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 | |
38 | #define MAX_QUEUE_DEPTH (MAX_SKBS / 2) | |
39 | ||
a1bcfacd | 40 | static struct sk_buff_head skb_pool; |
1da177e4 LT |
41 | |
42 | static atomic_t trapped; | |
43 | ||
2bdfe0ba | 44 | #define USEC_PER_POLL 50 |
d9452e9f DM |
45 | #define NETPOLL_RX_ENABLED 1 |
46 | #define NETPOLL_RX_DROP 2 | |
1da177e4 LT |
47 | |
48 | #define MAX_SKB_SIZE \ | |
49 | (MAX_UDP_CHUNK + sizeof(struct udphdr) + \ | |
50 | sizeof(struct iphdr) + sizeof(struct ethhdr)) | |
51 | ||
3578b0c8 | 52 | static void zap_completion_queue(void); |
068c6e98 | 53 | static void arp_reply(struct sk_buff *skb); |
1da177e4 | 54 | |
bff38771 AV |
55 | static unsigned int carrier_timeout = 4; |
56 | module_param(carrier_timeout, uint, 0644); | |
57 | ||
c4028958 | 58 | static void queue_process(struct work_struct *work) |
1da177e4 | 59 | { |
4c1ac1b4 DH |
60 | struct netpoll_info *npinfo = |
61 | container_of(work, struct netpoll_info, tx_work.work); | |
1da177e4 | 62 | struct sk_buff *skb; |
3640543d | 63 | unsigned long flags; |
1da177e4 | 64 | |
6c43ff18 SH |
65 | while ((skb = skb_dequeue(&npinfo->txq))) { |
66 | struct net_device *dev = skb->dev; | |
00829823 | 67 | const struct net_device_ops *ops = dev->netdev_ops; |
fd2ea0a7 | 68 | struct netdev_queue *txq; |
1da177e4 | 69 | |
6c43ff18 SH |
70 | if (!netif_device_present(dev) || !netif_running(dev)) { |
71 | __kfree_skb(skb); | |
72 | continue; | |
73 | } | |
1da177e4 | 74 | |
fd2ea0a7 DM |
75 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
76 | ||
3640543d | 77 | local_irq_save(flags); |
fd2ea0a7 DM |
78 | __netif_tx_lock(txq, smp_processor_id()); |
79 | if (netif_tx_queue_stopped(txq) || | |
c3f26a26 | 80 | netif_tx_queue_frozen(txq) || |
00829823 | 81 | ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { |
6c43ff18 | 82 | skb_queue_head(&npinfo->txq, skb); |
fd2ea0a7 | 83 | __netif_tx_unlock(txq); |
3640543d | 84 | local_irq_restore(flags); |
1da177e4 | 85 | |
25442caf | 86 | schedule_delayed_work(&npinfo->tx_work, HZ/10); |
6c43ff18 SH |
87 | return; |
88 | } | |
fd2ea0a7 | 89 | __netif_tx_unlock(txq); |
3640543d | 90 | local_irq_restore(flags); |
1da177e4 | 91 | } |
1da177e4 LT |
92 | } |
93 | ||
b51655b9 AV |
94 | static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, |
95 | unsigned short ulen, __be32 saddr, __be32 daddr) | |
1da177e4 | 96 | { |
d6f5493c | 97 | __wsum psum; |
fb286bb2 | 98 | |
60476372 | 99 | if (uh->check == 0 || skb_csum_unnecessary(skb)) |
1da177e4 LT |
100 | return 0; |
101 | ||
fb286bb2 HX |
102 | psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); |
103 | ||
84fa7933 | 104 | if (skb->ip_summed == CHECKSUM_COMPLETE && |
d3bc23e7 | 105 | !csum_fold(csum_add(psum, skb->csum))) |
fb286bb2 | 106 | return 0; |
1da177e4 | 107 | |
fb286bb2 | 108 | skb->csum = psum; |
1da177e4 | 109 | |
fb286bb2 | 110 | return __skb_checksum_complete(skb); |
1da177e4 LT |
111 | } |
112 | ||
113 | /* | |
114 | * Check whether delayed processing was scheduled for our NIC. If so, | |
115 | * we attempt to grab the poll lock and use ->poll() to pump the card. | |
116 | * If this fails, either we've recursed in ->poll() or it's already | |
117 | * running on another CPU. | |
118 | * | |
119 | * Note: we don't mask interrupts with this lock because we're using | |
120 | * trylock here and interrupts are already disabled in the softirq | |
121 | * case. Further, we test the poll_owner to avoid recursion on UP | |
122 | * systems where the lock doesn't exist. | |
123 | * | |
124 | * In cases where there is bi-directional communications, reading only | |
125 | * one message at a time can lead to packets being dropped by the | |
126 | * network adapter, forcing superfluous retries and possibly timeouts. | |
127 | * Thus, we set our budget to greater than 1. | |
128 | */ | |
0a7606c1 DM |
129 | static int poll_one_napi(struct netpoll_info *npinfo, |
130 | struct napi_struct *napi, int budget) | |
131 | { | |
132 | int work; | |
133 | ||
134 | /* net_rx_action's ->poll() invocations and our's are | |
135 | * synchronized by this test which is only made while | |
136 | * holding the napi->poll_lock. | |
137 | */ | |
138 | if (!test_bit(NAPI_STATE_SCHED, &napi->state)) | |
139 | return budget; | |
140 | ||
d9452e9f | 141 | npinfo->rx_flags |= NETPOLL_RX_DROP; |
0a7606c1 | 142 | atomic_inc(&trapped); |
7b363e44 | 143 | set_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 DM |
144 | |
145 | work = napi->poll(napi, budget); | |
7d18f114 | 146 | trace_napi_poll(napi); |
0a7606c1 | 147 | |
7b363e44 | 148 | clear_bit(NAPI_STATE_NPSVC, &napi->state); |
0a7606c1 | 149 | atomic_dec(&trapped); |
d9452e9f | 150 | npinfo->rx_flags &= ~NETPOLL_RX_DROP; |
0a7606c1 DM |
151 | |
152 | return budget - work; | |
153 | } | |
154 | ||
5106930b | 155 | static void poll_napi(struct net_device *dev) |
1da177e4 | 156 | { |
bea3348e | 157 | struct napi_struct *napi; |
1da177e4 LT |
158 | int budget = 16; |
159 | ||
f13d493d | 160 | list_for_each_entry(napi, &dev->napi_list, dev_list) { |
0a7606c1 | 161 | if (napi->poll_owner != smp_processor_id() && |
bea3348e | 162 | spin_trylock(&napi->poll_lock)) { |
5106930b | 163 | budget = poll_one_napi(dev->npinfo, napi, budget); |
bea3348e | 164 | spin_unlock(&napi->poll_lock); |
0a7606c1 DM |
165 | |
166 | if (!budget) | |
167 | break; | |
bea3348e | 168 | } |
1da177e4 LT |
169 | } |
170 | } | |
171 | ||
068c6e98 NH |
172 | static void service_arp_queue(struct netpoll_info *npi) |
173 | { | |
5106930b SH |
174 | if (npi) { |
175 | struct sk_buff *skb; | |
068c6e98 | 176 | |
5106930b SH |
177 | while ((skb = skb_dequeue(&npi->arp_tx))) |
178 | arp_reply(skb); | |
068c6e98 | 179 | } |
068c6e98 NH |
180 | } |
181 | ||
0e34e931 | 182 | void netpoll_poll_dev(struct net_device *dev) |
1da177e4 | 183 | { |
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 | |
3578b0c8 | 200 | zap_completion_queue(); |
1da177e4 | 201 | } |
9e34a5b5 | 202 | EXPORT_SYMBOL(netpoll_poll_dev); |
1da177e4 | 203 | |
0e34e931 WC |
204 | void netpoll_poll(struct netpoll *np) |
205 | { | |
206 | netpoll_poll_dev(np->dev); | |
207 | } | |
9e34a5b5 | 208 | EXPORT_SYMBOL(netpoll_poll); |
0e34e931 | 209 | |
1da177e4 LT |
210 | static void refill_skbs(void) |
211 | { | |
212 | struct sk_buff *skb; | |
213 | unsigned long flags; | |
214 | ||
a1bcfacd SH |
215 | spin_lock_irqsave(&skb_pool.lock, flags); |
216 | while (skb_pool.qlen < MAX_SKBS) { | |
1da177e4 LT |
217 | skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); |
218 | if (!skb) | |
219 | break; | |
220 | ||
a1bcfacd | 221 | __skb_queue_tail(&skb_pool, skb); |
1da177e4 | 222 | } |
a1bcfacd | 223 | spin_unlock_irqrestore(&skb_pool.lock, flags); |
1da177e4 LT |
224 | } |
225 | ||
3578b0c8 DM |
226 | static void zap_completion_queue(void) |
227 | { | |
228 | unsigned long flags; | |
229 | struct softnet_data *sd = &get_cpu_var(softnet_data); | |
230 | ||
231 | if (sd->completion_queue) { | |
232 | struct sk_buff *clist; | |
233 | ||
234 | local_irq_save(flags); | |
235 | clist = sd->completion_queue; | |
236 | sd->completion_queue = NULL; | |
237 | local_irq_restore(flags); | |
238 | ||
239 | while (clist != NULL) { | |
240 | struct sk_buff *skb = clist; | |
241 | clist = clist->next; | |
242 | if (skb->destructor) { | |
243 | atomic_inc(&skb->users); | |
244 | dev_kfree_skb_any(skb); /* put this one back */ | |
245 | } else { | |
246 | __kfree_skb(skb); | |
247 | } | |
248 | } | |
249 | } | |
250 | ||
251 | put_cpu_var(softnet_data); | |
252 | } | |
253 | ||
a1bcfacd | 254 | static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) |
1da177e4 | 255 | { |
a1bcfacd SH |
256 | int count = 0; |
257 | struct sk_buff *skb; | |
1da177e4 | 258 | |
3578b0c8 | 259 | zap_completion_queue(); |
a1bcfacd | 260 | refill_skbs(); |
1da177e4 | 261 | repeat: |
1da177e4 LT |
262 | |
263 | skb = alloc_skb(len, GFP_ATOMIC); | |
a1bcfacd SH |
264 | if (!skb) |
265 | skb = skb_dequeue(&skb_pool); | |
1da177e4 LT |
266 | |
267 | if (!skb) { | |
a1bcfacd SH |
268 | if (++count < 10) { |
269 | netpoll_poll(np); | |
270 | goto repeat; | |
1da177e4 | 271 | } |
a1bcfacd | 272 | return NULL; |
1da177e4 LT |
273 | } |
274 | ||
275 | atomic_set(&skb->users, 1); | |
276 | skb_reserve(skb, reserve); | |
277 | return skb; | |
278 | } | |
279 | ||
bea3348e SH |
280 | static int netpoll_owner_active(struct net_device *dev) |
281 | { | |
282 | struct napi_struct *napi; | |
283 | ||
284 | list_for_each_entry(napi, &dev->napi_list, dev_list) { | |
285 | if (napi->poll_owner == smp_processor_id()) | |
286 | return 1; | |
287 | } | |
288 | return 0; | |
289 | } | |
290 | ||
c2355e1a NH |
291 | void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, |
292 | struct net_device *dev) | |
1da177e4 | 293 | { |
2bdfe0ba SH |
294 | int status = NETDEV_TX_BUSY; |
295 | unsigned long tries; | |
00829823 | 296 | const struct net_device_ops *ops = dev->netdev_ops; |
de85d99e | 297 | /* It is up to the caller to keep npinfo alive. */ |
4ec93edb | 298 | struct netpoll_info *npinfo = np->dev->npinfo; |
2bdfe0ba | 299 | |
4ec93edb YH |
300 | if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { |
301 | __kfree_skb(skb); | |
302 | return; | |
303 | } | |
2bdfe0ba SH |
304 | |
305 | /* don't get messages out of order, and no recursion */ | |
bea3348e | 306 | if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { |
fd2ea0a7 | 307 | struct netdev_queue *txq; |
a49f99ff AM |
308 | unsigned long flags; |
309 | ||
fd2ea0a7 DM |
310 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); |
311 | ||
a49f99ff | 312 | local_irq_save(flags); |
0db3dc73 SH |
313 | /* try until next clock tick */ |
314 | for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; | |
315 | tries > 0; --tries) { | |
fd2ea0a7 | 316 | if (__netif_tx_trylock(txq)) { |
08baf561 | 317 | if (!netif_tx_queue_stopped(txq)) { |
0e34e931 | 318 | dev->priv_flags |= IFF_IN_NETPOLL; |
00829823 | 319 | status = ops->ndo_start_xmit(skb, dev); |
0e34e931 | 320 | dev->priv_flags &= ~IFF_IN_NETPOLL; |
08baf561 ED |
321 | if (status == NETDEV_TX_OK) |
322 | txq_trans_update(txq); | |
323 | } | |
fd2ea0a7 | 324 | __netif_tx_unlock(txq); |
e37b8d93 AM |
325 | |
326 | if (status == NETDEV_TX_OK) | |
327 | break; | |
328 | ||
e37b8d93 | 329 | } |
0db3dc73 SH |
330 | |
331 | /* tickle device maybe there is some cleanup */ | |
332 | netpoll_poll(np); | |
333 | ||
334 | udelay(USEC_PER_POLL); | |
0db1d6fc | 335 | } |
79b1bee8 DD |
336 | |
337 | WARN_ONCE(!irqs_disabled(), | |
338 | "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", | |
339 | dev->name, ops->ndo_start_xmit); | |
340 | ||
a49f99ff | 341 | local_irq_restore(flags); |
1da177e4 | 342 | } |
1da177e4 | 343 | |
2bdfe0ba | 344 | if (status != NETDEV_TX_OK) { |
5de4a473 | 345 | skb_queue_tail(&npinfo->txq, skb); |
4c1ac1b4 | 346 | schedule_delayed_work(&npinfo->tx_work,0); |
1da177e4 | 347 | } |
1da177e4 | 348 | } |
c2355e1a | 349 | EXPORT_SYMBOL(netpoll_send_skb_on_dev); |
1da177e4 LT |
350 | |
351 | void netpoll_send_udp(struct netpoll *np, const char *msg, int len) | |
352 | { | |
353 | int total_len, eth_len, ip_len, udp_len; | |
354 | struct sk_buff *skb; | |
355 | struct udphdr *udph; | |
356 | struct iphdr *iph; | |
357 | struct ethhdr *eth; | |
358 | ||
359 | udp_len = len + sizeof(*udph); | |
360 | ip_len = eth_len = udp_len + sizeof(*iph); | |
361 | total_len = eth_len + ETH_HLEN + NET_IP_ALIGN; | |
362 | ||
363 | skb = find_skb(np, total_len, total_len - len); | |
364 | if (!skb) | |
365 | return; | |
366 | ||
27d7ff46 | 367 | skb_copy_to_linear_data(skb, msg, len); |
1da177e4 LT |
368 | skb->len += len; |
369 | ||
4bedb452 ACM |
370 | skb_push(skb, sizeof(*udph)); |
371 | skb_reset_transport_header(skb); | |
372 | udph = udp_hdr(skb); | |
1da177e4 LT |
373 | udph->source = htons(np->local_port); |
374 | udph->dest = htons(np->remote_port); | |
375 | udph->len = htons(udp_len); | |
376 | udph->check = 0; | |
e7557af5 HH |
377 | udph->check = csum_tcpudp_magic(np->local_ip, |
378 | np->remote_ip, | |
8e365eec | 379 | udp_len, IPPROTO_UDP, |
07f0757a | 380 | csum_partial(udph, udp_len, 0)); |
8e365eec | 381 | if (udph->check == 0) |
5e57dff2 | 382 | udph->check = CSUM_MANGLED_0; |
1da177e4 | 383 | |
e2d1bca7 ACM |
384 | skb_push(skb, sizeof(*iph)); |
385 | skb_reset_network_header(skb); | |
eddc9ec5 | 386 | iph = ip_hdr(skb); |
1da177e4 LT |
387 | |
388 | /* iph->version = 4; iph->ihl = 5; */ | |
389 | put_unaligned(0x45, (unsigned char *)iph); | |
390 | iph->tos = 0; | |
391 | put_unaligned(htons(ip_len), &(iph->tot_len)); | |
392 | iph->id = 0; | |
393 | iph->frag_off = 0; | |
394 | iph->ttl = 64; | |
395 | iph->protocol = IPPROTO_UDP; | |
396 | iph->check = 0; | |
e7557af5 HH |
397 | put_unaligned(np->local_ip, &(iph->saddr)); |
398 | put_unaligned(np->remote_ip, &(iph->daddr)); | |
1da177e4 LT |
399 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); |
400 | ||
401 | eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); | |
459a98ed | 402 | skb_reset_mac_header(skb); |
206daaf7 | 403 | skb->protocol = eth->h_proto = htons(ETH_P_IP); |
09538641 SH |
404 | memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); |
405 | memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); | |
1da177e4 LT |
406 | |
407 | skb->dev = np->dev; | |
408 | ||
409 | netpoll_send_skb(np, skb); | |
410 | } | |
9e34a5b5 | 411 | EXPORT_SYMBOL(netpoll_send_udp); |
1da177e4 LT |
412 | |
413 | static void arp_reply(struct sk_buff *skb) | |
414 | { | |
115c1d6e | 415 | struct netpoll_info *npinfo = skb->dev->npinfo; |
1da177e4 LT |
416 | struct arphdr *arp; |
417 | unsigned char *arp_ptr; | |
418 | int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; | |
252e3346 | 419 | __be32 sip, tip; |
47bbec02 | 420 | unsigned char *sha; |
1da177e4 | 421 | struct sk_buff *send_skb; |
508e14b4 DB |
422 | struct netpoll *np, *tmp; |
423 | unsigned long flags; | |
424 | int hits = 0; | |
425 | ||
426 | if (list_empty(&npinfo->rx_np)) | |
427 | return; | |
428 | ||
429 | /* Before checking the packet, we do some early | |
430 | inspection whether this is interesting at all */ | |
431 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
432 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
433 | if (np->dev == skb->dev) | |
434 | hits++; | |
435 | } | |
436 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 | 437 | |
508e14b4 DB |
438 | /* No netpoll struct is using this dev */ |
439 | if (!hits) | |
115c1d6e | 440 | return; |
1da177e4 LT |
441 | |
442 | /* No arp on this interface */ | |
443 | if (skb->dev->flags & IFF_NOARP) | |
444 | return; | |
445 | ||
988b7050 | 446 | if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) |
1da177e4 LT |
447 | return; |
448 | ||
c1d2bbe1 | 449 | skb_reset_network_header(skb); |
badff6d0 | 450 | skb_reset_transport_header(skb); |
d0a92be0 | 451 | arp = arp_hdr(skb); |
1da177e4 LT |
452 | |
453 | if ((arp->ar_hrd != htons(ARPHRD_ETHER) && | |
454 | arp->ar_hrd != htons(ARPHRD_IEEE802)) || | |
455 | arp->ar_pro != htons(ETH_P_IP) || | |
456 | arp->ar_op != htons(ARPOP_REQUEST)) | |
457 | return; | |
458 | ||
47bbec02 NH |
459 | arp_ptr = (unsigned char *)(arp+1); |
460 | /* save the location of the src hw addr */ | |
461 | sha = arp_ptr; | |
462 | arp_ptr += skb->dev->addr_len; | |
1da177e4 | 463 | memcpy(&sip, arp_ptr, 4); |
47bbec02 | 464 | arp_ptr += 4; |
508e14b4 DB |
465 | /* If we actually cared about dst hw addr, |
466 | it would get copied here */ | |
47bbec02 | 467 | arp_ptr += skb->dev->addr_len; |
1da177e4 LT |
468 | memcpy(&tip, arp_ptr, 4); |
469 | ||
470 | /* Should we ignore arp? */ | |
508e14b4 | 471 | if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) |
1da177e4 LT |
472 | return; |
473 | ||
988b7050 | 474 | size = arp_hdr_len(skb->dev); |
1da177e4 | 475 | |
508e14b4 DB |
476 | spin_lock_irqsave(&npinfo->rx_lock, flags); |
477 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { | |
478 | if (tip != np->local_ip) | |
479 | continue; | |
1da177e4 | 480 | |
508e14b4 DB |
481 | send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev), |
482 | LL_RESERVED_SPACE(np->dev)); | |
483 | if (!send_skb) | |
484 | continue; | |
1da177e4 | 485 | |
508e14b4 DB |
486 | skb_reset_network_header(send_skb); |
487 | arp = (struct arphdr *) skb_put(send_skb, size); | |
488 | send_skb->dev = skb->dev; | |
489 | send_skb->protocol = htons(ETH_P_ARP); | |
1da177e4 | 490 | |
508e14b4 DB |
491 | /* Fill the device header for the ARP frame */ |
492 | if (dev_hard_header(send_skb, skb->dev, ptype, | |
493 | sha, np->dev->dev_addr, | |
494 | send_skb->len) < 0) { | |
495 | kfree_skb(send_skb); | |
496 | continue; | |
497 | } | |
1da177e4 | 498 | |
508e14b4 DB |
499 | /* |
500 | * Fill out the arp protocol part. | |
501 | * | |
502 | * we only support ethernet device type, | |
503 | * which (according to RFC 1390) should | |
504 | * always equal 1 (Ethernet). | |
505 | */ | |
1da177e4 | 506 | |
508e14b4 DB |
507 | arp->ar_hrd = htons(np->dev->type); |
508 | arp->ar_pro = htons(ETH_P_IP); | |
509 | arp->ar_hln = np->dev->addr_len; | |
510 | arp->ar_pln = 4; | |
511 | arp->ar_op = htons(type); | |
512 | ||
513 | arp_ptr = (unsigned char *)(arp + 1); | |
514 | memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); | |
515 | arp_ptr += np->dev->addr_len; | |
516 | memcpy(arp_ptr, &tip, 4); | |
517 | arp_ptr += 4; | |
518 | memcpy(arp_ptr, sha, np->dev->addr_len); | |
519 | arp_ptr += np->dev->addr_len; | |
520 | memcpy(arp_ptr, &sip, 4); | |
521 | ||
522 | netpoll_send_skb(np, send_skb); | |
523 | ||
524 | /* If there are several rx_hooks for the same address, | |
525 | we're fine by sending a single reply */ | |
526 | break; | |
527 | } | |
528 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
1da177e4 LT |
529 | } |
530 | ||
531 | int __netpoll_rx(struct sk_buff *skb) | |
532 | { | |
533 | int proto, len, ulen; | |
508e14b4 | 534 | int hits = 0; |
1da177e4 LT |
535 | struct iphdr *iph; |
536 | struct udphdr *uh; | |
508e14b4 DB |
537 | struct netpoll_info *npinfo = skb->dev->npinfo; |
538 | struct netpoll *np, *tmp; | |
068c6e98 | 539 | |
508e14b4 | 540 | if (list_empty(&npinfo->rx_np)) |
1da177e4 | 541 | goto out; |
508e14b4 | 542 | |
1da177e4 LT |
543 | if (skb->dev->type != ARPHRD_ETHER) |
544 | goto out; | |
545 | ||
d9452e9f | 546 | /* check if netpoll clients need ARP */ |
724800d6 | 547 | if (skb->protocol == htons(ETH_P_ARP) && |
1da177e4 | 548 | atomic_read(&trapped)) { |
508e14b4 | 549 | skb_queue_tail(&npinfo->arp_tx, skb); |
1da177e4 LT |
550 | return 1; |
551 | } | |
552 | ||
553 | proto = ntohs(eth_hdr(skb)->h_proto); | |
554 | if (proto != ETH_P_IP) | |
555 | goto out; | |
556 | if (skb->pkt_type == PACKET_OTHERHOST) | |
557 | goto out; | |
558 | if (skb_shared(skb)) | |
559 | goto out; | |
560 | ||
561 | iph = (struct iphdr *)skb->data; | |
562 | if (!pskb_may_pull(skb, sizeof(struct iphdr))) | |
563 | goto out; | |
564 | if (iph->ihl < 5 || iph->version != 4) | |
565 | goto out; | |
566 | if (!pskb_may_pull(skb, iph->ihl*4)) | |
567 | goto out; | |
568 | if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) | |
569 | goto out; | |
570 | ||
571 | len = ntohs(iph->tot_len); | |
572 | if (skb->len < len || len < iph->ihl*4) | |
573 | goto out; | |
574 | ||
5e7d7fa5 AL |
575 | /* |
576 | * Our transport medium may have padded the buffer out. | |
577 | * Now We trim to the true length of the frame. | |
578 | */ | |
579 | if (pskb_trim_rcsum(skb, len)) | |
580 | goto out; | |
581 | ||
1da177e4 LT |
582 | if (iph->protocol != IPPROTO_UDP) |
583 | goto out; | |
584 | ||
585 | len -= iph->ihl*4; | |
586 | uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); | |
587 | ulen = ntohs(uh->len); | |
588 | ||
589 | if (ulen != len) | |
590 | goto out; | |
fb286bb2 | 591 | if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) |
1da177e4 | 592 | goto out; |
1da177e4 | 593 | |
508e14b4 DB |
594 | list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { |
595 | if (np->local_ip && np->local_ip != iph->daddr) | |
596 | continue; | |
597 | if (np->remote_ip && np->remote_ip != iph->saddr) | |
598 | continue; | |
599 | if (np->local_port && np->local_port != ntohs(uh->dest)) | |
600 | continue; | |
601 | ||
602 | np->rx_hook(np, ntohs(uh->source), | |
603 | (char *)(uh+1), | |
604 | ulen - sizeof(struct udphdr)); | |
605 | hits++; | |
606 | } | |
607 | ||
608 | if (!hits) | |
609 | goto out; | |
1da177e4 LT |
610 | |
611 | kfree_skb(skb); | |
612 | return 1; | |
613 | ||
614 | out: | |
615 | if (atomic_read(&trapped)) { | |
616 | kfree_skb(skb); | |
617 | return 1; | |
618 | } | |
619 | ||
620 | return 0; | |
621 | } | |
622 | ||
0bcc1816 SS |
623 | void netpoll_print_options(struct netpoll *np) |
624 | { | |
625 | printk(KERN_INFO "%s: local port %d\n", | |
626 | np->name, np->local_port); | |
e7557af5 HH |
627 | printk(KERN_INFO "%s: local IP %pI4\n", |
628 | np->name, &np->local_ip); | |
5fc05f87 | 629 | printk(KERN_INFO "%s: interface '%s'\n", |
0bcc1816 SS |
630 | np->name, np->dev_name); |
631 | printk(KERN_INFO "%s: remote port %d\n", | |
632 | np->name, np->remote_port); | |
e7557af5 HH |
633 | printk(KERN_INFO "%s: remote IP %pI4\n", |
634 | np->name, &np->remote_ip); | |
e174961c JB |
635 | printk(KERN_INFO "%s: remote ethernet address %pM\n", |
636 | np->name, np->remote_mac); | |
0bcc1816 | 637 | } |
9e34a5b5 | 638 | EXPORT_SYMBOL(netpoll_print_options); |
0bcc1816 | 639 | |
1da177e4 LT |
640 | int netpoll_parse_options(struct netpoll *np, char *opt) |
641 | { | |
642 | char *cur=opt, *delim; | |
643 | ||
c68b9070 | 644 | if (*cur != '@') { |
1da177e4 LT |
645 | if ((delim = strchr(cur, '@')) == NULL) |
646 | goto parse_failed; | |
c68b9070 DM |
647 | *delim = 0; |
648 | np->local_port = simple_strtol(cur, NULL, 10); | |
649 | cur = delim; | |
1da177e4 LT |
650 | } |
651 | cur++; | |
1da177e4 | 652 | |
c68b9070 | 653 | if (*cur != '/') { |
1da177e4 LT |
654 | if ((delim = strchr(cur, '/')) == NULL) |
655 | goto parse_failed; | |
c68b9070 | 656 | *delim = 0; |
e7557af5 | 657 | np->local_ip = in_aton(cur); |
c68b9070 | 658 | cur = delim; |
1da177e4 LT |
659 | } |
660 | cur++; | |
661 | ||
c68b9070 | 662 | if (*cur != ',') { |
1da177e4 LT |
663 | /* parse out dev name */ |
664 | if ((delim = strchr(cur, ',')) == NULL) | |
665 | goto parse_failed; | |
c68b9070 | 666 | *delim = 0; |
1da177e4 | 667 | strlcpy(np->dev_name, cur, sizeof(np->dev_name)); |
c68b9070 | 668 | cur = delim; |
1da177e4 LT |
669 | } |
670 | cur++; | |
671 | ||
c68b9070 | 672 | if (*cur != '@') { |
1da177e4 LT |
673 | /* dst port */ |
674 | if ((delim = strchr(cur, '@')) == NULL) | |
675 | goto parse_failed; | |
c68b9070 | 676 | *delim = 0; |
5fc05f87 AW |
677 | if (*cur == ' ' || *cur == '\t') |
678 | printk(KERN_INFO "%s: warning: whitespace" | |
679 | "is not allowed\n", np->name); | |
c68b9070 DM |
680 | np->remote_port = simple_strtol(cur, NULL, 10); |
681 | cur = delim; | |
1da177e4 LT |
682 | } |
683 | cur++; | |
1da177e4 LT |
684 | |
685 | /* dst ip */ | |
686 | if ((delim = strchr(cur, '/')) == NULL) | |
687 | goto parse_failed; | |
c68b9070 | 688 | *delim = 0; |
e7557af5 | 689 | np->remote_ip = in_aton(cur); |
c68b9070 | 690 | cur = delim + 1; |
1da177e4 | 691 | |
c68b9070 | 692 | if (*cur != 0) { |
1da177e4 LT |
693 | /* MAC address */ |
694 | if ((delim = strchr(cur, ':')) == NULL) | |
695 | goto parse_failed; | |
c68b9070 DM |
696 | *delim = 0; |
697 | np->remote_mac[0] = simple_strtol(cur, NULL, 16); | |
698 | cur = delim + 1; | |
1da177e4 LT |
699 | if ((delim = strchr(cur, ':')) == NULL) |
700 | goto parse_failed; | |
c68b9070 DM |
701 | *delim = 0; |
702 | np->remote_mac[1] = simple_strtol(cur, NULL, 16); | |
703 | cur = delim + 1; | |
1da177e4 LT |
704 | if ((delim = strchr(cur, ':')) == NULL) |
705 | goto parse_failed; | |
c68b9070 DM |
706 | *delim = 0; |
707 | np->remote_mac[2] = simple_strtol(cur, NULL, 16); | |
708 | cur = delim + 1; | |
1da177e4 LT |
709 | if ((delim = strchr(cur, ':')) == NULL) |
710 | goto parse_failed; | |
c68b9070 DM |
711 | *delim = 0; |
712 | np->remote_mac[3] = simple_strtol(cur, NULL, 16); | |
713 | cur = delim + 1; | |
1da177e4 LT |
714 | if ((delim = strchr(cur, ':')) == NULL) |
715 | goto parse_failed; | |
c68b9070 DM |
716 | *delim = 0; |
717 | np->remote_mac[4] = simple_strtol(cur, NULL, 16); | |
718 | cur = delim + 1; | |
719 | np->remote_mac[5] = simple_strtol(cur, NULL, 16); | |
1da177e4 LT |
720 | } |
721 | ||
0bcc1816 | 722 | netpoll_print_options(np); |
1da177e4 LT |
723 | |
724 | return 0; | |
725 | ||
726 | parse_failed: | |
5fc05f87 | 727 | printk(KERN_INFO "%s: couldn't parse config at '%s'!\n", |
1da177e4 LT |
728 | np->name, cur); |
729 | return -1; | |
730 | } | |
9e34a5b5 | 731 | EXPORT_SYMBOL(netpoll_parse_options); |
1da177e4 | 732 | |
8fdd95ec | 733 | int __netpoll_setup(struct netpoll *np) |
1da177e4 | 734 | { |
8fdd95ec | 735 | struct net_device *ndev = np->dev; |
115c1d6e | 736 | struct netpoll_info *npinfo; |
4247e161 | 737 | const struct net_device_ops *ops; |
fbeec2e1 | 738 | unsigned long flags; |
b41848b6 | 739 | int err; |
1da177e4 | 740 | |
8fdd95ec HX |
741 | if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || |
742 | !ndev->netdev_ops->ndo_poll_controller) { | |
743 | printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n", | |
744 | np->name, np->dev_name); | |
745 | err = -ENOTSUPP; | |
746 | goto out; | |
747 | } | |
748 | ||
749 | if (!ndev->npinfo) { | |
750 | npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL); | |
751 | if (!npinfo) { | |
752 | err = -ENOMEM; | |
753 | goto out; | |
754 | } | |
755 | ||
756 | npinfo->rx_flags = 0; | |
757 | INIT_LIST_HEAD(&npinfo->rx_np); | |
758 | ||
759 | spin_lock_init(&npinfo->rx_lock); | |
760 | skb_queue_head_init(&npinfo->arp_tx); | |
761 | skb_queue_head_init(&npinfo->txq); | |
762 | INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); | |
763 | ||
764 | atomic_set(&npinfo->refcnt, 1); | |
765 | ||
766 | ops = np->dev->netdev_ops; | |
767 | if (ops->ndo_netpoll_setup) { | |
768 | err = ops->ndo_netpoll_setup(ndev, npinfo); | |
769 | if (err) | |
770 | goto free_npinfo; | |
771 | } | |
772 | } else { | |
773 | npinfo = ndev->npinfo; | |
774 | atomic_inc(&npinfo->refcnt); | |
775 | } | |
776 | ||
777 | npinfo->netpoll = np; | |
778 | ||
779 | if (np->rx_hook) { | |
780 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
781 | npinfo->rx_flags |= NETPOLL_RX_ENABLED; | |
782 | list_add_tail(&np->rx, &npinfo->rx_np); | |
783 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
784 | } | |
785 | ||
786 | /* last thing to do is link it to the net device structure */ | |
787 | rcu_assign_pointer(ndev->npinfo, npinfo); | |
8fdd95ec HX |
788 | |
789 | return 0; | |
790 | ||
791 | free_npinfo: | |
792 | kfree(npinfo); | |
793 | out: | |
794 | return err; | |
795 | } | |
796 | EXPORT_SYMBOL_GPL(__netpoll_setup); | |
797 | ||
798 | int netpoll_setup(struct netpoll *np) | |
799 | { | |
800 | struct net_device *ndev = NULL; | |
801 | struct in_device *in_dev; | |
802 | int err; | |
803 | ||
1da177e4 | 804 | if (np->dev_name) |
881d966b | 805 | ndev = dev_get_by_name(&init_net, np->dev_name); |
1da177e4 LT |
806 | if (!ndev) { |
807 | printk(KERN_ERR "%s: %s doesn't exist, aborting.\n", | |
808 | np->name, np->dev_name); | |
b41848b6 | 809 | return -ENODEV; |
1da177e4 LT |
810 | } |
811 | ||
1da177e4 LT |
812 | if (!netif_running(ndev)) { |
813 | unsigned long atmost, atleast; | |
814 | ||
815 | printk(KERN_INFO "%s: device %s not up yet, forcing it\n", | |
816 | np->name, np->dev_name); | |
817 | ||
6756ae4b | 818 | rtnl_lock(); |
b41848b6 SH |
819 | err = dev_open(ndev); |
820 | rtnl_unlock(); | |
821 | ||
822 | if (err) { | |
1da177e4 | 823 | printk(KERN_ERR "%s: failed to open %s\n", |
b41848b6 | 824 | np->name, ndev->name); |
dbaa1541 | 825 | goto put; |
1da177e4 | 826 | } |
1da177e4 LT |
827 | |
828 | atleast = jiffies + HZ/10; | |
bff38771 | 829 | atmost = jiffies + carrier_timeout * HZ; |
1da177e4 LT |
830 | while (!netif_carrier_ok(ndev)) { |
831 | if (time_after(jiffies, atmost)) { | |
832 | printk(KERN_NOTICE | |
833 | "%s: timeout waiting for carrier\n", | |
834 | np->name); | |
835 | break; | |
836 | } | |
1b614fb9 | 837 | msleep(1); |
1da177e4 LT |
838 | } |
839 | ||
840 | /* If carrier appears to come up instantly, we don't | |
841 | * trust it and pause so that we don't pump all our | |
842 | * queued console messages into the bitbucket. | |
843 | */ | |
844 | ||
845 | if (time_before(jiffies, atleast)) { | |
846 | printk(KERN_NOTICE "%s: carrier detect appears" | |
847 | " untrustworthy, waiting 4 seconds\n", | |
848 | np->name); | |
849 | msleep(4000); | |
850 | } | |
851 | } | |
852 | ||
1da177e4 LT |
853 | if (!np->local_ip) { |
854 | rcu_read_lock(); | |
e5ed6399 | 855 | in_dev = __in_dev_get_rcu(ndev); |
1da177e4 LT |
856 | |
857 | if (!in_dev || !in_dev->ifa_list) { | |
858 | rcu_read_unlock(); | |
859 | printk(KERN_ERR "%s: no IP address for %s, aborting\n", | |
860 | np->name, np->dev_name); | |
b41848b6 | 861 | err = -EDESTADDRREQ; |
dbaa1541 | 862 | goto put; |
1da177e4 LT |
863 | } |
864 | ||
e7557af5 | 865 | np->local_ip = in_dev->ifa_list->ifa_local; |
1da177e4 | 866 | rcu_read_unlock(); |
e7557af5 | 867 | printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip); |
1da177e4 LT |
868 | } |
869 | ||
dbaa1541 HX |
870 | np->dev = ndev; |
871 | ||
872 | /* fill up the skb queue */ | |
873 | refill_skbs(); | |
874 | ||
875 | rtnl_lock(); | |
8fdd95ec | 876 | err = __netpoll_setup(np); |
dbaa1541 | 877 | rtnl_unlock(); |
53fb95d3 | 878 | |
8fdd95ec HX |
879 | if (err) |
880 | goto put; | |
881 | ||
1da177e4 LT |
882 | return 0; |
883 | ||
21edbb22 | 884 | put: |
1da177e4 | 885 | dev_put(ndev); |
b41848b6 | 886 | return err; |
1da177e4 | 887 | } |
9e34a5b5 | 888 | EXPORT_SYMBOL(netpoll_setup); |
1da177e4 | 889 | |
c68b9070 DM |
890 | static int __init netpoll_init(void) |
891 | { | |
a1bcfacd SH |
892 | skb_queue_head_init(&skb_pool); |
893 | return 0; | |
894 | } | |
895 | core_initcall(netpoll_init); | |
896 | ||
8fdd95ec | 897 | void __netpoll_cleanup(struct netpoll *np) |
1da177e4 | 898 | { |
fbeec2e1 JM |
899 | struct netpoll_info *npinfo; |
900 | unsigned long flags; | |
901 | ||
8fdd95ec HX |
902 | npinfo = np->dev->npinfo; |
903 | if (!npinfo) | |
dbaa1541 | 904 | return; |
93ec2c72 | 905 | |
8fdd95ec HX |
906 | if (!list_empty(&npinfo->rx_np)) { |
907 | spin_lock_irqsave(&npinfo->rx_lock, flags); | |
908 | list_del(&np->rx); | |
909 | if (list_empty(&npinfo->rx_np)) | |
910 | npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; | |
911 | spin_unlock_irqrestore(&npinfo->rx_lock, flags); | |
912 | } | |
de85d99e | 913 | |
8fdd95ec HX |
914 | if (atomic_dec_and_test(&npinfo->refcnt)) { |
915 | const struct net_device_ops *ops; | |
de85d99e | 916 | |
8fdd95ec HX |
917 | ops = np->dev->netdev_ops; |
918 | if (ops->ndo_netpoll_cleanup) | |
919 | ops->ndo_netpoll_cleanup(np->dev); | |
de85d99e | 920 | |
8fdd95ec | 921 | rcu_assign_pointer(np->dev->npinfo, NULL); |
de85d99e | 922 | |
dbaa1541 HX |
923 | /* avoid racing with NAPI reading npinfo */ |
924 | synchronize_rcu_bh(); | |
93ec2c72 | 925 | |
dbaa1541 HX |
926 | skb_queue_purge(&npinfo->arp_tx); |
927 | skb_queue_purge(&npinfo->txq); | |
928 | cancel_rearming_delayed_work(&npinfo->tx_work); | |
93ec2c72 | 929 | |
dbaa1541 HX |
930 | /* clean after last, unfinished work */ |
931 | __skb_queue_purge(&npinfo->txq); | |
932 | kfree(npinfo); | |
115c1d6e | 933 | } |
8fdd95ec HX |
934 | } |
935 | EXPORT_SYMBOL_GPL(__netpoll_cleanup); | |
fbeec2e1 | 936 | |
8fdd95ec HX |
937 | void netpoll_cleanup(struct netpoll *np) |
938 | { | |
939 | if (!np->dev) | |
940 | return; | |
dbaa1541 | 941 | |
8fdd95ec HX |
942 | rtnl_lock(); |
943 | __netpoll_cleanup(np); | |
944 | rtnl_unlock(); | |
945 | ||
946 | dev_put(np->dev); | |
1da177e4 LT |
947 | np->dev = NULL; |
948 | } | |
9e34a5b5 | 949 | EXPORT_SYMBOL(netpoll_cleanup); |
1da177e4 LT |
950 | |
951 | int netpoll_trap(void) | |
952 | { | |
953 | return atomic_read(&trapped); | |
954 | } | |
9e34a5b5 | 955 | EXPORT_SYMBOL(netpoll_trap); |
1da177e4 LT |
956 | |
957 | void netpoll_set_trap(int trap) | |
958 | { | |
959 | if (trap) | |
960 | atomic_inc(&trapped); | |
961 | else | |
962 | atomic_dec(&trapped); | |
963 | } | |
1da177e4 | 964 | EXPORT_SYMBOL(netpoll_set_trap); |