net/core/netpoll.c

   1 /*
   2  * Common framework for low-level network console, dump, and debugger code
   3  *
   4  * Sep 8 2003  Matt Mackall <mpm@selenic.com>
   5  *
   6  * based on the netconsole code from:
   7  *
   8  * Copyright (C) 2001  Ingo Molnar <mingo@redhat.com>
   9  * Copyright (C) 2002  Red Hat, Inc.
  10  */
  11
  12 #include <linux/smp_lock.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/etherdevice.h>
  15 #include <linux/string.h>
  16 #include <linux/inetdevice.h>
  17 #include <linux/inet.h>
  18 #include <linux/interrupt.h>
  19 #include <linux/netpoll.h>
  20 #include <linux/sched.h>
  21 #include <linux/nmi.h>
  22 #include <net/tcp.h>
  23 #include <net/udp.h>
  24
  25 /*
  26  * We maintain a small pool of fully-sized skbs, to make sure the
  27  * message gets out even in extreme OOM situations.
  28  */
  29
  30 #define MAX_SKBS 32
  31 #define MAX_UDP_CHUNK 1460
  32
  33 #define NETPOLL_RX_ENABLED  1
  34 #define NETPOLL_RX_DROP     2
  35
  36 static spinlock_t skb_list_lock = SPIN_LOCK_UNLOCKED;
  37 static int nr_skbs;
  38 static struct sk_buff *skbs;
  39
  40 static spinlock_t rx_list_lock = SPIN_LOCK_UNLOCKED;
  41 static LIST_HEAD(rx_list);
  42
  43 static int trapped;
  44
  45 extern void (*netdump_func) (struct pt_regs *regs);
  46
  47 #define MAX_SKB_SIZE \
  48                 (MAX_UDP_CHUNK + sizeof(struct udphdr) + \
  49                                 sizeof(struct iphdr) + sizeof(struct ethhdr))
  50
  51 static void zap_completion_queue(void);
  52
  53 static int checksum_udp(struct sk_buff *skb, struct udphdr *uh,
  54                              unsigned short ulen, u32 saddr, u32 daddr)
  55 {
  56         if (uh->check == 0)
  57                 return 0;
  58
  59         if (skb->ip_summed == CHECKSUM_HW)
  60                 return csum_tcpudp_magic(
  61                         saddr, daddr, ulen, IPPROTO_UDP, skb->csum);
  62
  63         skb->csum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
  64
  65         return csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
  66 }
  67
  68 void netpoll_poll(struct netpoll *np)
  69 {
  70         int budget = netdump_mode ? 64 : 16;
  71
  72         if(!np->dev || !netif_running(np->dev) || !np->dev->poll_controller)
  73                 return;
  74
  75         /* Process pending work on NIC */
  76         np->dev->poll_controller(np->dev);
  77
  78         /* If scheduling is stopped, tickle NAPI bits */
  79         if (np->dev->poll &&
  80             test_bit(__LINK_STATE_RX_SCHED, &np->dev->state)) {
  81                 np->dev->netpoll_rx |= NETPOLL_RX_DROP;
  82                 if (trapped) {
  83                         np->dev->poll(np->dev, &budget);
  84                 } else {
  85                         trapped = 1;
  86                         np->dev->poll(np->dev, &budget);
  87                         trapped = 0;
  88                 }
  89                 np->dev->netpoll_rx &= ~NETPOLL_RX_DROP;
  90         }
  91
  92         zap_completion_queue();
  93 }
  94
  95 static void refill_skbs(void)
  96 {
  97         struct sk_buff *skb;
  98         unsigned long flags;
  99
 100         spin_lock_irqsave(&skb_list_lock, flags);
 101         while (nr_skbs < MAX_SKBS) {
 102                 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 103                 if (!skb)
 104                         break;
 105
 106                 skb->next = skbs;
 107                 skbs = skb;
 108                 nr_skbs++;
 109         }
 110         spin_unlock_irqrestore(&skb_list_lock, flags);
 111 }
 112
 113 static void zap_completion_queue(void)
 114 {
 115         unsigned long flags;
 116         struct softnet_data *sd = &get_cpu_var(softnet_data);
 117
 118         if (sd->completion_queue) {
 119                 struct sk_buff *clist;
 120
 121                 local_irq_save(flags);
 122                 clist = sd->completion_queue;
 123                 sd->completion_queue = NULL;
 124                 local_irq_restore(flags);
 125
 126                 while (clist != NULL) {
 127                         struct sk_buff *skb = clist;
 128                         clist = clist->next;
 129                         __kfree_skb(skb);
 130                 }
 131         }
 132
 133         put_cpu_var(softnet_data);
 134         touch_nmi_watchdog();
 135 }
 136
 137 static struct sk_buff * find_skb(struct netpoll *np, int len, int reserve)
 138 {
 139         int once = 1, count = 0;
 140         unsigned long flags;
 141         struct sk_buff *skb = NULL;
 142
 143         zap_completion_queue();
 144 repeat:
 145         if (nr_skbs < MAX_SKBS)
 146                 refill_skbs();
 147
 148         skb = alloc_skb(len, GFP_ATOMIC);
 149
 150         if (!skb) {
 151                 spin_lock_irqsave(&skb_list_lock, flags);
 152                 skb = skbs;
 153                 if (skb)
 154                         skbs = skb->next;
 155                 skb->next = NULL;
 156                 nr_skbs--;
 157                 spin_unlock_irqrestore(&skb_list_lock, flags);
 158         }
 159
 160         if(!skb) {
 161                 count++;
 162                 if (once && (count == 1000000)) {
 163                         printk("out of netpoll skbs!\n");
 164                         once = 0;
 165                 }
 166                 netpoll_poll(np);
 167                 goto repeat;
 168         }
 169
 170         atomic_set(&skb->users, 1);
 171         skb_reserve(skb, reserve);
 172         return skb;
 173 }
 174
 175 void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 176 {
 177         int status;
 178
 179 repeat:
 180         if(!np || !np->dev || !netif_running(np->dev)) {
 181                 __kfree_skb(skb);
 182                 return;
 183         }
 184
 185         spin_lock(&np->dev->xmit_lock);
 186         np->dev->xmit_lock_owner = smp_processor_id();
 187
 188         status = np->dev->hard_start_xmit(skb, np->dev);
 189         np->dev->xmit_lock_owner = -1;
 190         spin_unlock(&np->dev->xmit_lock);
 191
 192         /* transmit busy */
 193         if(status) {
 194                 netpoll_poll(np);
 195                 goto repeat;
 196         }
 197 }
 198
 199 void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 200 {
 201         int total_len, eth_len, ip_len, udp_len;
 202         struct sk_buff *skb;
 203         struct udphdr *udph;
 204         struct iphdr *iph;
 205         struct ethhdr *eth;
 206
 207         udp_len = len + sizeof(*udph);
 208         ip_len = eth_len = udp_len + sizeof(*iph);
 209         total_len = eth_len + ETH_HLEN;
 210
 211         skb = find_skb(np, total_len, total_len - len);
 212         if (!skb)
 213                 return;
 214
 215         memcpy(skb->data, msg, len);
 216         skb->len += len;
 217
 218         udph = (struct udphdr *) skb_push(skb, sizeof(*udph));
 219         udph->source = htons(np->local_port);
 220         udph->dest = htons(np->remote_port);
 221         udph->len = htons(udp_len);
 222         udph->check = 0;
 223
 224         iph = (struct iphdr *)skb_push(skb, sizeof(*iph));
 225
 226         iph->version  = 4;
 227         iph->ihl      = 5;
 228         iph->tos      = 0;
 229         iph->tot_len  = htons(ip_len);
 230         iph->id       = 0;
 231         iph->frag_off = 0;
 232         iph->ttl      = 64;
 233         iph->protocol = IPPROTO_UDP;
 234         iph->check    = 0;
 235         iph->saddr    = htonl(np->local_ip);
 236         iph->daddr    = htonl(np->remote_ip);
 237         iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 238
 239         eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 240
 241         eth->h_proto = htons(ETH_P_IP);
 242         memcpy(eth->h_source, np->local_mac, 6);
 243         memcpy(eth->h_dest, np->remote_mac, 6);
 244
 245         netpoll_send_skb(np, skb);
 246 }
 247
 248 static void arp_reply(struct sk_buff *skb)
 249 {
 250         struct arphdr *arp;
 251         unsigned char *arp_ptr;
 252         int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 253         u32 sip, tip;
 254         struct sk_buff *send_skb;
 255         unsigned long flags;
 256         struct list_head *p;
 257         struct netpoll *np = NULL;
 258
 259         spin_lock_irqsave(&rx_list_lock, flags);
 260         list_for_each(p, &rx_list) {
 261                 np = list_entry(p, struct netpoll, rx_list);
 262                 if ( np->dev == skb->dev )
 263                         break;
 264                 np = NULL;
 265         }
 266         spin_unlock_irqrestore(&rx_list_lock, flags);
 267
 268         if (!np) return;
 269
 270         /* No arp on this interface */
 271         if (skb->dev->flags & IFF_NOARP)
 272                 return;
 273
 274         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
 275                                  (2 * skb->dev->addr_len) +
 276                                  (2 * sizeof(u32)))))
 277                 return;
 278
 279         skb->h.raw = skb->nh.raw = skb->data;
 280         arp = skb->nh.arph;
 281
 282         if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 283              arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 284             arp->ar_pro != htons(ETH_P_IP) ||
 285             arp->ar_op != htons(ARPOP_REQUEST))
 286                 return;
 287
 288         arp_ptr = (unsigned char *)(arp+1) + skb->dev->addr_len;
 289         memcpy(&sip, arp_ptr, 4);
 290         arp_ptr += 4 + skb->dev->addr_len;
 291         memcpy(&tip, arp_ptr, 4);
 292
 293         /* Should we ignore arp? */
 294         if (tip != htonl(np->local_ip) || LOOPBACK(tip) || MULTICAST(tip))
 295                 return;
 296
 297         size = sizeof(struct arphdr) + 2 * (skb->dev->addr_len + 4);
 298         send_skb = find_skb(np, size + LL_RESERVED_SPACE(np->dev),
 299                             LL_RESERVED_SPACE(np->dev));
 300
 301         if (!send_skb)
 302                 return;
 303
 304         send_skb->nh.raw = send_skb->data;
 305         arp = (struct arphdr *) skb_put(send_skb, size);
 306         send_skb->dev = skb->dev;
 307         send_skb->protocol = htons(ETH_P_ARP);
 308
 309         /* Fill the device header for the ARP frame */
 310
 311         if (np->dev->hard_header &&
 312             np->dev->hard_header(send_skb, skb->dev, ptype,
 313                                        np->remote_mac, np->local_mac,
 314                                        send_skb->len) < 0) {
 315                 kfree_skb(send_skb);
 316                 return;
 317         }
 318
 319         /*
 320          * Fill out the arp protocol part.
 321          *
 322          * we only support ethernet device type,
 323          * which (according to RFC 1390) should always equal 1 (Ethernet).
 324          */
 325
 326         arp->ar_hrd = htons(np->dev->type);
 327         arp->ar_pro = htons(ETH_P_IP);
 328         arp->ar_hln = np->dev->addr_len;
 329         arp->ar_pln = 4;
 330         arp->ar_op = htons(type);
 331
 332         arp_ptr=(unsigned char *)(arp + 1);
 333         memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 334         arp_ptr += np->dev->addr_len;
 335         memcpy(arp_ptr, &tip, 4);
 336         arp_ptr += 4;
 337         memcpy(arp_ptr, np->remote_mac, np->dev->addr_len);
 338         arp_ptr += np->dev->addr_len;
 339         memcpy(arp_ptr, &sip, 4);
 340
 341         netpoll_send_skb(np, send_skb);
 342 }
 343
 344 int netpoll_rx(struct sk_buff *skb)
 345 {
 346         int proto, len, ulen;
 347         struct iphdr *iph;
 348         struct udphdr *uh;
 349         struct netpoll *np;
 350         struct list_head *p;
 351         unsigned long flags;
 352
 353         if (!(skb->dev->netpoll_rx & NETPOLL_RX_ENABLED))
 354                 return 1;
 355
 356         if (skb->dev->type != ARPHRD_ETHER)
 357                 goto out;
 358
 359         /* check if netpoll clients need ARP */
 360         if (skb->protocol == __constant_htons(ETH_P_ARP) && trapped) {
 361                 arp_reply(skb);
 362                 return 1;
 363         }
 364
 365         proto = ntohs(skb->mac.ethernet->h_proto);
 366         if (proto != ETH_P_IP)
 367                 goto out;
 368         if (skb->pkt_type == PACKET_OTHERHOST)
 369                 goto out;
 370         if (skb_shared(skb))
 371                 goto out;
 372
 373         iph = (struct iphdr *)skb->data;
 374         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 375                 goto out;
 376         if (iph->ihl < 5 || iph->version != 4)
 377                 goto out;
 378         if (!pskb_may_pull(skb, iph->ihl*4))
 379                 goto out;
 380         if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 381                 goto out;
 382
 383         len = ntohs(iph->tot_len);
 384         if (skb->len < len || len < iph->ihl*4)
 385                 goto out;
 386
 387         if (iph->protocol != IPPROTO_UDP)
 388                 goto out;
 389
 390         len -= iph->ihl*4;
 391         uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 392         ulen = ntohs(uh->len);
 393
 394         if (ulen != len)
 395                 goto out;
 396         if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr) < 0)
 397                 goto out;
 398
 399         spin_lock_irqsave(&rx_list_lock, flags);
 400         list_for_each(p, &rx_list) {
 401                 np = list_entry(p, struct netpoll, rx_list);
 402                 if (np->dev && np->dev != skb->dev)
 403                         continue;
 404                 if (np->local_ip && np->local_ip != ntohl(iph->daddr))
 405                         continue;
 406                 if (np->remote_ip && np->remote_ip != ntohl(iph->saddr))
 407                         continue;
 408                 if (np->local_port && np->local_port != ntohs(uh->dest))
 409                         continue;
 410
 411                 spin_unlock_irqrestore(&rx_list_lock, flags);
 412
 413                 if (np->rx_hook)
 414                         np->rx_hook(np, ntohs(uh->source),
 415                                     (char *)(uh+1),
 416                                     ulen - sizeof(struct udphdr));
 417
 418                 return 1;
 419         }
 420         spin_unlock_irqrestore(&rx_list_lock, flags);
 421
 422 out:
 423         return trapped;
 424 }
 425
 426 int netpoll_parse_options(struct netpoll *np, char *opt)
 427 {
 428         char *cur=opt, *delim;
 429
 430         if(*cur != '@') {
 431                 if ((delim = strchr(cur, '@')) == NULL)
 432                         goto parse_failed;
 433                 *delim=0;
 434                 np->local_port=simple_strtol(cur, NULL, 10);
 435                 cur=delim;
 436         }
 437         cur++;
 438         printk(KERN_INFO "%s: local port %d\n", np->name, np->local_port);
 439
 440         if(*cur != '/') {
 441                 if ((delim = strchr(cur, '/')) == NULL)
 442                         goto parse_failed;
 443                 *delim=0;
 444                 np->local_ip=ntohl(in_aton(cur));
 445                 cur=delim;
 446
 447                 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 448                        np->name, HIPQUAD(np->local_ip));
 449         }
 450         cur++;
 451
 452         if ( *cur != ',') {
 453                 /* parse out dev name */
 454                 if ((delim = strchr(cur, ',')) == NULL)
 455                         goto parse_failed;
 456                 *delim=0;
 457                 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 458                 cur=delim;
 459         }
 460         cur++;
 461
 462         printk(KERN_INFO "%s: interface %s\n", np->name, np->dev_name);
 463
 464         if ( *cur != '@' ) {
 465                 /* dst port */
 466                 if ((delim = strchr(cur, '@')) == NULL)
 467                         goto parse_failed;
 468                 *delim=0;
 469                 np->remote_port=simple_strtol(cur, NULL, 10);
 470                 cur=delim;
 471         }
 472         cur++;
 473         printk(KERN_INFO "%s: remote port %d\n", np->name, np->remote_port);
 474
 475         /* dst ip */
 476         if ((delim = strchr(cur, '/')) == NULL)
 477                 goto parse_failed;
 478         *delim=0;
 479         np->remote_ip=ntohl(in_aton(cur));
 480         cur=delim+1;
 481
 482         printk(KERN_INFO "%s: remote IP %d.%d.%d.%d\n",
 483                        np->name, HIPQUAD(np->remote_ip));
 484
 485         if( *cur != 0 )
 486         {
 487                 /* MAC address */
 488                 if ((delim = strchr(cur, ':')) == NULL)
 489                         goto parse_failed;
 490                 *delim=0;
 491                 np->remote_mac[0]=simple_strtol(cur, NULL, 16);
 492                 cur=delim+1;
 493                 if ((delim = strchr(cur, ':')) == NULL)
 494                         goto parse_failed;
 495                 *delim=0;
 496                 np->remote_mac[1]=simple_strtol(cur, NULL, 16);
 497                 cur=delim+1;
 498                 if ((delim = strchr(cur, ':')) == NULL)
 499                         goto parse_failed;
 500                 *delim=0;
 501                 np->remote_mac[2]=simple_strtol(cur, NULL, 16);
 502                 cur=delim+1;
 503                 if ((delim = strchr(cur, ':')) == NULL)
 504                         goto parse_failed;
 505                 *delim=0;
 506                 np->remote_mac[3]=simple_strtol(cur, NULL, 16);
 507                 cur=delim+1;
 508                 if ((delim = strchr(cur, ':')) == NULL)
 509                         goto parse_failed;
 510                 *delim=0;
 511                 np->remote_mac[4]=simple_strtol(cur, NULL, 16);
 512                 cur=delim+1;
 513                 np->remote_mac[5]=simple_strtol(cur, NULL, 16);
 514         }
 515
 516         printk(KERN_INFO "%s: remote ethernet address "
 517                "%02x:%02x:%02x:%02x:%02x:%02x\n",
 518                np->name,
 519                np->remote_mac[0],
 520                np->remote_mac[1],
 521                np->remote_mac[2],
 522                np->remote_mac[3],
 523                np->remote_mac[4],
 524                np->remote_mac[5]);
 525
 526         return 0;
 527
 528  parse_failed:
 529         printk(KERN_INFO "%s: couldn't parse config at %s!\n",
 530                np->name, cur);
 531         return -1;
 532 }
 533
 534 int netpoll_setup(struct netpoll *np)
 535 {
 536         struct net_device *ndev = NULL;
 537         struct in_device *in_dev;
 538
 539         if (np->dev_name)
 540                 ndev = dev_get_by_name(np->dev_name);
 541         if (!ndev) {
 542                 printk(KERN_ERR "%s: %s doesn't exist, aborting.\n",
 543                        np->name, np->dev_name);
 544                 return -1;
 545         }
 546         if (!ndev->poll_controller) {
 547                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
 548                        np->name, np->dev_name);
 549                 goto release;
 550         }
 551
 552         if (!(ndev->flags & IFF_UP)) {
 553                 unsigned short oflags;
 554                 unsigned long atmost, atleast;
 555
 556                 printk(KERN_INFO "%s: device %s not up yet, forcing it\n",
 557                        np->name, np->dev_name);
 558
 559                 oflags = ndev->flags;
 560
 561                 rtnl_shlock();
 562                 if (dev_change_flags(ndev, oflags | IFF_UP) < 0) {
 563                         printk(KERN_ERR "%s: failed to open %s\n",
 564                                np->name, np->dev_name);
 565                         rtnl_shunlock();
 566                         goto release;
 567                 }
 568                 rtnl_shunlock();
 569
 570                 atleast = jiffies + HZ/10;
 571                 atmost = jiffies + 10*HZ;
 572                 while (!netif_carrier_ok(ndev)) {
 573                         if (time_after(jiffies, atmost)) {
 574                                 printk(KERN_NOTICE
 575                                        "%s: timeout waiting for carrier\n",
 576                                        np->name);
 577                                 break;
 578                         }
 579                         cond_resched();
 580                 }
 581
 582                 if (time_before(jiffies, atleast)) {
 583                         printk(KERN_NOTICE "%s: carrier detect appears flaky,"
 584                                " waiting 10 seconds\n",
 585                                np->name);
 586                         while (time_before(jiffies, atmost))
 587                                 cond_resched();
 588                 }
 589         }
 590
 591         if (!memcmp(np->local_mac, "\0\0\0\0\0\0", 6) && ndev->dev_addr)
 592                 memcpy(np->local_mac, ndev->dev_addr, 6);
 593
 594         if (!np->local_ip) {
 595                 in_dev = in_dev_get(ndev);
 596
 597                 if (!in_dev) {
 598                         printk(KERN_ERR "%s: no IP address for %s, aborting\n",
 599                                np->name, np->dev_name);
 600                         goto release;
 601                 }
 602
 603                 np->local_ip = ntohl(in_dev->ifa_list->ifa_local);
 604                 in_dev_put(in_dev);
 605                 printk(KERN_INFO "%s: local IP %d.%d.%d.%d\n",
 606                        np->name, HIPQUAD(np->local_ip));
 607         }
 608
 609         np->dev = ndev;
 610
 611         if(np->rx_hook) {
 612                 unsigned long flags;
 613
 614                 np->dev->netpoll_rx = NETPOLL_RX_ENABLED;
 615
 616                 spin_lock_irqsave(&rx_list_lock, flags);
 617                 list_add(&np->rx_list, &rx_list);
 618                 spin_unlock_irqrestore(&rx_list_lock, flags);
 619         }
 620
 621         if(np->dump_func)
 622                 netdump_func = np->dump_func;
 623
 624         return 0;
 625  release:
 626         dev_put(ndev);
 627         return -1;
 628 }
 629
 630 void netpoll_cleanup(struct netpoll *np)
 631 {
 632         if(np->rx_hook) {
 633                 unsigned long flags;
 634
 635                 spin_lock_irqsave(&rx_list_lock, flags);
 636                 list_del(&np->rx_list);
 637                 np->dev->netpoll_rx = 0;
 638                 spin_unlock_irqrestore(&rx_list_lock, flags);
 639         }
 640
 641         dev_put(np->dev);
 642         np->dev = NULL;
 643 }
 644
 645 int netpoll_trap(void)
 646 {
 647         return trapped;
 648 }
 649
 650 void netpoll_set_trap(int trap)
 651 {
 652         trapped = trap;
 653 }
 654
 655 void netpoll_reset_locks(struct netpoll *np)
 656 {
 657         spin_lock_init(&rx_list_lock);
 658         spin_lock_init(&skb_list_lock);
 659         spin_lock_init(&np->dev->xmit_lock);
 660 }
 661
 662 EXPORT_SYMBOL(netpoll_set_trap);
 663 EXPORT_SYMBOL(netpoll_trap);
 664 EXPORT_SYMBOL(netpoll_parse_options);
 665 EXPORT_SYMBOL(netpoll_setup);
 666 EXPORT_SYMBOL(netpoll_cleanup);
 667 EXPORT_SYMBOL(netpoll_send_skb);
 668 EXPORT_SYMBOL(netpoll_send_udp);
 669 EXPORT_SYMBOL(netpoll_poll);
 670 EXPORT_SYMBOL_GPL(netpoll_reset_locks);