netpoll.c revision 2cde6acd49daca58b96f1fbc697492825511ad31
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 <net/addrconf.h> 33#include <net/ndisc.h> 34#include <net/ip6_checksum.h> 35#include <asm/unaligned.h> 36#include <trace/events/napi.h> 37 38/* 39 * We maintain a small pool of fully-sized skbs, to make sure the 40 * message gets out even in extreme OOM situations. 41 */ 42 43#define MAX_UDP_CHUNK 1460 44#define MAX_SKBS 32 45 46static struct sk_buff_head skb_pool; 47 48static atomic_t trapped; 49 50static struct srcu_struct netpoll_srcu; 51 52#define USEC_PER_POLL 50 53#define NETPOLL_RX_ENABLED 1 54#define NETPOLL_RX_DROP 2 55 56#define MAX_SKB_SIZE \ 57 (sizeof(struct ethhdr) + \ 58 sizeof(struct iphdr) + \ 59 sizeof(struct udphdr) + \ 60 MAX_UDP_CHUNK) 61 62static void zap_completion_queue(void); 63static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo); 64static void netpoll_async_cleanup(struct work_struct *work); 65 66static unsigned int carrier_timeout = 4; 67module_param(carrier_timeout, uint, 0644); 68 69#define np_info(np, fmt, ...) \ 70 pr_info("%s: " fmt, np->name, ##__VA_ARGS__) 71#define np_err(np, fmt, ...) \ 72 pr_err("%s: " fmt, np->name, ##__VA_ARGS__) 73#define np_notice(np, fmt, ...) \ 74 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__) 75 76static void queue_process(struct work_struct *work) 77{ 78 struct netpoll_info *npinfo = 79 container_of(work, struct netpoll_info, tx_work.work); 80 struct sk_buff *skb; 81 unsigned long flags; 82 83 while ((skb = skb_dequeue(&npinfo->txq))) { 84 struct net_device *dev = skb->dev; 85 const struct net_device_ops *ops = dev->netdev_ops; 86 struct netdev_queue *txq; 87 88 if (!netif_device_present(dev) || !netif_running(dev)) { 89 __kfree_skb(skb); 90 continue; 91 } 92 93 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); 94 95 local_irq_save(flags); 96 __netif_tx_lock(txq, smp_processor_id()); 97 if (netif_xmit_frozen_or_stopped(txq) || 98 ops->ndo_start_xmit(skb, dev) != NETDEV_TX_OK) { 99 skb_queue_head(&npinfo->txq, skb); 100 __netif_tx_unlock(txq); 101 local_irq_restore(flags); 102 103 schedule_delayed_work(&npinfo->tx_work, HZ/10); 104 return; 105 } 106 __netif_tx_unlock(txq); 107 local_irq_restore(flags); 108 } 109} 110 111static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh, 112 unsigned short ulen, __be32 saddr, __be32 daddr) 113{ 114 __wsum psum; 115 116 if (uh->check == 0 || skb_csum_unnecessary(skb)) 117 return 0; 118 119 psum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0); 120 121 if (skb->ip_summed == CHECKSUM_COMPLETE && 122 !csum_fold(csum_add(psum, skb->csum))) 123 return 0; 124 125 skb->csum = psum; 126 127 return __skb_checksum_complete(skb); 128} 129 130/* 131 * Check whether delayed processing was scheduled for our NIC. If so, 132 * we attempt to grab the poll lock and use ->poll() to pump the card. 133 * If this fails, either we've recursed in ->poll() or it's already 134 * running on another CPU. 135 * 136 * Note: we don't mask interrupts with this lock because we're using 137 * trylock here and interrupts are already disabled in the softirq 138 * case. Further, we test the poll_owner to avoid recursion on UP 139 * systems where the lock doesn't exist. 140 * 141 * In cases where there is bi-directional communications, reading only 142 * one message at a time can lead to packets being dropped by the 143 * network adapter, forcing superfluous retries and possibly timeouts. 144 * Thus, we set our budget to greater than 1. 145 */ 146static int poll_one_napi(struct netpoll_info *npinfo, 147 struct napi_struct *napi, int budget) 148{ 149 int work; 150 151 /* net_rx_action's ->poll() invocations and our's are 152 * synchronized by this test which is only made while 153 * holding the napi->poll_lock. 154 */ 155 if (!test_bit(NAPI_STATE_SCHED, &napi->state)) 156 return budget; 157 158 npinfo->rx_flags |= NETPOLL_RX_DROP; 159 atomic_inc(&trapped); 160 set_bit(NAPI_STATE_NPSVC, &napi->state); 161 162 work = napi->poll(napi, budget); 163 trace_napi_poll(napi); 164 165 clear_bit(NAPI_STATE_NPSVC, &napi->state); 166 atomic_dec(&trapped); 167 npinfo->rx_flags &= ~NETPOLL_RX_DROP; 168 169 return budget - work; 170} 171 172static void poll_napi(struct net_device *dev) 173{ 174 struct napi_struct *napi; 175 int budget = 16; 176 177 list_for_each_entry(napi, &dev->napi_list, dev_list) { 178 if (napi->poll_owner != smp_processor_id() && 179 spin_trylock(&napi->poll_lock)) { 180 budget = poll_one_napi(rcu_dereference_bh(dev->npinfo), 181 napi, budget); 182 spin_unlock(&napi->poll_lock); 183 184 if (!budget) 185 break; 186 } 187 } 188} 189 190static void service_neigh_queue(struct netpoll_info *npi) 191{ 192 if (npi) { 193 struct sk_buff *skb; 194 195 while ((skb = skb_dequeue(&npi->neigh_tx))) 196 netpoll_neigh_reply(skb, npi); 197 } 198} 199 200static void netpoll_poll_dev(struct net_device *dev) 201{ 202 const struct net_device_ops *ops; 203 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo); 204 205 /* Don't do any rx activity if the dev_lock mutex is held 206 * the dev_open/close paths use this to block netpoll activity 207 * while changing device state 208 */ 209 if (!mutex_trylock(&dev->npinfo->dev_lock)) 210 return; 211 212 if (!dev || !netif_running(dev)) 213 return; 214 215 ops = dev->netdev_ops; 216 if (!ops->ndo_poll_controller) 217 return; 218 219 /* Process pending work on NIC */ 220 ops->ndo_poll_controller(dev); 221 222 poll_napi(dev); 223 224 mutex_unlock(&dev->npinfo->dev_lock); 225 226 if (dev->flags & IFF_SLAVE) { 227 if (ni) { 228 struct net_device *bond_dev; 229 struct sk_buff *skb; 230 struct netpoll_info *bond_ni; 231 232 bond_dev = netdev_master_upper_dev_get_rcu(dev); 233 bond_ni = rcu_dereference_bh(bond_dev->npinfo); 234 while ((skb = skb_dequeue(&ni->neigh_tx))) { 235 skb->dev = bond_dev; 236 skb_queue_tail(&bond_ni->neigh_tx, skb); 237 } 238 } 239 } 240 241 service_neigh_queue(ni); 242 243 zap_completion_queue(); 244} 245 246int netpoll_rx_disable(struct net_device *dev) 247{ 248 struct netpoll_info *ni; 249 int idx; 250 might_sleep(); 251 idx = srcu_read_lock(&netpoll_srcu); 252 ni = srcu_dereference(dev->npinfo, &netpoll_srcu); 253 if (ni) 254 mutex_lock(&ni->dev_lock); 255 srcu_read_unlock(&netpoll_srcu, idx); 256 return 0; 257} 258EXPORT_SYMBOL(netpoll_rx_disable); 259 260void netpoll_rx_enable(struct net_device *dev) 261{ 262 struct netpoll_info *ni; 263 rcu_read_lock(); 264 ni = rcu_dereference(dev->npinfo); 265 if (ni) 266 mutex_unlock(&ni->dev_lock); 267 rcu_read_unlock(); 268} 269EXPORT_SYMBOL(netpoll_rx_enable); 270 271static void refill_skbs(void) 272{ 273 struct sk_buff *skb; 274 unsigned long flags; 275 276 spin_lock_irqsave(&skb_pool.lock, flags); 277 while (skb_pool.qlen < MAX_SKBS) { 278 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); 279 if (!skb) 280 break; 281 282 __skb_queue_tail(&skb_pool, skb); 283 } 284 spin_unlock_irqrestore(&skb_pool.lock, flags); 285} 286 287static void zap_completion_queue(void) 288{ 289 unsigned long flags; 290 struct softnet_data *sd = &get_cpu_var(softnet_data); 291 292 if (sd->completion_queue) { 293 struct sk_buff *clist; 294 295 local_irq_save(flags); 296 clist = sd->completion_queue; 297 sd->completion_queue = NULL; 298 local_irq_restore(flags); 299 300 while (clist != NULL) { 301 struct sk_buff *skb = clist; 302 clist = clist->next; 303 if (skb->destructor) { 304 atomic_inc(&skb->users); 305 dev_kfree_skb_any(skb); /* put this one back */ 306 } else { 307 __kfree_skb(skb); 308 } 309 } 310 } 311 312 put_cpu_var(softnet_data); 313} 314 315static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) 316{ 317 int count = 0; 318 struct sk_buff *skb; 319 320 zap_completion_queue(); 321 refill_skbs(); 322repeat: 323 324 skb = alloc_skb(len, GFP_ATOMIC); 325 if (!skb) 326 skb = skb_dequeue(&skb_pool); 327 328 if (!skb) { 329 if (++count < 10) { 330 netpoll_poll_dev(np->dev); 331 goto repeat; 332 } 333 return NULL; 334 } 335 336 atomic_set(&skb->users, 1); 337 skb_reserve(skb, reserve); 338 return skb; 339} 340 341static int netpoll_owner_active(struct net_device *dev) 342{ 343 struct napi_struct *napi; 344 345 list_for_each_entry(napi, &dev->napi_list, dev_list) { 346 if (napi->poll_owner == smp_processor_id()) 347 return 1; 348 } 349 return 0; 350} 351 352/* call with IRQ disabled */ 353void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, 354 struct net_device *dev) 355{ 356 int status = NETDEV_TX_BUSY; 357 unsigned long tries; 358 const struct net_device_ops *ops = dev->netdev_ops; 359 /* It is up to the caller to keep npinfo alive. */ 360 struct netpoll_info *npinfo; 361 362 WARN_ON_ONCE(!irqs_disabled()); 363 364 npinfo = rcu_dereference_bh(np->dev->npinfo); 365 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { 366 __kfree_skb(skb); 367 return; 368 } 369 370 /* don't get messages out of order, and no recursion */ 371 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { 372 struct netdev_queue *txq; 373 374 txq = netdev_pick_tx(dev, skb); 375 376 /* try until next clock tick */ 377 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; 378 tries > 0; --tries) { 379 if (__netif_tx_trylock(txq)) { 380 if (!netif_xmit_stopped(txq)) { 381 if (vlan_tx_tag_present(skb) && 382 !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) { 383 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); 384 if (unlikely(!skb)) 385 break; 386 skb->vlan_tci = 0; 387 } 388 389 status = ops->ndo_start_xmit(skb, dev); 390 if (status == NETDEV_TX_OK) 391 txq_trans_update(txq); 392 } 393 __netif_tx_unlock(txq); 394 395 if (status == NETDEV_TX_OK) 396 break; 397 398 } 399 400 /* tickle device maybe there is some cleanup */ 401 netpoll_poll_dev(np->dev); 402 403 udelay(USEC_PER_POLL); 404 } 405 406 WARN_ONCE(!irqs_disabled(), 407 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n", 408 dev->name, ops->ndo_start_xmit); 409 410 } 411 412 if (status != NETDEV_TX_OK) { 413 skb_queue_tail(&npinfo->txq, skb); 414 schedule_delayed_work(&npinfo->tx_work,0); 415 } 416} 417EXPORT_SYMBOL(netpoll_send_skb_on_dev); 418 419void netpoll_send_udp(struct netpoll *np, const char *msg, int len) 420{ 421 int total_len, ip_len, udp_len; 422 struct sk_buff *skb; 423 struct udphdr *udph; 424 struct iphdr *iph; 425 struct ethhdr *eth; 426 static atomic_t ip_ident; 427 struct ipv6hdr *ip6h; 428 429 udp_len = len + sizeof(*udph); 430 if (np->ipv6) 431 ip_len = udp_len + sizeof(*ip6h); 432 else 433 ip_len = udp_len + sizeof(*iph); 434 435 total_len = ip_len + LL_RESERVED_SPACE(np->dev); 436 437 skb = find_skb(np, total_len + np->dev->needed_tailroom, 438 total_len - len); 439 if (!skb) 440 return; 441 442 skb_copy_to_linear_data(skb, msg, len); 443 skb_put(skb, len); 444 445 skb_push(skb, sizeof(*udph)); 446 skb_reset_transport_header(skb); 447 udph = udp_hdr(skb); 448 udph->source = htons(np->local_port); 449 udph->dest = htons(np->remote_port); 450 udph->len = htons(udp_len); 451 452 if (np->ipv6) { 453 udph->check = 0; 454 udph->check = csum_ipv6_magic(&np->local_ip.in6, 455 &np->remote_ip.in6, 456 udp_len, IPPROTO_UDP, 457 csum_partial(udph, udp_len, 0)); 458 if (udph->check == 0) 459 udph->check = CSUM_MANGLED_0; 460 461 skb_push(skb, sizeof(*ip6h)); 462 skb_reset_network_header(skb); 463 ip6h = ipv6_hdr(skb); 464 465 /* ip6h->version = 6; ip6h->priority = 0; */ 466 put_unaligned(0x60, (unsigned char *)ip6h); 467 ip6h->flow_lbl[0] = 0; 468 ip6h->flow_lbl[1] = 0; 469 ip6h->flow_lbl[2] = 0; 470 471 ip6h->payload_len = htons(sizeof(struct udphdr) + len); 472 ip6h->nexthdr = IPPROTO_UDP; 473 ip6h->hop_limit = 32; 474 ip6h->saddr = np->local_ip.in6; 475 ip6h->daddr = np->remote_ip.in6; 476 477 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 478 skb_reset_mac_header(skb); 479 skb->protocol = eth->h_proto = htons(ETH_P_IPV6); 480 } else { 481 udph->check = 0; 482 udph->check = csum_tcpudp_magic(np->local_ip.ip, 483 np->remote_ip.ip, 484 udp_len, IPPROTO_UDP, 485 csum_partial(udph, udp_len, 0)); 486 if (udph->check == 0) 487 udph->check = CSUM_MANGLED_0; 488 489 skb_push(skb, sizeof(*iph)); 490 skb_reset_network_header(skb); 491 iph = ip_hdr(skb); 492 493 /* iph->version = 4; iph->ihl = 5; */ 494 put_unaligned(0x45, (unsigned char *)iph); 495 iph->tos = 0; 496 put_unaligned(htons(ip_len), &(iph->tot_len)); 497 iph->id = htons(atomic_inc_return(&ip_ident)); 498 iph->frag_off = 0; 499 iph->ttl = 64; 500 iph->protocol = IPPROTO_UDP; 501 iph->check = 0; 502 put_unaligned(np->local_ip.ip, &(iph->saddr)); 503 put_unaligned(np->remote_ip.ip, &(iph->daddr)); 504 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 505 506 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 507 skb_reset_mac_header(skb); 508 skb->protocol = eth->h_proto = htons(ETH_P_IP); 509 } 510 511 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); 512 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); 513 514 skb->dev = np->dev; 515 516 netpoll_send_skb(np, skb); 517} 518EXPORT_SYMBOL(netpoll_send_udp); 519 520static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo) 521{ 522 int size, type = ARPOP_REPLY; 523 __be32 sip, tip; 524 unsigned char *sha; 525 struct sk_buff *send_skb; 526 struct netpoll *np, *tmp; 527 unsigned long flags; 528 int hlen, tlen; 529 int hits = 0, proto; 530 531 if (list_empty(&npinfo->rx_np)) 532 return; 533 534 /* Before checking the packet, we do some early 535 inspection whether this is interesting at all */ 536 spin_lock_irqsave(&npinfo->rx_lock, flags); 537 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 538 if (np->dev == skb->dev) 539 hits++; 540 } 541 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 542 543 /* No netpoll struct is using this dev */ 544 if (!hits) 545 return; 546 547 proto = ntohs(eth_hdr(skb)->h_proto); 548 if (proto == ETH_P_IP) { 549 struct arphdr *arp; 550 unsigned char *arp_ptr; 551 /* No arp on this interface */ 552 if (skb->dev->flags & IFF_NOARP) 553 return; 554 555 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) 556 return; 557 558 skb_reset_network_header(skb); 559 skb_reset_transport_header(skb); 560 arp = arp_hdr(skb); 561 562 if ((arp->ar_hrd != htons(ARPHRD_ETHER) && 563 arp->ar_hrd != htons(ARPHRD_IEEE802)) || 564 arp->ar_pro != htons(ETH_P_IP) || 565 arp->ar_op != htons(ARPOP_REQUEST)) 566 return; 567 568 arp_ptr = (unsigned char *)(arp+1); 569 /* save the location of the src hw addr */ 570 sha = arp_ptr; 571 arp_ptr += skb->dev->addr_len; 572 memcpy(&sip, arp_ptr, 4); 573 arp_ptr += 4; 574 /* If we actually cared about dst hw addr, 575 it would get copied here */ 576 arp_ptr += skb->dev->addr_len; 577 memcpy(&tip, arp_ptr, 4); 578 579 /* Should we ignore arp? */ 580 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) 581 return; 582 583 size = arp_hdr_len(skb->dev); 584 585 spin_lock_irqsave(&npinfo->rx_lock, flags); 586 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 587 if (tip != np->local_ip.ip) 588 continue; 589 590 hlen = LL_RESERVED_SPACE(np->dev); 591 tlen = np->dev->needed_tailroom; 592 send_skb = find_skb(np, size + hlen + tlen, hlen); 593 if (!send_skb) 594 continue; 595 596 skb_reset_network_header(send_skb); 597 arp = (struct arphdr *) skb_put(send_skb, size); 598 send_skb->dev = skb->dev; 599 send_skb->protocol = htons(ETH_P_ARP); 600 601 /* Fill the device header for the ARP frame */ 602 if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP, 603 sha, np->dev->dev_addr, 604 send_skb->len) < 0) { 605 kfree_skb(send_skb); 606 continue; 607 } 608 609 /* 610 * Fill out the arp protocol part. 611 * 612 * we only support ethernet device type, 613 * which (according to RFC 1390) should 614 * always equal 1 (Ethernet). 615 */ 616 617 arp->ar_hrd = htons(np->dev->type); 618 arp->ar_pro = htons(ETH_P_IP); 619 arp->ar_hln = np->dev->addr_len; 620 arp->ar_pln = 4; 621 arp->ar_op = htons(type); 622 623 arp_ptr = (unsigned char *)(arp + 1); 624 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); 625 arp_ptr += np->dev->addr_len; 626 memcpy(arp_ptr, &tip, 4); 627 arp_ptr += 4; 628 memcpy(arp_ptr, sha, np->dev->addr_len); 629 arp_ptr += np->dev->addr_len; 630 memcpy(arp_ptr, &sip, 4); 631 632 netpoll_send_skb(np, send_skb); 633 634 /* If there are several rx_hooks for the same address, 635 we're fine by sending a single reply */ 636 break; 637 } 638 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 639 } else if( proto == ETH_P_IPV6) { 640#if IS_ENABLED(CONFIG_IPV6) 641 struct nd_msg *msg; 642 u8 *lladdr = NULL; 643 struct ipv6hdr *hdr; 644 struct icmp6hdr *icmp6h; 645 const struct in6_addr *saddr; 646 const struct in6_addr *daddr; 647 struct inet6_dev *in6_dev = NULL; 648 struct in6_addr *target; 649 650 in6_dev = in6_dev_get(skb->dev); 651 if (!in6_dev || !in6_dev->cnf.accept_ra) 652 return; 653 654 if (!pskb_may_pull(skb, skb->len)) 655 return; 656 657 msg = (struct nd_msg *)skb_transport_header(skb); 658 659 __skb_push(skb, skb->data - skb_transport_header(skb)); 660 661 if (ipv6_hdr(skb)->hop_limit != 255) 662 return; 663 if (msg->icmph.icmp6_code != 0) 664 return; 665 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION) 666 return; 667 668 saddr = &ipv6_hdr(skb)->saddr; 669 daddr = &ipv6_hdr(skb)->daddr; 670 671 size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr); 672 673 spin_lock_irqsave(&npinfo->rx_lock, flags); 674 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 675 if (!ipv6_addr_equal(daddr, &np->local_ip.in6)) 676 continue; 677 678 hlen = LL_RESERVED_SPACE(np->dev); 679 tlen = np->dev->needed_tailroom; 680 send_skb = find_skb(np, size + hlen + tlen, hlen); 681 if (!send_skb) 682 continue; 683 684 send_skb->protocol = htons(ETH_P_IPV6); 685 send_skb->dev = skb->dev; 686 687 skb_reset_network_header(send_skb); 688 skb_put(send_skb, sizeof(struct ipv6hdr)); 689 hdr = ipv6_hdr(send_skb); 690 691 *(__be32*)hdr = htonl(0x60000000); 692 693 hdr->payload_len = htons(size); 694 hdr->nexthdr = IPPROTO_ICMPV6; 695 hdr->hop_limit = 255; 696 hdr->saddr = *saddr; 697 hdr->daddr = *daddr; 698 699 send_skb->transport_header = send_skb->tail; 700 skb_put(send_skb, size); 701 702 icmp6h = (struct icmp6hdr *)skb_transport_header(skb); 703 icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 704 icmp6h->icmp6_router = 0; 705 icmp6h->icmp6_solicited = 1; 706 target = (struct in6_addr *)skb_transport_header(send_skb) + sizeof(struct icmp6hdr); 707 *target = msg->target; 708 icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size, 709 IPPROTO_ICMPV6, 710 csum_partial(icmp6h, 711 size, 0)); 712 713 if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6, 714 lladdr, np->dev->dev_addr, 715 send_skb->len) < 0) { 716 kfree_skb(send_skb); 717 continue; 718 } 719 720 netpoll_send_skb(np, send_skb); 721 722 /* If there are several rx_hooks for the same address, 723 we're fine by sending a single reply */ 724 break; 725 } 726 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 727#endif 728 } 729} 730 731static bool pkt_is_ns(struct sk_buff *skb) 732{ 733 struct nd_msg *msg; 734 struct ipv6hdr *hdr; 735 736 if (skb->protocol != htons(ETH_P_ARP)) 737 return false; 738 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg))) 739 return false; 740 741 msg = (struct nd_msg *)skb_transport_header(skb); 742 __skb_push(skb, skb->data - skb_transport_header(skb)); 743 hdr = ipv6_hdr(skb); 744 745 if (hdr->nexthdr != IPPROTO_ICMPV6) 746 return false; 747 if (hdr->hop_limit != 255) 748 return false; 749 if (msg->icmph.icmp6_code != 0) 750 return false; 751 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION) 752 return false; 753 754 return true; 755} 756 757int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo) 758{ 759 int proto, len, ulen; 760 int hits = 0; 761 const struct iphdr *iph; 762 struct udphdr *uh; 763 struct netpoll *np, *tmp; 764 765 if (list_empty(&npinfo->rx_np)) 766 goto out; 767 768 if (skb->dev->type != ARPHRD_ETHER) 769 goto out; 770 771 /* check if netpoll clients need ARP */ 772 if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) { 773 skb_queue_tail(&npinfo->neigh_tx, skb); 774 return 1; 775 } else if (pkt_is_ns(skb) && atomic_read(&trapped)) { 776 skb_queue_tail(&npinfo->neigh_tx, skb); 777 return 1; 778 } 779 780 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { 781 skb = vlan_untag(skb); 782 if (unlikely(!skb)) 783 goto out; 784 } 785 786 proto = ntohs(eth_hdr(skb)->h_proto); 787 if (proto != ETH_P_IP && proto != ETH_P_IPV6) 788 goto out; 789 if (skb->pkt_type == PACKET_OTHERHOST) 790 goto out; 791 if (skb_shared(skb)) 792 goto out; 793 794 if (proto == ETH_P_IP) { 795 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 796 goto out; 797 iph = (struct iphdr *)skb->data; 798 if (iph->ihl < 5 || iph->version != 4) 799 goto out; 800 if (!pskb_may_pull(skb, iph->ihl*4)) 801 goto out; 802 iph = (struct iphdr *)skb->data; 803 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) 804 goto out; 805 806 len = ntohs(iph->tot_len); 807 if (skb->len < len || len < iph->ihl*4) 808 goto out; 809 810 /* 811 * Our transport medium may have padded the buffer out. 812 * Now We trim to the true length of the frame. 813 */ 814 if (pskb_trim_rcsum(skb, len)) 815 goto out; 816 817 iph = (struct iphdr *)skb->data; 818 if (iph->protocol != IPPROTO_UDP) 819 goto out; 820 821 len -= iph->ihl*4; 822 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); 823 ulen = ntohs(uh->len); 824 825 if (ulen != len) 826 goto out; 827 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) 828 goto out; 829 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 830 if (np->local_ip.ip && np->local_ip.ip != iph->daddr) 831 continue; 832 if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr) 833 continue; 834 if (np->local_port && np->local_port != ntohs(uh->dest)) 835 continue; 836 837 np->rx_hook(np, ntohs(uh->source), 838 (char *)(uh+1), 839 ulen - sizeof(struct udphdr)); 840 hits++; 841 } 842 } else { 843#if IS_ENABLED(CONFIG_IPV6) 844 const struct ipv6hdr *ip6h; 845 846 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) 847 goto out; 848 ip6h = (struct ipv6hdr *)skb->data; 849 if (ip6h->version != 6) 850 goto out; 851 len = ntohs(ip6h->payload_len); 852 if (!len) 853 goto out; 854 if (len + sizeof(struct ipv6hdr) > skb->len) 855 goto out; 856 if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr))) 857 goto out; 858 ip6h = ipv6_hdr(skb); 859 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 860 goto out; 861 uh = udp_hdr(skb); 862 ulen = ntohs(uh->len); 863 if (ulen != skb->len) 864 goto out; 865 if (udp6_csum_init(skb, uh, IPPROTO_UDP)) 866 goto out; 867 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 868 if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr)) 869 continue; 870 if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr)) 871 continue; 872 if (np->local_port && np->local_port != ntohs(uh->dest)) 873 continue; 874 875 np->rx_hook(np, ntohs(uh->source), 876 (char *)(uh+1), 877 ulen - sizeof(struct udphdr)); 878 hits++; 879 } 880#endif 881 } 882 883 if (!hits) 884 goto out; 885 886 kfree_skb(skb); 887 return 1; 888 889out: 890 if (atomic_read(&trapped)) { 891 kfree_skb(skb); 892 return 1; 893 } 894 895 return 0; 896} 897 898void netpoll_print_options(struct netpoll *np) 899{ 900 np_info(np, "local port %d\n", np->local_port); 901 if (np->ipv6) 902 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6); 903 else 904 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip); 905 np_info(np, "interface '%s'\n", np->dev_name); 906 np_info(np, "remote port %d\n", np->remote_port); 907 if (np->ipv6) 908 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6); 909 else 910 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip); 911 np_info(np, "remote ethernet address %pM\n", np->remote_mac); 912} 913EXPORT_SYMBOL(netpoll_print_options); 914 915static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr) 916{ 917 const char *end; 918 919 if (!strchr(str, ':') && 920 in4_pton(str, -1, (void *)addr, -1, &end) > 0) { 921 if (!*end) 922 return 0; 923 } 924 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) { 925#if IS_ENABLED(CONFIG_IPV6) 926 if (!*end) 927 return 1; 928#else 929 return -1; 930#endif 931 } 932 return -1; 933} 934 935int netpoll_parse_options(struct netpoll *np, char *opt) 936{ 937 char *cur=opt, *delim; 938 int ipv6; 939 940 if (*cur != '@') { 941 if ((delim = strchr(cur, '@')) == NULL) 942 goto parse_failed; 943 *delim = 0; 944 if (kstrtou16(cur, 10, &np->local_port)) 945 goto parse_failed; 946 cur = delim; 947 } 948 cur++; 949 950 if (*cur != '/') { 951 if ((delim = strchr(cur, '/')) == NULL) 952 goto parse_failed; 953 *delim = 0; 954 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip); 955 if (ipv6 < 0) 956 goto parse_failed; 957 else 958 np->ipv6 = (bool)ipv6; 959 cur = delim; 960 } 961 cur++; 962 963 if (*cur != ',') { 964 /* parse out dev name */ 965 if ((delim = strchr(cur, ',')) == NULL) 966 goto parse_failed; 967 *delim = 0; 968 strlcpy(np->dev_name, cur, sizeof(np->dev_name)); 969 cur = delim; 970 } 971 cur++; 972 973 if (*cur != '@') { 974 /* dst port */ 975 if ((delim = strchr(cur, '@')) == NULL) 976 goto parse_failed; 977 *delim = 0; 978 if (*cur == ' ' || *cur == '\t') 979 np_info(np, "warning: whitespace is not allowed\n"); 980 if (kstrtou16(cur, 10, &np->remote_port)) 981 goto parse_failed; 982 cur = delim; 983 } 984 cur++; 985 986 /* dst ip */ 987 if ((delim = strchr(cur, '/')) == NULL) 988 goto parse_failed; 989 *delim = 0; 990 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip); 991 if (ipv6 < 0) 992 goto parse_failed; 993 else if (np->ipv6 != (bool)ipv6) 994 goto parse_failed; 995 else 996 np->ipv6 = (bool)ipv6; 997 cur = delim + 1; 998 999 if (*cur != 0) { 1000 /* MAC address */ 1001 if (!mac_pton(cur, np->remote_mac)) 1002 goto parse_failed; 1003 } 1004 1005 netpoll_print_options(np); 1006 1007 return 0; 1008 1009 parse_failed: 1010 np_info(np, "couldn't parse config at '%s'!\n", cur); 1011 return -1; 1012} 1013EXPORT_SYMBOL(netpoll_parse_options); 1014 1015int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp) 1016{ 1017 struct netpoll_info *npinfo; 1018 const struct net_device_ops *ops; 1019 unsigned long flags; 1020 int err; 1021 1022 np->dev = ndev; 1023 strlcpy(np->dev_name, ndev->name, IFNAMSIZ); 1024 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup); 1025 1026 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || 1027 !ndev->netdev_ops->ndo_poll_controller) { 1028 np_err(np, "%s doesn't support polling, aborting\n", 1029 np->dev_name); 1030 err = -ENOTSUPP; 1031 goto out; 1032 } 1033 1034 if (!ndev->npinfo) { 1035 npinfo = kmalloc(sizeof(*npinfo), gfp); 1036 if (!npinfo) { 1037 err = -ENOMEM; 1038 goto out; 1039 } 1040 1041 npinfo->rx_flags = 0; 1042 INIT_LIST_HEAD(&npinfo->rx_np); 1043 1044 spin_lock_init(&npinfo->rx_lock); 1045 mutex_init(&npinfo->dev_lock); 1046 skb_queue_head_init(&npinfo->neigh_tx); 1047 skb_queue_head_init(&npinfo->txq); 1048 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); 1049 1050 atomic_set(&npinfo->refcnt, 1); 1051 1052 ops = np->dev->netdev_ops; 1053 if (ops->ndo_netpoll_setup) { 1054 err = ops->ndo_netpoll_setup(ndev, npinfo, gfp); 1055 if (err) 1056 goto free_npinfo; 1057 } 1058 } else { 1059 npinfo = ndev->npinfo; 1060 atomic_inc(&npinfo->refcnt); 1061 } 1062 1063 npinfo->netpoll = np; 1064 1065 if (np->rx_hook) { 1066 spin_lock_irqsave(&npinfo->rx_lock, flags); 1067 npinfo->rx_flags |= NETPOLL_RX_ENABLED; 1068 list_add_tail(&np->rx, &npinfo->rx_np); 1069 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 1070 } 1071 1072 /* last thing to do is link it to the net device structure */ 1073 rcu_assign_pointer(ndev->npinfo, npinfo); 1074 1075 return 0; 1076 1077free_npinfo: 1078 kfree(npinfo); 1079out: 1080 return err; 1081} 1082EXPORT_SYMBOL_GPL(__netpoll_setup); 1083 1084int netpoll_setup(struct netpoll *np) 1085{ 1086 struct net_device *ndev = NULL; 1087 struct in_device *in_dev; 1088 int err; 1089 1090 rtnl_lock(); 1091 if (np->dev_name) { 1092 struct net *net = current->nsproxy->net_ns; 1093 ndev = __dev_get_by_name(net, np->dev_name); 1094 } 1095 if (!ndev) { 1096 np_err(np, "%s doesn't exist, aborting\n", np->dev_name); 1097 err = -ENODEV; 1098 goto unlock; 1099 } 1100 dev_hold(ndev); 1101 1102 if (netdev_master_upper_dev_get(ndev)) { 1103 np_err(np, "%s is a slave device, aborting\n", np->dev_name); 1104 err = -EBUSY; 1105 goto put; 1106 } 1107 1108 if (!netif_running(ndev)) { 1109 unsigned long atmost, atleast; 1110 1111 np_info(np, "device %s not up yet, forcing it\n", np->dev_name); 1112 1113 err = dev_open(ndev); 1114 1115 if (err) { 1116 np_err(np, "failed to open %s\n", ndev->name); 1117 goto put; 1118 } 1119 1120 rtnl_unlock(); 1121 atleast = jiffies + HZ/10; 1122 atmost = jiffies + carrier_timeout * HZ; 1123 while (!netif_carrier_ok(ndev)) { 1124 if (time_after(jiffies, atmost)) { 1125 np_notice(np, "timeout waiting for carrier\n"); 1126 break; 1127 } 1128 msleep(1); 1129 } 1130 1131 /* If carrier appears to come up instantly, we don't 1132 * trust it and pause so that we don't pump all our 1133 * queued console messages into the bitbucket. 1134 */ 1135 1136 if (time_before(jiffies, atleast)) { 1137 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n"); 1138 msleep(4000); 1139 } 1140 rtnl_lock(); 1141 } 1142 1143 if (!np->local_ip.ip) { 1144 if (!np->ipv6) { 1145 in_dev = __in_dev_get_rtnl(ndev); 1146 1147 if (!in_dev || !in_dev->ifa_list) { 1148 np_err(np, "no IP address for %s, aborting\n", 1149 np->dev_name); 1150 err = -EDESTADDRREQ; 1151 goto put; 1152 } 1153 1154 np->local_ip.ip = in_dev->ifa_list->ifa_local; 1155 np_info(np, "local IP %pI4\n", &np->local_ip.ip); 1156 } else { 1157#if IS_ENABLED(CONFIG_IPV6) 1158 struct inet6_dev *idev; 1159 1160 err = -EDESTADDRREQ; 1161 idev = __in6_dev_get(ndev); 1162 if (idev) { 1163 struct inet6_ifaddr *ifp; 1164 1165 read_lock_bh(&idev->lock); 1166 list_for_each_entry(ifp, &idev->addr_list, if_list) { 1167 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) 1168 continue; 1169 np->local_ip.in6 = ifp->addr; 1170 err = 0; 1171 break; 1172 } 1173 read_unlock_bh(&idev->lock); 1174 } 1175 if (err) { 1176 np_err(np, "no IPv6 address for %s, aborting\n", 1177 np->dev_name); 1178 goto put; 1179 } else 1180 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6); 1181#else 1182 np_err(np, "IPv6 is not supported %s, aborting\n", 1183 np->dev_name); 1184 err = -EINVAL; 1185 goto put; 1186#endif 1187 } 1188 } 1189 1190 /* fill up the skb queue */ 1191 refill_skbs(); 1192 1193 err = __netpoll_setup(np, ndev, GFP_KERNEL); 1194 if (err) 1195 goto put; 1196 1197 rtnl_unlock(); 1198 return 0; 1199 1200put: 1201 dev_put(ndev); 1202unlock: 1203 rtnl_unlock(); 1204 return err; 1205} 1206EXPORT_SYMBOL(netpoll_setup); 1207 1208static int __init netpoll_init(void) 1209{ 1210 skb_queue_head_init(&skb_pool); 1211 init_srcu_struct(&netpoll_srcu); 1212 return 0; 1213} 1214core_initcall(netpoll_init); 1215 1216static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head) 1217{ 1218 struct netpoll_info *npinfo = 1219 container_of(rcu_head, struct netpoll_info, rcu); 1220 1221 skb_queue_purge(&npinfo->neigh_tx); 1222 skb_queue_purge(&npinfo->txq); 1223 1224 /* we can't call cancel_delayed_work_sync here, as we are in softirq */ 1225 cancel_delayed_work(&npinfo->tx_work); 1226 1227 /* clean after last, unfinished work */ 1228 __skb_queue_purge(&npinfo->txq); 1229 /* now cancel it again */ 1230 cancel_delayed_work(&npinfo->tx_work); 1231 kfree(npinfo); 1232} 1233 1234void __netpoll_cleanup(struct netpoll *np) 1235{ 1236 struct netpoll_info *npinfo; 1237 unsigned long flags; 1238 1239 npinfo = np->dev->npinfo; 1240 if (!npinfo) 1241 return; 1242 1243 if (!list_empty(&npinfo->rx_np)) { 1244 spin_lock_irqsave(&npinfo->rx_lock, flags); 1245 list_del(&np->rx); 1246 if (list_empty(&npinfo->rx_np)) 1247 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; 1248 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 1249 } 1250 1251 synchronize_srcu(&netpoll_srcu); 1252 1253 if (atomic_dec_and_test(&npinfo->refcnt)) { 1254 const struct net_device_ops *ops; 1255 1256 ops = np->dev->netdev_ops; 1257 if (ops->ndo_netpoll_cleanup) 1258 ops->ndo_netpoll_cleanup(np->dev); 1259 1260 rcu_assign_pointer(np->dev->npinfo, NULL); 1261 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info); 1262 } 1263} 1264EXPORT_SYMBOL_GPL(__netpoll_cleanup); 1265 1266static void netpoll_async_cleanup(struct work_struct *work) 1267{ 1268 struct netpoll *np = container_of(work, struct netpoll, cleanup_work); 1269 1270 rtnl_lock(); 1271 __netpoll_cleanup(np); 1272 rtnl_unlock(); 1273 kfree(np); 1274} 1275 1276void __netpoll_free_async(struct netpoll *np) 1277{ 1278 schedule_work(&np->cleanup_work); 1279} 1280EXPORT_SYMBOL_GPL(__netpoll_free_async); 1281 1282void netpoll_cleanup(struct netpoll *np) 1283{ 1284 if (!np->dev) 1285 return; 1286 1287 rtnl_lock(); 1288 __netpoll_cleanup(np); 1289 rtnl_unlock(); 1290 1291 dev_put(np->dev); 1292 np->dev = NULL; 1293} 1294EXPORT_SYMBOL(netpoll_cleanup); 1295 1296int netpoll_trap(void) 1297{ 1298 return atomic_read(&trapped); 1299} 1300EXPORT_SYMBOL(netpoll_trap); 1301 1302void netpoll_set_trap(int trap) 1303{ 1304 if (trap) 1305 atomic_inc(&trapped); 1306 else 1307 atomic_dec(&trapped); 1308} 1309EXPORT_SYMBOL(netpoll_set_trap); 1310