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 #include <trace/events/napi.h>
  28
  29 /*
  30  * We maintain a small pool of fully-sized skbs, to make sure the
  31  * message gets out even in extreme OOM situations.
  32  */
  33
  34 #define MAX_UDP_CHUNK 1460
  35 #define MAX_SKBS 32
  36 #define MAX_QUEUE_DEPTH (MAX_SKBS / 2)
  37
  38 static struct sk_buff_head skb_pool;
  39
  40 static atomic_t trapped;
  41
  42 #define USEC_PER_POLL   50
  43 #define NETPOLL_RX_ENABLED  1
  44 #define NETPOLL_RX_DROP     2
  45
  46 #define MAX_SKB_SIZE \
  47                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
  48                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
  49
  50 static void zap_completion_queue(void);
  51 static void arp_reply(struct sk_buff *skb);
  52
  53 static void queue_process(struct work_struct *work)
  54 {
  55         struct netpoll_info *npinfo =
  56                 container_of(work, struct netpoll_info, tx_work.work);
  57         struct sk_buff *skb;
  58         unsigned long flags;
  59
  60         while ((skb = skb_dequeue(&npinfo->txq))) {
  61                 struct net_device *dev = skb->dev;
  62                 const struct net_device_ops *ops = dev->netdev_ops;
  63                 struct netdev_queue *txq;
  64
  65                 if (!netif_device_present(dev) || !netif_running(dev)) {
  66                         __kfree_skb(skb);
  67                         continue;
  68                 }
  69
  70                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  71
  72                 local_irq_save(flags);
  73                 __netif_tx_lock(txq, smp_processor_id());
  74                 if (netif_tx_queue_stopped(txq) ||
  75                     netif_tx_queue_frozen(txq) ||
  76                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
  77                         skb_queue_head(&npinfo->txq, skb);
  78                         __netif_tx_unlock(txq);
  79                         local_irq_restore(flags);
  80
  81                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  82                         return;
  83                 }
  84                 __netif_tx_unlock(txq);
  85                 local_irq_restore(flags);
  86         }
  87 }
  88
  89 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  90                             unsigned short ulen, __be32 saddr, __be32 daddr)
  91 {
  92         __wsum psum;
  93
  94         if (uh->check == 0 || skb_csum_unnecessary(skb))
  95                 return 0;
  96
  97         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
  98
  99         if (skb->ip_summed == CHECKSUM_COMPLETE &&
 100             !csum_fold(csum_add(psum, skb->csum)))
 101                 return 0;
 102
 103         skb->csum = psum;
 104
 105         return __skb_checksum_complete(skb);
 106 }
 107
 108 /*
 109  * Check whether delayed processing was scheduled for our NIC. If so,
 110  * we attempt to grab the poll lock and use ->poll() to pump the card.
 111  * If this fails, either we've recursed in ->poll() or it's already
 112  * running on another CPU.
 113  *
 114  * Note: we don't mask interrupts with this lock because we're using
 115  * trylock here and interrupts are already disabled in the softirq
 116  * case. Further, we test the poll_owner to avoid recursion on UP
 117  * systems where the lock doesn't exist.
 118  *
 119  * In cases where there is bi-directional communications, reading only
 120  * one message at a time can lead to packets being dropped by the
 121  * network adapter, forcing superfluous retries and possibly timeouts.
 122  * Thus, we set our budget to greater than 1.
 123  */
 124 static int poll_one_napi(struct netpoll_info *npinfo,
 125                          struct napi_struct *napi, int budget)
 126 {
 127         int work;
 128
 129         /* net_rx_action's ->poll() invocations and our's are
 130          * synchronized by this test which is only made while
 131          * holding the napi->poll_lock.
 132          */
 133         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 134                 return budget;
 135
 136         npinfo->rx_flags |= NETPOLL_RX_DROP;
 137         atomic_inc(&trapped);
 138         set_bit(NAPI_STATE_NPSVC, &napi->state);
 139
 140         work = napi->poll(napi, budget);
 141         trace_napi_poll(napi);
 142
 143         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 144         atomic_dec(&trapped);
 145         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 146
 147         return budget - work;
 148 }
 149
 150 static void poll_napi(struct net_device *dev)
 151 {
 152         struct napi_struct *napi;
 153         int budget = 16;
 154
 155         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 156                 if (napi->poll_owner != smp_processor_id() &&
 157                     spin_trylock(&napi->poll_lock)) {
 158                         budget = poll_one_napi(dev->npinfo, napi, budget);
 159                         spin_unlock(&napi->poll_lock);
 160
 161                         if (!budget)
 162                                 break;
 163                 }
 164         }
 165 }
 166
 167 static void service_arp_queue(struct netpoll_info *npi)
 168 {
 169         if (npi) {
 170                 struct sk_buff *skb;
 171
 172                 while ((skb = skb_dequeue(&npi->arp_tx)))
 173                         arp_reply(skb);
 174         }
 175 }
 176
 177 void netpoll_poll(struct netpoll *np)
 178 {
 179         struct net_device *dev = np->dev;
 180         const struct net_device_ops *ops;
 181
 182         if (!dev || !netif_running(dev))
 183                 return;
 184
 185         ops = dev->netdev_ops;
 186         if (!ops->ndo_poll_controller)
 187                 return;
 188
 189         /* Process pending work on NIC */
 190         ops->ndo_poll_controller(dev);
 191
 192         poll_napi(dev);
 193
 194         service_arp_queue(dev->npinfo);
 195
 196         zap_completion_queue();
 197 }
 198
 199 static void refill_skbs(void)
 200 {
 201         struct sk_buff *skb;
 202         unsigned long flags;
 203
 204         spin_lock_irqsave(&skb_pool.lock, flags);
 205         while (skb_pool.qlen < MAX_SKBS) {
 206                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 207                 if (!skb)
 208                         break;
 209
 210                 __skb_queue_tail(&skb_pool, skb);
 211         }
 212         spin_unlock_irqrestore(&skb_pool.lock, flags);
 213 }
 214
 215 static void zap_completion_queue(void)
 216 {
 217         unsigned long flags;
 218         struct softnet_data *sd = &get_cpu_var(softnet_data);
 219
 220         if (sd->completion_queue) {
 221                 struct sk_buff *clist;
 222
 223                 local_irq_save(flags);
 224                 clist = sd->completion_queue;
 225                 sd->completion_queue = NULL;
 226                 local_irq_restore(flags);
 227
 228                 while (clist != NULL) {
 229                         struct sk_buff *skb = clist;
 230                         clist = clist->next;
 231                         if (skb->destructor) {
 232                                 atomic_inc(&skb->users);
 233                                 dev_kfree_skb_any(skb); /* put this one back */
 234                         } else {
 235                                 __kfree_skb(skb);
 236                         }
 237                 }
 238         }
 239
 240         put_cpu_var(softnet_data);
 241 }
 242
 243 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 244 {
 245         int count = 0;
 246         struct sk_buff *skb;
 247
 248         zap_completion_queue();
 249         refill_skbs();
 250 repeat:
 251
 252         skb = alloc_skb(len, GFP_ATOMIC);
 253         if (!skb)
 254                 skb = skb_dequeue(&skb_pool);
 255
 256         if (!skb) {
 257                 if (++count < 10) {
 258                         netpoll_poll(np);
 259                         goto repeat;
 260                 }
 261                 return NULL;
 262         }
 263
 264         atomic_set(&skb->users, 1);
 265         skb_reserve(skb, reserve);
 266         return skb;
 267 }
 268
 269 static int netpoll_owner_active(struct net_device *dev)
 270 {
 271         struct napi_struct *napi;
 272
 273         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 274                 if (napi->poll_owner == smp_processor_id())
 275                         return 1;
 276         }
 277         return 0;
 278 }
 279
 280 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 281 {
 282         int status = NETDEV_TX_BUSY;
 283         unsigned long tries;
 284         struct net_device *dev = np->dev;
 285         const struct net_device_ops *ops = dev->netdev_ops;
 286         struct netpoll_info *npinfo = np->dev->npinfo;
 287
 288         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 289                 __kfree_skb(skb);
 290                 return;
 291         }
 292
 293         /* don't get messages out of order, and no recursion */
 294         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 295                 struct netdev_queue *txq;
 296                 unsigned long flags;
 297
 298                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 299
 300                 local_irq_save(flags);
 301                 /* try until next clock tick */
 302                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 303                      tries > 0; --tries) {
 304                         if (__netif_tx_trylock(txq)) {
 305                                 if (!netif_tx_queue_stopped(txq)) {
 306                                         status = ops->ndo_start_xmit(skb, dev);
 307                                         if (status == NETDEV_TX_OK)
 308                                                 txq_trans_update(txq);
 309                                 }
 310                                 __netif_tx_unlock(txq);
 311
 312                                 if (status == NETDEV_TX_OK)
 313                                         break;
 314
 315                         }
 316
 317                         /* tickle device maybe there is some cleanup */
 318                         netpoll_poll(np);
 319
 320                         udelay(USEC_PER_POLL);
 321                 }
 322                 local_irq_restore(flags);
 323         }
 324
 325         if (status != NETDEV_TX_OK) {
 326                 skb_queue_tail(&npinfo->txq, skb);
 327                 schedule_delayed_work(&npinfo->tx_work,0);
 328         }
 329 }
 330
 331 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 332 {
 333         int total_len, eth_len, ip_len, udp_len;
 334         struct sk_buff *skb;
 335         struct udphdr *udph;
 336         struct iphdr *iph;
 337         struct ethhdr *eth;
 338
 339         udp_len = len + sizeof(*udph);
 340         ip_len = eth_len = udp_len + sizeof(*iph);
 341         total_len = eth_len + ETH_HLEN + NET_IP_ALIGN;
 342
 343         skb = find_skb(np, total_len, total_len - len);
 344         if (!skb)
 345                 return;
 346
 347         skb_copy_to_linear_data(skb, msg, len);
 348         skb->len += len;
 349
 350         skb_push(skb, sizeof(*udph));
 351         skb_reset_transport_header(skb);
 352         udph = udp_hdr(skb);
 353         udph->source = htons(np->local_port);
 354         udph->dest = htons(np->remote_port);
 355         udph->len = htons(udp_len);
 356         udph->check = 0;
 357         udph->check = csum_tcpudp_magic(np->local_ip,
 358                                         np->remote_ip,
 359                                         udp_len, IPPROTO_UDP,
 360                                         csum_partial(udph, udp_len, 0));
 361         if (udph->check == 0)
 362                 udph->check = CSUM_MANGLED_0;
 363
 364         skb_push(skb, sizeof(*iph));
 365         skb_reset_network_header(skb);
 366         iph = ip_hdr(skb);
 367
 368         /* iph->version = 4; iph->ihl = 5; */
 369         put_unaligned(0x45, (unsigned char *)iph);
 370         iph->tos      = 0;
 371         put_unaligned(htons(ip_len), &(iph->tot_len));
 372         iph->id       = 0;
 373         iph->frag_off = 0;
 374         iph->ttl      = 64;
 375         iph->protocol = IPPROTO_UDP;
 376         iph->check    = 0;
 377         put_unaligned(np->local_ip, &(iph->saddr));
 378         put_unaligned(np->remote_ip, &(iph->daddr));
 379         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 380
 381         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 382         skb_reset_mac_header(skb);
 383         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 384         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 385         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 386
 387         skb->dev = np->dev;
 388
 389         netpoll_send_skb(np, skb);
 390 }
 391
 392 static void arp_reply(struct sk_buff *skb)
 393 {
 394         struct netpoll_info *npinfo = skb->dev->npinfo;
 395         struct arphdr *arp;
 396         unsigned char *arp_ptr;
 397         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 398         __be32 sip, tip;
 399         unsigned char *sha;
 400         struct sk_buff *send_skb;
 401         struct netpoll *np = NULL;
 402
 403         if (npinfo->rx_np && npinfo->rx_np->dev == skb->dev)
 404                 np = npinfo->rx_np;
 405         if (!np)
 406                 return;
 407
 408         /* No arp on this interface */
 409         if (skb->dev->flags & IFF_NOARP)
 410                 return;
 411
 412         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 413                 return;
 414
 415         skb_reset_network_header(skb);
 416         skb_reset_transport_header(skb);
 417         arp = arp_hdr(skb);
 418
 419         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 420              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 421             arp->ar_pro != htons(ETH_P_IP) ||
 422             arp->ar_op != htons(ARPOP_REQUEST))
 423                 return;
 424
 425         arp_ptr = (unsigned char *)(arp+1);
 426         /* save the location of the src hw addr */
 427         sha = arp_ptr;
 428         arp_ptr += skb->dev->addr_len;
 429         memcpy(&sip, arp_ptr, 4);
 430         arp_ptr += 4;
 431         /* if we actually cared about dst hw addr, it would get copied here */
 432         arp_ptr += skb->dev->addr_len;
 433         memcpy(&tip, arp_ptr, 4);
 434
 435         /* Should we ignore arp? */
 436         if (tip != np->local_ip ||
 437             ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 438                 return;
 439
 440         size = arp_hdr_len(skb->dev);
 441         send_skb = find_skb(np, size + LL_ALLOCATED_SPACE(np->dev),
 442                             LL_RESERVED_SPACE(np->dev));
 443
 444         if (!send_skb)
 445                 return;
 446
 447         skb_reset_network_header(send_skb);
 448         arp = (struct arphdr *) skb_put(send_skb, size);
 449         send_skb->dev = skb->dev;
 450         send_skb->protocol = htons(ETH_P_ARP);
 451
 452         /* Fill the device header for the ARP frame */
 453         if (dev_hard_header(send_skb, skb->dev, ptype,
 454                             sha, np->dev->dev_addr,
 455                             send_skb->len) < 0) {
 456                 kfree_skb(send_skb);
 457                 return;
 458         }
 459
 460         /*
 461          * Fill out the arp protocol part.
 462          *
 463          * we only support ethernet device type,
 464          * which (according to RFC 1390) should always equal 1 (Ethernet).
 465          */
 466
 467         arp->ar_hrd = htons(np->dev->type);
 468         arp->ar_pro = htons(ETH_P_IP);
 469         arp->ar_hln = np->dev->addr_len;
 470         arp->ar_pln = 4;
 471         arp->ar_op = htons(type);
 472
 473         arp_ptr=(unsigned char *)(arp + 1);
 474         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 475         arp_ptr += np->dev->addr_len;
 476         memcpy(arp_ptr, &tip, 4);
 477         arp_ptr += 4;
 478         memcpy(arp_ptr, sha, np->dev->addr_len);
 479         arp_ptr += np->dev->addr_len;
 480         memcpy(arp_ptr, &sip, 4);
 481
 482         netpoll_send_skb(np, send_skb);
 483 }
 484
 485 int __netpoll_rx(struct sk_buff *skb)
 486 {
 487         int proto, len, ulen;
 488         struct iphdr *iph;
 489         struct udphdr *uh;
 490         struct netpoll_info *npi = skb->dev->npinfo;
 491         struct netpoll *np = npi->rx_np;
 492
 493         if (!np)
 494                 goto out;
 495         if (skb->dev->type != ARPHRD_ETHER)
 496                 goto out;
 497
 498         /* check if netpoll clients need ARP */
 499         if (skb->protocol == htons(ETH_P_ARP) &&
 500             atomic_read(&trapped)) {
 501                 skb_queue_tail(&npi->arp_tx, skb);
 502                 return 1;
 503         }
 504
 505         proto = ntohs(eth_hdr(skb)->h_proto);
 506         if (proto != ETH_P_IP)
 507                 goto out;
 508         if (skb->pkt_type == PACKET_OTHERHOST)
 509                 goto out;
 510         if (skb_shared(skb))
 511                 goto out;
 512
 513         iph = (struct iphdr *)skb->data;
 514         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 515                 goto out;
 516         if (iph->ihl < 5 || iph->version != 4)
 517                 goto out;
 518         if (!pskb_may_pull(skb, iph->ihl*4))
 519                 goto out;
 520         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 521                 goto out;
 522
 523         len = ntohs(iph->tot_len);
 524         if (skb->len < len || len < iph->ihl*4)
 525                 goto out;
 526
 527         /*
 528          * Our transport medium may have padded the buffer out.
 529          * Now We trim to the true length of the frame.
 530          */
 531         if (pskb_trim_rcsum(skb, len))
 532                 goto out;
 533
 534         if (iph->protocol != IPPROTO_UDP)
 535                 goto out;
 536
 537         len -= iph->ihl*4;
 538         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 539         ulen = ntohs(uh->len);
 540
 541         if (ulen != len)
 542                 goto out;
 543         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 544                 goto out;
 545         if (np->local_ip && np->local_ip != iph->daddr)
 546                 goto out;
 547         if (np->remote_ip && np->remote_ip != iph->saddr)
 548                 goto out;
 549         if (np->local_port && np->local_port != ntohs(uh->dest))
 550                 goto out;
 551
 552         np->rx_hook(np, ntohs(uh->source),
 553                     (char *)(uh+1),
 554                     ulen - sizeof(struct udphdr));
 555
 556         kfree_skb(skb);
 557         return 1;
 558
 559 out:
 560         if (atomic_read(&trapped)) {
 561                 kfree_skb(skb);
 562                 return 1;
 563         }
 564
 565         return 0;
 566 }
 567
 568 void netpoll_print_options(struct netpoll *np)
 569 {
 570         printk(KERN_INFO "%s: local port %d\n",
 571                          np->name, np->local_port);
 572         printk(KERN_INFO "%s: local IP %pI4\n",
 573                          np->name, &np->local_ip);
 574         printk(KERN_INFO "%s: interface %s\n",
 575                          np->name, np->dev_name);
 576         printk(KERN_INFO "%s: remote port %d\n",
 577                          np->name, np->remote_port);
 578         printk(KERN_INFO "%s: remote IP %pI4\n",
 579                          np->name, &np->remote_ip);
 580         printk(KERN_INFO "%s: remote ethernet address %pM\n",
 581                          np->name, np->remote_mac);
 582 }
 583
 584 int netpoll_parse_options(struct netpoll *np, char *opt)
 585 {
 586         char *cur=opt, *delim;
 587
 588         if (*cur != '@') {
 589                 if ((delim = strchr(cur, '@')) == NULL)
 590                         goto parse_failed;
 591                 *delim = 0;
 592                 np->local_port = simple_strtol(cur, NULL, 10);
 593                 cur = delim;
 594         }
 595         cur++;
 596
 597         if (*cur != '/') {
 598                 if ((delim = strchr(cur, '/')) == NULL)
 599                         goto parse_failed;
 600                 *delim = 0;
 601                 np->local_ip = in_aton(cur);
 602                 cur = delim;
 603         }
 604         cur++;
 605
 606         if (*cur != ',') {
 607                 /* parse out dev name */
 608                 if ((delim = strchr(cur, ',')) == NULL)
 609                         goto parse_failed;
 610                 *delim = 0;
 611                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 612                 cur = delim;
 613         }
 614         cur++;
 615
 616         if (*cur != '@') {
 617                 /* dst port */
 618                 if ((delim = strchr(cur, '@')) == NULL)
 619                         goto parse_failed;
 620                 *delim = 0;
 621                 np->remote_port = simple_strtol(cur, NULL, 10);
 622                 cur = delim;
 623         }
 624         cur++;
 625
 626         /* dst ip */
 627         if ((delim = strchr(cur, '/')) == NULL)
 628                 goto parse_failed;
 629         *delim = 0;
 630         np->remote_ip = in_aton(cur);
 631         cur = delim + 1;
 632
 633         if (*cur != 0) {
 634                 /* MAC address */
 635                 if ((delim = strchr(cur, ':')) == NULL)
 636                         goto parse_failed;
 637                 *delim = 0;
 638                 np->remote_mac[0] = simple_strtol(cur, NULL, 16);
 639                 cur = delim + 1;
 640                 if ((delim = strchr(cur, ':')) == NULL)
 641                         goto parse_failed;
 642                 *delim = 0;
 643                 np->remote_mac[1] = simple_strtol(cur, NULL, 16);
 644                 cur = delim + 1;
 645                 if ((delim = strchr(cur, ':')) == NULL)
 646                         goto parse_failed;
 647                 *delim = 0;
 648                 np->remote_mac[2] = simple_strtol(cur, NULL, 16);
 649                 cur = delim + 1;
 650                 if ((delim = strchr(cur, ':')) == NULL)
 651                         goto parse_failed;
 652                 *delim = 0;
 653                 np->remote_mac[3] = simple_strtol(cur, NULL, 16);
 654                 cur = delim + 1;
 655                 if ((delim = strchr(cur, ':')) == NULL)
 656                         goto parse_failed;
 657                 *delim = 0;
 658                 np->remote_mac[4] = simple_strtol(cur, NULL, 16);
 659                 cur = delim + 1;
 660                 np->remote_mac[5] = simple_strtol(cur, NULL, 16);
 661         }
 662
 663         netpoll_print_options(np);
 664
 665         return 0;
 666
 667  parse_failed:
 668         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 669                np->name, cur);
 670         return -1;
 671 }
 672
 673 int netpoll_setup(struct netpoll *np)
 674 {
 675         struct net_device *ndev = NULL;
 676         struct in_device *in_dev;
 677         struct netpoll_info *npinfo;
 678         unsigned long flags;
 679         int err;
 680
 681         if (np->dev_name)
 682                 ndev = dev_get_by_name(&init_net, np->dev_name);
 683         if (!ndev) {
 684                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 685                        np->name, np->dev_name);
 686                 return -ENODEV;
 687         }
 688
 689         np->dev = ndev;
 690         if (!ndev->npinfo) {
 691                 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
 692                 if (!npinfo) {
 693                         err = -ENOMEM;
 694                         goto release;
 695                 }
 696
 697                 npinfo->rx_flags = 0;
 698                 npinfo->rx_np = NULL;
 699
 700                 spin_lock_init(&npinfo->rx_lock);
 701                 skb_queue_head_init(&npinfo->arp_tx);
 702                 skb_queue_head_init(&npinfo->txq);
 703                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 704
 705                 atomic_set(&npinfo->refcnt, 1);
 706         } else {
 707                 npinfo = ndev->npinfo;
 708                 atomic_inc(&npinfo->refcnt);
 709         }
 710
 711         if (!ndev->netdev_ops->ndo_poll_controller) {
 712                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 713                        np->name, np->dev_name);
 714                 err = -ENOTSUPP;
 715                 goto release;
 716         }
 717
 718         if (!netif_running(ndev)) {
 719                 unsigned long atmost, atleast;
 720
 721                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 722                        np->name, np->dev_name);
 723
 724                 rtnl_lock();
 725                 err = dev_open(ndev);
 726                 rtnl_unlock();
 727
 728                 if (err) {
 729                         printk(KERN_ERR "%s: failed to open %s\n",
 730                                np->name, ndev->name);
 731                         goto release;
 732                 }
 733
 734                 atleast = jiffies + HZ/10;
 735                 atmost = jiffies + 4*HZ;
 736                 while (!netif_carrier_ok(ndev)) {
 737                         if (time_after(jiffies, atmost)) {
 738                                 printk(KERN_NOTICE
 739                                        "%s: timeout waiting for carrier\n",
 740                                        np->name);
 741                                 break;
 742                         }
 743                         cond_resched();
 744                 }
 745
 746                 /* If carrier appears to come up instantly, we don't
 747                  * trust it and pause so that we don't pump all our
 748                  * queued console messages into the bitbucket.
 749                  */
 750
 751                 if (time_before(jiffies, atleast)) {
 752                         printk(KERN_NOTICE "%s: carrier detect appears"
 753                                " untrustworthy, waiting 4 seconds\n",
 754                                np->name);
 755                         msleep(4000);
 756                 }
 757         }
 758
 759         if (!np->local_ip) {
 760                 rcu_read_lock();
 761                 in_dev = __in_dev_get_rcu(ndev);
 762
 763                 if (!in_dev || !in_dev->ifa_list) {
 764                         rcu_read_unlock();
 765                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 766                                np->name, np->dev_name);
 767                         err = -EDESTADDRREQ;
 768                         goto release;
 769                 }
 770
 771                 np->local_ip = in_dev->ifa_list->ifa_local;
 772                 rcu_read_unlock();
 773                 printk(KERN_INFO "%s: local IP %pI4\n", np->name, &np->local_ip);
 774         }
 775
 776         if (np->rx_hook) {
 777                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 778                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 779                 npinfo->rx_np = np;
 780                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 781         }
 782
 783         /* fill up the skb queue */
 784         refill_skbs();
 785
 786         /* last thing to do is link it to the net device structure */
 787         ndev->npinfo = npinfo;
 788
 789         /* avoid racing with NAPI reading npinfo */
 790         synchronize_rcu();
 791
 792         return 0;
 793
 794  release:
 795         if (!ndev->npinfo)
 796                 kfree(npinfo);
 797         np->dev = NULL;
 798         dev_put(ndev);
 799         return err;
 800 }
 801
 802 static int __init netpoll_init(void)
 803 {
 804         skb_queue_head_init(&skb_pool);
 805         return 0;
 806 }
 807 core_initcall(netpoll_init);
 808
 809 void netpoll_cleanup(struct netpoll *np)
 810 {
 811         struct netpoll_info *npinfo;
 812         unsigned long flags;
 813
 814         if (np->dev) {
 815                 npinfo = np->dev->npinfo;
 816                 if (npinfo) {
 817                         if (npinfo->rx_np == np) {
 818                                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 819                                 npinfo->rx_np = NULL;
 820                                 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 821                                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 822                         }
 823
 824                         if (atomic_dec_and_test(&npinfo->refcnt)) {
 825                                 skb_queue_purge(&npinfo->arp_tx);
 826                                 skb_queue_purge(&npinfo->txq);
 827                                 cancel_rearming_delayed_work(&npinfo->tx_work);
 828
 829                                 /* clean after last, unfinished work */
 830                                 __skb_queue_purge(&npinfo->txq);
 831                                 kfree(npinfo);
 832                                 np->dev->npinfo = NULL;
 833                         }
 834                 }
 835
 836                 dev_put(np->dev);
 837         }
 838
 839         np->dev = NULL;
 840 }
 841
 842 int netpoll_trap(void)
 843 {
 844         return atomic_read(&trapped);
 845 }
 846
 847 void netpoll_set_trap(int trap)
 848 {
 849         if (trap)
 850                 atomic_inc(&trapped);
 851         else
 852                 atomic_dec(&trapped);
 853 }
 854
 855 EXPORT_SYMBOL(netpoll_set_trap);
 856 EXPORT_SYMBOL(netpoll_trap);
 857 EXPORT_SYMBOL(netpoll_print_options);
 858 EXPORT_SYMBOL(netpoll_parse_options);
 859 EXPORT_SYMBOL(netpoll_setup);
 860 EXPORT_SYMBOL(netpoll_cleanup);
 861 EXPORT_SYMBOL(netpoll_send_udp);
 862 EXPORT_SYMBOL(netpoll_poll);