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