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 12 #include <linux/moduleparam.h> 13 #include <linux/netdevice.h> 14 #include <linux/etherdevice.h> 15 #include <linux/string.h> 16 #include <linux/if_arp.h> 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 <linux/slab.h> 26 #include <net/tcp.h> 27 #include <net/udp.h> 28 #include <asm/unaligned.h> 29 #include <trace/events/napi.h> 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 40 static struct sk_buff_head skb_pool; 41 42 static atomic_t trapped; 43 44 #define USEC_PER_POLL 50 45 #define NETPOLL_RX_ENABLED 1 46 #define NETPOLL_RX_DROP 2 47 48 #define MAX_SKB_SIZE \ 49 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \ 50 sizeof(struct iphdr) + sizeof(struct ethhdr)) 51 52 static void zap_completion_queue(void); 53 static void arp_reply(struct sk_buff *skb); 54 55 static unsigned int carrier_timeout = 4; 56 module_param(carrier_timeout, uint, 0644); 57 58 static void queue_process(struct work_struct *work) 59 { 60 struct netpoll_info *npinfo = 61 container_of(work, struct netpoll_info, tx_work.work); 62 struct sk_buff *skb; 63 unsigned long flags; 64 65 while ((skb = skb_dequeue(&npinfo->txq))) { 66 struct net_device *dev = skb->dev; 67 const struct net_device_ops *ops = dev->netdev_ops; 68 struct netdev_queue *txq; 69 70 if (!netif_device_present(dev) || !netif_running(dev)) { 71 __kfree_skb(skb); 72 continue; 73 } 74 75 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); 76 77 local_irq_save(flags); 78 __netif_tx_lock(txq, smp_processor_id()); 79 if (netif_tx_queue_stopped(txq) || 80 netif_tx_queue_frozen(txq) || 81 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { 82 skb_queue_head(&npinfo->txq, skb); 83 __netif_tx_unlock(txq); 84 local_irq_restore(flags); 85 86 schedule_delayed_work(&npinfo->tx_work, HZ/10); 87 return; 88 } 89 __netif_tx_unlock(txq); 90 local_irq_restore(flags); 91 } 92 } 93 94 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, 95 unsigned short ulen, __be32 saddr, __be32 daddr) 96 { 97 __wsum psum; 98 99 if (uh->check == 0 || skb_csum_unnecessary(skb)) 100 return 0; 101 102 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); 103 104 if (skb->ip_summed == CHECKSUM_COMPLETE && 105 !csum_fold(csum_add(psum, skb->csum))) 106 return 0; 107 108 skb->csum = psum; 109 110 return __skb_checksum_complete(skb); 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 */ 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; 140 141 npinfo->rx_flags |= NETPOLL_RX_DROP; 142 atomic_inc(&trapped); 143 set_bit(NAPI_STATE_NPSVC, &napi->state); 144 145 work = napi->poll(napi, budget); 146 trace_napi_poll(napi); 147 148 clear_bit(NAPI_STATE_NPSVC, &napi->state); 149 atomic_dec(&trapped); 150 npinfo->rx_flags &= ~NETPOLL_RX_DROP; 151 152 return budget - work; 153 } 154 155 static void poll_napi(struct net_device *dev) 156 { 157 struct napi_struct *napi; 158 int budget = 16; 159 160 list_for_each_entry(napi, &dev->napi_list, dev_list) { 161 if (napi->poll_owner != smp_processor_id() && 162 spin_trylock(&napi->poll_lock)) { 163 budget = poll_one_napi(dev->npinfo, napi, budget); 164 spin_unlock(&napi->poll_lock); 165 166 if (!budget) 167 break; 168 } 169 } 170 } 171 172 static void service_arp_queue(struct netpoll_info *npi) 173 { 174 if (npi) { 175 struct sk_buff *skb; 176 177 while ((skb = skb_dequeue(&npi->arp_tx))) 178 arp_reply(skb); 179 } 180 } 181 182 void netpoll_poll_dev(struct net_device *dev) 183 { 184 const struct net_device_ops *ops; 185 186 if (!dev || !netif_running(dev)) 187 return; 188 189 ops = dev->netdev_ops; 190 if (!ops->ndo_poll_controller) 191 return; 192 193 /* Process pending work on NIC */ 194 ops->ndo_poll_controller(dev); 195 196 poll_napi(dev); 197 198 service_arp_queue(dev->npinfo); 199 200 zap_completion_queue(); 201 } 202 EXPORT_SYMBOL(netpoll_poll_dev); 203 204 void netpoll_poll(struct netpoll *np) 205 { 206 netpoll_poll_dev(np->dev); 207 } 208 EXPORT_SYMBOL(netpoll_poll); 209 210 static void refill_skbs(void) 211 { 212 struct sk_buff *skb; 213 unsigned long flags; 214 215 spin_lock_irqsave(&skb_pool.lock, flags); 216 while (skb_pool.qlen < MAX_SKBS) { 217 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); 218 if (!skb) 219 break; 220 221 __skb_queue_tail(&skb_pool, skb); 222 } 223 spin_unlock_irqrestore(&skb_pool.lock, flags); 224 } 225 226 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 } 253 254 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) 255 { 256 int count = 0; 257 struct sk_buff *skb; 258 259 zap_completion_queue(); 260 refill_skbs(); 261 repeat: 262 263 skb = alloc_skb(len, GFP_ATOMIC); 264 if (!skb) 265 skb = skb_dequeue(&skb_pool); 266 267 if (!skb) { 268 if (++count < 10) { 269 netpoll_poll(np); 270 goto repeat; 271 } 272 return NULL; 273 } 274 275 atomic_set(&skb->users, 1); 276 skb_reserve(skb, reserve); 277 return skb; 278 } 279 280 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 } 290 291 void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) 292 { 293 int status = NETDEV_TX_BUSY; 294 unsigned long tries; 295 struct net_device *dev = np->dev; 296 const struct net_device_ops *ops = dev->netdev_ops; 297 /* It is up to the caller to keep npinfo alive. */ 298 struct netpoll_info *npinfo = np->dev->npinfo; 299 300 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { 301 __kfree_skb(skb); 302 return; 303 } 304 305 /* don't get messages out of order, and no recursion */ 306 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { 307 struct netdev_queue *txq; 308 unsigned long flags; 309 310 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); 311 312 local_irq_save(flags); 313 /* try until next clock tick */ 314 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; 315 tries > 0; --tries) { 316 if (__netif_tx_trylock(txq)) { 317 if (!netif_tx_queue_stopped(txq)) { 318 dev->priv_flags |= IFF_IN_NETPOLL; 319 status = ops->ndo_start_xmit(skb, dev); 320 dev->priv_flags &= ~IFF_IN_NETPOLL; 321 if (status == NETDEV_TX_OK) 322 txq_trans_update(txq); 323 } 324 __netif_tx_unlock(txq); 325 326 if (status == NETDEV_TX_OK) 327 break; 328 329 } 330 331 /* tickle device maybe there is some cleanup */ 332 netpoll_poll(np); 333 334 udelay(USEC_PER_POLL); 335 } 336 337 WARN_ONCE(!irqs_disabled(), 338 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", 339 dev->name, ops->ndo_start_xmit); 340 341 local_irq_restore(flags); 342 } 343 344 if (status != NETDEV_TX_OK) { 345 skb_queue_tail(&npinfo->txq, skb); 346 schedule_delayed_work(&npinfo->tx_work,0); 347 } 348 } 349 EXPORT_SYMBOL(netpoll_send_skb); 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; 366 367 skb_copy_to_linear_data(skb, msg, len); 368 skb->len += len; 369 370 skb_push(skb, sizeof(*udph)); 371 skb_reset_transport_header(skb); 372 udph = udp_hdr(skb); 373 udph->source = htons(np->local_port); 374 udph->dest = htons(np->remote_port); 375 udph->len = htons(udp_len); 376 udph->check = 0; 377 udph->check = csum_tcpudp_magic(np->local_ip, 378 np->remote_ip, 379 udp_len, IPPROTO_UDP, 380 csum_partial(udph, udp_len, 0)); 381 if (udph->check == 0) 382 udph->check = CSUM_MANGLED_0; 383 384 skb_push(skb, sizeof(*iph)); 385 skb_reset_network_header(skb); 386 iph = ip_hdr(skb); 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; 397 put_unaligned(np->local_ip, &(iph->saddr)); 398 put_unaligned(np->remote_ip, &(iph->daddr)); 399 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 400 401 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 402 skb_reset_mac_header(skb); 403 skb->protocol = eth->h_proto = htons(ETH_P_IP); 404 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); 405 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); 406 407 skb->dev = np->dev; 408 409 netpoll_send_skb(np, skb); 410 } 411 EXPORT_SYMBOL(netpoll_send_udp); 412 413 static void arp_reply(struct sk_buff *skb) 414 { 415 struct netpoll_info *npinfo = skb->dev->npinfo; 416 struct arphdr *arp; 417 unsigned char *arp_ptr; 418 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; 419 __be32 sip, tip; 420 unsigned char *sha; 421 struct sk_buff *send_skb; 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); 437 438 /* No netpoll struct is using this dev */ 439 if (!hits) 440 return; 441 442 /* No arp on this interface */ 443 if (skb->dev->flags & IFF_NOARP) 444 return; 445 446 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) 447 return; 448 449 skb_reset_network_header(skb); 450 skb_reset_transport_header(skb); 451 arp = arp_hdr(skb); 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; 458 459 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; 463 memcpy(&sip, arp_ptr, 4); 464 arp_ptr += 4; 465 /* If we actually cared about dst hw addr, 466 it would get copied here */ 467 arp_ptr += skb->dev->addr_len; 468 memcpy(&tip, arp_ptr, 4); 469 470 /* Should we ignore arp? */ 471 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) 472 return; 473 474 size = arp_hdr_len(skb->dev); 475 476 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; 480 481 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev), 482 LL_RESERVED_SPACE(np->dev)); 483 if (!send_skb) 484 continue; 485 486 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); 490 491 /* 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 } 498 499 /* 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 */ 506 507 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); 529 } 530 531 int __netpoll_rx(struct sk_buff *skb) 532 { 533 int proto, len, ulen; 534 int hits = 0; 535 struct iphdr *iph; 536 struct udphdr *uh; 537 struct netpoll_info *npinfo = skb->dev->npinfo; 538 struct netpoll *np, *tmp; 539 540 if (list_empty(&npinfo->rx_np)) 541 goto out; 542 543 if (skb->dev->type != ARPHRD_ETHER) 544 goto out; 545 546 /* check if netpoll clients need ARP */ 547 if (skb->protocol == htons(ETH_P_ARP) && 548 atomic_read(&trapped)) { 549 skb_queue_tail(&npinfo->arp_tx, skb); 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; 574 575 /* 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; 581 582 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; 591 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) 592 goto out; 593 594 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; 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 } 622 623 void netpoll_print_options(struct netpoll *np) 624 { 625 printk(KERN_INFO "%s: local port %d\n", 626 np->name, np->local_port); 627 printk(KERN_INFO "%s: local IP %pI4\n", 628 np->name, &np->local_ip); 629 printk(KERN_INFO "%s: interface '%s'\n", 630 np->name, np->dev_name); 631 printk(KERN_INFO "%s: remote port %d\n", 632 np->name, np->remote_port); 633 printk(KERN_INFO "%s: remote IP %pI4\n", 634 np->name, &np->remote_ip); 635 printk(KERN_INFO "%s: remote ethernet address %pM\n", 636 np->name, np->remote_mac); 637 } 638 EXPORT_SYMBOL(netpoll_print_options); 639 640 int netpoll_parse_options(struct netpoll *np, char *opt) 641 { 642 char *cur=opt, *delim; 643 644 if (*cur != '@') { 645 if ((delim = strchr(cur, '@')) == NULL) 646 goto parse_failed; 647 *delim = 0; 648 np->local_port = simple_strtol(cur, NULL, 10); 649 cur = delim; 650 } 651 cur++; 652 653 if (*cur != '/') { 654 if ((delim = strchr(cur, '/')) == NULL) 655 goto parse_failed; 656 *delim = 0; 657 np->local_ip = in_aton(cur); 658 cur = delim; 659 } 660 cur++; 661 662 if (*cur != ',') { 663 /* parse out dev name */ 664 if ((delim = strchr(cur, ',')) == NULL) 665 goto parse_failed; 666 *delim = 0; 667 strlcpy(np->dev_name, cur, sizeof(np->dev_name)); 668 cur = delim; 669 } 670 cur++; 671 672 if (*cur != '@') { 673 /* dst port */ 674 if ((delim = strchr(cur, '@')) == NULL) 675 goto parse_failed; 676 *delim = 0; 677 if (*cur == ' ' || *cur == '\t') 678 printk(KERN_INFO "%s: warning: whitespace" 679 "is not allowed\n", np->name); 680 np->remote_port = simple_strtol(cur, NULL, 10); 681 cur = delim; 682 } 683 cur++; 684 685 /* dst ip */ 686 if ((delim = strchr(cur, '/')) == NULL) 687 goto parse_failed; 688 *delim = 0; 689 np->remote_ip = in_aton(cur); 690 cur = delim + 1; 691 692 if (*cur != 0) { 693 /* MAC address */ 694 if ((delim = strchr(cur, ':')) == NULL) 695 goto parse_failed; 696 *delim = 0; 697 np->remote_mac[0] = simple_strtol(cur, NULL, 16); 698 cur = delim + 1; 699 if ((delim = strchr(cur, ':')) == NULL) 700 goto parse_failed; 701 *delim = 0; 702 np->remote_mac[1] = simple_strtol(cur, NULL, 16); 703 cur = delim + 1; 704 if ((delim = strchr(cur, ':')) == NULL) 705 goto parse_failed; 706 *delim = 0; 707 np->remote_mac[2] = simple_strtol(cur, NULL, 16); 708 cur = delim + 1; 709 if ((delim = strchr(cur, ':')) == NULL) 710 goto parse_failed; 711 *delim = 0; 712 np->remote_mac[3] = simple_strtol(cur, NULL, 16); 713 cur = delim + 1; 714 if ((delim = strchr(cur, ':')) == NULL) 715 goto parse_failed; 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); 720 } 721 722 netpoll_print_options(np); 723 724 return 0; 725 726 parse_failed: 727 printk(KERN_INFO "%s: couldn't parse config at '%s'!\n", 728 np->name, cur); 729 return -1; 730 } 731 EXPORT_SYMBOL(netpoll_parse_options); 732 733 int __netpoll_setup(struct netpoll *np) 734 { 735 struct net_device *ndev = np->dev; 736 struct netpoll_info *npinfo; 737 const struct net_device_ops *ops; 738 unsigned long flags; 739 int err; 740 741 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); 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; 803 804 if (np->dev_name) 805 ndev = dev_get_by_name(&init_net, np->dev_name); 806 if (!ndev) { 807 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n", 808 np->name, np->dev_name); 809 return -ENODEV; 810 } 811 812 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); 817 818 rtnl_lock(); 819 err = dev_open(ndev); 820 rtnl_unlock(); 821 822 if (err) { 823 printk(KERN_ERR "%s: failed to open %s\n", 824 np->name, ndev->name); 825 goto put; 826 } 827 828 atleast = jiffies + HZ/10; 829 atmost = jiffies + carrier_timeout * HZ; 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 } 837 msleep(1); 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 } 852 853 if (!np->local_ip) { 854 rcu_read_lock(); 855 in_dev = __in_dev_get_rcu(ndev); 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); 861 err = -EDESTADDRREQ; 862 goto put; 863 } 864 865 np->local_ip = in_dev->ifa_list->ifa_local; 866 rcu_read_unlock(); 867 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip); 868 } 869 870 np->dev = ndev; 871 872 /* fill up the skb queue */ 873 refill_skbs(); 874 875 rtnl_lock(); 876 err = __netpoll_setup(np); 877 rtnl_unlock(); 878 879 if (err) 880 goto put; 881 882 return 0; 883 884 put: 885 dev_put(ndev); 886 return err; 887 } 888 EXPORT_SYMBOL(netpoll_setup); 889 890 static int __init netpoll_init(void) 891 { 892 skb_queue_head_init(&skb_pool); 893 return 0; 894 } 895 core_initcall(netpoll_init); 896 897 void __netpoll_cleanup(struct netpoll *np) 898 { 899 struct netpoll_info *npinfo; 900 unsigned long flags; 901 902 npinfo = np->dev->npinfo; 903 if (!npinfo) 904 return; 905 906 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 } 913 914 if (atomic_dec_and_test(&npinfo->refcnt)) { 915 const struct net_device_ops *ops; 916 917 ops = np->dev->netdev_ops; 918 if (ops->ndo_netpoll_cleanup) 919 ops->ndo_netpoll_cleanup(np->dev); 920 921 rcu_assign_pointer(np->dev->npinfo, NULL); 922 923 /* avoid racing with NAPI reading npinfo */ 924 synchronize_rcu_bh(); 925 926 skb_queue_purge(&npinfo->arp_tx); 927 skb_queue_purge(&npinfo->txq); 928 cancel_rearming_delayed_work(&npinfo->tx_work); 929 930 /* clean after last, unfinished work */ 931 __skb_queue_purge(&npinfo->txq); 932 kfree(npinfo); 933 } 934 } 935 EXPORT_SYMBOL_GPL(__netpoll_cleanup); 936 937 void netpoll_cleanup(struct netpoll *np) 938 { 939 if (!np->dev) 940 return; 941 942 rtnl_lock(); 943 __netpoll_cleanup(np); 944 rtnl_unlock(); 945 946 dev_put(np->dev); 947 np->dev = NULL; 948 } 949 EXPORT_SYMBOL(netpoll_cleanup); 950 951 int netpoll_trap(void) 952 { 953 return atomic_read(&trapped); 954 } 955 EXPORT_SYMBOL(netpoll_trap); 956 957 void netpoll_set_trap(int trap) 958 { 959 if (trap) 960 atomic_inc(&trapped); 961 else 962 atomic_dec(&trapped); 963 } 964 EXPORT_SYMBOL(netpoll_set_trap); 965