netpoll.c revision bc3b2d7fb9b014d75ebb79ba371a763dbab5e8cf
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 <linux/export.h> 27#include <net/tcp.h> 28#include <net/udp.h> 29#include <asm/unaligned.h> 30#include <trace/events/napi.h> 31 32/* 33 * We maintain a small pool of fully-sized skbs, to make sure the 34 * message gets out even in extreme OOM situations. 35 */ 36 37#define MAX_UDP_CHUNK 1460 38#define MAX_SKBS 32 39 40static struct sk_buff_head skb_pool; 41 42static 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 52static void zap_completion_queue(void); 53static void arp_reply(struct sk_buff *skb); 54 55static unsigned int carrier_timeout = 4; 56module_param(carrier_timeout, uint, 0644); 57 58static 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_frozen_or_stopped(txq) || 80 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { 81 skb_queue_head(&npinfo->txq, skb); 82 __netif_tx_unlock(txq); 83 local_irq_restore(flags); 84 85 schedule_delayed_work(&npinfo->tx_work, HZ/10); 86 return; 87 } 88 __netif_tx_unlock(txq); 89 local_irq_restore(flags); 90 } 91} 92 93static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, 94 unsigned short ulen, __be32 saddr, __be32 daddr) 95{ 96 __wsum psum; 97 98 if (uh->check == 0 || skb_csum_unnecessary(skb)) 99 return 0; 100 101 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); 102 103 if (skb->ip_summed == CHECKSUM_COMPLETE && 104 !csum_fold(csum_add(psum, skb->csum))) 105 return 0; 106 107 skb->csum = psum; 108 109 return __skb_checksum_complete(skb); 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 */ 128static 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; 139 140 npinfo->rx_flags |= NETPOLL_RX_DROP; 141 atomic_inc(&trapped); 142 set_bit(NAPI_STATE_NPSVC, &napi->state); 143 144 work = napi->poll(napi, budget); 145 trace_napi_poll(napi); 146 147 clear_bit(NAPI_STATE_NPSVC, &napi->state); 148 atomic_dec(&trapped); 149 npinfo->rx_flags &= ~NETPOLL_RX_DROP; 150 151 return budget - work; 152} 153 154static void poll_napi(struct net_device *dev) 155{ 156 struct napi_struct *napi; 157 int budget = 16; 158 159 list_for_each_entry(napi, &dev->napi_list, dev_list) { 160 if (napi->poll_owner != smp_processor_id() && 161 spin_trylock(&napi->poll_lock)) { 162 budget = poll_one_napi(dev->npinfo, napi, budget); 163 spin_unlock(&napi->poll_lock); 164 165 if (!budget) 166 break; 167 } 168 } 169} 170 171static void service_arp_queue(struct netpoll_info *npi) 172{ 173 if (npi) { 174 struct sk_buff *skb; 175 176 while ((skb = skb_dequeue(&npi->arp_tx))) 177 arp_reply(skb); 178 } 179} 180 181static void netpoll_poll_dev(struct net_device *dev) 182{ 183 const struct net_device_ops *ops; 184 185 if (!dev || !netif_running(dev)) 186 return; 187 188 ops = dev->netdev_ops; 189 if (!ops->ndo_poll_controller) 190 return; 191 192 /* Process pending work on NIC */ 193 ops->ndo_poll_controller(dev); 194 195 poll_napi(dev); 196 197 if (dev->priv_flags & IFF_SLAVE) { 198 if (dev->npinfo) { 199 struct net_device *bond_dev = dev->master; 200 struct sk_buff *skb; 201 while ((skb = skb_dequeue(&dev->npinfo->arp_tx))) { 202 skb->dev = bond_dev; 203 skb_queue_tail(&bond_dev->npinfo->arp_tx, skb); 204 } 205 } 206 } 207 208 service_arp_queue(dev->npinfo); 209 210 zap_completion_queue(); 211} 212 213static void refill_skbs(void) 214{ 215 struct sk_buff *skb; 216 unsigned long flags; 217 218 spin_lock_irqsave(&skb_pool.lock, flags); 219 while (skb_pool.qlen < MAX_SKBS) { 220 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); 221 if (!skb) 222 break; 223 224 __skb_queue_tail(&skb_pool, skb); 225 } 226 spin_unlock_irqrestore(&skb_pool.lock, flags); 227} 228 229static void zap_completion_queue(void) 230{ 231 unsigned long flags; 232 struct softnet_data *sd = &get_cpu_var(softnet_data); 233 234 if (sd->completion_queue) { 235 struct sk_buff *clist; 236 237 local_irq_save(flags); 238 clist = sd->completion_queue; 239 sd->completion_queue = NULL; 240 local_irq_restore(flags); 241 242 while (clist != NULL) { 243 struct sk_buff *skb = clist; 244 clist = clist->next; 245 if (skb->destructor) { 246 atomic_inc(&skb->users); 247 dev_kfree_skb_any(skb); /* put this one back */ 248 } else { 249 __kfree_skb(skb); 250 } 251 } 252 } 253 254 put_cpu_var(softnet_data); 255} 256 257static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) 258{ 259 int count = 0; 260 struct sk_buff *skb; 261 262 zap_completion_queue(); 263 refill_skbs(); 264repeat: 265 266 skb = alloc_skb(len, GFP_ATOMIC); 267 if (!skb) 268 skb = skb_dequeue(&skb_pool); 269 270 if (!skb) { 271 if (++count < 10) { 272 netpoll_poll_dev(np->dev); 273 goto repeat; 274 } 275 return NULL; 276 } 277 278 atomic_set(&skb->users, 1); 279 skb_reserve(skb, reserve); 280 return skb; 281} 282 283static int netpoll_owner_active(struct net_device *dev) 284{ 285 struct napi_struct *napi; 286 287 list_for_each_entry(napi, &dev->napi_list, dev_list) { 288 if (napi->poll_owner == smp_processor_id()) 289 return 1; 290 } 291 return 0; 292} 293 294void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, 295 struct net_device *dev) 296{ 297 int status = NETDEV_TX_BUSY; 298 unsigned long tries; 299 const struct net_device_ops *ops = dev->netdev_ops; 300 /* It is up to the caller to keep npinfo alive. */ 301 struct netpoll_info *npinfo = np->dev->npinfo; 302 303 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { 304 __kfree_skb(skb); 305 return; 306 } 307 308 /* don't get messages out of order, and no recursion */ 309 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { 310 struct netdev_queue *txq; 311 unsigned long flags; 312 313 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); 314 315 local_irq_save(flags); 316 /* try until next clock tick */ 317 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; 318 tries > 0; --tries) { 319 if (__netif_tx_trylock(txq)) { 320 if (!netif_tx_queue_stopped(txq)) { 321 status = ops->ndo_start_xmit(skb, dev); 322 if (status == NETDEV_TX_OK) 323 txq_trans_update(txq); 324 } 325 __netif_tx_unlock(txq); 326 327 if (status == NETDEV_TX_OK) 328 break; 329 330 } 331 332 /* tickle device maybe there is some cleanup */ 333 netpoll_poll_dev(np->dev); 334 335 udelay(USEC_PER_POLL); 336 } 337 338 WARN_ONCE(!irqs_disabled(), 339 "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n", 340 dev->name, ops->ndo_start_xmit); 341 342 local_irq_restore(flags); 343 } 344 345 if (status != NETDEV_TX_OK) { 346 skb_queue_tail(&npinfo->txq, skb); 347 schedule_delayed_work(&npinfo->tx_work,0); 348 } 349} 350EXPORT_SYMBOL(netpoll_send_skb_on_dev); 351 352void netpoll_send_udp(struct netpoll *np, const char *msg, int len) 353{ 354 int total_len, eth_len, ip_len, udp_len; 355 struct sk_buff *skb; 356 struct udphdr *udph; 357 struct iphdr *iph; 358 struct ethhdr *eth; 359 360 udp_len = len + sizeof(*udph); 361 ip_len = eth_len = udp_len + sizeof(*iph); 362 total_len = eth_len + ETH_HLEN + NET_IP_ALIGN; 363 364 skb = find_skb(np, total_len, total_len - len); 365 if (!skb) 366 return; 367 368 skb_copy_to_linear_data(skb, msg, len); 369 skb->len += len; 370 371 skb_push(skb, sizeof(*udph)); 372 skb_reset_transport_header(skb); 373 udph = udp_hdr(skb); 374 udph->source = htons(np->local_port); 375 udph->dest = htons(np->remote_port); 376 udph->len = htons(udp_len); 377 udph->check = 0; 378 udph->check = csum_tcpudp_magic(np->local_ip, 379 np->remote_ip, 380 udp_len, IPPROTO_UDP, 381 csum_partial(udph, udp_len, 0)); 382 if (udph->check == 0) 383 udph->check = CSUM_MANGLED_0; 384 385 skb_push(skb, sizeof(*iph)); 386 skb_reset_network_header(skb); 387 iph = ip_hdr(skb); 388 389 /* iph->version = 4; iph->ihl = 5; */ 390 put_unaligned(0x45, (unsigned char *)iph); 391 iph->tos = 0; 392 put_unaligned(htons(ip_len), &(iph->tot_len)); 393 iph->id = 0; 394 iph->frag_off = 0; 395 iph->ttl = 64; 396 iph->protocol = IPPROTO_UDP; 397 iph->check = 0; 398 put_unaligned(np->local_ip, &(iph->saddr)); 399 put_unaligned(np->remote_ip, &(iph->daddr)); 400 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 401 402 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 403 skb_reset_mac_header(skb); 404 skb->protocol = eth->h_proto = htons(ETH_P_IP); 405 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); 406 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); 407 408 skb->dev = np->dev; 409 410 netpoll_send_skb(np, skb); 411} 412EXPORT_SYMBOL(netpoll_send_udp); 413 414static void arp_reply(struct sk_buff *skb) 415{ 416 struct netpoll_info *npinfo = skb->dev->npinfo; 417 struct arphdr *arp; 418 unsigned char *arp_ptr; 419 int size, type = ARPOP_REPLY, ptype = ETH_P_ARP; 420 __be32 sip, tip; 421 unsigned char *sha; 422 struct sk_buff *send_skb; 423 struct netpoll *np, *tmp; 424 unsigned long flags; 425 int hits = 0; 426 427 if (list_empty(&npinfo->rx_np)) 428 return; 429 430 /* Before checking the packet, we do some early 431 inspection whether this is interesting at all */ 432 spin_lock_irqsave(&npinfo->rx_lock, flags); 433 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 434 if (np->dev == skb->dev) 435 hits++; 436 } 437 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 438 439 /* No netpoll struct is using this dev */ 440 if (!hits) 441 return; 442 443 /* No arp on this interface */ 444 if (skb->dev->flags & IFF_NOARP) 445 return; 446 447 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) 448 return; 449 450 skb_reset_network_header(skb); 451 skb_reset_transport_header(skb); 452 arp = arp_hdr(skb); 453 454 if ((arp->ar_hrd != htons(ARPHRD_ETHER) && 455 arp->ar_hrd != htons(ARPHRD_IEEE802)) || 456 arp->ar_pro != htons(ETH_P_IP) || 457 arp->ar_op != htons(ARPOP_REQUEST)) 458 return; 459 460 arp_ptr = (unsigned char *)(arp+1); 461 /* save the location of the src hw addr */ 462 sha = arp_ptr; 463 arp_ptr += skb->dev->addr_len; 464 memcpy(&sip, arp_ptr, 4); 465 arp_ptr += 4; 466 /* If we actually cared about dst hw addr, 467 it would get copied here */ 468 arp_ptr += skb->dev->addr_len; 469 memcpy(&tip, arp_ptr, 4); 470 471 /* Should we ignore arp? */ 472 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) 473 return; 474 475 size = arp_hdr_len(skb->dev); 476 477 spin_lock_irqsave(&npinfo->rx_lock, flags); 478 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 479 if (tip != np->local_ip) 480 continue; 481 482 send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev), 483 LL_RESERVED_SPACE(np->dev)); 484 if (!send_skb) 485 continue; 486 487 skb_reset_network_header(send_skb); 488 arp = (struct arphdr *) skb_put(send_skb, size); 489 send_skb->dev = skb->dev; 490 send_skb->protocol = htons(ETH_P_ARP); 491 492 /* Fill the device header for the ARP frame */ 493 if (dev_hard_header(send_skb, skb->dev, ptype, 494 sha, np->dev->dev_addr, 495 send_skb->len) < 0) { 496 kfree_skb(send_skb); 497 continue; 498 } 499 500 /* 501 * Fill out the arp protocol part. 502 * 503 * we only support ethernet device type, 504 * which (according to RFC 1390) should 505 * always equal 1 (Ethernet). 506 */ 507 508 arp->ar_hrd = htons(np->dev->type); 509 arp->ar_pro = htons(ETH_P_IP); 510 arp->ar_hln = np->dev->addr_len; 511 arp->ar_pln = 4; 512 arp->ar_op = htons(type); 513 514 arp_ptr = (unsigned char *)(arp + 1); 515 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); 516 arp_ptr += np->dev->addr_len; 517 memcpy(arp_ptr, &tip, 4); 518 arp_ptr += 4; 519 memcpy(arp_ptr, sha, np->dev->addr_len); 520 arp_ptr += np->dev->addr_len; 521 memcpy(arp_ptr, &sip, 4); 522 523 netpoll_send_skb(np, send_skb); 524 525 /* If there are several rx_hooks for the same address, 526 we're fine by sending a single reply */ 527 break; 528 } 529 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 530} 531 532int __netpoll_rx(struct sk_buff *skb) 533{ 534 int proto, len, ulen; 535 int hits = 0; 536 const struct iphdr *iph; 537 struct udphdr *uh; 538 struct netpoll_info *npinfo = skb->dev->npinfo; 539 struct netpoll *np, *tmp; 540 541 if (list_empty(&npinfo->rx_np)) 542 goto out; 543 544 if (skb->dev->type != ARPHRD_ETHER) 545 goto out; 546 547 /* check if netpoll clients need ARP */ 548 if (skb->protocol == htons(ETH_P_ARP) && 549 atomic_read(&trapped)) { 550 skb_queue_tail(&npinfo->arp_tx, skb); 551 return 1; 552 } 553 554 proto = ntohs(eth_hdr(skb)->h_proto); 555 if (proto != ETH_P_IP) 556 goto out; 557 if (skb->pkt_type == PACKET_OTHERHOST) 558 goto out; 559 if (skb_shared(skb)) 560 goto out; 561 562 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 563 goto out; 564 iph = (struct iphdr *)skb->data; 565 if (iph->ihl < 5 || iph->version != 4) 566 goto out; 567 if (!pskb_may_pull(skb, iph->ihl*4)) 568 goto out; 569 iph = (struct iphdr *)skb->data; 570 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) 571 goto out; 572 573 len = ntohs(iph->tot_len); 574 if (skb->len < len || len < iph->ihl*4) 575 goto out; 576 577 /* 578 * Our transport medium may have padded the buffer out. 579 * Now We trim to the true length of the frame. 580 */ 581 if (pskb_trim_rcsum(skb, len)) 582 goto out; 583 584 iph = (struct iphdr *)skb->data; 585 if (iph->protocol != IPPROTO_UDP) 586 goto out; 587 588 len -= iph->ihl*4; 589 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); 590 ulen = ntohs(uh->len); 591 592 if (ulen != len) 593 goto out; 594 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) 595 goto out; 596 597 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 598 if (np->local_ip && np->local_ip != iph->daddr) 599 continue; 600 if (np->remote_ip && np->remote_ip != iph->saddr) 601 continue; 602 if (np->local_port && np->local_port != ntohs(uh->dest)) 603 continue; 604 605 np->rx_hook(np, ntohs(uh->source), 606 (char *)(uh+1), 607 ulen - sizeof(struct udphdr)); 608 hits++; 609 } 610 611 if (!hits) 612 goto out; 613 614 kfree_skb(skb); 615 return 1; 616 617out: 618 if (atomic_read(&trapped)) { 619 kfree_skb(skb); 620 return 1; 621 } 622 623 return 0; 624} 625 626void netpoll_print_options(struct netpoll *np) 627{ 628 printk(KERN_INFO "%s: local port %d\n", 629 np->name, np->local_port); 630 printk(KERN_INFO "%s: local IP %pI4\n", 631 np->name, &np->local_ip); 632 printk(KERN_INFO "%s: interface '%s'\n", 633 np->name, np->dev_name); 634 printk(KERN_INFO "%s: remote port %d\n", 635 np->name, np->remote_port); 636 printk(KERN_INFO "%s: remote IP %pI4\n", 637 np->name, &np->remote_ip); 638 printk(KERN_INFO "%s: remote ethernet address %pM\n", 639 np->name, np->remote_mac); 640} 641EXPORT_SYMBOL(netpoll_print_options); 642 643int netpoll_parse_options(struct netpoll *np, char *opt) 644{ 645 char *cur=opt, *delim; 646 647 if (*cur != '@') { 648 if ((delim = strchr(cur, '@')) == NULL) 649 goto parse_failed; 650 *delim = 0; 651 np->local_port = simple_strtol(cur, NULL, 10); 652 cur = delim; 653 } 654 cur++; 655 656 if (*cur != '/') { 657 if ((delim = strchr(cur, '/')) == NULL) 658 goto parse_failed; 659 *delim = 0; 660 np->local_ip = in_aton(cur); 661 cur = delim; 662 } 663 cur++; 664 665 if (*cur != ',') { 666 /* parse out dev name */ 667 if ((delim = strchr(cur, ',')) == NULL) 668 goto parse_failed; 669 *delim = 0; 670 strlcpy(np->dev_name, cur, sizeof(np->dev_name)); 671 cur = delim; 672 } 673 cur++; 674 675 if (*cur != '@') { 676 /* dst port */ 677 if ((delim = strchr(cur, '@')) == NULL) 678 goto parse_failed; 679 *delim = 0; 680 if (*cur == ' ' || *cur == '\t') 681 printk(KERN_INFO "%s: warning: whitespace" 682 "is not allowed\n", np->name); 683 np->remote_port = simple_strtol(cur, NULL, 10); 684 cur = delim; 685 } 686 cur++; 687 688 /* dst ip */ 689 if ((delim = strchr(cur, '/')) == NULL) 690 goto parse_failed; 691 *delim = 0; 692 np->remote_ip = in_aton(cur); 693 cur = delim + 1; 694 695 if (*cur != 0) { 696 /* MAC address */ 697 if (!mac_pton(cur, np->remote_mac)) 698 goto parse_failed; 699 } 700 701 netpoll_print_options(np); 702 703 return 0; 704 705 parse_failed: 706 printk(KERN_INFO "%s: couldn't parse config at '%s'!\n", 707 np->name, cur); 708 return -1; 709} 710EXPORT_SYMBOL(netpoll_parse_options); 711 712int __netpoll_setup(struct netpoll *np) 713{ 714 struct net_device *ndev = np->dev; 715 struct netpoll_info *npinfo; 716 const struct net_device_ops *ops; 717 unsigned long flags; 718 int err; 719 720 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || 721 !ndev->netdev_ops->ndo_poll_controller) { 722 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n", 723 np->name, np->dev_name); 724 err = -ENOTSUPP; 725 goto out; 726 } 727 728 if (!ndev->npinfo) { 729 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL); 730 if (!npinfo) { 731 err = -ENOMEM; 732 goto out; 733 } 734 735 npinfo->rx_flags = 0; 736 INIT_LIST_HEAD(&npinfo->rx_np); 737 738 spin_lock_init(&npinfo->rx_lock); 739 skb_queue_head_init(&npinfo->arp_tx); 740 skb_queue_head_init(&npinfo->txq); 741 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); 742 743 atomic_set(&npinfo->refcnt, 1); 744 745 ops = np->dev->netdev_ops; 746 if (ops->ndo_netpoll_setup) { 747 err = ops->ndo_netpoll_setup(ndev, npinfo); 748 if (err) 749 goto free_npinfo; 750 } 751 } else { 752 npinfo = ndev->npinfo; 753 atomic_inc(&npinfo->refcnt); 754 } 755 756 npinfo->netpoll = np; 757 758 if (np->rx_hook) { 759 spin_lock_irqsave(&npinfo->rx_lock, flags); 760 npinfo->rx_flags |= NETPOLL_RX_ENABLED; 761 list_add_tail(&np->rx, &npinfo->rx_np); 762 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 763 } 764 765 /* last thing to do is link it to the net device structure */ 766 RCU_INIT_POINTER(ndev->npinfo, npinfo); 767 768 return 0; 769 770free_npinfo: 771 kfree(npinfo); 772out: 773 return err; 774} 775EXPORT_SYMBOL_GPL(__netpoll_setup); 776 777int netpoll_setup(struct netpoll *np) 778{ 779 struct net_device *ndev = NULL; 780 struct in_device *in_dev; 781 int err; 782 783 if (np->dev_name) 784 ndev = dev_get_by_name(&init_net, np->dev_name); 785 if (!ndev) { 786 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n", 787 np->name, np->dev_name); 788 return -ENODEV; 789 } 790 791 if (ndev->master) { 792 printk(KERN_ERR "%s: %s is a slave device, aborting.\n", 793 np->name, np->dev_name); 794 err = -EBUSY; 795 goto put; 796 } 797 798 if (!netif_running(ndev)) { 799 unsigned long atmost, atleast; 800 801 printk(KERN_INFO "%s: device %s not up yet, forcing it\n", 802 np->name, np->dev_name); 803 804 rtnl_lock(); 805 err = dev_open(ndev); 806 rtnl_unlock(); 807 808 if (err) { 809 printk(KERN_ERR "%s: failed to open %s\n", 810 np->name, ndev->name); 811 goto put; 812 } 813 814 atleast = jiffies + HZ/10; 815 atmost = jiffies + carrier_timeout * HZ; 816 while (!netif_carrier_ok(ndev)) { 817 if (time_after(jiffies, atmost)) { 818 printk(KERN_NOTICE 819 "%s: timeout waiting for carrier\n", 820 np->name); 821 break; 822 } 823 msleep(1); 824 } 825 826 /* If carrier appears to come up instantly, we don't 827 * trust it and pause so that we don't pump all our 828 * queued console messages into the bitbucket. 829 */ 830 831 if (time_before(jiffies, atleast)) { 832 printk(KERN_NOTICE "%s: carrier detect appears" 833 " untrustworthy, waiting 4 seconds\n", 834 np->name); 835 msleep(4000); 836 } 837 } 838 839 if (!np->local_ip) { 840 rcu_read_lock(); 841 in_dev = __in_dev_get_rcu(ndev); 842 843 if (!in_dev || !in_dev->ifa_list) { 844 rcu_read_unlock(); 845 printk(KERN_ERR "%s: no IP address for %s, aborting\n", 846 np->name, np->dev_name); 847 err = -EDESTADDRREQ; 848 goto put; 849 } 850 851 np->local_ip = in_dev->ifa_list->ifa_local; 852 rcu_read_unlock(); 853 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip); 854 } 855 856 np->dev = ndev; 857 858 /* fill up the skb queue */ 859 refill_skbs(); 860 861 rtnl_lock(); 862 err = __netpoll_setup(np); 863 rtnl_unlock(); 864 865 if (err) 866 goto put; 867 868 return 0; 869 870put: 871 dev_put(ndev); 872 return err; 873} 874EXPORT_SYMBOL(netpoll_setup); 875 876static int __init netpoll_init(void) 877{ 878 skb_queue_head_init(&skb_pool); 879 return 0; 880} 881core_initcall(netpoll_init); 882 883void __netpoll_cleanup(struct netpoll *np) 884{ 885 struct netpoll_info *npinfo; 886 unsigned long flags; 887 888 npinfo = np->dev->npinfo; 889 if (!npinfo) 890 return; 891 892 if (!list_empty(&npinfo->rx_np)) { 893 spin_lock_irqsave(&npinfo->rx_lock, flags); 894 list_del(&np->rx); 895 if (list_empty(&npinfo->rx_np)) 896 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; 897 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 898 } 899 900 if (atomic_dec_and_test(&npinfo->refcnt)) { 901 const struct net_device_ops *ops; 902 903 ops = np->dev->netdev_ops; 904 if (ops->ndo_netpoll_cleanup) 905 ops->ndo_netpoll_cleanup(np->dev); 906 907 RCU_INIT_POINTER(np->dev->npinfo, NULL); 908 909 /* avoid racing with NAPI reading npinfo */ 910 synchronize_rcu_bh(); 911 912 skb_queue_purge(&npinfo->arp_tx); 913 skb_queue_purge(&npinfo->txq); 914 cancel_delayed_work_sync(&npinfo->tx_work); 915 916 /* clean after last, unfinished work */ 917 __skb_queue_purge(&npinfo->txq); 918 kfree(npinfo); 919 } 920} 921EXPORT_SYMBOL_GPL(__netpoll_cleanup); 922 923void netpoll_cleanup(struct netpoll *np) 924{ 925 if (!np->dev) 926 return; 927 928 rtnl_lock(); 929 __netpoll_cleanup(np); 930 rtnl_unlock(); 931 932 dev_put(np->dev); 933 np->dev = NULL; 934} 935EXPORT_SYMBOL(netpoll_cleanup); 936 937int netpoll_trap(void) 938{ 939 return atomic_read(&trapped); 940} 941EXPORT_SYMBOL(netpoll_trap); 942 943void netpoll_set_trap(int trap) 944{ 945 if (trap) 946 atomic_inc(&trapped); 947 else 948 atomic_dec(&trapped); 949} 950EXPORT_SYMBOL(netpoll_set_trap); 951