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