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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13
  14 #include <linux/moduleparam.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/etherdevice.h>
  17 #include <linux/string.h>
  18 #include <linux/if_arp.h>
  19 #include <linux/inetdevice.h>
  20 #include <linux/inet.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/netpoll.h>
  23 #include <linux/sched.h>
  24 #include <linux/delay.h>
  25 #include <linux/rcupdate.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/slab.h>
  28 #include <linux/export.h>
  29 #include <net/tcp.h>
  30 #include <net/udp.h>
  31 #include <asm/unaligned.h>
  32 #include <trace/events/napi.h>
  33
  34 /*
  35  * We maintain a small pool of fully-sized skbs, to make sure the
  36  * message gets out even in extreme OOM situations.
  37  */
  38
  39 #define MAX_UDP_CHUNK 1460
  40 #define MAX_SKBS 32
  41
  42 static struct sk_buff_head skb_pool;
  43
  44 static atomic_t trapped;
  45
  46 #define USEC_PER_POLL   50
  47 #define NETPOLL_RX_ENABLED  1
  48 #define NETPOLL_RX_DROP     2
  49
  50 #define MAX_SKB_SIZE                                                    \
  51         (sizeof(struct ethhdr) +                                        \
  52          sizeof(struct iphdr) +                                         \
  53          sizeof(struct udphdr) +                                        \
  54          MAX_UDP_CHUNK)
  55
  56 static void zap_completion_queue(void);
  57 static void arp_reply(struct sk_buff *skb);
  58
  59 static unsigned int carrier_timeout = 4;
  60 module_param(carrier_timeout, uint, 0644);
  61
  62 #define np_info(np, fmt, ...)                           \
  63         pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
  64 #define np_err(np, fmt, ...)                            \
  65         pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
  66 #define np_notice(np, fmt, ...)                         \
  67         pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
  68
  69 static void queue_process(struct work_struct *work)
  70 {
  71         struct netpoll_info *npinfo =
  72                 container_of(work, struct netpoll_info, tx_work.work);
  73         struct sk_buff *skb;
  74         unsigned long flags;
  75
  76         while ((skb = skb_dequeue(&npinfo->txq))) {
  77                 struct net_device *dev = skb->dev;
  78                 const struct net_device_ops *ops = dev->netdev_ops;
  79                 struct netdev_queue *txq;
  80
  81                 if (!netif_device_present(dev) || !netif_running(dev)) {
  82                         __kfree_skb(skb);
  83                         continue;
  84                 }
  85
  86                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  87
  88                 local_irq_save(flags);
  89                 __netif_tx_lock(txq, smp_processor_id());
  90                 if (netif_xmit_frozen_or_stopped(txq) ||
  91                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
  92                         skb_queue_head(&npinfo->txq, skb);
  93                         __netif_tx_unlock(txq);
  94                         local_irq_restore(flags);
  95
  96                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  97                         return;
  98                 }
  99                 __netif_tx_unlock(txq);
 100                 local_irq_restore(flags);
 101         }
 102 }
 103
 104 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 105                             unsigned short ulen, __be32 saddr, __be32 daddr)
 106 {
 107         __wsum psum;
 108
 109         if (uh->check == 0 || skb_csum_unnecessary(skb))
 110                 return 0;
 111
 112         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 113
 114         if (skb->ip_summed == CHECKSUM_COMPLETE &&
 115             !csum_fold(csum_add(psum, skb->csum)))
 116                 return 0;
 117
 118         skb->csum = psum;
 119
 120         return __skb_checksum_complete(skb);
 121 }
 122
 123 /*
 124  * Check whether delayed processing was scheduled for our NIC. If so,
 125  * we attempt to grab the poll lock and use ->poll() to pump the card.
 126  * If this fails, either we've recursed in ->poll() or it's already
 127  * running on another CPU.
 128  *
 129  * Note: we don't mask interrupts with this lock because we're using
 130  * trylock here and interrupts are already disabled in the softirq
 131  * case. Further, we test the poll_owner to avoid recursion on UP
 132  * systems where the lock doesn't exist.
 133  *
 134  * In cases where there is bi-directional communications, reading only
 135  * one message at a time can lead to packets being dropped by the
 136  * network adapter, forcing superfluous retries and possibly timeouts.
 137  * Thus, we set our budget to greater than 1.
 138  */
 139 static int poll_one_napi(struct netpoll_info *npinfo,
 140                          struct napi_struct *napi, int budget)
 141 {
 142         int work;
 143
 144         /* net_rx_action's ->poll() invocations and our's are
 145          * synchronized by this test which is only made while
 146          * holding the napi->poll_lock.
 147          */
 148         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 149                 return budget;
 150
 151         npinfo->rx_flags |= NETPOLL_RX_DROP;
 152         atomic_inc(&trapped);
 153         set_bit(NAPI_STATE_NPSVC, &napi->state);
 154
 155         work = napi->poll(napi, budget);
 156         trace_napi_poll(napi);
 157
 158         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 159         atomic_dec(&trapped);
 160         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 161
 162         return budget - work;
 163 }
 164
 165 static void poll_napi(struct net_device *dev)
 166 {
 167         struct napi_struct *napi;
 168         int budget = 16;
 169
 170         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 171                 if (napi->poll_owner != smp_processor_id() &&
 172                     spin_trylock(&napi->poll_lock)) {
 173                         budget = poll_one_napi(dev->npinfo, napi, budget);
 174                         spin_unlock(&napi->poll_lock);
 175
 176                         if (!budget)
 177                                 break;
 178                 }
 179         }
 180 }
 181
 182 static void service_arp_queue(struct netpoll_info *npi)
 183 {
 184         if (npi) {
 185                 struct sk_buff *skb;
 186
 187                 while ((skb = skb_dequeue(&npi->arp_tx)))
 188                         arp_reply(skb);
 189         }
 190 }
 191
 192 static void netpoll_poll_dev(struct net_device *dev)
 193 {
 194         const struct net_device_ops *ops;
 195
 196         if (!dev || !netif_running(dev))
 197                 return;
 198
 199         ops = dev->netdev_ops;
 200         if (!ops->ndo_poll_controller)
 201                 return;
 202
 203         /* Process pending work on NIC */
 204         ops->ndo_poll_controller(dev);
 205
 206         poll_napi(dev);
 207
 208         if (dev->flags & IFF_SLAVE) {
 209                 if (dev->npinfo) {
 210                         struct net_device *bond_dev = dev->master;
 211                         struct sk_buff *skb;
 212                         while ((skb = skb_dequeue(&dev->npinfo->arp_tx))) {
 213                                 skb->dev = bond_dev;
 214                                 skb_queue_tail(&bond_dev->npinfo->arp_tx, skb);
 215                         }
 216                 }
 217         }
 218
 219         service_arp_queue(dev->npinfo);
 220
 221         zap_completion_queue();
 222 }
 223
 224 static void refill_skbs(void)
 225 {
 226         struct sk_buff *skb;
 227         unsigned long flags;
 228
 229         spin_lock_irqsave(&skb_pool.lock, flags);
 230         while (skb_pool.qlen < MAX_SKBS) {
 231                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 232                 if (!skb)
 233                         break;
 234
 235                 __skb_queue_tail(&skb_pool, skb);
 236         }
 237         spin_unlock_irqrestore(&skb_pool.lock, flags);
 238 }
 239
 240 static void zap_completion_queue(void)
 241 {
 242         unsigned long flags;
 243         struct softnet_data *sd = &get_cpu_var(softnet_data);
 244
 245         if (sd->completion_queue) {
 246                 struct sk_buff *clist;
 247
 248                 local_irq_save(flags);
 249                 clist = sd->completion_queue;
 250                 sd->completion_queue = NULL;
 251                 local_irq_restore(flags);
 252
 253                 while (clist != NULL) {
 254                         struct sk_buff *skb = clist;
 255                         clist = clist->next;
 256                         if (skb->destructor) {
 257                                 atomic_inc(&skb->users);
 258                                 dev_kfree_skb_any(skb); /* put this one back */
 259                         } else {
 260                                 __kfree_skb(skb);
 261                         }
 262                 }
 263         }
 264
 265         put_cpu_var(softnet_data);
 266 }
 267
 268 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 269 {
 270         int count = 0;
 271         struct sk_buff *skb;
 272
 273         zap_completion_queue();
 274         refill_skbs();
 275 repeat:
 276
 277         skb = alloc_skb(len, GFP_ATOMIC);
 278         if (!skb)
 279                 skb = skb_dequeue(&skb_pool);
 280
 281         if (!skb) {
 282                 if (++count < 10) {
 283                         netpoll_poll_dev(np->dev);
 284                         goto repeat;
 285                 }
 286                 return NULL;
 287         }
 288
 289         atomic_set(&skb->users, 1);
 290         skb_reserve(skb, reserve);
 291         return skb;
 292 }
 293
 294 static int netpoll_owner_active(struct net_device *dev)
 295 {
 296         struct napi_struct *napi;
 297
 298         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 299                 if (napi->poll_owner == smp_processor_id())
 300                         return 1;
 301         }
 302         return 0;
 303 }
 304
 305 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 306                              struct net_device *dev)
 307 {
 308         int status = NETDEV_TX_BUSY;
 309         unsigned long tries;
 310         const struct net_device_ops *ops = dev->netdev_ops;
 311         /* It is up to the caller to keep npinfo alive. */
 312         struct netpoll_info *npinfo = np->dev->npinfo;
 313
 314         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 315                 __kfree_skb(skb);
 316                 return;
 317         }
 318
 319         /* don't get messages out of order, and no recursion */
 320         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 321                 struct netdev_queue *txq;
 322                 unsigned long flags;
 323
 324                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 325
 326                 local_irq_save(flags);
 327                 /* try until next clock tick */
 328                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 329                      tries > 0; --tries) {
 330                         if (__netif_tx_trylock(txq)) {
 331                                 if (!netif_xmit_stopped(txq)) {
 332                                         status = ops->ndo_start_xmit(skb, dev);
 333                                         if (status == NETDEV_TX_OK)
 334                                                 txq_trans_update(txq);
 335                                 }
 336                                 __netif_tx_unlock(txq);
 337
 338                                 if (status == NETDEV_TX_OK)
 339                                         break;
 340
 341                         }
 342
 343                         /* tickle device maybe there is some cleanup */
 344                         netpoll_poll_dev(np->dev);
 345
 346                         udelay(USEC_PER_POLL);
 347                 }
 348
 349                 WARN_ONCE(!irqs_disabled(),
 350                         "netpoll_send_skb(): %s enabled interrupts in poll (%pF)\n",
 351                         dev->name, ops->ndo_start_xmit);
 352
 353                 local_irq_restore(flags);
 354         }
 355
 356         if (status != NETDEV_TX_OK) {
 357                 skb_queue_tail(&npinfo->txq, skb);
 358                 schedule_delayed_work(&npinfo->tx_work,0);
 359         }
 360 }
 361 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 362
 363 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 364 {
 365         int total_len, ip_len, udp_len;
 366         struct sk_buff *skb;
 367         struct udphdr *udph;
 368         struct iphdr *iph;
 369         struct ethhdr *eth;
 370
 371         udp_len = len + sizeof(*udph);
 372         ip_len = udp_len + sizeof(*iph);
 373         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 374
 375         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 376                        total_len - len);
 377         if (!skb)
 378                 return;
 379
 380         skb_copy_to_linear_data(skb, msg, len);
 381         skb_put(skb, len);
 382
 383         skb_push(skb, sizeof(*udph));
 384         skb_reset_transport_header(skb);
 385         udph = udp_hdr(skb);
 386         udph->source = htons(np->local_port);
 387         udph->dest = htons(np->remote_port);
 388         udph->len = htons(udp_len);
 389         udph->check = 0;
 390         udph->check = csum_tcpudp_magic(np->local_ip,
 391                                         np->remote_ip,
 392                                         udp_len, IPPROTO_UDP,
 393                                         csum_partial(udph, udp_len, 0));
 394         if (udph->check == 0)
 395                 udph->check = CSUM_MANGLED_0;
 396
 397         skb_push(skb, sizeof(*iph));
 398         skb_reset_network_header(skb);
 399         iph = ip_hdr(skb);
 400
 401         /* iph->version = 4; iph->ihl = 5; */
 402         put_unaligned(0x45, (unsigned char *)iph);
 403         iph->tos      = 0;
 404         put_unaligned(htons(ip_len), &(iph->tot_len));
 405         iph->id       = 0;
 406         iph->frag_off = 0;
 407         iph->ttl      = 64;
 408         iph->protocol = IPPROTO_UDP;
 409         iph->check    = 0;
 410         put_unaligned(np->local_ip, &(iph->saddr));
 411         put_unaligned(np->remote_ip, &(iph->daddr));
 412         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 413
 414         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 415         skb_reset_mac_header(skb);
 416         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 417         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 418         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 419
 420         skb->dev = np->dev;
 421
 422         netpoll_send_skb(np, skb);
 423 }
 424 EXPORT_SYMBOL(netpoll_send_udp);
 425
 426 static void arp_reply(struct sk_buff *skb)
 427 {
 428         struct netpoll_info *npinfo = skb->dev->npinfo;
 429         struct arphdr *arp;
 430         unsigned char *arp_ptr;
 431         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 432         __be32 sip, tip;
 433         unsigned char *sha;
 434         struct sk_buff *send_skb;
 435         struct netpoll *np, *tmp;
 436         unsigned long flags;
 437         int hlen, tlen;
 438         int hits = 0;
 439
 440         if (list_empty(&npinfo->rx_np))
 441                 return;
 442
 443         /* Before checking the packet, we do some early
 444            inspection whether this is interesting at all */
 445         spin_lock_irqsave(&npinfo->rx_lock, flags);
 446         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 447                 if (np->dev == skb->dev)
 448                         hits++;
 449         }
 450         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 451
 452         /* No netpoll struct is using this dev */
 453         if (!hits)
 454                 return;
 455
 456         /* No arp on this interface */
 457         if (skb->dev->flags & IFF_NOARP)
 458                 return;
 459
 460         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 461                 return;
 462
 463         skb_reset_network_header(skb);
 464         skb_reset_transport_header(skb);
 465         arp = arp_hdr(skb);
 466
 467         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 468              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 469             arp->ar_pro != htons(ETH_P_IP) ||
 470             arp->ar_op != htons(ARPOP_REQUEST))
 471                 return;
 472
 473         arp_ptr = (unsigned char *)(arp+1);
 474         /* save the location of the src hw addr */
 475         sha = arp_ptr;
 476         arp_ptr += skb->dev->addr_len;
 477         memcpy(&sip, arp_ptr, 4);
 478         arp_ptr += 4;
 479         /* If we actually cared about dst hw addr,
 480            it would get copied here */
 481         arp_ptr += skb->dev->addr_len;
 482         memcpy(&tip, arp_ptr, 4);
 483
 484         /* Should we ignore arp? */
 485         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 486                 return;
 487
 488         size = arp_hdr_len(skb->dev);
 489
 490         spin_lock_irqsave(&npinfo->rx_lock, flags);
 491         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 492                 if (tip != np->local_ip)
 493                         continue;
 494
 495                 hlen = LL_RESERVED_SPACE(np->dev);
 496                 tlen = np->dev->needed_tailroom;
 497                 send_skb = find_skb(np, size + hlen + tlen, hlen);
 498                 if (!send_skb)
 499                         continue;
 500
 501                 skb_reset_network_header(send_skb);
 502                 arp = (struct arphdr *) skb_put(send_skb, size);
 503                 send_skb->dev = skb->dev;
 504                 send_skb->protocol = htons(ETH_P_ARP);
 505
 506                 /* Fill the device header for the ARP frame */
 507                 if (dev_hard_header(send_skb, skb->dev, ptype,
 508                                     sha, np->dev->dev_addr,
 509                                     send_skb->len) < 0) {
 510                         kfree_skb(send_skb);
 511                         continue;
 512                 }
 513
 514                 /*
 515                  * Fill out the arp protocol part.
 516                  *
 517                  * we only support ethernet device type,
 518                  * which (according to RFC 1390) should
 519                  * always equal 1 (Ethernet).
 520                  */
 521
 522                 arp->ar_hrd = htons(np->dev->type);
 523                 arp->ar_pro = htons(ETH_P_IP);
 524                 arp->ar_hln = np->dev->addr_len;
 525                 arp->ar_pln = 4;
 526                 arp->ar_op = htons(type);
 527
 528                 arp_ptr = (unsigned char *)(arp + 1);
 529                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 530                 arp_ptr += np->dev->addr_len;
 531                 memcpy(arp_ptr, &tip, 4);
 532                 arp_ptr += 4;
 533                 memcpy(arp_ptr, sha, np->dev->addr_len);
 534                 arp_ptr += np->dev->addr_len;
 535                 memcpy(arp_ptr, &sip, 4);
 536
 537                 netpoll_send_skb(np, send_skb);
 538
 539                 /* If there are several rx_hooks for the same address,
 540                    we're fine by sending a single reply */
 541                 break;
 542         }
 543         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 544 }
 545
 546 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
 547 {
 548         int proto, len, ulen;
 549         int hits = 0;
 550         const struct iphdr *iph;
 551         struct udphdr *uh;
 552         struct netpoll *np, *tmp;
 553
 554         if (list_empty(&npinfo->rx_np))
 555                 goto out;
 556
 557         if (skb->dev->type != ARPHRD_ETHER)
 558                 goto out;
 559
 560         /* check if netpoll clients need ARP */
 561         if (skb->protocol == htons(ETH_P_ARP) &&
 562             atomic_read(&trapped)) {
 563                 skb_queue_tail(&npinfo->arp_tx, skb);
 564                 return 1;
 565         }
 566
 567         proto = ntohs(eth_hdr(skb)->h_proto);
 568         if (proto != ETH_P_IP)
 569                 goto out;
 570         if (skb->pkt_type == PACKET_OTHERHOST)
 571                 goto out;
 572         if (skb_shared(skb))
 573                 goto out;
 574
 575         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 576                 goto out;
 577         iph = (struct iphdr *)skb->data;
 578         if (iph->ihl < 5 || iph->version != 4)
 579                 goto out;
 580         if (!pskb_may_pull(skb, iph->ihl*4))
 581                 goto out;
 582         iph = (struct iphdr *)skb->data;
 583         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 584                 goto out;
 585
 586         len = ntohs(iph->tot_len);
 587         if (skb->len < len || len < iph->ihl*4)
 588                 goto out;
 589
 590         /*
 591          * Our transport medium may have padded the buffer out.
 592          * Now We trim to the true length of the frame.
 593          */
 594         if (pskb_trim_rcsum(skb, len))
 595                 goto out;
 596
 597         iph = (struct iphdr *)skb->data;
 598         if (iph->protocol != IPPROTO_UDP)
 599                 goto out;
 600
 601         len -= iph->ihl*4;
 602         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 603         ulen = ntohs(uh->len);
 604
 605         if (ulen != len)
 606                 goto out;
 607         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 608                 goto out;
 609
 610         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 611                 if (np->local_ip && np->local_ip != iph->daddr)
 612                         continue;
 613                 if (np->remote_ip && np->remote_ip != iph->saddr)
 614                         continue;
 615                 if (np->local_port && np->local_port != ntohs(uh->dest))
 616                         continue;
 617
 618                 np->rx_hook(np, ntohs(uh->source),
 619                                (char *)(uh+1),
 620                                ulen - sizeof(struct udphdr));
 621                 hits++;
 622         }
 623
 624         if (!hits)
 625                 goto out;
 626
 627         kfree_skb(skb);
 628         return 1;
 629
 630 out:
 631         if (atomic_read(&trapped)) {
 632                 kfree_skb(skb);
 633                 return 1;
 634         }
 635
 636         return 0;
 637 }
 638
 639 void netpoll_print_options(struct netpoll *np)
 640 {
 641         np_info(np, "local port %d\n", np->local_port);
 642         np_info(np, "local IP %pI4\n", &np->local_ip);
 643         np_info(np, "interface '%s'\n", np->dev_name);
 644         np_info(np, "remote port %d\n", np->remote_port);
 645         np_info(np, "remote IP %pI4\n", &np->remote_ip);
 646         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 647 }
 648 EXPORT_SYMBOL(netpoll_print_options);
 649
 650 int netpoll_parse_options(struct netpoll *np, char *opt)
 651 {
 652         char *cur=opt, *delim;
 653
 654         if (*cur != '@') {
 655                 if ((delim = strchr(cur, '@')) == NULL)
 656                         goto parse_failed;
 657                 *delim = 0;
 658                 np->local_port = simple_strtol(cur, NULL, 10);
 659                 cur = delim;
 660         }
 661         cur++;
 662
 663         if (*cur != '/') {
 664                 if ((delim = strchr(cur, '/')) == NULL)
 665                         goto parse_failed;
 666                 *delim = 0;
 667                 np->local_ip = in_aton(cur);
 668                 cur = delim;
 669         }
 670         cur++;
 671
 672         if (*cur != ',') {
 673                 /* parse out dev name */
 674                 if ((delim = strchr(cur, ',')) == NULL)
 675                         goto parse_failed;
 676                 *delim = 0;
 677                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 678                 cur = delim;
 679         }
 680         cur++;
 681
 682         if (*cur != '@') {
 683                 /* dst port */
 684                 if ((delim = strchr(cur, '@')) == NULL)
 685                         goto parse_failed;
 686                 *delim = 0;
 687                 if (*cur == ' ' || *cur == '\t')
 688                         np_info(np, "warning: whitespace is not allowed\n");
 689                 np->remote_port = simple_strtol(cur, NULL, 10);
 690                 cur = delim;
 691         }
 692         cur++;
 693
 694         /* dst ip */
 695         if ((delim = strchr(cur, '/')) == NULL)
 696                 goto parse_failed;
 697         *delim = 0;
 698         np->remote_ip = in_aton(cur);
 699         cur = delim + 1;
 700
 701         if (*cur != 0) {
 702                 /* MAC address */
 703                 if (!mac_pton(cur, np->remote_mac))
 704                         goto parse_failed;
 705         }
 706
 707         netpoll_print_options(np);
 708
 709         return 0;
 710
 711  parse_failed:
 712         np_info(np, "couldn't parse config at '%s'!\n", cur);
 713         return -1;
 714 }
 715 EXPORT_SYMBOL(netpoll_parse_options);
 716
 717 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
 718 {
 719         struct netpoll_info *npinfo;
 720         const struct net_device_ops *ops;
 721         unsigned long flags;
 722         int err;
 723
 724         np->dev = ndev;
 725         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 726
 727         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 728             !ndev->netdev_ops->ndo_poll_controller) {
 729                 np_err(np, "%s doesn't support polling, aborting\n",
 730                        np->dev_name);
 731                 err = -ENOTSUPP;
 732                 goto out;
 733         }
 734
 735         if (!ndev->npinfo) {
 736                 npinfo = kmalloc(sizeof(*npinfo), gfp);
 737                 if (!npinfo) {
 738                         err = -ENOMEM;
 739                         goto out;
 740                 }
 741
 742                 npinfo->rx_flags = 0;
 743                 INIT_LIST_HEAD(&npinfo->rx_np);
 744
 745                 spin_lock_init(&npinfo->rx_lock);
 746                 skb_queue_head_init(&npinfo->arp_tx);
 747                 skb_queue_head_init(&npinfo->txq);
 748                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 749
 750                 atomic_set(&npinfo->refcnt, 1);
 751
 752                 ops = np->dev->netdev_ops;
 753                 if (ops->ndo_netpoll_setup) {
 754                         err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
 755                         if (err)
 756                                 goto free_npinfo;
 757                 }
 758         } else {
 759                 npinfo = ndev->npinfo;
 760                 atomic_inc(&npinfo->refcnt);
 761         }
 762
 763         npinfo->netpoll = np;
 764
 765         if (np->rx_hook) {
 766                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 767                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 768                 list_add_tail(&np->rx, &npinfo->rx_np);
 769                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 770         }
 771
 772         /* last thing to do is link it to the net device structure */
 773         rcu_assign_pointer(ndev->npinfo, npinfo);
 774
 775         return 0;
 776
 777 free_npinfo:
 778         kfree(npinfo);
 779 out:
 780         return err;
 781 }
 782 EXPORT_SYMBOL_GPL(__netpoll_setup);
 783
 784 int netpoll_setup(struct netpoll *np)
 785 {
 786         struct net_device *ndev = NULL;
 787         struct in_device *in_dev;
 788         int err;
 789
 790         if (np->dev_name)
 791                 ndev = dev_get_by_name(&init_net, np->dev_name);
 792         if (!ndev) {
 793                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 794                 return -ENODEV;
 795         }
 796
 797         if (ndev->master) {
 798                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 799                 err = -EBUSY;
 800                 goto put;
 801         }
 802
 803         if (!netif_running(ndev)) {
 804                 unsigned long atmost, atleast;
 805
 806                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 807
 808                 rtnl_lock();
 809                 err = dev_open(ndev);
 810                 rtnl_unlock();
 811
 812                 if (err) {
 813                         np_err(np, "failed to open %s\n", ndev->name);
 814                         goto put;
 815                 }
 816
 817                 atleast = jiffies + HZ/10;
 818                 atmost = jiffies + carrier_timeout * HZ;
 819                 while (!netif_carrier_ok(ndev)) {
 820                         if (time_after(jiffies, atmost)) {
 821                                 np_notice(np, "timeout waiting for carrier\n");
 822                                 break;
 823                         }
 824                         msleep(1);
 825                 }
 826
 827                 /* If carrier appears to come up instantly, we don't
 828                  * trust it and pause so that we don't pump all our
 829                  * queued console messages into the bitbucket.
 830                  */
 831
 832                 if (time_before(jiffies, atleast)) {
 833                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 834                         msleep(4000);
 835                 }
 836         }
 837
 838         if (!np->local_ip) {
 839                 rcu_read_lock();
 840                 in_dev = __in_dev_get_rcu(ndev);
 841
 842                 if (!in_dev || !in_dev->ifa_list) {
 843                         rcu_read_unlock();
 844                         np_err(np, "no IP address for %s, aborting\n",
 845                                np->dev_name);
 846                         err = -EDESTADDRREQ;
 847                         goto put;
 848                 }
 849
 850                 np->local_ip = in_dev->ifa_list->ifa_local;
 851                 rcu_read_unlock();
 852                 np_info(np, "local IP %pI4\n", &np->local_ip);
 853         }
 854
 855         /* fill up the skb queue */
 856         refill_skbs();
 857
 858         rtnl_lock();
 859         err = __netpoll_setup(np, ndev, GFP_KERNEL);
 860         rtnl_unlock();
 861
 862         if (err)
 863                 goto put;
 864
 865         return 0;
 866
 867 put:
 868         dev_put(ndev);
 869         return err;
 870 }
 871 EXPORT_SYMBOL(netpoll_setup);
 872
 873 static int __init netpoll_init(void)
 874 {
 875         skb_queue_head_init(&skb_pool);
 876         return 0;
 877 }
 878 core_initcall(netpoll_init);
 879
 880 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 881 {
 882         struct netpoll_info *npinfo =
 883                         container_of(rcu_head, struct netpoll_info, rcu);
 884
 885         skb_queue_purge(&npinfo->arp_tx);
 886         skb_queue_purge(&npinfo->txq);
 887
 888         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 889         cancel_delayed_work(&npinfo->tx_work);
 890
 891         /* clean after last, unfinished work */
 892         __skb_queue_purge(&npinfo->txq);
 893         /* now cancel it again */
 894         cancel_delayed_work(&npinfo->tx_work);
 895         kfree(npinfo);
 896 }
 897
 898 void __netpoll_cleanup(struct netpoll *np)
 899 {
 900         struct netpoll_info *npinfo;
 901         unsigned long flags;
 902
 903         npinfo = np->dev->npinfo;
 904         if (!npinfo)
 905                 return;
 906
 907         if (!list_empty(&npinfo->rx_np)) {
 908                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 909                 list_del(&np->rx);
 910                 if (list_empty(&npinfo->rx_np))
 911                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 912                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 913         }
 914
 915         if (atomic_dec_and_test(&npinfo->refcnt)) {
 916                 const struct net_device_ops *ops;
 917
 918                 ops = np->dev->netdev_ops;
 919                 if (ops->ndo_netpoll_cleanup)
 920                         ops->ndo_netpoll_cleanup(np->dev);
 921
 922                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 923                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
 924         }
 925 }
 926 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 927
 928 static void rcu_cleanup_netpoll(struct rcu_head *rcu_head)
 929 {
 930         struct netpoll *np = container_of(rcu_head, struct netpoll, rcu);
 931
 932         __netpoll_cleanup(np);
 933         kfree(np);
 934 }
 935
 936 void __netpoll_free_rcu(struct netpoll *np)
 937 {
 938         call_rcu_bh(&np->rcu, rcu_cleanup_netpoll);
 939 }
 940 EXPORT_SYMBOL_GPL(__netpoll_free_rcu);
 941
 942 void netpoll_cleanup(struct netpoll *np)
 943 {
 944         if (!np->dev)
 945                 return;
 946
 947         rtnl_lock();
 948         __netpoll_cleanup(np);
 949         rtnl_unlock();
 950
 951         dev_put(np->dev);
 952         np->dev = NULL;
 953 }
 954 EXPORT_SYMBOL(netpoll_cleanup);
 955
 956 int netpoll_trap(void)
 957 {
 958         return atomic_read(&trapped);
 959 }
 960 EXPORT_SYMBOL(netpoll_trap);
 961
 962 void netpoll_set_trap(int trap)
 963 {
 964         if (trap)
 965                 atomic_inc(&trapped);
 966         else
 967                 atomic_dec(&trapped);
 968 }
 969 EXPORT_SYMBOL(netpoll_set_trap);