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
;
108 unsigned int q_index
;
110 if (!netif_device_present(dev
) || !netif_running(dev
)) {
115 local_irq_save(flags
);
116 /* check if skb->queue_mapping is still valid */
117 q_index
= skb_get_queue_mapping(skb
);
118 if (unlikely(q_index
>= dev
->real_num_tx_queues
)) {
119 q_index
= q_index
% dev
->real_num_tx_queues
;
120 skb_set_queue_mapping(skb
, q_index
);
122 txq
= netdev_get_tx_queue(dev
, q_index
);
123 HARD_TX_LOCK(dev
, txq
, smp_processor_id());
124 if (netif_xmit_frozen_or_stopped(txq
) ||
125 netpoll_start_xmit(skb
, dev
, txq
) != NETDEV_TX_OK
) {
126 skb_queue_head(&npinfo
->txq
, skb
);
127 HARD_TX_UNLOCK(dev
, txq
);
128 local_irq_restore(flags
);
130 schedule_delayed_work(&npinfo
->tx_work
, HZ
/10);
133 HARD_TX_UNLOCK(dev
, txq
);
134 local_irq_restore(flags
);
139 * Check whether delayed processing was scheduled for our NIC. If so,
140 * we attempt to grab the poll lock and use ->poll() to pump the card.
141 * If this fails, either we've recursed in ->poll() or it's already
142 * running on another CPU.
144 * Note: we don't mask interrupts with this lock because we're using
145 * trylock here and interrupts are already disabled in the softirq
146 * case. Further, we test the poll_owner to avoid recursion on UP
147 * systems where the lock doesn't exist.
149 static void poll_one_napi(struct napi_struct
*napi
)
153 /* net_rx_action's ->poll() invocations and our's are
154 * synchronized by this test which is only made while
155 * holding the napi->poll_lock.
157 if (!test_bit(NAPI_STATE_SCHED
, &napi
->state
))
160 /* If we set this bit but see that it has already been set,
161 * that indicates that napi has been disabled and we need
162 * to abort this operation
164 if (test_and_set_bit(NAPI_STATE_NPSVC
, &napi
->state
))
167 /* We explicilty pass the polling call a budget of 0 to
168 * indicate that we are clearing the Tx path only.
170 work
= napi
->poll(napi
, 0);
171 WARN_ONCE(work
, "%pF exceeded budget in poll\n", napi
->poll
);
172 trace_napi_poll(napi
, work
, 0);
174 clear_bit(NAPI_STATE_NPSVC
, &napi
->state
);
177 static void poll_napi(struct net_device
*dev
)
179 struct napi_struct
*napi
;
180 int cpu
= smp_processor_id();
182 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
183 if (cmpxchg(&napi
->poll_owner
, -1, cpu
) == -1) {
185 smp_store_release(&napi
->poll_owner
, -1);
190 static void netpoll_poll_dev(struct net_device
*dev
)
192 const struct net_device_ops
*ops
;
193 struct netpoll_info
*ni
= rcu_dereference_bh(dev
->npinfo
);
195 /* Don't do any rx activity if the dev_lock mutex is held
196 * the dev_open/close paths use this to block netpoll activity
197 * while changing device state
199 if (down_trylock(&ni
->dev_lock
))
202 if (!netif_running(dev
)) {
207 ops
= dev
->netdev_ops
;
208 if (!ops
->ndo_poll_controller
) {
213 /* Process pending work on NIC */
214 ops
->ndo_poll_controller(dev
);
220 zap_completion_queue();
223 void netpoll_poll_disable(struct net_device
*dev
)
225 struct netpoll_info
*ni
;
228 idx
= srcu_read_lock(&netpoll_srcu
);
229 ni
= srcu_dereference(dev
->npinfo
, &netpoll_srcu
);
232 srcu_read_unlock(&netpoll_srcu
, idx
);
234 EXPORT_SYMBOL(netpoll_poll_disable
);
236 void netpoll_poll_enable(struct net_device
*dev
)
238 struct netpoll_info
*ni
;
240 ni
= rcu_dereference(dev
->npinfo
);
245 EXPORT_SYMBOL(netpoll_poll_enable
);
247 static void refill_skbs(void)
252 spin_lock_irqsave(&skb_pool
.lock
, flags
);
253 while (skb_pool
.qlen
< MAX_SKBS
) {
254 skb
= alloc_skb(MAX_SKB_SIZE
, GFP_ATOMIC
);
258 __skb_queue_tail(&skb_pool
, skb
);
260 spin_unlock_irqrestore(&skb_pool
.lock
, flags
);
263 static void zap_completion_queue(void)
266 struct softnet_data
*sd
= &get_cpu_var(softnet_data
);
268 if (sd
->completion_queue
) {
269 struct sk_buff
*clist
;
271 local_irq_save(flags
);
272 clist
= sd
->completion_queue
;
273 sd
->completion_queue
= NULL
;
274 local_irq_restore(flags
);
276 while (clist
!= NULL
) {
277 struct sk_buff
*skb
= clist
;
279 if (!skb_irq_freeable(skb
)) {
280 refcount_set(&skb
->users
, 1);
281 dev_kfree_skb_any(skb
); /* put this one back */
288 put_cpu_var(softnet_data
);
291 static struct sk_buff
*find_skb(struct netpoll
*np
, int len
, int reserve
)
296 zap_completion_queue();
300 skb
= alloc_skb(len
, GFP_ATOMIC
);
302 skb
= skb_dequeue(&skb_pool
);
306 netpoll_poll_dev(np
->dev
);
312 refcount_set(&skb
->users
, 1);
313 skb_reserve(skb
, reserve
);
317 static int netpoll_owner_active(struct net_device
*dev
)
319 struct napi_struct
*napi
;
321 list_for_each_entry(napi
, &dev
->napi_list
, dev_list
) {
322 if (napi
->poll_owner
== smp_processor_id())
328 /* call with IRQ disabled */
329 void netpoll_send_skb_on_dev(struct netpoll
*np
, struct sk_buff
*skb
,
330 struct net_device
*dev
)
332 int status
= NETDEV_TX_BUSY
;
334 /* It is up to the caller to keep npinfo alive. */
335 struct netpoll_info
*npinfo
;
337 lockdep_assert_irqs_disabled();
339 npinfo
= rcu_dereference_bh(np
->dev
->npinfo
);
340 if (!npinfo
|| !netif_running(dev
) || !netif_device_present(dev
)) {
341 dev_kfree_skb_irq(skb
);
345 /* don't get messages out of order, and no recursion */
346 if (skb_queue_len(&npinfo
->txq
) == 0 && !netpoll_owner_active(dev
)) {
347 struct netdev_queue
*txq
;
349 txq
= netdev_pick_tx(dev
, skb
, NULL
);
351 /* try until next clock tick */
352 for (tries
= jiffies_to_usecs(1)/USEC_PER_POLL
;
353 tries
> 0; --tries
) {
354 if (HARD_TX_TRYLOCK(dev
, txq
)) {
355 if (!netif_xmit_stopped(txq
))
356 status
= netpoll_start_xmit(skb
, dev
, txq
);
358 HARD_TX_UNLOCK(dev
, txq
);
360 if (status
== NETDEV_TX_OK
)
365 /* tickle device maybe there is some cleanup */
366 netpoll_poll_dev(np
->dev
);
368 udelay(USEC_PER_POLL
);
371 WARN_ONCE(!irqs_disabled(),
372 "netpoll_send_skb_on_dev(): %s enabled interrupts in poll (%pF)\n",
373 dev
->name
, dev
->netdev_ops
->ndo_start_xmit
);
377 if (status
!= NETDEV_TX_OK
) {
378 skb_queue_tail(&npinfo
->txq
, skb
);
379 schedule_delayed_work(&npinfo
->tx_work
,0);
382 EXPORT_SYMBOL(netpoll_send_skb_on_dev
);
384 void netpoll_send_udp(struct netpoll
*np
, const char *msg
, int len
)
386 int total_len
, ip_len
, udp_len
;
391 static atomic_t ip_ident
;
392 struct ipv6hdr
*ip6h
;
394 WARN_ON_ONCE(!irqs_disabled());
396 udp_len
= len
+ sizeof(*udph
);
398 ip_len
= udp_len
+ sizeof(*ip6h
);
400 ip_len
= udp_len
+ sizeof(*iph
);
402 total_len
= ip_len
+ LL_RESERVED_SPACE(np
->dev
);
404 skb
= find_skb(np
, total_len
+ np
->dev
->needed_tailroom
,
409 skb_copy_to_linear_data(skb
, msg
, len
);
412 skb_push(skb
, sizeof(*udph
));
413 skb_reset_transport_header(skb
);
415 udph
->source
= htons(np
->local_port
);
416 udph
->dest
= htons(np
->remote_port
);
417 udph
->len
= htons(udp_len
);
421 udph
->check
= csum_ipv6_magic(&np
->local_ip
.in6
,
423 udp_len
, IPPROTO_UDP
,
424 csum_partial(udph
, udp_len
, 0));
425 if (udph
->check
== 0)
426 udph
->check
= CSUM_MANGLED_0
;
428 skb_push(skb
, sizeof(*ip6h
));
429 skb_reset_network_header(skb
);
430 ip6h
= ipv6_hdr(skb
);
432 /* ip6h->version = 6; ip6h->priority = 0; */
433 put_unaligned(0x60, (unsigned char *)ip6h
);
434 ip6h
->flow_lbl
[0] = 0;
435 ip6h
->flow_lbl
[1] = 0;
436 ip6h
->flow_lbl
[2] = 0;
438 ip6h
->payload_len
= htons(sizeof(struct udphdr
) + len
);
439 ip6h
->nexthdr
= IPPROTO_UDP
;
440 ip6h
->hop_limit
= 32;
441 ip6h
->saddr
= np
->local_ip
.in6
;
442 ip6h
->daddr
= np
->remote_ip
.in6
;
444 eth
= skb_push(skb
, ETH_HLEN
);
445 skb_reset_mac_header(skb
);
446 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IPV6
);
449 udph
->check
= csum_tcpudp_magic(np
->local_ip
.ip
,
451 udp_len
, IPPROTO_UDP
,
452 csum_partial(udph
, udp_len
, 0));
453 if (udph
->check
== 0)
454 udph
->check
= CSUM_MANGLED_0
;
456 skb_push(skb
, sizeof(*iph
));
457 skb_reset_network_header(skb
);
460 /* iph->version = 4; iph->ihl = 5; */
461 put_unaligned(0x45, (unsigned char *)iph
);
463 put_unaligned(htons(ip_len
), &(iph
->tot_len
));
464 iph
->id
= htons(atomic_inc_return(&ip_ident
));
467 iph
->protocol
= IPPROTO_UDP
;
469 put_unaligned(np
->local_ip
.ip
, &(iph
->saddr
));
470 put_unaligned(np
->remote_ip
.ip
, &(iph
->daddr
));
471 iph
->check
= ip_fast_csum((unsigned char *)iph
, iph
->ihl
);
473 eth
= skb_push(skb
, ETH_HLEN
);
474 skb_reset_mac_header(skb
);
475 skb
->protocol
= eth
->h_proto
= htons(ETH_P_IP
);
478 ether_addr_copy(eth
->h_source
, np
->dev
->dev_addr
);
479 ether_addr_copy(eth
->h_dest
, np
->remote_mac
);
483 netpoll_send_skb(np
, skb
);
485 EXPORT_SYMBOL(netpoll_send_udp
);
487 void netpoll_print_options(struct netpoll
*np
)
489 np_info(np
, "local port %d\n", np
->local_port
);
491 np_info(np
, "local IPv6 address %pI6c\n", &np
->local_ip
.in6
);
493 np_info(np
, "local IPv4 address %pI4\n", &np
->local_ip
.ip
);
494 np_info(np
, "interface '%s'\n", np
->dev_name
);
495 np_info(np
, "remote port %d\n", np
->remote_port
);
497 np_info(np
, "remote IPv6 address %pI6c\n", &np
->remote_ip
.in6
);
499 np_info(np
, "remote IPv4 address %pI4\n", &np
->remote_ip
.ip
);
500 np_info(np
, "remote ethernet address %pM\n", np
->remote_mac
);
502 EXPORT_SYMBOL(netpoll_print_options
);
504 static int netpoll_parse_ip_addr(const char *str
, union inet_addr
*addr
)
508 if (!strchr(str
, ':') &&
509 in4_pton(str
, -1, (void *)addr
, -1, &end
) > 0) {
513 if (in6_pton(str
, -1, addr
->in6
.s6_addr
, -1, &end
) > 0) {
514 #if IS_ENABLED(CONFIG_IPV6)
524 int netpoll_parse_options(struct netpoll
*np
, char *opt
)
526 char *cur
=opt
, *delim
;
528 bool ipversion_set
= false;
531 if ((delim
= strchr(cur
, '@')) == NULL
)
534 if (kstrtou16(cur
, 10, &np
->local_port
))
541 ipversion_set
= true;
542 if ((delim
= strchr(cur
, '/')) == NULL
)
545 ipv6
= netpoll_parse_ip_addr(cur
, &np
->local_ip
);
549 np
->ipv6
= (bool)ipv6
;
555 /* parse out dev name */
556 if ((delim
= strchr(cur
, ',')) == NULL
)
559 strlcpy(np
->dev_name
, cur
, sizeof(np
->dev_name
));
566 if ((delim
= strchr(cur
, '@')) == NULL
)
569 if (*cur
== ' ' || *cur
== '\t')
570 np_info(np
, "warning: whitespace is not allowed\n");
571 if (kstrtou16(cur
, 10, &np
->remote_port
))
578 if ((delim
= strchr(cur
, '/')) == NULL
)
581 ipv6
= netpoll_parse_ip_addr(cur
, &np
->remote_ip
);
584 else if (ipversion_set
&& np
->ipv6
!= (bool)ipv6
)
587 np
->ipv6
= (bool)ipv6
;
592 if (!mac_pton(cur
, np
->remote_mac
))
596 netpoll_print_options(np
);
601 np_info(np
, "couldn't parse config at '%s'!\n", cur
);
604 EXPORT_SYMBOL(netpoll_parse_options
);
606 int __netpoll_setup(struct netpoll
*np
, struct net_device
*ndev
)
608 struct netpoll_info
*npinfo
;
609 const struct net_device_ops
*ops
;
613 strlcpy(np
->dev_name
, ndev
->name
, IFNAMSIZ
);
614 INIT_WORK(&np
->cleanup_work
, netpoll_async_cleanup
);
616 if ((ndev
->priv_flags
& IFF_DISABLE_NETPOLL
) ||
617 !ndev
->netdev_ops
->ndo_poll_controller
) {
618 np_err(np
, "%s doesn't support polling, aborting\n",
625 npinfo
= kmalloc(sizeof(*npinfo
), GFP_KERNEL
);
631 sema_init(&npinfo
->dev_lock
, 1);
632 skb_queue_head_init(&npinfo
->txq
);
633 INIT_DELAYED_WORK(&npinfo
->tx_work
, queue_process
);
635 refcount_set(&npinfo
->refcnt
, 1);
637 ops
= np
->dev
->netdev_ops
;
638 if (ops
->ndo_netpoll_setup
) {
639 err
= ops
->ndo_netpoll_setup(ndev
, npinfo
);
644 npinfo
= rtnl_dereference(ndev
->npinfo
);
645 refcount_inc(&npinfo
->refcnt
);
648 npinfo
->netpoll
= np
;
650 /* last thing to do is link it to the net device structure */
651 rcu_assign_pointer(ndev
->npinfo
, npinfo
);
660 EXPORT_SYMBOL_GPL(__netpoll_setup
);
662 int netpoll_setup(struct netpoll
*np
)
664 struct net_device
*ndev
= NULL
;
665 struct in_device
*in_dev
;
669 if (np
->dev_name
[0]) {
670 struct net
*net
= current
->nsproxy
->net_ns
;
671 ndev
= __dev_get_by_name(net
, np
->dev_name
);
674 np_err(np
, "%s doesn't exist, aborting\n", np
->dev_name
);
680 if (netdev_master_upper_dev_get(ndev
)) {
681 np_err(np
, "%s is a slave device, aborting\n", np
->dev_name
);
686 if (!netif_running(ndev
)) {
687 unsigned long atmost
, atleast
;
689 np_info(np
, "device %s not up yet, forcing it\n", np
->dev_name
);
691 err
= dev_open(ndev
);
694 np_err(np
, "failed to open %s\n", ndev
->name
);
699 atleast
= jiffies
+ HZ
/10;
700 atmost
= jiffies
+ carrier_timeout
* HZ
;
701 while (!netif_carrier_ok(ndev
)) {
702 if (time_after(jiffies
, atmost
)) {
703 np_notice(np
, "timeout waiting for carrier\n");
709 /* If carrier appears to come up instantly, we don't
710 * trust it and pause so that we don't pump all our
711 * queued console messages into the bitbucket.
714 if (time_before(jiffies
, atleast
)) {
715 np_notice(np
, "carrier detect appears untrustworthy, waiting 4 seconds\n");
721 if (!np
->local_ip
.ip
) {
723 in_dev
= __in_dev_get_rtnl(ndev
);
725 if (!in_dev
|| !in_dev
->ifa_list
) {
726 np_err(np
, "no IP address for %s, aborting\n",
732 np
->local_ip
.ip
= in_dev
->ifa_list
->ifa_local
;
733 np_info(np
, "local IP %pI4\n", &np
->local_ip
.ip
);
735 #if IS_ENABLED(CONFIG_IPV6)
736 struct inet6_dev
*idev
;
739 idev
= __in6_dev_get(ndev
);
741 struct inet6_ifaddr
*ifp
;
743 read_lock_bh(&idev
->lock
);
744 list_for_each_entry(ifp
, &idev
->addr_list
, if_list
) {
745 if (ipv6_addr_type(&ifp
->addr
) & IPV6_ADDR_LINKLOCAL
)
747 np
->local_ip
.in6
= ifp
->addr
;
751 read_unlock_bh(&idev
->lock
);
754 np_err(np
, "no IPv6 address for %s, aborting\n",
758 np_info(np
, "local IPv6 %pI6c\n", &np
->local_ip
.in6
);
760 np_err(np
, "IPv6 is not supported %s, aborting\n",
768 /* fill up the skb queue */
771 err
= __netpoll_setup(np
, ndev
);
784 EXPORT_SYMBOL(netpoll_setup
);
786 static int __init
netpoll_init(void)
788 skb_queue_head_init(&skb_pool
);
791 core_initcall(netpoll_init
);
793 static void rcu_cleanup_netpoll_info(struct rcu_head
*rcu_head
)
795 struct netpoll_info
*npinfo
=
796 container_of(rcu_head
, struct netpoll_info
, rcu
);
798 skb_queue_purge(&npinfo
->txq
);
800 /* we can't call cancel_delayed_work_sync here, as we are in softirq */
801 cancel_delayed_work(&npinfo
->tx_work
);
803 /* clean after last, unfinished work */
804 __skb_queue_purge(&npinfo
->txq
);
805 /* now cancel it again */
806 cancel_delayed_work(&npinfo
->tx_work
);
810 void __netpoll_cleanup(struct netpoll
*np
)
812 struct netpoll_info
*npinfo
;
814 /* rtnl_dereference would be preferable here but
815 * rcu_cleanup_netpoll path can put us in here safely without
816 * holding the rtnl, so plain rcu_dereference it is
818 npinfo
= rtnl_dereference(np
->dev
->npinfo
);
822 synchronize_srcu(&netpoll_srcu
);
824 if (refcount_dec_and_test(&npinfo
->refcnt
)) {
825 const struct net_device_ops
*ops
;
827 ops
= np
->dev
->netdev_ops
;
828 if (ops
->ndo_netpoll_cleanup
)
829 ops
->ndo_netpoll_cleanup(np
->dev
);
831 RCU_INIT_POINTER(np
->dev
->npinfo
, NULL
);
832 call_rcu_bh(&npinfo
->rcu
, rcu_cleanup_netpoll_info
);
834 RCU_INIT_POINTER(np
->dev
->npinfo
, NULL
);
836 EXPORT_SYMBOL_GPL(__netpoll_cleanup
);
838 static void netpoll_async_cleanup(struct work_struct
*work
)
840 struct netpoll
*np
= container_of(work
, struct netpoll
, cleanup_work
);
843 __netpoll_cleanup(np
);
848 void __netpoll_free_async(struct netpoll
*np
)
850 schedule_work(&np
->cleanup_work
);
852 EXPORT_SYMBOL_GPL(__netpoll_free_async
);
854 void netpoll_cleanup(struct netpoll
*np
)
859 __netpoll_cleanup(np
);
865 EXPORT_SYMBOL(netpoll_cleanup
);