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         WARN_ON_ONCE(!irqs_disabled());
 172
 173         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 174                 local_irq_enable();
 175                 if (napi->poll_owner != smp_processor_id() &&
 176                     spin_trylock(&napi->poll_lock)) {
 177                         rcu_read_lock_bh();
 178                         budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
 179                                                napi, budget);
 180                         rcu_read_unlock_bh();
 181                         spin_unlock(&napi->poll_lock);
 182
 183                         if (!budget) {
 184                                 local_irq_disable();
 185                                 break;
 186                         }
 187                 }
 188                 local_irq_disable();
 189         }
 190 }
 191
 192 static void service_arp_queue(struct netpoll_info *npi)
 193 {
 194         if (npi) {
 195                 struct sk_buff *skb;
 196
 197                 while ((skb = skb_dequeue(&npi->arp_tx)))
 198                         netpoll_arp_reply(skb, npi);
 199         }
 200 }
 201
 202 static void netpoll_poll_dev(struct net_device *dev)
 203 {
 204         const struct net_device_ops *ops;
 205         struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 206
 207         if (!dev || !netif_running(dev))
 208                 return;
 209
 210         ops = dev->netdev_ops;
 211         if (!ops->ndo_poll_controller)
 212                 return;
 213
 214         /* Process pending work on NIC */
 215         ops->ndo_poll_controller(dev);
 216
 217         poll_napi(dev);
 218
 219         if (dev->flags & IFF_SLAVE) {
 220                 if (ni) {
 221                         struct net_device *bond_dev = dev->master;
 222                         struct sk_buff *skb;
 223                         struct netpoll_info *bond_ni = rcu_dereference_bh(bond_dev->npinfo);
 224                         while ((skb = skb_dequeue(&ni->arp_tx))) {
 225                                 skb->dev = bond_dev;
 226                                 skb_queue_tail(&bond_ni->arp_tx, skb);
 227                         }
 228                 }
 229         }
 230
 231         service_arp_queue(ni);
 232
 233         zap_completion_queue();
 234 }
 235
 236 static void refill_skbs(void)
 237 {
 238         struct sk_buff *skb;
 239         unsigned long flags;
 240
 241         spin_lock_irqsave(&skb_pool.lock, flags);
 242         while (skb_pool.qlen < MAX_SKBS) {
 243                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 244                 if (!skb)
 245                         break;
 246
 247                 __skb_queue_tail(&skb_pool, skb);
 248         }
 249         spin_unlock_irqrestore(&skb_pool.lock, flags);
 250 }
 251
 252 static void zap_completion_queue(void)
 253 {
 254         unsigned long flags;
 255         struct softnet_data *sd = &get_cpu_var(softnet_data);
 256
 257         if (sd->completion_queue) {
 258                 struct sk_buff *clist;
 259
 260                 local_irq_save(flags);
 261                 clist = sd->completion_queue;
 262                 sd->completion_queue = NULL;
 263                 local_irq_restore(flags);
 264
 265                 while (clist != NULL) {
 266                         struct sk_buff *skb = clist;
 267                         clist = clist->next;
 268                         if (skb->destructor) {
 269                                 atomic_inc(&skb->users);
 270                                 dev_kfree_skb_any(skb); /* put this one back */
 271                         } else {
 272                                 __kfree_skb(skb);
 273                         }
 274                 }
 275         }
 276
 277         put_cpu_var(softnet_data);
 278 }
 279
 280 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 281 {
 282         int count = 0;
 283         struct sk_buff *skb;
 284
 285         zap_completion_queue();
 286         refill_skbs();
 287 repeat:
 288
 289         skb = alloc_skb(len, GFP_ATOMIC);
 290         if (!skb)
 291                 skb = skb_dequeue(&skb_pool);
 292
 293         if (!skb) {
 294                 if (++count < 10) {
 295                         netpoll_poll_dev(np->dev);
 296                         goto repeat;
 297                 }
 298                 return NULL;
 299         }
 300
 301         atomic_set(&skb->users, 1);
 302         skb_reserve(skb, reserve);
 303         return skb;
 304 }
 305
 306 static int netpoll_owner_active(struct net_device *dev)
 307 {
 308         struct napi_struct *napi;
 309
 310         list_for_each_entry(napi, &dev->napi_list, dev_list) {
 311                 if (napi->poll_owner == smp_processor_id())
 312                         return 1;
 313         }
 314         return 0;
 315 }
 316
 317 /* call with IRQ disabled */
 318 void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 319                              struct net_device *dev)
 320 {
 321         int status = NETDEV_TX_BUSY;
 322         unsigned long tries;
 323         const struct net_device_ops *ops = dev->netdev_ops;
 324         /* It is up to the caller to keep npinfo alive. */
 325         struct netpoll_info *npinfo;
 326
 327         WARN_ON_ONCE(!irqs_disabled());
 328
 329         npinfo = rcu_dereference_bh(np->dev->npinfo);
 330         if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 331                 __kfree_skb(skb);
 332                 return;
 333         }
 334
 335         /* don't get messages out of order, and no recursion */
 336         if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 337                 struct netdev_queue *txq;
 338
 339                 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 340
 341                 /* try until next clock tick */
 342                 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 343                      tries > 0; --tries) {
 344                         if (__netif_tx_trylock(txq)) {
 345                                 if (!netif_xmit_stopped(txq)) {
 346                                         if (vlan_tx_tag_present(skb) &&
 347                                             !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) {
 348                                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 349                                                 if (unlikely(!skb))
 350                                                         break;
 351                                                 skb->vlan_tci = 0;
 352                                         }
 353
 354                                         status = ops->ndo_start_xmit(skb, dev);
 355                                         if (status == NETDEV_TX_OK)
 356                                                 txq_trans_update(txq);
 357                                 }
 358                                 __netif_tx_unlock(txq);
 359
 360                                 if (status == NETDEV_TX_OK)
 361                                         break;
 362
 363                         }
 364
 365                         /* tickle device maybe there is some cleanup */
 366                         netpoll_poll_dev(np->dev);
 367
 368                         udelay(USEC_PER_POLL);
 369                 }
 370
 371                 WARN_ONCE(!irqs_disabled(),
 372                         "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
 373                         dev->name, ops->ndo_start_xmit);
 374
 375         }
 376
 377         if (status != NETDEV_TX_OK) {
 378                 skb_queue_tail(&npinfo->txq, skb);
 379                 schedule_delayed_work(&npinfo->tx_work,0);
 380         }
 381 }
 382 EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 383
 384 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 385 {
 386         int total_len, ip_len, udp_len;
 387         struct sk_buff *skb;
 388         struct udphdr *udph;
 389         struct iphdr *iph;
 390         struct ethhdr *eth;
 391
 392         udp_len = len + sizeof(*udph);
 393         ip_len = udp_len + sizeof(*iph);
 394         total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 395
 396         skb = find_skb(np, total_len + np->dev->needed_tailroom,
 397                        total_len - len);
 398         if (!skb)
 399                 return;
 400
 401         skb_copy_to_linear_data(skb, msg, len);
 402         skb_put(skb, len);
 403
 404         skb_push(skb, sizeof(*udph));
 405         skb_reset_transport_header(skb);
 406         udph = udp_hdr(skb);
 407         udph->source = htons(np->local_port);
 408         udph->dest = htons(np->remote_port);
 409         udph->len = htons(udp_len);
 410         udph->check = 0;
 411         udph->check = csum_tcpudp_magic(np->local_ip,
 412                                         np->remote_ip,
 413                                         udp_len, IPPROTO_UDP,
 414                                         csum_partial(udph, udp_len, 0));
 415         if (udph->check == 0)
 416                 udph->check = CSUM_MANGLED_0;
 417
 418         skb_push(skb, sizeof(*iph));
 419         skb_reset_network_header(skb);
 420         iph = ip_hdr(skb);
 421
 422         /* iph->version = 4; iph->ihl = 5; */
 423         put_unaligned(0x45, (unsigned char *)iph);
 424         iph->tos      = 0;
 425         put_unaligned(htons(ip_len), &(iph->tot_len));
 426         iph->id       = 0;
 427         iph->frag_off = 0;
 428         iph->ttl      = 64;
 429         iph->protocol = IPPROTO_UDP;
 430         iph->check    = 0;
 431         put_unaligned(np->local_ip, &(iph->saddr));
 432         put_unaligned(np->remote_ip, &(iph->daddr));
 433         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 434
 435         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 436         skb_reset_mac_header(skb);
 437         skb->protocol = eth->h_proto = htons(ETH_P_IP);
 438         memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 439         memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 440
 441         skb->dev = np->dev;
 442
 443         netpoll_send_skb(np, skb);
 444 }
 445 EXPORT_SYMBOL(netpoll_send_udp);
 446
 447 static void netpoll_arp_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
 448 {
 449         struct arphdr *arp;
 450         unsigned char *arp_ptr;
 451         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 452         __be32 sip, tip;
 453         unsigned char *sha;
 454         struct sk_buff *send_skb;
 455         struct netpoll *np, *tmp;
 456         unsigned long flags;
 457         int hlen, tlen;
 458         int hits = 0;
 459
 460         if (list_empty(&npinfo->rx_np))
 461                 return;
 462
 463         /* Before checking the packet, we do some early
 464            inspection whether this is interesting at all */
 465         spin_lock_irqsave(&npinfo->rx_lock, flags);
 466         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 467                 if (np->dev == skb->dev)
 468                         hits++;
 469         }
 470         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 471
 472         /* No netpoll struct is using this dev */
 473         if (!hits)
 474                 return;
 475
 476         /* No arp on this interface */
 477         if (skb->dev->flags & IFF_NOARP)
 478                 return;
 479
 480         if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 481                 return;
 482
 483         skb_reset_network_header(skb);
 484         skb_reset_transport_header(skb);
 485         arp = arp_hdr(skb);
 486
 487         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 488              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 489             arp->ar_pro != htons(ETH_P_IP) ||
 490             arp->ar_op != htons(ARPOP_REQUEST))
 491                 return;
 492
 493         arp_ptr = (unsigned char *)(arp+1);
 494         /* save the location of the src hw addr */
 495         sha = arp_ptr;
 496         arp_ptr += skb->dev->addr_len;
 497         memcpy(&sip, arp_ptr, 4);
 498         arp_ptr += 4;
 499         /* If we actually cared about dst hw addr,
 500            it would get copied here */
 501         arp_ptr += skb->dev->addr_len;
 502         memcpy(&tip, arp_ptr, 4);
 503
 504         /* Should we ignore arp? */
 505         if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 506                 return;
 507
 508         size = arp_hdr_len(skb->dev);
 509
 510         spin_lock_irqsave(&npinfo->rx_lock, flags);
 511         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 512                 if (tip != np->local_ip)
 513                         continue;
 514
 515                 hlen = LL_RESERVED_SPACE(np->dev);
 516                 tlen = np->dev->needed_tailroom;
 517                 send_skb = find_skb(np, size + hlen + tlen, hlen);
 518                 if (!send_skb)
 519                         continue;
 520
 521                 skb_reset_network_header(send_skb);
 522                 arp = (struct arphdr *) skb_put(send_skb, size);
 523                 send_skb->dev = skb->dev;
 524                 send_skb->protocol = htons(ETH_P_ARP);
 525
 526                 /* Fill the device header for the ARP frame */
 527                 if (dev_hard_header(send_skb, skb->dev, ptype,
 528                                     sha, np->dev->dev_addr,
 529                                     send_skb->len) < 0) {
 530                         kfree_skb(send_skb);
 531                         continue;
 532                 }
 533
 534                 /*
 535                  * Fill out the arp protocol part.
 536                  *
 537                  * we only support ethernet device type,
 538                  * which (according to RFC 1390) should
 539                  * always equal 1 (Ethernet).
 540                  */
 541
 542                 arp->ar_hrd = htons(np->dev->type);
 543                 arp->ar_pro = htons(ETH_P_IP);
 544                 arp->ar_hln = np->dev->addr_len;
 545                 arp->ar_pln = 4;
 546                 arp->ar_op = htons(type);
 547
 548                 arp_ptr = (unsigned char *)(arp + 1);
 549                 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 550                 arp_ptr += np->dev->addr_len;
 551                 memcpy(arp_ptr, &tip, 4);
 552                 arp_ptr += 4;
 553                 memcpy(arp_ptr, sha, np->dev->addr_len);
 554                 arp_ptr += np->dev->addr_len;
 555                 memcpy(arp_ptr, &sip, 4);
 556
 557                 netpoll_send_skb(np, send_skb);
 558
 559                 /* If there are several rx_hooks for the same address,
 560                    we're fine by sending a single reply */
 561                 break;
 562         }
 563         spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 564 }
 565
 566 int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
 567 {
 568         int proto, len, ulen;
 569         int hits = 0;
 570         const struct iphdr *iph;
 571         struct udphdr *uh;
 572         struct netpoll *np, *tmp;
 573
 574         if (list_empty(&npinfo->rx_np))
 575                 goto out;
 576
 577         if (skb->dev->type != ARPHRD_ETHER)
 578                 goto out;
 579
 580         /* check if netpoll clients need ARP */
 581         if (skb->protocol == htons(ETH_P_ARP) &&
 582             atomic_read(&trapped)) {
 583                 skb_queue_tail(&npinfo->arp_tx, skb);
 584                 return 1;
 585         }
 586
 587         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
 588                 skb = vlan_untag(skb);
 589                 if (unlikely(!skb))
 590                         goto out;
 591         }
 592
 593         proto = ntohs(eth_hdr(skb)->h_proto);
 594         if (proto != ETH_P_IP)
 595                 goto out;
 596         if (skb->pkt_type == PACKET_OTHERHOST)
 597                 goto out;
 598         if (skb_shared(skb))
 599                 goto out;
 600
 601         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 602                 goto out;
 603         iph = (struct iphdr *)skb->data;
 604         if (iph->ihl < 5 || iph->version != 4)
 605                 goto out;
 606         if (!pskb_may_pull(skb, iph->ihl*4))
 607                 goto out;
 608         iph = (struct iphdr *)skb->data;
 609         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 610                 goto out;
 611
 612         len = ntohs(iph->tot_len);
 613         if (skb->len < len || len < iph->ihl*4)
 614                 goto out;
 615
 616         /*
 617          * Our transport medium may have padded the buffer out.
 618          * Now We trim to the true length of the frame.
 619          */
 620         if (pskb_trim_rcsum(skb, len))
 621                 goto out;
 622
 623         iph = (struct iphdr *)skb->data;
 624         if (iph->protocol != IPPROTO_UDP)
 625                 goto out;
 626
 627         len -= iph->ihl*4;
 628         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 629         ulen = ntohs(uh->len);
 630
 631         if (ulen != len)
 632                 goto out;
 633         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 634                 goto out;
 635
 636         list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 637                 if (np->local_ip && np->local_ip != iph->daddr)
 638                         continue;
 639                 if (np->remote_ip && np->remote_ip != iph->saddr)
 640                         continue;
 641                 if (np->local_port && np->local_port != ntohs(uh->dest))
 642                         continue;
 643
 644                 np->rx_hook(np, ntohs(uh->source),
 645                                (char *)(uh+1),
 646                                ulen - sizeof(struct udphdr));
 647                 hits++;
 648         }
 649
 650         if (!hits)
 651                 goto out;
 652
 653         kfree_skb(skb);
 654         return 1;
 655
 656 out:
 657         if (atomic_read(&trapped)) {
 658                 kfree_skb(skb);
 659                 return 1;
 660         }
 661
 662         return 0;
 663 }
 664
 665 void netpoll_print_options(struct netpoll *np)
 666 {
 667         np_info(np, "local port %d\n", np->local_port);
 668         np_info(np, "local IP %pI4\n", &np->local_ip);
 669         np_info(np, "interface '%s'\n", np->dev_name);
 670         np_info(np, "remote port %d\n", np->remote_port);
 671         np_info(np, "remote IP %pI4\n", &np->remote_ip);
 672         np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 673 }
 674 EXPORT_SYMBOL(netpoll_print_options);
 675
 676 int netpoll_parse_options(struct netpoll *np, char *opt)
 677 {
 678         char *cur=opt, *delim;
 679
 680         if (*cur != '@') {
 681                 if ((delim = strchr(cur, '@')) == NULL)
 682                         goto parse_failed;
 683                 *delim = 0;
 684                 np->local_port = simple_strtol(cur, NULL, 10);
 685                 cur = delim;
 686         }
 687         cur++;
 688
 689         if (*cur != '/') {
 690                 if ((delim = strchr(cur, '/')) == NULL)
 691                         goto parse_failed;
 692                 *delim = 0;
 693                 np->local_ip = in_aton(cur);
 694                 cur = delim;
 695         }
 696         cur++;
 697
 698         if (*cur != ',') {
 699                 /* parse out dev name */
 700                 if ((delim = strchr(cur, ',')) == NULL)
 701                         goto parse_failed;
 702                 *delim = 0;
 703                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 704                 cur = delim;
 705         }
 706         cur++;
 707
 708         if (*cur != '@') {
 709                 /* dst port */
 710                 if ((delim = strchr(cur, '@')) == NULL)
 711                         goto parse_failed;
 712                 *delim = 0;
 713                 if (*cur == ' ' || *cur == '\t')
 714                         np_info(np, "warning: whitespace is not allowed\n");
 715                 np->remote_port = simple_strtol(cur, NULL, 10);
 716                 cur = delim;
 717         }
 718         cur++;
 719
 720         /* dst ip */
 721         if ((delim = strchr(cur, '/')) == NULL)
 722                 goto parse_failed;
 723         *delim = 0;
 724         np->remote_ip = in_aton(cur);
 725         cur = delim + 1;
 726
 727         if (*cur != 0) {
 728                 /* MAC address */
 729                 if (!mac_pton(cur, np->remote_mac))
 730                         goto parse_failed;
 731         }
 732
 733         netpoll_print_options(np);
 734
 735         return 0;
 736
 737  parse_failed:
 738         np_info(np, "couldn't parse config at '%s'!\n", cur);
 739         return -1;
 740 }
 741 EXPORT_SYMBOL(netpoll_parse_options);
 742
 743 int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
 744 {
 745         struct netpoll_info *npinfo;
 746         const struct net_device_ops *ops;
 747         unsigned long flags;
 748         int err;
 749
 750         np->dev = ndev;
 751         strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 752
 753         if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 754             !ndev->netdev_ops->ndo_poll_controller) {
 755                 np_err(np, "%s doesn't support polling, aborting\n",
 756                        np->dev_name);
 757                 err = -ENOTSUPP;
 758                 goto out;
 759         }
 760
 761         if (!ndev->npinfo) {
 762                 npinfo = kmalloc(sizeof(*npinfo), gfp);
 763                 if (!npinfo) {
 764                         err = -ENOMEM;
 765                         goto out;
 766                 }
 767
 768                 npinfo->rx_flags = 0;
 769                 INIT_LIST_HEAD(&npinfo->rx_np);
 770
 771                 spin_lock_init(&npinfo->rx_lock);
 772                 skb_queue_head_init(&npinfo->arp_tx);
 773                 skb_queue_head_init(&npinfo->txq);
 774                 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 775
 776                 atomic_set(&npinfo->refcnt, 1);
 777
 778                 ops = np->dev->netdev_ops;
 779                 if (ops->ndo_netpoll_setup) {
 780                         err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
 781                         if (err)
 782                                 goto free_npinfo;
 783                 }
 784         } else {
 785                 npinfo = ndev->npinfo;
 786                 atomic_inc(&npinfo->refcnt);
 787         }
 788
 789         npinfo->netpoll = np;
 790
 791         if (np->rx_hook) {
 792                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 793                 npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 794                 list_add_tail(&np->rx, &npinfo->rx_np);
 795                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 796         }
 797
 798         /* last thing to do is link it to the net device structure */
 799         rcu_assign_pointer(ndev->npinfo, npinfo);
 800
 801         return 0;
 802
 803 free_npinfo:
 804         kfree(npinfo);
 805 out:
 806         return err;
 807 }
 808 EXPORT_SYMBOL_GPL(__netpoll_setup);
 809
 810 int netpoll_setup(struct netpoll *np)
 811 {
 812         struct net_device *ndev = NULL;
 813         struct in_device *in_dev;
 814         int err;
 815
 816         if (np->dev_name)
 817                 ndev = dev_get_by_name(&init_net, np->dev_name);
 818         if (!ndev) {
 819                 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 820                 return -ENODEV;
 821         }
 822
 823         if (ndev->master) {
 824                 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 825                 err = -EBUSY;
 826                 goto put;
 827         }
 828
 829         if (!netif_running(ndev)) {
 830                 unsigned long atmost, atleast;
 831
 832                 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 833
 834                 rtnl_lock();
 835                 err = dev_open(ndev);
 836                 rtnl_unlock();
 837
 838                 if (err) {
 839                         np_err(np, "failed to open %s\n", ndev->name);
 840                         goto put;
 841                 }
 842
 843                 atleast = jiffies + HZ/10;
 844                 atmost = jiffies + carrier_timeout * HZ;
 845                 while (!netif_carrier_ok(ndev)) {
 846                         if (time_after(jiffies, atmost)) {
 847                                 np_notice(np, "timeout waiting for carrier\n");
 848                                 break;
 849                         }
 850                         msleep(1);
 851                 }
 852
 853                 /* If carrier appears to come up instantly, we don't
 854                  * trust it and pause so that we don't pump all our
 855                  * queued console messages into the bitbucket.
 856                  */
 857
 858                 if (time_before(jiffies, atleast)) {
 859                         np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 860                         msleep(4000);
 861                 }
 862         }
 863
 864         if (!np->local_ip) {
 865                 rcu_read_lock();
 866                 in_dev = __in_dev_get_rcu(ndev);
 867
 868                 if (!in_dev || !in_dev->ifa_list) {
 869                         rcu_read_unlock();
 870                         np_err(np, "no IP address for %s, aborting\n",
 871                                np->dev_name);
 872                         err = -EDESTADDRREQ;
 873                         goto put;
 874                 }
 875
 876                 np->local_ip = in_dev->ifa_list->ifa_local;
 877                 rcu_read_unlock();
 878                 np_info(np, "local IP %pI4\n", &np->local_ip);
 879         }
 880
 881         /* fill up the skb queue */
 882         refill_skbs();
 883
 884         rtnl_lock();
 885         err = __netpoll_setup(np, ndev, GFP_KERNEL);
 886         rtnl_unlock();
 887
 888         if (err)
 889                 goto put;
 890
 891         return 0;
 892
 893 put:
 894         dev_put(ndev);
 895         return err;
 896 }
 897 EXPORT_SYMBOL(netpoll_setup);
 898
 899 static int __init netpoll_init(void)
 900 {
 901         skb_queue_head_init(&skb_pool);
 902         return 0;
 903 }
 904 core_initcall(netpoll_init);
 905
 906 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 907 {
 908         struct netpoll_info *npinfo =
 909                         container_of(rcu_head, struct netpoll_info, rcu);
 910
 911         skb_queue_purge(&npinfo->arp_tx);
 912         skb_queue_purge(&npinfo->txq);
 913
 914         /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 915         cancel_delayed_work(&npinfo->tx_work);
 916
 917         /* clean after last, unfinished work */
 918         __skb_queue_purge(&npinfo->txq);
 919         /* now cancel it again */
 920         cancel_delayed_work(&npinfo->tx_work);
 921         kfree(npinfo);
 922 }
 923
 924 void __netpoll_cleanup(struct netpoll *np)
 925 {
 926         struct netpoll_info *npinfo;
 927         unsigned long flags;
 928
 929         npinfo = np->dev->npinfo;
 930         if (!npinfo)
 931                 return;
 932
 933         if (!list_empty(&npinfo->rx_np)) {
 934                 spin_lock_irqsave(&npinfo->rx_lock, flags);
 935                 list_del(&np->rx);
 936                 if (list_empty(&npinfo->rx_np))
 937                         npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 938                 spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 939         }
 940
 941         if (atomic_dec_and_test(&npinfo->refcnt)) {
 942                 const struct net_device_ops *ops;
 943
 944                 ops = np->dev->netdev_ops;
 945                 if (ops->ndo_netpoll_cleanup)
 946                         ops->ndo_netpoll_cleanup(np->dev);
 947
 948                 RCU_INIT_POINTER(np->dev->npinfo, NULL);
 949                 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
 950         }
 951 }
 952 EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 953
 954 static void rcu_cleanup_netpoll(struct rcu_head *rcu_head)
 955 {
 956         struct netpoll *np = container_of(rcu_head, struct netpoll, rcu);
 957
 958         __netpoll_cleanup(np);
 959         kfree(np);
 960 }
 961
 962 void __netpoll_free_rcu(struct netpoll *np)
 963 {
 964         call_rcu_bh(&np->rcu, rcu_cleanup_netpoll);
 965 }
 966 EXPORT_SYMBOL_GPL(__netpoll_free_rcu);
 967
 968 void netpoll_cleanup(struct netpoll *np)
 969 {
 970         if (!np->dev)
 971                 return;
 972
 973         rtnl_lock();
 974         __netpoll_cleanup(np);
 975         rtnl_unlock();
 976
 977         dev_put(np->dev);
 978         np->dev = NULL;
 979 }
 980 EXPORT_SYMBOL(netpoll_cleanup);
 981
 982 int netpoll_trap(void)
 983 {
 984         return atomic_read(&trapped);
 985 }
 986 EXPORT_SYMBOL(netpoll_trap);
 987
 988 void netpoll_set_trap(int trap)
 989 {
 990         if (trap)
 991                 atomic_inc(&trapped);
 992         else
 993                 atomic_dec(&trapped);
 994 }
 995 EXPORT_SYMBOL(netpoll_set_trap);