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 <asm/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 | |
49 | static struct sk_buff_head skb_pool; |
50 | |
51 | DEFINE_STATIC_SRCU(netpoll_srcu); |
52 | |
53 | #define USEC_PER_POLL 50 |
54 | |
55 | #define MAX_SKB_SIZE \ |
56 | (sizeof(struct ethhdr) + \ |
57 | sizeof(struct iphdr) + \ |
58 | sizeof(struct udphdr) + \ |
59 | MAX_UDP_CHUNK) |
60 | |
61 | static void zap_completion_queue(void); |
62 | |
63 | static unsigned int carrier_timeout = 4; |
64 | module_param(carrier_timeout, uint, 0644); |
65 | |
66 | #define np_info(np, fmt, ...) \ |
67 | pr_info("%s: " fmt, np->name, ##__VA_ARGS__) |
68 | #define np_err(np, fmt, ...) \ |
69 | pr_err("%s: " fmt, np->name, ##__VA_ARGS__) |
70 | #define np_notice(np, fmt, ...) \ |
71 | pr_notice("%s: " fmt, np->name, ##__VA_ARGS__) |
72 | |
73 | static netdev_tx_t netpoll_start_xmit(struct sk_buff *skb, |
74 | struct net_device *dev, |
75 | struct netdev_queue *txq) |
76 | { |
77 | netdev_tx_t status = NETDEV_TX_OK; |
78 | netdev_features_t features; |
79 | |
80 | features = netif_skb_features(skb); |
81 | |
82 | if (skb_vlan_tag_present(skb) && |
83 | !vlan_hw_offload_capable(features, proto: skb->vlan_proto)) { |
84 | skb = __vlan_hwaccel_push_inside(skb); |
85 | if (unlikely(!skb)) { |
86 | /* This is actually a packet drop, but we |
87 | * don't want the code that calls this |
88 | * function to try and operate on a NULL skb. |
89 | */ |
90 | goto out; |
91 | } |
92 | } |
93 | |
94 | status = netdev_start_xmit(skb, dev, txq, more: false); |
95 | |
96 | out: |
97 | return status; |
98 | } |
99 | |
100 | static void queue_process(struct work_struct *work) |
101 | { |
102 | struct netpoll_info *npinfo = |
103 | container_of(work, struct netpoll_info, tx_work.work); |
104 | struct sk_buff *skb; |
105 | unsigned long flags; |
106 | |
107 | while ((skb = skb_dequeue(list: &npinfo->txq))) { |
108 | struct net_device *dev = skb->dev; |
109 | struct netdev_queue *txq; |
110 | unsigned int q_index; |
111 | |
112 | if (!netif_device_present(dev) || !netif_running(dev)) { |
113 | kfree_skb(skb); |
114 | continue; |
115 | } |
116 | |
117 | local_irq_save(flags); |
118 | /* check if skb->queue_mapping is still valid */ |
119 | q_index = skb_get_queue_mapping(skb); |
120 | if (unlikely(q_index >= dev->real_num_tx_queues)) { |
121 | q_index = q_index % dev->real_num_tx_queues; |
122 | skb_set_queue_mapping(skb, queue_mapping: q_index); |
123 | } |
124 | txq = netdev_get_tx_queue(dev, index: q_index); |
125 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
126 | if (netif_xmit_frozen_or_stopped(dev_queue: txq) || |
127 | !dev_xmit_complete(rc: netpoll_start_xmit(skb, dev, txq))) { |
128 | skb_queue_head(list: &npinfo->txq, newsk: skb); |
129 | HARD_TX_UNLOCK(dev, txq); |
130 | local_irq_restore(flags); |
131 | |
132 | schedule_delayed_work(dwork: &npinfo->tx_work, HZ/10); |
133 | return; |
134 | } |
135 | HARD_TX_UNLOCK(dev, txq); |
136 | local_irq_restore(flags); |
137 | } |
138 | } |
139 | |
140 | static int netif_local_xmit_active(struct net_device *dev) |
141 | { |
142 | int i; |
143 | |
144 | for (i = 0; i < dev->num_tx_queues; i++) { |
145 | struct netdev_queue *txq = netdev_get_tx_queue(dev, index: i); |
146 | |
147 | if (READ_ONCE(txq->xmit_lock_owner) == smp_processor_id()) |
148 | return 1; |
149 | } |
150 | |
151 | return 0; |
152 | } |
153 | |
154 | static void poll_one_napi(struct napi_struct *napi) |
155 | { |
156 | int work; |
157 | |
158 | /* If we set this bit but see that it has already been set, |
159 | * that indicates that napi has been disabled and we need |
160 | * to abort this operation |
161 | */ |
162 | if (test_and_set_bit(nr: NAPI_STATE_NPSVC, addr: &napi->state)) |
163 | return; |
164 | |
165 | /* We explicilty pass the polling call a budget of 0 to |
166 | * indicate that we are clearing the Tx path only. |
167 | */ |
168 | work = napi->poll(napi, 0); |
169 | WARN_ONCE(work, "%pS exceeded budget in poll\n" , napi->poll); |
170 | trace_napi_poll(napi, work, budget: 0); |
171 | |
172 | clear_bit(nr: NAPI_STATE_NPSVC, addr: &napi->state); |
173 | } |
174 | |
175 | static void poll_napi(struct net_device *dev) |
176 | { |
177 | struct napi_struct *napi; |
178 | int cpu = smp_processor_id(); |
179 | |
180 | list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) { |
181 | if (cmpxchg(&napi->poll_owner, -1, cpu) == -1) { |
182 | poll_one_napi(napi); |
183 | smp_store_release(&napi->poll_owner, -1); |
184 | } |
185 | } |
186 | } |
187 | |
188 | void netpoll_poll_dev(struct net_device *dev) |
189 | { |
190 | struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo); |
191 | const struct net_device_ops *ops; |
192 | |
193 | /* Don't do any rx activity if the dev_lock mutex is held |
194 | * the dev_open/close paths use this to block netpoll activity |
195 | * while changing device state |
196 | */ |
197 | if (!ni || down_trylock(sem: &ni->dev_lock)) |
198 | return; |
199 | |
200 | /* Some drivers will take the same locks in poll and xmit, |
201 | * we can't poll if local CPU is already in xmit. |
202 | */ |
203 | if (!netif_running(dev) || netif_local_xmit_active(dev)) { |
204 | up(sem: &ni->dev_lock); |
205 | return; |
206 | } |
207 | |
208 | ops = dev->netdev_ops; |
209 | if (ops->ndo_poll_controller) |
210 | ops->ndo_poll_controller(dev); |
211 | |
212 | poll_napi(dev); |
213 | |
214 | up(sem: &ni->dev_lock); |
215 | |
216 | zap_completion_queue(); |
217 | } |
218 | EXPORT_SYMBOL(netpoll_poll_dev); |
219 | |
220 | void netpoll_poll_disable(struct net_device *dev) |
221 | { |
222 | struct netpoll_info *ni; |
223 | int idx; |
224 | might_sleep(); |
225 | idx = srcu_read_lock(ssp: &netpoll_srcu); |
226 | ni = srcu_dereference(dev->npinfo, &netpoll_srcu); |
227 | if (ni) |
228 | down(sem: &ni->dev_lock); |
229 | srcu_read_unlock(ssp: &netpoll_srcu, idx); |
230 | } |
231 | EXPORT_SYMBOL(netpoll_poll_disable); |
232 | |
233 | void netpoll_poll_enable(struct net_device *dev) |
234 | { |
235 | struct netpoll_info *ni; |
236 | rcu_read_lock(); |
237 | ni = rcu_dereference(dev->npinfo); |
238 | if (ni) |
239 | up(sem: &ni->dev_lock); |
240 | rcu_read_unlock(); |
241 | } |
242 | EXPORT_SYMBOL(netpoll_poll_enable); |
243 | |
244 | static void refill_skbs(void) |
245 | { |
246 | struct sk_buff *skb; |
247 | unsigned long flags; |
248 | |
249 | spin_lock_irqsave(&skb_pool.lock, flags); |
250 | while (skb_pool.qlen < MAX_SKBS) { |
251 | skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC); |
252 | if (!skb) |
253 | break; |
254 | |
255 | __skb_queue_tail(list: &skb_pool, newsk: skb); |
256 | } |
257 | spin_unlock_irqrestore(lock: &skb_pool.lock, flags); |
258 | } |
259 | |
260 | static void zap_completion_queue(void) |
261 | { |
262 | unsigned long flags; |
263 | struct softnet_data *sd = &get_cpu_var(softnet_data); |
264 | |
265 | if (sd->completion_queue) { |
266 | struct sk_buff *clist; |
267 | |
268 | local_irq_save(flags); |
269 | clist = sd->completion_queue; |
270 | sd->completion_queue = NULL; |
271 | local_irq_restore(flags); |
272 | |
273 | while (clist != NULL) { |
274 | struct sk_buff *skb = clist; |
275 | clist = clist->next; |
276 | if (!skb_irq_freeable(skb)) { |
277 | refcount_set(r: &skb->users, n: 1); |
278 | dev_kfree_skb_any(skb); /* put this one back */ |
279 | } else { |
280 | __kfree_skb(skb); |
281 | } |
282 | } |
283 | } |
284 | |
285 | put_cpu_var(softnet_data); |
286 | } |
287 | |
288 | static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve) |
289 | { |
290 | int count = 0; |
291 | struct sk_buff *skb; |
292 | |
293 | zap_completion_queue(); |
294 | refill_skbs(); |
295 | repeat: |
296 | |
297 | skb = alloc_skb(size: len, GFP_ATOMIC); |
298 | if (!skb) |
299 | skb = skb_dequeue(list: &skb_pool); |
300 | |
301 | if (!skb) { |
302 | if (++count < 10) { |
303 | netpoll_poll_dev(np->dev); |
304 | goto repeat; |
305 | } |
306 | return NULL; |
307 | } |
308 | |
309 | refcount_set(r: &skb->users, n: 1); |
310 | skb_reserve(skb, len: reserve); |
311 | return skb; |
312 | } |
313 | |
314 | static int netpoll_owner_active(struct net_device *dev) |
315 | { |
316 | struct napi_struct *napi; |
317 | |
318 | list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) { |
319 | if (napi->poll_owner == smp_processor_id()) |
320 | return 1; |
321 | } |
322 | return 0; |
323 | } |
324 | |
325 | /* call with IRQ disabled */ |
326 | static netdev_tx_t __netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) |
327 | { |
328 | netdev_tx_t status = NETDEV_TX_BUSY; |
329 | struct net_device *dev; |
330 | unsigned long tries; |
331 | /* It is up to the caller to keep npinfo alive. */ |
332 | struct netpoll_info *npinfo; |
333 | |
334 | lockdep_assert_irqs_disabled(); |
335 | |
336 | dev = np->dev; |
337 | npinfo = rcu_dereference_bh(dev->npinfo); |
338 | |
339 | if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) { |
340 | dev_kfree_skb_irq(skb); |
341 | return NET_XMIT_DROP; |
342 | } |
343 | |
344 | /* don't get messages out of order, and no recursion */ |
345 | if (skb_queue_len(list_: &npinfo->txq) == 0 && !netpoll_owner_active(dev)) { |
346 | struct netdev_queue *txq; |
347 | |
348 | txq = netdev_core_pick_tx(dev, skb, NULL); |
349 | |
350 | /* try until next clock tick */ |
351 | for (tries = jiffies_to_usecs(j: 1)/USEC_PER_POLL; |
352 | tries > 0; --tries) { |
353 | if (HARD_TX_TRYLOCK(dev, txq)) { |
354 | if (!netif_xmit_stopped(dev_queue: txq)) |
355 | status = netpoll_start_xmit(skb, dev, txq); |
356 | |
357 | HARD_TX_UNLOCK(dev, txq); |
358 | |
359 | if (dev_xmit_complete(rc: status)) |
360 | break; |
361 | |
362 | } |
363 | |
364 | /* tickle device maybe there is some cleanup */ |
365 | netpoll_poll_dev(np->dev); |
366 | |
367 | udelay(USEC_PER_POLL); |
368 | } |
369 | |
370 | WARN_ONCE(!irqs_disabled(), |
371 | "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pS)\n" , |
372 | dev->name, dev->netdev_ops->ndo_start_xmit); |
373 | |
374 | } |
375 | |
376 | if (!dev_xmit_complete(rc: status)) { |
377 | skb_queue_tail(list: &npinfo->txq, newsk: skb); |
378 | schedule_delayed_work(dwork: &npinfo->tx_work,delay: 0); |
379 | } |
380 | return NETDEV_TX_OK; |
381 | } |
382 | |
383 | netdev_tx_t netpoll_send_skb(struct netpoll *np, struct sk_buff *skb) |
384 | { |
385 | unsigned long flags; |
386 | netdev_tx_t ret; |
387 | |
388 | if (unlikely(!np)) { |
389 | dev_kfree_skb_irq(skb); |
390 | ret = NET_XMIT_DROP; |
391 | } else { |
392 | local_irq_save(flags); |
393 | ret = __netpoll_send_skb(np, skb); |
394 | local_irq_restore(flags); |
395 | } |
396 | return ret; |
397 | } |
398 | EXPORT_SYMBOL(netpoll_send_skb); |
399 | |
400 | void netpoll_send_udp(struct netpoll *np, const char *msg, int len) |
401 | { |
402 | int total_len, ip_len, udp_len; |
403 | struct sk_buff *skb; |
404 | struct udphdr *udph; |
405 | struct iphdr *iph; |
406 | struct ethhdr *eth; |
407 | static atomic_t ip_ident; |
408 | struct ipv6hdr *ip6h; |
409 | |
410 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
411 | WARN_ON_ONCE(!irqs_disabled()); |
412 | |
413 | udp_len = len + sizeof(*udph); |
414 | if (np->ipv6) |
415 | ip_len = udp_len + sizeof(*ip6h); |
416 | else |
417 | ip_len = udp_len + sizeof(*iph); |
418 | |
419 | total_len = ip_len + LL_RESERVED_SPACE(np->dev); |
420 | |
421 | skb = find_skb(np, len: total_len + np->dev->needed_tailroom, |
422 | reserve: total_len - len); |
423 | if (!skb) |
424 | return; |
425 | |
426 | skb_copy_to_linear_data(skb, from: msg, len); |
427 | skb_put(skb, len); |
428 | |
429 | skb_push(skb, len: sizeof(*udph)); |
430 | skb_reset_transport_header(skb); |
431 | udph = udp_hdr(skb); |
432 | udph->source = htons(np->local_port); |
433 | udph->dest = htons(np->remote_port); |
434 | udph->len = htons(udp_len); |
435 | |
436 | if (np->ipv6) { |
437 | udph->check = 0; |
438 | udph->check = csum_ipv6_magic(saddr: &np->local_ip.in6, |
439 | daddr: &np->remote_ip.in6, |
440 | len: udp_len, IPPROTO_UDP, |
441 | sum: csum_partial(buff: udph, len: udp_len, sum: 0)); |
442 | if (udph->check == 0) |
443 | udph->check = CSUM_MANGLED_0; |
444 | |
445 | skb_push(skb, len: sizeof(*ip6h)); |
446 | skb_reset_network_header(skb); |
447 | ip6h = ipv6_hdr(skb); |
448 | |
449 | /* ip6h->version = 6; ip6h->priority = 0; */ |
450 | *(unsigned char *)ip6h = 0x60; |
451 | ip6h->flow_lbl[0] = 0; |
452 | ip6h->flow_lbl[1] = 0; |
453 | ip6h->flow_lbl[2] = 0; |
454 | |
455 | ip6h->payload_len = htons(sizeof(struct udphdr) + len); |
456 | ip6h->nexthdr = IPPROTO_UDP; |
457 | ip6h->hop_limit = 32; |
458 | ip6h->saddr = np->local_ip.in6; |
459 | ip6h->daddr = np->remote_ip.in6; |
460 | |
461 | eth = skb_push(skb, ETH_HLEN); |
462 | skb_reset_mac_header(skb); |
463 | skb->protocol = eth->h_proto = htons(ETH_P_IPV6); |
464 | } else { |
465 | udph->check = 0; |
466 | udph->check = csum_tcpudp_magic(saddr: np->local_ip.ip, |
467 | daddr: np->remote_ip.ip, |
468 | len: udp_len, IPPROTO_UDP, |
469 | sum: csum_partial(buff: udph, len: udp_len, sum: 0)); |
470 | if (udph->check == 0) |
471 | udph->check = CSUM_MANGLED_0; |
472 | |
473 | skb_push(skb, len: sizeof(*iph)); |
474 | skb_reset_network_header(skb); |
475 | iph = ip_hdr(skb); |
476 | |
477 | /* iph->version = 4; iph->ihl = 5; */ |
478 | *(unsigned char *)iph = 0x45; |
479 | iph->tos = 0; |
480 | put_unaligned(htons(ip_len), &(iph->tot_len)); |
481 | iph->id = htons(atomic_inc_return(&ip_ident)); |
482 | iph->frag_off = 0; |
483 | iph->ttl = 64; |
484 | iph->protocol = IPPROTO_UDP; |
485 | iph->check = 0; |
486 | put_unaligned(np->local_ip.ip, &(iph->saddr)); |
487 | put_unaligned(np->remote_ip.ip, &(iph->daddr)); |
488 | iph->check = ip_fast_csum(iph: (unsigned char *)iph, ihl: iph->ihl); |
489 | |
490 | eth = skb_push(skb, ETH_HLEN); |
491 | skb_reset_mac_header(skb); |
492 | skb->protocol = eth->h_proto = htons(ETH_P_IP); |
493 | } |
494 | |
495 | ether_addr_copy(dst: eth->h_source, src: np->dev->dev_addr); |
496 | ether_addr_copy(dst: eth->h_dest, src: np->remote_mac); |
497 | |
498 | skb->dev = np->dev; |
499 | |
500 | netpoll_send_skb(np, skb); |
501 | } |
502 | EXPORT_SYMBOL(netpoll_send_udp); |
503 | |
504 | void netpoll_print_options(struct netpoll *np) |
505 | { |
506 | np_info(np, "local port %d\n" , np->local_port); |
507 | if (np->ipv6) |
508 | np_info(np, "local IPv6 address %pI6c\n" , &np->local_ip.in6); |
509 | else |
510 | np_info(np, "local IPv4 address %pI4\n" , &np->local_ip.ip); |
511 | np_info(np, "interface '%s'\n" , np->dev_name); |
512 | np_info(np, "remote port %d\n" , np->remote_port); |
513 | if (np->ipv6) |
514 | np_info(np, "remote IPv6 address %pI6c\n" , &np->remote_ip.in6); |
515 | else |
516 | np_info(np, "remote IPv4 address %pI4\n" , &np->remote_ip.ip); |
517 | np_info(np, "remote ethernet address %pM\n" , np->remote_mac); |
518 | } |
519 | EXPORT_SYMBOL(netpoll_print_options); |
520 | |
521 | static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr) |
522 | { |
523 | const char *end; |
524 | |
525 | if (!strchr(str, ':') && |
526 | in4_pton(src: str, srclen: -1, dst: (void *)addr, delim: -1, end: &end) > 0) { |
527 | if (!*end) |
528 | return 0; |
529 | } |
530 | if (in6_pton(src: str, srclen: -1, dst: addr->in6.s6_addr, delim: -1, end: &end) > 0) { |
531 | #if IS_ENABLED(CONFIG_IPV6) |
532 | if (!*end) |
533 | return 1; |
534 | #else |
535 | return -1; |
536 | #endif |
537 | } |
538 | return -1; |
539 | } |
540 | |
541 | int netpoll_parse_options(struct netpoll *np, char *opt) |
542 | { |
543 | char *cur=opt, *delim; |
544 | int ipv6; |
545 | bool ipversion_set = false; |
546 | |
547 | if (*cur != '@') { |
548 | if ((delim = strchr(cur, '@')) == NULL) |
549 | goto parse_failed; |
550 | *delim = 0; |
551 | if (kstrtou16(s: cur, base: 10, res: &np->local_port)) |
552 | goto parse_failed; |
553 | cur = delim; |
554 | } |
555 | cur++; |
556 | |
557 | if (*cur != '/') { |
558 | ipversion_set = true; |
559 | if ((delim = strchr(cur, '/')) == NULL) |
560 | goto parse_failed; |
561 | *delim = 0; |
562 | ipv6 = netpoll_parse_ip_addr(str: cur, addr: &np->local_ip); |
563 | if (ipv6 < 0) |
564 | goto parse_failed; |
565 | else |
566 | np->ipv6 = (bool)ipv6; |
567 | cur = delim; |
568 | } |
569 | cur++; |
570 | |
571 | if (*cur != ',') { |
572 | /* parse out dev name */ |
573 | if ((delim = strchr(cur, ',')) == NULL) |
574 | goto parse_failed; |
575 | *delim = 0; |
576 | strscpy(np->dev_name, cur, sizeof(np->dev_name)); |
577 | cur = delim; |
578 | } |
579 | cur++; |
580 | |
581 | if (*cur != '@') { |
582 | /* dst port */ |
583 | if ((delim = strchr(cur, '@')) == NULL) |
584 | goto parse_failed; |
585 | *delim = 0; |
586 | if (*cur == ' ' || *cur == '\t') |
587 | np_info(np, "warning: whitespace is not allowed\n" ); |
588 | if (kstrtou16(s: cur, base: 10, res: &np->remote_port)) |
589 | goto parse_failed; |
590 | cur = delim; |
591 | } |
592 | cur++; |
593 | |
594 | /* dst ip */ |
595 | if ((delim = strchr(cur, '/')) == NULL) |
596 | goto parse_failed; |
597 | *delim = 0; |
598 | ipv6 = netpoll_parse_ip_addr(str: cur, addr: &np->remote_ip); |
599 | if (ipv6 < 0) |
600 | goto parse_failed; |
601 | else if (ipversion_set && np->ipv6 != (bool)ipv6) |
602 | goto parse_failed; |
603 | else |
604 | np->ipv6 = (bool)ipv6; |
605 | cur = delim + 1; |
606 | |
607 | if (*cur != 0) { |
608 | /* MAC address */ |
609 | if (!mac_pton(s: cur, mac: np->remote_mac)) |
610 | goto parse_failed; |
611 | } |
612 | |
613 | netpoll_print_options(np); |
614 | |
615 | return 0; |
616 | |
617 | parse_failed: |
618 | np_info(np, "couldn't parse config at '%s'!\n" , cur); |
619 | return -1; |
620 | } |
621 | EXPORT_SYMBOL(netpoll_parse_options); |
622 | |
623 | int __netpoll_setup(struct netpoll *np, struct net_device *ndev) |
624 | { |
625 | struct netpoll_info *npinfo; |
626 | const struct net_device_ops *ops; |
627 | int err; |
628 | |
629 | np->dev = ndev; |
630 | strscpy(np->dev_name, ndev->name, IFNAMSIZ); |
631 | |
632 | if (ndev->priv_flags & IFF_DISABLE_NETPOLL) { |
633 | np_err(np, "%s doesn't support polling, aborting\n" , |
634 | np->dev_name); |
635 | err = -ENOTSUPP; |
636 | goto out; |
637 | } |
638 | |
639 | if (!ndev->npinfo) { |
640 | npinfo = kmalloc(size: sizeof(*npinfo), GFP_KERNEL); |
641 | if (!npinfo) { |
642 | err = -ENOMEM; |
643 | goto out; |
644 | } |
645 | |
646 | sema_init(sem: &npinfo->dev_lock, val: 1); |
647 | skb_queue_head_init(list: &npinfo->txq); |
648 | INIT_DELAYED_WORK(&npinfo->tx_work, queue_process); |
649 | |
650 | refcount_set(r: &npinfo->refcnt, n: 1); |
651 | |
652 | ops = np->dev->netdev_ops; |
653 | if (ops->ndo_netpoll_setup) { |
654 | err = ops->ndo_netpoll_setup(ndev, npinfo); |
655 | if (err) |
656 | goto free_npinfo; |
657 | } |
658 | } else { |
659 | npinfo = rtnl_dereference(ndev->npinfo); |
660 | refcount_inc(r: &npinfo->refcnt); |
661 | } |
662 | |
663 | npinfo->netpoll = np; |
664 | |
665 | /* last thing to do is link it to the net device structure */ |
666 | rcu_assign_pointer(ndev->npinfo, npinfo); |
667 | |
668 | return 0; |
669 | |
670 | free_npinfo: |
671 | kfree(objp: npinfo); |
672 | out: |
673 | return err; |
674 | } |
675 | EXPORT_SYMBOL_GPL(__netpoll_setup); |
676 | |
677 | int netpoll_setup(struct netpoll *np) |
678 | { |
679 | struct net_device *ndev = NULL; |
680 | struct in_device *in_dev; |
681 | int err; |
682 | |
683 | rtnl_lock(); |
684 | if (np->dev_name[0]) { |
685 | struct net *net = current->nsproxy->net_ns; |
686 | ndev = __dev_get_by_name(net, name: np->dev_name); |
687 | } |
688 | if (!ndev) { |
689 | np_err(np, "%s doesn't exist, aborting\n" , np->dev_name); |
690 | err = -ENODEV; |
691 | goto unlock; |
692 | } |
693 | netdev_hold(dev: ndev, tracker: &np->dev_tracker, GFP_KERNEL); |
694 | |
695 | if (netdev_master_upper_dev_get(dev: ndev)) { |
696 | np_err(np, "%s is a slave device, aborting\n" , np->dev_name); |
697 | err = -EBUSY; |
698 | goto put; |
699 | } |
700 | |
701 | if (!netif_running(dev: ndev)) { |
702 | unsigned long atmost; |
703 | |
704 | np_info(np, "device %s not up yet, forcing it\n" , np->dev_name); |
705 | |
706 | err = dev_open(dev: ndev, NULL); |
707 | |
708 | if (err) { |
709 | np_err(np, "failed to open %s\n" , ndev->name); |
710 | goto put; |
711 | } |
712 | |
713 | rtnl_unlock(); |
714 | atmost = jiffies + carrier_timeout * HZ; |
715 | while (!netif_carrier_ok(dev: ndev)) { |
716 | if (time_after(jiffies, atmost)) { |
717 | np_notice(np, "timeout waiting for carrier\n" ); |
718 | break; |
719 | } |
720 | msleep(msecs: 1); |
721 | } |
722 | |
723 | rtnl_lock(); |
724 | } |
725 | |
726 | if (!np->local_ip.ip) { |
727 | if (!np->ipv6) { |
728 | const struct in_ifaddr *ifa; |
729 | |
730 | in_dev = __in_dev_get_rtnl(dev: ndev); |
731 | if (!in_dev) |
732 | goto put_noaddr; |
733 | |
734 | ifa = rtnl_dereference(in_dev->ifa_list); |
735 | if (!ifa) { |
736 | put_noaddr: |
737 | np_err(np, "no IP address for %s, aborting\n" , |
738 | np->dev_name); |
739 | err = -EDESTADDRREQ; |
740 | goto put; |
741 | } |
742 | |
743 | np->local_ip.ip = ifa->ifa_local; |
744 | np_info(np, "local IP %pI4\n" , &np->local_ip.ip); |
745 | } else { |
746 | #if IS_ENABLED(CONFIG_IPV6) |
747 | struct inet6_dev *idev; |
748 | |
749 | err = -EDESTADDRREQ; |
750 | idev = __in6_dev_get(dev: ndev); |
751 | if (idev) { |
752 | struct inet6_ifaddr *ifp; |
753 | |
754 | read_lock_bh(&idev->lock); |
755 | list_for_each_entry(ifp, &idev->addr_list, if_list) { |
756 | if (!!(ipv6_addr_type(addr: &ifp->addr) & IPV6_ADDR_LINKLOCAL) != |
757 | !!(ipv6_addr_type(addr: &np->remote_ip.in6) & IPV6_ADDR_LINKLOCAL)) |
758 | continue; |
759 | np->local_ip.in6 = ifp->addr; |
760 | err = 0; |
761 | break; |
762 | } |
763 | read_unlock_bh(&idev->lock); |
764 | } |
765 | if (err) { |
766 | np_err(np, "no IPv6 address for %s, aborting\n" , |
767 | np->dev_name); |
768 | goto put; |
769 | } else |
770 | np_info(np, "local IPv6 %pI6c\n" , &np->local_ip.in6); |
771 | #else |
772 | np_err(np, "IPv6 is not supported %s, aborting\n" , |
773 | np->dev_name); |
774 | err = -EINVAL; |
775 | goto put; |
776 | #endif |
777 | } |
778 | } |
779 | |
780 | /* fill up the skb queue */ |
781 | refill_skbs(); |
782 | |
783 | err = __netpoll_setup(np, ndev); |
784 | if (err) |
785 | goto put; |
786 | rtnl_unlock(); |
787 | return 0; |
788 | |
789 | put: |
790 | netdev_put(dev: ndev, tracker: &np->dev_tracker); |
791 | unlock: |
792 | rtnl_unlock(); |
793 | return err; |
794 | } |
795 | EXPORT_SYMBOL(netpoll_setup); |
796 | |
797 | static int __init netpoll_init(void) |
798 | { |
799 | skb_queue_head_init(list: &skb_pool); |
800 | return 0; |
801 | } |
802 | core_initcall(netpoll_init); |
803 | |
804 | static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head) |
805 | { |
806 | struct netpoll_info *npinfo = |
807 | container_of(rcu_head, struct netpoll_info, rcu); |
808 | |
809 | skb_queue_purge(list: &npinfo->txq); |
810 | |
811 | /* we can't call cancel_delayed_work_sync here, as we are in softirq */ |
812 | cancel_delayed_work(dwork: &npinfo->tx_work); |
813 | |
814 | /* clean after last, unfinished work */ |
815 | __skb_queue_purge(list: &npinfo->txq); |
816 | /* now cancel it again */ |
817 | cancel_delayed_work(dwork: &npinfo->tx_work); |
818 | kfree(objp: npinfo); |
819 | } |
820 | |
821 | void __netpoll_cleanup(struct netpoll *np) |
822 | { |
823 | struct netpoll_info *npinfo; |
824 | |
825 | npinfo = rtnl_dereference(np->dev->npinfo); |
826 | if (!npinfo) |
827 | return; |
828 | |
829 | synchronize_srcu(ssp: &netpoll_srcu); |
830 | |
831 | if (refcount_dec_and_test(r: &npinfo->refcnt)) { |
832 | const struct net_device_ops *ops; |
833 | |
834 | ops = np->dev->netdev_ops; |
835 | if (ops->ndo_netpoll_cleanup) |
836 | ops->ndo_netpoll_cleanup(np->dev); |
837 | |
838 | RCU_INIT_POINTER(np->dev->npinfo, NULL); |
839 | call_rcu(head: &npinfo->rcu, func: rcu_cleanup_netpoll_info); |
840 | } else |
841 | RCU_INIT_POINTER(np->dev->npinfo, NULL); |
842 | } |
843 | EXPORT_SYMBOL_GPL(__netpoll_cleanup); |
844 | |
845 | void __netpoll_free(struct netpoll *np) |
846 | { |
847 | ASSERT_RTNL(); |
848 | |
849 | /* Wait for transmitting packets to finish before freeing. */ |
850 | synchronize_rcu(); |
851 | __netpoll_cleanup(np); |
852 | kfree(objp: np); |
853 | } |
854 | EXPORT_SYMBOL_GPL(__netpoll_free); |
855 | |
856 | void netpoll_cleanup(struct netpoll *np) |
857 | { |
858 | rtnl_lock(); |
859 | if (!np->dev) |
860 | goto out; |
861 | __netpoll_cleanup(np); |
862 | netdev_put(dev: np->dev, tracker: &np->dev_tracker); |
863 | np->dev = NULL; |
864 | out: |
865 | rtnl_unlock(); |
866 | } |
867 | EXPORT_SYMBOL(netpoll_cleanup); |
868 | |