1; | |
601 | ||
602 | out: | |
603 | if (atomic_read(&trapped)) { | |
604 | kfree_skb(skb); | |
605 | return 1; | |
606 | } | |
607 | ||
608 | return 0; | |
609 | } | |
610 | ||
0bcc1816 SS |
611 | void netpoll_print_options(struct netpoll *np) |
612 | { | |
613 | printk(KERN_INFO "%s: local port %d\n", | |
614 | np->name, np->local_port); | |
e7557af5 HH |
615 | printk(KERN_INFO "%s: local IP %pI4\n", |
616 | np->name, &np->local_ip); | |
0bcc1816 SS |
617 | printk(KERN_INFO "%s: interface %s\n", |
618 | np->name, np->dev_name); | |
619 | printk(KERN_INFO "%s: remote port %d\n", | |
620 | np->name, np->remote_port); | |
e7557af5 HH |
621 | printk(KERN_INFO "%s: remote IP %pI4\n", |
622 | np->name, &np->remote_ip); | |
e174961c JB |
623 | printk(KERN_INFO "%s: remote ethernet address %pM\n", |
624 | np->name, np->remote_mac); | |
0bcc1816 SS |
625 | } |
626 | ||
1da177e4 LT |
627 | int netpoll_parse_options(struct netpoll *np, char *opt) |
628 | { | |
629 | char *cur=opt, *delim; | |
630 | ||
c68b9070 | 631 | if (*cur != '@') { |
1da177e4 LT |
632 | if ((delim = strchr(cur, '@')) == NULL) |
633 | goto parse_failed; | |
c68b9070 DM |
634 | *delim = 0; |
635 | np->local_port = simple_strtol(cur, NULL, 10); | |
636 | cur = delim; | |
1da177e4 LT |
637 | } |
638 | cur++; | |
1da177e4 | 639 | |
c68b9070 | 640 | if (*cur != '/') { |
1da177e4 LT |
641 | if ((delim = strchr(cur, '/')) == NULL) |
642 | goto parse_failed; | |
c68b9070 | 643 | *delim = 0; |
e7557af5 | 644 | np->local_ip = in_aton(cur); |
c68b9070 | 645 | cur = delim; |
1da177e4 LT |
646 | } |
647 | cur++; | |
648 | ||
c68b9070 | 649 | if (*cur != ',') { |
1da177e4 LT |
650 | /* parse out dev name */ |
651 | if ((delim = strchr(cur, ',')) == NULL) | |
652 | goto parse_failed; | |
c68b9070 | 653 | *delim = 0; |
1da177e4 | 654 | strlcpy(np->dev_name, cur, sizeof(np->dev_name)); |
c68b9070 | 655 | cur = delim; |
1da177e4 LT |
656 | } |
657 | cur++; | |
658 | ||
c68b9070 | 659 | if (*cur != '@') { |
1da177e4 LT |
660 | /* dst port */ |
661 | if ((delim = strchr(cur, '@')) == NULL) | |
662 | goto parse_failed; | |
c68b9070 DM |
663 | *delim = 0; |
664 | np->remote_port = simple_strtol(cur, NULL, 10); | |
665 | cur = delim; | |
1da177e4 LT |
666 | } |
667 | cur++; | |
1da177e4 LT |
668 | |
669 | /* dst ip */ | |
670 | if ((delim = strchr(cur, '/')) == NULL) | |
671 | goto parse_failed; | |
c68b9070 | 672 | *delim = 0; |
e7557af5 | 673 | np->remote_ip = in_aton(cur); |
c68b9070 | 674 | cur = delim + 1; |
1da177e4 | 675 | |
c68b9070 | 676 | if (*cur != 0) { |
1da177e4 LT |
677 | /* MAC address */ |
678 | if ((delim = strchr(cur, ':')) == NULL) | |
679 | goto parse_failed; | |
c68b9070 DM |
680 | *delim = 0; |
681 | np->remote_mac[0] = simple_strtol(cur, NULL, 16); | |
682 | cur = delim + 1; | |
1da177e4 LT |
683 | if ((delim = strchr(cur, ':')) == NULL) |
684 | goto parse_failed; | |
c68b9070 DM |
685 | *delim = 0; |
686 | np->remote_mac[1] = simple_strtol(cur, NULL, 16); | |
687 | cur = delim + 1; | |
1da177e4 LT |
688 | if ((delim = strchr(cur, ':')) == NULL) |
689 | goto parse_failed; | |
c68b9070 DM |
690 | *delim = 0; |
691 | np->remote_mac[2] = simple_strtol(cur, NULL, 16); | |
692 | cur = delim + 1; | |
1da177e4 LT |
693 | if ((delim = strchr(cur, ':')) == NULL) |
694 | goto parse_failed; | |
c68b9070 DM |
695 | *delim = 0; |
696 | np->remote_mac[3] = simple_strtol(cur, NULL, 16); | |
697 | cur = delim + 1; | |
1da177e4 LT |
698 | if ((delim = strchr(cur, ':')) == NULL) |
699 | goto parse_failed; | |
c68b9070 DM |
700 | *delim = 0; |
701 | np->remote_mac[4] = simple_strtol(cur, NULL, 16); | |
702 | cur = delim + 1; | |
703 | np->remote_mac[5] = simple_strtol(cur, NULL, 16); | |
1da177e4 LT |
704 | } |
705 | ||
0bcc1816 | 706 | netpoll_print_options(np); |
1da177e4 LT |
707 | |
708 | return 0; | |
709 | ||
710 | parse_failed: | |
711 | printk(KERN_INFO "%s: couldn't parse config at %s!\n", | |
712 | np->name, cur); | |
713 | return -1; | |
714 | } | |
715 | ||
716 | int netpoll_setup(struct netpoll *np) | |
717 | { | |
718 | struct net_device *ndev = NULL; | |
719 | struct in_device *in_dev; | |
115c1d6e | 720 | struct netpoll_info *npinfo; |
508e14b4 | 721 | struct netpoll *npe, *tmp; |
fbeec2e1 | 722 | unsigned long flags; |
b41848b6 | 723 | int err; |
1da177e4 LT |
724 | |
725 | if (np->dev_name) | |
881d966b | 726 | ndev = dev_get_by_name(&init_net, np->dev_name); |
1da177e4 LT |
727 | if (!ndev) { |
728 | printk(KERN_ERR "%s: %s doesn't exist, aborting.\n", | |
729 | np->name, np->dev_name); | |
b41848b6 | 730 | return -ENODEV; |
1da177e4 LT |
731 | } |
732 | ||
733 | np->dev = ndev; | |
115c1d6e JM |
734 | if (!ndev->npinfo) { |
735 | npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL); | |
b41848b6 SH |
736 | if (!npinfo) { |
737 | err = -ENOMEM; | |
21edbb22 | 738 | goto put; |
b41848b6 | 739 | } |
115c1d6e | 740 | |
d9452e9f | 741 | npinfo->rx_flags = 0; |
508e14b4 | 742 | INIT_LIST_HEAD(&npinfo->rx_np); |
2bdfe0ba | 743 | |
a9f6a0dd | 744 | spin_lock_init(&npinfo->rx_lock); |
068c6e98 | 745 | skb_queue_head_init(&npinfo->arp_tx); |
b6cd27ed | 746 | skb_queue_head_init(&npinfo->txq); |
4c1ac1b4 | 747 | INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); |
b6cd27ed | 748 | |
93ec2c72 SH |
749 | atomic_set(&npinfo->refcnt, 1); |
750 | } else { | |
115c1d6e | 751 | npinfo = ndev->npinfo; |
93ec2c72 SH |
752 | atomic_inc(&npinfo->refcnt); |
753 | } | |
1da177e4 | 754 | |
d314774c | 755 | if (!ndev->netdev_ops->ndo_poll_controller) { |
1da177e4 LT |
756 | printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n", |
757 | np->name, np->dev_name); | |
b41848b6 | 758 | err = -ENOTSUPP; |
1da177e4 LT |
759 | goto release; |
760 | } | |
761 | ||
762 | if (!netif_running(ndev)) { | |
763 | unsigned long atmost, atleast; | |
764 | ||
765 | printk(KERN_INFO "%s: device %s not up yet, forcing it\n", | |
766 | np->name, np->dev_name); | |
767 | ||
6756ae4b | 768 | rtnl_lock(); |
b41848b6 SH |
769 | err = dev_open(ndev); |
770 | rtnl_unlock(); | |
771 | ||
772 | if (err) { | |
1da177e4 | 773 | printk(KERN_ERR "%s: failed to open %s\n", |
b41848b6 | 774 | np->name, ndev->name); |
1da177e4 LT |
775 | goto release; |
776 | } | |
1da177e4 LT |
777 | |
778 | atleast = jiffies + HZ/10; | |
bff38771 | 779 | atmost = jiffies + carrier_timeout * HZ; |
1da177e4 LT |
780 | while (!netif_carrier_ok(ndev)) { |
781 | if (time_after(jiffies, atmost)) { | |
782 | printk(KERN_NOTICE | |
783 | "%s: timeout waiting for carrier\n", | |
784 | np->name); | |
785 | break; | |
786 | } | |
1b614fb9 | 787 | msleep(1); |
1da177e4 LT |
788 | } |
789 | ||
790 | /* If carrier appears to come up instantly, we don't | |
791 | * trust it and pause so that we don't pump all our | |
792 | * queued console messages into the bitbucket. | |
793 | */ | |
794 | ||
795 | if (time_before(jiffies, atleast)) { | |
796 | printk(KERN_NOTICE "%s: carrier detect appears" | |
797 | " untrustworthy, waiting 4 seconds\n", | |
798 | np->name); | |
799 | msleep(4000); | |
800 | } | |
801 | } | |
802 | ||
1da177e4 LT |
803 | if (!np->local_ip) { |
804 | rcu_read_lock(); | |
e5ed6399 | 805 | in_dev = __in_dev_get_rcu(ndev); |
1da177e4 LT |
806 | |
807 | if (!in_dev || !in_dev->ifa_list) { | |
808 | rcu_read_unlock(); | |
809 | printk(KERN_ERR "%s: no IP address for %s, aborting\n", | |
810 | np->name, np->dev_name); | |
b41848b6 | 811 | err = -EDESTADDRREQ; |
1da177e4 LT |
812 | goto release; |
813 | } | |
814 | ||
e7557af5 | 815 | np->local_ip = in_dev->ifa_list->ifa_local; |
1da177e4 | 816 | rcu_read_unlock(); |
e7557af5 | 817 | printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip); |
1da177e4 LT |
818 | } |
819 | ||
fbeec2e1 JM |
820 | if (np->rx_hook) { |
821 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
d9452e9f | 822 | npinfo->rx_flags |= NETPOLL_RX_ENABLED; |
508e14b4 | 823 | list_add_tail(&np->rx, &npinfo->rx_np); |
fbeec2e1 JM |
824 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); |
825 | } | |
26520765 IM |
826 | |
827 | /* fill up the skb queue */ | |
828 | refill_skbs(); | |
829 | ||
fbeec2e1 | 830 | /* last thing to do is link it to the net device structure */ |
115c1d6e | 831 | ndev->npinfo = npinfo; |
1da177e4 | 832 | |
53fb95d3 MM |
833 | /* avoid racing with NAPI reading npinfo */ |
834 | synchronize_rcu(); | |
835 | ||
1da177e4 LT |
836 | return 0; |
837 | ||
838 | release: | |
508e14b4 DB |
839 | if (!ndev->npinfo) { |
840 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
841 | list_for_each_entry_safe(npe, tmp, &npinfo->rx_np, rx) { | |
842 | npe->dev = NULL; | |
843 | } | |
844 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
845 | ||
115c1d6e | 846 | kfree(npinfo); |
508e14b4 | 847 | } |
21edbb22 | 848 | put: |
1da177e4 | 849 | dev_put(ndev); |
b41848b6 | 850 | return err; |
1da177e4 LT |
851 | } |
852 | ||
c68b9070 DM |
853 | static int __init netpoll_init(void) |
854 | { | |
a1bcfacd SH |
855 | skb_queue_head_init(&skb_pool); |
856 | return 0; | |
857 | } | |
858 | core_initcall(netpoll_init); | |
859 | ||
1da177e4 LT |
860 | void netpoll_cleanup(struct netpoll *np) |
861 | { | |
fbeec2e1 JM |
862 | struct netpoll_info *npinfo; |
863 | unsigned long flags; | |
864 | ||
115c1d6e | 865 | if (np->dev) { |
fbeec2e1 | 866 | npinfo = np->dev->npinfo; |
93ec2c72 | 867 | if (npinfo) { |
508e14b4 | 868 | if (!list_empty(&npinfo->rx_np)) { |
93ec2c72 | 869 | spin_lock_irqsave(&npinfo->rx_lock, flags); |
508e14b4 DB |
870 | list_del(&np->rx); |
871 | if (list_empty(&npinfo->rx_np)) | |
872 | npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; | |
93ec2c72 SH |
873 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); |
874 | } | |
875 | ||
93ec2c72 SH |
876 | if (atomic_dec_and_test(&npinfo->refcnt)) { |
877 | skb_queue_purge(&npinfo->arp_tx); | |
4ec93edb | 878 | skb_queue_purge(&npinfo->txq); |
25442caf | 879 | cancel_rearming_delayed_work(&npinfo->tx_work); |
93ec2c72 | 880 | |
17200811 | 881 | /* clean after last, unfinished work */ |
0adc9add | 882 | __skb_queue_purge(&npinfo->txq); |
93ec2c72 | 883 | kfree(npinfo); |
1498b3f1 | 884 | np->dev->npinfo = NULL; |
93ec2c72 | 885 | } |
fbeec2e1 | 886 | } |
93ec2c72 | 887 | |
115c1d6e JM |
888 | dev_put(np->dev); |
889 | } | |
fbeec2e1 | 890 | |
1da177e4 LT |
891 | np->dev = NULL; |
892 | } | |
893 | ||
894 | int netpoll_trap(void) | |
895 | { | |
896 | return atomic_read(&trapped); | |
897 | } | |
898 | ||
899 | void netpoll_set_trap(int trap) | |
900 | { | |
901 | if (trap) | |
902 | atomic_inc(&trapped); | |
903 | else | |
904 | atomic_dec(&trapped); | |
905 | } | |
906 | ||
907 | EXPORT_SYMBOL(netpoll_set_trap); | |
908 | EXPORT_SYMBOL(netpoll_trap); | |
0bcc1816 | 909 | EXPORT_SYMBOL(netpoll_print_options); |
1da177e4 LT |
910 | EXPORT_SYMBOL(netpoll_parse_options); |
911 | EXPORT_SYMBOL(netpoll_setup); | |
912 | EXPORT_SYMBOL(netpoll_cleanup); | |
913 | EXPORT_SYMBOL(netpoll_send_udp); | |
914 | EXPORT_SYMBOL(netpoll_poll); |