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