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 <linux/if_vlan.h>
  30 #include <net/tcp.h>
  31 #include <net/udp.h>
  32 #include <asm/unaligned.h>
  33 #include <trace/events/napi.h>
  34
  35 /*
  36  * We maintain a small pool of fully-sized skbs, to make sure the
  37  * message gets out even in extreme OOM situations.
  38  */
  39
  40 #define MAX_UDP_CHUNK 1460
  41 #define MAX_SKBS 32
  42
  43 static struct sk_buff_head skb_pool;
  44
  45 static atomic_t trapped;
  46
  47 #define USEC_PER_POLL   50
  48 #define NETPOLL_RX_ENABLED  1
  49 #define NETPOLL_RX_DROP     2
  50
  51 #define MAX_SKB_SIZE                                                    \
  52         (sizeof(struct ethhdr) +                                        \
  53          sizeof(struct iphdr) +                                         \
  54          sizeof(struct udphdr) +                                        \
  55          MAX_UDP_CHUNK)
  56
  57 static void zap_completion_queue(void);
  58 static void netpoll_arp_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
  59
  60 static unsigned int carrier_timeout = 4;
  61 module_param(carrier_timeout, uint, 0644);
  62
  63 #define np_info(np, fmt, ...)                           \
  64         pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
  65 #define np_err(np, fmt, ...)                            \
  66         pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
  67 #define np_notice(np, fmt, ...)                         \
  68         pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
  69
  70 static void queue_process(struct work_struct *work)
  71 {
  72         struct netpoll_info *npinfo =
  73                 container_of(work, struct netpoll_info, tx_work.work);
  74         struct sk_buff *skb;
  75         unsigned long flags;
  76
  77         while ((skb = skb_dequeue(&npinfo->txq))) {
  78                 struct net_device *dev = skb->dev;
  79                 const struct net_device_ops *ops = dev->netdev_ops;
  80                 struct netdev_queue *txq;
  81
  82                 if (!netif_device_present(dev) || !netif_running(dev)) {
  83                         __kfree_skb(skb);
  84                         continue;
  85                 }
  86
  87                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  88
  89                 local_irq_save(flags);
  90                 __netif_tx_lock(txq, smp_processor_id());
  91                 if (netif_xmit_frozen_or_stopped(txq) ||
  92                     ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) {
  93                         skb_queue_head(&npinfo->txq, skb);
  94                         __netif_tx_unlock(txq);
  95                         local_irq_restore(flags);
  96
  97                         schedule_delayed_work(&npinfo->tx_work, HZ/10);
  98                         return;
  99                 }
 100                 __netif_tx_unlock(txq);
 101                 local_irq_restore(flags);
 102         }
 103 }
 104
 105 static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 106                             unsigned short ulen, __be32 saddr, __be32 daddr)
 107 {
 108         __wsum psum;
 109
 110         if (uh->check == 0 || skb_csum_unnecessary(skb))
 111                 return 0;
 112
 113         psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 114
 115         if (skb->ip_summed == CHECKSUM_COMPLETE &&
 116             !csum_fold(csum_add(psum, skb->csum)))
 117                 return 0;
 118
 119         skb->csum = psum;
 120
 121         return __skb_checksum_complete(skb);
 122 }
 123
 124 /*
 125  * Check whether delayed processing was scheduled for our NIC. If so,
 126  * we attempt to grab the poll lock and use ->poll() to pump the card.
 127  * If this fails, either we've recursed in ->poll() or it's already
 128  * running on another CPU.
 129  *
 130  * Note: we don't mask interrupts with this lock because we're using
 131  * trylock here and interrupts are already disabled in the softirq
 132  * case. Further, we test the poll_owner to avoid recursion on UP
 133  * systems where the lock doesn't exist.
 134  *
 135  * In cases where there is bi-directional communications, reading only
 136  * one message at a time can lead to packets being dropped by the
 137  * network adapter, forcing superfluous retries and possibly timeouts.
 138  * Thus, we set our budget to greater than 1.
 139  */
 140 static int poll_one_napi(struct netpoll_info *npinfo,
 141                          struct napi_struct *napi, int budget)
 142 {
 143         int work;
 144
 145         /* net_rx_action's ->poll() invocations and our's are
 146          * synchronized by this test which is only made while
 147          * holding the napi->poll_lock.
 148          */
 149         if (!test_bit(NAPI_STATE_SCHED, &napi->state))
 150                 return budget;
 151
 152         npinfo->rx_flags |= NETPOLL_RX_DROP;
 153         atomic_inc(&trapped);
 154         set_bit(NAPI_STATE_NPSVC, &napi->state);
 155
 156         work = napi->poll(napi, budget);
 157         trace_napi_poll(napi);
 158
 159         clear_bit(NAPI_STATE_NPSVC, &napi->state);
 160         atomic_dec(&trapped);
 161         npinfo->rx_flags &= ~NETPOLL_RX_DROP;
 162
 163         return budget - work;
 164 }
 165
 166 static void poll_napi(struct net_device *dev)
 167 {
 168         struct napi_struct *napi;
 169         int budget = 16;
 170
 171         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 172                 if (napi->poll_owner != smp_processor_id() &&
 173                     spin_trylock(&napi->poll_lock)) {
 174                         budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
 175                                                napi, budget);
 176                         spin_unlock(&napi->poll_lock);
 177
 178                         if (!budget)
 179                                 break;
 180                 }
 181         }
 182 }
 183
 184 static void service_arp_queue(struct netpoll_info *npi)
 185 {
 186         if (npi) {
 187                 struct sk_buff *skb;
 188
 189                 while ((skb = skb_dequeue(&npi->arp_tx)))
 190                         netpoll_arp_reply(skb, npi);
 191         }
 192 }
 193
 194 static void netpoll_poll_dev(struct net_device *dev)
 195 {
 196         const struct net_device_ops *ops;
 197         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 198
 199         if (!dev || !netif_running(dev))
 200                 return;
 201
 202         ops = dev->netdev_ops;
 203         if (!ops->ndo_poll_controller)
 204                 return;
 205
 206         /* Process pending work on NIC */
 207         ops->ndo_poll_controller(dev);
 208
 209         poll_napi(dev);
 210
 211         if (dev->flags & IFF_SLAVE) {
 212                 if (ni) {
 213                         struct net_device *bond_dev = dev->master;
 214                         struct sk_buff *skb;
 215                         struct netpoll_info *bond_ni = rcu_dereference_bh(bond_dev->npinfo);
 216                         while ((skb = skb_dequeue(&ni->arp_tx))) {
 217                                 skb->dev = bond_dev;
 218                                 skb_queue_tail(&bond_ni->arp_tx, skb);
 219                         }
 220                 }
 221         }
 222
 223         service_arp_queue(ni);
 224
 225         zap_completion_queue();
 226 }
 227
 228 static void refill_skbs(void)
 229 {
 230         struct sk_buff *skb;
 231         unsigned long flags;
 232
 233         spin_lock_irqsave(&skb_pool.lock, flags);
 234         while (skb_pool.qlen < MAX_SKBS) {
 235                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 236                 if (!skb)
 237                         break;
 238
 239                 __skb_queue_tail(&skb_pool, skb);
 240         }
 241         spin_unlock_irqrestore(&skb_pool.lock, flags);
 242 }
 243
 244 static void zap_completion_queue(void)
 245 {
 246         unsigned long flags;
 247         struct softnet_data *sd = &get_cpu_var(softnet_data);
 248
 249         if (sd->completion_queue) {
 250                 struct sk_buff *clist;
 251
 252                 local_irq_save(flags);
 253                 clist = sd->completion_queue;
 254                 sd->completion_queue = NULL;
 255                 local_irq_restore(flags);
 256
 257                 while (clist != NULL) {
 258                         struct sk_buff *skb = clist;
 259                         clist = clist->next;
 260                         if (skb->destructor) {
 261                                 atomic_inc(&skb->users);
 262                                 dev_kfree_skb_any(skb); /* put this one back */
 263                         } else {
 264                                 __kfree_skb(skb);
 265                         }
 266                 }
 267         }
 268
 269         put_cpu_var(softnet_data);
 270 }
 271
 272 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 273 {
 274         int count = 0;
 275         struct sk_buff *skb;
 276
 277         zap_completion_queue();
 278         refill_skbs();
 279 repeat:
 280
 281         skb = alloc_skb(len, GFP_ATOMIC);
 282         if (!skb)
 283                 skb = skb_dequeue(&skb_pool);
 284
 285         if (!skb) {
 286                 if (++count < 10) {
 287                         netpoll_poll_dev(np->dev);
 288                         goto repeat;
 289                 }
 290                 return NULL;
 291         }
 292
 293         atomic_set(&skb->users, 1);
 294         skb_reserve(skb, reserve);
 295         return skb;
 296 }
 297
 298 static int netpoll_owner_active(struct net_device *dev)
 299 {
 300         struct napi_struct *napi;
 301
 302         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 303                 if (napi->poll_owner == smp_processor_id())
 304                         return 1;
 305         }
 306         return 0;
 307 }
 308
 309 /* call with IRQ disabled */
 310 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 311                              struct net_device *dev)
 312 {
 313         int status = NETDEV_TX_BUSY;
 314         unsigned long tries;
 315         const struct net_device_ops *ops = dev->netdev_ops;
 316         /* It is up to the caller to keep npinfo alive. */
 317         struct netpoll_info *npinfo;
 318
 319         WARN_ON_ONCE(!irqs_disabled());
 320
 321         npinfo = rcu_dereference_bh(np->dev->npinfo);
 322         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 323                 __kfree_skb(skb);
 324                 return;
 325         }
 326
 327         /* don't get messages out of order, and no recursion */
 328         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 329                 struct netdev_queue *txq;
 330
 331                 txq = netdev_pick_tx(dev, skb);
 332
 333                 /* try until next clock tick */
 334                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 335                      tries > 0; --tries) {
 336                         if (__netif_tx_trylock(txq)) {
 337                                 if (!netif_xmit_stopped(txq)) {
 338                                         if (vlan_tx_tag_present(skb) &&
 339                                             !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) {
 340                                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 341                                                 if (unlikely(!skb))
 342                                                         break;
 343                                                 skb->vlan_tci = 0;
 344                                         }
 345
 346                                         status = ops->ndo_start_xmit(skb, dev);
 347                                         if (status == NETDEV_TX_OK)
 348                                                 txq_trans_update(txq);
 349                                 }
 350                                 __netif_tx_unlock(txq);
 351
 352                                 if (status == NETDEV_TX_OK)
 353                                         break;
 354
 355                         }
 356
 357                         /* tickle device maybe there is some cleanup */
 358                         netpoll_poll_dev(np->dev);
 359
 360                         udelay(USEC_PER_POLL);
 361                 }
 362
 363                 WARN_ONCE(!irqs_disabled(),
 364                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
 365                         dev->name, ops->ndo_start_xmit);
 366
 367         }
 368
 369         if (status != NETDEV_TX_OK) {
 370                 skb_queue_tail(&npinfo->txq, skb);
 371                 schedule_delayed_work(&npinfo->tx_work,0);
 372         }
 373 }
 374 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 375
 376 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 377 {
 378         int total_len, ip_len, udp_len;
 379         struct sk_buff *skb;
 380         struct udphdr *udph;
 381         struct iphdr *iph;
 382         struct ethhdr *eth;
 383         static atomic_t ip_ident;
 384
 385         udp_len = len + sizeof(*udph);
 386         ip_len = udp_len + sizeof(*iph);
 387         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 388
 389         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 390                        total_len - len);
 391         if (!skb)
 392                 return;
 393
 394         skb_copy_to_linear_data(skb, msg, len);
 395         skb_put(skb, len);
 396
 397         skb_push(skb, sizeof(*udph));
 398         skb_reset_transport_header(skb);
 399         udph = udp_hdr(skb);
 400         udph->source = htons(np->local_port);
 401         udph->dest = htons(np->remote_port);
 402         udph->len = htons(udp_len);
 403         udph->check = 0;
 404         udph->check = csum_tcpudp_magic(np->local_ip,
 405                                         np->remote_ip,
 406                                         udp_len, IPPROTO_UDP,
 407                                         csum_partial(udph, udp_len, 0));
 408         if (udph->check == 0)
 409                 udph->check = CSUM_MANGLED_0;
 410
 411         skb_push(skb, sizeof(*iph));
 412         skb_reset_network_header(skb);
 413         iph = ip_hdr(skb);
 414
 415         /* iph->version = 4; iph->ihl = 5; */
 416         put_unaligned(0x45, (unsigned char *)iph);
 417         iph->tos      = 0;
 418         put_unaligned(htons(ip_len), &(iph->tot_len));
 419         iph->id       = htons(atomic_inc_return(&ip_ident));
 420         iph->frag_off = 0;
 421         iph->ttl      = 64;
 422         iph->protocol = IPPROTO_UDP;
 423         iph->check    = 0;
 424         put_unaligned(np->local_ip, &(iph->saddr));
 425         put_unaligned(np->remote_ip, &(iph->daddr));
 426         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 427
 428         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 429         skb_reset_mac_header(skb);
 430         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 431         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 432         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 433
 434         skb->dev = np->dev;
 435
 436         netpoll_send_skb(np, skb);
 437 }
 438 EXPORT_SYMBOL(netpoll_send_udp);
 439
 440 static void netpoll_arp_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
 441 {
 442         struct arphdr *arp;
 443         unsigned char *arp_ptr;
 444         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 445         __be32 sip, tip;
 446         unsigned char *sha;
 447         struct sk_buff *send_skb;
 448         struct netpoll *np, *tmp;
 449         unsigned long flags;
 450         int hlen, tlen;
 451         int hits = 0;
 452
 453         if (list_empty(&npinfo->rx_np))
 454                 return;
 455
 456         /* Before checking the packet, we do some early
 457            inspection whether this is interesting at all */
 458         spin_lock_irqsave(&npinfo->rx_lock, flags);
 459         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 460                 if (np->dev == skb->dev)
 461                         hits++;
 462         }
 463         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 464
 465         /* No netpoll struct is using this dev */
 466         if (!hits)
 467                 return;
 468
 469         /* No arp on this interface */
 470         if (skb->dev->flags & IFF_NOARP)
 471                 return;
 472
 473         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 474                 return;
 475
 476         skb_reset_network_header(skb);
 477         skb_reset_transport_header(skb);
 478         arp = arp_hdr(skb);
 479
 480         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 481              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 482             arp->ar_pro != htons(ETH_P_IP) ||
 483             arp->ar_op != htons(ARPOP_REQUEST))
 484                 return;
 485
 486         arp_ptr = (unsigned char *)(arp+1);
 487         /* save the location of the src hw addr */
 488         sha = arp_ptr;
 489         arp_ptr += skb->dev->addr_len;
 490         memcpy(&sip, arp_ptr, 4);
 491         arp_ptr += 4;
 492         /* If we actually cared about dst hw addr,
 493            it would get copied here */
 494         arp_ptr += skb->dev->addr_len;
 495         memcpy(&tip, arp_ptr, 4);
 496
 497         /* Should we ignore arp? */
 498         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 499                 return;
 500
 501         size = arp_hdr_len(skb->dev);
 502
 503         spin_lock_irqsave(&npinfo->rx_lock, flags);
 504         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 505                 if (tip != np->local_ip)
 506                         continue;
 507
 508                 hlen = LL_RESERVED_SPACE(np->dev);
 509                 tlen = np->dev->needed_tailroom;
 510                 send_skb = find_skb(np, size + hlen + tlen, hlen);
 511                 if (!send_skb)
 512                         continue;
 513
 514                 skb_reset_network_header(send_skb);
 515                 arp = (struct arphdr *) skb_put(send_skb, size);
 516                 send_skb->dev = skb->dev;
 517                 send_skb->protocol = htons(ETH_P_ARP);
 518
 519                 /* Fill the device header for the ARP frame */
 520                 if (dev_hard_header(send_skb, skb->dev, ptype,
 521                                     sha, np->dev->dev_addr,
 522                                     send_skb->len) < 0) {
 523                         kfree_skb(send_skb);
 524                         continue;
 525                 }
 526
 527                 /*
 528                  * Fill out the arp protocol part.
 529                  *
 530                  * we only support ethernet device type,
 531                  * which (according to RFC 1390) should
 532                  * always equal 1 (Ethernet).
 533                  */
 534
 535                 arp->ar_hrd = htons(np->dev->type);
 536                 arp->ar_pro = htons(ETH_P_IP);
 537                 arp->ar_hln = np->dev->addr_len;
 538                 arp->ar_pln = 4;
 539                 arp->ar_op = htons(type);
 540
 541                 arp_ptr = (unsigned char *)(arp + 1);
 542                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 543                 arp_ptr += np->dev->addr_len;
 544                 memcpy(arp_ptr, &tip, 4);
 545                 arp_ptr += 4;
 546                 memcpy(arp_ptr, sha, np->dev->addr_len);
 547                 arp_ptr += np->dev->addr_len;
 548                 memcpy(arp_ptr, &sip, 4);
 549
 550                 netpoll_send_skb(np, send_skb);
 551
 552                 /* If there are several rx_hooks for the same address,
 553                    we're fine by sending a single reply */
 554                 break;
 555         }
 556         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 557 }
 558
 559 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
 560 {
 561         int proto, len, ulen;
 562         int hits = 0;
 563         const struct iphdr *iph;
 564         struct udphdr *uh;
 565         struct netpoll *np, *tmp;
 566
 567         if (list_empty(&npinfo->rx_np))
 568                 goto out;
 569
 570         if (skb->dev->type != ARPHRD_ETHER)
 571                 goto out;
 572
 573         /* check if netpoll clients need ARP */
 574         if (skb->protocol == htons(ETH_P_ARP) &&
 575             atomic_read(&trapped)) {
 576                 skb_queue_tail(&npinfo->arp_tx, skb);
 577                 return 1;
 578         }
 579
 580         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
 581                 skb = vlan_untag(skb);
 582                 if (unlikely(!skb))
 583                         goto out;
 584         }
 585
 586         proto = ntohs(eth_hdr(skb)->h_proto);
 587         if (proto != ETH_P_IP)
 588                 goto out;
 589         if (skb->pkt_type == PACKET_OTHERHOST)
 590                 goto out;
 591         if (skb_shared(skb))
 592                 goto out;
 593
 594         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 595                 goto out;
 596         iph = (struct iphdr *)skb->data;
 597         if (iph->ihl < 5 || iph->version != 4)
 598                 goto out;
 599         if (!pskb_may_pull(skb, iph->ihl*4))
 600                 goto out;
 601         iph = (struct iphdr *)skb->data;
 602         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 603                 goto out;
 604
 605         len = ntohs(iph->tot_len);
 606         if (skb->len < len || len < iph->ihl*4)
 607                 goto out;
 608
 609         /*
 610          * Our transport medium may have padded the buffer out.
 611          * Now We trim to the true length of the frame.
 612          */
 613         if (pskb_trim_rcsum(skb, len))
 614                 goto out;
 615
 616         iph = (struct iphdr *)skb->data;
 617         if (iph->protocol != IPPROTO_UDP)
 618                 goto out;
 619
 620         len -= iph->ihl*4;
 621         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 622         ulen = ntohs(uh->len);
 623
 624         if (ulen != len)
 625                 goto out;
 626         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 627                 goto out;
 628
 629         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 630                 if (np->local_ip && np->local_ip != iph->daddr)
 631                         continue;
 632                 if (np->remote_ip && np->remote_ip != iph->saddr)
 633                         continue;
 634                 if (np->local_port && np->local_port != ntohs(uh->dest))
 635                         continue;
 636
 637                 np->rx_hook(np, ntohs(uh->source),
 638                                (char *)(uh+1),
 639                                ulen - sizeof(struct udphdr));
 640                 hits++;
 641         }
 642
 643         if (!hits)
 644                 goto out;
 645
 646         kfree_skb(skb);
 647         return 1;
 648
 649 out:
 650         if (atomic_read(&trapped)) {
 651                 kfree_skb(skb);
 652                 return 1;
 653         }
 654
 655         return 0;
 656 }
 657
 658 void netpoll_print_options(struct netpoll *np)
 659 {
 660         np_info(np, "local port %d\n", np->local_port);
 661         np_info(np, "local IP %pI4\n", &np->local_ip);
 662         np_info(np, "interface '%s'\n", np->dev_name);
 663         np_info(np, "remote port %d\n", np->remote_port);
 664         np_info(np, "remote IP %pI4\n", &np->remote_ip);
 665         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 666 }
 667 EXPORT_SYMBOL(netpoll_print_options);
 668
 669 int netpoll_parse_options(struct netpoll *np, char *opt)
 670 {
 671         char *cur=opt, *delim;
 672
 673         if (*cur != '@') {
 674                 if ((delim = strchr(cur, '@')) == NULL)
 675                         goto parse_failed;
 676                 *delim = 0;
 677                 np->local_port = simple_strtol(cur, NULL, 10);
 678                 cur = delim;
 679         }
 680         cur++;
 681
 682         if (*cur != '/') {
 683                 if ((delim = strchr(cur, '/')) == NULL)
 684                         goto parse_failed;
 685                 *delim = 0;
 686                 np->local_ip = in_aton(cur);
 687                 cur = delim;
 688         }
 689         cur++;
 690
 691         if (*cur != ',') {
 692                 /* parse out dev name */
 693                 if ((delim = strchr(cur, ',')) == NULL)
 694                         goto parse_failed;
 695                 *delim = 0;
 696                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 697                 cur = delim;
 698         }
 699         cur++;
 700
 701         if (*cur != '@') {
 702                 /* dst port */
 703                 if ((delim = strchr(cur, '@')) == NULL)
 704                         goto parse_failed;
 705                 *delim = 0;
 706                 if (*cur == ' ' || *cur == '\t')
 707                         np_info(np, "warning: whitespace is not allowed\n");
 708                 np->remote_port = simple_strtol(cur, NULL, 10);
 709                 cur = delim;
 710         }
 711         cur++;
 712
 713         /* dst ip */
 714         if ((delim = strchr(cur, '/')) == NULL)
 715                 goto parse_failed;
 716         *delim = 0;
 717         np->remote_ip = in_aton(cur);
 718         cur = delim + 1;
 719
 720         if (*cur != 0) {
 721                 /* MAC address */
 722                 if (!mac_pton(cur, np->remote_mac))
 723                         goto parse_failed;
 724         }
 725
 726         netpoll_print_options(np);
 727
 728         return 0;
 729
 730  parse_failed:
 731         np_info(np, "couldn't parse config at '%s'!\n", cur);
 732         return -1;
 733 }
 734 EXPORT_SYMBOL(netpoll_parse_options);
 735
 736 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
 737 {
 738         struct netpoll_info *npinfo;
 739         const struct net_device_ops *ops;
 740         unsigned long flags;
 741         int err;
 742
 743         np->dev = ndev;
 744         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 745
 746         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 747             !ndev->netdev_ops->ndo_poll_controller) {
 748                 np_err(np, "%s doesn't support polling, aborting\n",
 749                        np->dev_name);
 750                 err = -ENOTSUPP;
 751                 goto out;
 752         }
 753
 754         if (!ndev->npinfo) {
 755                 npinfo = kmalloc(sizeof(*npinfo), gfp);
 756                 if (!npinfo) {
 757                         err = -ENOMEM;
 758                         goto out;
 759                 }
 760
 761                 npinfo->rx_flags = 0;
 762                 INIT_LIST_HEAD(&npinfo->rx_np);
 763
 764                 spin_lock_init(&npinfo->rx_lock);
 765                 skb_queue_head_init(&npinfo->arp_tx);
 766                 skb_queue_head_init(&npinfo->txq);
 767                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 768
 769                 atomic_set(&npinfo->refcnt, 1);
 770
 771                 ops = np->dev->netdev_ops;
 772                 if (ops->ndo_netpoll_setup) {
 773                         err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
 774                         if (err)
 775                                 goto free_npinfo;
 776                 }
 777         } else {
 778                 npinfo = ndev->npinfo;
 779                 atomic_inc(&npinfo->refcnt);
 780         }
 781
 782         npinfo->netpoll = np;
 783
 784         if (np->rx_hook) {
 785                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 786                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 787                 list_add_tail(&np->rx, &npinfo->rx_np);
 788                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 789         }
 790
 791         /* last thing to do is link it to the net device structure */
 792         rcu_assign_pointer(ndev->npinfo, npinfo);
 793
 794         return 0;
 795
 796 free_npinfo:
 797         kfree(npinfo);
 798 out:
 799         return err;
 800 }
 801 EXPORT_SYMBOL_GPL(__netpoll_setup);
 802
 803 int netpoll_setup(struct netpoll *np)
 804 {
 805         struct net_device *ndev = NULL;
 806         struct in_device *in_dev;
 807         int err;
 808
 809         if (np->dev_name)
 810                 ndev = dev_get_by_name(&init_net, np->dev_name);
 811         if (!ndev) {
 812                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 813                 return -ENODEV;
 814         }
 815
 816         if (ndev->master) {
 817                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 818                 err = -EBUSY;
 819                 goto put;
 820         }
 821
 822         if (!netif_running(ndev)) {
 823                 unsigned long atmost, atleast;
 824
 825                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 826
 827                 rtnl_lock();
 828                 err = dev_open(ndev);
 829                 rtnl_unlock();
 830
 831                 if (err) {
 832                         np_err(np, "failed to open %s\n", ndev->name);
 833                         goto put;
 834                 }
 835
 836                 atleast = jiffies + HZ/10;
 837                 atmost = jiffies + carrier_timeout * HZ;
 838                 while (!netif_carrier_ok(ndev)) {
 839                         if (time_after(jiffies, atmost)) {
 840                                 np_notice(np, "timeout waiting for carrier\n");
 841                                 break;
 842                         }
 843                         msleep(1);
 844                 }
 845
 846                 /* If carrier appears to come up instantly, we don't
 847                  * trust it and pause so that we don't pump all our
 848                  * queued console messages into the bitbucket.
 849                  */
 850
 851                 if (time_before(jiffies, atleast)) {
 852                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 853                         msleep(4000);
 854                 }
 855         }
 856
 857         if (!np->local_ip) {
 858                 rcu_read_lock();
 859                 in_dev = __in_dev_get_rcu(ndev);
 860
 861                 if (!in_dev || !in_dev->ifa_list) {
 862                         rcu_read_unlock();
 863                         np_err(np, "no IP address for %s, aborting\n",
 864                                np->dev_name);
 865                         err = -EDESTADDRREQ;
 866                         goto put;
 867                 }
 868
 869                 np->local_ip = in_dev->ifa_list->ifa_local;
 870                 rcu_read_unlock();
 871                 np_info(np, "local IP %pI4\n", &np->local_ip);
 872         }
 873
 874         /* fill up the skb queue */
 875         refill_skbs();
 876
 877         rtnl_lock();
 878         err = __netpoll_setup(np, ndev, GFP_KERNEL);
 879         rtnl_unlock();
 880
 881         if (err)
 882                 goto put;
 883
 884         return 0;
 885
 886 put:
 887         dev_put(ndev);
 888         return err;
 889 }
 890 EXPORT_SYMBOL(netpoll_setup);
 891
 892 static int __init netpoll_init(void)
 893 {
 894         skb_queue_head_init(&skb_pool);
 895         return 0;
 896 }
 897 core_initcall(netpoll_init);
 898
 899 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 900 {
 901         struct netpoll_info *npinfo =
 902                         container_of(rcu_head, struct netpoll_info, rcu);
 903
 904         skb_queue_purge(&npinfo->arp_tx);
 905         skb_queue_purge(&npinfo->txq);
 906
 907         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 908         cancel_delayed_work(&npinfo->tx_work);
 909
 910         /* clean after last, unfinished work */
 911         __skb_queue_purge(&npinfo->txq);
 912         /* now cancel it again */
 913         cancel_delayed_work(&npinfo->tx_work);
 914         kfree(npinfo);
 915 }
 916
 917 void __netpoll_cleanup(struct netpoll *np)
 918 {
 919         struct netpoll_info *npinfo;
 920         unsigned long flags;
 921
 922         npinfo = np->dev->npinfo;
 923         if (!npinfo)
 924                 return;
 925
 926         if (!list_empty(&npinfo->rx_np)) {
 927                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 928                 list_del(&np->rx);
 929                 if (list_empty(&npinfo->rx_np))
 930                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 931                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 932         }
 933
 934         if (atomic_dec_and_test(&npinfo->refcnt)) {
 935                 const struct net_device_ops *ops;
 936
 937                 ops = np->dev->netdev_ops;
 938                 if (ops->ndo_netpoll_cleanup)
 939                         ops->ndo_netpoll_cleanup(np->dev);
 940
 941                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 942                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
 943         }
 944 }
 945 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 946
 947 static void rcu_cleanup_netpoll(struct rcu_head *rcu_head)
 948 {
 949         struct netpoll *np = container_of(rcu_head, struct netpoll, rcu);
 950
 951         __netpoll_cleanup(np);
 952         kfree(np);
 953 }
 954
 955 void __netpoll_free_rcu(struct netpoll *np)
 956 {
 957         call_rcu_bh(&np->rcu, rcu_cleanup_netpoll);
 958 }
 959 EXPORT_SYMBOL_GPL(__netpoll_free_rcu);
 960
 961 void netpoll_cleanup(struct netpoll *np)
 962 {
 963         if (!np->dev)
 964                 return;
 965
 966         rtnl_lock();
 967         __netpoll_cleanup(np);
 968         rtnl_unlock();
 969
 970         dev_put(np->dev);
 971         np->dev = NULL;
 972 }
 973 EXPORT_SYMBOL(netpoll_cleanup);
 974
 975 int netpoll_trap(void)
 976 {
 977         return atomic_read(&trapped);
 978 }
 979 EXPORT_SYMBOL(netpoll_trap);
 980
 981 void netpoll_set_trap(int trap)
 982 {
 983         if (trap)
 984                 atomic_inc(&trapped);
 985         else
 986                 atomic_dec(&trapped);
 987 }
 988 EXPORT_SYMBOL(netpoll_set_trap);