2 * Common framework for low-level network console, dump, and debugger code
4 * Sep 8 2003 Matt Mackall <mpm@selenic.com>
6 * based on the netconsole code from:
8 * Copyright (C) 2001 Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2002 Red Hat, Inc.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/moduleparam.h>
15 #include <linux/kernel.h>
16 #include <linux/netdevice.h>
17 #include <linux/etherdevice.h>
18 #include <linux/string.h>
19 #include <linux/if_arp.h>
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>
28 #include <linux/slab.h>
29 #include <linux/export.h>
30 #include <linux/if_vlan.h>
33 #include <net/addrconf.h>
34 #include <net/ndisc.h>
35 #include <net/ip6_checksum.h>
36 #include <asm/unaligned.h>
37 #include <trace/events/napi.h>
40 * We maintain a small pool of fully-sized skbs, to make sure the
41 * message gets out even in extreme OOM situations.
44 #define MAX_UDP_CHUNK 1460
47 static struct sk_buff_head skb_pool
;
49 DEFINE_STATIC_SRCU(netpoll_srcu
);
51 #define USEC_PER_POLL 50
53 #define MAX_SKB_SIZE \
54 (sizeof(struct ethhdr) + \
55 sizeof(struct iphdr) + \
56 sizeof(struct udphdr) + \
59 static void zap_completion_queue(void);
60 static void netpoll_async_cleanup(struct work_struct
*work
);
62 static unsigned int carrier_timeout
= 4;
63 module_param(carrier_timeout
, uint
, 0644);
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__)
72 static int netpoll_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
73 struct netdev_queue
*txq
)
75 int status
= NETDEV_TX_OK
;
76 netdev_features_t features
;
78 features
= netif_skb_features(skb
);
80 if (skb_vlan_tag_present(skb
) &&
81 !vlan_hw_offload_capable(features
, skb
->vlan_proto
)) {
82 skb
= __vlan_hwaccel_push_inside(skb
);
84 /* This is actually a packet drop, but we
85 * don't want the code that calls this
86 * function to try and operate on a NULL skb.
92 status
= netdev_start_xmit(skb
, dev
, txq
, false);
98 static void queue_process(struct work_struct
*work
)
100 struct netpoll_info
*npinfo
=
101 container_of(work
, struct netpoll_info
, tx_work
.work
);
105 while ((skb
= skb_dequeue(&npinfo
->txq
))) {
106 struct net_device
*dev
= skb
->dev
;
107 struct netdev_queue
*txq
;
109 if (!netif_device_present(dev
) || !netif_running(dev
)) {
114 txq
= skb_get_tx_queue(dev
, skb
);
116 local_irq_save(flags
);
117 HARD_TX_LOCK(dev
, txq
, smp_processor_id());
118 if (netif_xmit_frozen_or_stopped(txq
) ||
119 netpoll_start_xmit(skb
, dev
, txq
) != NETDEV_TX_OK
) {
120 skb_queue_head(&npinfo
->txq
, skb
);
121 HARD_TX_UNLOCK(dev
, txq
);
122 local_irq_restore(flags
);
124 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
127 HARD_TX_UNLOCK(dev
, txq
);
128 local_irq_restore(flags
);
133 * Check whether delayed processing was scheduled for our NIC. If so,
134 * we attempt to grab the poll lock and use ->poll() to pump the card.
135 * If this fails, either we've recursed in ->poll() or it's already
136 * running on another CPU.
138 * Note: we don't mask interrupts with this lock because we're using
139 * trylock here and interrupts are already disabled in the softirq
140 * case. Further, we test the poll_owner to avoid recursion on UP
141 * systems where the lock doesn't exist.
143 static void poll_one_napi(struct napi_struct
*napi
)
147 /* net_rx_action's ->poll() invocations and our's are
148 * synchronized by this test which is only made while
149 * holding the napi->poll_lock.
151 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
154 /* If we set this bit but see that it has already been set,
155 * that indicates that napi has been disabled and we need
156 * to abort this operation
158 if (test_and_set_bit(NAPI_STATE_NPSVC
, &napi
->state
))
161 /* We explicilty pass the polling call a budget of 0 to
162 * indicate that we are clearing the Tx path only.
164 work
= napi
->poll(napi
, 0);
165 WARN_ONCE(work
, "%pF exceeded budget in poll\n", napi
->poll
);
166 trace_napi_poll(napi
, work
, 0);
168 clear_bit(NAPI_STATE_NPSVC
, &napi
->state
);
171 static void poll_napi(struct net_device
*dev
)
173 struct napi_struct
*napi
;
175 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
176 if (napi
->poll_owner
!= smp_processor_id() &&
177 spin_trylock(&napi
->poll_lock
)) {
179 spin_unlock(&napi
->poll_lock
);
184 static void netpoll_poll_dev(struct net_device
*dev
)
186 const struct net_device_ops
*ops
;
187 struct netpoll_info
*ni
= rcu_dereference_bh(dev
->npinfo
);
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
193 if (down_trylock(&ni
->dev_lock
))
196 if (!netif_running(dev
)) {
201 ops
= dev
->netdev_ops
;
202 if (!ops
->ndo_poll_controller
) {
207 /* Process pending work on NIC */
208 ops
->ndo_poll_controller(dev
);
214 zap_completion_queue();
217 void netpoll_poll_disable(struct net_device
*dev
)
219 struct netpoll_info
*ni
;
222 idx
= srcu_read_lock(&netpoll_srcu
);
223 ni
= srcu_dereference(dev
->npinfo
, &netpoll_srcu
);
226 srcu_read_unlock(&netpoll_srcu
, idx
);
228 EXPORT_SYMBOL(netpoll_poll_disable
);
230 void netpoll_poll_enable(struct net_device
*dev
)
232 struct netpoll_info
*ni
;
234 ni
= rcu_dereference(dev
->npinfo
);
239 EXPORT_SYMBOL(netpoll_poll_enable
);
241 static void refill_skbs(void)
246 spin_lock_irqsave(&skb_pool
.lock
, flags
);
247 while (skb_pool
.qlen
< MAX_SKBS
) {
248 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
252 __skb_queue_tail(&skb_pool
, skb
);
254 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
257 static void zap_completion_queue(void)
260 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
262 if (sd
->completion_queue
) {
263 struct sk_buff
*clist
;
265 local_irq_save(flags
);
266 clist
= sd
->completion_queue
;
267 sd
->completion_queue
= NULL
;
268 local_irq_restore(flags
);
270 while (clist
!= NULL
) {
271 struct sk_buff
*skb
= clist
;
273 if (!skb_irq_freeable(skb
)) {
274 atomic_inc(&skb
->users
);
275 dev_kfree_skb_any(skb
); /* put this one back */
282 put_cpu_var(softnet_data
);
285 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
290 zap_completion_queue();
294 skb
= alloc_skb(len
, GFP_ATOMIC
);
296 skb
= skb_dequeue(&skb_pool
);
300 netpoll_poll_dev(np
->dev
);
306 atomic_set(&skb
->users
, 1);
307 skb_reserve(skb
, reserve
);
311 static int netpoll_owner_active(struct net_device
*dev
)
313 struct napi_struct
*napi
;
315 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
316 if (napi
->poll_owner
== smp_processor_id())
322 /* call with IRQ disabled */
323 void netpoll_send_skb_on_dev(struct netpoll
*np
, struct sk_buff
*skb
,
324 struct net_device
*dev
)
326 int status
= NETDEV_TX_BUSY
;
328 /* It is up to the caller to keep npinfo alive. */
329 struct netpoll_info
*npinfo
;
331 WARN_ON_ONCE(!irqs_disabled());
333 npinfo
= rcu_dereference_bh(np
->dev
->npinfo
);
334 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
335 dev_kfree_skb_irq(skb
);
339 /* don't get messages out of order, and no recursion */
340 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
341 struct netdev_queue
*txq
;
343 txq
= netdev_pick_tx(dev
, skb
, NULL
);
345 /* try until next clock tick */
346 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
347 tries
> 0; --tries
) {
348 if (HARD_TX_TRYLOCK(dev
, txq
)) {
349 if (!netif_xmit_stopped(txq
))
350 status
= netpoll_start_xmit(skb
, dev
, txq
);
352 HARD_TX_UNLOCK(dev
, txq
);
354 if (status
== NETDEV_TX_OK
)
359 /* tickle device maybe there is some cleanup */
360 netpoll_poll_dev(np
->dev
);
362 udelay(USEC_PER_POLL
);
365 WARN_ONCE(!irqs_disabled(),
366 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
367 dev
->name
, dev
->netdev_ops
->ndo_start_xmit
);
371 if (status
!= NETDEV_TX_OK
) {
372 skb_queue_tail(&npinfo
->txq
, skb
);
373 schedule_delayed_work(&npinfo
->tx_work
,0);
376 EXPORT_SYMBOL(netpoll_send_skb_on_dev
);
378 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
380 int total_len
, ip_len
, udp_len
;
385 static atomic_t ip_ident
;
386 struct ipv6hdr
*ip6h
;
388 WARN_ON_ONCE(!irqs_disabled());
390 udp_len
= len
+ sizeof(*udph
);
392 ip_len
= udp_len
+ sizeof(*ip6h
);
394 ip_len
= udp_len
+ sizeof(*iph
);
396 total_len
= ip_len
+ LL_RESERVED_SPACE(np
->dev
);
398 skb
= find_skb(np
, total_len
+ np
->dev
->needed_tailroom
,
403 skb_copy_to_linear_data(skb
, msg
, len
);
406 skb_push(skb
, sizeof(*udph
));
407 skb_reset_transport_header(skb
);
409 udph
->source
= htons(np
->local_port
);
410 udph
->dest
= htons(np
->remote_port
);
411 udph
->len
= htons(udp_len
);
415 udph
->check
= csum_ipv6_magic(&np
->local_ip
.in6
,
417 udp_len
, IPPROTO_UDP
,
418 csum_partial(udph
, udp_len
, 0));
419 if (udph
->check
== 0)
420 udph
->check
= CSUM_MANGLED_0
;
422 skb_push(skb
, sizeof(*ip6h
));
423 skb_reset_network_header(skb
);
424 ip6h
= ipv6_hdr(skb
);
426 /* ip6h->version = 6; ip6h->priority = 0; */
427 put_unaligned(0x60, (unsigned char *)ip6h
);
428 ip6h
->flow_lbl
[0] = 0;
429 ip6h
->flow_lbl
[1] = 0;
430 ip6h
->flow_lbl
[2] = 0;
432 ip6h
->payload_len
= htons(sizeof(struct udphdr
) + len
);
433 ip6h
->nexthdr
= IPPROTO_UDP
;
434 ip6h
->hop_limit
= 32;
435 ip6h
->saddr
= np
->local_ip
.in6
;
436 ip6h
->daddr
= np
->remote_ip
.in6
;
438 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
439 skb_reset_mac_header(skb
);
440 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IPV6
);
443 udph
->check
= csum_tcpudp_magic(np
->local_ip
.ip
,
445 udp_len
, IPPROTO_UDP
,
446 csum_partial(udph
, udp_len
, 0));
447 if (udph
->check
== 0)
448 udph
->check
= CSUM_MANGLED_0
;
450 skb_push(skb
, sizeof(*iph
));
451 skb_reset_network_header(skb
);
454 /* iph->version = 4; iph->ihl = 5; */
455 put_unaligned(0x45, (unsigned char *)iph
);
457 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
458 iph
->id
= htons(atomic_inc_return(&ip_ident
));
461 iph
->protocol
= IPPROTO_UDP
;
463 put_unaligned(np
->local_ip
.ip
, &(iph
->saddr
));
464 put_unaligned(np
->remote_ip
.ip
, &(iph
->daddr
));
465 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
467 eth
= (struct ethhdr
*) skb_push(skb
, ETH_HLEN
);
468 skb_reset_mac_header(skb
);
469 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
472 ether_addr_copy(eth
->h_source
, np
->dev
->dev_addr
);
473 ether_addr_copy(eth
->h_dest
, np
->remote_mac
);
477 netpoll_send_skb(np
, skb
);
479 EXPORT_SYMBOL(netpoll_send_udp
);
481 void netpoll_print_options(struct netpoll
*np
)
483 np_info(np
, "local port %d\n", np
->local_port
);
485 np_info(np
, "local IPv6 address %pI6c\n", &np
->local_ip
.in6
);
487 np_info(np
, "local IPv4 address %pI4\n", &np
->local_ip
.ip
);
488 np_info(np
, "interface '%s'\n", np
->dev_name
);
489 np_info(np
, "remote port %d\n", np
->remote_port
);
491 np_info(np
, "remote IPv6 address %pI6c\n", &np
->remote_ip
.in6
);
493 np_info(np
, "remote IPv4 address %pI4\n", &np
->remote_ip
.ip
);
494 np_info(np
, "remote ethernet address %pM\n", np
->remote_mac
);
496 EXPORT_SYMBOL(netpoll_print_options
);
498 static int netpoll_parse_ip_addr(const char *str
, union inet_addr
*addr
)
502 if (!strchr(str
, ':') &&
503 in4_pton(str
, -1, (void *)addr
, -1, &end
) > 0) {
507 if (in6_pton(str
, -1, addr
->in6
.s6_addr
, -1, &end
) > 0) {
508 #if IS_ENABLED(CONFIG_IPV6)
518 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
520 char *cur
=opt
, *delim
;
522 bool ipversion_set
= false;
525 if ((delim
= strchr(cur
, '@')) == NULL
)
528 if (kstrtou16(cur
, 10, &np
->local_port
))
535 ipversion_set
= true;
536 if ((delim
= strchr(cur
, '/')) == NULL
)
539 ipv6
= netpoll_parse_ip_addr(cur
, &np
->local_ip
);
543 np
->ipv6
= (bool)ipv6
;
549 /* parse out dev name */
550 if ((delim
= strchr(cur
, ',')) == NULL
)
553 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
560 if ((delim
= strchr(cur
, '@')) == NULL
)
563 if (*cur
== ' ' || *cur
== '\t')
564 np_info(np
, "warning: whitespace is not allowed\n");
565 if (kstrtou16(cur
, 10, &np
->remote_port
))
572 if ((delim
= strchr(cur
, '/')) == NULL
)
575 ipv6
= netpoll_parse_ip_addr(cur
, &np
->remote_ip
);
578 else if (ipversion_set
&& np
->ipv6
!= (bool)ipv6
)
581 np
->ipv6
= (bool)ipv6
;
586 if (!mac_pton(cur
, np
->remote_mac
))
590 netpoll_print_options(np
);
595 np_info(np
, "couldn't parse config at '%s'!\n", cur
);
598 EXPORT_SYMBOL(netpoll_parse_options
);
600 int __netpoll_setup(struct netpoll
*np
, struct net_device
*ndev
)
602 struct netpoll_info
*npinfo
;
603 const struct net_device_ops
*ops
;
607 strlcpy(np
->dev_name
, ndev
->name
, IFNAMSIZ
);
608 INIT_WORK(&np
->cleanup_work
, netpoll_async_cleanup
);
610 if ((ndev
->priv_flags
& IFF_DISABLE_NETPOLL
) ||
611 !ndev
->netdev_ops
->ndo_poll_controller
) {
612 np_err(np
, "%s doesn't support polling, aborting\n",
619 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
625 sema_init(&npinfo
->dev_lock
, 1);
626 skb_queue_head_init(&npinfo
->txq
);
627 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
629 atomic_set(&npinfo
->refcnt
, 1);
631 ops
= np
->dev
->netdev_ops
;
632 if (ops
->ndo_netpoll_setup
) {
633 err
= ops
->ndo_netpoll_setup(ndev
, npinfo
);
638 npinfo
= rtnl_dereference(ndev
->npinfo
);
639 atomic_inc(&npinfo
->refcnt
);
642 npinfo
->netpoll
= np
;
644 /* last thing to do is link it to the net device structure */
645 rcu_assign_pointer(ndev
->npinfo
, npinfo
);
654 EXPORT_SYMBOL_GPL(__netpoll_setup
);
656 int netpoll_setup(struct netpoll
*np
)
658 struct net_device
*ndev
= NULL
;
659 struct in_device
*in_dev
;
664 struct net
*net
= current
->nsproxy
->net_ns
;
665 ndev
= __dev_get_by_name(net
, np
->dev_name
);
668 np_err(np
, "%s doesn't exist, aborting\n", np
->dev_name
);
674 if (netdev_master_upper_dev_get(ndev
)) {
675 np_err(np
, "%s is a slave device, aborting\n", np
->dev_name
);
680 if (!netif_running(ndev
)) {
681 unsigned long atmost
, atleast
;
683 np_info(np
, "device %s not up yet, forcing it\n", np
->dev_name
);
685 err
= dev_open(ndev
);
688 np_err(np
, "failed to open %s\n", ndev
->name
);
693 atleast
= jiffies
+ HZ
/10;
694 atmost
= jiffies
+ carrier_timeout
* HZ
;
695 while (!netif_carrier_ok(ndev
)) {
696 if (time_after(jiffies
, atmost
)) {
697 np_notice(np
, "timeout waiting for carrier\n");
703 /* If carrier appears to come up instantly, we don't
704 * trust it and pause so that we don't pump all our
705 * queued console messages into the bitbucket.
708 if (time_before(jiffies
, atleast
)) {
709 np_notice(np
, "carrier detect appears untrustworthy, waiting 4 seconds\n");
715 if (!np
->local_ip
.ip
) {
717 in_dev
= __in_dev_get_rtnl(ndev
);
719 if (!in_dev
|| !in_dev
->ifa_list
) {
720 np_err(np
, "no IP address for %s, aborting\n",
726 np
->local_ip
.ip
= in_dev
->ifa_list
->ifa_local
;
727 np_info(np
, "local IP %pI4\n", &np
->local_ip
.ip
);
729 #if IS_ENABLED(CONFIG_IPV6)
730 struct inet6_dev
*idev
;
733 idev
= __in6_dev_get(ndev
);
735 struct inet6_ifaddr
*ifp
;
737 read_lock_bh(&idev
->lock
);
738 list_for_each_entry(ifp
, &idev
->addr_list
, if_list
) {
739 if (ipv6_addr_type(&ifp
->addr
) & IPV6_ADDR_LINKLOCAL
)
741 np
->local_ip
.in6
= ifp
->addr
;
745 read_unlock_bh(&idev
->lock
);
748 np_err(np
, "no IPv6 address for %s, aborting\n",
752 np_info(np
, "local IPv6 %pI6c\n", &np
->local_ip
.in6
);
754 np_err(np
, "IPv6 is not supported %s, aborting\n",
762 /* fill up the skb queue */
765 err
= __netpoll_setup(np
, ndev
);
778 EXPORT_SYMBOL(netpoll_setup
);
780 static int __init
netpoll_init(void)
782 skb_queue_head_init(&skb_pool
);
785 core_initcall(netpoll_init
);
787 static void rcu_cleanup_netpoll_info(struct rcu_head
*rcu_head
)
789 struct netpoll_info
*npinfo
=
790 container_of(rcu_head
, struct netpoll_info
, rcu
);
792 skb_queue_purge(&npinfo
->txq
);
794 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
795 cancel_delayed_work(&npinfo
->tx_work
);
797 /* clean after last, unfinished work */
798 __skb_queue_purge(&npinfo
->txq
);
799 /* now cancel it again */
800 cancel_delayed_work(&npinfo
->tx_work
);
804 void __netpoll_cleanup(struct netpoll
*np
)
806 struct netpoll_info
*npinfo
;
808 /* rtnl_dereference would be preferable here but
809 * rcu_cleanup_netpoll path can put us in here safely without
810 * holding the rtnl, so plain rcu_dereference it is
812 npinfo
= rtnl_dereference(np
->dev
->npinfo
);
816 synchronize_srcu(&netpoll_srcu
);
818 if (atomic_dec_and_test(&npinfo
->refcnt
)) {
819 const struct net_device_ops
*ops
;
821 ops
= np
->dev
->netdev_ops
;
822 if (ops
->ndo_netpoll_cleanup
)
823 ops
->ndo_netpoll_cleanup(np
->dev
);
825 RCU_INIT_POINTER(np
->dev
->npinfo
, NULL
);
826 call_rcu_bh(&npinfo
->rcu
, rcu_cleanup_netpoll_info
);
828 RCU_INIT_POINTER(np
->dev
->npinfo
, NULL
);
830 EXPORT_SYMBOL_GPL(__netpoll_cleanup
);
832 static void netpoll_async_cleanup(struct work_struct
*work
)
834 struct netpoll
*np
= container_of(work
, struct netpoll
, cleanup_work
);
837 __netpoll_cleanup(np
);
842 void __netpoll_free_async(struct netpoll
*np
)
844 schedule_work(&np
->cleanup_work
);
846 EXPORT_SYMBOL_GPL(__netpoll_free_async
);
848 void netpoll_cleanup(struct netpoll
*np
)
853 __netpoll_cleanup(np
);
859 EXPORT_SYMBOL(netpoll_cleanup
);