net/core/netpoll.c

   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 <net/tcp.h>
  26 #include <net/udp.h>
  27 #include <asm/unaligned.h>
  28 #include <trace/events/napi.h>
  29
  30 /*
  31  * We maintain a small pool of fully-sized skbs, to make sure the
  32  * message gets out even in extreme OOM situations.
  33  */
  34
  35 #define MAX_UDP_CHUNK 1460
  36 #define MAX_SKBS 32
  37 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
  38
  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_stopped(txq) ||
  79                     netif_tx_queue_frozen(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
  93 static __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  */
 128 static int poll_one_napi(struct netpoll_info *npinfo,
 129                          struct napi_struct *napi, int budget)
 130 {
 131         int work;
 132
 133         /* net_rx_action's ->poll() invocations and our's are
 134          * synchronized by this test which is only made while
 135          * holding the napi->poll_lock.
 136          */
 137         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 138                 return budget;
 139
 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
 154 static 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
 171 static 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
 181 void netpoll_poll(struct netpoll *np)
 182 {
 183         struct net_device *dev = np->dev;
 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
 203 static void refill_skbs(void)
 204 {
 205         struct sk_buff *skb;
 206         unsigned long flags;
 207
 208         spin_lock_irqsave(&skb_pool.lock, flags);
 209         while (skb_pool.qlen < MAX_SKBS) {
 210                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 211                 if (!skb)
 212                         break;
 213
 214                 __skb_queue_tail(&skb_pool, skb);
 215         }
 216         spin_unlock_irqrestore(&skb_pool.lock, flags);
 217 }
 218
 219 static void zap_completion_queue(void)
 220 {
 221         unsigned long flags;
 222         struct softnet_data *sd = &get_cpu_var(softnet_data);
 223
 224         if (sd->completion_queue) {
 225                 struct sk_buff *clist;
 226
 227                 local_irq_save(flags);
 228                 clist = sd->completion_queue;
 229                 sd->completion_queue = NULL;
 230                 local_irq_restore(flags);
 231
 232                 while (clist != NULL) {
 233                         struct sk_buff *skb = clist;
 234                         clist = clist->next;
 235                         if (skb->destructor) {
 236                                 atomic_inc(&skb->users);
 237                                 dev_kfree_skb_any(skb); /* put this one back */
 238                         } else {
 239                                 __kfree_skb(skb);
 240                         }
 241                 }
 242         }
 243
 244         put_cpu_var(softnet_data);
 245 }
 246
 247 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 248 {
 249         int count = 0;
 250         struct sk_buff *skb;
 251
 252         zap_completion_queue();
 253         refill_skbs();
 254 repeat:
 255
 256         skb = alloc_skb(len, GFP_ATOMIC);
 257         if (!skb)
 258                 skb = skb_dequeue(&skb_pool);
 259
 260         if (!skb) {
 261                 if (++count < 10) {
 262                         netpoll_poll(np);
 263                         goto repeat;
 264                 }
 265                 return NULL;
 266         }
 267
 268         atomic_set(&skb->users, 1);
 269         skb_reserve(skb, reserve);
 270         return skb;
 271 }
 272
 273 static int netpoll_owner_active(struct net_device *dev)
 274 {
 275         struct napi_struct *napi;
 276
 277         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 278                 if (napi->poll_owner == smp_processor_id())
 279                         return 1;
 280         }
 281         return 0;
 282 }
 283
 284 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 285 {
 286         int status = NETDEV_TX_BUSY;
 287         unsigned long tries;
 288         struct net_device *dev = np->dev;
 289         const struct net_device_ops *ops = dev->netdev_ops;
 290         struct netpoll_info *npinfo = np->dev->npinfo;
 291
 292         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 293                 __kfree_skb(skb);
 294                 return;
 295         }
 296
 297         /* don't get messages out of order, and no recursion */
 298         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 299                 struct netdev_queue *txq;
 300                 unsigned long flags;
 301
 302                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 303
 304                 local_irq_save(flags);
 305                 /* try until next clock tick */
 306                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 307                      tries > 0; --tries) {
 308                         if (__netif_tx_trylock(txq)) {
 309                                 if (!netif_tx_queue_stopped(txq)) {
 310                                         status = ops->ndo_start_xmit(skb, dev);
 311                                         if (status == NETDEV_TX_OK)
 312                                                 txq_trans_update(txq);
 313                                 }
 314                                 __netif_tx_unlock(txq);
 315
 316                                 if (status == NETDEV_TX_OK)
 317                                         break;
 318
 319                         }
 320
 321                         /* tickle device maybe there is some cleanup */
 322                         netpoll_poll(np);
 323
 324                         udelay(USEC_PER_POLL);
 325                 }
 326                 local_irq_restore(flags);
 327         }
 328
 329         if (status != NETDEV_TX_OK) {
 330                 skb_queue_tail(&npinfo->txq, skb);
 331                 schedule_delayed_work(&npinfo->tx_work,0);
 332         }
 333 }
 334
 335 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 336 {
 337         int total_len, eth_len, ip_len, udp_len;
 338         struct sk_buff *skb;
 339         struct udphdr *udph;
 340         struct iphdr *iph;
 341         struct ethhdr *eth;
 342
 343         udp_len = len + sizeof(*udph);
 344         ip_len = eth_len = udp_len + sizeof(*iph);
 345         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 346
 347         skb = find_skb(np, total_len, total_len - len);
 348         if (!skb)
 349                 return;
 350
 351         skb_copy_to_linear_data(skb, msg, len);
 352         skb->len += len;
 353
 354         skb_push(skb, sizeof(*udph));
 355         skb_reset_transport_header(skb);
 356         udph = udp_hdr(skb);
 357         udph->source = htons(np->local_port);
 358         udph->dest = htons(np->remote_port);
 359         udph->len = htons(udp_len);
 360         udph->check = 0;
 361         udph->check = csum_tcpudp_magic(np->local_ip,
 362                                         np->remote_ip,
 363                                         udp_len, IPPROTO_UDP,
 364                                         csum_partial(udph, udp_len, 0));
 365         if (udph->check == 0)
 366                 udph->check = CSUM_MANGLED_0;
 367
 368         skb_push(skb, sizeof(*iph));
 369         skb_reset_network_header(skb);
 370         iph = ip_hdr(skb);
 371
 372         /* iph->version = 4; iph->ihl = 5; */
 373         put_unaligned(0x45, (unsigned char *)iph);
 374         iph->tos      = 0;
 375         put_unaligned(htons(ip_len), &(iph->tot_len));
 376         iph->id       = 0;
 377         iph->frag_off = 0;
 378         iph->ttl      = 64;
 379         iph->protocol = IPPROTO_UDP;
 380         iph->check    = 0;
 381         put_unaligned(np->local_ip, &(iph->saddr));
 382         put_unaligned(np->remote_ip, &(iph->daddr));
 383         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 384
 385         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 386         skb_reset_mac_header(skb);
 387         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 388         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 389         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 390
 391         skb->dev = np->dev;
 392
 393         netpoll_send_skb(np, skb);
 394 }
 395
 396 static void arp_reply(struct sk_buff *skb)
 397 {
 398         struct netpoll_info *npinfo = skb->dev->npinfo;
 399         struct arphdr *arp;
 400         unsigned char *arp_ptr;
 401         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 402         __be32 sip, tip;
 403         unsigned char *sha;
 404         struct sk_buff *send_skb;
 405         struct netpoll *np = NULL;
 406
 407         if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
 408                 np = npinfo->rx_np;
 409         if (!np)
 410                 return;
 411
 412         /* No arp on this interface */
 413         if (skb->dev->flags & IFF_NOARP)
 414                 return;
 415
 416         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 417                 return;
 418
 419         skb_reset_network_header(skb);
 420         skb_reset_transport_header(skb);
 421         arp = arp_hdr(skb);
 422
 423         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 424              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 425             arp->ar_pro != htons(ETH_P_IP) ||
 426             arp->ar_op != htons(ARPOP_REQUEST))
 427                 return;
 428
 429         arp_ptr = (unsigned char *)(arp+1);
 430         /* save the location of the src hw addr */
 431         sha = arp_ptr;
 432         arp_ptr += skb->dev->addr_len;
 433         memcpy(&sip, arp_ptr, 4);
 434         arp_ptr += 4;
 435         /* if we actually cared about dst hw addr, it would get copied here */
 436         arp_ptr += skb->dev->addr_len;
 437         memcpy(&tip, arp_ptr, 4);
 438
 439         /* Should we ignore arp? */
 440         if (tip != np->local_ip ||
 441             ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 442                 return;
 443
 444         size = arp_hdr_len(skb->dev);
 445         send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
 446                             LL_RESERVED_SPACE(np->dev));
 447
 448         if (!send_skb)
 449                 return;
 450
 451         skb_reset_network_header(send_skb);
 452         arp = (struct arphdr *) skb_put(send_skb, size);
 453         send_skb->dev = skb->dev;
 454         send_skb->protocol = htons(ETH_P_ARP);
 455
 456         /* Fill the device header for the ARP frame */
 457         if (dev_hard_header(send_skb, skb->dev, ptype,
 458                             sha, np->dev->dev_addr,
 459                             send_skb->len) < 0) {
 460                 kfree_skb(send_skb);
 461                 return;
 462         }
 463
 464         /*
 465          * Fill out the arp protocol part.
 466          *
 467          * we only support ethernet device type,
 468          * which (according to RFC 1390) should always equal 1 (Ethernet).
 469          */
 470
 471         arp->ar_hrd = htons(np->dev->type);
 472         arp->ar_pro = htons(ETH_P_IP);
 473         arp->ar_hln = np->dev->addr_len;
 474         arp->ar_pln = 4;
 475         arp->ar_op = htons(type);
 476
 477         arp_ptr=(unsigned char *)(arp + 1);
 478         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 479         arp_ptr += np->dev->addr_len;
 480         memcpy(arp_ptr, &tip, 4);
 481         arp_ptr += 4;
 482         memcpy(arp_ptr, sha, np->dev->addr_len);
 483         arp_ptr += np->dev->addr_len;
 484         memcpy(arp_ptr, &sip, 4);
 485
 486         netpoll_send_skb(np, send_skb);
 487 }
 488
 489 int __netpoll_rx(struct sk_buff *skb)
 490 {
 491         int proto, len, ulen;
 492         struct iphdr *iph;
 493         struct udphdr *uh;
 494         struct netpoll_info *npi = skb->dev->npinfo;
 495         struct netpoll *np = npi->rx_np;
 496
 497         if (!np)
 498                 goto out;
 499         if (skb->dev->type != ARPHRD_ETHER)
 500                 goto out;
 501
 502         /* check if netpoll clients need ARP */
 503         if (skb->protocol == htons(ETH_P_ARP) &&
 504             atomic_read(&trapped)) {
 505                 skb_queue_tail(&npi->arp_tx, skb);
 506                 return 1;
 507         }
 508
 509         proto = ntohs(eth_hdr(skb)->h_proto);
 510         if (proto != ETH_P_IP)
 511                 goto out;
 512         if (skb->pkt_type == PACKET_OTHERHOST)
 513                 goto out;
 514         if (skb_shared(skb))
 515                 goto out;
 516
 517         iph = (struct iphdr *)skb->data;
 518         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 519                 goto out;
 520         if (iph->ihl < 5 || iph->version != 4)
 521                 goto out;
 522         if (!pskb_may_pull(skb, iph->ihl*4))
 523                 goto out;
 524         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 525                 goto out;
 526
 527         len = ntohs(iph->tot_len);
 528         if (skb->len < len || len < iph->ihl*4)
 529                 goto out;
 530
 531         /*
 532          * Our transport medium may have padded the buffer out.
 533          * Now We trim to the true length of the frame.
 534          */
 535         if (pskb_trim_rcsum(skb, len))
 536                 goto out;
 537
 538         if (iph->protocol != IPPROTO_UDP)
 539                 goto out;
 540
 541         len -= iph->ihl*4;
 542         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 543         ulen = ntohs(uh->len);
 544
 545         if (ulen != len)
 546                 goto out;
 547         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 548                 goto out;
 549         if (np->local_ip && np->local_ip != iph->daddr)
 550                 goto out;
 551         if (np->remote_ip && np->remote_ip != iph->saddr)
 552                 goto out;
 553         if (np->local_port && np->local_port != ntohs(uh->dest))
 554                 goto out;
 555
 556         np->rx_hook(np, ntohs(uh->source),
 557                     (char *)(uh+1),
 558                     ulen - sizeof(struct udphdr));
 559
 560         kfree_skb(skb);
 561         return 1;
 562
 563 out:
 564         if (atomic_read(&trapped)) {
 565                 kfree_skb(skb);
 566                 return 1;
 567         }
 568
 569         return 0;
 570 }
 571
 572 void netpoll_print_options(struct netpoll *np)
 573 {
 574         printk(KERN_INFO "%s: local port %d\n",
 575                          np->name, np->local_port);
 576         printk(KERN_INFO "%s: local IP %pI4\n",
 577                          np->name, &np->local_ip);
 578         printk(KERN_INFO "%s: interface %s\n",
 579                          np->name, np->dev_name);
 580         printk(KERN_INFO "%s: remote port %d\n",
 581                          np->name, np->remote_port);
 582         printk(KERN_INFO "%s: remote IP %pI4\n",
 583                          np->name, &np->remote_ip);
 584         printk(KERN_INFO "%s: remote ethernet address %pM\n",
 585                          np->name, np->remote_mac);
 586 }
 587
 588 int netpoll_parse_options(struct netpoll *np, char *opt)
 589 {
 590         char *cur=opt, *delim;
 591
 592         if (*cur != '@') {
 593                 if ((delim = strchr(cur, '@')) == NULL)
 594                         goto parse_failed;
 595                 *delim = 0;
 596                 np->local_port = simple_strtol(cur, NULL, 10);
 597                 cur = delim;
 598         }
 599         cur++;
 600
 601         if (*cur != '/') {
 602                 if ((delim = strchr(cur, '/')) == NULL)
 603                         goto parse_failed;
 604                 *delim = 0;
 605                 np->local_ip = in_aton(cur);
 606                 cur = delim;
 607         }
 608         cur++;
 609
 610         if (*cur != ',') {
 611                 /* parse out dev name */
 612                 if ((delim = strchr(cur, ',')) == NULL)
 613                         goto parse_failed;
 614                 *delim = 0;
 615                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 616                 cur = delim;
 617         }
 618         cur++;
 619
 620         if (*cur != '@') {
 621                 /* dst port */
 622                 if ((delim = strchr(cur, '@')) == NULL)
 623                         goto parse_failed;
 624                 *delim = 0;
 625                 np->remote_port = simple_strtol(cur, NULL, 10);
 626                 cur = delim;
 627         }
 628         cur++;
 629
 630         /* dst ip */
 631         if ((delim = strchr(cur, '/')) == NULL)
 632                 goto parse_failed;
 633         *delim = 0;
 634         np->remote_ip = in_aton(cur);
 635         cur = delim + 1;
 636
 637         if (*cur != 0) {
 638                 /* MAC address */
 639                 if ((delim = strchr(cur, ':')) == NULL)
 640                         goto parse_failed;
 641                 *delim = 0;
 642                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
 643                 cur = delim + 1;
 644                 if ((delim = strchr(cur, ':')) == NULL)
 645                         goto parse_failed;
 646                 *delim = 0;
 647                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
 648                 cur = delim + 1;
 649                 if ((delim = strchr(cur, ':')) == NULL)
 650                         goto parse_failed;
 651                 *delim = 0;
 652                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
 653                 cur = delim + 1;
 654                 if ((delim = strchr(cur, ':')) == NULL)
 655                         goto parse_failed;
 656                 *delim = 0;
 657                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
 658                 cur = delim + 1;
 659                 if ((delim = strchr(cur, ':')) == NULL)
 660                         goto parse_failed;
 661                 *delim = 0;
 662                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 663                 cur = delim + 1;
 664                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 665         }
 666
 667         netpoll_print_options(np);
 668
 669         return 0;
 670
 671  parse_failed:
 672         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 673                np->name, cur);
 674         return -1;
 675 }
 676
 677 int netpoll_setup(struct netpoll *np)
 678 {
 679         struct net_device *ndev = NULL;
 680         struct in_device *in_dev;
 681         struct netpoll_info *npinfo;
 682         unsigned long flags;
 683         int err;
 684
 685         if (np->dev_name)
 686                 ndev = dev_get_by_name(&init_net, np->dev_name);
 687         if (!ndev) {
 688                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 689                        np->name, np->dev_name);
 690                 return -ENODEV;
 691         }
 692
 693         np->dev = ndev;
 694         if (!ndev->npinfo) {
 695                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 696                 if (!npinfo) {
 697                         err = -ENOMEM;
 698                         goto release;
 699                 }
 700
 701                 npinfo->rx_flags = 0;
 702                 npinfo->rx_np = NULL;
 703
 704                 spin_lock_init(&npinfo->rx_lock);
 705                 skb_queue_head_init(&npinfo->arp_tx);
 706                 skb_queue_head_init(&npinfo->txq);
 707                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 708
 709                 atomic_set(&npinfo->refcnt, 1);
 710         } else {
 711                 npinfo = ndev->npinfo;
 712                 atomic_inc(&npinfo->refcnt);
 713         }
 714
 715         if (!ndev->netdev_ops->ndo_poll_controller) {
 716                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 717                        np->name, np->dev_name);
 718                 err = -ENOTSUPP;
 719                 goto release;
 720         }
 721
 722         if (!netif_running(ndev)) {
 723                 unsigned long atmost, atleast;
 724
 725                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 726                        np->name, np->dev_name);
 727
 728                 rtnl_lock();
 729                 err = dev_open(ndev);
 730                 rtnl_unlock();
 731
 732                 if (err) {
 733                         printk(KERN_ERR "%s: failed to open %s\n",
 734                                np->name, ndev->name);
 735                         goto release;
 736                 }
 737
 738                 atleast = jiffies + HZ/10;
 739                 atmost = jiffies + carrier_timeout * HZ;
 740                 while (!netif_carrier_ok(ndev)) {
 741                         if (time_after(jiffies, atmost)) {
 742                                 printk(KERN_NOTICE
 743                                        "%s: timeout waiting for carrier\n",
 744                                        np->name);
 745                                 break;
 746                         }
 747                         msleep(1);
 748                 }
 749
 750                 /* If carrier appears to come up instantly, we don't
 751                  * trust it and pause so that we don't pump all our
 752                  * queued console messages into the bitbucket.
 753                  */
 754
 755                 if (time_before(jiffies, atleast)) {
 756                         printk(KERN_NOTICE "%s: carrier detect appears"
 757                                " untrustworthy, waiting 4 seconds\n",
 758                                np->name);
 759                         msleep(4000);
 760                 }
 761         }
 762
 763         if (!np->local_ip) {
 764                 rcu_read_lock();
 765                 in_dev = __in_dev_get_rcu(ndev);
 766
 767                 if (!in_dev || !in_dev->ifa_list) {
 768                         rcu_read_unlock();
 769                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 770                                np->name, np->dev_name);
 771                         err = -EDESTADDRREQ;
 772                         goto release;
 773                 }
 774
 775                 np->local_ip = in_dev->ifa_list->ifa_local;
 776                 rcu_read_unlock();
 777                 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
 778         }
 779
 780         if (np->rx_hook) {
 781                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 782                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 783                 npinfo->rx_np = np;
 784                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 785         }
 786
 787         /* fill up the skb queue */
 788         refill_skbs();
 789
 790         /* last thing to do is link it to the net device structure */
 791         ndev->npinfo = npinfo;
 792
 793         /* avoid racing with NAPI reading npinfo */
 794         synchronize_rcu();
 795
 796         return 0;
 797
 798  release:
 799         if (!ndev->npinfo)
 800                 kfree(npinfo);
 801         np->dev = NULL;
 802         dev_put(ndev);
 803         return err;
 804 }
 805
 806 static int __init netpoll_init(void)
 807 {
 808         skb_queue_head_init(&skb_pool);
 809         return 0;
 810 }
 811 core_initcall(netpoll_init);
 812
 813 void netpoll_cleanup(struct netpoll *np)
 814 {
 815         struct netpoll_info *npinfo;
 816         unsigned long flags;
 817
 818         if (np->dev) {
 819                 npinfo = np->dev->npinfo;
 820                 if (npinfo) {
 821                         if (npinfo->rx_np == np) {
 822                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 823                                 npinfo->rx_np = NULL;
 824                                 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 825                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 826                         }
 827
 828                         if (atomic_dec_and_test(&npinfo->refcnt)) {
 829                                 skb_queue_purge(&npinfo->arp_tx);
 830                                 skb_queue_purge(&npinfo->txq);
 831                                 cancel_rearming_delayed_work(&npinfo->tx_work);
 832
 833                                 /* clean after last, unfinished work */
 834                                 __skb_queue_purge(&npinfo->txq);
 835                                 kfree(npinfo);
 836                                 np->dev->npinfo = NULL;
 837                         }
 838                 }
 839
 840                 dev_put(np->dev);
 841         }
 842
 843         np->dev = NULL;
 844 }
 845
 846 int netpoll_trap(void)
 847 {
 848         return atomic_read(&trapped);
 849 }
 850
 851 void netpoll_set_trap(int trap)
 852 {
 853         if (trap)
 854                 atomic_inc(&trapped);
 855         else
 856                 atomic_dec(&trapped);
 857 }
 858
 859 EXPORT_SYMBOL(netpoll_set_trap);
 860 EXPORT_SYMBOL(netpoll_trap);
 861 EXPORT_SYMBOL(netpoll_print_options);
 862 EXPORT_SYMBOL(netpoll_parse_options);
 863 EXPORT_SYMBOL(netpoll_setup);
 864 EXPORT_SYMBOL(netpoll_cleanup);
 865 EXPORT_SYMBOL(netpoll_send_udp);
 866 EXPORT_SYMBOL(netpoll_poll);