]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/core/netpoll.c
xen-netback: Adding debugfs "io_ring_qX" files
[mirror_ubuntu-bionic-kernel.git] / net / core / netpoll.c
CommitLineData
1da177e4
LT
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
e6ec2693
JP
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
bff38771 14#include <linux/moduleparam.h>
4cd5773a 15#include <linux/kernel.h>
1da177e4
LT
16#include <linux/netdevice.h>
17#include <linux/etherdevice.h>
18#include <linux/string.h>
14c85021 19#include <linux/if_arp.h>
1da177e4
LT
20#include <linux/inetdevice.h>
21#include <linux/inet.h>
22#include <linux/interrupt.h>
23#include <linux/netpoll.h>
24#include <linux/sched.h>
25#include <linux/delay.h>
26#include <linux/rcupdate.h>
27#include <linux/workqueue.h>
5a0e3ad6 28#include <linux/slab.h>
bc3b2d7f 29#include <linux/export.h>
689971b4 30#include <linux/if_vlan.h>
1da177e4
LT
31#include <net/tcp.h>
32#include <net/udp.h>
b3d936f3
CW
33#include <net/addrconf.h>
34#include <net/ndisc.h>
35#include <net/ip6_checksum.h>
1da177e4 36#include <asm/unaligned.h>
9cbc1cb8 37#include <trace/events/napi.h>
1da177e4
LT
38
39/*
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
42 */
43
44#define MAX_UDP_CHUNK 1460
45#define MAX_SKBS 32
1da177e4 46
a1bcfacd 47static struct sk_buff_head skb_pool;
1da177e4 48
7f9421c2 49DEFINE_STATIC_SRCU(netpoll_srcu);
ca99ca14 50
2bdfe0ba 51#define USEC_PER_POLL 50
1da177e4 52
6f706245
JP
53#define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
57 MAX_UDP_CHUNK)
1da177e4 58
3578b0c8 59static void zap_completion_queue(void);
2cde6acd 60static void netpoll_async_cleanup(struct work_struct *work);
1da177e4 61
bff38771
AV
62static unsigned int carrier_timeout = 4;
63module_param(carrier_timeout, uint, 0644);
64
e6ec2693
JP
65#define np_info(np, fmt, ...) \
66 pr_info("%s: " fmt, np->name, ##__VA_ARGS__)
67#define np_err(np, fmt, ...) \
68 pr_err("%s: " fmt, np->name, ##__VA_ARGS__)
69#define np_notice(np, fmt, ...) \
70 pr_notice("%s: " fmt, np->name, ##__VA_ARGS__)
71
944e2948
EB
72static int netpoll_start_xmit(struct sk_buff *skb, struct net_device *dev,
73 struct netdev_queue *txq)
74{
75 const struct net_device_ops *ops = dev->netdev_ops;
76 int status = NETDEV_TX_OK;
77 netdev_features_t features;
78
79 features = netif_skb_features(skb);
80
81 if (vlan_tx_tag_present(skb) &&
82 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
83 skb = __vlan_put_tag(skb, skb->vlan_proto,
84 vlan_tx_tag_get(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 skb->vlan_tci = 0;
93 }
94
95 status = ops->ndo_start_xmit(skb, dev);
96 if (status == NETDEV_TX_OK)
97 txq_trans_update(txq);
98
99out:
100 return status;
101}
102
c4028958 103static void queue_process(struct work_struct *work)
1da177e4 104{
4c1ac1b4
DH
105 struct netpoll_info *npinfo =
106 container_of(work, struct netpoll_info, tx_work.work);
1da177e4 107 struct sk_buff *skb;
3640543d 108 unsigned long flags;
1da177e4 109
6c43ff18
SH
110 while ((skb = skb_dequeue(&npinfo->txq))) {
111 struct net_device *dev = skb->dev;
fd2ea0a7 112 struct netdev_queue *txq;
1da177e4 113
6c43ff18 114 if (!netif_device_present(dev) || !netif_running(dev)) {
080b3c19 115 kfree_skb(skb);
6c43ff18
SH
116 continue;
117 }
1da177e4 118
fd2ea0a7
DM
119 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
120
3640543d 121 local_irq_save(flags);
5efeac44 122 HARD_TX_LOCK(dev, txq, smp_processor_id());
73466498 123 if (netif_xmit_frozen_or_stopped(txq) ||
944e2948 124 netpoll_start_xmit(skb, dev, txq) != NETDEV_TX_OK) {
6c43ff18 125 skb_queue_head(&npinfo->txq, skb);
5efeac44 126 HARD_TX_UNLOCK(dev, txq);
3640543d 127 local_irq_restore(flags);
1da177e4 128
25442caf 129 schedule_delayed_work(&npinfo->tx_work, HZ/10);
6c43ff18
SH
130 return;
131 }
5efeac44 132 HARD_TX_UNLOCK(dev, txq);
3640543d 133 local_irq_restore(flags);
1da177e4 134 }
1da177e4
LT
135}
136
1da177e4
LT
137/*
138 * Check whether delayed processing was scheduled for our NIC. If so,
139 * we attempt to grab the poll lock and use ->poll() to pump the card.
140 * If this fails, either we've recursed in ->poll() or it's already
141 * running on another CPU.
142 *
143 * Note: we don't mask interrupts with this lock because we're using
144 * trylock here and interrupts are already disabled in the softirq
145 * case. Further, we test the poll_owner to avoid recursion on UP
146 * systems where the lock doesn't exist.
1da177e4 147 */
b249b51b 148static int poll_one_napi(struct napi_struct *napi, int budget)
0a7606c1
DM
149{
150 int work;
151
152 /* net_rx_action's ->poll() invocations and our's are
153 * synchronized by this test which is only made while
154 * holding the napi->poll_lock.
155 */
156 if (!test_bit(NAPI_STATE_SCHED, &napi->state))
157 return budget;
158
7b363e44 159 set_bit(NAPI_STATE_NPSVC, &napi->state);
0a7606c1
DM
160
161 work = napi->poll(napi, budget);
e97dc3fc 162 WARN_ONCE(work > budget, "%pF exceeded budget in poll\n", napi->poll);
7d18f114 163 trace_napi_poll(napi);
0a7606c1 164
7b363e44 165 clear_bit(NAPI_STATE_NPSVC, &napi->state);
0a7606c1
DM
166
167 return budget - work;
168}
169
9852fbec 170static void poll_napi(struct net_device *dev, int budget)
1da177e4 171{
bea3348e 172 struct napi_struct *napi;
1da177e4 173
f13d493d 174 list_for_each_entry(napi, &dev->napi_list, dev_list) {
0a7606c1 175 if (napi->poll_owner != smp_processor_id() &&
bea3348e 176 spin_trylock(&napi->poll_lock)) {
b249b51b 177 budget = poll_one_napi(napi, budget);
bea3348e
SH
178 spin_unlock(&napi->poll_lock);
179 }
1da177e4
LT
180 }
181}
182
234b921d 183static void netpoll_poll_dev(struct net_device *dev)
1da177e4 184{
5e392739 185 const struct net_device_ops *ops;
2899656b 186 struct netpoll_info *ni = rcu_dereference_bh(dev->npinfo);
9c62a68d 187 int budget = 0;
5106930b 188
ca99ca14
NH
189 /* Don't do any rx activity if the dev_lock mutex is held
190 * the dev_open/close paths use this to block netpoll activity
191 * while changing device state
192 */
a3dbbc2b 193 if (down_trylock(&ni->dev_lock))
ca99ca14
NH
194 return;
195
959d5fde 196 if (!netif_running(dev)) {
bd7c4b60 197 up(&ni->dev_lock);
5e392739 198 return;
959d5fde 199 }
5e392739
PE
200
201 ops = dev->netdev_ops;
959d5fde 202 if (!ops->ndo_poll_controller) {
bd7c4b60 203 up(&ni->dev_lock);
1da177e4 204 return;
959d5fde 205 }
1da177e4
LT
206
207 /* Process pending work on NIC */
d314774c 208 ops->ndo_poll_controller(dev);
5106930b 209
9852fbec 210 poll_napi(dev, budget);
1da177e4 211
bd7c4b60 212 up(&ni->dev_lock);
ca99ca14 213
3578b0c8 214 zap_completion_queue();
1da177e4
LT
215}
216
66b5552f 217void netpoll_poll_disable(struct net_device *dev)
ca99ca14
NH
218{
219 struct netpoll_info *ni;
220 int idx;
221 might_sleep();
222 idx = srcu_read_lock(&netpoll_srcu);
223 ni = srcu_dereference(dev->npinfo, &netpoll_srcu);
224 if (ni)
bd7c4b60 225 down(&ni->dev_lock);
ca99ca14 226 srcu_read_unlock(&netpoll_srcu, idx);
ca99ca14 227}
66b5552f 228EXPORT_SYMBOL(netpoll_poll_disable);
ca99ca14 229
66b5552f 230void netpoll_poll_enable(struct net_device *dev)
ca99ca14
NH
231{
232 struct netpoll_info *ni;
233 rcu_read_lock();
234 ni = rcu_dereference(dev->npinfo);
235 if (ni)
bd7c4b60 236 up(&ni->dev_lock);
ca99ca14
NH
237 rcu_read_unlock();
238}
66b5552f 239EXPORT_SYMBOL(netpoll_poll_enable);
ca99ca14 240
1da177e4
LT
241static void refill_skbs(void)
242{
243 struct sk_buff *skb;
244 unsigned long flags;
245
a1bcfacd
SH
246 spin_lock_irqsave(&skb_pool.lock, flags);
247 while (skb_pool.qlen < MAX_SKBS) {
1da177e4
LT
248 skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
249 if (!skb)
250 break;
251
a1bcfacd 252 __skb_queue_tail(&skb_pool, skb);
1da177e4 253 }
a1bcfacd 254 spin_unlock_irqrestore(&skb_pool.lock, flags);
1da177e4
LT
255}
256
3578b0c8
DM
257static void zap_completion_queue(void)
258{
259 unsigned long flags;
260 struct softnet_data *sd = &get_cpu_var(softnet_data);
261
262 if (sd->completion_queue) {
263 struct sk_buff *clist;
264
265 local_irq_save(flags);
266 clist = sd->completion_queue;
267 sd->completion_queue = NULL;
268 local_irq_restore(flags);
269
270 while (clist != NULL) {
271 struct sk_buff *skb = clist;
272 clist = clist->next;
b1586f09 273 if (!skb_irq_freeable(skb)) {
3578b0c8
DM
274 atomic_inc(&skb->users);
275 dev_kfree_skb_any(skb); /* put this one back */
276 } else {
277 __kfree_skb(skb);
278 }
279 }
280 }
281
282 put_cpu_var(softnet_data);
283}
284
a1bcfacd 285static struct sk_buff *find_skb(struct netpoll *np, int len, int reserve)
1da177e4 286{
a1bcfacd
SH
287 int count = 0;
288 struct sk_buff *skb;
1da177e4 289
3578b0c8 290 zap_completion_queue();
a1bcfacd 291 refill_skbs();
1da177e4 292repeat:
1da177e4
LT
293
294 skb = alloc_skb(len, GFP_ATOMIC);
a1bcfacd
SH
295 if (!skb)
296 skb = skb_dequeue(&skb_pool);
1da177e4
LT
297
298 if (!skb) {
a1bcfacd 299 if (++count < 10) {
2a49e001 300 netpoll_poll_dev(np->dev);
a1bcfacd 301 goto repeat;
1da177e4 302 }
a1bcfacd 303 return NULL;
1da177e4
LT
304 }
305
306 atomic_set(&skb->users, 1);
307 skb_reserve(skb, reserve);
308 return skb;
309}
310
bea3348e
SH
311static int netpoll_owner_active(struct net_device *dev)
312{
313 struct napi_struct *napi;
314
315 list_for_each_entry(napi, &dev->napi_list, dev_list) {
316 if (napi->poll_owner == smp_processor_id())
317 return 1;
318 }
319 return 0;
320}
321
2899656b 322/* call with IRQ disabled */
c2355e1a
NH
323void netpoll_send_skb_on_dev(struct netpoll *np, struct sk_buff *skb,
324 struct net_device *dev)
1da177e4 325{
2bdfe0ba
SH
326 int status = NETDEV_TX_BUSY;
327 unsigned long tries;
de85d99e 328 /* It is up to the caller to keep npinfo alive. */
2899656b 329 struct netpoll_info *npinfo;
2bdfe0ba 330
2899656b
AW
331 WARN_ON_ONCE(!irqs_disabled());
332
333 npinfo = rcu_dereference_bh(np->dev->npinfo);
4ec93edb 334 if (!npinfo || !netif_running(dev) || !netif_device_present(dev)) {
080b3c19 335 dev_kfree_skb_irq(skb);
4ec93edb
YH
336 return;
337 }
2bdfe0ba
SH
338
339 /* don't get messages out of order, and no recursion */
bea3348e 340 if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
fd2ea0a7 341 struct netdev_queue *txq;
a49f99ff 342
f663dd9a 343 txq = netdev_pick_tx(dev, skb, NULL);
fd2ea0a7 344
0db3dc73
SH
345 /* try until next clock tick */
346 for (tries = jiffies_to_usecs(1)/USEC_PER_POLL;
347 tries > 0; --tries) {
5efeac44 348 if (HARD_TX_TRYLOCK(dev, txq)) {
944e2948
EB
349 if (!netif_xmit_stopped(txq))
350 status = netpoll_start_xmit(skb, dev, txq);
351
5efeac44 352 HARD_TX_UNLOCK(dev, txq);
e37b8d93
AM
353
354 if (status == NETDEV_TX_OK)
355 break;
356
e37b8d93 357 }
0db3dc73
SH
358
359 /* tickle device maybe there is some cleanup */
2a49e001 360 netpoll_poll_dev(np->dev);
0db3dc73
SH
361
362 udelay(USEC_PER_POLL);
0db1d6fc 363 }
79b1bee8
DD
364
365 WARN_ONCE(!irqs_disabled(),
2899656b 366 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
944e2948 367 dev->name, dev->netdev_ops->ndo_start_xmit);
79b1bee8 368
1da177e4 369 }
1da177e4 370
2bdfe0ba 371 if (status != NETDEV_TX_OK) {
5de4a473 372 skb_queue_tail(&npinfo->txq, skb);
4c1ac1b4 373 schedule_delayed_work(&npinfo->tx_work,0);
1da177e4 374 }
1da177e4 375}
c2355e1a 376EXPORT_SYMBOL(netpoll_send_skb_on_dev);
1da177e4
LT
377
378void netpoll_send_udp(struct netpoll *np, const char *msg, int len)
379{
954fba02 380 int total_len, ip_len, udp_len;
1da177e4
LT
381 struct sk_buff *skb;
382 struct udphdr *udph;
383 struct iphdr *iph;
384 struct ethhdr *eth;
ee130409 385 static atomic_t ip_ident;
b3d936f3 386 struct ipv6hdr *ip6h;
1da177e4
LT
387
388 udp_len = len + sizeof(*udph);
b3d936f3
CW
389 if (np->ipv6)
390 ip_len = udp_len + sizeof(*ip6h);
391 else
b7394d24
CW
392 ip_len = udp_len + sizeof(*iph);
393
954fba02 394 total_len = ip_len + LL_RESERVED_SPACE(np->dev);
1da177e4 395
954fba02
ED
396 skb = find_skb(np, total_len + np->dev->needed_tailroom,
397 total_len - len);
1da177e4
LT
398 if (!skb)
399 return;
400
27d7ff46 401 skb_copy_to_linear_data(skb, msg, len);
954fba02 402 skb_put(skb, len);
1da177e4 403
4bedb452
ACM
404 skb_push(skb, sizeof(*udph));
405 skb_reset_transport_header(skb);
406 udph = udp_hdr(skb);
1da177e4
LT
407 udph->source = htons(np->local_port);
408 udph->dest = htons(np->remote_port);
409 udph->len = htons(udp_len);
b7394d24 410
b3d936f3
CW
411 if (np->ipv6) {
412 udph->check = 0;
413 udph->check = csum_ipv6_magic(&np->local_ip.in6,
414 &np->remote_ip.in6,
415 udp_len, IPPROTO_UDP,
416 csum_partial(udph, udp_len, 0));
417 if (udph->check == 0)
418 udph->check = CSUM_MANGLED_0;
419
420 skb_push(skb, sizeof(*ip6h));
421 skb_reset_network_header(skb);
422 ip6h = ipv6_hdr(skb);
423
424 /* ip6h->version = 6; ip6h->priority = 0; */
425 put_unaligned(0x60, (unsigned char *)ip6h);
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 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
437 skb_reset_mac_header(skb);
438 skb->protocol = eth->h_proto = htons(ETH_P_IPV6);
439 } else {
b7394d24
CW
440 udph->check = 0;
441 udph->check = csum_tcpudp_magic(np->local_ip.ip,
442 np->remote_ip.ip,
443 udp_len, IPPROTO_UDP,
444 csum_partial(udph, udp_len, 0));
445 if (udph->check == 0)
446 udph->check = CSUM_MANGLED_0;
447
448 skb_push(skb, sizeof(*iph));
449 skb_reset_network_header(skb);
450 iph = ip_hdr(skb);
451
452 /* iph->version = 4; iph->ihl = 5; */
453 put_unaligned(0x45, (unsigned char *)iph);
454 iph->tos = 0;
455 put_unaligned(htons(ip_len), &(iph->tot_len));
456 iph->id = htons(atomic_inc_return(&ip_ident));
457 iph->frag_off = 0;
458 iph->ttl = 64;
459 iph->protocol = IPPROTO_UDP;
460 iph->check = 0;
461 put_unaligned(np->local_ip.ip, &(iph->saddr));
462 put_unaligned(np->remote_ip.ip, &(iph->daddr));
463 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
464
465 eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
466 skb_reset_mac_header(skb);
467 skb->protocol = eth->h_proto = htons(ETH_P_IP);
468 }
469
c62326ab
JP
470 ether_addr_copy(eth->h_source, np->dev->dev_addr);
471 ether_addr_copy(eth->h_dest, np->remote_mac);
1da177e4
LT
472
473 skb->dev = np->dev;
474
475 netpoll_send_skb(np, skb);
476}
9e34a5b5 477EXPORT_SYMBOL(netpoll_send_udp);
1da177e4 478
0bcc1816
SS
479void netpoll_print_options(struct netpoll *np)
480{
e6ec2693 481 np_info(np, "local port %d\n", np->local_port);
b3d936f3
CW
482 if (np->ipv6)
483 np_info(np, "local IPv6 address %pI6c\n", &np->local_ip.in6);
484 else
b7394d24 485 np_info(np, "local IPv4 address %pI4\n", &np->local_ip.ip);
e6ec2693
JP
486 np_info(np, "interface '%s'\n", np->dev_name);
487 np_info(np, "remote port %d\n", np->remote_port);
b3d936f3
CW
488 if (np->ipv6)
489 np_info(np, "remote IPv6 address %pI6c\n", &np->remote_ip.in6);
490 else
b7394d24 491 np_info(np, "remote IPv4 address %pI4\n", &np->remote_ip.ip);
e6ec2693 492 np_info(np, "remote ethernet address %pM\n", np->remote_mac);
0bcc1816 493}
9e34a5b5 494EXPORT_SYMBOL(netpoll_print_options);
0bcc1816 495
b7394d24
CW
496static int netpoll_parse_ip_addr(const char *str, union inet_addr *addr)
497{
498 const char *end;
499
500 if (!strchr(str, ':') &&
501 in4_pton(str, -1, (void *)addr, -1, &end) > 0) {
502 if (!*end)
503 return 0;
504 }
505 if (in6_pton(str, -1, addr->in6.s6_addr, -1, &end) > 0) {
506#if IS_ENABLED(CONFIG_IPV6)
507 if (!*end)
508 return 1;
509#else
510 return -1;
511#endif
512 }
513 return -1;
514}
515
1da177e4
LT
516int netpoll_parse_options(struct netpoll *np, char *opt)
517{
518 char *cur=opt, *delim;
b7394d24 519 int ipv6;
00fe11b3 520 bool ipversion_set = false;
1da177e4 521
c68b9070 522 if (*cur != '@') {
1da177e4
LT
523 if ((delim = strchr(cur, '@')) == NULL)
524 goto parse_failed;
c68b9070 525 *delim = 0;
4b5511eb
AP
526 if (kstrtou16(cur, 10, &np->local_port))
527 goto parse_failed;
c68b9070 528 cur = delim;
1da177e4
LT
529 }
530 cur++;
1da177e4 531
c68b9070 532 if (*cur != '/') {
00fe11b3 533 ipversion_set = true;
1da177e4
LT
534 if ((delim = strchr(cur, '/')) == NULL)
535 goto parse_failed;
c68b9070 536 *delim = 0;
b7394d24
CW
537 ipv6 = netpoll_parse_ip_addr(cur, &np->local_ip);
538 if (ipv6 < 0)
539 goto parse_failed;
540 else
541 np->ipv6 = (bool)ipv6;
c68b9070 542 cur = delim;
1da177e4
LT
543 }
544 cur++;
545
c68b9070 546 if (*cur != ',') {
1da177e4
LT
547 /* parse out dev name */
548 if ((delim = strchr(cur, ',')) == NULL)
549 goto parse_failed;
c68b9070 550 *delim = 0;
1da177e4 551 strlcpy(np->dev_name, cur, sizeof(np->dev_name));
c68b9070 552 cur = delim;
1da177e4
LT
553 }
554 cur++;
555
c68b9070 556 if (*cur != '@') {
1da177e4
LT
557 /* dst port */
558 if ((delim = strchr(cur, '@')) == NULL)
559 goto parse_failed;
c68b9070 560 *delim = 0;
5fc05f87 561 if (*cur == ' ' || *cur == '\t')
e6ec2693 562 np_info(np, "warning: whitespace is not allowed\n");
4b5511eb
AP
563 if (kstrtou16(cur, 10, &np->remote_port))
564 goto parse_failed;
c68b9070 565 cur = delim;
1da177e4
LT
566 }
567 cur++;
1da177e4
LT
568
569 /* dst ip */
570 if ((delim = strchr(cur, '/')) == NULL)
571 goto parse_failed;
c68b9070 572 *delim = 0;
b7394d24
CW
573 ipv6 = netpoll_parse_ip_addr(cur, &np->remote_ip);
574 if (ipv6 < 0)
575 goto parse_failed;
00fe11b3 576 else if (ipversion_set && np->ipv6 != (bool)ipv6)
b7394d24
CW
577 goto parse_failed;
578 else
579 np->ipv6 = (bool)ipv6;
c68b9070 580 cur = delim + 1;
1da177e4 581
c68b9070 582 if (*cur != 0) {
1da177e4 583 /* MAC address */
4940fc88 584 if (!mac_pton(cur, np->remote_mac))
1da177e4 585 goto parse_failed;
1da177e4
LT
586 }
587
0bcc1816 588 netpoll_print_options(np);
1da177e4
LT
589
590 return 0;
591
592 parse_failed:
e6ec2693 593 np_info(np, "couldn't parse config at '%s'!\n", cur);
1da177e4
LT
594 return -1;
595}
9e34a5b5 596EXPORT_SYMBOL(netpoll_parse_options);
1da177e4 597
a8779ec1 598int __netpoll_setup(struct netpoll *np, struct net_device *ndev)
1da177e4 599{
115c1d6e 600 struct netpoll_info *npinfo;
4247e161 601 const struct net_device_ops *ops;
b41848b6 602 int err;
1da177e4 603
30fdd8a0
JP
604 np->dev = ndev;
605 strlcpy(np->dev_name, ndev->name, IFNAMSIZ);
2cde6acd 606 INIT_WORK(&np->cleanup_work, netpoll_async_cleanup);
30fdd8a0 607
8fdd95ec
HX
608 if ((ndev->priv_flags & IFF_DISABLE_NETPOLL) ||
609 !ndev->netdev_ops->ndo_poll_controller) {
e6ec2693
JP
610 np_err(np, "%s doesn't support polling, aborting\n",
611 np->dev_name);
8fdd95ec
HX
612 err = -ENOTSUPP;
613 goto out;
614 }
615
616 if (!ndev->npinfo) {
a8779ec1 617 npinfo = kmalloc(sizeof(*npinfo), GFP_KERNEL);
8fdd95ec
HX
618 if (!npinfo) {
619 err = -ENOMEM;
620 goto out;
621 }
622
bd7c4b60 623 sema_init(&npinfo->dev_lock, 1);
8fdd95ec
HX
624 skb_queue_head_init(&npinfo->txq);
625 INIT_DELAYED_WORK(&npinfo->tx_work, queue_process);
626
627 atomic_set(&npinfo->refcnt, 1);
628
629 ops = np->dev->netdev_ops;
630 if (ops->ndo_netpoll_setup) {
a8779ec1 631 err = ops->ndo_netpoll_setup(ndev, npinfo);
8fdd95ec
HX
632 if (err)
633 goto free_npinfo;
634 }
635 } else {
0790bbb6 636 npinfo = rtnl_dereference(ndev->npinfo);
8fdd95ec
HX
637 atomic_inc(&npinfo->refcnt);
638 }
639
640 npinfo->netpoll = np;
641
8fdd95ec 642 /* last thing to do is link it to the net device structure */
cf778b00 643 rcu_assign_pointer(ndev->npinfo, npinfo);
8fdd95ec
HX
644
645 return 0;
646
647free_npinfo:
648 kfree(npinfo);
649out:
650 return err;
651}
652EXPORT_SYMBOL_GPL(__netpoll_setup);
653
654int netpoll_setup(struct netpoll *np)
655{
656 struct net_device *ndev = NULL;
657 struct in_device *in_dev;
658 int err;
659
f92d3180 660 rtnl_lock();
556e6256
CW
661 if (np->dev_name) {
662 struct net *net = current->nsproxy->net_ns;
663 ndev = __dev_get_by_name(net, np->dev_name);
664 }
1da177e4 665 if (!ndev) {
e6ec2693 666 np_err(np, "%s doesn't exist, aborting\n", np->dev_name);
f92d3180
CW
667 err = -ENODEV;
668 goto unlock;
1da177e4 669 }
5bd30d39 670 dev_hold(ndev);
1da177e4 671
49bd8fb0 672 if (netdev_master_upper_dev_get(ndev)) {
e6ec2693 673 np_err(np, "%s is a slave device, aborting\n", np->dev_name);
83fe32de
DC
674 err = -EBUSY;
675 goto put;
0c1ad04a
WC
676 }
677
1da177e4
LT
678 if (!netif_running(ndev)) {
679 unsigned long atmost, atleast;
680
e6ec2693 681 np_info(np, "device %s not up yet, forcing it\n", np->dev_name);
1da177e4 682
b41848b6 683 err = dev_open(ndev);
b41848b6
SH
684
685 if (err) {
e6ec2693 686 np_err(np, "failed to open %s\n", ndev->name);
dbaa1541 687 goto put;
1da177e4 688 }
1da177e4 689
f92d3180 690 rtnl_unlock();
1da177e4 691 atleast = jiffies + HZ/10;
bff38771 692 atmost = jiffies + carrier_timeout * HZ;
1da177e4
LT
693 while (!netif_carrier_ok(ndev)) {
694 if (time_after(jiffies, atmost)) {
e6ec2693 695 np_notice(np, "timeout waiting for carrier\n");
1da177e4
LT
696 break;
697 }
1b614fb9 698 msleep(1);
1da177e4
LT
699 }
700
701 /* If carrier appears to come up instantly, we don't
702 * trust it and pause so that we don't pump all our
703 * queued console messages into the bitbucket.
704 */
705
706 if (time_before(jiffies, atleast)) {
e6ec2693 707 np_notice(np, "carrier detect appears untrustworthy, waiting 4 seconds\n");
1da177e4
LT
708 msleep(4000);
709 }
f92d3180 710 rtnl_lock();
1da177e4
LT
711 }
712
b7394d24
CW
713 if (!np->local_ip.ip) {
714 if (!np->ipv6) {
f92d3180 715 in_dev = __in_dev_get_rtnl(ndev);
b7394d24
CW
716
717 if (!in_dev || !in_dev->ifa_list) {
b7394d24
CW
718 np_err(np, "no IP address for %s, aborting\n",
719 np->dev_name);
720 err = -EDESTADDRREQ;
721 goto put;
722 }
723
724 np->local_ip.ip = in_dev->ifa_list->ifa_local;
b7394d24 725 np_info(np, "local IP %pI4\n", &np->local_ip.ip);
b3d936f3
CW
726 } else {
727#if IS_ENABLED(CONFIG_IPV6)
728 struct inet6_dev *idev;
729
730 err = -EDESTADDRREQ;
b3d936f3
CW
731 idev = __in6_dev_get(ndev);
732 if (idev) {
733 struct inet6_ifaddr *ifp;
734
735 read_lock_bh(&idev->lock);
736 list_for_each_entry(ifp, &idev->addr_list, if_list) {
737 if (ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL)
738 continue;
739 np->local_ip.in6 = ifp->addr;
740 err = 0;
741 break;
742 }
743 read_unlock_bh(&idev->lock);
744 }
b3d936f3
CW
745 if (err) {
746 np_err(np, "no IPv6 address for %s, aborting\n",
747 np->dev_name);
748 goto put;
749 } else
750 np_info(np, "local IPv6 %pI6c\n", &np->local_ip.in6);
751#else
752 np_err(np, "IPv6 is not supported %s, aborting\n",
753 np->dev_name);
e39363a9 754 err = -EINVAL;
b3d936f3
CW
755 goto put;
756#endif
1da177e4 757 }
1da177e4
LT
758 }
759
dbaa1541
HX
760 /* fill up the skb queue */
761 refill_skbs();
762
a8779ec1 763 err = __netpoll_setup(np, ndev);
8fdd95ec
HX
764 if (err)
765 goto put;
766
f92d3180 767 rtnl_unlock();
1da177e4
LT
768 return 0;
769
21edbb22 770put:
1da177e4 771 dev_put(ndev);
f92d3180
CW
772unlock:
773 rtnl_unlock();
b41848b6 774 return err;
1da177e4 775}
9e34a5b5 776EXPORT_SYMBOL(netpoll_setup);
1da177e4 777
c68b9070
DM
778static int __init netpoll_init(void)
779{
a1bcfacd
SH
780 skb_queue_head_init(&skb_pool);
781 return 0;
782}
783core_initcall(netpoll_init);
784
38e6bc18
AW
785static void rcu_cleanup_netpoll_info(struct rcu_head *rcu_head)
786{
787 struct netpoll_info *npinfo =
788 container_of(rcu_head, struct netpoll_info, rcu);
789
38e6bc18
AW
790 skb_queue_purge(&npinfo->txq);
791
792 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
793 cancel_delayed_work(&npinfo->tx_work);
794
795 /* clean after last, unfinished work */
796 __skb_queue_purge(&npinfo->txq);
797 /* now cancel it again */
798 cancel_delayed_work(&npinfo->tx_work);
799 kfree(npinfo);
800}
801
8fdd95ec 802void __netpoll_cleanup(struct netpoll *np)
1da177e4 803{
fbeec2e1 804 struct netpoll_info *npinfo;
fbeec2e1 805
0790bbb6
NH
806 /* rtnl_dereference would be preferable here but
807 * rcu_cleanup_netpoll path can put us in here safely without
808 * holding the rtnl, so plain rcu_dereference it is
809 */
810 npinfo = rtnl_dereference(np->dev->npinfo);
8fdd95ec 811 if (!npinfo)
dbaa1541 812 return;
93ec2c72 813
ca99ca14
NH
814 synchronize_srcu(&netpoll_srcu);
815
8fdd95ec
HX
816 if (atomic_dec_and_test(&npinfo->refcnt)) {
817 const struct net_device_ops *ops;
de85d99e 818
8fdd95ec
HX
819 ops = np->dev->netdev_ops;
820 if (ops->ndo_netpoll_cleanup)
821 ops->ndo_netpoll_cleanup(np->dev);
de85d99e 822
fcb144b5 823 RCU_INIT_POINTER(np->dev->npinfo, NULL);
38e6bc18
AW
824 call_rcu_bh(&npinfo->rcu, rcu_cleanup_netpoll_info);
825 }
826}
827EXPORT_SYMBOL_GPL(__netpoll_cleanup);
de85d99e 828
2cde6acd 829static void netpoll_async_cleanup(struct work_struct *work)
38e6bc18 830{
2cde6acd 831 struct netpoll *np = container_of(work, struct netpoll, cleanup_work);
93ec2c72 832
2cde6acd 833 rtnl_lock();
38e6bc18 834 __netpoll_cleanup(np);
2cde6acd 835 rtnl_unlock();
38e6bc18
AW
836 kfree(np);
837}
93ec2c72 838
2cde6acd 839void __netpoll_free_async(struct netpoll *np)
38e6bc18 840{
2cde6acd 841 schedule_work(&np->cleanup_work);
8fdd95ec 842}
2cde6acd 843EXPORT_SYMBOL_GPL(__netpoll_free_async);
fbeec2e1 844
8fdd95ec
HX
845void netpoll_cleanup(struct netpoll *np)
846{
8fdd95ec 847 rtnl_lock();
d0fe8c88
NA
848 if (!np->dev)
849 goto out;
8fdd95ec 850 __netpoll_cleanup(np);
8fdd95ec 851 dev_put(np->dev);
1da177e4 852 np->dev = NULL;
d0fe8c88
NA
853out:
854 rtnl_unlock();
1da177e4 855}
9e34a5b5 856EXPORT_SYMBOL(netpoll_cleanup);