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/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/string.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/inetdevice.h>
  17 #include <linux/inet.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/netpoll.h>
  20 #include <linux/sched.h>
  21 #include <linux/delay.h>
  22 #include <linux/rcupdate.h>
  23 #include <linux/workqueue.h>
  24 #include <net/tcp.h>
  25 #include <net/udp.h>
  26 #include <asm/unaligned.h>
  27
  28 /*
  29  * We maintain a small pool of fully-sized skbs, to make sure the
  30  * message gets out even in extreme OOM situations.
  31  */
  32
  33 #define MAX_UDP_CHUNK 1460
  34 #define MAX_SKBS 32
  35 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
  36
  37 static struct sk_buff_head skb_pool;
  38
  39 static atomic_t trapped;
  40
  41 #define USEC_PER_POLL   50
  42 #define NETPOLL_RX_ENABLED  1
  43 #define NETPOLL_RX_DROP     2
  44
  45 #define MAX_SKB_SIZE \
  46                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
  47                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
  48
  49 static void zap_completion_queue(void);
  50 static void arp_reply(struct sk_buff *skb);
  51
  52 static void queue_process(struct work_struct *work)
  53 {
  54         struct netpoll_info *npinfo =
  55                 container_of(work, struct netpoll_info, tx_work.work);
  56         struct sk_buff *skb;
  57         unsigned long flags;
  58
  59         while ((skb = skb_dequeue(&npinfo->txq))) {
  60                 struct net_device *dev = skb->dev;
  61
  62                 if (!netif_device_present(dev) || !netif_running(dev)) {
  63                         __kfree_skb(skb);
  64                         continue;
  65                 }
  66
  67                 local_irq_save(flags);
  68                 netif_tx_lock(dev);
  69                 if ((netif_queue_stopped(dev) ||
  70                      netif_subqueue_stopped(dev, skb->queue_mapping)) ||
  71                      dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
  72                         skb_queue_head(&npinfo->txq, skb);
  73                         netif_tx_unlock(dev);
  74                         local_irq_restore(flags);
  75
  76                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  77                         return;
  78                 }
  79                 netif_tx_unlock(dev);
  80                 local_irq_restore(flags);
  81         }
  82 }
  83
  84 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  85                             unsigned short ulen, __be32 saddr, __be32 daddr)
  86 {
  87         __wsum psum;
  88
  89         if (uh->check == 0 || skb_csum_unnecessary(skb))
  90                 return 0;
  91
  92         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
  93
  94         if (skb->ip_summed == CHECKSUM_COMPLETE &&
  95             !csum_fold(csum_add(psum, skb->csum)))
  96                 return 0;
  97
  98         skb->csum = psum;
  99
 100         return __skb_checksum_complete(skb);
 101 }
 102
 103 /*
 104  * Check whether delayed processing was scheduled for our NIC. If so,
 105  * we attempt to grab the poll lock and use ->poll() to pump the card.
 106  * If this fails, either we've recursed in ->poll() or it's already
 107  * running on another CPU.
 108  *
 109  * Note: we don't mask interrupts with this lock because we're using
 110  * trylock here and interrupts are already disabled in the softirq
 111  * case. Further, we test the poll_owner to avoid recursion on UP
 112  * systems where the lock doesn't exist.
 113  *
 114  * In cases where there is bi-directional communications, reading only
 115  * one message at a time can lead to packets being dropped by the
 116  * network adapter, forcing superfluous retries and possibly timeouts.
 117  * Thus, we set our budget to greater than 1.
 118  */
 119 static void poll_napi(struct netpoll *np)
 120 {
 121         struct netpoll_info *npinfo = np->dev->npinfo;
 122         struct napi_struct *napi;
 123         int budget = 16;
 124
 125         list_for_each_entry(napi, &np->dev->napi_list, dev_list) {
 126                 if (test_bit(NAPI_STATE_SCHED, &napi->state) &&
 127                     napi->poll_owner != smp_processor_id() &&
 128                     spin_trylock(&napi->poll_lock)) {
 129                         npinfo->rx_flags |= NETPOLL_RX_DROP;
 130                         atomic_inc(&trapped);
 131
 132                         napi->poll(napi, budget);
 133
 134                         atomic_dec(&trapped);
 135                         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 136                         spin_unlock(&napi->poll_lock);
 137                 }
 138         }
 139 }
 140
 141 static void service_arp_queue(struct netpoll_info *npi)
 142 {
 143         struct sk_buff *skb;
 144
 145         if (unlikely(!npi))
 146                 return;
 147
 148         skb = skb_dequeue(&npi->arp_tx);
 149
 150         while (skb != NULL) {
 151                 arp_reply(skb);
 152                 skb = skb_dequeue(&npi->arp_tx);
 153         }
 154 }
 155
 156 void netpoll_poll(struct netpoll *np)
 157 {
 158         if (!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
 159                 return;
 160
 161         /* Process pending work on NIC */
 162         np->dev->poll_controller(np->dev);
 163         if (!list_empty(&np->dev->napi_list))
 164                 poll_napi(np);
 165
 166         service_arp_queue(np->dev->npinfo);
 167
 168         zap_completion_queue();
 169 }
 170
 171 static void refill_skbs(void)
 172 {
 173         struct sk_buff *skb;
 174         unsigned long flags;
 175
 176         spin_lock_irqsave(&skb_pool.lock, flags);
 177         while (skb_pool.qlen < MAX_SKBS) {
 178                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 179                 if (!skb)
 180                         break;
 181
 182                 __skb_queue_tail(&skb_pool, skb);
 183         }
 184         spin_unlock_irqrestore(&skb_pool.lock, flags);
 185 }
 186
 187 static void zap_completion_queue(void)
 188 {
 189         unsigned long flags;
 190         struct softnet_data *sd = &get_cpu_var(softnet_data);
 191
 192         if (sd->completion_queue) {
 193                 struct sk_buff *clist;
 194
 195                 local_irq_save(flags);
 196                 clist = sd->completion_queue;
 197                 sd->completion_queue = NULL;
 198                 local_irq_restore(flags);
 199
 200                 while (clist != NULL) {
 201                         struct sk_buff *skb = clist;
 202                         clist = clist->next;
 203                         if (skb->destructor)
 204                                 dev_kfree_skb_any(skb); /* put this one back */
 205                         else
 206                                 __kfree_skb(skb);
 207                 }
 208         }
 209
 210         put_cpu_var(softnet_data);
 211 }
 212
 213 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 214 {
 215         int count = 0;
 216         struct sk_buff *skb;
 217
 218         zap_completion_queue();
 219         refill_skbs();
 220 repeat:
 221
 222         skb = alloc_skb(len, GFP_ATOMIC);
 223         if (!skb)
 224                 skb = skb_dequeue(&skb_pool);
 225
 226         if (!skb) {
 227                 if (++count < 10) {
 228                         netpoll_poll(np);
 229                         goto repeat;
 230                 }
 231                 return NULL;
 232         }
 233
 234         atomic_set(&skb->users, 1);
 235         skb_reserve(skb, reserve);
 236         return skb;
 237 }
 238
 239 static int netpoll_owner_active(struct net_device *dev)
 240 {
 241         struct napi_struct *napi;
 242
 243         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 244                 if (napi->poll_owner == smp_processor_id())
 245                         return 1;
 246         }
 247         return 0;
 248 }
 249
 250 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 251 {
 252         int status = NETDEV_TX_BUSY;
 253         unsigned long tries;
 254         struct net_device *dev = np->dev;
 255         struct netpoll_info *npinfo = np->dev->npinfo;
 256
 257         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 258                 __kfree_skb(skb);
 259                 return;
 260         }
 261
 262         /* don't get messages out of order, and no recursion */
 263         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 264                 unsigned long flags;
 265
 266                 local_irq_save(flags);
 267                 /* try until next clock tick */
 268                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 269                      tries > 0; --tries) {
 270                         if (netif_tx_trylock(dev)) {
 271                                 if (!netif_queue_stopped(dev) &&
 272                                     !netif_subqueue_stopped(dev, skb->queue_mapping))
 273                                         status = dev->hard_start_xmit(skb, dev);
 274                                 netif_tx_unlock(dev);
 275
 276                                 if (status == NETDEV_TX_OK)
 277                                         break;
 278
 279                         }
 280
 281                         /* tickle device maybe there is some cleanup */
 282                         netpoll_poll(np);
 283
 284                         udelay(USEC_PER_POLL);
 285                 }
 286                 local_irq_restore(flags);
 287         }
 288
 289         if (status != NETDEV_TX_OK) {
 290                 skb_queue_tail(&npinfo->txq, skb);
 291                 schedule_delayed_work(&npinfo->tx_work,0);
 292         }
 293 }
 294
 295 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 296 {
 297         int total_len, eth_len, ip_len, udp_len;
 298         struct sk_buff *skb;
 299         struct udphdr *udph;
 300         struct iphdr *iph;
 301         struct ethhdr *eth;
 302
 303         udp_len = len + sizeof(*udph);
 304         ip_len = eth_len = udp_len + sizeof(*iph);
 305         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 306
 307         skb = find_skb(np, total_len, total_len - len);
 308         if (!skb)
 309                 return;
 310
 311         skb_copy_to_linear_data(skb, msg, len);
 312         skb->len += len;
 313
 314         skb_push(skb, sizeof(*udph));
 315         skb_reset_transport_header(skb);
 316         udph = udp_hdr(skb);
 317         udph->source = htons(np->local_port);
 318         udph->dest = htons(np->remote_port);
 319         udph->len = htons(udp_len);
 320         udph->check = 0;
 321         udph->check = csum_tcpudp_magic(htonl(np->local_ip),
 322                                         htonl(np->remote_ip),
 323                                         udp_len, IPPROTO_UDP,
 324                                         csum_partial((unsigned char *)udph, udp_len, 0));
 325         if (udph->check == 0)
 326                 udph->check = CSUM_MANGLED_0;
 327
 328         skb_push(skb, sizeof(*iph));
 329         skb_reset_network_header(skb);
 330         iph = ip_hdr(skb);
 331
 332         /* iph->version = 4; iph->ihl = 5; */
 333         put_unaligned(0x45, (unsigned char *)iph);
 334         iph->tos      = 0;
 335         put_unaligned(htons(ip_len), &(iph->tot_len));
 336         iph->id       = 0;
 337         iph->frag_off = 0;
 338         iph->ttl      = 64;
 339         iph->protocol = IPPROTO_UDP;
 340         iph->check    = 0;
 341         put_unaligned(htonl(np->local_ip), &(iph->saddr));
 342         put_unaligned(htonl(np->remote_ip), &(iph->daddr));
 343         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 344
 345         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 346         skb_reset_mac_header(skb);
 347         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 348         memcpy(eth->h_source, np->local_mac, 6);
 349         memcpy(eth->h_dest, np->remote_mac, 6);
 350
 351         skb->dev = np->dev;
 352
 353         netpoll_send_skb(np, skb);
 354 }
 355
 356 static void arp_reply(struct sk_buff *skb)
 357 {
 358         struct netpoll_info *npinfo = skb->dev->npinfo;
 359         struct arphdr *arp;
 360         unsigned char *arp_ptr;
 361         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 362         __be32 sip, tip;
 363         unsigned char *sha;
 364         struct sk_buff *send_skb;
 365         struct netpoll *np = NULL;
 366
 367         if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
 368                 np = npinfo->rx_np;
 369         if (!np)
 370                 return;
 371
 372         /* No arp on this interface */
 373         if (skb->dev->flags & IFF_NOARP)
 374                 return;
 375
 376         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
 377                                  (2 * skb->dev->addr_len) +
 378                                  (2 * sizeof(u32)))))
 379                 return;
 380
 381         skb_reset_network_header(skb);
 382         skb_reset_transport_header(skb);
 383         arp = arp_hdr(skb);
 384
 385         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 386              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 387             arp->ar_pro != htons(ETH_P_IP) ||
 388             arp->ar_op != htons(ARPOP_REQUEST))
 389                 return;
 390
 391         arp_ptr = (unsigned char *)(arp+1);
 392         /* save the location of the src hw addr */
 393         sha = arp_ptr;
 394         arp_ptr += skb->dev->addr_len;
 395         memcpy(&sip, arp_ptr, 4);
 396         arp_ptr += 4;
 397         /* if we actually cared about dst hw addr, it would get copied here */
 398         arp_ptr += skb->dev->addr_len;
 399         memcpy(&tip, arp_ptr, 4);
 400
 401         /* Should we ignore arp? */
 402         if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
 403                 return;
 404
 405         size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
 406         send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
 407                             LL_RESERVED_SPACE(np->dev));
 408
 409         if (!send_skb)
 410                 return;
 411
 412         skb_reset_network_header(send_skb);
 413         arp = (struct arphdr *) skb_put(send_skb, size);
 414         send_skb->dev = skb->dev;
 415         send_skb->protocol = htons(ETH_P_ARP);
 416
 417         /* Fill the device header for the ARP frame */
 418         if (dev_hard_header(send_skb, skb->dev, ptype,
 419                             sha, np->local_mac,
 420                             send_skb->len) < 0) {
 421                 kfree_skb(send_skb);
 422                 return;
 423         }
 424
 425         /*
 426          * Fill out the arp protocol part.
 427          *
 428          * we only support ethernet device type,
 429          * which (according to RFC 1390) should always equal 1 (Ethernet).
 430          */
 431
 432         arp->ar_hrd = htons(np->dev->type);
 433         arp->ar_pro = htons(ETH_P_IP);
 434         arp->ar_hln = np->dev->addr_len;
 435         arp->ar_pln = 4;
 436         arp->ar_op = htons(type);
 437
 438         arp_ptr=(unsigned char *)(arp + 1);
 439         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 440         arp_ptr += np->dev->addr_len;
 441         memcpy(arp_ptr, &tip, 4);
 442         arp_ptr += 4;
 443         memcpy(arp_ptr, sha, np->dev->addr_len);
 444         arp_ptr += np->dev->addr_len;
 445         memcpy(arp_ptr, &sip, 4);
 446
 447         netpoll_send_skb(np, send_skb);
 448 }
 449
 450 int __netpoll_rx(struct sk_buff *skb)
 451 {
 452         int proto, len, ulen;
 453         struct iphdr *iph;
 454         struct udphdr *uh;
 455         struct netpoll_info *npi = skb->dev->npinfo;
 456         struct netpoll *np = npi->rx_np;
 457
 458         if (!np)
 459                 goto out;
 460         if (skb->dev->type != ARPHRD_ETHER)
 461                 goto out;
 462
 463         /* check if netpoll clients need ARP */
 464         if (skb->protocol == htons(ETH_P_ARP) &&
 465             atomic_read(&trapped)) {
 466                 skb_queue_tail(&npi->arp_tx, skb);
 467                 return 1;
 468         }
 469
 470         proto = ntohs(eth_hdr(skb)->h_proto);
 471         if (proto != ETH_P_IP)
 472                 goto out;
 473         if (skb->pkt_type == PACKET_OTHERHOST)
 474                 goto out;
 475         if (skb_shared(skb))
 476                 goto out;
 477
 478         iph = (struct iphdr *)skb->data;
 479         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 480                 goto out;
 481         if (iph->ihl < 5 || iph->version != 4)
 482                 goto out;
 483         if (!pskb_may_pull(skb, iph->ihl*4))
 484                 goto out;
 485         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 486                 goto out;
 487
 488         len = ntohs(iph->tot_len);
 489         if (skb->len < len || len < iph->ihl*4)
 490                 goto out;
 491
 492         /*
 493          * Our transport medium may have padded the buffer out.
 494          * Now We trim to the true length of the frame.
 495          */
 496         if (pskb_trim_rcsum(skb, len))
 497                 goto out;
 498
 499         if (iph->protocol != IPPROTO_UDP)
 500                 goto out;
 501
 502         len -= iph->ihl*4;
 503         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 504         ulen = ntohs(uh->len);
 505
 506         if (ulen != len)
 507                 goto out;
 508         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 509                 goto out;
 510         if (np->local_ip && np->local_ip != ntohl(iph->daddr))
 511                 goto out;
 512         if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
 513                 goto out;
 514         if (np->local_port && np->local_port != ntohs(uh->dest))
 515                 goto out;
 516
 517         np->rx_hook(np, ntohs(uh->source),
 518                     (char *)(uh+1),
 519                     ulen - sizeof(struct udphdr));
 520
 521         kfree_skb(skb);
 522         return 1;
 523
 524 out:
 525         if (atomic_read(&trapped)) {
 526                 kfree_skb(skb);
 527                 return 1;
 528         }
 529
 530         return 0;
 531 }
 532
 533 void netpoll_print_options(struct netpoll *np)
 534 {
 535         DECLARE_MAC_BUF(mac);
 536         printk(KERN_INFO "%s: local port %d\n",
 537                          np->name, np->local_port);
 538         printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 539                          np->name, HIPQUAD(np->local_ip));
 540         printk(KERN_INFO "%s: interface %s\n",
 541                          np->name, np->dev_name);
 542         printk(KERN_INFO "%s: remote port %d\n",
 543                          np->name, np->remote_port);
 544         printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
 545                          np->name, HIPQUAD(np->remote_ip));
 546         printk(KERN_INFO "%s: remote ethernet address %s\n",
 547                          np->name, print_mac(mac, np->remote_mac));
 548 }
 549
 550 int netpoll_parse_options(struct netpoll *np, char *opt)
 551 {
 552         char *cur=opt, *delim;
 553
 554         if (*cur != '@') {
 555                 if ((delim = strchr(cur, '@')) == NULL)
 556                         goto parse_failed;
 557                 *delim = 0;
 558                 np->local_port = simple_strtol(cur, NULL, 10);
 559                 cur = delim;
 560         }
 561         cur++;
 562
 563         if (*cur != '/') {
 564                 if ((delim = strchr(cur, '/')) == NULL)
 565                         goto parse_failed;
 566                 *delim = 0;
 567                 np->local_ip = ntohl(in_aton(cur));
 568                 cur = delim;
 569         }
 570         cur++;
 571
 572         if (*cur != ',') {
 573                 /* parse out dev name */
 574                 if ((delim = strchr(cur, ',')) == NULL)
 575                         goto parse_failed;
 576                 *delim = 0;
 577                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 578                 cur = delim;
 579         }
 580         cur++;
 581
 582         if (*cur != '@') {
 583                 /* dst port */
 584                 if ((delim = strchr(cur, '@')) == NULL)
 585                         goto parse_failed;
 586                 *delim = 0;
 587                 np->remote_port = simple_strtol(cur, NULL, 10);
 588                 cur = delim;
 589         }
 590         cur++;
 591
 592         /* dst ip */
 593         if ((delim = strchr(cur, '/')) == NULL)
 594                 goto parse_failed;
 595         *delim = 0;
 596         np->remote_ip = ntohl(in_aton(cur));
 597         cur = delim + 1;
 598
 599         if (*cur != 0) {
 600                 /* MAC address */
 601                 if ((delim = strchr(cur, ':')) == NULL)
 602                         goto parse_failed;
 603                 *delim = 0;
 604                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
 605                 cur = delim + 1;
 606                 if ((delim = strchr(cur, ':')) == NULL)
 607                         goto parse_failed;
 608                 *delim = 0;
 609                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
 610                 cur = delim + 1;
 611                 if ((delim = strchr(cur, ':')) == NULL)
 612                         goto parse_failed;
 613                 *delim = 0;
 614                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
 615                 cur = delim + 1;
 616                 if ((delim = strchr(cur, ':')) == NULL)
 617                         goto parse_failed;
 618                 *delim = 0;
 619                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
 620                 cur = delim + 1;
 621                 if ((delim = strchr(cur, ':')) == NULL)
 622                         goto parse_failed;
 623                 *delim = 0;
 624                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 625                 cur = delim + 1;
 626                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 627         }
 628
 629         netpoll_print_options(np);
 630
 631         return 0;
 632
 633  parse_failed:
 634         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 635                np->name, cur);
 636         return -1;
 637 }
 638
 639 int netpoll_setup(struct netpoll *np)
 640 {
 641         struct net_device *ndev = NULL;
 642         struct in_device *in_dev;
 643         struct netpoll_info *npinfo;
 644         unsigned long flags;
 645         int err;
 646
 647         if (np->dev_name)
 648                 ndev = dev_get_by_name(&init_net, np->dev_name);
 649         if (!ndev) {
 650                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 651                        np->name, np->dev_name);
 652                 return -ENODEV;
 653         }
 654
 655         np->dev = ndev;
 656         if (!ndev->npinfo) {
 657                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 658                 if (!npinfo) {
 659                         err = -ENOMEM;
 660                         goto release;
 661                 }
 662
 663                 npinfo->rx_flags = 0;
 664                 npinfo->rx_np = NULL;
 665
 666                 spin_lock_init(&npinfo->rx_lock);
 667                 skb_queue_head_init(&npinfo->arp_tx);
 668                 skb_queue_head_init(&npinfo->txq);
 669                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 670
 671                 atomic_set(&npinfo->refcnt, 1);
 672         } else {
 673                 npinfo = ndev->npinfo;
 674                 atomic_inc(&npinfo->refcnt);
 675         }
 676
 677         if (!ndev->poll_controller) {
 678                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 679                        np->name, np->dev_name);
 680                 err = -ENOTSUPP;
 681                 goto release;
 682         }
 683
 684         if (!netif_running(ndev)) {
 685                 unsigned long atmost, atleast;
 686
 687                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 688                        np->name, np->dev_name);
 689
 690                 rtnl_lock();
 691                 err = dev_open(ndev);
 692                 rtnl_unlock();
 693
 694                 if (err) {
 695                         printk(KERN_ERR "%s: failed to open %s\n",
 696                                np->name, ndev->name);
 697                         goto release;
 698                 }
 699
 700                 atleast = jiffies + HZ/10;
 701                 atmost = jiffies + 4*HZ;
 702                 while (!netif_carrier_ok(ndev)) {
 703                         if (time_after(jiffies, atmost)) {
 704                                 printk(KERN_NOTICE
 705                                        "%s: timeout waiting for carrier\n",
 706                                        np->name);
 707                                 break;
 708                         }
 709                         cond_resched();
 710                 }
 711
 712                 /* If carrier appears to come up instantly, we don't
 713                  * trust it and pause so that we don't pump all our
 714                  * queued console messages into the bitbucket.
 715                  */
 716
 717                 if (time_before(jiffies, atleast)) {
 718                         printk(KERN_NOTICE "%s: carrier detect appears"
 719                                " untrustworthy, waiting 4 seconds\n",
 720                                np->name);
 721                         msleep(4000);
 722                 }
 723         }
 724
 725         if (is_zero_ether_addr(np->local_mac) && ndev->dev_addr)
 726                 memcpy(np->local_mac, ndev->dev_addr, 6);
 727
 728         if (!np->local_ip) {
 729                 rcu_read_lock();
 730                 in_dev = __in_dev_get_rcu(ndev);
 731
 732                 if (!in_dev || !in_dev->ifa_list) {
 733                         rcu_read_unlock();
 734                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 735                                np->name, np->dev_name);
 736                         err = -EDESTADDRREQ;
 737                         goto release;
 738                 }
 739
 740                 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
 741                 rcu_read_unlock();
 742                 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 743                        np->name, HIPQUAD(np->local_ip));
 744         }
 745
 746         if (np->rx_hook) {
 747                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 748                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 749                 npinfo->rx_np = np;
 750                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 751         }
 752
 753         /* fill up the skb queue */
 754         refill_skbs();
 755
 756         /* last thing to do is link it to the net device structure */
 757         ndev->npinfo = npinfo;
 758
 759         /* avoid racing with NAPI reading npinfo */
 760         synchronize_rcu();
 761
 762         return 0;
 763
 764  release:
 765         if (!ndev->npinfo)
 766                 kfree(npinfo);
 767         np->dev = NULL;
 768         dev_put(ndev);
 769         return err;
 770 }
 771
 772 static int __init netpoll_init(void)
 773 {
 774         skb_queue_head_init(&skb_pool);
 775         return 0;
 776 }
 777 core_initcall(netpoll_init);
 778
 779 void netpoll_cleanup(struct netpoll *np)
 780 {
 781         struct netpoll_info *npinfo;
 782         unsigned long flags;
 783
 784         if (np->dev) {
 785                 npinfo = np->dev->npinfo;
 786                 if (npinfo) {
 787                         if (npinfo->rx_np == np) {
 788                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 789                                 npinfo->rx_np = NULL;
 790                                 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 791                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 792                         }
 793
 794                         if (atomic_dec_and_test(&npinfo->refcnt)) {
 795                                 skb_queue_purge(&npinfo->arp_tx);
 796                                 skb_queue_purge(&npinfo->txq);
 797                                 cancel_rearming_delayed_work(&npinfo->tx_work);
 798
 799                                 /* clean after last, unfinished work */
 800                                 if (!skb_queue_empty(&npinfo->txq)) {
 801                                         struct sk_buff *skb;
 802                                         skb = __skb_dequeue(&npinfo->txq);
 803                                         kfree_skb(skb);
 804                                 }
 805                                 kfree(npinfo);
 806                                 np->dev->npinfo = NULL;
 807                         }
 808                 }
 809
 810                 dev_put(np->dev);
 811         }
 812
 813         np->dev = NULL;
 814 }
 815
 816 int netpoll_trap(void)
 817 {
 818         return atomic_read(&trapped);
 819 }
 820
 821 void netpoll_set_trap(int trap)
 822 {
 823         if (trap)
 824                 atomic_inc(&trapped);
 825         else
 826                 atomic_dec(&trapped);
 827 }
 828
 829 EXPORT_SYMBOL(netpoll_set_trap);
 830 EXPORT_SYMBOL(netpoll_trap);
 831 EXPORT_SYMBOL(netpoll_print_options);
 832 EXPORT_SYMBOL(netpoll_parse_options);
 833 EXPORT_SYMBOL(netpoll_setup);
 834 EXPORT_SYMBOL(netpoll_cleanup);
 835 EXPORT_SYMBOL(netpoll_send_udp);
 836 EXPORT_SYMBOL(netpoll_poll);