netpoll.c revision 7f9421c264f8a6e6137027a45ae576517f66fa56
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 50DEFINE_STATIC_SRCU(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(&ni->dev_lock)) 210 return; 211 212 if (!netif_running(dev)) { 213 mutex_unlock(&ni->dev_lock); 214 return; 215 } 216 217 ops = dev->netdev_ops; 218 if (!ops->ndo_poll_controller) { 219 mutex_unlock(&ni->dev_lock); 220 return; 221 } 222 223 /* Process pending work on NIC */ 224 ops->ndo_poll_controller(dev); 225 226 poll_napi(dev); 227 228 mutex_unlock(&ni->dev_lock); 229 230 if (dev->flags & IFF_SLAVE) { 231 if (ni) { 232 struct net_device *bond_dev; 233 struct sk_buff *skb; 234 struct netpoll_info *bond_ni; 235 236 bond_dev = netdev_master_upper_dev_get_rcu(dev); 237 bond_ni = rcu_dereference_bh(bond_dev->npinfo); 238 while ((skb = skb_dequeue(&ni->neigh_tx))) { 239 skb->dev = bond_dev; 240 skb_queue_tail(&bond_ni->neigh_tx, skb); 241 } 242 } 243 } 244 245 service_neigh_queue(ni); 246 247 zap_completion_queue(); 248} 249 250int netpoll_rx_disable(struct net_device *dev) 251{ 252 struct netpoll_info *ni; 253 int idx; 254 might_sleep(); 255 idx = srcu_read_lock(&netpoll_srcu); 256 ni = srcu_dereference(dev->npinfo, &netpoll_srcu); 257 if (ni) 258 mutex_lock(&ni->dev_lock); 259 srcu_read_unlock(&netpoll_srcu, idx); 260 return 0; 261} 262EXPORT_SYMBOL(netpoll_rx_disable); 263 264void netpoll_rx_enable(struct net_device *dev) 265{ 266 struct netpoll_info *ni; 267 rcu_read_lock(); 268 ni = rcu_dereference(dev->npinfo); 269 if (ni) 270 mutex_unlock(&ni->dev_lock); 271 rcu_read_unlock(); 272} 273EXPORT_SYMBOL(netpoll_rx_enable); 274 275static void refill_skbs(void) 276{ 277 struct sk_buff *skb; 278 unsigned long flags; 279 280 spin_lock_irqsave(&skb_pool.lock, flags); 281 while (skb_pool.qlen < MAX_SKBS) { 282 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); 283 if (!skb) 284 break; 285 286 __skb_queue_tail(&skb_pool, skb); 287 } 288 spin_unlock_irqrestore(&skb_pool.lock, flags); 289} 290 291static void zap_completion_queue(void) 292{ 293 unsigned long flags; 294 struct softnet_data *sd = &get_cpu_var(softnet_data); 295 296 if (sd->completion_queue) { 297 struct sk_buff *clist; 298 299 local_irq_save(flags); 300 clist = sd->completion_queue; 301 sd->completion_queue = NULL; 302 local_irq_restore(flags); 303 304 while (clist != NULL) { 305 struct sk_buff *skb = clist; 306 clist = clist->next; 307 if (skb->destructor) { 308 atomic_inc(&skb->users); 309 dev_kfree_skb_any(skb); /* put this one back */ 310 } else { 311 __kfree_skb(skb); 312 } 313 } 314 } 315 316 put_cpu_var(softnet_data); 317} 318 319static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) 320{ 321 int count = 0; 322 struct sk_buff *skb; 323 324 zap_completion_queue(); 325 refill_skbs(); 326repeat: 327 328 skb = alloc_skb(len, GFP_ATOMIC); 329 if (!skb) 330 skb = skb_dequeue(&skb_pool); 331 332 if (!skb) { 333 if (++count < 10) { 334 netpoll_poll_dev(np->dev); 335 goto repeat; 336 } 337 return NULL; 338 } 339 340 atomic_set(&skb->users, 1); 341 skb_reserve(skb, reserve); 342 return skb; 343} 344 345static int netpoll_owner_active(struct net_device *dev) 346{ 347 struct napi_struct *napi; 348 349 list_for_each_entry(napi, &dev->napi_list, dev_list) { 350 if (napi->poll_owner == smp_processor_id()) 351 return 1; 352 } 353 return 0; 354} 355 356/* call with IRQ disabled */ 357void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb, 358 struct net_device *dev) 359{ 360 int status = NETDEV_TX_BUSY; 361 unsigned long tries; 362 const struct net_device_ops *ops = dev->netdev_ops; 363 /* It is up to the caller to keep npinfo alive. */ 364 struct netpoll_info *npinfo; 365 366 WARN_ON_ONCE(!irqs_disabled()); 367 368 npinfo = rcu_dereference_bh(np->dev->npinfo); 369 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { 370 __kfree_skb(skb); 371 return; 372 } 373 374 /* don't get messages out of order, and no recursion */ 375 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { 376 struct netdev_queue *txq; 377 378 txq = netdev_pick_tx(dev, skb); 379 380 /* try until next clock tick */ 381 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; 382 tries > 0; --tries) { 383 if (__netif_tx_trylock(txq)) { 384 if (!netif_xmit_stopped(txq)) { 385 if (vlan_tx_tag_present(skb) && 386 !(netif_skb_features(skb) & NETIF_F_HW_VLAN_TX)) { 387 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb)); 388 if (unlikely(!skb)) 389 break; 390 skb->vlan_tci = 0; 391 } 392 393 status = ops->ndo_start_xmit(skb, dev); 394 if (status == NETDEV_TX_OK) 395 txq_trans_update(txq); 396 } 397 __netif_tx_unlock(txq); 398 399 if (status == NETDEV_TX_OK) 400 break; 401 402 } 403 404 /* tickle device maybe there is some cleanup */ 405 netpoll_poll_dev(np->dev); 406 407 udelay(USEC_PER_POLL); 408 } 409 410 WARN_ONCE(!irqs_disabled(), 411 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n", 412 dev->name, ops->ndo_start_xmit); 413 414 } 415 416 if (status != NETDEV_TX_OK) { 417 skb_queue_tail(&npinfo->txq, skb); 418 schedule_delayed_work(&npinfo->tx_work,0); 419 } 420} 421EXPORT_SYMBOL(netpoll_send_skb_on_dev); 422 423void netpoll_send_udp(struct netpoll *np, const char *msg, int len) 424{ 425 int total_len, ip_len, udp_len; 426 struct sk_buff *skb; 427 struct udphdr *udph; 428 struct iphdr *iph; 429 struct ethhdr *eth; 430 static atomic_t ip_ident; 431 struct ipv6hdr *ip6h; 432 433 udp_len = len + sizeof(*udph); 434 if (np->ipv6) 435 ip_len = udp_len + sizeof(*ip6h); 436 else 437 ip_len = udp_len + sizeof(*iph); 438 439 total_len = ip_len + LL_RESERVED_SPACE(np->dev); 440 441 skb = find_skb(np, total_len + np->dev->needed_tailroom, 442 total_len - len); 443 if (!skb) 444 return; 445 446 skb_copy_to_linear_data(skb, msg, len); 447 skb_put(skb, len); 448 449 skb_push(skb, sizeof(*udph)); 450 skb_reset_transport_header(skb); 451 udph = udp_hdr(skb); 452 udph->source = htons(np->local_port); 453 udph->dest = htons(np->remote_port); 454 udph->len = htons(udp_len); 455 456 if (np->ipv6) { 457 udph->check = 0; 458 udph->check = csum_ipv6_magic(&np->local_ip.in6, 459 &np->remote_ip.in6, 460 udp_len, IPPROTO_UDP, 461 csum_partial(udph, udp_len, 0)); 462 if (udph->check == 0) 463 udph->check = CSUM_MANGLED_0; 464 465 skb_push(skb, sizeof(*ip6h)); 466 skb_reset_network_header(skb); 467 ip6h = ipv6_hdr(skb); 468 469 /* ip6h->version = 6; ip6h->priority = 0; */ 470 put_unaligned(0x60, (unsigned char *)ip6h); 471 ip6h->flow_lbl[0] = 0; 472 ip6h->flow_lbl[1] = 0; 473 ip6h->flow_lbl[2] = 0; 474 475 ip6h->payload_len = htons(sizeof(struct udphdr) + len); 476 ip6h->nexthdr = IPPROTO_UDP; 477 ip6h->hop_limit = 32; 478 ip6h->saddr = np->local_ip.in6; 479 ip6h->daddr = np->remote_ip.in6; 480 481 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 482 skb_reset_mac_header(skb); 483 skb->protocol = eth->h_proto = htons(ETH_P_IPV6); 484 } else { 485 udph->check = 0; 486 udph->check = csum_tcpudp_magic(np->local_ip.ip, 487 np->remote_ip.ip, 488 udp_len, IPPROTO_UDP, 489 csum_partial(udph, udp_len, 0)); 490 if (udph->check == 0) 491 udph->check = CSUM_MANGLED_0; 492 493 skb_push(skb, sizeof(*iph)); 494 skb_reset_network_header(skb); 495 iph = ip_hdr(skb); 496 497 /* iph->version = 4; iph->ihl = 5; */ 498 put_unaligned(0x45, (unsigned char *)iph); 499 iph->tos = 0; 500 put_unaligned(htons(ip_len), &(iph->tot_len)); 501 iph->id = htons(atomic_inc_return(&ip_ident)); 502 iph->frag_off = 0; 503 iph->ttl = 64; 504 iph->protocol = IPPROTO_UDP; 505 iph->check = 0; 506 put_unaligned(np->local_ip.ip, &(iph->saddr)); 507 put_unaligned(np->remote_ip.ip, &(iph->daddr)); 508 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 509 510 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN); 511 skb_reset_mac_header(skb); 512 skb->protocol = eth->h_proto = htons(ETH_P_IP); 513 } 514 515 memcpy(eth->h_source, np->dev->dev_addr, ETH_ALEN); 516 memcpy(eth->h_dest, np->remote_mac, ETH_ALEN); 517 518 skb->dev = np->dev; 519 520 netpoll_send_skb(np, skb); 521} 522EXPORT_SYMBOL(netpoll_send_udp); 523 524static void netpoll_neigh_reply(struct sk_buff *skb, struct netpoll_info *npinfo) 525{ 526 int size, type = ARPOP_REPLY; 527 __be32 sip, tip; 528 unsigned char *sha; 529 struct sk_buff *send_skb; 530 struct netpoll *np, *tmp; 531 unsigned long flags; 532 int hlen, tlen; 533 int hits = 0, proto; 534 535 if (list_empty(&npinfo->rx_np)) 536 return; 537 538 /* Before checking the packet, we do some early 539 inspection whether this is interesting at all */ 540 spin_lock_irqsave(&npinfo->rx_lock, flags); 541 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 542 if (np->dev == skb->dev) 543 hits++; 544 } 545 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 546 547 /* No netpoll struct is using this dev */ 548 if (!hits) 549 return; 550 551 proto = ntohs(eth_hdr(skb)->h_proto); 552 if (proto == ETH_P_IP) { 553 struct arphdr *arp; 554 unsigned char *arp_ptr; 555 /* No arp on this interface */ 556 if (skb->dev->flags & IFF_NOARP) 557 return; 558 559 if (!pskb_may_pull(skb, arp_hdr_len(skb->dev))) 560 return; 561 562 skb_reset_network_header(skb); 563 skb_reset_transport_header(skb); 564 arp = arp_hdr(skb); 565 566 if ((arp->ar_hrd != htons(ARPHRD_ETHER) && 567 arp->ar_hrd != htons(ARPHRD_IEEE802)) || 568 arp->ar_pro != htons(ETH_P_IP) || 569 arp->ar_op != htons(ARPOP_REQUEST)) 570 return; 571 572 arp_ptr = (unsigned char *)(arp+1); 573 /* save the location of the src hw addr */ 574 sha = arp_ptr; 575 arp_ptr += skb->dev->addr_len; 576 memcpy(&sip, arp_ptr, 4); 577 arp_ptr += 4; 578 /* If we actually cared about dst hw addr, 579 it would get copied here */ 580 arp_ptr += skb->dev->addr_len; 581 memcpy(&tip, arp_ptr, 4); 582 583 /* Should we ignore arp? */ 584 if (ipv4_is_loopback(tip) || ipv4_is_multicast(tip)) 585 return; 586 587 size = arp_hdr_len(skb->dev); 588 589 spin_lock_irqsave(&npinfo->rx_lock, flags); 590 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 591 if (tip != np->local_ip.ip) 592 continue; 593 594 hlen = LL_RESERVED_SPACE(np->dev); 595 tlen = np->dev->needed_tailroom; 596 send_skb = find_skb(np, size + hlen + tlen, hlen); 597 if (!send_skb) 598 continue; 599 600 skb_reset_network_header(send_skb); 601 arp = (struct arphdr *) skb_put(send_skb, size); 602 send_skb->dev = skb->dev; 603 send_skb->protocol = htons(ETH_P_ARP); 604 605 /* Fill the device header for the ARP frame */ 606 if (dev_hard_header(send_skb, skb->dev, ETH_P_ARP, 607 sha, np->dev->dev_addr, 608 send_skb->len) < 0) { 609 kfree_skb(send_skb); 610 continue; 611 } 612 613 /* 614 * Fill out the arp protocol part. 615 * 616 * we only support ethernet device type, 617 * which (according to RFC 1390) should 618 * always equal 1 (Ethernet). 619 */ 620 621 arp->ar_hrd = htons(np->dev->type); 622 arp->ar_pro = htons(ETH_P_IP); 623 arp->ar_hln = np->dev->addr_len; 624 arp->ar_pln = 4; 625 arp->ar_op = htons(type); 626 627 arp_ptr = (unsigned char *)(arp + 1); 628 memcpy(arp_ptr, np->dev->dev_addr, np->dev->addr_len); 629 arp_ptr += np->dev->addr_len; 630 memcpy(arp_ptr, &tip, 4); 631 arp_ptr += 4; 632 memcpy(arp_ptr, sha, np->dev->addr_len); 633 arp_ptr += np->dev->addr_len; 634 memcpy(arp_ptr, &sip, 4); 635 636 netpoll_send_skb(np, send_skb); 637 638 /* If there are several rx_hooks for the same address, 639 we're fine by sending a single reply */ 640 break; 641 } 642 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 643 } else if( proto == ETH_P_IPV6) { 644#if IS_ENABLED(CONFIG_IPV6) 645 struct nd_msg *msg; 646 u8 *lladdr = NULL; 647 struct ipv6hdr *hdr; 648 struct icmp6hdr *icmp6h; 649 const struct in6_addr *saddr; 650 const struct in6_addr *daddr; 651 struct inet6_dev *in6_dev = NULL; 652 struct in6_addr *target; 653 654 in6_dev = in6_dev_get(skb->dev); 655 if (!in6_dev || !in6_dev->cnf.accept_ra) 656 return; 657 658 if (!pskb_may_pull(skb, skb->len)) 659 return; 660 661 msg = (struct nd_msg *)skb_transport_header(skb); 662 663 __skb_push(skb, skb->data - skb_transport_header(skb)); 664 665 if (ipv6_hdr(skb)->hop_limit != 255) 666 return; 667 if (msg->icmph.icmp6_code != 0) 668 return; 669 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION) 670 return; 671 672 saddr = &ipv6_hdr(skb)->saddr; 673 daddr = &ipv6_hdr(skb)->daddr; 674 675 size = sizeof(struct icmp6hdr) + sizeof(struct in6_addr); 676 677 spin_lock_irqsave(&npinfo->rx_lock, flags); 678 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 679 if (!ipv6_addr_equal(daddr, &np->local_ip.in6)) 680 continue; 681 682 hlen = LL_RESERVED_SPACE(np->dev); 683 tlen = np->dev->needed_tailroom; 684 send_skb = find_skb(np, size + hlen + tlen, hlen); 685 if (!send_skb) 686 continue; 687 688 send_skb->protocol = htons(ETH_P_IPV6); 689 send_skb->dev = skb->dev; 690 691 skb_reset_network_header(send_skb); 692 skb_put(send_skb, sizeof(struct ipv6hdr)); 693 hdr = ipv6_hdr(send_skb); 694 695 *(__be32*)hdr = htonl(0x60000000); 696 697 hdr->payload_len = htons(size); 698 hdr->nexthdr = IPPROTO_ICMPV6; 699 hdr->hop_limit = 255; 700 hdr->saddr = *saddr; 701 hdr->daddr = *daddr; 702 703 send_skb->transport_header = send_skb->tail; 704 skb_put(send_skb, size); 705 706 icmp6h = (struct icmp6hdr *)skb_transport_header(skb); 707 icmp6h->icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 708 icmp6h->icmp6_router = 0; 709 icmp6h->icmp6_solicited = 1; 710 target = (struct in6_addr *)(skb_transport_header(send_skb) + sizeof(struct icmp6hdr)); 711 *target = msg->target; 712 icmp6h->icmp6_cksum = csum_ipv6_magic(saddr, daddr, size, 713 IPPROTO_ICMPV6, 714 csum_partial(icmp6h, 715 size, 0)); 716 717 if (dev_hard_header(send_skb, skb->dev, ETH_P_IPV6, 718 lladdr, np->dev->dev_addr, 719 send_skb->len) < 0) { 720 kfree_skb(send_skb); 721 continue; 722 } 723 724 netpoll_send_skb(np, send_skb); 725 726 /* If there are several rx_hooks for the same address, 727 we're fine by sending a single reply */ 728 break; 729 } 730 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 731#endif 732 } 733} 734 735static bool pkt_is_ns(struct sk_buff *skb) 736{ 737 struct nd_msg *msg; 738 struct ipv6hdr *hdr; 739 740 if (skb->protocol != htons(ETH_P_ARP)) 741 return false; 742 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr) + sizeof(struct nd_msg))) 743 return false; 744 745 msg = (struct nd_msg *)skb_transport_header(skb); 746 __skb_push(skb, skb->data - skb_transport_header(skb)); 747 hdr = ipv6_hdr(skb); 748 749 if (hdr->nexthdr != IPPROTO_ICMPV6) 750 return false; 751 if (hdr->hop_limit != 255) 752 return false; 753 if (msg->icmph.icmp6_code != 0) 754 return false; 755 if (msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION) 756 return false; 757 758 return true; 759} 760 761int __netpoll_rx(struct sk_buff *skb, struct netpoll_info *npinfo) 762{ 763 int proto, len, ulen; 764 int hits = 0; 765 const struct iphdr *iph; 766 struct udphdr *uh; 767 struct netpoll *np, *tmp; 768 769 if (list_empty(&npinfo->rx_np)) 770 goto out; 771 772 if (skb->dev->type != ARPHRD_ETHER) 773 goto out; 774 775 /* check if netpoll clients need ARP */ 776 if (skb->protocol == htons(ETH_P_ARP) && atomic_read(&trapped)) { 777 skb_queue_tail(&npinfo->neigh_tx, skb); 778 return 1; 779 } else if (pkt_is_ns(skb) && atomic_read(&trapped)) { 780 skb_queue_tail(&npinfo->neigh_tx, skb); 781 return 1; 782 } 783 784 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) { 785 skb = vlan_untag(skb); 786 if (unlikely(!skb)) 787 goto out; 788 } 789 790 proto = ntohs(eth_hdr(skb)->h_proto); 791 if (proto != ETH_P_IP && proto != ETH_P_IPV6) 792 goto out; 793 if (skb->pkt_type == PACKET_OTHERHOST) 794 goto out; 795 if (skb_shared(skb)) 796 goto out; 797 798 if (proto == ETH_P_IP) { 799 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 800 goto out; 801 iph = (struct iphdr *)skb->data; 802 if (iph->ihl < 5 || iph->version != 4) 803 goto out; 804 if (!pskb_may_pull(skb, iph->ihl*4)) 805 goto out; 806 iph = (struct iphdr *)skb->data; 807 if (ip_fast_csum((u8 *)iph, iph->ihl) != 0) 808 goto out; 809 810 len = ntohs(iph->tot_len); 811 if (skb->len < len || len < iph->ihl*4) 812 goto out; 813 814 /* 815 * Our transport medium may have padded the buffer out. 816 * Now We trim to the true length of the frame. 817 */ 818 if (pskb_trim_rcsum(skb, len)) 819 goto out; 820 821 iph = (struct iphdr *)skb->data; 822 if (iph->protocol != IPPROTO_UDP) 823 goto out; 824 825 len -= iph->ihl*4; 826 uh = (struct udphdr *)(((char *)iph) + iph->ihl*4); 827 ulen = ntohs(uh->len); 828 829 if (ulen != len) 830 goto out; 831 if (checksum_udp(skb, uh, ulen, iph->saddr, iph->daddr)) 832 goto out; 833 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 834 if (np->local_ip.ip && np->local_ip.ip != iph->daddr) 835 continue; 836 if (np->remote_ip.ip && np->remote_ip.ip != iph->saddr) 837 continue; 838 if (np->local_port && np->local_port != ntohs(uh->dest)) 839 continue; 840 841 np->rx_hook(np, ntohs(uh->source), 842 (char *)(uh+1), 843 ulen - sizeof(struct udphdr)); 844 hits++; 845 } 846 } else { 847#if IS_ENABLED(CONFIG_IPV6) 848 const struct ipv6hdr *ip6h; 849 850 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) 851 goto out; 852 ip6h = (struct ipv6hdr *)skb->data; 853 if (ip6h->version != 6) 854 goto out; 855 len = ntohs(ip6h->payload_len); 856 if (!len) 857 goto out; 858 if (len + sizeof(struct ipv6hdr) > skb->len) 859 goto out; 860 if (pskb_trim_rcsum(skb, len + sizeof(struct ipv6hdr))) 861 goto out; 862 ip6h = ipv6_hdr(skb); 863 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 864 goto out; 865 uh = udp_hdr(skb); 866 ulen = ntohs(uh->len); 867 if (ulen != skb->len) 868 goto out; 869 if (udp6_csum_init(skb, uh, IPPROTO_UDP)) 870 goto out; 871 list_for_each_entry_safe(np, tmp, &npinfo->rx_np, rx) { 872 if (!ipv6_addr_equal(&np->local_ip.in6, &ip6h->daddr)) 873 continue; 874 if (!ipv6_addr_equal(&np->remote_ip.in6, &ip6h->saddr)) 875 continue; 876 if (np->local_port && np->local_port != ntohs(uh->dest)) 877 continue; 878 879 np->rx_hook(np, ntohs(uh->source), 880 (char *)(uh+1), 881 ulen - sizeof(struct udphdr)); 882 hits++; 883 } 884#endif 885 } 886 887 if (!hits) 888 goto out; 889 890 kfree_skb(skb); 891 return 1; 892 893out: 894 if (atomic_read(&trapped)) { 895 kfree_skb(skb); 896 return 1; 897 } 898 899 return 0; 900} 901 902void netpoll_print_options(struct netpoll *np) 903{ 904 np_info(np, "local port %d\n", np->local_port); 905 if (np->ipv6) 906 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6); 907 else 908 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip); 909 np_info(np, "interface '%s'\n", np->dev_name); 910 np_info(np, "remote port %d\n", np->remote_port); 911 if (np->ipv6) 912 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6); 913 else 914 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip); 915 np_info(np, "remote ethernet address %pM\n", np->remote_mac); 916} 917EXPORT_SYMBOL(netpoll_print_options); 918 919static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr) 920{ 921 const char *end; 922 923 if (!strchr(str, ':') && 924 in4_pton(str, -1, (void *)addr, -1, &end) > 0) { 925 if (!*end) 926 return 0; 927 } 928 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) { 929#if IS_ENABLED(CONFIG_IPV6) 930 if (!*end) 931 return 1; 932#else 933 return -1; 934#endif 935 } 936 return -1; 937} 938 939int netpoll_parse_options(struct netpoll *np, char *opt) 940{ 941 char *cur=opt, *delim; 942 int ipv6; 943 944 if (*cur != '@') { 945 if ((delim = strchr(cur, '@')) == NULL) 946 goto parse_failed; 947 *delim = 0; 948 if (kstrtou16(cur, 10, &np->local_port)) 949 goto parse_failed; 950 cur = delim; 951 } 952 cur++; 953 954 if (*cur != '/') { 955 if ((delim = strchr(cur, '/')) == NULL) 956 goto parse_failed; 957 *delim = 0; 958 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip); 959 if (ipv6 < 0) 960 goto parse_failed; 961 else 962 np->ipv6 = (bool)ipv6; 963 cur = delim; 964 } 965 cur++; 966 967 if (*cur != ',') { 968 /* parse out dev name */ 969 if ((delim = strchr(cur, ',')) == NULL) 970 goto parse_failed; 971 *delim = 0; 972 strlcpy(np->dev_name, cur, sizeof(np->dev_name)); 973 cur = delim; 974 } 975 cur++; 976 977 if (*cur != '@') { 978 /* dst port */ 979 if ((delim = strchr(cur, '@')) == NULL) 980 goto parse_failed; 981 *delim = 0; 982 if (*cur == ' ' || *cur == '\t') 983 np_info(np, "warning: whitespace is not allowed\n"); 984 if (kstrtou16(cur, 10, &np->remote_port)) 985 goto parse_failed; 986 cur = delim; 987 } 988 cur++; 989 990 /* dst ip */ 991 if ((delim = strchr(cur, '/')) == NULL) 992 goto parse_failed; 993 *delim = 0; 994 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip); 995 if (ipv6 < 0) 996 goto parse_failed; 997 else if (np->ipv6 != (bool)ipv6) 998 goto parse_failed; 999 else 1000 np->ipv6 = (bool)ipv6; 1001 cur = delim + 1; 1002 1003 if (*cur != 0) { 1004 /* MAC address */ 1005 if (!mac_pton(cur, np->remote_mac)) 1006 goto parse_failed; 1007 } 1008 1009 netpoll_print_options(np); 1010 1011 return 0; 1012 1013 parse_failed: 1014 np_info(np, "couldn't parse config at '%s'!\n", cur); 1015 return -1; 1016} 1017EXPORT_SYMBOL(netpoll_parse_options); 1018 1019int __netpoll_setup(struct netpoll *np, struct net_device *ndev, gfp_t gfp) 1020{ 1021 struct netpoll_info *npinfo; 1022 const struct net_device_ops *ops; 1023 unsigned long flags; 1024 int err; 1025 1026 np->dev = ndev; 1027 strlcpy(np->dev_name, ndev->name, IFNAMSIZ); 1028 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup); 1029 1030 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) || 1031 !ndev->netdev_ops->ndo_poll_controller) { 1032 np_err(np, "%s doesn't support polling, aborting\n", 1033 np->dev_name); 1034 err = -ENOTSUPP; 1035 goto out; 1036 } 1037 1038 if (!ndev->npinfo) { 1039 npinfo = kmalloc(sizeof(*npinfo), gfp); 1040 if (!npinfo) { 1041 err = -ENOMEM; 1042 goto out; 1043 } 1044 1045 npinfo->rx_flags = 0; 1046 INIT_LIST_HEAD(&npinfo->rx_np); 1047 1048 spin_lock_init(&npinfo->rx_lock); 1049 mutex_init(&npinfo->dev_lock); 1050 skb_queue_head_init(&npinfo->neigh_tx); 1051 skb_queue_head_init(&npinfo->txq); 1052 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); 1053 1054 atomic_set(&npinfo->refcnt, 1); 1055 1056 ops = np->dev->netdev_ops; 1057 if (ops->ndo_netpoll_setup) { 1058 err = ops->ndo_netpoll_setup(ndev, npinfo, gfp); 1059 if (err) 1060 goto free_npinfo; 1061 } 1062 } else { 1063 npinfo = rtnl_dereference(ndev->npinfo); 1064 atomic_inc(&npinfo->refcnt); 1065 } 1066 1067 npinfo->netpoll = np; 1068 1069 if (np->rx_hook) { 1070 spin_lock_irqsave(&npinfo->rx_lock, flags); 1071 npinfo->rx_flags |= NETPOLL_RX_ENABLED; 1072 list_add_tail(&np->rx, &npinfo->rx_np); 1073 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 1074 } 1075 1076 /* last thing to do is link it to the net device structure */ 1077 rcu_assign_pointer(ndev->npinfo, npinfo); 1078 1079 return 0; 1080 1081free_npinfo: 1082 kfree(npinfo); 1083out: 1084 return err; 1085} 1086EXPORT_SYMBOL_GPL(__netpoll_setup); 1087 1088int netpoll_setup(struct netpoll *np) 1089{ 1090 struct net_device *ndev = NULL; 1091 struct in_device *in_dev; 1092 int err; 1093 1094 rtnl_lock(); 1095 if (np->dev_name) { 1096 struct net *net = current->nsproxy->net_ns; 1097 ndev = __dev_get_by_name(net, np->dev_name); 1098 } 1099 if (!ndev) { 1100 np_err(np, "%s doesn't exist, aborting\n", np->dev_name); 1101 err = -ENODEV; 1102 goto unlock; 1103 } 1104 dev_hold(ndev); 1105 1106 if (netdev_master_upper_dev_get(ndev)) { 1107 np_err(np, "%s is a slave device, aborting\n", np->dev_name); 1108 err = -EBUSY; 1109 goto put; 1110 } 1111 1112 if (!netif_running(ndev)) { 1113 unsigned long atmost, atleast; 1114 1115 np_info(np, "device %s not up yet, forcing it\n", np->dev_name); 1116 1117 err = dev_open(ndev); 1118 1119 if (err) { 1120 np_err(np, "failed to open %s\n", ndev->name); 1121 goto put; 1122 } 1123 1124 rtnl_unlock(); 1125 atleast = jiffies + HZ/10; 1126 atmost = jiffies + carrier_timeout * HZ; 1127 while (!netif_carrier_ok(ndev)) { 1128 if (time_after(jiffies, atmost)) { 1129 np_notice(np, "timeout waiting for carrier\n"); 1130 break; 1131 } 1132 msleep(1); 1133 } 1134 1135 /* If carrier appears to come up instantly, we don't 1136 * trust it and pause so that we don't pump all our 1137 * queued console messages into the bitbucket. 1138 */ 1139 1140 if (time_before(jiffies, atleast)) { 1141 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n"); 1142 msleep(4000); 1143 } 1144 rtnl_lock(); 1145 } 1146 1147 if (!np->local_ip.ip) { 1148 if (!np->ipv6) { 1149 in_dev = __in_dev_get_rtnl(ndev); 1150 1151 if (!in_dev || !in_dev->ifa_list) { 1152 np_err(np, "no IP address for %s, aborting\n", 1153 np->dev_name); 1154 err = -EDESTADDRREQ; 1155 goto put; 1156 } 1157 1158 np->local_ip.ip = in_dev->ifa_list->ifa_local; 1159 np_info(np, "local IP %pI4\n", &np->local_ip.ip); 1160 } else { 1161#if IS_ENABLED(CONFIG_IPV6) 1162 struct inet6_dev *idev; 1163 1164 err = -EDESTADDRREQ; 1165 idev = __in6_dev_get(ndev); 1166 if (idev) { 1167 struct inet6_ifaddr *ifp; 1168 1169 read_lock_bh(&idev->lock); 1170 list_for_each_entry(ifp, &idev->addr_list, if_list) { 1171 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) 1172 continue; 1173 np->local_ip.in6 = ifp->addr; 1174 err = 0; 1175 break; 1176 } 1177 read_unlock_bh(&idev->lock); 1178 } 1179 if (err) { 1180 np_err(np, "no IPv6 address for %s, aborting\n", 1181 np->dev_name); 1182 goto put; 1183 } else 1184 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6); 1185#else 1186 np_err(np, "IPv6 is not supported %s, aborting\n", 1187 np->dev_name); 1188 err = -EINVAL; 1189 goto put; 1190#endif 1191 } 1192 } 1193 1194 /* fill up the skb queue */ 1195 refill_skbs(); 1196 1197 err = __netpoll_setup(np, ndev, GFP_KERNEL); 1198 if (err) 1199 goto put; 1200 1201 rtnl_unlock(); 1202 return 0; 1203 1204put: 1205 dev_put(ndev); 1206unlock: 1207 rtnl_unlock(); 1208 return err; 1209} 1210EXPORT_SYMBOL(netpoll_setup); 1211 1212static int __init netpoll_init(void) 1213{ 1214 skb_queue_head_init(&skb_pool); 1215 return 0; 1216} 1217core_initcall(netpoll_init); 1218 1219static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head) 1220{ 1221 struct netpoll_info *npinfo = 1222 container_of(rcu_head, struct netpoll_info, rcu); 1223 1224 skb_queue_purge(&npinfo->neigh_tx); 1225 skb_queue_purge(&npinfo->txq); 1226 1227 /* we can't call cancel_delayed_work_sync here, as we are in softirq */ 1228 cancel_delayed_work(&npinfo->tx_work); 1229 1230 /* clean after last, unfinished work */ 1231 __skb_queue_purge(&npinfo->txq); 1232 /* now cancel it again */ 1233 cancel_delayed_work(&npinfo->tx_work); 1234 kfree(npinfo); 1235} 1236 1237void __netpoll_cleanup(struct netpoll *np) 1238{ 1239 struct netpoll_info *npinfo; 1240 unsigned long flags; 1241 1242 /* rtnl_dereference would be preferable here but 1243 * rcu_cleanup_netpoll path can put us in here safely without 1244 * holding the rtnl, so plain rcu_dereference it is 1245 */ 1246 npinfo = rtnl_dereference(np->dev->npinfo); 1247 if (!npinfo) 1248 return; 1249 1250 if (!list_empty(&npinfo->rx_np)) { 1251 spin_lock_irqsave(&npinfo->rx_lock, flags); 1252 list_del(&np->rx); 1253 if (list_empty(&npinfo->rx_np)) 1254 npinfo->rx_flags &= ~NETPOLL_RX_ENABLED; 1255 spin_unlock_irqrestore(&npinfo->rx_lock, flags); 1256 } 1257 1258 synchronize_srcu(&netpoll_srcu); 1259 1260 if (atomic_dec_and_test(&npinfo->refcnt)) { 1261 const struct net_device_ops *ops; 1262 1263 ops = np->dev->netdev_ops; 1264 if (ops->ndo_netpoll_cleanup) 1265 ops->ndo_netpoll_cleanup(np->dev); 1266 1267 rcu_assign_pointer(np->dev->npinfo, NULL); 1268 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info); 1269 } 1270} 1271EXPORT_SYMBOL_GPL(__netpoll_cleanup); 1272 1273static void netpoll_async_cleanup(struct work_struct *work) 1274{ 1275 struct netpoll *np = container_of(work, struct netpoll, cleanup_work); 1276 1277 rtnl_lock(); 1278 __netpoll_cleanup(np); 1279 rtnl_unlock(); 1280 kfree(np); 1281} 1282 1283void __netpoll_free_async(struct netpoll *np) 1284{ 1285 schedule_work(&np->cleanup_work); 1286} 1287EXPORT_SYMBOL_GPL(__netpoll_free_async); 1288 1289void netpoll_cleanup(struct netpoll *np) 1290{ 1291 if (!np->dev) 1292 return; 1293 1294 rtnl_lock(); 1295 __netpoll_cleanup(np); 1296 rtnl_unlock(); 1297 1298 dev_put(np->dev); 1299 np->dev = NULL; 1300} 1301EXPORT_SYMBOL(netpoll_cleanup); 1302 1303int netpoll_trap(void) 1304{ 1305 return atomic_read(&trapped); 1306} 1307EXPORT_SYMBOL(netpoll_trap); 1308 1309void netpoll_set_trap(int trap) 1310{ 1311 if (trap) 1312 atomic_inc(&trapped); 1313 else 1314 atomic_dec(&trapped); 1315} 1316EXPORT_SYMBOL(netpoll_set_trap); 1317