1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Common framework for low-level network console, dump, and debugger code 4 * 5 * Sep 8 2003 Matt Mackall <mpm@selenic.com> 6 * 7 * based on the netconsole code from: 8 * 9 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com> 10 * Copyright (C) 2002 Red Hat, Inc. 11 */ 12 13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 14 15 #include <linux/moduleparam.h> 16 #include <linux/kernel.h> 17 #include <linux/netdevice.h> 18 #include <linux/etherdevice.h> 19 #include <linux/string.h> 20 #include <linux/if_arp.h> 21 #include <linux/inetdevice.h> 22 #include <linux/inet.h> 23 #include <linux/interrupt.h> 24 #include <linux/netpoll.h> 25 #include <linux/sched.h> 26 #include <linux/delay.h> 27 #include <linux/rcupdate.h> 28 #include <linux/workqueue.h> 29 #include <linux/slab.h> 30 #include <linux/export.h> 31 #include <linux/if_vlan.h> 32 #include <net/tcp.h> 33 #include <net/udp.h> 34 #include <net/addrconf.h> 35 #include <net/ndisc.h> 36 #include <net/ip6_checksum.h> 37 #include <linux/unaligned.h> 38 #include <trace/events/napi.h> 39 #include <linux/kconfig.h> 40 41 /* 42 * We maintain a small pool of fully-sized skbs, to make sure the 43 * message gets out even in extreme OOM situations. 44 */ 45 46 #define MAX_UDP_CHUNK 1460 47 #define MAX_SKBS 32 48 #define USEC_PER_POLL 50 49 50 #define MAX_SKB_SIZE \ 51 (sizeof(struct ethhdr) + \ 52 sizeof(struct iphdr) + \ 53 sizeof(struct udphdr) + \ 54 MAX_UDP_CHUNK) 55 56 static void zap_completion_queue(void); 57 58 static unsigned int carrier_timeout = 4; 59 module_param(carrier_timeout, uint, 0644); 60 61 static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb, 62 struct net_device *dev, 63 struct netdev_queue *txq) 64 { 65 netdev_tx_t status = NETDEV_TX_OK; 66 netdev_features_t features; 67 68 features = netif_skb_features(skb); 69 70 if (skb_vlan_tag_present(skb) && 71 !vlan_hw_offload_capable(features, skb->vlan_proto)) { 72 skb = __vlan_hwaccel_push_inside(skb); 73 if (unlikely(!skb)) { 74 /* This is actually a packet drop, but we 75 * don't want the code that calls this 76 * function to try and operate on a NULL skb. 77 */ 78 goto out; 79 } 80 } 81 82 status = netdev_start_xmit(skb, dev, txq, false); 83 84 out: 85 return status; 86 } 87 88 static void queue_process(struct work_struct *work) 89 { 90 struct netpoll_info *npinfo = 91 container_of(work, struct netpoll_info, tx_work.work); 92 struct sk_buff *skb; 93 unsigned long flags; 94 95 while ((skb = skb_dequeue(&npinfo->txq))) { 96 struct net_device *dev = skb->dev; 97 struct netdev_queue *txq; 98 unsigned int q_index; 99 100 if (!netif_device_present(dev) || !netif_running(dev)) { 101 kfree_skb(skb); 102 continue; 103 } 104 105 local_irq_save(flags); 106 /* check if skb->queue_mapping is still valid */ 107 q_index = skb_get_queue_mapping(skb); 108 if (unlikely(q_index >= dev->real_num_tx_queues)) { 109 q_index = q_index % dev->real_num_tx_queues; 110 skb_set_queue_mapping(skb, q_index); 111 } 112 txq = netdev_get_tx_queue(dev, q_index); 113 HARD_TX_LOCK(dev, txq, smp_processor_id()); 114 if (netif_xmit_frozen_or_stopped(txq) || 115 !dev_xmit_complete(netpoll_start_xmit(skb, dev, txq))) { 116 skb_queue_head(&npinfo->txq, skb); 117 HARD_TX_UNLOCK(dev, txq); 118 local_irq_restore(flags); 119 120 schedule_delayed_work(&npinfo->tx_work, HZ/10); 121 return; 122 } 123 HARD_TX_UNLOCK(dev, txq); 124 local_irq_restore(flags); 125 } 126 } 127 128 static int netif_local_xmit_active(struct net_device *dev) 129 { 130 int i; 131 132 for (i = 0; i < dev->num_tx_queues; i++) { 133 struct netdev_queue *txq = netdev_get_tx_queue(dev, i); 134 135 if (netif_tx_owned(txq, smp_processor_id())) 136 return 1; 137 } 138 139 return 0; 140 } 141 142 static void poll_one_napi(struct napi_struct *napi) 143 { 144 int work; 145 146 /* If we set this bit but see that it has already been set, 147 * that indicates that napi has been disabled and we need 148 * to abort this operation 149 */ 150 if (test_and_set_bit(NAPI_STATE_NPSVC, &napi->state)) 151 return; 152 153 /* We explicitly pass the polling call a budget of 0 to 154 * indicate that we are clearing the Tx path only. 155 */ 156 work = napi->poll(napi, 0); 157 WARN_ONCE(work, "%pS exceeded budget in poll\n", napi->poll); 158 trace_napi_poll(napi, work, 0); 159 160 clear_bit(NAPI_STATE_NPSVC, &napi->state); 161 } 162 163 static void poll_napi(struct net_device *dev) 164 { 165 struct napi_struct *napi; 166 int cpu = smp_processor_id(); 167 168 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) { 169 if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) { 170 poll_one_napi(napi); 171 smp_store_release(&napi->poll_owner, -1); 172 } 173 } 174 } 175 176 void netpoll_poll_dev(struct net_device *dev) 177 { 178 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo); 179 const struct net_device_ops *ops; 180 181 /* Don't do any rx activity if the dev_lock mutex is held 182 * the dev_open/close paths use this to block netpoll activity 183 * while changing device state 184 */ 185 if (!ni || down_trylock(&ni->dev_lock)) 186 return; 187 188 /* Some drivers will take the same locks in poll and xmit, 189 * we can't poll if local CPU is already in xmit. 190 */ 191 if (!netif_running(dev) || netif_local_xmit_active(dev)) { 192 up(&ni->dev_lock); 193 return; 194 } 195 196 ops = dev->netdev_ops; 197 if (ops->ndo_poll_controller) 198 ops->ndo_poll_controller(dev); 199 200 poll_napi(dev); 201 202 up(&ni->dev_lock); 203 204 zap_completion_queue(); 205 } 206 EXPORT_SYMBOL(netpoll_poll_dev); 207 208 void netpoll_poll_disable(struct net_device *dev) 209 { 210 struct netpoll_info *ni; 211 212 might_sleep(); 213 ni = rtnl_dereference(dev->npinfo); 214 if (ni) 215 down(&ni->dev_lock); 216 } 217 218 void netpoll_poll_enable(struct net_device *dev) 219 { 220 struct netpoll_info *ni; 221 222 ni = rtnl_dereference(dev->npinfo); 223 if (ni) 224 up(&ni->dev_lock); 225 } 226 227 static void refill_skbs(struct netpoll *np) 228 { 229 struct sk_buff_head *skb_pool; 230 struct sk_buff *skb; 231 232 skb_pool = &np->skb_pool; 233 234 while (READ_ONCE(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 } 242 243 static void zap_completion_queue(void) 244 { 245 unsigned long flags; 246 struct softnet_data *sd = &get_cpu_var(softnet_data); 247 248 if (sd->completion_queue) { 249 struct sk_buff *clist; 250 251 local_irq_save(flags); 252 clist = sd->completion_queue; 253 sd->completion_queue = NULL; 254 local_irq_restore(flags); 255 256 while (clist != NULL) { 257 struct sk_buff *skb = clist; 258 clist = clist->next; 259 if (!skb_irq_freeable(skb)) { 260 refcount_set(&skb->users, 1); 261 dev_kfree_skb_any(skb); /* put this one back */ 262 } else { 263 __kfree_skb(skb); 264 } 265 } 266 } 267 268 put_cpu_var(softnet_data); 269 } 270 271 static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) 272 { 273 int count = 0; 274 struct sk_buff *skb; 275 276 zap_completion_queue(); 277 repeat: 278 279 skb = alloc_skb(len, GFP_ATOMIC); 280 if (!skb) { 281 skb = skb_dequeue(&np->skb_pool); 282 schedule_work(&np->refill_wq); 283 } 284 285 if (!skb) { 286 if (++count < 10) { 287 netpoll_poll_dev(np->dev); 288 goto repeat; 289 } 290 return NULL; 291 } 292 293 refcount_set(&skb->users, 1); 294 skb_reserve(skb, reserve); 295 return skb; 296 } 297 298 static int netpoll_owner_active(struct net_device *dev) 299 { 300 struct napi_struct *napi; 301 302 list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) { 303 if (READ_ONCE(napi->poll_owner) == smp_processor_id()) 304 return 1; 305 } 306 return 0; 307 } 308 309 /* call with IRQ disabled */ 310 static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) 311 { 312 netdev_tx_t status = NETDEV_TX_BUSY; 313 netdev_tx_t ret = NET_XMIT_DROP; 314 struct net_device *dev; 315 unsigned long tries; 316 /* It is up to the caller to keep npinfo alive. */ 317 struct netpoll_info *npinfo; 318 319 lockdep_assert_irqs_disabled(); 320 321 dev = np->dev; 322 /* npinfo->txq belongs to np->dev, so retries must stay bound to it. */ 323 skb->dev = dev; 324 rcu_read_lock(); 325 npinfo = rcu_dereference_bh(dev->npinfo); 326 327 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { 328 dev_kfree_skb_irq(skb); 329 goto out; 330 } 331 332 /* don't get messages out of order, and no recursion */ 333 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) { 334 struct netdev_queue *txq; 335 336 txq = netdev_core_pick_tx(dev, skb, NULL); 337 338 /* try until next clock tick */ 339 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; 340 tries > 0; --tries) { 341 if (HARD_TX_TRYLOCK(dev, txq)) { 342 if (!netif_xmit_stopped(txq)) 343 status = netpoll_start_xmit(skb, dev, txq); 344 345 HARD_TX_UNLOCK(dev, txq); 346 347 if (dev_xmit_complete(status)) 348 break; 349 350 } 351 352 /* tickle device maybe there is some cleanup */ 353 netpoll_poll_dev(np->dev); 354 355 udelay(USEC_PER_POLL); 356 } 357 358 WARN_ONCE(!irqs_disabled(), 359 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n", 360 dev->name, dev->netdev_ops->ndo_start_xmit); 361 362 } 363 364 if (!dev_xmit_complete(status)) { 365 skb_queue_tail(&npinfo->txq, skb); 366 schedule_delayed_work(&npinfo->tx_work,0); 367 } 368 ret = NETDEV_TX_OK; 369 out: 370 rcu_read_unlock(); 371 return ret; 372 } 373 374 static void netpoll_udp_checksum(struct netpoll *np, struct sk_buff *skb, 375 int len) 376 { 377 struct udphdr *udph; 378 int udp_len; 379 380 udp_len = len + sizeof(struct udphdr); 381 udph = udp_hdr(skb); 382 383 /* check needs to be set, since it will be consumed in csum_partial */ 384 udph->check = 0; 385 if (np->ipv6) 386 udph->check = csum_ipv6_magic(&np->local_ip.in6, 387 &np->remote_ip.in6, 388 udp_len, IPPROTO_UDP, 389 csum_partial(udph, udp_len, 0)); 390 else 391 udph->check = csum_tcpudp_magic(np->local_ip.ip, 392 np->remote_ip.ip, 393 udp_len, IPPROTO_UDP, 394 csum_partial(udph, udp_len, 0)); 395 if (udph->check == 0) 396 udph->check = CSUM_MANGLED_0; 397 } 398 399 netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) 400 { 401 unsigned long flags; 402 netdev_tx_t ret; 403 404 if (unlikely(!np)) { 405 dev_kfree_skb_irq(skb); 406 ret = NET_XMIT_DROP; 407 } else { 408 local_irq_save(flags); 409 ret = __netpoll_send_skb(np, skb); 410 local_irq_restore(flags); 411 } 412 return ret; 413 } 414 EXPORT_SYMBOL(netpoll_send_skb); 415 416 static void push_ipv6(struct netpoll *np, struct sk_buff *skb, int len) 417 { 418 struct ipv6hdr *ip6h; 419 420 skb_push(skb, sizeof(struct ipv6hdr)); 421 skb_reset_network_header(skb); 422 ip6h = ipv6_hdr(skb); 423 424 /* ip6h->version = 6; ip6h->priority = 0; */ 425 *(unsigned char *)ip6h = 0x60; 426 ip6h->flow_lbl[0] = 0; 427 ip6h->flow_lbl[1] = 0; 428 ip6h->flow_lbl[2] = 0; 429 430 ip6h->payload_len = htons(sizeof(struct udphdr) + len); 431 ip6h->nexthdr = IPPROTO_UDP; 432 ip6h->hop_limit = 32; 433 ip6h->saddr = np->local_ip.in6; 434 ip6h->daddr = np->remote_ip.in6; 435 436 skb->protocol = htons(ETH_P_IPV6); 437 } 438 439 static void push_ipv4(struct netpoll *np, struct sk_buff *skb, int len) 440 { 441 static atomic_t ip_ident; 442 struct iphdr *iph; 443 int ip_len; 444 445 ip_len = len + sizeof(struct udphdr) + sizeof(struct iphdr); 446 447 skb_push(skb, sizeof(struct iphdr)); 448 skb_reset_network_header(skb); 449 iph = ip_hdr(skb); 450 451 /* iph->version = 4; iph->ihl = 5; */ 452 *(unsigned char *)iph = 0x45; 453 iph->tos = 0; 454 put_unaligned(htons(ip_len), &iph->tot_len); 455 iph->id = htons(atomic_inc_return(&ip_ident)); 456 iph->frag_off = 0; 457 iph->ttl = 64; 458 iph->protocol = IPPROTO_UDP; 459 iph->check = 0; 460 put_unaligned(np->local_ip.ip, &iph->saddr); 461 put_unaligned(np->remote_ip.ip, &iph->daddr); 462 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); 463 skb->protocol = htons(ETH_P_IP); 464 } 465 466 static void push_udp(struct netpoll *np, struct sk_buff *skb, int len) 467 { 468 struct udphdr *udph; 469 int udp_len; 470 471 udp_len = len + sizeof(struct udphdr); 472 473 skb_push(skb, sizeof(struct udphdr)); 474 skb_reset_transport_header(skb); 475 476 udph = udp_hdr(skb); 477 udph->source = htons(np->local_port); 478 udph->dest = htons(np->remote_port); 479 udph->len = htons(udp_len); 480 481 netpoll_udp_checksum(np, skb, len); 482 } 483 484 static void push_eth(struct netpoll *np, struct sk_buff *skb) 485 { 486 struct ethhdr *eth; 487 488 eth = skb_push(skb, ETH_HLEN); 489 skb_reset_mac_header(skb); 490 ether_addr_copy(eth->h_source, np->dev->dev_addr); 491 ether_addr_copy(eth->h_dest, np->remote_mac); 492 if (np->ipv6) 493 eth->h_proto = htons(ETH_P_IPV6); 494 else 495 eth->h_proto = htons(ETH_P_IP); 496 } 497 498 int netpoll_send_udp(struct netpoll *np, const char *msg, int len) 499 { 500 int total_len, ip_len, udp_len; 501 struct sk_buff *skb; 502 503 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) 504 WARN_ON_ONCE(!irqs_disabled()); 505 506 udp_len = len + sizeof(struct udphdr); 507 if (np->ipv6) 508 ip_len = udp_len + sizeof(struct ipv6hdr); 509 else 510 ip_len = udp_len + sizeof(struct iphdr); 511 512 total_len = ip_len + LL_RESERVED_SPACE(np->dev); 513 514 skb = find_skb(np, total_len + np->dev->needed_tailroom, 515 total_len - len); 516 if (!skb) 517 return -ENOMEM; 518 519 skb_copy_to_linear_data(skb, msg, len); 520 skb_put(skb, len); 521 522 push_udp(np, skb, len); 523 if (np->ipv6) 524 push_ipv6(np, skb, len); 525 else 526 push_ipv4(np, skb, len); 527 push_eth(np, skb); 528 skb->dev = np->dev; 529 530 return (int)netpoll_send_skb(np, skb); 531 } 532 EXPORT_SYMBOL(netpoll_send_udp); 533 534 535 static void skb_pool_flush(struct netpoll *np) 536 { 537 struct sk_buff_head *skb_pool; 538 539 cancel_work_sync(&np->refill_wq); 540 skb_pool = &np->skb_pool; 541 skb_queue_purge_reason(skb_pool, SKB_CONSUMED); 542 } 543 544 static void refill_skbs_work_handler(struct work_struct *work) 545 { 546 struct netpoll *np = 547 container_of(work, struct netpoll, refill_wq); 548 549 refill_skbs(np); 550 } 551 552 int __netpoll_setup(struct netpoll *np, struct net_device *ndev) 553 { 554 struct netpoll_info *npinfo; 555 const struct net_device_ops *ops; 556 int err; 557 558 skb_queue_head_init(&np->skb_pool); 559 INIT_WORK(&np->refill_wq, refill_skbs_work_handler); 560 561 if (ndev->priv_flags & IFF_DISABLE_NETPOLL) { 562 np_err(np, "%s doesn't support polling, aborting\n", 563 ndev->name); 564 err = -ENOTSUPP; 565 goto out; 566 } 567 568 npinfo = rtnl_dereference(ndev->npinfo); 569 if (!npinfo) { 570 npinfo = kmalloc_obj(*npinfo); 571 if (!npinfo) { 572 err = -ENOMEM; 573 goto out; 574 } 575 576 sema_init(&npinfo->dev_lock, 1); 577 skb_queue_head_init(&npinfo->txq); 578 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); 579 580 refcount_set(&npinfo->refcnt, 1); 581 582 ops = ndev->netdev_ops; 583 if (ops->ndo_netpoll_setup) { 584 err = ops->ndo_netpoll_setup(ndev); 585 if (err) 586 goto free_npinfo; 587 } 588 } else { 589 refcount_inc(&npinfo->refcnt); 590 } 591 592 np->dev = ndev; 593 strscpy(np->dev_name, ndev->name, IFNAMSIZ); 594 595 /* fill up the skb queue */ 596 refill_skbs(np); 597 598 /* last thing to do is link it to the net device structure */ 599 rcu_assign_pointer(ndev->npinfo, npinfo); 600 601 return 0; 602 603 free_npinfo: 604 kfree(npinfo); 605 out: 606 return err; 607 } 608 EXPORT_SYMBOL_GPL(__netpoll_setup); 609 610 /* 611 * Returns a pointer to a string representation of the identifier used 612 * to select the egress interface for the given netpoll instance. buf 613 * is used to format np->dev_mac when np->dev_name is empty; bufsz must 614 * be at least MAC_ADDR_STR_LEN + 1 to fit the formatted MAC address 615 * and its NUL terminator. 616 */ 617 static char *egress_dev(struct netpoll *np, char *buf, size_t bufsz) 618 { 619 if (np->dev_name[0]) 620 return np->dev_name; 621 622 snprintf(buf, bufsz, "%pM", np->dev_mac); 623 return buf; 624 } 625 626 static void netpoll_wait_carrier(struct netpoll *np, struct net_device *ndev, 627 unsigned int timeout) 628 { 629 unsigned long atmost; 630 631 atmost = jiffies + timeout * HZ; 632 while (!netif_carrier_ok(ndev)) { 633 if (time_after(jiffies, atmost)) { 634 np_notice(np, "timeout waiting for carrier\n"); 635 break; 636 } 637 msleep(1); 638 } 639 } 640 641 /* 642 * Take the IPv6 from ndev and populate local_ip structure in netpoll 643 */ 644 static int netpoll_take_ipv6(struct netpoll *np, struct net_device *ndev) 645 { 646 char buf[MAC_ADDR_STR_LEN + 1]; 647 int err = -EDESTADDRREQ; 648 struct inet6_dev *idev; 649 650 if (!IS_ENABLED(CONFIG_IPV6)) { 651 np_err(np, "IPv6 is not supported %s, aborting\n", 652 egress_dev(np, buf, sizeof(buf))); 653 return -EINVAL; 654 } 655 656 idev = __in6_dev_get(ndev); 657 if (idev) { 658 struct inet6_ifaddr *ifp; 659 660 read_lock_bh(&idev->lock); 661 list_for_each_entry(ifp, &idev->addr_list, if_list) { 662 if (!!(ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL) != 663 !!(ipv6_addr_type(&np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL)) 664 continue; 665 /* Got the IP, let's return */ 666 np->local_ip.in6 = ifp->addr; 667 err = 0; 668 break; 669 } 670 read_unlock_bh(&idev->lock); 671 } 672 if (err) { 673 np_err(np, "no IPv6 address for %s, aborting\n", 674 egress_dev(np, buf, sizeof(buf))); 675 return err; 676 } 677 678 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6); 679 return 0; 680 } 681 682 /* 683 * Take the IPv4 from ndev and populate local_ip structure in netpoll 684 */ 685 static int netpoll_take_ipv4(struct netpoll *np, struct net_device *ndev) 686 { 687 char buf[MAC_ADDR_STR_LEN + 1]; 688 const struct in_ifaddr *ifa; 689 struct in_device *in_dev; 690 691 in_dev = __in_dev_get_rtnl(ndev); 692 if (!in_dev) { 693 np_err(np, "no IP address for %s, aborting\n", 694 egress_dev(np, buf, sizeof(buf))); 695 return -EDESTADDRREQ; 696 } 697 698 ifa = rtnl_dereference(in_dev->ifa_list); 699 if (!ifa) { 700 np_err(np, "no IP address for %s, aborting\n", 701 egress_dev(np, buf, sizeof(buf))); 702 return -EDESTADDRREQ; 703 } 704 705 np->local_ip.ip = ifa->ifa_local; 706 np_info(np, "local IP %pI4\n", &np->local_ip.ip); 707 708 return 0; 709 } 710 711 /* 712 * Test whether the caller left np->local_ip unset, so that 713 * netpoll_setup() should auto-populate it from the egress device. 714 * 715 * np->local_ip is a union of __be32 (IPv4) and struct in6_addr (IPv6), 716 * so an IPv6 address whose first 4 bytes are zero (e.g. ::1, ::2, 717 * IPv4-mapped ::ffff:a.b.c.d) must not be tested via the IPv4 arm — 718 * doing so would misclassify a caller-supplied address as unset and 719 * silently overwrite it with whatever address the device exposes. 720 */ 721 static bool netpoll_local_ip_unset(const struct netpoll *np) 722 { 723 if (np->ipv6) 724 return ipv6_addr_any(&np->local_ip.in6); 725 return !np->local_ip.ip; 726 } 727 728 int netpoll_setup(struct netpoll *np) 729 { 730 struct net *net = current->nsproxy->net_ns; 731 char buf[MAC_ADDR_STR_LEN + 1]; 732 struct net_device *ndev = NULL; 733 bool ip_overwritten = false; 734 int err; 735 736 rtnl_lock(); 737 if (np->dev_name[0]) 738 ndev = __dev_get_by_name(net, np->dev_name); 739 else if (is_valid_ether_addr(np->dev_mac)) 740 ndev = dev_getbyhwaddr(net, ARPHRD_ETHER, np->dev_mac); 741 742 if (!ndev) { 743 np_err(np, "%s doesn't exist, aborting\n", 744 egress_dev(np, buf, sizeof(buf))); 745 err = -ENODEV; 746 goto unlock; 747 } 748 netdev_hold(ndev, &np->dev_tracker, GFP_KERNEL); 749 750 if (netdev_master_upper_dev_get(ndev)) { 751 np_err(np, "%s is a slave device, aborting\n", 752 egress_dev(np, buf, sizeof(buf))); 753 err = -EBUSY; 754 goto put; 755 } 756 757 if (!netif_running(ndev)) { 758 np_info(np, "device %s not up yet, forcing it\n", 759 egress_dev(np, buf, sizeof(buf))); 760 761 err = dev_open(ndev, NULL); 762 if (err) { 763 np_err(np, "failed to open %s\n", ndev->name); 764 goto put; 765 } 766 767 rtnl_unlock(); 768 netpoll_wait_carrier(np, ndev, carrier_timeout); 769 rtnl_lock(); 770 } 771 772 if (netpoll_local_ip_unset(np)) { 773 if (!np->ipv6) { 774 err = netpoll_take_ipv4(np, ndev); 775 if (err) 776 goto put; 777 } else { 778 err = netpoll_take_ipv6(np, ndev); 779 if (err) 780 goto put; 781 } 782 ip_overwritten = true; 783 } 784 785 err = __netpoll_setup(np, ndev); 786 if (err) 787 goto flush; 788 rtnl_unlock(); 789 790 /* Make sure all NAPI polls which started before dev->npinfo 791 * was visible have exited before we start calling NAPI poll. 792 * NAPI skips locking if dev->npinfo is NULL. 793 */ 794 synchronize_rcu(); 795 796 return 0; 797 798 flush: 799 skb_pool_flush(np); 800 put: 801 DEBUG_NET_WARN_ON_ONCE(np->dev); 802 if (ip_overwritten) 803 memset(&np->local_ip, 0, sizeof(np->local_ip)); 804 netdev_put(ndev, &np->dev_tracker); 805 unlock: 806 rtnl_unlock(); 807 return err; 808 } 809 EXPORT_SYMBOL(netpoll_setup); 810 811 static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head) 812 { 813 struct netpoll_info *npinfo = 814 container_of(rcu_head, struct netpoll_info, rcu); 815 816 skb_queue_purge(&npinfo->txq); 817 818 /* we can't call cancel_delayed_work_sync here, as we are in softirq */ 819 cancel_delayed_work(&npinfo->tx_work); 820 821 /* clean after last, unfinished work */ 822 __skb_queue_purge(&npinfo->txq); 823 /* now cancel it again */ 824 cancel_delayed_work(&npinfo->tx_work); 825 kfree(npinfo); 826 } 827 828 static void __netpoll_cleanup(struct netpoll *np) 829 { 830 struct netpoll_info *npinfo; 831 832 npinfo = rtnl_dereference(np->dev->npinfo); 833 if (!npinfo) 834 return; 835 836 /* At this point, there is a single npinfo instance per netdevice, and 837 * its refcnt tracks how many netpoll structures are linked to it. We 838 * only perform npinfo cleanup when the refcnt decrements to zero. 839 */ 840 if (refcount_dec_and_test(&npinfo->refcnt)) { 841 const struct net_device_ops *ops; 842 843 ops = np->dev->netdev_ops; 844 if (ops->ndo_netpoll_cleanup) 845 ops->ndo_netpoll_cleanup(np->dev); 846 847 RCU_INIT_POINTER(np->dev->npinfo, NULL); 848 call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info); 849 } 850 851 skb_pool_flush(np); 852 } 853 854 void __netpoll_free(struct netpoll *np) 855 { 856 ASSERT_RTNL(); 857 858 /* Wait for transmitting packets to finish before freeing. */ 859 synchronize_net(); 860 __netpoll_cleanup(np); 861 kfree(np); 862 } 863 EXPORT_SYMBOL_GPL(__netpoll_free); 864 865 void do_netpoll_cleanup(struct netpoll *np) 866 { 867 __netpoll_cleanup(np); 868 netdev_put(np->dev, &np->dev_tracker); 869 np->dev = NULL; 870 } 871 EXPORT_SYMBOL(do_netpoll_cleanup); 872 873 void netpoll_cleanup(struct netpoll *np) 874 { 875 rtnl_lock(); 876 if (!np->dev) 877 goto out; 878 do_netpoll_cleanup(np); 879 out: 880 rtnl_unlock(); 881 } 882 EXPORT_SYMBOL(netpoll_cleanup); 883