linux/net/core/netpoll.c
<<
>>
Prefs
   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
  43static struct sk_buff_head skb_pool;
  44
  45static 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
  57static void zap_completion_queue(void);
  58static void netpoll_arp_reply(struct sk_buff *skb, struct netpoll_info *npinfo);
  59
  60static unsigned int carrier_timeout = 4;
  61module_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
  70static 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
 105static __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 */
 140static 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
 166static void poll_napi(struct net_device *dev)
 167{
 168        struct napi_struct *napi;
 169        int budget = 16;
 170
 171        list_for_each_entry(napi, &dev->napi_list, dev_list) {
 172                if (napi->poll_owner != smp_processor_id() &&
 173                    spin_trylock(&napi->poll_lock)) {
 174                        budget = poll_one_napi(rcu_dereference_bh(dev->npinfo),
 175                                               napi, budget);
 176                        spin_unlock(&napi->poll_lock);
 177
 178                        if (!budget)
 179                                break;
 180                }
 181        }
 182}
 183
 184static void service_arp_queue(struct netpoll_info *npi)
 185{
 186        if (npi) {
 187                struct sk_buff *skb;
 188
 189                while ((skb = skb_dequeue(&npi->arp_tx)))
 190                        netpoll_arp_reply(skb, npi);
 191        }
 192}
 193
 194static void netpoll_poll_dev(struct net_device *dev)
 195{
 196        const struct net_device_ops *ops;
 197        struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
 198
 199        if (!dev || !netif_running(dev))
 200                return;
 201
 202        ops = dev->netdev_ops;
 203        if (!ops->ndo_poll_controller)
 204                return;
 205
 206        /* Process pending work on NIC */
 207        ops->ndo_poll_controller(dev);
 208
 209        poll_napi(dev);
 210
 211        if (dev->flags & IFF_SLAVE) {
 212                if (ni) {
 213                        struct net_device *bond_dev = dev->master;
 214                        struct sk_buff *skb;
 215                        struct netpoll_info *bond_ni = rcu_dereference_bh(bond_dev->npinfo);
 216                        while ((skb = skb_dequeue(&ni->arp_tx))) {
 217                                skb->dev = bond_dev;
 218                                skb_queue_tail(&bond_ni->arp_tx, skb);
 219                        }
 220                }
 221        }
 222
 223        service_arp_queue(ni);
 224
 225        zap_completion_queue();
 226}
 227
 228static void refill_skbs(void)
 229{
 230        struct sk_buff *skb;
 231        unsigned long flags;
 232
 233        spin_lock_irqsave(&skb_pool.lock, flags);
 234        while (skb_pool.qlen < MAX_SKBS) {
 235                skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
 236                if (!skb)
 237                        break;
 238
 239                __skb_queue_tail(&skb_pool, skb);
 240        }
 241        spin_unlock_irqrestore(&skb_pool.lock, flags);
 242}
 243
 244static void zap_completion_queue(void)
 245{
 246        unsigned long flags;
 247        struct softnet_data *sd = &get_cpu_var(softnet_data);
 248
 249        if (sd->completion_queue) {
 250                struct sk_buff *clist;
 251
 252                local_irq_save(flags);
 253                clist = sd->completion_queue;
 254                sd->completion_queue = NULL;
 255                local_irq_restore(flags);
 256
 257                while (clist != NULL) {
 258                        struct sk_buff *skb = clist;
 259                        clist = clist->next;
 260                        if (skb->destructor) {
 261                                atomic_inc(&skb->users);
 262                                dev_kfree_skb_any(skb); /* put this one back */
 263                        } else {
 264                                __kfree_skb(skb);
 265                        }
 266                }
 267        }
 268
 269        put_cpu_var(softnet_data);
 270}
 271
 272static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
 273{
 274        int count = 0;
 275        struct sk_buff *skb;
 276
 277        zap_completion_queue();
 278        refill_skbs();
 279repeat:
 280
 281        skb = alloc_skb(len, GFP_ATOMIC);
 282        if (!skb)
 283                skb = skb_dequeue(&skb_pool);
 284
 285        if (!skb) {
 286                if (++count < 10) {
 287                        netpoll_poll_dev(np->dev);
 288                        goto repeat;
 289                }
 290                return NULL;
 291        }
 292
 293        atomic_set(&skb->users, 1);
 294        skb_reserve(skb, reserve);
 295        return skb;
 296}
 297
 298static int netpoll_owner_active(struct net_device *dev)
 299{
 300        struct napi_struct *napi;
 301
 302        list_for_each_entry(napi, &dev->napi_list, dev_list) {
 303                if (napi->poll_owner == smp_processor_id())
 304                        return 1;
 305        }
 306        return 0;
 307}
 308
 309/* call with IRQ disabled */
 310void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
 311                             struct net_device *dev)
 312{
 313        int status = NETDEV_TX_BUSY;
 314        unsigned long tries;
 315        const struct net_device_ops *ops = dev->netdev_ops;
 316        /* It is up to the caller to keep npinfo alive. */
 317        struct netpoll_info *npinfo;
 318
 319        WARN_ON_ONCE(!irqs_disabled());
 320
 321        npinfo = rcu_dereference_bh(np->dev->npinfo);
 322        if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
 323                __kfree_skb(skb);
 324                return;
 325        }
 326
 327        /* don't get messages out of order, and no recursion */
 328        if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 329                struct netdev_queue *txq;
 330
 331                txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 332
 333                /* try until next clock tick */
 334                for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
 335                     tries > 0; --tries) {
 336                        if (__netif_tx_trylock(txq)) {
 337                                if (!netif_xmit_stopped(txq)) {
 338                                        if (vlan_tx_tag_present(skb) &&
 339                                            !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) {
 340                                                skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
 341                                                if (unlikely(!skb))
 342                                                        break;
 343                                                skb->vlan_tci = 0;
 344                                        }
 345
 346                                        status = ops->ndo_start_xmit(skb, dev);
 347                                        if (status == NETDEV_TX_OK)
 348                                                txq_trans_update(txq);
 349                                }
 350                                __netif_tx_unlock(txq);
 351
 352                                if (status == NETDEV_TX_OK)
 353                                        break;
 354
 355                        }
 356
 357                        /* tickle device maybe there is some cleanup */
 358                        netpoll_poll_dev(np->dev);
 359
 360                        udelay(USEC_PER_POLL);
 361                }
 362
 363                WARN_ONCE(!irqs_disabled(),
 364                        "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
 365                        dev->name, ops->ndo_start_xmit);
 366
 367        }
 368
 369        if (status != NETDEV_TX_OK) {
 370                skb_queue_tail(&npinfo->txq, skb);
 371                schedule_delayed_work(&npinfo->tx_work,0);
 372        }
 373}
 374EXPORT_SYMBOL(netpoll_send_skb_on_dev);
 375
 376void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
 377{
 378        int total_len, ip_len, udp_len;
 379        struct sk_buff *skb;
 380        struct udphdr *udph;
 381        struct iphdr *iph;
 382        struct ethhdr *eth;
 383
 384        udp_len = len + sizeof(*udph);
 385        ip_len = udp_len + sizeof(*iph);
 386        total_len = ip_len + LL_RESERVED_SPACE(np->dev);
 387
 388        skb = find_skb(np, total_len + np->dev->needed_tailroom,
 389                       total_len - len);
 390        if (!skb)
 391                return;
 392
 393        skb_copy_to_linear_data(skb, msg, len);
 394        skb_put(skb, len);
 395
 396        skb_push(skb, sizeof(*udph));
 397        skb_reset_transport_header(skb);
 398        udph = udp_hdr(skb);
 399        udph->source = htons(np->local_port);
 400        udph->dest = htons(np->remote_port);
 401        udph->len = htons(udp_len);
 402        udph->check = 0;
 403        udph->check = csum_tcpudp_magic(np->local_ip,
 404                                        np->remote_ip,
 405                                        udp_len, IPPROTO_UDP,
 406                                        csum_partial(udph, udp_len, 0));
 407        if (udph->check == 0)
 408                udph->check = CSUM_MANGLED_0;
 409
 410        skb_push(skb, sizeof(*iph));
 411        skb_reset_network_header(skb);
 412        iph = ip_hdr(skb);
 413
 414        /* iph->version = 4; iph->ihl = 5; */
 415        put_unaligned(0x45, (unsigned char *)iph);
 416        iph->tos      = 0;
 417        put_unaligned(htons(ip_len), &(iph->tot_len));
 418        iph->id       = 0;
 419        iph->frag_off = 0;
 420        iph->ttl      = 64;
 421        iph->protocol = IPPROTO_UDP;
 422        iph->check    = 0;
 423        put_unaligned(np->local_ip, &(iph->saddr));
 424        put_unaligned(np->remote_ip, &(iph->daddr));
 425        iph->check    = ip_fast_csum((unsigned char *)iph, iph->ihl);
 426
 427        eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
 428        skb_reset_mac_header(skb);
 429        skb->protocol = eth->h_proto = htons(ETH_P_IP);
 430        memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN);
 431        memcpy(eth->h_dest, np->remote_mac, ETH_ALEN);
 432
 433        skb->dev = np->dev;
 434
 435        netpoll_send_skb(np, skb);
 436}
 437EXPORT_SYMBOL(netpoll_send_udp);
 438
 439static void netpoll_arp_reply(struct sk_buff *skb, struct netpoll_info *npinfo)
 440{
 441        struct arphdr *arp;
 442        unsigned char *arp_ptr;
 443        int size, type = ARPOP_REPLY, ptype = ETH_P_ARP;
 444        __be32 sip, tip;
 445        unsigned char *sha;
 446        struct sk_buff *send_skb;
 447        struct netpoll *np, *tmp;
 448        unsigned long flags;
 449        int hlen, tlen;
 450        int hits = 0;
 451
 452        if (list_empty(&npinfo->rx_np))
 453                return;
 454
 455        /* Before checking the packet, we do some early
 456           inspection whether this is interesting at all */
 457        spin_lock_irqsave(&npinfo->rx_lock, flags);
 458        list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 459                if (np->dev == skb->dev)
 460                        hits++;
 461        }
 462        spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 463
 464        /* No netpoll struct is using this dev */
 465        if (!hits)
 466                return;
 467
 468        /* No arp on this interface */
 469        if (skb->dev->flags & IFF_NOARP)
 470                return;
 471
 472        if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
 473                return;
 474
 475        skb_reset_network_header(skb);
 476        skb_reset_transport_header(skb);
 477        arp = arp_hdr(skb);
 478
 479        if ((arp->ar_hrd != htons(ARPHRD_ETHER) &&
 480             arp->ar_hrd != htons(ARPHRD_IEEE802)) ||
 481            arp->ar_pro != htons(ETH_P_IP) ||
 482            arp->ar_op != htons(ARPOP_REQUEST))
 483                return;
 484
 485        arp_ptr = (unsigned char *)(arp+1);
 486        /* save the location of the src hw addr */
 487        sha = arp_ptr;
 488        arp_ptr += skb->dev->addr_len;
 489        memcpy(&sip, arp_ptr, 4);
 490        arp_ptr += 4;
 491        /* If we actually cared about dst hw addr,
 492           it would get copied here */
 493        arp_ptr += skb->dev->addr_len;
 494        memcpy(&tip, arp_ptr, 4);
 495
 496        /* Should we ignore arp? */
 497        if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip))
 498                return;
 499
 500        size = arp_hdr_len(skb->dev);
 501
 502        spin_lock_irqsave(&npinfo->rx_lock, flags);
 503        list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 504                if (tip != np->local_ip)
 505                        continue;
 506
 507                hlen = LL_RESERVED_SPACE(np->dev);
 508                tlen = np->dev->needed_tailroom;
 509                send_skb = find_skb(np, size + hlen + tlen, hlen);
 510                if (!send_skb)
 511                        continue;
 512
 513                skb_reset_network_header(send_skb);
 514                arp = (struct arphdr *) skb_put(send_skb, size);
 515                send_skb->dev = skb->dev;
 516                send_skb->protocol = htons(ETH_P_ARP);
 517
 518                /* Fill the device header for the ARP frame */
 519                if (dev_hard_header(send_skb, skb->dev, ptype,
 520                                    sha, np->dev->dev_addr,
 521                                    send_skb->len) < 0) {
 522                        kfree_skb(send_skb);
 523                        continue;
 524                }
 525
 526                /*
 527                 * Fill out the arp protocol part.
 528                 *
 529                 * we only support ethernet device type,
 530                 * which (according to RFC 1390) should
 531                 * always equal 1 (Ethernet).
 532                 */
 533
 534                arp->ar_hrd = htons(np->dev->type);
 535                arp->ar_pro = htons(ETH_P_IP);
 536                arp->ar_hln = np->dev->addr_len;
 537                arp->ar_pln = 4;
 538                arp->ar_op = htons(type);
 539
 540                arp_ptr = (unsigned char *)(arp + 1);
 541                memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len);
 542                arp_ptr += np->dev->addr_len;
 543                memcpy(arp_ptr, &tip, 4);
 544                arp_ptr += 4;
 545                memcpy(arp_ptr, sha, np->dev->addr_len);
 546                arp_ptr += np->dev->addr_len;
 547                memcpy(arp_ptr, &sip, 4);
 548
 549                netpoll_send_skb(np, send_skb);
 550
 551                /* If there are several rx_hooks for the same address,
 552                   we're fine by sending a single reply */
 553                break;
 554        }
 555        spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 556}
 557
 558int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo)
 559{
 560        int proto, len, ulen;
 561        int hits = 0;
 562        const struct iphdr *iph;
 563        struct udphdr *uh;
 564        struct netpoll *np, *tmp;
 565
 566        if (list_empty(&npinfo->rx_np))
 567                goto out;
 568
 569        if (skb->dev->type != ARPHRD_ETHER)
 570                goto out;
 571
 572        /* check if netpoll clients need ARP */
 573        if (skb->protocol == htons(ETH_P_ARP) &&
 574            atomic_read(&trapped)) {
 575                skb_queue_tail(&npinfo->arp_tx, skb);
 576                return 1;
 577        }
 578
 579        if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
 580                skb = vlan_untag(skb);
 581                if (unlikely(!skb))
 582                        goto out;
 583        }
 584
 585        proto = ntohs(eth_hdr(skb)->h_proto);
 586        if (proto != ETH_P_IP)
 587                goto out;
 588        if (skb->pkt_type == PACKET_OTHERHOST)
 589                goto out;
 590        if (skb_shared(skb))
 591                goto out;
 592
 593        if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 594                goto out;
 595        iph = (struct iphdr *)skb->data;
 596        if (iph->ihl < 5 || iph->version != 4)
 597                goto out;
 598        if (!pskb_may_pull(skb, iph->ihl*4))
 599                goto out;
 600        iph = (struct iphdr *)skb->data;
 601        if (ip_fast_csum((u8 *)iph, iph->ihl) != 0)
 602                goto out;
 603
 604        len = ntohs(iph->tot_len);
 605        if (skb->len < len || len < iph->ihl*4)
 606                goto out;
 607
 608        /*
 609         * Our transport medium may have padded the buffer out.
 610         * Now We trim to the true length of the frame.
 611         */
 612        if (pskb_trim_rcsum(skb, len))
 613                goto out;
 614
 615        iph = (struct iphdr *)skb->data;
 616        if (iph->protocol != IPPROTO_UDP)
 617                goto out;
 618
 619        len -= iph->ihl*4;
 620        uh = (struct udphdr *)(((char *)iph) + iph->ihl*4);
 621        ulen = ntohs(uh->len);
 622
 623        if (ulen != len)
 624                goto out;
 625        if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr))
 626                goto out;
 627
 628        list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) {
 629                if (np->local_ip && np->local_ip != iph->daddr)
 630                        continue;
 631                if (np->remote_ip && np->remote_ip != iph->saddr)
 632                        continue;
 633                if (np->local_port && np->local_port != ntohs(uh->dest))
 634                        continue;
 635
 636                np->rx_hook(np, ntohs(uh->source),
 637                               (char *)(uh+1),
 638                               ulen - sizeof(struct udphdr));
 639                hits++;
 640        }
 641
 642        if (!hits)
 643                goto out;
 644
 645        kfree_skb(skb);
 646        return 1;
 647
 648out:
 649        if (atomic_read(&trapped)) {
 650                kfree_skb(skb);
 651                return 1;
 652        }
 653
 654        return 0;
 655}
 656
 657void netpoll_print_options(struct netpoll *np)
 658{
 659        np_info(np, "local port %d\n", np->local_port);
 660        np_info(np, "local IP %pI4\n", &np->local_ip);
 661        np_info(np, "interface '%s'\n", np->dev_name);
 662        np_info(np, "remote port %d\n", np->remote_port);
 663        np_info(np, "remote IP %pI4\n", &np->remote_ip);
 664        np_info(np, "remote ethernet address %pM\n", np->remote_mac);
 665}
 666EXPORT_SYMBOL(netpoll_print_options);
 667
 668int netpoll_parse_options(struct netpoll *np, char *opt)
 669{
 670        char *cur=opt, *delim;
 671
 672        if (*cur != '@') {
 673                if ((delim = strchr(cur, '@')) == NULL)
 674                        goto parse_failed;
 675                *delim = 0;
 676                np->local_port = simple_strtol(cur, NULL, 10);
 677                cur = delim;
 678        }
 679        cur++;
 680
 681        if (*cur != '/') {
 682                if ((delim = strchr(cur, '/')) == NULL)
 683                        goto parse_failed;
 684                *delim = 0;
 685                np->local_ip = in_aton(cur);
 686                cur = delim;
 687        }
 688        cur++;
 689
 690        if (*cur != ',') {
 691                /* parse out dev name */
 692                if ((delim = strchr(cur, ',')) == NULL)
 693                        goto parse_failed;
 694                *delim = 0;
 695                strlcpy(np->dev_name, cur, sizeof(np->dev_name));
 696                cur = delim;
 697        }
 698        cur++;
 699
 700        if (*cur != '@') {
 701                /* dst port */
 702                if ((delim = strchr(cur, '@')) == NULL)
 703                        goto parse_failed;
 704                *delim = 0;
 705                if (*cur == ' ' || *cur == '\t')
 706                        np_info(np, "warning: whitespace is not allowed\n");
 707                np->remote_port = simple_strtol(cur, NULL, 10);
 708                cur = delim;
 709        }
 710        cur++;
 711
 712        /* dst ip */
 713        if ((delim = strchr(cur, '/')) == NULL)
 714                goto parse_failed;
 715        *delim = 0;
 716        np->remote_ip = in_aton(cur);
 717        cur = delim + 1;
 718
 719        if (*cur != 0) {
 720                /* MAC address */
 721                if (!mac_pton(cur, np->remote_mac))
 722                        goto parse_failed;
 723        }
 724
 725        netpoll_print_options(np);
 726
 727        return 0;
 728
 729 parse_failed:
 730        np_info(np, "couldn't parse config at '%s'!\n", cur);
 731        return -1;
 732}
 733EXPORT_SYMBOL(netpoll_parse_options);
 734
 735int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp)
 736{
 737        struct netpoll_info *npinfo;
 738        const struct net_device_ops *ops;
 739        unsigned long flags;
 740        int err;
 741
 742        np->dev = ndev;
 743        strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
 744
 745        if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
 746            !ndev->netdev_ops->ndo_poll_controller) {
 747                np_err(np, "%s doesn't support polling, aborting\n",
 748                       np->dev_name);
 749                err = -ENOTSUPP;
 750                goto out;
 751        }
 752
 753        if (!ndev->npinfo) {
 754                npinfo = kmalloc(sizeof(*npinfo), gfp);
 755                if (!npinfo) {
 756                        err = -ENOMEM;
 757                        goto out;
 758                }
 759
 760                npinfo->rx_flags = 0;
 761                INIT_LIST_HEAD(&npinfo->rx_np);
 762
 763                spin_lock_init(&npinfo->rx_lock);
 764                skb_queue_head_init(&npinfo->arp_tx);
 765                skb_queue_head_init(&npinfo->txq);
 766                INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
 767
 768                atomic_set(&npinfo->refcnt, 1);
 769
 770                ops = np->dev->netdev_ops;
 771                if (ops->ndo_netpoll_setup) {
 772                        err = ops->ndo_netpoll_setup(ndev, npinfo, gfp);
 773                        if (err)
 774                                goto free_npinfo;
 775                }
 776        } else {
 777                npinfo = ndev->npinfo;
 778                atomic_inc(&npinfo->refcnt);
 779        }
 780
 781        npinfo->netpoll = np;
 782
 783        if (np->rx_hook) {
 784                spin_lock_irqsave(&npinfo->rx_lock, flags);
 785                npinfo->rx_flags |= NETPOLL_RX_ENABLED;
 786                list_add_tail(&np->rx, &npinfo->rx_np);
 787                spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 788        }
 789
 790        /* last thing to do is link it to the net device structure */
 791        rcu_assign_pointer(ndev->npinfo, npinfo);
 792
 793        return 0;
 794
 795free_npinfo:
 796        kfree(npinfo);
 797out:
 798        return err;
 799}
 800EXPORT_SYMBOL_GPL(__netpoll_setup);
 801
 802int netpoll_setup(struct netpoll *np)
 803{
 804        struct net_device *ndev = NULL;
 805        struct in_device *in_dev;
 806        int err;
 807
 808        if (np->dev_name)
 809                ndev = dev_get_by_name(&init_net, np->dev_name);
 810        if (!ndev) {
 811                np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
 812                return -ENODEV;
 813        }
 814
 815        if (ndev->master) {
 816                np_err(np, "%s is a slave device, aborting\n", np->dev_name);
 817                err = -EBUSY;
 818                goto put;
 819        }
 820
 821        if (!netif_running(ndev)) {
 822                unsigned long atmost, atleast;
 823
 824                np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
 825
 826                rtnl_lock();
 827                err = dev_open(ndev);
 828                rtnl_unlock();
 829
 830                if (err) {
 831                        np_err(np, "failed to open %s\n", ndev->name);
 832                        goto put;
 833                }
 834
 835                atleast = jiffies + HZ/10;
 836                atmost = jiffies + carrier_timeout * HZ;
 837                while (!netif_carrier_ok(ndev)) {
 838                        if (time_after(jiffies, atmost)) {
 839                                np_notice(np, "timeout waiting for carrier\n");
 840                                break;
 841                        }
 842                        msleep(1);
 843                }
 844
 845                /* If carrier appears to come up instantly, we don't
 846                 * trust it and pause so that we don't pump all our
 847                 * queued console messages into the bitbucket.
 848                 */
 849
 850                if (time_before(jiffies, atleast)) {
 851                        np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
 852                        msleep(4000);
 853                }
 854        }
 855
 856        if (!np->local_ip) {
 857                rcu_read_lock();
 858                in_dev = __in_dev_get_rcu(ndev);
 859
 860                if (!in_dev || !in_dev->ifa_list) {
 861                        rcu_read_unlock();
 862                        np_err(np, "no IP address for %s, aborting\n",
 863                               np->dev_name);
 864                        err = -EDESTADDRREQ;
 865                        goto put;
 866                }
 867
 868                np->local_ip = in_dev->ifa_list->ifa_local;
 869                rcu_read_unlock();
 870                np_info(np, "local IP %pI4\n", &np->local_ip);
 871        }
 872
 873        /* fill up the skb queue */
 874        refill_skbs();
 875
 876        rtnl_lock();
 877        err = __netpoll_setup(np, ndev, GFP_KERNEL);
 878        rtnl_unlock();
 879
 880        if (err)
 881                goto put;
 882
 883        return 0;
 884
 885put:
 886        dev_put(ndev);
 887        return err;
 888}
 889EXPORT_SYMBOL(netpoll_setup);
 890
 891static int __init netpoll_init(void)
 892{
 893        skb_queue_head_init(&skb_pool);
 894        return 0;
 895}
 896core_initcall(netpoll_init);
 897
 898static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
 899{
 900        struct netpoll_info *npinfo =
 901                        container_of(rcu_head, struct netpoll_info, rcu);
 902
 903        skb_queue_purge(&npinfo->arp_tx);
 904        skb_queue_purge(&npinfo->txq);
 905
 906        /* we can't call cancel_delayed_work_sync here, as we are in softirq */
 907        cancel_delayed_work(&npinfo->tx_work);
 908
 909        /* clean after last, unfinished work */
 910        __skb_queue_purge(&npinfo->txq);
 911        /* now cancel it again */
 912        cancel_delayed_work(&npinfo->tx_work);
 913        kfree(npinfo);
 914}
 915
 916void __netpoll_cleanup(struct netpoll *np)
 917{
 918        struct netpoll_info *npinfo;
 919        unsigned long flags;
 920
 921        npinfo = np->dev->npinfo;
 922        if (!npinfo)
 923                return;
 924
 925        if (!list_empty(&npinfo->rx_np)) {
 926                spin_lock_irqsave(&npinfo->rx_lock, flags);
 927                list_del(&np->rx);
 928                if (list_empty(&npinfo->rx_np))
 929                        npinfo->rx_flags &= ~NETPOLL_RX_ENABLED;
 930                spin_unlock_irqrestore(&npinfo->rx_lock, flags);
 931        }
 932
 933        if (atomic_dec_and_test(&npinfo->refcnt)) {
 934                const struct net_device_ops *ops;
 935
 936                ops = np->dev->netdev_ops;
 937                if (ops->ndo_netpoll_cleanup)
 938                        ops->ndo_netpoll_cleanup(np->dev);
 939
 940                RCU_INIT_POINTER(np->dev->npinfo, NULL);
 941                call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
 942        }
 943}
 944EXPORT_SYMBOL_GPL(__netpoll_cleanup);
 945
 946static void rcu_cleanup_netpoll(struct rcu_head *rcu_head)
 947{
 948        struct netpoll *np = container_of(rcu_head, struct netpoll, rcu);
 949
 950        __netpoll_cleanup(np);
 951        kfree(np);
 952}
 953
 954void __netpoll_free_rcu(struct netpoll *np)
 955{
 956        call_rcu_bh(&np->rcu, rcu_cleanup_netpoll);
 957}
 958EXPORT_SYMBOL_GPL(__netpoll_free_rcu);
 959
 960void netpoll_cleanup(struct netpoll *np)
 961{
 962        if (!np->dev)
 963                return;
 964
 965        rtnl_lock();
 966        __netpoll_cleanup(np);
 967        rtnl_unlock();
 968
 969        dev_put(np->dev);
 970        np->dev = NULL;
 971}
 972EXPORT_SYMBOL(netpoll_cleanup);
 973
 974int netpoll_trap(void)
 975{
 976        return atomic_read(&trapped);
 977}
 978EXPORT_SYMBOL(netpoll_trap);
 979
 980void netpoll_set_trap(int trap)
 981{
 982        if (trap)
 983                atomic_inc(&trapped);
 984        else
 985                atomic_dec(&trapped);
 986}
 987EXPORT_SYMBOL(netpoll_set_trap);
 988
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.