1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75#include <linux/uaccess.h>
76#include <linux/bitops.h>
77#include <linux/capability.h>
78#include <linux/cpu.h>
79#include <linux/types.h>
80#include <linux/kernel.h>
81#include <linux/hash.h>
82#include <linux/slab.h>
83#include <linux/sched.h>
84#include <linux/sched/mm.h>
85#include <linux/mutex.h>
86#include <linux/string.h>
87#include <linux/mm.h>
88#include <linux/socket.h>
89#include <linux/sockios.h>
90#include <linux/errno.h>
91#include <linux/interrupt.h>
92#include <linux/if_ether.h>
93#include <linux/netdevice.h>
94#include <linux/etherdevice.h>
95#include <linux/ethtool.h>
96#include <linux/notifier.h>
97#include <linux/skbuff.h>
98#include <linux/bpf.h>
99#include <linux/bpf_trace.h>
100#include <net/net_namespace.h>
101#include <net/sock.h>
102#include <net/busy_poll.h>
103#include <linux/rtnetlink.h>
104#include <linux/stat.h>
105#include <net/dst.h>
106#include <net/dst_metadata.h>
107#include <net/pkt_sched.h>
108#include <net/pkt_cls.h>
109#include <net/checksum.h>
110#include <net/xfrm.h>
111#include <linux/highmem.h>
112#include <linux/init.h>
113#include <linux/module.h>
114#include <linux/netpoll.h>
115#include <linux/rcupdate.h>
116#include <linux/delay.h>
117#include <net/iw_handler.h>
118#include <asm/current.h>
119#include <linux/audit.h>
120#include <linux/dmaengine.h>
121#include <linux/err.h>
122#include <linux/ctype.h>
123#include <linux/if_arp.h>
124#include <linux/if_vlan.h>
125#include <linux/ip.h>
126#include <net/ip.h>
127#include <net/mpls.h>
128#include <linux/ipv6.h>
129#include <linux/in.h>
130#include <linux/jhash.h>
131#include <linux/random.h>
132#include <trace/events/napi.h>
133#include <trace/events/net.h>
134#include <trace/events/skb.h>
135#include <linux/pci.h>
136#include <linux/inetdevice.h>
137#include <linux/cpu_rmap.h>
138#include <linux/static_key.h>
139#include <linux/hashtable.h>
140#include <linux/vmalloc.h>
141#include <linux/if_macvlan.h>
142#include <linux/errqueue.h>
143#include <linux/hrtimer.h>
144#include <linux/netfilter_ingress.h>
145#include <linux/crash_dump.h>
146#include <linux/sctp.h>
147
148#include "net-sysfs.h"
149
150
151#define MAX_GRO_SKBS 8
152
153
154#define GRO_MAX_HEAD (MAX_HEADER + 128)
155
156static DEFINE_SPINLOCK(ptype_lock);
157static DEFINE_SPINLOCK(offload_lock);
158struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
159struct list_head ptype_all __read_mostly;
160static struct list_head offload_base __read_mostly;
161
162static int netif_rx_internal(struct sk_buff *skb);
163static int call_netdevice_notifiers_info(unsigned long val,
164 struct net_device *dev,
165 struct netdev_notifier_info *info);
166static struct napi_struct *napi_by_id(unsigned int napi_id);
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187DEFINE_RWLOCK(dev_base_lock);
188EXPORT_SYMBOL(dev_base_lock);
189
190
191static DEFINE_SPINLOCK(napi_hash_lock);
192
193static unsigned int napi_gen_id = NR_CPUS;
194static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
195
196static seqcount_t devnet_rename_seq;
197
198static inline void dev_base_seq_inc(struct net *net)
199{
200 while (++net->dev_base_seq == 0)
201 ;
202}
203
204static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
205{
206 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
207
208 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
209}
210
211static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
212{
213 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
214}
215
216static inline void rps_lock(struct softnet_data *sd)
217{
218#ifdef CONFIG_RPS
219 spin_lock(&sd->input_pkt_queue.lock);
220#endif
221}
222
223static inline void rps_unlock(struct softnet_data *sd)
224{
225#ifdef CONFIG_RPS
226 spin_unlock(&sd->input_pkt_queue.lock);
227#endif
228}
229
230
231static void list_netdevice(struct net_device *dev)
232{
233 struct net *net = dev_net(dev);
234
235 ASSERT_RTNL();
236
237 write_lock_bh(&dev_base_lock);
238 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
239 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
240 hlist_add_head_rcu(&dev->index_hlist,
241 dev_index_hash(net, dev->ifindex));
242 write_unlock_bh(&dev_base_lock);
243
244 dev_base_seq_inc(net);
245}
246
247
248
249
250static void unlist_netdevice(struct net_device *dev)
251{
252 ASSERT_RTNL();
253
254
255 write_lock_bh(&dev_base_lock);
256 list_del_rcu(&dev->dev_list);
257 hlist_del_rcu(&dev->name_hlist);
258 hlist_del_rcu(&dev->index_hlist);
259 write_unlock_bh(&dev_base_lock);
260
261 dev_base_seq_inc(dev_net(dev));
262}
263
264
265
266
267
268static RAW_NOTIFIER_HEAD(netdev_chain);
269
270
271
272
273
274
275DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
276EXPORT_PER_CPU_SYMBOL(softnet_data);
277
278#ifdef CONFIG_LOCKDEP
279
280
281
282
283static const unsigned short netdev_lock_type[] = {
284 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
285 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
286 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
287 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
288 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
289 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
290 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
291 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
292 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
293 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
294 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
295 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
296 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
297 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
298 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
299
300static const char *const netdev_lock_name[] = {
301 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
302 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
303 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
304 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
305 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
306 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
307 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
308 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
309 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
310 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
311 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
312 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
313 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
314 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
315 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
316
317static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
318static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
319
320static inline unsigned short netdev_lock_pos(unsigned short dev_type)
321{
322 int i;
323
324 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
325 if (netdev_lock_type[i] == dev_type)
326 return i;
327
328 return ARRAY_SIZE(netdev_lock_type) - 1;
329}
330
331static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
332 unsigned short dev_type)
333{
334 int i;
335
336 i = netdev_lock_pos(dev_type);
337 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
338 netdev_lock_name[i]);
339}
340
341static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
342{
343 int i;
344
345 i = netdev_lock_pos(dev->type);
346 lockdep_set_class_and_name(&dev->addr_list_lock,
347 &netdev_addr_lock_key[i],
348 netdev_lock_name[i]);
349}
350#else
351static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
352 unsigned short dev_type)
353{
354}
355static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356{
357}
358#endif
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383static inline struct list_head *ptype_head(const struct packet_type *pt)
384{
385 if (pt->type == htons(ETH_P_ALL))
386 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
387 else
388 return pt->dev ? &pt->dev->ptype_specific :
389 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
390}
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405void dev_add_pack(struct packet_type *pt)
406{
407 struct list_head *head = ptype_head(pt);
408
409 spin_lock(&ptype_lock);
410 list_add_rcu(&pt->list, head);
411 spin_unlock(&ptype_lock);
412}
413EXPORT_SYMBOL(dev_add_pack);
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428void __dev_remove_pack(struct packet_type *pt)
429{
430 struct list_head *head = ptype_head(pt);
431 struct packet_type *pt1;
432
433 spin_lock(&ptype_lock);
434
435 list_for_each_entry(pt1, head, list) {
436 if (pt == pt1) {
437 list_del_rcu(&pt->list);
438 goto out;
439 }
440 }
441
442 pr_warn("dev_remove_pack: %p not found\n", pt);
443out:
444 spin_unlock(&ptype_lock);
445}
446EXPORT_SYMBOL(__dev_remove_pack);
447
448
449
450
451
452
453
454
455
456
457
458
459
460void dev_remove_pack(struct packet_type *pt)
461{
462 __dev_remove_pack(pt);
463
464 synchronize_net();
465}
466EXPORT_SYMBOL(dev_remove_pack);
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481void dev_add_offload(struct packet_offload *po)
482{
483 struct packet_offload *elem;
484
485 spin_lock(&offload_lock);
486 list_for_each_entry(elem, &offload_base, list) {
487 if (po->priority < elem->priority)
488 break;
489 }
490 list_add_rcu(&po->list, elem->list.prev);
491 spin_unlock(&offload_lock);
492}
493EXPORT_SYMBOL(dev_add_offload);
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508static void __dev_remove_offload(struct packet_offload *po)
509{
510 struct list_head *head = &offload_base;
511 struct packet_offload *po1;
512
513 spin_lock(&offload_lock);
514
515 list_for_each_entry(po1, head, list) {
516 if (po == po1) {
517 list_del_rcu(&po->list);
518 goto out;
519 }
520 }
521
522 pr_warn("dev_remove_offload: %p not found\n", po);
523out:
524 spin_unlock(&offload_lock);
525}
526
527
528
529
530
531
532
533
534
535
536
537
538
539void dev_remove_offload(struct packet_offload *po)
540{
541 __dev_remove_offload(po);
542
543 synchronize_net();
544}
545EXPORT_SYMBOL(dev_remove_offload);
546
547
548
549
550
551
552
553
554static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
555
556
557
558
559
560
561
562
563
564
565static int netdev_boot_setup_add(char *name, struct ifmap *map)
566{
567 struct netdev_boot_setup *s;
568 int i;
569
570 s = dev_boot_setup;
571 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
572 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
573 memset(s[i].name, 0, sizeof(s[i].name));
574 strlcpy(s[i].name, name, IFNAMSIZ);
575 memcpy(&s[i].map, map, sizeof(s[i].map));
576 break;
577 }
578 }
579
580 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
581}
582
583
584
585
586
587
588
589
590
591
592int netdev_boot_setup_check(struct net_device *dev)
593{
594 struct netdev_boot_setup *s = dev_boot_setup;
595 int i;
596
597 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
598 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
599 !strcmp(dev->name, s[i].name)) {
600 dev->irq = s[i].map.irq;
601 dev->base_addr = s[i].map.base_addr;
602 dev->mem_start = s[i].map.mem_start;
603 dev->mem_end = s[i].map.mem_end;
604 return 1;
605 }
606 }
607 return 0;
608}
609EXPORT_SYMBOL(netdev_boot_setup_check);
610
611
612
613
614
615
616
617
618
619
620
621
622unsigned long netdev_boot_base(const char *prefix, int unit)
623{
624 const struct netdev_boot_setup *s = dev_boot_setup;
625 char name[IFNAMSIZ];
626 int i;
627
628 sprintf(name, "%s%d", prefix, unit);
629
630
631
632
633
634 if (__dev_get_by_name(&init_net, name))
635 return 1;
636
637 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
638 if (!strcmp(name, s[i].name))
639 return s[i].map.base_addr;
640 return 0;
641}
642
643
644
645
646int __init netdev_boot_setup(char *str)
647{
648 int ints[5];
649 struct ifmap map;
650
651 str = get_options(str, ARRAY_SIZE(ints), ints);
652 if (!str || !*str)
653 return 0;
654
655
656 memset(&map, 0, sizeof(map));
657 if (ints[0] > 0)
658 map.irq = ints[1];
659 if (ints[0] > 1)
660 map.base_addr = ints[2];
661 if (ints[0] > 2)
662 map.mem_start = ints[3];
663 if (ints[0] > 3)
664 map.mem_end = ints[4];
665
666
667 return netdev_boot_setup_add(str, &map);
668}
669
670__setup("netdev=", netdev_boot_setup);
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686int dev_get_iflink(const struct net_device *dev)
687{
688 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
689 return dev->netdev_ops->ndo_get_iflink(dev);
690
691 return dev->ifindex;
692}
693EXPORT_SYMBOL(dev_get_iflink);
694
695
696
697
698
699
700
701
702
703
704int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
705{
706 struct ip_tunnel_info *info;
707
708 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
709 return -EINVAL;
710
711 info = skb_tunnel_info_unclone(skb);
712 if (!info)
713 return -ENOMEM;
714 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
715 return -EINVAL;
716
717 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
718}
719EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
720
721
722
723
724
725
726
727
728
729
730
731
732
733struct net_device *__dev_get_by_name(struct net *net, const char *name)
734{
735 struct net_device *dev;
736 struct hlist_head *head = dev_name_hash(net, name);
737
738 hlist_for_each_entry(dev, head, name_hlist)
739 if (!strncmp(dev->name, name, IFNAMSIZ))
740 return dev;
741
742 return NULL;
743}
744EXPORT_SYMBOL(__dev_get_by_name);
745
746
747
748
749
750
751
752
753
754
755
756
757
758struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
759{
760 struct net_device *dev;
761 struct hlist_head *head = dev_name_hash(net, name);
762
763 hlist_for_each_entry_rcu(dev, head, name_hlist)
764 if (!strncmp(dev->name, name, IFNAMSIZ))
765 return dev;
766
767 return NULL;
768}
769EXPORT_SYMBOL(dev_get_by_name_rcu);
770
771
772
773
774
775
776
777
778
779
780
781
782
783struct net_device *dev_get_by_name(struct net *net, const char *name)
784{
785 struct net_device *dev;
786
787 rcu_read_lock();
788 dev = dev_get_by_name_rcu(net, name);
789 if (dev)
790 dev_hold(dev);
791 rcu_read_unlock();
792 return dev;
793}
794EXPORT_SYMBOL(dev_get_by_name);
795
796
797
798
799
800
801
802
803
804
805
806
807
808struct net_device *__dev_get_by_index(struct net *net, int ifindex)
809{
810 struct net_device *dev;
811 struct hlist_head *head = dev_index_hash(net, ifindex);
812
813 hlist_for_each_entry(dev, head, index_hlist)
814 if (dev->ifindex == ifindex)
815 return dev;
816
817 return NULL;
818}
819EXPORT_SYMBOL(__dev_get_by_index);
820
821
822
823
824
825
826
827
828
829
830
831
832struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
833{
834 struct net_device *dev;
835 struct hlist_head *head = dev_index_hash(net, ifindex);
836
837 hlist_for_each_entry_rcu(dev, head, index_hlist)
838 if (dev->ifindex == ifindex)
839 return dev;
840
841 return NULL;
842}
843EXPORT_SYMBOL(dev_get_by_index_rcu);
844
845
846
847
848
849
850
851
852
853
854
855
856
857struct net_device *dev_get_by_index(struct net *net, int ifindex)
858{
859 struct net_device *dev;
860
861 rcu_read_lock();
862 dev = dev_get_by_index_rcu(net, ifindex);
863 if (dev)
864 dev_hold(dev);
865 rcu_read_unlock();
866 return dev;
867}
868EXPORT_SYMBOL(dev_get_by_index);
869
870
871
872
873
874
875
876
877
878
879
880struct net_device *dev_get_by_napi_id(unsigned int napi_id)
881{
882 struct napi_struct *napi;
883
884 WARN_ON_ONCE(!rcu_read_lock_held());
885
886 if (napi_id < MIN_NAPI_ID)
887 return NULL;
888
889 napi = napi_by_id(napi_id);
890
891 return napi ? napi->dev : NULL;
892}
893EXPORT_SYMBOL(dev_get_by_napi_id);
894
895
896
897
898
899
900
901
902
903
904
905int netdev_get_name(struct net *net, char *name, int ifindex)
906{
907 struct net_device *dev;
908 unsigned int seq;
909
910retry:
911 seq = raw_seqcount_begin(&devnet_rename_seq);
912 rcu_read_lock();
913 dev = dev_get_by_index_rcu(net, ifindex);
914 if (!dev) {
915 rcu_read_unlock();
916 return -ENODEV;
917 }
918
919 strcpy(name, dev->name);
920 rcu_read_unlock();
921 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
922 cond_resched();
923 goto retry;
924 }
925
926 return 0;
927}
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
944 const char *ha)
945{
946 struct net_device *dev;
947
948 for_each_netdev_rcu(net, dev)
949 if (dev->type == type &&
950 !memcmp(dev->dev_addr, ha, dev->addr_len))
951 return dev;
952
953 return NULL;
954}
955EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
956
957struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
958{
959 struct net_device *dev;
960
961 ASSERT_RTNL();
962 for_each_netdev(net, dev)
963 if (dev->type == type)
964 return dev;
965
966 return NULL;
967}
968EXPORT_SYMBOL(__dev_getfirstbyhwtype);
969
970struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
971{
972 struct net_device *dev, *ret = NULL;
973
974 rcu_read_lock();
975 for_each_netdev_rcu(net, dev)
976 if (dev->type == type) {
977 dev_hold(dev);
978 ret = dev;
979 break;
980 }
981 rcu_read_unlock();
982 return ret;
983}
984EXPORT_SYMBOL(dev_getfirstbyhwtype);
985
986
987
988
989
990
991
992
993
994
995
996
997struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
998 unsigned short mask)
999{
1000 struct net_device *dev, *ret;
1001
1002 ASSERT_RTNL();
1003
1004 ret = NULL;
1005 for_each_netdev(net, dev) {
1006 if (((dev->flags ^ if_flags) & mask) == 0) {
1007 ret = dev;
1008 break;
1009 }
1010 }
1011 return ret;
1012}
1013EXPORT_SYMBOL(__dev_get_by_flags);
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023bool dev_valid_name(const char *name)
1024{
1025 if (*name == '\0')
1026 return false;
1027 if (strlen(name) >= IFNAMSIZ)
1028 return false;
1029 if (!strcmp(name, ".") || !strcmp(name, ".."))
1030 return false;
1031
1032 while (*name) {
1033 if (*name == '/' || *name == ':' || isspace(*name))
1034 return false;
1035 name++;
1036 }
1037 return true;
1038}
1039EXPORT_SYMBOL(dev_valid_name);
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1057{
1058 int i = 0;
1059 const char *p;
1060 const int max_netdevices = 8*PAGE_SIZE;
1061 unsigned long *inuse;
1062 struct net_device *d;
1063
1064 p = strnchr(name, IFNAMSIZ-1, '%');
1065 if (p) {
1066
1067
1068
1069
1070
1071 if (p[1] != 'd' || strchr(p + 2, '%'))
1072 return -EINVAL;
1073
1074
1075 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1076 if (!inuse)
1077 return -ENOMEM;
1078
1079 for_each_netdev(net, d) {
1080 if (!sscanf(d->name, name, &i))
1081 continue;
1082 if (i < 0 || i >= max_netdevices)
1083 continue;
1084
1085
1086 snprintf(buf, IFNAMSIZ, name, i);
1087 if (!strncmp(buf, d->name, IFNAMSIZ))
1088 set_bit(i, inuse);
1089 }
1090
1091 i = find_first_zero_bit(inuse, max_netdevices);
1092 free_page((unsigned long) inuse);
1093 }
1094
1095 if (buf != name)
1096 snprintf(buf, IFNAMSIZ, name, i);
1097 if (!__dev_get_by_name(net, buf))
1098 return i;
1099
1100
1101
1102
1103
1104 return -ENFILE;
1105}
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121int dev_alloc_name(struct net_device *dev, const char *name)
1122{
1123 char buf[IFNAMSIZ];
1124 struct net *net;
1125 int ret;
1126
1127 BUG_ON(!dev_net(dev));
1128 net = dev_net(dev);
1129 ret = __dev_alloc_name(net, name, buf);
1130 if (ret >= 0)
1131 strlcpy(dev->name, buf, IFNAMSIZ);
1132 return ret;
1133}
1134EXPORT_SYMBOL(dev_alloc_name);
1135
1136static int dev_alloc_name_ns(struct net *net,
1137 struct net_device *dev,
1138 const char *name)
1139{
1140 char buf[IFNAMSIZ];
1141 int ret;
1142
1143 ret = __dev_alloc_name(net, name, buf);
1144 if (ret >= 0)
1145 strlcpy(dev->name, buf, IFNAMSIZ);
1146 return ret;
1147}
1148
1149static int dev_get_valid_name(struct net *net,
1150 struct net_device *dev,
1151 const char *name)
1152{
1153 BUG_ON(!net);
1154
1155 if (!dev_valid_name(name))
1156 return -EINVAL;
1157
1158 if (strchr(name, '%'))
1159 return dev_alloc_name_ns(net, dev, name);
1160 else if (__dev_get_by_name(net, name))
1161 return -EEXIST;
1162 else if (dev->name != name)
1163 strlcpy(dev->name, name, IFNAMSIZ);
1164
1165 return 0;
1166}
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176int dev_change_name(struct net_device *dev, const char *newname)
1177{
1178 unsigned char old_assign_type;
1179 char oldname[IFNAMSIZ];
1180 int err = 0;
1181 int ret;
1182 struct net *net;
1183
1184 ASSERT_RTNL();
1185 BUG_ON(!dev_net(dev));
1186
1187 net = dev_net(dev);
1188 if (dev->flags & IFF_UP)
1189 return -EBUSY;
1190
1191 write_seqcount_begin(&devnet_rename_seq);
1192
1193 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1194 write_seqcount_end(&devnet_rename_seq);
1195 return 0;
1196 }
1197
1198 memcpy(oldname, dev->name, IFNAMSIZ);
1199
1200 err = dev_get_valid_name(net, dev, newname);
1201 if (err < 0) {
1202 write_seqcount_end(&devnet_rename_seq);
1203 return err;
1204 }
1205
1206 if (oldname[0] && !strchr(oldname, '%'))
1207 netdev_info(dev, "renamed from %s\n", oldname);
1208
1209 old_assign_type = dev->name_assign_type;
1210 dev->name_assign_type = NET_NAME_RENAMED;
1211
1212rollback:
1213 ret = device_rename(&dev->dev, dev->name);
1214 if (ret) {
1215 memcpy(dev->name, oldname, IFNAMSIZ);
1216 dev->name_assign_type = old_assign_type;
1217 write_seqcount_end(&devnet_rename_seq);
1218 return ret;
1219 }
1220
1221 write_seqcount_end(&devnet_rename_seq);
1222
1223 netdev_adjacent_rename_links(dev, oldname);
1224
1225 write_lock_bh(&dev_base_lock);
1226 hlist_del_rcu(&dev->name_hlist);
1227 write_unlock_bh(&dev_base_lock);
1228
1229 synchronize_rcu();
1230
1231 write_lock_bh(&dev_base_lock);
1232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1233 write_unlock_bh(&dev_base_lock);
1234
1235 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1236 ret = notifier_to_errno(ret);
1237
1238 if (ret) {
1239
1240 if (err >= 0) {
1241 err = ret;
1242 write_seqcount_begin(&devnet_rename_seq);
1243 memcpy(dev->name, oldname, IFNAMSIZ);
1244 memcpy(oldname, newname, IFNAMSIZ);
1245 dev->name_assign_type = old_assign_type;
1246 old_assign_type = NET_NAME_RENAMED;
1247 goto rollback;
1248 } else {
1249 pr_err("%s: name change rollback failed: %d\n",
1250 dev->name, ret);
1251 }
1252 }
1253
1254 return err;
1255}
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1266{
1267 char *new_ifalias;
1268
1269 ASSERT_RTNL();
1270
1271 if (len >= IFALIASZ)
1272 return -EINVAL;
1273
1274 if (!len) {
1275 kfree(dev->ifalias);
1276 dev->ifalias = NULL;
1277 return 0;
1278 }
1279
1280 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1281 if (!new_ifalias)
1282 return -ENOMEM;
1283 dev->ifalias = new_ifalias;
1284 memcpy(dev->ifalias, alias, len);
1285 dev->ifalias[len] = 0;
1286
1287 return len;
1288}
1289
1290
1291
1292
1293
1294
1295
1296
1297void netdev_features_change(struct net_device *dev)
1298{
1299 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1300}
1301EXPORT_SYMBOL(netdev_features_change);
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311void netdev_state_change(struct net_device *dev)
1312{
1313 if (dev->flags & IFF_UP) {
1314 struct netdev_notifier_change_info change_info;
1315
1316 change_info.flags_changed = 0;
1317 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1318 &change_info.info);
1319 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1320 }
1321}
1322EXPORT_SYMBOL(netdev_state_change);
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334void netdev_notify_peers(struct net_device *dev)
1335{
1336 rtnl_lock();
1337 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1338 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1339 rtnl_unlock();
1340}
1341EXPORT_SYMBOL(netdev_notify_peers);
1342
1343static int __dev_open(struct net_device *dev)
1344{
1345 const struct net_device_ops *ops = dev->netdev_ops;
1346 int ret;
1347
1348 ASSERT_RTNL();
1349
1350 if (!netif_device_present(dev))
1351 return -ENODEV;
1352
1353
1354
1355
1356
1357 netpoll_poll_disable(dev);
1358
1359 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1360 ret = notifier_to_errno(ret);
1361 if (ret)
1362 return ret;
1363
1364 set_bit(__LINK_STATE_START, &dev->state);
1365
1366 if (ops->ndo_validate_addr)
1367 ret = ops->ndo_validate_addr(dev);
1368
1369 if (!ret && ops->ndo_open)
1370 ret = ops->ndo_open(dev);
1371
1372 netpoll_poll_enable(dev);
1373
1374 if (ret)
1375 clear_bit(__LINK_STATE_START, &dev->state);
1376 else {
1377 dev->flags |= IFF_UP;
1378 dev_set_rx_mode(dev);
1379 dev_activate(dev);
1380 add_device_randomness(dev->dev_addr, dev->addr_len);
1381 }
1382
1383 return ret;
1384}
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398int dev_open(struct net_device *dev)
1399{
1400 int ret;
1401
1402 if (dev->flags & IFF_UP)
1403 return 0;
1404
1405 ret = __dev_open(dev);
1406 if (ret < 0)
1407 return ret;
1408
1409 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1410 call_netdevice_notifiers(NETDEV_UP, dev);
1411
1412 return ret;
1413}
1414EXPORT_SYMBOL(dev_open);
1415
1416static int __dev_close_many(struct list_head *head)
1417{
1418 struct net_device *dev;
1419
1420 ASSERT_RTNL();
1421 might_sleep();
1422
1423 list_for_each_entry(dev, head, close_list) {
1424
1425 netpoll_poll_disable(dev);
1426
1427 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1428
1429 clear_bit(__LINK_STATE_START, &dev->state);
1430
1431
1432
1433
1434
1435
1436
1437 smp_mb__after_atomic();
1438 }
1439
1440 dev_deactivate_many(head);
1441
1442 list_for_each_entry(dev, head, close_list) {
1443 const struct net_device_ops *ops = dev->netdev_ops;
1444
1445
1446
1447
1448
1449
1450
1451
1452 if (ops->ndo_stop)
1453 ops->ndo_stop(dev);
1454
1455 dev->flags &= ~IFF_UP;
1456 netpoll_poll_enable(dev);
1457 }
1458
1459 return 0;
1460}
1461
1462static int __dev_close(struct net_device *dev)
1463{
1464 int retval;
1465 LIST_HEAD(single);
1466
1467 list_add(&dev->close_list, &single);
1468 retval = __dev_close_many(&single);
1469 list_del(&single);
1470
1471 return retval;
1472}
1473
1474int dev_close_many(struct list_head *head, bool unlink)
1475{
1476 struct net_device *dev, *tmp;
1477
1478
1479 list_for_each_entry_safe(dev, tmp, head, close_list)
1480 if (!(dev->flags & IFF_UP))
1481 list_del_init(&dev->close_list);
1482
1483 __dev_close_many(head);
1484
1485 list_for_each_entry_safe(dev, tmp, head, close_list) {
1486 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1487 call_netdevice_notifiers(NETDEV_DOWN, dev);
1488 if (unlink)
1489 list_del_init(&dev->close_list);
1490 }
1491
1492 return 0;
1493}
1494EXPORT_SYMBOL(dev_close_many);
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505int dev_close(struct net_device *dev)
1506{
1507 if (dev->flags & IFF_UP) {
1508 LIST_HEAD(single);
1509
1510 list_add(&dev->close_list, &single);
1511 dev_close_many(&single, true);
1512 list_del(&single);
1513 }
1514 return 0;
1515}
1516EXPORT_SYMBOL(dev_close);
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527void dev_disable_lro(struct net_device *dev)
1528{
1529 struct net_device *lower_dev;
1530 struct list_head *iter;
1531
1532 dev->wanted_features &= ~NETIF_F_LRO;
1533 netdev_update_features(dev);
1534
1535 if (unlikely(dev->features & NETIF_F_LRO))
1536 netdev_WARN(dev, "failed to disable LRO!\n");
1537
1538 netdev_for_each_lower_dev(dev, lower_dev, iter)
1539 dev_disable_lro(lower_dev);
1540}
1541EXPORT_SYMBOL(dev_disable_lro);
1542
1543static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1544 struct net_device *dev)
1545{
1546 struct netdev_notifier_info info;
1547
1548 netdev_notifier_info_init(&info, dev);
1549 return nb->notifier_call(nb, val, &info);
1550}
1551
1552static int dev_boot_phase = 1;
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568int register_netdevice_notifier(struct notifier_block *nb)
1569{
1570 struct net_device *dev;
1571 struct net_device *last;
1572 struct net *net;
1573 int err;
1574
1575 rtnl_lock();
1576 err = raw_notifier_chain_register(&netdev_chain, nb);
1577 if (err)
1578 goto unlock;
1579 if (dev_boot_phase)
1580 goto unlock;
1581 for_each_net(net) {
1582 for_each_netdev(net, dev) {
1583 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1584 err = notifier_to_errno(err);
1585 if (err)
1586 goto rollback;
1587
1588 if (!(dev->flags & IFF_UP))
1589 continue;
1590
1591 call_netdevice_notifier(nb, NETDEV_UP, dev);
1592 }
1593 }
1594
1595unlock:
1596 rtnl_unlock();
1597 return err;
1598
1599rollback:
1600 last = dev;
1601 for_each_net(net) {
1602 for_each_netdev(net, dev) {
1603 if (dev == last)
1604 goto outroll;
1605
1606 if (dev->flags & IFF_UP) {
1607 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1608 dev);
1609 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1610 }
1611 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1612 }
1613 }
1614
1615outroll:
1616 raw_notifier_chain_unregister(&netdev_chain, nb);
1617 goto unlock;
1618}
1619EXPORT_SYMBOL(register_netdevice_notifier);
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635int unregister_netdevice_notifier(struct notifier_block *nb)
1636{
1637 struct net_device *dev;
1638 struct net *net;
1639 int err;
1640
1641 rtnl_lock();
1642 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1643 if (err)
1644 goto unlock;
1645
1646 for_each_net(net) {
1647 for_each_netdev(net, dev) {
1648 if (dev->flags & IFF_UP) {
1649 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1650 dev);
1651 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1652 }
1653 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1654 }
1655 }
1656unlock:
1657 rtnl_unlock();
1658 return err;
1659}
1660EXPORT_SYMBOL(unregister_netdevice_notifier);
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672static int call_netdevice_notifiers_info(unsigned long val,
1673 struct net_device *dev,
1674 struct netdev_notifier_info *info)
1675{
1676 ASSERT_RTNL();
1677 netdev_notifier_info_init(info, dev);
1678 return raw_notifier_call_chain(&netdev_chain, val, info);
1679}
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1691{
1692 struct netdev_notifier_info info;
1693
1694 return call_netdevice_notifiers_info(val, dev, &info);
1695}
1696EXPORT_SYMBOL(call_netdevice_notifiers);
1697
1698#ifdef CONFIG_NET_INGRESS
1699static struct static_key ingress_needed __read_mostly;
1700
1701void net_inc_ingress_queue(void)
1702{
1703 static_key_slow_inc(&ingress_needed);
1704}
1705EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1706
1707void net_dec_ingress_queue(void)
1708{
1709 static_key_slow_dec(&ingress_needed);
1710}
1711EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1712#endif
1713
1714#ifdef CONFIG_NET_EGRESS
1715static struct static_key egress_needed __read_mostly;
1716
1717void net_inc_egress_queue(void)
1718{
1719 static_key_slow_inc(&egress_needed);
1720}
1721EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1722
1723void net_dec_egress_queue(void)
1724{
1725 static_key_slow_dec(&egress_needed);
1726}
1727EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1728#endif
1729
1730static struct static_key netstamp_needed __read_mostly;
1731#ifdef HAVE_JUMP_LABEL
1732static atomic_t netstamp_needed_deferred;
1733static atomic_t netstamp_wanted;
1734static void netstamp_clear(struct work_struct *work)
1735{
1736 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1737 int wanted;
1738
1739 wanted = atomic_add_return(deferred, &netstamp_wanted);
1740 if (wanted > 0)
1741 static_key_enable(&netstamp_needed);
1742 else
1743 static_key_disable(&netstamp_needed);
1744}
1745static DECLARE_WORK(netstamp_work, netstamp_clear);
1746#endif
1747
1748void net_enable_timestamp(void)
1749{
1750#ifdef HAVE_JUMP_LABEL
1751 int wanted;
1752
1753 while (1) {
1754 wanted = atomic_read(&netstamp_wanted);
1755 if (wanted <= 0)
1756 break;
1757 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1758 return;
1759 }
1760 atomic_inc(&netstamp_needed_deferred);
1761 schedule_work(&netstamp_work);
1762#else
1763 static_key_slow_inc(&netstamp_needed);
1764#endif
1765}
1766EXPORT_SYMBOL(net_enable_timestamp);
1767
1768void net_disable_timestamp(void)
1769{
1770#ifdef HAVE_JUMP_LABEL
1771 int wanted;
1772
1773 while (1) {
1774 wanted = atomic_read(&netstamp_wanted);
1775 if (wanted <= 1)
1776 break;
1777 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1778 return;
1779 }
1780 atomic_dec(&netstamp_needed_deferred);
1781 schedule_work(&netstamp_work);
1782#else
1783 static_key_slow_dec(&netstamp_needed);
1784#endif
1785}
1786EXPORT_SYMBOL(net_disable_timestamp);
1787
1788static inline void net_timestamp_set(struct sk_buff *skb)
1789{
1790 skb->tstamp = 0;
1791 if (static_key_false(&netstamp_needed))
1792 __net_timestamp(skb);
1793}
1794
1795#define net_timestamp_check(COND, SKB) \
1796 if (static_key_false(&netstamp_needed)) { \
1797 if ((COND) && !(SKB)->tstamp) \
1798 __net_timestamp(SKB); \
1799 } \
1800
1801bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1802{
1803 unsigned int len;
1804
1805 if (!(dev->flags & IFF_UP))
1806 return false;
1807
1808 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1809 if (skb->len <= len)
1810 return true;
1811
1812
1813
1814
1815 if (skb_is_gso(skb))
1816 return true;
1817
1818 return false;
1819}
1820EXPORT_SYMBOL_GPL(is_skb_forwardable);
1821
1822int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1823{
1824 int ret = ____dev_forward_skb(dev, skb);
1825
1826 if (likely(!ret)) {
1827 skb->protocol = eth_type_trans(skb, dev);
1828 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1829 }
1830
1831 return ret;
1832}
1833EXPORT_SYMBOL_GPL(__dev_forward_skb);
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1854{
1855 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1856}
1857EXPORT_SYMBOL_GPL(dev_forward_skb);
1858
1859static inline int deliver_skb(struct sk_buff *skb,
1860 struct packet_type *pt_prev,
1861 struct net_device *orig_dev)
1862{
1863 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1864 return -ENOMEM;
1865 refcount_inc(&skb->users);
1866 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1867}
1868
1869static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1870 struct packet_type **pt,
1871 struct net_device *orig_dev,
1872 __be16 type,
1873 struct list_head *ptype_list)
1874{
1875 struct packet_type *ptype, *pt_prev = *pt;
1876
1877 list_for_each_entry_rcu(ptype, ptype_list, list) {
1878 if (ptype->type != type)
1879 continue;
1880 if (pt_prev)
1881 deliver_skb(skb, pt_prev, orig_dev);
1882 pt_prev = ptype;
1883 }
1884 *pt = pt_prev;
1885}
1886
1887static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1888{
1889 if (!ptype->af_packet_priv || !skb->sk)
1890 return false;
1891
1892 if (ptype->id_match)
1893 return ptype->id_match(ptype, skb->sk);
1894 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1895 return true;
1896
1897 return false;
1898}
1899
1900
1901
1902
1903
1904
1905void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1906{
1907 struct packet_type *ptype;
1908 struct sk_buff *skb2 = NULL;
1909 struct packet_type *pt_prev = NULL;
1910 struct list_head *ptype_list = &ptype_all;
1911
1912 rcu_read_lock();
1913again:
1914 list_for_each_entry_rcu(ptype, ptype_list, list) {
1915
1916
1917
1918 if (skb_loop_sk(ptype, skb))
1919 continue;
1920
1921 if (pt_prev) {
1922 deliver_skb(skb2, pt_prev, skb->dev);
1923 pt_prev = ptype;
1924 continue;
1925 }
1926
1927
1928 skb2 = skb_clone(skb, GFP_ATOMIC);
1929 if (!skb2)
1930 goto out_unlock;
1931
1932 net_timestamp_set(skb2);
1933
1934
1935
1936
1937
1938 skb_reset_mac_header(skb2);
1939
1940 if (skb_network_header(skb2) < skb2->data ||
1941 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1942 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1943 ntohs(skb2->protocol),
1944 dev->name);
1945 skb_reset_network_header(skb2);
1946 }
1947
1948 skb2->transport_header = skb2->network_header;
1949 skb2->pkt_type = PACKET_OUTGOING;
1950 pt_prev = ptype;
1951 }
1952
1953 if (ptype_list == &ptype_all) {
1954 ptype_list = &dev->ptype_all;
1955 goto again;
1956 }
1957out_unlock:
1958 if (pt_prev)
1959 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1960 rcu_read_unlock();
1961}
1962EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1978{
1979 int i;
1980 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1981
1982
1983 if (tc->offset + tc->count > txq) {
1984 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1985 dev->num_tc = 0;
1986 return;
1987 }
1988
1989
1990 for (i = 1; i < TC_BITMASK + 1; i++) {
1991 int q = netdev_get_prio_tc_map(dev, i);
1992
1993 tc = &dev->tc_to_txq[q];
1994 if (tc->offset + tc->count > txq) {
1995 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1996 i, q);
1997 netdev_set_prio_tc_map(dev, i, 0);
1998 }
1999 }
2000}
2001
2002int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2003{
2004 if (dev->num_tc) {
2005 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2006 int i;
2007
2008 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2009 if ((txq - tc->offset) < tc->count)
2010 return i;
2011 }
2012
2013 return -1;
2014 }
2015
2016 return 0;
2017}
2018
2019#ifdef CONFIG_XPS
2020static DEFINE_MUTEX(xps_map_mutex);
2021#define xmap_dereference(P) \
2022 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2023
2024static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2025 int tci, u16 index)
2026{
2027 struct xps_map *map = NULL;
2028 int pos;
2029
2030 if (dev_maps)
2031 map = xmap_dereference(dev_maps->cpu_map[tci]);
2032 if (!map)
2033 return false;
2034
2035 for (pos = map->len; pos--;) {
2036 if (map->queues[pos] != index)
2037 continue;
2038
2039 if (map->len > 1) {
2040 map->queues[pos] = map->queues[--map->len];
2041 break;
2042 }
2043
2044 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
2045 kfree_rcu(map, rcu);
2046 return false;
2047 }
2048
2049 return true;
2050}
2051
2052static bool remove_xps_queue_cpu(struct net_device *dev,
2053 struct xps_dev_maps *dev_maps,
2054 int cpu, u16 offset, u16 count)
2055{
2056 int num_tc = dev->num_tc ? : 1;
2057 bool active = false;
2058 int tci;
2059
2060 for (tci = cpu * num_tc; num_tc--; tci++) {
2061 int i, j;
2062
2063 for (i = count, j = offset; i--; j++) {
2064 if (!remove_xps_queue(dev_maps, cpu, j))
2065 break;
2066 }
2067
2068 active |= i < 0;
2069 }
2070
2071 return active;
2072}
2073
2074static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2075 u16 count)
2076{
2077 struct xps_dev_maps *dev_maps;
2078 int cpu, i;
2079 bool active = false;
2080
2081 mutex_lock(&xps_map_mutex);
2082 dev_maps = xmap_dereference(dev->xps_maps);
2083
2084 if (!dev_maps)
2085 goto out_no_maps;
2086
2087 for_each_possible_cpu(cpu)
2088 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2089 offset, count);
2090
2091 if (!active) {
2092 RCU_INIT_POINTER(dev->xps_maps, NULL);
2093 kfree_rcu(dev_maps, rcu);
2094 }
2095
2096 for (i = offset + (count - 1); count--; i--)
2097 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2098 NUMA_NO_NODE);
2099
2100out_no_maps:
2101 mutex_unlock(&xps_map_mutex);
2102}
2103
2104static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2105{
2106 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2107}
2108
2109static struct xps_map *expand_xps_map(struct xps_map *map,
2110 int cpu, u16 index)
2111{
2112 struct xps_map *new_map;
2113 int alloc_len = XPS_MIN_MAP_ALLOC;
2114 int i, pos;
2115
2116 for (pos = 0; map && pos < map->len; pos++) {
2117 if (map->queues[pos] != index)
2118 continue;
2119 return map;
2120 }
2121
2122
2123 if (map) {
2124 if (pos < map->alloc_len)
2125 return map;
2126
2127 alloc_len = map->alloc_len * 2;
2128 }
2129
2130
2131 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2132 cpu_to_node(cpu));
2133 if (!new_map)
2134 return NULL;
2135
2136 for (i = 0; i < pos; i++)
2137 new_map->queues[i] = map->queues[i];
2138 new_map->alloc_len = alloc_len;
2139 new_map->len = pos;
2140
2141 return new_map;
2142}
2143
2144int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2145 u16 index)
2146{
2147 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2148 int i, cpu, tci, numa_node_id = -2;
2149 int maps_sz, num_tc = 1, tc = 0;
2150 struct xps_map *map, *new_map;
2151 bool active = false;
2152
2153 if (dev->num_tc) {
2154 num_tc = dev->num_tc;
2155 tc = netdev_txq_to_tc(dev, index);
2156 if (tc < 0)
2157 return -EINVAL;
2158 }
2159
2160 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2161 if (maps_sz < L1_CACHE_BYTES)
2162 maps_sz = L1_CACHE_BYTES;
2163
2164 mutex_lock(&xps_map_mutex);
2165
2166 dev_maps = xmap_dereference(dev->xps_maps);
2167
2168
2169 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2170 if (!new_dev_maps)
2171 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2172 if (!new_dev_maps) {
2173 mutex_unlock(&xps_map_mutex);
2174 return -ENOMEM;
2175 }
2176
2177 tci = cpu * num_tc + tc;
2178 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2179 NULL;
2180
2181 map = expand_xps_map(map, cpu, index);
2182 if (!map)
2183 goto error;
2184
2185 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2186 }
2187
2188 if (!new_dev_maps)
2189 goto out_no_new_maps;
2190
2191 for_each_possible_cpu(cpu) {
2192
2193 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2194
2195 map = xmap_dereference(dev_maps->cpu_map[tci]);
2196 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2197 }
2198
2199
2200
2201
2202 tci = cpu * num_tc + tc;
2203
2204 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2205
2206 int pos = 0;
2207
2208 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2209 while ((pos < map->len) && (map->queues[pos] != index))
2210 pos++;
2211
2212 if (pos == map->len)
2213 map->queues[map->len++] = index;
2214#ifdef CONFIG_NUMA
2215 if (numa_node_id == -2)
2216 numa_node_id = cpu_to_node(cpu);
2217 else if (numa_node_id != cpu_to_node(cpu))
2218 numa_node_id = -1;
2219#endif
2220 } else if (dev_maps) {
2221
2222 map = xmap_dereference(dev_maps->cpu_map[tci]);
2223 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2224 }
2225
2226
2227 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2228
2229 map = xmap_dereference(dev_maps->cpu_map[tci]);
2230 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2231 }
2232 }
2233
2234 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2235
2236
2237 if (!dev_maps)
2238 goto out_no_old_maps;
2239
2240 for_each_possible_cpu(cpu) {
2241 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2242 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2243 map = xmap_dereference(dev_maps->cpu_map[tci]);
2244 if (map && map != new_map)
2245 kfree_rcu(map, rcu);
2246 }
2247 }
2248
2249 kfree_rcu(dev_maps, rcu);
2250
2251out_no_old_maps:
2252 dev_maps = new_dev_maps;
2253 active = true;
2254
2255out_no_new_maps:
2256
2257 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2258 (numa_node_id >= 0) ? numa_node_id :
2259 NUMA_NO_NODE);
2260
2261 if (!dev_maps)
2262 goto out_no_maps;
2263
2264
2265 for_each_possible_cpu(cpu) {
2266 for (i = tc, tci = cpu * num_tc; i--; tci++)
2267 active |= remove_xps_queue(dev_maps, tci, index);
2268 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2269 active |= remove_xps_queue(dev_maps, tci, index);
2270 for (i = num_tc - tc, tci++; --i; tci++)
2271 active |= remove_xps_queue(dev_maps, tci, index);
2272 }
2273
2274
2275 if (!active) {
2276 RCU_INIT_POINTER(dev->xps_maps, NULL);
2277 kfree_rcu(dev_maps, rcu);
2278 }
2279
2280out_no_maps:
2281 mutex_unlock(&xps_map_mutex);
2282
2283 return 0;
2284error:
2285
2286 for_each_possible_cpu(cpu) {
2287 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2288 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2289 map = dev_maps ?
2290 xmap_dereference(dev_maps->cpu_map[tci]) :
2291 NULL;
2292 if (new_map && new_map != map)
2293 kfree(new_map);
2294 }
2295 }
2296
2297 mutex_unlock(&xps_map_mutex);
2298
2299 kfree(new_dev_maps);
2300 return -ENOMEM;
2301}
2302EXPORT_SYMBOL(netif_set_xps_queue);
2303
2304#endif
2305void netdev_reset_tc(struct net_device *dev)
2306{
2307#ifdef CONFIG_XPS
2308 netif_reset_xps_queues_gt(dev, 0);
2309#endif
2310 dev->num_tc = 0;
2311 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2312 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2313}
2314EXPORT_SYMBOL(netdev_reset_tc);
2315
2316int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2317{
2318 if (tc >= dev->num_tc)
2319 return -EINVAL;
2320
2321#ifdef CONFIG_XPS
2322 netif_reset_xps_queues(dev, offset, count);
2323#endif
2324 dev->tc_to_txq[tc].count = count;
2325 dev->tc_to_txq[tc].offset = offset;
2326 return 0;
2327}
2328EXPORT_SYMBOL(netdev_set_tc_queue);
2329
2330int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2331{
2332 if (num_tc > TC_MAX_QUEUE)
2333 return -EINVAL;
2334
2335#ifdef CONFIG_XPS
2336 netif_reset_xps_queues_gt(dev, 0);
2337#endif
2338 dev->num_tc = num_tc;
2339 return 0;
2340}
2341EXPORT_SYMBOL(netdev_set_num_tc);
2342
2343
2344
2345
2346
2347int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2348{
2349 int rc;
2350
2351 if (txq < 1 || txq > dev->num_tx_queues)
2352 return -EINVAL;
2353
2354 if (dev->reg_state == NETREG_REGISTERED ||
2355 dev->reg_state == NETREG_UNREGISTERING) {
2356 ASSERT_RTNL();
2357
2358 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2359 txq);
2360 if (rc)
2361 return rc;
2362
2363 if (dev->num_tc)
2364 netif_setup_tc(dev, txq);
2365
2366 if (txq < dev->real_num_tx_queues) {
2367 qdisc_reset_all_tx_gt(dev, txq);
2368#ifdef CONFIG_XPS
2369 netif_reset_xps_queues_gt(dev, txq);
2370#endif
2371 }
2372 }
2373
2374 dev->real_num_tx_queues = txq;
2375 return 0;
2376}
2377EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2378
2379#ifdef CONFIG_SYSFS
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2391{
2392 int rc;
2393
2394 if (rxq < 1 || rxq > dev->num_rx_queues)
2395 return -EINVAL;
2396
2397 if (dev->reg_state == NETREG_REGISTERED) {
2398 ASSERT_RTNL();
2399
2400 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2401 rxq);
2402 if (rc)
2403 return rc;
2404 }
2405
2406 dev->real_num_rx_queues = rxq;
2407 return 0;
2408}
2409EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2410#endif
2411
2412
2413
2414
2415
2416
2417
2418int netif_get_num_default_rss_queues(void)
2419{
2420 return is_kdump_kernel() ?
2421 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2422}
2423EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2424
2425static void __netif_reschedule(struct Qdisc *q)
2426{
2427 struct softnet_data *sd;
2428 unsigned long flags;
2429
2430 local_irq_save(flags);
2431 sd = this_cpu_ptr(&softnet_data);
2432 q->next_sched = NULL;
2433 *sd->output_queue_tailp = q;
2434 sd->output_queue_tailp = &q->next_sched;
2435 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2436 local_irq_restore(flags);
2437}
2438
2439void __netif_schedule(struct Qdisc *q)
2440{
2441 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2442 __netif_reschedule(q);
2443}
2444EXPORT_SYMBOL(__netif_schedule);
2445
2446struct dev_kfree_skb_cb {
2447 enum skb_free_reason reason;
2448};
2449
2450static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2451{
2452 return (struct dev_kfree_skb_cb *)skb->cb;
2453}
2454
2455void netif_schedule_queue(struct netdev_queue *txq)
2456{
2457 rcu_read_lock();
2458 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2459 struct Qdisc *q = rcu_dereference(txq->qdisc);
2460
2461 __netif_schedule(q);
2462 }
2463 rcu_read_unlock();
2464}
2465EXPORT_SYMBOL(netif_schedule_queue);
2466
2467void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2468{
2469 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2470 struct Qdisc *q;
2471
2472 rcu_read_lock();
2473 q = rcu_dereference(dev_queue->qdisc);
2474 __netif_schedule(q);
2475 rcu_read_unlock();
2476 }
2477}
2478EXPORT_SYMBOL(netif_tx_wake_queue);
2479
2480void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2481{
2482 unsigned long flags;
2483
2484 if (unlikely(!skb))
2485 return;
2486
2487 if (likely(refcount_read(&skb->users) == 1)) {
2488 smp_rmb();
2489 refcount_set(&skb->users, 0);
2490 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2491 return;
2492 }
2493 get_kfree_skb_cb(skb)->reason = reason;
2494 local_irq_save(flags);
2495 skb->next = __this_cpu_read(softnet_data.completion_queue);
2496 __this_cpu_write(softnet_data.completion_queue, skb);
2497 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2498 local_irq_restore(flags);
2499}
2500EXPORT_SYMBOL(__dev_kfree_skb_irq);
2501
2502void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2503{
2504 if (in_irq() || irqs_disabled())
2505 __dev_kfree_skb_irq(skb, reason);
2506 else
2507 dev_kfree_skb(skb);
2508}
2509EXPORT_SYMBOL(__dev_kfree_skb_any);
2510
2511
2512
2513
2514
2515
2516
2517
2518void netif_device_detach(struct net_device *dev)
2519{
2520 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2521 netif_running(dev)) {
2522 netif_tx_stop_all_queues(dev);
2523 }
2524}
2525EXPORT_SYMBOL(netif_device_detach);
2526
2527
2528
2529
2530
2531
2532
2533void netif_device_attach(struct net_device *dev)
2534{
2535 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2536 netif_running(dev)) {
2537 netif_tx_wake_all_queues(dev);
2538 __netdev_watchdog_up(dev);
2539 }
2540}
2541EXPORT_SYMBOL(netif_device_attach);
2542
2543
2544
2545
2546
2547u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2548 unsigned int num_tx_queues)
2549{
2550 u32 hash;
2551 u16 qoffset = 0;
2552 u16 qcount = num_tx_queues;
2553
2554 if (skb_rx_queue_recorded(skb)) {
2555 hash = skb_get_rx_queue(skb);
2556 while (unlikely(hash >= num_tx_queues))
2557 hash -= num_tx_queues;
2558 return hash;
2559 }
2560
2561 if (dev->num_tc) {
2562 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2563
2564 qoffset = dev->tc_to_txq[tc].offset;
2565 qcount = dev->tc_to_txq[tc].count;
2566 }
2567
2568 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2569}
2570EXPORT_SYMBOL(__skb_tx_hash);
2571
2572static void skb_warn_bad_offload(const struct sk_buff *skb)
2573{
2574 static const netdev_features_t null_features;
2575 struct net_device *dev = skb->dev;
2576 const char *name = "";
2577
2578 if (!net_ratelimit())
2579 return;
2580
2581 if (dev) {
2582 if (dev->dev.parent)
2583 name = dev_driver_string(dev->dev.parent);
2584 else
2585 name = netdev_name(dev);
2586 }
2587 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2588 "gso_type=%d ip_summed=%d\n",
2589 name, dev ? &dev->features : &null_features,
2590 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2591 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2592 skb_shinfo(skb)->gso_type, skb->ip_summed);
2593}
2594
2595
2596
2597
2598
2599int skb_checksum_help(struct sk_buff *skb)
2600{
2601 __wsum csum;
2602 int ret = 0, offset;
2603
2604 if (skb->ip_summed == CHECKSUM_COMPLETE)
2605 goto out_set_summed;
2606
2607 if (unlikely(skb_shinfo(skb)->gso_size)) {
2608 skb_warn_bad_offload(skb);
2609 return -EINVAL;
2610 }
2611
2612
2613
2614
2615 if (skb_has_shared_frag(skb)) {
2616 ret = __skb_linearize(skb);
2617 if (ret)
2618 goto out;
2619 }
2620
2621 offset = skb_checksum_start_offset(skb);
2622 BUG_ON(offset >= skb_headlen(skb));
2623 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2624
2625 offset += skb->csum_offset;
2626 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2627
2628 if (skb_cloned(skb) &&
2629 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2630 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2631 if (ret)
2632 goto out;
2633 }
2634
2635 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2636out_set_summed:
2637 skb->ip_summed = CHECKSUM_NONE;
2638out:
2639 return ret;
2640}
2641EXPORT_SYMBOL(skb_checksum_help);
2642
2643int skb_crc32c_csum_help(struct sk_buff *skb)
2644{
2645 __le32 crc32c_csum;
2646 int ret = 0, offset, start;
2647
2648 if (skb->ip_summed != CHECKSUM_PARTIAL)
2649 goto out;
2650
2651 if (unlikely(skb_is_gso(skb)))
2652 goto out;
2653
2654
2655
2656
2657 if (unlikely(skb_has_shared_frag(skb))) {
2658 ret = __skb_linearize(skb);
2659 if (ret)
2660 goto out;
2661 }
2662 start = skb_checksum_start_offset(skb);
2663 offset = start + offsetof(struct sctphdr, checksum);
2664 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2665 ret = -EINVAL;
2666 goto out;
2667 }
2668 if (skb_cloned(skb) &&
2669 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2670 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2671 if (ret)
2672 goto out;
2673 }
2674 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2675 skb->len - start, ~(__u32)0,
2676 crc32c_csum_stub));
2677 *(__le32 *)(skb->data + offset) = crc32c_csum;
2678 skb->ip_summed = CHECKSUM_NONE;
2679 skb->csum_not_inet = 0;
2680out:
2681 return ret;
2682}
2683
2684__be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2685{
2686 __be16 type = skb->protocol;
2687
2688
2689 if (type == htons(ETH_P_TEB)) {
2690 struct ethhdr *eth;
2691
2692 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2693 return 0;
2694
2695 eth = (struct ethhdr *)skb_mac_header(skb);
2696 type = eth->h_proto;
2697 }
2698
2699 return __vlan_get_protocol(skb, type, depth);
2700}
2701
2702
2703
2704
2705
2706
2707struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2708 netdev_features_t features)
2709{
2710 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2711 struct packet_offload *ptype;
2712 int vlan_depth = skb->mac_len;
2713 __be16 type = skb_network_protocol(skb, &vlan_depth);
2714
2715 if (unlikely(!type))
2716 return ERR_PTR(-EINVAL);
2717
2718 __skb_pull(skb, vlan_depth);
2719
2720 rcu_read_lock();
2721 list_for_each_entry_rcu(ptype, &offload_base, list) {
2722 if (ptype->type == type && ptype->callbacks.gso_segment) {
2723 segs = ptype->callbacks.gso_segment(skb, features);
2724 break;
2725 }
2726 }
2727 rcu_read_unlock();
2728
2729 __skb_push(skb, skb->data - skb_mac_header(skb));
2730
2731 return segs;
2732}
2733EXPORT_SYMBOL(skb_mac_gso_segment);
2734
2735
2736
2737
2738static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2739{
2740 if (tx_path)
2741 return skb->ip_summed != CHECKSUM_PARTIAL &&
2742 skb->ip_summed != CHECKSUM_UNNECESSARY;
2743
2744 return skb->ip_summed == CHECKSUM_NONE;
2745}
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2761 netdev_features_t features, bool tx_path)
2762{
2763 struct sk_buff *segs;
2764
2765 if (unlikely(skb_needs_check(skb, tx_path))) {
2766 int err;
2767
2768
2769 err = skb_cow_head(skb, 0);
2770 if (err < 0)
2771 return ERR_PTR(err);
2772 }
2773
2774
2775
2776
2777
2778 if (features & NETIF_F_GSO_PARTIAL) {
2779 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2780 struct net_device *dev = skb->dev;
2781
2782 partial_features |= dev->features & dev->gso_partial_features;
2783 if (!skb_gso_ok(skb, features | partial_features))
2784 features &= ~NETIF_F_GSO_PARTIAL;
2785 }
2786
2787 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2788 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2789
2790 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2791 SKB_GSO_CB(skb)->encap_level = 0;
2792
2793 skb_reset_mac_header(skb);
2794 skb_reset_mac_len(skb);
2795
2796 segs = skb_mac_gso_segment(skb, features);
2797
2798 if (unlikely(skb_needs_check(skb, tx_path)))
2799 skb_warn_bad_offload(skb);
2800
2801 return segs;
2802}
2803EXPORT_SYMBOL(__skb_gso_segment);
2804
2805
2806#ifdef CONFIG_BUG
2807void netdev_rx_csum_fault(struct net_device *dev)
2808{
2809 if (net_ratelimit()) {
2810 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2811 dump_stack();
2812 }
2813}
2814EXPORT_SYMBOL(netdev_rx_csum_fault);
2815#endif
2816
2817
2818
2819
2820
2821
2822static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2823{
2824#ifdef CONFIG_HIGHMEM
2825 int i;
2826
2827 if (!(dev->features & NETIF_F_HIGHDMA)) {
2828 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2829 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2830
2831 if (PageHighMem(skb_frag_page(frag)))
2832 return 1;
2833 }
2834 }
2835
2836 if (PCI_DMA_BUS_IS_PHYS) {
2837 struct device *pdev = dev->dev.parent;
2838
2839 if (!pdev)
2840 return 0;
2841 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2842 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2843 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2844
2845 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2846 return 1;
2847 }
2848 }
2849#endif
2850 return 0;
2851}
2852
2853
2854
2855
2856#if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2857static netdev_features_t net_mpls_features(struct sk_buff *skb,
2858 netdev_features_t features,
2859 __be16 type)
2860{
2861 if (eth_p_mpls(type))
2862 features &= skb->dev->mpls_features;
2863
2864 return features;
2865}
2866#else
2867static netdev_features_t net_mpls_features(struct sk_buff *skb,
2868 netdev_features_t features,
2869 __be16 type)
2870{
2871 return features;
2872}
2873#endif
2874
2875static netdev_features_t harmonize_features(struct sk_buff *skb,
2876 netdev_features_t features)
2877{
2878 int tmp;
2879 __be16 type;
2880
2881 type = skb_network_protocol(skb, &tmp);
2882 features = net_mpls_features(skb, features, type);
2883
2884 if (skb->ip_summed != CHECKSUM_NONE &&
2885 !can_checksum_protocol(features, type)) {
2886 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2887 }
2888 if (illegal_highdma(skb->dev, skb))
2889 features &= ~NETIF_F_SG;
2890
2891 return features;
2892}
2893
2894netdev_features_t passthru_features_check(struct sk_buff *skb,
2895 struct net_device *dev,
2896 netdev_features_t features)
2897{
2898 return features;
2899}
2900EXPORT_SYMBOL(passthru_features_check);
2901
2902static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2903 struct net_device *dev,
2904 netdev_features_t features)
2905{
2906 return vlan_features_check(skb, features);
2907}
2908
2909static netdev_features_t gso_features_check(const struct sk_buff *skb,
2910 struct net_device *dev,
2911 netdev_features_t features)
2912{
2913 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2914
2915 if (gso_segs > dev->gso_max_segs)
2916 return features & ~NETIF_F_GSO_MASK;
2917
2918
2919
2920
2921
2922
2923
2924 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2925 features &= ~dev->gso_partial_features;
2926
2927
2928
2929
2930 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2931 struct iphdr *iph = skb->encapsulation ?
2932 inner_ip_hdr(skb) : ip_hdr(skb);
2933
2934 if (!(iph->frag_off & htons(IP_DF)))
2935 features &= ~NETIF_F_TSO_MANGLEID;
2936 }
2937
2938 return features;
2939}
2940
2941netdev_features_t netif_skb_features(struct sk_buff *skb)
2942{
2943 struct net_device *dev = skb->dev;
2944 netdev_features_t features = dev->features;
2945
2946 if (skb_is_gso(skb))
2947 features = gso_features_check(skb, dev, features);
2948
2949
2950
2951
2952
2953 if (skb->encapsulation)
2954 features &= dev->hw_enc_features;
2955
2956 if (skb_vlan_tagged(skb))
2957 features = netdev_intersect_features(features,
2958 dev->vlan_features |
2959 NETIF_F_HW_VLAN_CTAG_TX |
2960 NETIF_F_HW_VLAN_STAG_TX);
2961
2962 if (dev->netdev_ops->ndo_features_check)
2963 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2964 features);
2965 else
2966 features &= dflt_features_check(skb, dev, features);
2967
2968 return harmonize_features(skb, features);
2969}
2970EXPORT_SYMBOL(netif_skb_features);
2971
2972static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2973 struct netdev_queue *txq, bool more)
2974{
2975 unsigned int len;
2976 int rc;
2977
2978 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2979 dev_queue_xmit_nit(skb, dev);
2980
2981 len = skb->len;
2982 trace_net_dev_start_xmit(skb, dev);
2983 rc = netdev_start_xmit(skb, dev, txq, more);
2984 trace_net_dev_xmit(skb, rc, dev, len);
2985
2986 return rc;
2987}
2988
2989struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2990 struct netdev_queue *txq, int *ret)
2991{
2992 struct sk_buff *skb = first;
2993 int rc = NETDEV_TX_OK;
2994
2995 while (skb) {
2996 struct sk_buff *next = skb->next;
2997
2998 skb->next = NULL;
2999 rc = xmit_one(skb, dev, txq, next != NULL);
3000 if (unlikely(!dev_xmit_complete(rc))) {
3001 skb->next = next;
3002 goto out;
3003 }
3004
3005 skb = next;
3006 if (netif_xmit_stopped(txq) && skb) {
3007 rc = NETDEV_TX_BUSY;
3008 break;
3009 }
3010 }
3011
3012out:
3013 *ret = rc;
3014 return skb;
3015}
3016
3017static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3018 netdev_features_t features)
3019{
3020 if (skb_vlan_tag_present(skb) &&
3021 !vlan_hw_offload_capable(features, skb->vlan_proto))
3022 skb = __vlan_hwaccel_push_inside(skb);
3023 return skb;
3024}
3025
3026int skb_csum_hwoffload_help(struct sk_buff *skb,
3027 const netdev_features_t features)
3028{
3029 if (unlikely(skb->csum_not_inet))
3030 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3031 skb_crc32c_csum_help(skb);
3032
3033 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3034}
3035EXPORT_SYMBOL(skb_csum_hwoffload_help);
3036
3037static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
3038{
3039 netdev_features_t features;
3040
3041 features = netif_skb_features(skb);
3042 skb = validate_xmit_vlan(skb, features);
3043 if (unlikely(!skb))
3044 goto out_null;
3045
3046 if (netif_needs_gso(skb, features)) {
3047 struct sk_buff *segs;
3048
3049 segs = skb_gso_segment(skb, features);
3050 if (IS_ERR(segs)) {
3051 goto out_kfree_skb;
3052 } else if (segs) {
3053 consume_skb(skb);
3054 skb = segs;
3055 }
3056 } else {
3057 if (skb_needs_linearize(skb, features) &&
3058 __skb_linearize(skb))
3059 goto out_kfree_skb;
3060
3061 if (validate_xmit_xfrm(skb, features))
3062 goto out_kfree_skb;
3063
3064
3065
3066
3067
3068 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3069 if (skb->encapsulation)
3070 skb_set_inner_transport_header(skb,
3071 skb_checksum_start_offset(skb));
3072 else
3073 skb_set_transport_header(skb,
3074 skb_checksum_start_offset(skb));
3075 if (skb_csum_hwoffload_help(skb, features))
3076 goto out_kfree_skb;
3077 }
3078 }
3079
3080 return skb;
3081
3082out_kfree_skb:
3083 kfree_skb(skb);
3084out_null:
3085 atomic_long_inc(&dev->tx_dropped);
3086 return NULL;
3087}
3088
3089struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3090{
3091 struct sk_buff *next, *head = NULL, *tail;
3092
3093 for (; skb != NULL; skb = next) {
3094 next = skb->next;
3095 skb->next = NULL;
3096
3097
3098 skb->prev = skb;
3099
3100 skb = validate_xmit_skb(skb, dev);
3101 if (!skb)
3102 continue;
3103
3104 if (!head)
3105 head = skb;
3106 else
3107 tail->next = skb;
3108
3109
3110
3111 tail = skb->prev;
3112 }
3113 return head;
3114}
3115EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3116
3117static void qdisc_pkt_len_init(struct sk_buff *skb)
3118{
3119 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3120
3121 qdisc_skb_cb(skb)->pkt_len = skb->len;
3122
3123
3124
3125
3126 if (shinfo->gso_size) {
3127 unsigned int hdr_len;
3128 u16 gso_segs = shinfo->gso_segs;
3129
3130
3131 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3132
3133
3134 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3135 hdr_len += tcp_hdrlen(skb);
3136 else
3137 hdr_len += sizeof(struct udphdr);
3138
3139 if (shinfo->gso_type & SKB_GSO_DODGY)
3140 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3141 shinfo->gso_size);
3142
3143 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3144 }
3145}
3146
3147static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3148 struct net_device *dev,
3149 struct netdev_queue *txq)
3150{
3151 spinlock_t *root_lock = qdisc_lock(q);
3152 struct sk_buff *to_free = NULL;
3153 bool contended;
3154 int rc;
3155
3156 qdisc_calculate_pkt_len(skb, q);
3157
3158
3159
3160
3161
3162
3163 contended = qdisc_is_running(q);
3164 if (unlikely(contended))
3165 spin_lock(&q->busylock);
3166
3167 spin_lock(root_lock);
3168 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3169 __qdisc_drop(skb, &to_free);
3170 rc = NET_XMIT_DROP;
3171 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3172 qdisc_run_begin(q)) {
3173
3174
3175
3176
3177
3178
3179 qdisc_bstats_update(q, skb);
3180
3181 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3182 if (unlikely(contended)) {
3183 spin_unlock(&q->busylock);
3184 contended = false;
3185 }
3186 __qdisc_run(q);
3187 } else
3188 qdisc_run_end(q);
3189
3190 rc = NET_XMIT_SUCCESS;
3191 } else {
3192 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3193 if (qdisc_run_begin(q)) {
3194 if (unlikely(contended)) {
3195 spin_unlock(&q->busylock);
3196 contended = false;
3197 }
3198 __qdisc_run(q);
3199 }
3200 }
3201 spin_unlock(root_lock);
3202 if (unlikely(to_free))
3203 kfree_skb_list(to_free);
3204 if (unlikely(contended))
3205 spin_unlock(&q->busylock);
3206 return rc;
3207}
3208
3209#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3210static void skb_update_prio(struct sk_buff *skb)
3211{
3212 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3213
3214 if (!skb->priority && skb->sk && map) {
3215 unsigned int prioidx =
3216 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3217
3218 if (prioidx < map->priomap_len)
3219 skb->priority = map->priomap[prioidx];
3220 }
3221}
3222#else
3223#define skb_update_prio(skb)
3224#endif
3225
3226DEFINE_PER_CPU(int, xmit_recursion);
3227EXPORT_SYMBOL(xmit_recursion);
3228
3229
3230
3231
3232
3233
3234
3235int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3236{
3237 skb_reset_mac_header(skb);
3238 __skb_pull(skb, skb_network_offset(skb));
3239 skb->pkt_type = PACKET_LOOPBACK;
3240 skb->ip_summed = CHECKSUM_UNNECESSARY;
3241 WARN_ON(!skb_dst(skb));
3242 skb_dst_force(skb);
3243 netif_rx_ni(skb);
3244 return 0;
3245}
3246EXPORT_SYMBOL(dev_loopback_xmit);
3247
3248#ifdef CONFIG_NET_EGRESS
3249static struct sk_buff *
3250sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3251{
3252 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3253 struct tcf_result cl_res;
3254
3255 if (!cl)
3256 return skb;
3257
3258
3259 qdisc_bstats_cpu_update(cl->q, skb);
3260
3261 switch (tcf_classify(skb, cl, &cl_res, false)) {
3262 case TC_ACT_OK:
3263 case TC_ACT_RECLASSIFY:
3264 skb->tc_index = TC_H_MIN(cl_res.classid);
3265 break;
3266 case TC_ACT_SHOT:
3267 qdisc_qstats_cpu_drop(cl->q);
3268 *ret = NET_XMIT_DROP;
3269 kfree_skb(skb);
3270 return NULL;
3271 case TC_ACT_STOLEN:
3272 case TC_ACT_QUEUED:
3273 case TC_ACT_TRAP:
3274 *ret = NET_XMIT_SUCCESS;
3275 consume_skb(skb);
3276 return NULL;
3277 case TC_ACT_REDIRECT:
3278
3279 skb_do_redirect(skb);
3280 *ret = NET_XMIT_SUCCESS;
3281 return NULL;
3282 default:
3283 break;
3284 }
3285
3286 return skb;
3287}
3288#endif
3289
3290static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3291{
3292#ifdef CONFIG_XPS
3293 struct xps_dev_maps *dev_maps;
3294 struct xps_map *map;
3295 int queue_index = -1;
3296
3297 rcu_read_lock();
3298 dev_maps = rcu_dereference(dev->xps_maps);
3299 if (dev_maps) {
3300 unsigned int tci = skb->sender_cpu - 1;
3301
3302 if (dev->num_tc) {
3303 tci *= dev->num_tc;
3304 tci += netdev_get_prio_tc_map(dev, skb->priority);
3305 }
3306
3307 map = rcu_dereference(dev_maps->cpu_map[tci]);
3308 if (map) {
3309 if (map->len == 1)
3310 queue_index = map->queues[0];
3311 else
3312 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3313 map->len)];
3314 if (unlikely(queue_index >= dev->real_num_tx_queues))
3315 queue_index = -1;
3316 }
3317 }
3318 rcu_read_unlock();
3319
3320 return queue_index;
3321#else
3322 return -1;
3323#endif
3324}
3325
3326static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3327{
3328 struct sock *sk = skb->sk;
3329 int queue_index = sk_tx_queue_get(sk);
3330
3331 if (queue_index < 0 || skb->ooo_okay ||
3332 queue_index >= dev->real_num_tx_queues) {
3333 int new_index = get_xps_queue(dev, skb);
3334
3335 if (new_index < 0)
3336 new_index = skb_tx_hash(dev, skb);
3337
3338 if (queue_index != new_index && sk &&
3339 sk_fullsock(sk) &&
3340 rcu_access_pointer(sk->sk_dst_cache))
3341 sk_tx_queue_set(sk, new_index);
3342
3343 queue_index = new_index;
3344 }
3345
3346 return queue_index;
3347}
3348
3349struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3350 struct sk_buff *skb,
3351 void *accel_priv)
3352{
3353 int queue_index = 0;
3354
3355#ifdef CONFIG_XPS
3356 u32 sender_cpu = skb->sender_cpu - 1;
3357
3358 if (sender_cpu >= (u32)NR_CPUS)
3359 skb->sender_cpu = raw_smp_processor_id() + 1;
3360#endif
3361
3362 if (dev->real_num_tx_queues != 1) {
3363 const struct net_device_ops *ops = dev->netdev_ops;
3364
3365 if (ops->ndo_select_queue)
3366 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3367 __netdev_pick_tx);
3368 else
3369 queue_index = __netdev_pick_tx(dev, skb);
3370
3371 if (!accel_priv)
3372 queue_index = netdev_cap_txqueue(dev, queue_index);
3373 }
3374
3375 skb_set_queue_mapping(skb, queue_index);
3376 return netdev_get_tx_queue(dev, queue_index);
3377}
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3406{
3407 struct net_device *dev = skb->dev;
3408 struct netdev_queue *txq;
3409 struct Qdisc *q;
3410 int rc = -ENOMEM;
3411
3412 skb_reset_mac_header(skb);
3413
3414 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3415 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3416
3417
3418
3419
3420 rcu_read_lock_bh();
3421
3422 skb_update_prio(skb);
3423
3424 qdisc_pkt_len_init(skb);
3425#ifdef CONFIG_NET_CLS_ACT
3426 skb->tc_at_ingress = 0;
3427# ifdef CONFIG_NET_EGRESS
3428 if (static_key_false(&egress_needed)) {
3429 skb = sch_handle_egress(skb, &rc, dev);
3430 if (!skb)
3431 goto out;
3432 }
3433# endif
3434#endif
3435
3436
3437
3438 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3439 skb_dst_drop(skb);
3440 else
3441 skb_dst_force(skb);
3442
3443 txq = netdev_pick_tx(dev, skb, accel_priv);
3444 q = rcu_dereference_bh(txq->qdisc);
3445
3446 trace_net_dev_queue(skb);
3447 if (q->enqueue) {
3448 rc = __dev_xmit_skb(skb, q, dev, txq);
3449 goto out;
3450 }
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464 if (dev->flags & IFF_UP) {
3465 int cpu = smp_processor_id();
3466
3467 if (txq->xmit_lock_owner != cpu) {
3468 if (unlikely(__this_cpu_read(xmit_recursion) >
3469 XMIT_RECURSION_LIMIT))
3470 goto recursion_alert;
3471
3472 skb = validate_xmit_skb(skb, dev);
3473 if (!skb)
3474 goto out;
3475
3476 HARD_TX_LOCK(dev, txq, cpu);
3477
3478 if (!netif_xmit_stopped(txq)) {
3479 __this_cpu_inc(xmit_recursion);
3480 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3481 __this_cpu_dec(xmit_recursion);
3482 if (dev_xmit_complete(rc)) {
3483 HARD_TX_UNLOCK(dev, txq);
3484 goto out;
3485 }
3486 }
3487 HARD_TX_UNLOCK(dev, txq);
3488 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3489 dev->name);
3490 } else {
3491
3492
3493
3494recursion_alert:
3495 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3496 dev->name);
3497 }
3498 }
3499
3500 rc = -ENETDOWN;
3501 rcu_read_unlock_bh();
3502
3503 atomic_long_inc(&dev->tx_dropped);
3504 kfree_skb_list(skb);
3505 return rc;
3506out:
3507 rcu_read_unlock_bh();
3508 return rc;
3509}
3510
3511int dev_queue_xmit(struct sk_buff *skb)
3512{
3513 return __dev_queue_xmit(skb, NULL);
3514}
3515EXPORT_SYMBOL(dev_queue_xmit);
3516
3517int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3518{
3519 return __dev_queue_xmit(skb, accel_priv);
3520}
3521EXPORT_SYMBOL(dev_queue_xmit_accel);
3522
3523
3524
3525
3526
3527
3528int netdev_max_backlog __read_mostly = 1000;
3529EXPORT_SYMBOL(netdev_max_backlog);
3530
3531int netdev_tstamp_prequeue __read_mostly = 1;
3532int netdev_budget __read_mostly = 300;
3533unsigned int __read_mostly netdev_budget_usecs = 2000;
3534int weight_p __read_mostly = 64;
3535int dev_weight_rx_bias __read_mostly = 1;
3536int dev_weight_tx_bias __read_mostly = 1;
3537int dev_rx_weight __read_mostly = 64;
3538int dev_tx_weight __read_mostly = 64;
3539
3540
3541static inline void ____napi_schedule(struct softnet_data *sd,
3542 struct napi_struct *napi)
3543{
3544 list_add_tail(&napi->poll_list, &sd->poll_list);
3545 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3546}
3547
3548#ifdef CONFIG_RPS
3549
3550
3551struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3552EXPORT_SYMBOL(rps_sock_flow_table);
3553u32 rps_cpu_mask __read_mostly;
3554EXPORT_SYMBOL(rps_cpu_mask);
3555
3556struct static_key rps_needed __read_mostly;
3557EXPORT_SYMBOL(rps_needed);
3558struct static_key rfs_needed __read_mostly;
3559EXPORT_SYMBOL(rfs_needed);
3560
3561static struct rps_dev_flow *
3562set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3563 struct rps_dev_flow *rflow, u16 next_cpu)
3564{
3565 if (next_cpu < nr_cpu_ids) {
3566#ifdef CONFIG_RFS_ACCEL
3567 struct netdev_rx_queue *rxqueue;
3568 struct rps_dev_flow_table *flow_table;
3569 struct rps_dev_flow *old_rflow;
3570 u32 flow_id;
3571 u16 rxq_index;
3572 int rc;
3573
3574
3575 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3576 !(dev->features & NETIF_F_NTUPLE))
3577 goto out;
3578 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3579 if (rxq_index == skb_get_rx_queue(skb))
3580 goto out;
3581
3582 rxqueue = dev->_rx + rxq_index;
3583 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3584 if (!flow_table)
3585 goto out;
3586 flow_id = skb_get_hash(skb) & flow_table->mask;
3587 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3588 rxq_index, flow_id);
3589 if (rc < 0)
3590 goto out;
3591 old_rflow = rflow;
3592 rflow = &flow_table->flows[flow_id];
3593 rflow->filter = rc;
3594 if (old_rflow->filter == rflow->filter)
3595 old_rflow->filter = RPS_NO_FILTER;
3596 out:
3597#endif
3598 rflow->last_qtail =
3599 per_cpu(softnet_data, next_cpu).input_queue_head;
3600 }
3601
3602 rflow->cpu = next_cpu;
3603 return rflow;
3604}
3605
3606
3607
3608
3609
3610
3611static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3612 struct rps_dev_flow **rflowp)
3613{
3614 const struct rps_sock_flow_table *sock_flow_table;
3615 struct netdev_rx_queue *rxqueue = dev->_rx;
3616 struct rps_dev_flow_table *flow_table;
3617 struct rps_map *map;
3618 int cpu = -1;
3619 u32 tcpu;
3620 u32 hash;
3621
3622 if (skb_rx_queue_recorded(skb)) {
3623 u16 index = skb_get_rx_queue(skb);
3624
3625 if (unlikely(index >= dev->real_num_rx_queues)) {
3626 WARN_ONCE(dev->real_num_rx_queues > 1,
3627 "%s received packet on queue %u, but number "
3628 "of RX queues is %u\n",
3629 dev->name, index, dev->real_num_rx_queues);
3630 goto done;
3631 }
3632 rxqueue += index;
3633 }
3634
3635
3636
3637 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3638 map = rcu_dereference(rxqueue->rps_map);
3639 if (!flow_table && !map)
3640 goto done;
3641
3642 skb_reset_network_header(skb);
3643 hash = skb_get_hash(skb);
3644 if (!hash)
3645 goto done;
3646
3647 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3648 if (flow_table && sock_flow_table) {
3649 struct rps_dev_flow *rflow;
3650 u32 next_cpu;
3651 u32 ident;
3652
3653
3654 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3655 if ((ident ^ hash) & ~rps_cpu_mask)
3656 goto try_rps;
3657
3658 next_cpu = ident & rps_cpu_mask;
3659
3660
3661
3662
3663 rflow = &flow_table->flows[hash & flow_table->mask];
3664 tcpu = rflow->cpu;
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677 if (unlikely(tcpu != next_cpu) &&
3678 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3679 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3680 rflow->last_qtail)) >= 0)) {
3681 tcpu = next_cpu;
3682 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3683 }
3684
3685 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3686 *rflowp = rflow;
3687 cpu = tcpu;
3688 goto done;
3689 }
3690 }
3691
3692try_rps:
3693
3694 if (map) {
3695 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3696 if (cpu_online(tcpu)) {
3697 cpu = tcpu;
3698 goto done;
3699 }
3700 }
3701
3702done:
3703 return cpu;
3704}
3705
3706#ifdef CONFIG_RFS_ACCEL
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3720 u32 flow_id, u16 filter_id)
3721{
3722 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3723 struct rps_dev_flow_table *flow_table;
3724 struct rps_dev_flow *rflow;
3725 bool expire = true;
3726 unsigned int cpu;
3727
3728 rcu_read_lock();
3729 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3730 if (flow_table && flow_id <= flow_table->mask) {
3731 rflow = &flow_table->flows[flow_id];
3732 cpu = ACCESS_ONCE(rflow->cpu);
3733 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3734 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3735 rflow->last_qtail) <
3736 (int)(10 * flow_table->mask)))
3737 expire = false;
3738 }
3739 rcu_read_unlock();
3740 return expire;
3741}
3742EXPORT_SYMBOL(rps_may_expire_flow);
3743
3744#endif
3745
3746
3747static void rps_trigger_softirq(void *data)
3748{
3749 struct softnet_data *sd = data;
3750
3751 ____napi_schedule(sd, &sd->backlog);
3752 sd->received_rps++;
3753}
3754
3755#endif
3756
3757
3758
3759
3760
3761
3762static int rps_ipi_queued(struct softnet_data *sd)
3763{
3764#ifdef CONFIG_RPS
3765 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3766
3767 if (sd != mysd) {
3768 sd->rps_ipi_next = mysd->rps_ipi_list;
3769 mysd->rps_ipi_list = sd;
3770
3771 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3772 return 1;
3773 }
3774#endif
3775 return 0;
3776}
3777
3778#ifdef CONFIG_NET_FLOW_LIMIT
3779int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3780#endif
3781
3782static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3783{
3784#ifdef CONFIG_NET_FLOW_LIMIT
3785 struct sd_flow_limit *fl;
3786 struct softnet_data *sd;
3787 unsigned int old_flow, new_flow;
3788
3789 if (qlen < (netdev_max_backlog >> 1))
3790 return false;
3791
3792 sd = this_cpu_ptr(&softnet_data);
3793
3794 rcu_read_lock();
3795 fl = rcu_dereference(sd->flow_limit);
3796 if (fl) {
3797 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3798 old_flow = fl->history[fl->history_head];
3799 fl->history[fl->history_head] = new_flow;
3800
3801 fl->history_head++;
3802 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3803
3804 if (likely(fl->buckets[old_flow]))
3805 fl->buckets[old_flow]--;
3806
3807 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3808 fl->count++;
3809 rcu_read_unlock();
3810 return true;
3811 }
3812 }
3813 rcu_read_unlock();
3814#endif
3815 return false;
3816}
3817
3818
3819
3820
3821
3822static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3823 unsigned int *qtail)
3824{
3825 struct softnet_data *sd;
3826 unsigned long flags;
3827 unsigned int qlen;
3828
3829 sd = &per_cpu(softnet_data, cpu);
3830
3831 local_irq_save(flags);
3832
3833 rps_lock(sd);
3834 if (!netif_running(skb->dev))
3835 goto drop;
3836 qlen = skb_queue_len(&sd->input_pkt_queue);
3837 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3838 if (qlen) {
3839enqueue:
3840 __skb_queue_tail(&sd->input_pkt_queue, skb);
3841 input_queue_tail_incr_save(sd, qtail);
3842 rps_unlock(sd);
3843 local_irq_restore(flags);
3844 return NET_RX_SUCCESS;
3845 }
3846
3847
3848
3849
3850 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3851 if (!rps_ipi_queued(sd))
3852 ____napi_schedule(sd, &sd->backlog);
3853 }
3854 goto enqueue;
3855 }
3856
3857drop:
3858 sd->dropped++;
3859 rps_unlock(sd);
3860
3861 local_irq_restore(flags);
3862
3863 atomic_long_inc(&skb->dev->rx_dropped);
3864 kfree_skb(skb);
3865 return NET_RX_DROP;
3866}
3867
3868static int netif_rx_internal(struct sk_buff *skb)
3869{
3870 int ret;
3871
3872 net_timestamp_check(netdev_tstamp_prequeue, skb);
3873
3874 trace_netif_rx(skb);
3875#ifdef CONFIG_RPS
3876 if (static_key_false(&rps_needed)) {
3877 struct rps_dev_flow voidflow, *rflow = &voidflow;
3878 int cpu;
3879
3880 preempt_disable();
3881 rcu_read_lock();
3882
3883 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3884 if (cpu < 0)
3885 cpu = smp_processor_id();
3886
3887 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3888
3889 rcu_read_unlock();
3890 preempt_enable();
3891 } else
3892#endif
3893 {
3894 unsigned int qtail;
3895
3896 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3897 put_cpu();
3898 }
3899 return ret;
3900}
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917int netif_rx(struct sk_buff *skb)
3918{
3919 trace_netif_rx_entry(skb);
3920
3921 return netif_rx_internal(skb);
3922}
3923EXPORT_SYMBOL(netif_rx);
3924
3925int netif_rx_ni(struct sk_buff *skb)
3926{
3927 int err;
3928
3929 trace_netif_rx_ni_entry(skb);
3930
3931 preempt_disable();
3932 err = netif_rx_internal(skb);
3933 if (local_softirq_pending())
3934 do_softirq();
3935 preempt_enable();
3936
3937 return err;
3938}
3939EXPORT_SYMBOL(netif_rx_ni);
3940
3941static __latent_entropy void net_tx_action(struct softirq_action *h)
3942{
3943 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3944
3945 if (sd->completion_queue) {
3946 struct sk_buff *clist;
3947
3948 local_irq_disable();
3949 clist = sd->completion_queue;
3950 sd->completion_queue = NULL;
3951 local_irq_enable();
3952
3953 while (clist) {
3954 struct sk_buff *skb = clist;
3955
3956 clist = clist->next;
3957
3958 WARN_ON(refcount_read(&skb->users));
3959 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3960 trace_consume_skb(skb);
3961 else
3962 trace_kfree_skb(skb, net_tx_action);
3963
3964 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3965 __kfree_skb(skb);
3966 else
3967 __kfree_skb_defer(skb);
3968 }
3969
3970 __kfree_skb_flush();
3971 }
3972
3973 if (sd->output_queue) {
3974 struct Qdisc *head;
3975
3976 local_irq_disable();
3977 head = sd->output_queue;
3978 sd->output_queue = NULL;
3979 sd->output_queue_tailp = &sd->output_queue;
3980 local_irq_enable();
3981
3982 while (head) {
3983 struct Qdisc *q = head;
3984 spinlock_t *root_lock;
3985
3986 head = head->next_sched;
3987
3988 root_lock = qdisc_lock(q);
3989 spin_lock(root_lock);
3990
3991
3992
3993 smp_mb__before_atomic();
3994 clear_bit(__QDISC_STATE_SCHED, &q->state);
3995 qdisc_run(q);
3996 spin_unlock(root_lock);
3997 }
3998 }
3999}
4000
4001#if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4002
4003int (*br_fdb_test_addr_hook)(struct net_device *dev,
4004 unsigned char *addr) __read_mostly;
4005EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4006#endif
4007
4008static inline struct sk_buff *
4009sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4010 struct net_device *orig_dev)
4011{
4012#ifdef CONFIG_NET_CLS_ACT
4013 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
4014 struct tcf_result cl_res;
4015
4016
4017
4018
4019
4020
4021 if (!cl)
4022 return skb;
4023 if (*pt_prev) {
4024 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4025 *pt_prev = NULL;
4026 }
4027
4028 qdisc_skb_cb(skb)->pkt_len = skb->len;
4029 skb->tc_at_ingress = 1;
4030 qdisc_bstats_cpu_update(cl->q, skb);
4031
4032 switch (tcf_classify(skb, cl, &cl_res, false)) {
4033 case TC_ACT_OK:
4034 case TC_ACT_RECLASSIFY:
4035 skb->tc_index = TC_H_MIN(cl_res.classid);
4036 break;
4037 case TC_ACT_SHOT:
4038 qdisc_qstats_cpu_drop(cl->q);
4039 kfree_skb(skb);
4040 return NULL;
4041 case TC_ACT_STOLEN:
4042 case TC_ACT_QUEUED:
4043 case TC_ACT_TRAP:
4044 consume_skb(skb);
4045 return NULL;
4046 case TC_ACT_REDIRECT:
4047
4048
4049
4050
4051 __skb_push(skb, skb->mac_len);
4052 skb_do_redirect(skb);
4053 return NULL;
4054 default:
4055 break;
4056 }
4057#endif
4058 return skb;
4059}
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070bool netdev_is_rx_handler_busy(struct net_device *dev)
4071{
4072 ASSERT_RTNL();
4073 return dev && rtnl_dereference(dev->rx_handler);
4074}
4075EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091int netdev_rx_handler_register(struct net_device *dev,
4092 rx_handler_func_t *rx_handler,
4093 void *rx_handler_data)
4094{
4095 if (netdev_is_rx_handler_busy(dev))
4096 return -EBUSY;
4097
4098
4099 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4100 rcu_assign_pointer(dev->rx_handler, rx_handler);
4101
4102 return 0;
4103}
4104EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114void netdev_rx_handler_unregister(struct net_device *dev)
4115{
4116
4117 ASSERT_RTNL();
4118 RCU_INIT_POINTER(dev->rx_handler, NULL);
4119
4120
4121
4122
4123 synchronize_net();
4124 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4125}
4126EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4127
4128
4129
4130
4131
4132static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4133{
4134 switch (skb->protocol) {
4135 case htons(ETH_P_ARP):
4136 case htons(ETH_P_IP):
4137 case htons(ETH_P_IPV6):
4138 case htons(ETH_P_8021Q):
4139 case htons(ETH_P_8021AD):
4140 return true;
4141 default:
4142 return false;
4143 }
4144}
4145
4146static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4147 int *ret, struct net_device *orig_dev)
4148{
4149#ifdef CONFIG_NETFILTER_INGRESS
4150 if (nf_hook_ingress_active(skb)) {
4151 int ingress_retval;
4152
4153 if (*pt_prev) {
4154 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4155 *pt_prev = NULL;
4156 }
4157
4158 rcu_read_lock();
4159 ingress_retval = nf_hook_ingress(skb);
4160 rcu_read_unlock();
4161 return ingress_retval;
4162 }
4163#endif
4164 return 0;
4165}
4166
4167static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4168{
4169 struct packet_type *ptype, *pt_prev;
4170 rx_handler_func_t *rx_handler;
4171 struct net_device *orig_dev;
4172 bool deliver_exact = false;
4173 int ret = NET_RX_DROP;
4174 __be16 type;
4175
4176 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4177
4178 trace_netif_receive_skb(skb);
4179
4180 orig_dev = skb->dev;
4181
4182 skb_reset_network_header(skb);
4183 if (!skb_transport_header_was_set(skb))
4184 skb_reset_transport_header(skb);
4185 skb_reset_mac_len(skb);
4186
4187 pt_prev = NULL;
4188
4189another_round:
4190 skb->skb_iif = skb->dev->ifindex;
4191
4192 __this_cpu_inc(softnet_data.processed);
4193
4194 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4195 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4196 skb = skb_vlan_untag(skb);
4197 if (unlikely(!skb))
4198 goto out;
4199 }
4200
4201 if (skb_skip_tc_classify(skb))
4202 goto skip_classify;
4203
4204 if (pfmemalloc)
4205 goto skip_taps;
4206
4207 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4208 if (pt_prev)
4209 ret = deliver_skb(skb, pt_prev, orig_dev);
4210 pt_prev = ptype;
4211 }
4212
4213 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4214 if (pt_prev)
4215 ret = deliver_skb(skb, pt_prev, orig_dev);
4216 pt_prev = ptype;
4217 }
4218
4219skip_taps:
4220#ifdef CONFIG_NET_INGRESS
4221 if (static_key_false(&ingress_needed)) {
4222 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4223 if (!skb)
4224 goto out;
4225
4226 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4227 goto out;
4228 }
4229#endif
4230 skb_reset_tc(skb);
4231skip_classify:
4232 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4233 goto drop;
4234
4235 if (skb_vlan_tag_present(skb)) {
4236 if (pt_prev) {
4237 ret = deliver_skb(skb, pt_prev, orig_dev);
4238 pt_prev = NULL;
4239 }
4240 if (vlan_do_receive(&skb))
4241 goto another_round;
4242 else if (unlikely(!skb))
4243 goto out;
4244 }
4245
4246 rx_handler = rcu_dereference(skb->dev->rx_handler);
4247 if (rx_handler) {
4248 if (pt_prev) {
4249 ret = deliver_skb(skb, pt_prev, orig_dev);
4250 pt_prev = NULL;
4251 }
4252 switch (rx_handler(&skb)) {
4253 case RX_HANDLER_CONSUMED:
4254 ret = NET_RX_SUCCESS;
4255 goto out;
4256 case RX_HANDLER_ANOTHER:
4257 goto another_round;
4258 case RX_HANDLER_EXACT:
4259 deliver_exact = true;
4260 case RX_HANDLER_PASS:
4261 break;
4262 default:
4263 BUG();
4264 }
4265 }
4266
4267 if (unlikely(skb_vlan_tag_present(skb))) {
4268 if (skb_vlan_tag_get_id(skb))
4269 skb->pkt_type = PACKET_OTHERHOST;
4270
4271
4272
4273
4274 skb->vlan_tci = 0;
4275 }
4276
4277 type = skb->protocol;
4278
4279
4280 if (likely(!deliver_exact)) {
4281 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4282 &ptype_base[ntohs(type) &
4283 PTYPE_HASH_MASK]);
4284 }
4285
4286 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4287 &orig_dev->ptype_specific);
4288
4289 if (unlikely(skb->dev != orig_dev)) {
4290 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4291 &skb->dev->ptype_specific);
4292 }
4293
4294 if (pt_prev) {
4295 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4296 goto drop;
4297 else
4298 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4299 } else {
4300drop:
4301 if (!deliver_exact)
4302 atomic_long_inc(&skb->dev->rx_dropped);
4303 else
4304 atomic_long_inc(&skb->dev->rx_nohandler);
4305 kfree_skb(skb);
4306
4307
4308
4309 ret = NET_RX_DROP;
4310 }
4311
4312out:
4313 return ret;
4314}
4315
4316static int __netif_receive_skb(struct sk_buff *skb)
4317{
4318 int ret;
4319
4320 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4321 unsigned int noreclaim_flag;
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332 noreclaim_flag = memalloc_noreclaim_save();
4333 ret = __netif_receive_skb_core(skb, true);
4334 memalloc_noreclaim_restore(noreclaim_flag);
4335 } else
4336 ret = __netif_receive_skb_core(skb, false);
4337
4338 return ret;
4339}
4340
4341static struct static_key generic_xdp_needed __read_mostly;
4342
4343static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp)
4344{
4345 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
4346 struct bpf_prog *new = xdp->prog;
4347 int ret = 0;
4348
4349 switch (xdp->command) {
4350 case XDP_SETUP_PROG:
4351 rcu_assign_pointer(dev->xdp_prog, new);
4352 if (old)
4353 bpf_prog_put(old);
4354
4355 if (old && !new) {
4356 static_key_slow_dec(&generic_xdp_needed);
4357 } else if (new && !old) {
4358 static_key_slow_inc(&generic_xdp_needed);
4359 dev_disable_lro(dev);
4360 }
4361 break;
4362
4363 case XDP_QUERY_PROG:
4364 xdp->prog_attached = !!old;
4365 xdp->prog_id = old ? old->aux->id : 0;
4366 break;
4367
4368 default:
4369 ret = -EINVAL;
4370 break;
4371 }
4372
4373 return ret;
4374}
4375
4376static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4377 struct bpf_prog *xdp_prog)
4378{
4379 struct xdp_buff xdp;
4380 u32 act = XDP_DROP;
4381 void *orig_data;
4382 int hlen, off;
4383 u32 mac_len;
4384
4385
4386
4387
4388 if (skb_cloned(skb))
4389 return XDP_PASS;
4390
4391 if (skb_linearize(skb))
4392 goto do_drop;
4393
4394
4395
4396
4397 mac_len = skb->data - skb_mac_header(skb);
4398 hlen = skb_headlen(skb) + mac_len;
4399 xdp.data = skb->data - mac_len;
4400 xdp.data_end = xdp.data + hlen;
4401 xdp.data_hard_start = skb->data - skb_headroom(skb);
4402 orig_data = xdp.data;
4403
4404 act = bpf_prog_run_xdp(xdp_prog, &xdp);
4405
4406 off = xdp.data - orig_data;
4407 if (off > 0)
4408 __skb_pull(skb, off);
4409 else if (off < 0)
4410 __skb_push(skb, -off);
4411
4412 switch (act) {
4413 case XDP_TX:
4414 __skb_push(skb, mac_len);
4415
4416 case XDP_PASS:
4417 break;
4418
4419 default:
4420 bpf_warn_invalid_xdp_action(act);
4421
4422 case XDP_ABORTED:
4423 trace_xdp_exception(skb->dev, xdp_prog, act);
4424
4425 case XDP_DROP:
4426 do_drop:
4427 kfree_skb(skb);
4428 break;
4429 }
4430
4431 return act;
4432}
4433
4434
4435
4436
4437static void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4438{
4439 struct net_device *dev = skb->dev;
4440 struct netdev_queue *txq;
4441 bool free_skb = true;
4442 int cpu, rc;
4443
4444 txq = netdev_pick_tx(dev, skb, NULL);
4445 cpu = smp_processor_id();
4446 HARD_TX_LOCK(dev, txq, cpu);
4447 if (!netif_xmit_stopped(txq)) {
4448 rc = netdev_start_xmit(skb, dev, txq, 0);
4449 if (dev_xmit_complete(rc))
4450 free_skb = false;
4451 }
4452 HARD_TX_UNLOCK(dev, txq);
4453 if (free_skb) {
4454 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4455 kfree_skb(skb);
4456 }
4457}
4458
4459static int netif_receive_skb_internal(struct sk_buff *skb)
4460{
4461 int ret;
4462
4463 net_timestamp_check(netdev_tstamp_prequeue, skb);
4464
4465 if (skb_defer_rx_timestamp(skb))
4466 return NET_RX_SUCCESS;
4467
4468 rcu_read_lock();
4469
4470 if (static_key_false(&generic_xdp_needed)) {
4471 struct bpf_prog *xdp_prog = rcu_dereference(skb->dev->xdp_prog);
4472
4473 if (xdp_prog) {
4474 u32 act = netif_receive_generic_xdp(skb, xdp_prog);
4475
4476 if (act != XDP_PASS) {
4477 rcu_read_unlock();
4478 if (act == XDP_TX)
4479 generic_xdp_tx(skb, xdp_prog);
4480 return NET_RX_DROP;
4481 }
4482 }
4483 }
4484
4485#ifdef CONFIG_RPS
4486 if (static_key_false(&rps_needed)) {
4487 struct rps_dev_flow voidflow, *rflow = &voidflow;
4488 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4489
4490 if (cpu >= 0) {
4491 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4492 rcu_read_unlock();
4493 return ret;
4494 }
4495 }
4496#endif
4497 ret = __netif_receive_skb(skb);
4498 rcu_read_unlock();
4499 return ret;
4500}
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517int netif_receive_skb(struct sk_buff *skb)
4518{
4519 trace_netif_receive_skb_entry(skb);
4520
4521 return netif_receive_skb_internal(skb);
4522}
4523EXPORT_SYMBOL(netif_receive_skb);
4524
4525DEFINE_PER_CPU(struct work_struct, flush_works);
4526
4527
4528static void flush_backlog(struct work_struct *work)
4529{
4530 struct sk_buff *skb, *tmp;
4531 struct softnet_data *sd;
4532
4533 local_bh_disable();
4534 sd = this_cpu_ptr(&softnet_data);
4535
4536 local_irq_disable();
4537 rps_lock(sd);
4538 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4539 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4540 __skb_unlink(skb, &sd->input_pkt_queue);
4541 kfree_skb(skb);
4542 input_queue_head_incr(sd);
4543 }
4544 }
4545 rps_unlock(sd);
4546 local_irq_enable();
4547
4548 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4549 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4550 __skb_unlink(skb, &sd->process_queue);
4551 kfree_skb(skb);
4552 input_queue_head_incr(sd);
4553 }
4554 }
4555 local_bh_enable();
4556}
4557
4558static void flush_all_backlogs(void)
4559{
4560 unsigned int cpu;
4561
4562 get_online_cpus();
4563
4564 for_each_online_cpu(cpu)
4565 queue_work_on(cpu, system_highpri_wq,
4566 per_cpu_ptr(&flush_works, cpu));
4567
4568 for_each_online_cpu(cpu)
4569 flush_work(per_cpu_ptr(&flush_works, cpu));
4570
4571 put_online_cpus();
4572}
4573
4574static int napi_gro_complete(struct sk_buff *skb)
4575{
4576 struct packet_offload *ptype;
4577 __be16 type = skb->protocol;
4578 struct list_head *head = &offload_base;
4579 int err = -ENOENT;
4580
4581 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4582
4583 if (NAPI_GRO_CB(skb)->count == 1) {
4584 skb_shinfo(skb)->gso_size = 0;
4585 goto out;
4586 }
4587
4588 rcu_read_lock();
4589 list_for_each_entry_rcu(ptype, head, list) {
4590 if (ptype->type != type || !ptype->callbacks.gro_complete)
4591 continue;
4592
4593 err = ptype->callbacks.gro_complete(skb, 0);
4594 break;
4595 }
4596 rcu_read_unlock();
4597
4598 if (err) {
4599 WARN_ON(&ptype->list == head);
4600 kfree_skb(skb);
4601 return NET_RX_SUCCESS;
4602 }
4603
4604out:
4605 return netif_receive_skb_internal(skb);
4606}
4607
4608
4609
4610
4611
4612void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4613{
4614 struct sk_buff *skb, *prev = NULL;
4615
4616
4617 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4618 skb->prev = prev;
4619 prev = skb;
4620 }
4621
4622 for (skb = prev; skb; skb = prev) {
4623 skb->next = NULL;
4624
4625 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4626 return;
4627
4628 prev = skb->prev;
4629 napi_gro_complete(skb);
4630 napi->gro_count--;
4631 }
4632
4633 napi->gro_list = NULL;
4634}
4635EXPORT_SYMBOL(napi_gro_flush);
4636
4637static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4638{
4639 struct sk_buff *p;
4640 unsigned int maclen = skb->dev->hard_header_len;
4641 u32 hash = skb_get_hash_raw(skb);
4642
4643 for (p = napi->gro_list; p; p = p->next) {
4644 unsigned long diffs;
4645
4646 NAPI_GRO_CB(p)->flush = 0;
4647
4648 if (hash != skb_get_hash_raw(p)) {
4649 NAPI_GRO_CB(p)->same_flow = 0;
4650 continue;
4651 }
4652
4653 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4654 diffs |= p->vlan_tci ^ skb->vlan_tci;
4655 diffs |= skb_metadata_dst_cmp(p, skb);
4656 if (maclen == ETH_HLEN)
4657 diffs |= compare_ether_header(skb_mac_header(p),
4658 skb_mac_header(skb));
4659 else if (!diffs)
4660 diffs = memcmp(skb_mac_header(p),
4661 skb_mac_header(skb),
4662 maclen);
4663 NAPI_GRO_CB(p)->same_flow = !diffs;
4664 }
4665}
4666
4667static void skb_gro_reset_offset(struct sk_buff *skb)
4668{
4669 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4670 const skb_frag_t *frag0 = &pinfo->frags[0];
4671
4672 NAPI_GRO_CB(skb)->data_offset = 0;
4673 NAPI_GRO_CB(skb)->frag0 = NULL;
4674 NAPI_GRO_CB(skb)->frag0_len = 0;
4675
4676 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4677 pinfo->nr_frags &&
4678 !PageHighMem(skb_frag_page(frag0))) {
4679 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4680 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4681 skb_frag_size(frag0),
4682 skb->end - skb->tail);
4683 }
4684}
4685
4686static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4687{
4688 struct skb_shared_info *pinfo = skb_shinfo(skb);
4689
4690 BUG_ON(skb->end - skb->tail < grow);
4691
4692 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4693
4694 skb->data_len -= grow;
4695 skb->tail += grow;
4696
4697 pinfo->frags[0].page_offset += grow;
4698 skb_frag_size_sub(&pinfo->frags[0], grow);
4699
4700 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4701 skb_frag_unref(skb, 0);
4702 memmove(pinfo->frags, pinfo->frags + 1,
4703 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4704 }
4705}
4706
4707static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4708{
4709 struct sk_buff **pp = NULL;
4710 struct packet_offload *ptype;
4711 __be16 type = skb->protocol;
4712 struct list_head *head = &offload_base;
4713 int same_flow;
4714 enum gro_result ret;
4715 int grow;
4716
4717 if (netif_elide_gro(skb->dev))
4718 goto normal;
4719
4720 gro_list_prepare(napi, skb);
4721
4722 rcu_read_lock();
4723 list_for_each_entry_rcu(ptype, head, list) {
4724 if (ptype->type != type || !ptype->callbacks.gro_receive)
4725 continue;
4726
4727 skb_set_network_header(skb, skb_gro_offset(skb));
4728 skb_reset_mac_len(skb);
4729 NAPI_GRO_CB(skb)->same_flow = 0;
4730 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4731 NAPI_GRO_CB(skb)->free = 0;
4732 NAPI_GRO_CB(skb)->encap_mark = 0;
4733 NAPI_GRO_CB(skb)->recursion_counter = 0;
4734 NAPI_GRO_CB(skb)->is_fou = 0;
4735 NAPI_GRO_CB(skb)->is_atomic = 1;
4736 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4737
4738
4739 switch (skb->ip_summed) {
4740 case CHECKSUM_COMPLETE:
4741 NAPI_GRO_CB(skb)->csum = skb->csum;
4742 NAPI_GRO_CB(skb)->csum_valid = 1;
4743 NAPI_GRO_CB(skb)->csum_cnt = 0;
4744 break;
4745 case CHECKSUM_UNNECESSARY:
4746 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4747 NAPI_GRO_CB(skb)->csum_valid = 0;
4748 break;
4749 default:
4750 NAPI_GRO_CB(skb)->csum_cnt = 0;
4751 NAPI_GRO_CB(skb)->csum_valid = 0;
4752 }
4753
4754 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4755 break;
4756 }
4757 rcu_read_unlock();
4758
4759 if (&ptype->list == head)
4760 goto normal;
4761
4762 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
4763 ret = GRO_CONSUMED;
4764 goto ok;
4765 }
4766
4767 same_flow = NAPI_GRO_CB(skb)->same_flow;
4768 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4769
4770 if (pp) {
4771 struct sk_buff *nskb = *pp;
4772
4773 *pp = nskb->next;
4774 nskb->next = NULL;
4775 napi_gro_complete(nskb);
4776 napi->gro_count--;
4777 }
4778
4779 if (same_flow)
4780 goto ok;
4781
4782 if (NAPI_GRO_CB(skb)->flush)
4783 goto normal;
4784
4785 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4786 struct sk_buff *nskb = napi->gro_list;
4787
4788
4789 while (nskb->next) {
4790 pp = &nskb->next;
4791 nskb = *pp;
4792 }
4793 *pp = NULL;
4794 nskb->next = NULL;
4795 napi_gro_complete(nskb);
4796 } else {
4797 napi->gro_count++;
4798 }
4799 NAPI_GRO_CB(skb)->count = 1;
4800 NAPI_GRO_CB(skb)->age = jiffies;
4801 NAPI_GRO_CB(skb)->last = skb;
4802 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4803 skb->next = napi->gro_list;
4804 napi->gro_list = skb;
4805 ret = GRO_HELD;
4806
4807pull:
4808 grow = skb_gro_offset(skb) - skb_headlen(skb);
4809 if (grow > 0)
4810 gro_pull_from_frag0(skb, grow);
4811ok:
4812 return ret;
4813
4814normal:
4815 ret = GRO_NORMAL;
4816 goto pull;
4817}
4818
4819struct packet_offload *gro_find_receive_by_type(__be16 type)
4820{
4821 struct list_head *offload_head = &offload_base;
4822 struct packet_offload *ptype;
4823
4824 list_for_each_entry_rcu(ptype, offload_head, list) {
4825 if (ptype->type != type || !ptype->callbacks.gro_receive)
4826 continue;
4827 return ptype;
4828 }
4829 return NULL;
4830}
4831EXPORT_SYMBOL(gro_find_receive_by_type);
4832
4833struct packet_offload *gro_find_complete_by_type(__be16 type)
4834{
4835 struct list_head *offload_head = &offload_base;
4836 struct packet_offload *ptype;
4837
4838 list_for_each_entry_rcu(ptype, offload_head, list) {
4839 if (ptype->type != type || !ptype->callbacks.gro_complete)
4840 continue;
4841 return ptype;
4842 }
4843 return NULL;
4844}
4845EXPORT_SYMBOL(gro_find_complete_by_type);
4846
4847static void napi_skb_free_stolen_head(struct sk_buff *skb)
4848{
4849 skb_dst_drop(skb);
4850 secpath_reset(skb);
4851 kmem_cache_free(skbuff_head_cache, skb);
4852}
4853
4854static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4855{
4856 switch (ret) {
4857 case GRO_NORMAL:
4858 if (netif_receive_skb_internal(skb))
4859 ret = GRO_DROP;
4860 break;
4861
4862 case GRO_DROP:
4863 kfree_skb(skb);
4864 break;
4865
4866 case GRO_MERGED_FREE:
4867 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4868 napi_skb_free_stolen_head(skb);
4869 else
4870 __kfree_skb(skb);
4871 break;
4872
4873 case GRO_HELD:
4874 case GRO_MERGED:
4875 case GRO_CONSUMED:
4876 break;
4877 }
4878
4879 return ret;
4880}
4881
4882gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4883{
4884 skb_mark_napi_id(skb, napi);
4885 trace_napi_gro_receive_entry(skb);
4886
4887 skb_gro_reset_offset(skb);
4888
4889 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4890}
4891EXPORT_SYMBOL(napi_gro_receive);
4892
4893static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4894{
4895 if (unlikely(skb->pfmemalloc)) {
4896 consume_skb(skb);
4897 return;
4898 }
4899 __skb_pull(skb, skb_headlen(skb));
4900
4901 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4902 skb->vlan_tci = 0;
4903 skb->dev = napi->dev;
4904 skb->skb_iif = 0;
4905 skb->encapsulation = 0;
4906 skb_shinfo(skb)->gso_type = 0;
4907 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4908 secpath_reset(skb);
4909
4910 napi->skb = skb;
4911}
4912
4913struct sk_buff *napi_get_frags(struct napi_struct *napi)
4914{
4915 struct sk_buff *skb = napi->skb;
4916
4917 if (!skb) {
4918 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4919 if (skb) {
4920 napi->skb = skb;
4921 skb_mark_napi_id(skb, napi);
4922 }
4923 }
4924 return skb;
4925}
4926EXPORT_SYMBOL(napi_get_frags);
4927
4928static gro_result_t napi_frags_finish(struct napi_struct *napi,
4929 struct sk_buff *skb,
4930 gro_result_t ret)
4931{
4932 switch (ret) {
4933 case GRO_NORMAL:
4934 case GRO_HELD:
4935 __skb_push(skb, ETH_HLEN);
4936 skb->protocol = eth_type_trans(skb, skb->dev);
4937 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4938 ret = GRO_DROP;
4939 break;
4940
4941 case GRO_DROP:
4942 napi_reuse_skb(napi, skb);
4943 break;
4944
4945 case GRO_MERGED_FREE:
4946 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4947 napi_skb_free_stolen_head(skb);
4948 else
4949 napi_reuse_skb(napi, skb);
4950 break;
4951
4952 case GRO_MERGED:
4953 case GRO_CONSUMED:
4954 break;
4955 }
4956
4957 return ret;
4958}
4959
4960
4961
4962
4963
4964static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4965{
4966 struct sk_buff *skb = napi->skb;
4967 const struct ethhdr *eth;
4968 unsigned int hlen = sizeof(*eth);
4969
4970 napi->skb = NULL;
4971
4972 skb_reset_mac_header(skb);
4973 skb_gro_reset_offset(skb);
4974
4975 eth = skb_gro_header_fast(skb, 0);
4976 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4977 eth = skb_gro_header_slow(skb, hlen, 0);
4978 if (unlikely(!eth)) {
4979 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4980 __func__, napi->dev->name);
4981 napi_reuse_skb(napi, skb);
4982 return NULL;
4983 }
4984 } else {
4985 gro_pull_from_frag0(skb, hlen);
4986 NAPI_GRO_CB(skb)->frag0 += hlen;
4987 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4988 }
4989 __skb_pull(skb, hlen);
4990
4991
4992
4993
4994
4995
4996 skb->protocol = eth->h_proto;
4997
4998 return skb;
4999}
5000
5001gro_result_t napi_gro_frags(struct napi_struct *napi)
5002{
5003 struct sk_buff *skb = napi_frags_skb(napi);
5004
5005 if (!skb)
5006 return GRO_DROP;
5007
5008 trace_napi_gro_frags_entry(skb);
5009
5010 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5011}
5012EXPORT_SYMBOL(napi_gro_frags);
5013
5014
5015
5016
5017__sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5018{
5019 __wsum wsum;
5020 __sum16 sum;
5021
5022 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5023
5024
5025 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5026 if (likely(!sum)) {
5027 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5028 !skb->csum_complete_sw)
5029 netdev_rx_csum_fault(skb->dev);
5030 }
5031
5032 NAPI_GRO_CB(skb)->csum = wsum;
5033 NAPI_GRO_CB(skb)->csum_valid = 1;
5034
5035 return sum;
5036}
5037EXPORT_SYMBOL(__skb_gro_checksum_complete);
5038
5039static void net_rps_send_ipi(struct softnet_data *remsd)
5040{
5041#ifdef CONFIG_RPS
5042 while (remsd) {
5043 struct softnet_data *next = remsd->rps_ipi_next;
5044
5045 if (cpu_online(remsd->cpu))
5046 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5047 remsd = next;
5048 }
5049#endif
5050}
5051
5052
5053
5054
5055
5056static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5057{
5058#ifdef CONFIG_RPS
5059 struct softnet_data *remsd = sd->rps_ipi_list;
5060
5061 if (remsd) {
5062 sd->rps_ipi_list = NULL;
5063
5064 local_irq_enable();
5065
5066
5067 net_rps_send_ipi(remsd);
5068 } else
5069#endif
5070 local_irq_enable();
5071}
5072
5073static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5074{
5075#ifdef CONFIG_RPS
5076 return sd->rps_ipi_list != NULL;
5077#else
5078 return false;
5079#endif
5080}
5081
5082static int process_backlog(struct napi_struct *napi, int quota)
5083{
5084 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5085 bool again = true;
5086 int work = 0;
5087
5088
5089
5090
5091 if (sd_has_rps_ipi_waiting(sd)) {
5092 local_irq_disable();
5093 net_rps_action_and_irq_enable(sd);
5094 }
5095
5096 napi->weight = dev_rx_weight;
5097 while (again) {
5098 struct sk_buff *skb;
5099
5100 while ((skb = __skb_dequeue(&sd->process_queue))) {
5101 rcu_read_lock();
5102 __netif_receive_skb(skb);
5103 rcu_read_unlock();
5104 input_queue_head_incr(sd);
5105 if (++work >= quota)
5106 return work;
5107
5108 }
5109
5110 local_irq_disable();
5111 rps_lock(sd);
5112 if (skb_queue_empty(&sd->input_pkt_queue)) {
5113
5114
5115
5116
5117
5118
5119
5120
5121 napi->state = 0;
5122 again = false;
5123 } else {
5124 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5125 &sd->process_queue);
5126 }
5127 rps_unlock(sd);
5128 local_irq_enable();
5129 }
5130
5131 return work;
5132}
5133
5134
5135
5136
5137
5138
5139
5140
5141void __napi_schedule(struct napi_struct *n)
5142{
5143 unsigned long flags;
5144
5145 local_irq_save(flags);
5146 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5147 local_irq_restore(flags);
5148}
5149EXPORT_SYMBOL(__napi_schedule);
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160bool napi_schedule_prep(struct napi_struct *n)
5161{
5162 unsigned long val, new;
5163
5164 do {
5165 val = READ_ONCE(n->state);
5166 if (unlikely(val & NAPIF_STATE_DISABLE))
5167 return false;
5168 new = val | NAPIF_STATE_SCHED;
5169
5170
5171
5172
5173
5174
5175
5176 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5177 NAPIF_STATE_MISSED;
5178 } while (cmpxchg(&n->state, val, new) != val);
5179
5180 return !(val & NAPIF_STATE_SCHED);
5181}
5182EXPORT_SYMBOL(napi_schedule_prep);
5183
5184
5185
5186
5187
5188
5189
5190void __napi_schedule_irqoff(struct napi_struct *n)
5191{
5192 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5193}
5194EXPORT_SYMBOL(__napi_schedule_irqoff);
5195
5196bool napi_complete_done(struct napi_struct *n, int work_done)
5197{
5198 unsigned long flags, val, new;
5199
5200
5201
5202
5203
5204
5205
5206 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5207 NAPIF_STATE_IN_BUSY_POLL)))
5208 return false;
5209
5210 if (n->gro_list) {
5211 unsigned long timeout = 0;
5212
5213 if (work_done)
5214 timeout = n->dev->gro_flush_timeout;
5215
5216 if (timeout)
5217 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5218 HRTIMER_MODE_REL_PINNED);
5219 else
5220 napi_gro_flush(n, false);
5221 }
5222 if (unlikely(!list_empty(&n->poll_list))) {
5223
5224 local_irq_save(flags);
5225 list_del_init(&n->poll_list);
5226 local_irq_restore(flags);
5227 }
5228
5229 do {
5230 val = READ_ONCE(n->state);
5231
5232 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5233
5234 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
5235
5236
5237
5238
5239
5240 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5241 NAPIF_STATE_SCHED;
5242 } while (cmpxchg(&n->state, val, new) != val);
5243
5244 if (unlikely(val & NAPIF_STATE_MISSED)) {
5245 __napi_schedule(n);
5246 return false;
5247 }
5248
5249 return true;
5250}
5251EXPORT_SYMBOL(napi_complete_done);
5252
5253
5254static struct napi_struct *napi_by_id(unsigned int napi_id)
5255{
5256 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5257 struct napi_struct *napi;
5258
5259 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5260 if (napi->napi_id == napi_id)
5261 return napi;
5262
5263 return NULL;
5264}
5265
5266#if defined(CONFIG_NET_RX_BUSY_POLL)
5267
5268#define BUSY_POLL_BUDGET 8
5269
5270static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
5271{
5272 int rc;
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283 clear_bit(NAPI_STATE_MISSED, &napi->state);
5284 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
5285
5286 local_bh_disable();
5287
5288
5289
5290
5291 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5292 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5293 netpoll_poll_unlock(have_poll_lock);
5294 if (rc == BUSY_POLL_BUDGET)
5295 __napi_schedule(napi);
5296 local_bh_enable();
5297}
5298
5299void napi_busy_loop(unsigned int napi_id,
5300 bool (*loop_end)(void *, unsigned long),
5301 void *loop_end_arg)
5302{
5303 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
5304 int (*napi_poll)(struct napi_struct *napi, int budget);
5305 void *have_poll_lock = NULL;
5306 struct napi_struct *napi;
5307
5308restart:
5309 napi_poll = NULL;
5310
5311 rcu_read_lock();
5312
5313 napi = napi_by_id(napi_id);
5314 if (!napi)
5315 goto out;
5316
5317 preempt_disable();
5318 for (;;) {
5319 int work = 0;
5320
5321 local_bh_disable();
5322 if (!napi_poll) {
5323 unsigned long val = READ_ONCE(napi->state);
5324
5325
5326
5327
5328 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5329 NAPIF_STATE_IN_BUSY_POLL))
5330 goto count;
5331 if (cmpxchg(&napi->state, val,
5332 val | NAPIF_STATE_IN_BUSY_POLL |
5333 NAPIF_STATE_SCHED) != val)
5334 goto count;
5335 have_poll_lock = netpoll_poll_lock(napi);
5336 napi_poll = napi->poll;
5337 }
5338 work = napi_poll(napi, BUSY_POLL_BUDGET);
5339 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
5340count:
5341 if (work > 0)
5342 __NET_ADD_STATS(dev_net(napi->dev),
5343 LINUX_MIB_BUSYPOLLRXPACKETS, work);
5344 local_bh_enable();
5345
5346 if (!loop_end || loop_end(loop_end_arg, start_time))
5347 break;
5348
5349 if (unlikely(need_resched())) {
5350 if (napi_poll)
5351 busy_poll_stop(napi, have_poll_lock);
5352 preempt_enable();
5353 rcu_read_unlock();
5354 cond_resched();
5355 if (loop_end(loop_end_arg, start_time))
5356 return;
5357 goto restart;
5358 }
5359 cpu_relax();
5360 }
5361 if (napi_poll)
5362 busy_poll_stop(napi, have_poll_lock);
5363 preempt_enable();
5364out:
5365 rcu_read_unlock();
5366}
5367EXPORT_SYMBOL(napi_busy_loop);
5368
5369#endif
5370
5371static void napi_hash_add(struct napi_struct *napi)
5372{
5373 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5374 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5375 return;
5376
5377 spin_lock(&napi_hash_lock);
5378
5379
5380 do {
5381 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
5382 napi_gen_id = MIN_NAPI_ID;
5383 } while (napi_by_id(napi_gen_id));
5384 napi->napi_id = napi_gen_id;
5385
5386 hlist_add_head_rcu(&napi->napi_hash_node,
5387 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5388
5389 spin_unlock(&napi_hash_lock);
5390}
5391
5392
5393
5394
5395bool napi_hash_del(struct napi_struct *napi)
5396{
5397 bool rcu_sync_needed = false;
5398
5399 spin_lock(&napi_hash_lock);
5400
5401 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5402 rcu_sync_needed = true;
5403 hlist_del_rcu(&napi->napi_hash_node);
5404 }
5405 spin_unlock(&napi_hash_lock);
5406 return rcu_sync_needed;
5407}
5408EXPORT_SYMBOL_GPL(napi_hash_del);
5409
5410static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5411{
5412 struct napi_struct *napi;
5413
5414 napi = container_of(timer, struct napi_struct, timer);
5415
5416
5417
5418
5419 if (napi->gro_list && !napi_disable_pending(napi) &&
5420 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
5421 __napi_schedule_irqoff(napi);
5422
5423 return HRTIMER_NORESTART;
5424}
5425
5426void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5427 int (*poll)(struct napi_struct *, int), int weight)
5428{
5429 INIT_LIST_HEAD(&napi->poll_list);
5430 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5431 napi->timer.function = napi_watchdog;
5432 napi->gro_count = 0;
5433 napi->gro_list = NULL;
5434 napi->skb = NULL;
5435 napi->poll = poll;
5436 if (weight > NAPI_POLL_WEIGHT)
5437 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5438 weight, dev->name);
5439 napi->weight = weight;
5440 list_add(&napi->dev_list, &dev->napi_list);
5441 napi->dev = dev;
5442#ifdef CONFIG_NETPOLL
5443 napi->poll_owner = -1;
5444#endif
5445 set_bit(NAPI_STATE_SCHED, &napi->state);
5446 napi_hash_add(napi);
5447}
5448EXPORT_SYMBOL(netif_napi_add);
5449
5450void napi_disable(struct napi_struct *n)
5451{
5452 might_sleep();
5453 set_bit(NAPI_STATE_DISABLE, &n->state);
5454
5455 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5456 msleep(1);
5457 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5458 msleep(1);
5459
5460 hrtimer_cancel(&n->timer);
5461
5462 clear_bit(NAPI_STATE_DISABLE, &n->state);
5463}
5464EXPORT_SYMBOL(napi_disable);
5465
5466
5467void netif_napi_del(struct napi_struct *napi)
5468{
5469 might_sleep();
5470 if (napi_hash_del(napi))
5471 synchronize_net();
5472 list_del_init(&napi->dev_list);
5473 napi_free_frags(napi);
5474
5475 kfree_skb_list(napi->gro_list);
5476 napi->gro_list = NULL;
5477 napi->gro_count = 0;
5478}
5479EXPORT_SYMBOL(netif_napi_del);
5480
5481static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5482{
5483 void *have;
5484 int work, weight;
5485
5486 list_del_init(&n->poll_list);
5487
5488 have = netpoll_poll_lock(n);
5489
5490 weight = n->weight;
5491
5492
5493
5494
5495
5496
5497
5498 work = 0;
5499 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5500 work = n->poll(n, weight);
5501 trace_napi_poll(n, work, weight);
5502 }
5503
5504 WARN_ON_ONCE(work > weight);
5505
5506 if (likely(work < weight))
5507 goto out_unlock;
5508
5509
5510
5511
5512
5513
5514 if (unlikely(napi_disable_pending(n))) {
5515 napi_complete(n);
5516 goto out_unlock;
5517 }
5518
5519 if (n->gro_list) {
5520
5521
5522
5523 napi_gro_flush(n, HZ >= 1000);
5524 }
5525
5526
5527
5528
5529 if (unlikely(!list_empty(&n->poll_list))) {
5530 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5531 n->dev ? n->dev->name : "backlog");
5532 goto out_unlock;
5533 }
5534
5535 list_add_tail(&n->poll_list, repoll);
5536
5537out_unlock:
5538 netpoll_poll_unlock(have);
5539
5540 return work;
5541}
5542
5543static __latent_entropy void net_rx_action(struct softirq_action *h)
5544{
5545 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5546 unsigned long time_limit = jiffies +
5547 usecs_to_jiffies(netdev_budget_usecs);
5548 int budget = netdev_budget;
5549 LIST_HEAD(list);
5550 LIST_HEAD(repoll);
5551
5552 local_irq_disable();
5553 list_splice_init(&sd->poll_list, &list);
5554 local_irq_enable();
5555
5556 for (;;) {
5557 struct napi_struct *n;
5558
5559 if (list_empty(&list)) {
5560 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5561 goto out;
5562 break;
5563 }
5564
5565 n = list_first_entry(&list, struct napi_struct, poll_list);
5566 budget -= napi_poll(n, &repoll);
5567
5568
5569
5570
5571
5572 if (unlikely(budget <= 0 ||
5573 time_after_eq(jiffies, time_limit))) {
5574 sd->time_squeeze++;
5575 break;
5576 }
5577 }
5578
5579 local_irq_disable();
5580
5581 list_splice_tail_init(&sd->poll_list, &list);
5582 list_splice_tail(&repoll, &list);
5583 list_splice(&list, &sd->poll_list);
5584 if (!list_empty(&sd->poll_list))
5585 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5586
5587 net_rps_action_and_irq_enable(sd);
5588out:
5589 __kfree_skb_flush();
5590}
5591
5592struct netdev_adjacent {
5593 struct net_device *dev;
5594
5595
5596 bool master;
5597
5598
5599 u16 ref_nr;
5600
5601
5602 void *private;
5603
5604 struct list_head list;
5605 struct rcu_head rcu;
5606};
5607
5608static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5609 struct list_head *adj_list)
5610{
5611 struct netdev_adjacent *adj;
5612
5613 list_for_each_entry(adj, adj_list, list) {
5614 if (adj->dev == adj_dev)
5615 return adj;
5616 }
5617 return NULL;
5618}
5619
5620static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5621{
5622 struct net_device *dev = data;
5623
5624 return upper_dev == dev;
5625}
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636bool netdev_has_upper_dev(struct net_device *dev,
5637 struct net_device *upper_dev)
5638{
5639 ASSERT_RTNL();
5640
5641 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5642 upper_dev);
5643}
5644EXPORT_SYMBOL(netdev_has_upper_dev);
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5657 struct net_device *upper_dev)
5658{
5659 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5660 upper_dev);
5661}
5662EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5663
5664
5665
5666
5667
5668
5669
5670
5671bool netdev_has_any_upper_dev(struct net_device *dev)
5672{
5673 ASSERT_RTNL();
5674
5675 return !list_empty(&dev->adj_list.upper);
5676}
5677EXPORT_SYMBOL(netdev_has_any_upper_dev);
5678
5679
5680
5681
5682
5683
5684
5685
5686struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5687{
5688 struct netdev_adjacent *upper;
5689
5690 ASSERT_RTNL();
5691
5692 if (list_empty(&dev->adj_list.upper))
5693 return NULL;
5694
5695 upper = list_first_entry(&dev->adj_list.upper,
5696 struct netdev_adjacent, list);
5697 if (likely(upper->master))
5698 return upper->dev;
5699 return NULL;
5700}
5701EXPORT_SYMBOL(netdev_master_upper_dev_get);
5702
5703
5704
5705
5706
5707
5708
5709
5710static bool netdev_has_any_lower_dev(struct net_device *dev)
5711{
5712 ASSERT_RTNL();
5713
5714 return !list_empty(&dev->adj_list.lower);
5715}
5716
5717void *netdev_adjacent_get_private(struct list_head *adj_list)
5718{
5719 struct netdev_adjacent *adj;
5720
5721 adj = list_entry(adj_list, struct netdev_adjacent, list);
5722
5723 return adj->private;
5724}
5725EXPORT_SYMBOL(netdev_adjacent_get_private);
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5736 struct list_head **iter)
5737{
5738 struct netdev_adjacent *upper;
5739
5740 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5741
5742 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5743
5744 if (&upper->list == &dev->adj_list.upper)
5745 return NULL;
5746
5747 *iter = &upper->list;
5748
5749 return upper->dev;
5750}
5751EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5752
5753static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5754 struct list_head **iter)
5755{
5756 struct netdev_adjacent *upper;
5757
5758 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5759
5760 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5761
5762 if (&upper->list == &dev->adj_list.upper)
5763 return NULL;
5764
5765 *iter = &upper->list;
5766
5767 return upper->dev;
5768}
5769
5770int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5771 int (*fn)(struct net_device *dev,
5772 void *data),
5773 void *data)
5774{
5775 struct net_device *udev;
5776 struct list_head *iter;
5777 int ret;
5778
5779 for (iter = &dev->adj_list.upper,
5780 udev = netdev_next_upper_dev_rcu(dev, &iter);
5781 udev;
5782 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5783
5784 ret = fn(udev, data);
5785 if (ret)
5786 return ret;
5787
5788
5789 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5790 if (ret)
5791 return ret;
5792 }
5793
5794 return 0;
5795}
5796EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809void *netdev_lower_get_next_private(struct net_device *dev,
5810 struct list_head **iter)
5811{
5812 struct netdev_adjacent *lower;
5813
5814 lower = list_entry(*iter, struct netdev_adjacent, list);
5815
5816 if (&lower->list == &dev->adj_list.lower)
5817 return NULL;
5818
5819 *iter = lower->list.next;
5820
5821 return lower->private;
5822}
5823EXPORT_SYMBOL(netdev_lower_get_next_private);
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5836 struct list_head **iter)
5837{
5838 struct netdev_adjacent *lower;
5839
5840 WARN_ON_ONCE(!rcu_read_lock_held());
5841
5842 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5843
5844 if (&lower->list == &dev->adj_list.lower)
5845 return NULL;
5846
5847 *iter = &lower->list;
5848
5849 return lower->private;
5850}
5851EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5865{
5866 struct netdev_adjacent *lower;
5867
5868 lower = list_entry(*iter, struct netdev_adjacent, list);
5869
5870 if (&lower->list == &dev->adj_list.lower)
5871 return NULL;
5872
5873 *iter = lower->list.next;
5874
5875 return lower->dev;
5876}
5877EXPORT_SYMBOL(netdev_lower_get_next);
5878
5879static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5880 struct list_head **iter)
5881{
5882 struct netdev_adjacent *lower;
5883
5884 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5885
5886 if (&lower->list == &dev->adj_list.lower)
5887 return NULL;
5888
5889 *iter = &lower->list;
5890
5891 return lower->dev;
5892}
5893
5894int netdev_walk_all_lower_dev(struct net_device *dev,
5895 int (*fn)(struct net_device *dev,
5896 void *data),
5897 void *data)
5898{
5899 struct net_device *ldev;
5900 struct list_head *iter;
5901 int ret;
5902
5903 for (iter = &dev->adj_list.lower,
5904 ldev = netdev_next_lower_dev(dev, &iter);
5905 ldev;
5906 ldev = netdev_next_lower_dev(dev, &iter)) {
5907
5908 ret = fn(ldev, data);
5909 if (ret)
5910 return ret;
5911
5912
5913 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5914 if (ret)
5915 return ret;
5916 }
5917
5918 return 0;
5919}
5920EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5921
5922static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5923 struct list_head **iter)
5924{
5925 struct netdev_adjacent *lower;
5926
5927 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5928 if (&lower->list == &dev->adj_list.lower)
5929 return NULL;
5930
5931 *iter = &lower->list;
5932
5933 return lower->dev;
5934}
5935
5936int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5937 int (*fn)(struct net_device *dev,
5938 void *data),
5939 void *data)
5940{
5941 struct net_device *ldev;
5942 struct list_head *iter;
5943 int ret;
5944
5945 for (iter = &dev->adj_list.lower,
5946 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5947 ldev;
5948 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5949
5950 ret = fn(ldev, data);
5951 if (ret)
5952 return ret;
5953
5954
5955 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5956 if (ret)
5957 return ret;
5958 }
5959
5960 return 0;
5961}
5962EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5974{
5975 struct netdev_adjacent *lower;
5976
5977 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5978 struct netdev_adjacent, list);
5979 if (lower)
5980 return lower->private;
5981 return NULL;
5982}
5983EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5984
5985
5986
5987
5988
5989
5990
5991
5992struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5993{
5994 struct netdev_adjacent *upper;
5995
5996 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5997 struct netdev_adjacent, list);
5998 if (upper && likely(upper->master))
5999 return upper->dev;
6000 return NULL;
6001}
6002EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6003
6004static int netdev_adjacent_sysfs_add(struct net_device *dev,
6005 struct net_device *adj_dev,
6006 struct list_head *dev_list)
6007{
6008 char linkname[IFNAMSIZ+7];
6009
6010 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6011 "upper_%s" : "lower_%s", adj_dev->name);
6012 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6013 linkname);
6014}
6015static void netdev_adjacent_sysfs_del(struct net_device *dev,
6016 char *name,
6017 struct list_head *dev_list)
6018{
6019 char linkname[IFNAMSIZ+7];
6020
6021 sprintf(linkname, dev_list == &dev->adj_list.upper ?
6022 "upper_%s" : "lower_%s", name);
6023 sysfs_remove_link(&(dev->dev.kobj), linkname);
6024}
6025
6026static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6027 struct net_device *adj_dev,
6028 struct list_head *dev_list)
6029{
6030 return (dev_list == &dev->adj_list.upper ||
6031 dev_list == &dev->adj_list.lower) &&
6032 net_eq(dev_net(dev), dev_net(adj_dev));
6033}
6034
6035static int __netdev_adjacent_dev_insert(struct net_device *dev,
6036 struct net_device *adj_dev,
6037 struct list_head *dev_list,
6038 void *private, bool master)
6039{
6040 struct netdev_adjacent *adj;
6041 int ret;
6042
6043 adj = __netdev_find_adj(adj_dev, dev_list);
6044
6045 if (adj) {
6046 adj->ref_nr += 1;
6047 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6048 dev->name, adj_dev->name, adj->ref_nr);
6049
6050 return 0;
6051 }
6052
6053 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6054 if (!adj)
6055 return -ENOMEM;
6056
6057 adj->dev = adj_dev;
6058 adj->master = master;
6059 adj->ref_nr = 1;
6060 adj->private = private;
6061 dev_hold(adj_dev);
6062
6063 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6064 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6065
6066 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6067 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6068 if (ret)
6069 goto free_adj;
6070 }
6071
6072
6073 if (master) {
6074 ret = sysfs_create_link(&(dev->dev.kobj),
6075 &(adj_dev->dev.kobj), "master");
6076 if (ret)
6077 goto remove_symlinks;
6078
6079 list_add_rcu(&adj->list, dev_list);
6080 } else {
6081 list_add_tail_rcu(&adj->list, dev_list);
6082 }
6083
6084 return 0;
6085
6086remove_symlinks:
6087 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6088 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6089free_adj:
6090 kfree(adj);
6091 dev_put(adj_dev);
6092
6093 return ret;
6094}
6095
6096static void __netdev_adjacent_dev_remove(struct net_device *dev,
6097 struct net_device *adj_dev,
6098 u16 ref_nr,
6099 struct list_head *dev_list)
6100{
6101 struct netdev_adjacent *adj;
6102
6103 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6104 dev->name, adj_dev->name, ref_nr);
6105
6106 adj = __netdev_find_adj(adj_dev, dev_list);
6107
6108 if (!adj) {
6109 pr_err("Adjacency does not exist for device %s from %s\n",
6110 dev->name, adj_dev->name);
6111 WARN_ON(1);
6112 return;
6113 }
6114
6115 if (adj->ref_nr > ref_nr) {
6116 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6117 dev->name, adj_dev->name, ref_nr,
6118 adj->ref_nr - ref_nr);
6119 adj->ref_nr -= ref_nr;
6120 return;
6121 }
6122
6123 if (adj->master)
6124 sysfs_remove_link(&(dev->dev.kobj), "master");
6125
6126 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6127 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6128
6129 list_del_rcu(&adj->list);
6130 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6131 adj_dev->name, dev->name, adj_dev->name);
6132 dev_put(adj_dev);
6133 kfree_rcu(adj, rcu);
6134}
6135
6136static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6137 struct net_device *upper_dev,
6138 struct list_head *up_list,
6139 struct list_head *down_list,
6140 void *private, bool master)
6141{
6142 int ret;
6143
6144 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6145 private, master);
6146 if (ret)
6147 return ret;
6148
6149 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6150 private, false);
6151 if (ret) {
6152 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6153 return ret;
6154 }
6155
6156 return 0;
6157}
6158
6159static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6160 struct net_device *upper_dev,
6161 u16 ref_nr,
6162 struct list_head *up_list,
6163 struct list_head *down_list)
6164{
6165 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6166 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6167}
6168
6169static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6170 struct net_device *upper_dev,
6171 void *private, bool master)
6172{
6173 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6174 &dev->adj_list.upper,
6175 &upper_dev->adj_list.lower,
6176 private, master);
6177}
6178
6179static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6180 struct net_device *upper_dev)
6181{
6182 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
6183 &dev->adj_list.upper,
6184 &upper_dev->adj_list.lower);
6185}
6186
6187static int __netdev_upper_dev_link(struct net_device *dev,
6188 struct net_device *upper_dev, bool master,
6189 void *upper_priv, void *upper_info)
6190{
6191 struct netdev_notifier_changeupper_info changeupper_info;
6192 int ret = 0;
6193
6194 ASSERT_RTNL();
6195
6196 if (dev == upper_dev)
6197 return -EBUSY;
6198
6199
6200 if (netdev_has_upper_dev(upper_dev, dev))
6201 return -EBUSY;
6202
6203 if (netdev_has_upper_dev(dev, upper_dev))
6204 return -EEXIST;
6205
6206 if (master && netdev_master_upper_dev_get(dev))
6207 return -EBUSY;
6208
6209 changeupper_info.upper_dev = upper_dev;
6210 changeupper_info.master = master;
6211 changeupper_info.linking = true;
6212 changeupper_info.upper_info = upper_info;
6213
6214 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6215 &changeupper_info.info);
6216 ret = notifier_to_errno(ret);
6217 if (ret)
6218 return ret;
6219
6220 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
6221 master);
6222 if (ret)
6223 return ret;
6224
6225 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6226 &changeupper_info.info);
6227 ret = notifier_to_errno(ret);
6228 if (ret)
6229 goto rollback;
6230
6231 return 0;
6232
6233rollback:
6234 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6235
6236 return ret;
6237}
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249int netdev_upper_dev_link(struct net_device *dev,
6250 struct net_device *upper_dev)
6251{
6252 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
6253}
6254EXPORT_SYMBOL(netdev_upper_dev_link);
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269int netdev_master_upper_dev_link(struct net_device *dev,
6270 struct net_device *upper_dev,
6271 void *upper_priv, void *upper_info)
6272{
6273 return __netdev_upper_dev_link(dev, upper_dev, true,
6274 upper_priv, upper_info);
6275}
6276EXPORT_SYMBOL(netdev_master_upper_dev_link);
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286void netdev_upper_dev_unlink(struct net_device *dev,
6287 struct net_device *upper_dev)
6288{
6289 struct netdev_notifier_changeupper_info changeupper_info;
6290
6291 ASSERT_RTNL();
6292
6293 changeupper_info.upper_dev = upper_dev;
6294 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6295 changeupper_info.linking = false;
6296
6297 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6298 &changeupper_info.info);
6299
6300 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6301
6302 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6303 &changeupper_info.info);
6304}
6305EXPORT_SYMBOL(netdev_upper_dev_unlink);
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315void netdev_bonding_info_change(struct net_device *dev,
6316 struct netdev_bonding_info *bonding_info)
6317{
6318 struct netdev_notifier_bonding_info info;
6319
6320 memcpy(&info.bonding_info, bonding_info,
6321 sizeof(struct netdev_bonding_info));
6322 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6323 &info.info);
6324}
6325EXPORT_SYMBOL(netdev_bonding_info_change);
6326
6327static void netdev_adjacent_add_links(struct net_device *dev)
6328{
6329 struct netdev_adjacent *iter;
6330
6331 struct net *net = dev_net(dev);
6332
6333 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6334 if (!net_eq(net, dev_net(iter->dev)))
6335 continue;
6336 netdev_adjacent_sysfs_add(iter->dev, dev,
6337 &iter->dev->adj_list.lower);
6338 netdev_adjacent_sysfs_add(dev, iter->dev,
6339 &dev->adj_list.upper);
6340 }
6341
6342 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6343 if (!net_eq(net, dev_net(iter->dev)))
6344 continue;
6345 netdev_adjacent_sysfs_add(iter->dev, dev,
6346 &iter->dev->adj_list.upper);
6347 netdev_adjacent_sysfs_add(dev, iter->dev,
6348 &dev->adj_list.lower);
6349 }
6350}
6351
6352static void netdev_adjacent_del_links(struct net_device *dev)
6353{
6354 struct netdev_adjacent *iter;
6355
6356 struct net *net = dev_net(dev);
6357
6358 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6359 if (!net_eq(net, dev_net(iter->dev)))
6360 continue;
6361 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6362 &iter->dev->adj_list.lower);
6363 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6364 &dev->adj_list.upper);
6365 }
6366
6367 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6368 if (!net_eq(net, dev_net(iter->dev)))
6369 continue;
6370 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6371 &iter->dev->adj_list.upper);
6372 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6373 &dev->adj_list.lower);
6374 }
6375}
6376
6377void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6378{
6379 struct netdev_adjacent *iter;
6380
6381 struct net *net = dev_net(dev);
6382
6383 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6384 if (!net_eq(net, dev_net(iter->dev)))
6385 continue;
6386 netdev_adjacent_sysfs_del(iter->dev, oldname,
6387 &iter->dev->adj_list.lower);
6388 netdev_adjacent_sysfs_add(iter->dev, dev,
6389 &iter->dev->adj_list.lower);
6390 }
6391
6392 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6393 if (!net_eq(net, dev_net(iter->dev)))
6394 continue;
6395 netdev_adjacent_sysfs_del(iter->dev, oldname,
6396 &iter->dev->adj_list.upper);
6397 netdev_adjacent_sysfs_add(iter->dev, dev,
6398 &iter->dev->adj_list.upper);
6399 }
6400}
6401
6402void *netdev_lower_dev_get_private(struct net_device *dev,
6403 struct net_device *lower_dev)
6404{
6405 struct netdev_adjacent *lower;
6406
6407 if (!lower_dev)
6408 return NULL;
6409 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6410 if (!lower)
6411 return NULL;
6412
6413 return lower->private;
6414}
6415EXPORT_SYMBOL(netdev_lower_dev_get_private);
6416
6417
6418int dev_get_nest_level(struct net_device *dev)
6419{
6420 struct net_device *lower = NULL;
6421 struct list_head *iter;
6422 int max_nest = -1;
6423 int nest;
6424
6425 ASSERT_RTNL();
6426
6427 netdev_for_each_lower_dev(dev, lower, iter) {
6428 nest = dev_get_nest_level(lower);
6429 if (max_nest < nest)
6430 max_nest = nest;
6431 }
6432
6433 return max_nest + 1;
6434}
6435EXPORT_SYMBOL(dev_get_nest_level);
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445void netdev_lower_state_changed(struct net_device *lower_dev,
6446 void *lower_state_info)
6447{
6448 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6449
6450 ASSERT_RTNL();
6451 changelowerstate_info.lower_state_info = lower_state_info;
6452 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6453 &changelowerstate_info.info);
6454}
6455EXPORT_SYMBOL(netdev_lower_state_changed);
6456
6457static void dev_change_rx_flags(struct net_device *dev, int flags)
6458{
6459 const struct net_device_ops *ops = dev->netdev_ops;
6460
6461 if (ops->ndo_change_rx_flags)
6462 ops->ndo_change_rx_flags(dev, flags);
6463}
6464
6465static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6466{
6467 unsigned int old_flags = dev->flags;
6468 kuid_t uid;
6469 kgid_t gid;
6470
6471 ASSERT_RTNL();
6472
6473 dev->flags |= IFF_PROMISC;
6474 dev->promiscuity += inc;
6475 if (dev->promiscuity == 0) {
6476
6477
6478
6479
6480 if (inc < 0)
6481 dev->flags &= ~IFF_PROMISC;
6482 else {
6483 dev->promiscuity -= inc;
6484 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6485 dev->name);
6486 return -EOVERFLOW;
6487 }
6488 }
6489 if (dev->flags != old_flags) {
6490 pr_info("device %s %s promiscuous mode\n",
6491 dev->name,
6492 dev->flags & IFF_PROMISC ? "entered" : "left");
6493 if (audit_enabled) {
6494 current_uid_gid(&uid, &gid);
6495 audit_log(current->audit_context, GFP_ATOMIC,
6496 AUDIT_ANOM_PROMISCUOUS,
6497 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6498 dev->name, (dev->flags & IFF_PROMISC),
6499 (old_flags & IFF_PROMISC),
6500 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6501 from_kuid(&init_user_ns, uid),
6502 from_kgid(&init_user_ns, gid),
6503 audit_get_sessionid(current));
6504 }
6505
6506 dev_change_rx_flags(dev, IFF_PROMISC);
6507 }
6508 if (notify)
6509 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6510 return 0;
6511}
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524int dev_set_promiscuity(struct net_device *dev, int inc)
6525{
6526 unsigned int old_flags = dev->flags;
6527 int err;
6528
6529 err = __dev_set_promiscuity(dev, inc, true);
6530 if (err < 0)
6531 return err;
6532 if (dev->flags != old_flags)
6533 dev_set_rx_mode(dev);
6534 return err;
6535}
6536EXPORT_SYMBOL(dev_set_promiscuity);
6537
6538static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6539{
6540 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6541
6542 ASSERT_RTNL();
6543
6544 dev->flags |= IFF_ALLMULTI;
6545 dev->allmulti += inc;
6546 if (dev->allmulti == 0) {
6547
6548
6549
6550
6551 if (inc < 0)
6552 dev->flags &= ~IFF_ALLMULTI;
6553 else {
6554 dev->allmulti -= inc;
6555 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6556 dev->name);
6557 return -EOVERFLOW;
6558 }
6559 }
6560 if (dev->flags ^ old_flags) {
6561 dev_change_rx_flags(dev, IFF_ALLMULTI);
6562 dev_set_rx_mode(dev);
6563 if (notify)
6564 __dev_notify_flags(dev, old_flags,
6565 dev->gflags ^ old_gflags);
6566 }
6567 return 0;
6568}
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583int dev_set_allmulti(struct net_device *dev, int inc)
6584{
6585 return __dev_set_allmulti(dev, inc, true);
6586}
6587EXPORT_SYMBOL(dev_set_allmulti);
6588
6589
6590
6591
6592
6593
6594
6595void __dev_set_rx_mode(struct net_device *dev)
6596{
6597 const struct net_device_ops *ops = dev->netdev_ops;
6598
6599
6600 if (!(dev->flags&IFF_UP))
6601 return;
6602
6603 if (!netif_device_present(dev))
6604 return;
6605
6606 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6607
6608
6609
6610 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6611 __dev_set_promiscuity(dev, 1, false);
6612 dev->uc_promisc = true;
6613 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6614 __dev_set_promiscuity(dev, -1, false);
6615 dev->uc_promisc = false;
6616 }
6617 }
6618
6619 if (ops->ndo_set_rx_mode)
6620 ops->ndo_set_rx_mode(dev);
6621}
6622
6623void dev_set_rx_mode(struct net_device *dev)
6624{
6625 netif_addr_lock_bh(dev);
6626 __dev_set_rx_mode(dev);
6627 netif_addr_unlock_bh(dev);
6628}
6629
6630
6631
6632
6633
6634
6635
6636unsigned int dev_get_flags(const struct net_device *dev)
6637{
6638 unsigned int flags;
6639
6640 flags = (dev->flags & ~(IFF_PROMISC |
6641 IFF_ALLMULTI |
6642 IFF_RUNNING |
6643 IFF_LOWER_UP |
6644 IFF_DORMANT)) |
6645 (dev->gflags & (IFF_PROMISC |
6646 IFF_ALLMULTI));
6647
6648 if (netif_running(dev)) {
6649 if (netif_oper_up(dev))
6650 flags |= IFF_RUNNING;
6651 if (netif_carrier_ok(dev))
6652 flags |= IFF_LOWER_UP;
6653 if (netif_dormant(dev))
6654 flags |= IFF_DORMANT;
6655 }
6656
6657 return flags;
6658}
6659EXPORT_SYMBOL(dev_get_flags);
6660
6661int __dev_change_flags(struct net_device *dev, unsigned int flags)
6662{
6663 unsigned int old_flags = dev->flags;
6664 int ret;
6665
6666 ASSERT_RTNL();
6667
6668
6669
6670
6671
6672 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6673 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6674 IFF_AUTOMEDIA)) |
6675 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6676 IFF_ALLMULTI));
6677
6678
6679
6680
6681
6682 if ((old_flags ^ flags) & IFF_MULTICAST)
6683 dev_change_rx_flags(dev, IFF_MULTICAST);
6684
6685 dev_set_rx_mode(dev);
6686
6687
6688
6689
6690
6691
6692
6693 ret = 0;
6694 if ((old_flags ^ flags) & IFF_UP)
6695 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6696
6697 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6698 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6699 unsigned int old_flags = dev->flags;
6700
6701 dev->gflags ^= IFF_PROMISC;
6702
6703 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6704 if (dev->flags != old_flags)
6705 dev_set_rx_mode(dev);
6706 }
6707
6708
6709
6710
6711
6712 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6713 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6714
6715 dev->gflags ^= IFF_ALLMULTI;
6716 __dev_set_allmulti(dev, inc, false);
6717 }
6718
6719 return ret;
6720}
6721
6722void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6723 unsigned int gchanges)
6724{
6725 unsigned int changes = dev->flags ^ old_flags;
6726
6727 if (gchanges)
6728 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6729
6730 if (changes & IFF_UP) {
6731 if (dev->flags & IFF_UP)
6732 call_netdevice_notifiers(NETDEV_UP, dev);
6733 else
6734 call_netdevice_notifiers(NETDEV_DOWN, dev);
6735 }
6736
6737 if (dev->flags & IFF_UP &&
6738 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6739 struct netdev_notifier_change_info change_info;
6740
6741 change_info.flags_changed = changes;
6742 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6743 &change_info.info);
6744 }
6745}
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755int dev_change_flags(struct net_device *dev, unsigned int flags)
6756{
6757 int ret;
6758 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6759
6760 ret = __dev_change_flags(dev, flags);
6761 if (ret < 0)
6762 return ret;
6763
6764 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6765 __dev_notify_flags(dev, old_flags, changes);
6766 return ret;
6767}
6768EXPORT_SYMBOL(dev_change_flags);
6769
6770int __dev_set_mtu(struct net_device *dev, int new_mtu)
6771{
6772 const struct net_device_ops *ops = dev->netdev_ops;
6773
6774 if (ops->ndo_change_mtu)
6775 return ops->ndo_change_mtu(dev, new_mtu);
6776
6777 dev->mtu = new_mtu;
6778 return 0;
6779}
6780EXPORT_SYMBOL(__dev_set_mtu);
6781
6782
6783
6784
6785
6786
6787
6788
6789int dev_set_mtu(struct net_device *dev, int new_mtu)
6790{
6791 int err, orig_mtu;
6792
6793 if (new_mtu == dev->mtu)
6794 return 0;
6795
6796
6797 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6798 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6799 dev->name, new_mtu, dev->min_mtu);
6800 return -EINVAL;
6801 }
6802
6803 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6804 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6805 dev->name, new_mtu, dev->max_mtu);
6806 return -EINVAL;
6807 }
6808
6809 if (!netif_device_present(dev))
6810 return -ENODEV;
6811
6812 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6813 err = notifier_to_errno(err);
6814 if (err)
6815 return err;
6816
6817 orig_mtu = dev->mtu;
6818 err = __dev_set_mtu(dev, new_mtu);
6819
6820 if (!err) {
6821 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6822 err = notifier_to_errno(err);
6823 if (err) {
6824
6825
6826
6827 __dev_set_mtu(dev, orig_mtu);
6828 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6829 }
6830 }
6831 return err;
6832}
6833EXPORT_SYMBOL(dev_set_mtu);
6834
6835
6836
6837
6838
6839
6840void dev_set_group(struct net_device *dev, int new_group)
6841{
6842 dev->group = new_group;
6843}
6844EXPORT_SYMBOL(dev_set_group);
6845
6846
6847
6848
6849
6850
6851
6852
6853int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6854{
6855 const struct net_device_ops *ops = dev->netdev_ops;
6856 int err;
6857
6858 if (!ops->ndo_set_mac_address)
6859 return -EOPNOTSUPP;
6860 if (sa->sa_family != dev->type)
6861 return -EINVAL;
6862 if (!netif_device_present(dev))
6863 return -ENODEV;
6864 err = ops->ndo_set_mac_address(dev, sa);
6865 if (err)
6866 return err;
6867 dev->addr_assign_type = NET_ADDR_SET;
6868 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6869 add_device_randomness(dev->dev_addr, dev->addr_len);
6870 return 0;
6871}
6872EXPORT_SYMBOL(dev_set_mac_address);
6873
6874
6875
6876
6877
6878
6879
6880
6881int dev_change_carrier(struct net_device *dev, bool new_carrier)
6882{
6883 const struct net_device_ops *ops = dev->netdev_ops;
6884
6885 if (!ops->ndo_change_carrier)
6886 return -EOPNOTSUPP;
6887 if (!netif_device_present(dev))
6888 return -ENODEV;
6889 return ops->ndo_change_carrier(dev, new_carrier);
6890}
6891EXPORT_SYMBOL(dev_change_carrier);
6892
6893
6894
6895
6896
6897
6898
6899
6900int dev_get_phys_port_id(struct net_device *dev,
6901 struct netdev_phys_item_id *ppid)
6902{
6903 const struct net_device_ops *ops = dev->netdev_ops;
6904
6905 if (!ops->ndo_get_phys_port_id)
6906 return -EOPNOTSUPP;
6907 return ops->ndo_get_phys_port_id(dev, ppid);
6908}
6909EXPORT_SYMBOL(dev_get_phys_port_id);
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919int dev_get_phys_port_name(struct net_device *dev,
6920 char *name, size_t len)
6921{
6922 const struct net_device_ops *ops = dev->netdev_ops;
6923
6924 if (!ops->ndo_get_phys_port_name)
6925 return -EOPNOTSUPP;
6926 return ops->ndo_get_phys_port_name(dev, name, len);
6927}
6928EXPORT_SYMBOL(dev_get_phys_port_name);
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938int dev_change_proto_down(struct net_device *dev, bool proto_down)
6939{
6940 const struct net_device_ops *ops = dev->netdev_ops;
6941
6942 if (!ops->ndo_change_proto_down)
6943 return -EOPNOTSUPP;
6944 if (!netif_device_present(dev))
6945 return -ENODEV;
6946 return ops->ndo_change_proto_down(dev, proto_down);
6947}
6948EXPORT_SYMBOL(dev_change_proto_down);
6949
6950u8 __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op, u32 *prog_id)
6951{
6952 struct netdev_xdp xdp;
6953
6954 memset(&xdp, 0, sizeof(xdp));
6955 xdp.command = XDP_QUERY_PROG;
6956
6957
6958 WARN_ON(xdp_op(dev, &xdp) < 0);
6959 if (prog_id)
6960 *prog_id = xdp.prog_id;
6961
6962 return xdp.prog_attached;
6963}
6964
6965static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op,
6966 struct netlink_ext_ack *extack, u32 flags,
6967 struct bpf_prog *prog)
6968{
6969 struct netdev_xdp xdp;
6970
6971 memset(&xdp, 0, sizeof(xdp));
6972 if (flags & XDP_FLAGS_HW_MODE)
6973 xdp.command = XDP_SETUP_PROG_HW;
6974 else
6975 xdp.command = XDP_SETUP_PROG;
6976 xdp.extack = extack;
6977 xdp.flags = flags;
6978 xdp.prog = prog;
6979
6980 return xdp_op(dev, &xdp);
6981}
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
6993 int fd, u32 flags)
6994{
6995 const struct net_device_ops *ops = dev->netdev_ops;
6996 struct bpf_prog *prog = NULL;
6997 xdp_op_t xdp_op, xdp_chk;
6998 int err;
6999
7000 ASSERT_RTNL();
7001
7002 xdp_op = xdp_chk = ops->ndo_xdp;
7003 if (!xdp_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7004 return -EOPNOTSUPP;
7005 if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE))
7006 xdp_op = generic_xdp_install;
7007 if (xdp_op == xdp_chk)
7008 xdp_chk = generic_xdp_install;
7009
7010 if (fd >= 0) {
7011 if (xdp_chk && __dev_xdp_attached(dev, xdp_chk, NULL))
7012 return -EEXIST;
7013 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7014 __dev_xdp_attached(dev, xdp_op, NULL))
7015 return -EBUSY;
7016
7017 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
7018 if (IS_ERR(prog))
7019 return PTR_ERR(prog);
7020 }
7021
7022 err = dev_xdp_install(dev, xdp_op, extack, flags, prog);
7023 if (err < 0 && prog)
7024 bpf_prog_put(prog);
7025
7026 return err;
7027}
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037static int dev_new_index(struct net *net)
7038{
7039 int ifindex = net->ifindex;
7040
7041 for (;;) {
7042 if (++ifindex <= 0)
7043 ifindex = 1;
7044 if (!__dev_get_by_index(net, ifindex))
7045 return net->ifindex = ifindex;
7046 }
7047}
7048
7049
7050static LIST_HEAD(net_todo_list);
7051DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7052
7053static void net_set_todo(struct net_device *dev)
7054{
7055 list_add_tail(&dev->todo_list, &net_todo_list);
7056 dev_net(dev)->dev_unreg_count++;
7057}
7058
7059static void rollback_registered_many(struct list_head *head)
7060{
7061 struct net_device *dev, *tmp;
7062 LIST_HEAD(close_head);
7063
7064 BUG_ON(dev_boot_phase);
7065 ASSERT_RTNL();
7066
7067 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
7068
7069
7070
7071
7072 if (dev->reg_state == NETREG_UNINITIALIZED) {
7073 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7074 dev->name, dev);
7075
7076 WARN_ON(1);
7077 list_del(&dev->unreg_list);
7078 continue;
7079 }
7080 dev->dismantle = true;
7081 BUG_ON(dev->reg_state != NETREG_REGISTERED);
7082 }
7083
7084
7085 list_for_each_entry(dev, head, unreg_list)
7086 list_add_tail(&dev->close_list, &close_head);
7087 dev_close_many(&close_head, true);
7088
7089 list_for_each_entry(dev, head, unreg_list) {
7090
7091 unlist_netdevice(dev);
7092
7093 dev->reg_state = NETREG_UNREGISTERING;
7094 }
7095 flush_all_backlogs();
7096
7097 synchronize_net();
7098
7099 list_for_each_entry(dev, head, unreg_list) {
7100 struct sk_buff *skb = NULL;
7101
7102
7103 dev_shutdown(dev);
7104
7105
7106
7107
7108
7109 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7110
7111 if (!dev->rtnl_link_ops ||
7112 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7113 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
7114 GFP_KERNEL);
7115
7116
7117
7118
7119 dev_uc_flush(dev);
7120 dev_mc_flush(dev);
7121
7122 if (dev->netdev_ops->ndo_uninit)
7123 dev->netdev_ops->ndo_uninit(dev);
7124
7125 if (skb)
7126 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
7127
7128
7129 WARN_ON(netdev_has_any_upper_dev(dev));
7130 WARN_ON(netdev_has_any_lower_dev(dev));
7131
7132
7133 netdev_unregister_kobject(dev);
7134#ifdef CONFIG_XPS
7135
7136 netif_reset_xps_queues_gt(dev, 0);
7137#endif
7138 }
7139
7140 synchronize_net();
7141
7142 list_for_each_entry(dev, head, unreg_list)
7143 dev_put(dev);
7144}
7145
7146static void rollback_registered(struct net_device *dev)
7147{
7148 LIST_HEAD(single);
7149
7150 list_add(&dev->unreg_list, &single);
7151 rollback_registered_many(&single);
7152 list_del(&single);
7153}
7154
7155static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
7156 struct net_device *upper, netdev_features_t features)
7157{
7158 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7159 netdev_features_t feature;
7160 int feature_bit;
7161
7162 for_each_netdev_feature(&upper_disables, feature_bit) {
7163 feature = __NETIF_F_BIT(feature_bit);
7164 if (!(upper->wanted_features & feature)
7165 && (features & feature)) {
7166 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
7167 &feature, upper->name);
7168 features &= ~feature;
7169 }
7170 }
7171
7172 return features;
7173}
7174
7175static void netdev_sync_lower_features(struct net_device *upper,
7176 struct net_device *lower, netdev_features_t features)
7177{
7178 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7179 netdev_features_t feature;
7180 int feature_bit;
7181
7182 for_each_netdev_feature(&upper_disables, feature_bit) {
7183 feature = __NETIF_F_BIT(feature_bit);
7184 if (!(features & feature) && (lower->features & feature)) {
7185 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
7186 &feature, lower->name);
7187 lower->wanted_features &= ~feature;
7188 netdev_update_features(lower);
7189
7190 if (unlikely(lower->features & feature))
7191 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
7192 &feature, lower->name);
7193 }
7194 }
7195}
7196
7197static netdev_features_t netdev_fix_features(struct net_device *dev,
7198 netdev_features_t features)
7199{
7200
7201 if ((features & NETIF_F_HW_CSUM) &&
7202 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
7203 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
7204 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
7205 }
7206
7207
7208 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
7209 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
7210 features &= ~NETIF_F_ALL_TSO;
7211 }
7212
7213 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
7214 !(features & NETIF_F_IP_CSUM)) {
7215 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
7216 features &= ~NETIF_F_TSO;
7217 features &= ~NETIF_F_TSO_ECN;
7218 }
7219
7220 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
7221 !(features & NETIF_F_IPV6_CSUM)) {
7222 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
7223 features &= ~NETIF_F_TSO6;
7224 }
7225
7226
7227 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
7228 features &= ~NETIF_F_TSO_MANGLEID;
7229
7230
7231 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
7232 features &= ~NETIF_F_TSO_ECN;
7233
7234
7235 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
7236 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
7237 features &= ~NETIF_F_GSO;
7238 }
7239
7240
7241 if (features & NETIF_F_UFO) {
7242
7243 if (!(features & NETIF_F_HW_CSUM) &&
7244 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
7245 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
7246 netdev_dbg(dev,
7247 "Dropping NETIF_F_UFO since no checksum offload features.\n");
7248 features &= ~NETIF_F_UFO;
7249 }
7250
7251 if (!(features & NETIF_F_SG)) {
7252 netdev_dbg(dev,
7253 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
7254 features &= ~NETIF_F_UFO;
7255 }
7256 }
7257
7258
7259 if ((features & dev->gso_partial_features) &&
7260 !(features & NETIF_F_GSO_PARTIAL)) {
7261 netdev_dbg(dev,
7262 "Dropping partially supported GSO features since no GSO partial.\n");
7263 features &= ~dev->gso_partial_features;
7264 }
7265
7266 return features;
7267}
7268
7269int __netdev_update_features(struct net_device *dev)
7270{
7271 struct net_device *upper, *lower;
7272 netdev_features_t features;
7273 struct list_head *iter;
7274 int err = -1;
7275
7276 ASSERT_RTNL();
7277
7278 features = netdev_get_wanted_features(dev);
7279
7280 if (dev->netdev_ops->ndo_fix_features)
7281 features = dev->netdev_ops->ndo_fix_features(dev, features);
7282
7283
7284 features = netdev_fix_features(dev, features);
7285
7286
7287 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7288 features = netdev_sync_upper_features(dev, upper, features);
7289
7290 if (dev->features == features)
7291 goto sync_lower;
7292
7293 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7294 &dev->features, &features);
7295
7296 if (dev->netdev_ops->ndo_set_features)
7297 err = dev->netdev_ops->ndo_set_features(dev, features);
7298 else
7299 err = 0;
7300
7301 if (unlikely(err < 0)) {
7302 netdev_err(dev,
7303 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7304 err, &features, &dev->features);
7305
7306
7307
7308 return -1;
7309 }
7310
7311sync_lower:
7312
7313
7314
7315 netdev_for_each_lower_dev(dev, lower, iter)
7316 netdev_sync_lower_features(dev, lower, features);
7317
7318 if (!err)
7319 dev->features = features;
7320
7321 return err < 0 ? 0 : 1;
7322}
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332void netdev_update_features(struct net_device *dev)
7333{
7334 if (__netdev_update_features(dev))
7335 netdev_features_change(dev);
7336}
7337EXPORT_SYMBOL(netdev_update_features);
7338
7339
7340
7341
7342
7343
7344
7345
7346
7347
7348
7349void netdev_change_features(struct net_device *dev)
7350{
7351 __netdev_update_features(dev);
7352 netdev_features_change(dev);
7353}
7354EXPORT_SYMBOL(netdev_change_features);
7355
7356
7357
7358
7359
7360
7361
7362
7363
7364
7365void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7366 struct net_device *dev)
7367{
7368 if (rootdev->operstate == IF_OPER_DORMANT)
7369 netif_dormant_on(dev);
7370 else
7371 netif_dormant_off(dev);
7372
7373 if (netif_carrier_ok(rootdev))
7374 netif_carrier_on(dev);
7375 else
7376 netif_carrier_off(dev);
7377}
7378EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7379
7380#ifdef CONFIG_SYSFS
7381static int netif_alloc_rx_queues(struct net_device *dev)
7382{
7383 unsigned int i, count = dev->num_rx_queues;
7384 struct netdev_rx_queue *rx;
7385 size_t sz = count * sizeof(*rx);
7386
7387 BUG_ON(count < 1);
7388
7389 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7390 if (!rx)
7391 return -ENOMEM;
7392
7393 dev->_rx = rx;
7394
7395 for (i = 0; i < count; i++)
7396 rx[i].dev = dev;
7397 return 0;
7398}
7399#endif
7400
7401static void netdev_init_one_queue(struct net_device *dev,
7402 struct netdev_queue *queue, void *_unused)
7403{
7404
7405 spin_lock_init(&queue->_xmit_lock);
7406 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7407 queue->xmit_lock_owner = -1;
7408 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7409 queue->dev = dev;
7410#ifdef CONFIG_BQL
7411 dql_init(&queue->dql, HZ);
7412#endif
7413}
7414
7415static void netif_free_tx_queues(struct net_device *dev)
7416{
7417 kvfree(dev->_tx);
7418}
7419
7420static int netif_alloc_netdev_queues(struct net_device *dev)
7421{
7422 unsigned int count = dev->num_tx_queues;
7423 struct netdev_queue *tx;
7424 size_t sz = count * sizeof(*tx);
7425
7426 if (count < 1 || count > 0xffff)
7427 return -EINVAL;
7428
7429 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7430 if (!tx)
7431 return -ENOMEM;
7432
7433 dev->_tx = tx;
7434
7435 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7436 spin_lock_init(&dev->tx_global_lock);
7437
7438 return 0;
7439}
7440
7441void netif_tx_stop_all_queues(struct net_device *dev)
7442{
7443 unsigned int i;
7444
7445 for (i = 0; i < dev->num_tx_queues; i++) {
7446 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7447
7448 netif_tx_stop_queue(txq);
7449 }
7450}
7451EXPORT_SYMBOL(netif_tx_stop_all_queues);
7452
7453
7454
7455
7456
7457
7458
7459
7460
7461
7462
7463
7464
7465
7466
7467
7468
7469
7470int register_netdevice(struct net_device *dev)
7471{
7472 int ret;
7473 struct net *net = dev_net(dev);
7474
7475 BUG_ON(dev_boot_phase);
7476 ASSERT_RTNL();
7477
7478 might_sleep();
7479
7480
7481 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7482 BUG_ON(!net);
7483
7484 spin_lock_init(&dev->addr_list_lock);
7485 netdev_set_addr_lockdep_class(dev);
7486
7487 ret = dev_get_valid_name(net, dev, dev->name);
7488 if (ret < 0)
7489 goto out;
7490
7491
7492 if (dev->netdev_ops->ndo_init) {
7493 ret = dev->netdev_ops->ndo_init(dev);
7494 if (ret) {
7495 if (ret > 0)
7496 ret = -EIO;
7497 goto out;
7498 }
7499 }
7500
7501 if (((dev->hw_features | dev->features) &
7502 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7503 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7504 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7505 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7506 ret = -EINVAL;
7507 goto err_uninit;
7508 }
7509
7510 ret = -EBUSY;
7511 if (!dev->ifindex)
7512 dev->ifindex = dev_new_index(net);
7513 else if (__dev_get_by_index(net, dev->ifindex))
7514 goto err_uninit;
7515
7516
7517
7518
7519 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7520 dev->features |= NETIF_F_SOFT_FEATURES;
7521 dev->wanted_features = dev->features & dev->hw_features;
7522
7523 if (!(dev->flags & IFF_LOOPBACK))
7524 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7525
7526
7527
7528
7529
7530
7531 if (dev->hw_features & NETIF_F_TSO)
7532 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7533 if (dev->vlan_features & NETIF_F_TSO)
7534 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7535 if (dev->mpls_features & NETIF_F_TSO)
7536 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7537 if (dev->hw_enc_features & NETIF_F_TSO)
7538 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7539
7540
7541
7542 dev->vlan_features |= NETIF_F_HIGHDMA;
7543
7544
7545
7546 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7547
7548
7549
7550 dev->mpls_features |= NETIF_F_SG;
7551
7552 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7553 ret = notifier_to_errno(ret);
7554 if (ret)
7555 goto err_uninit;
7556
7557 ret = netdev_register_kobject(dev);
7558 if (ret)
7559 goto err_uninit;
7560 dev->reg_state = NETREG_REGISTERED;
7561
7562 __netdev_update_features(dev);
7563
7564
7565
7566
7567
7568
7569 set_bit(__LINK_STATE_PRESENT, &dev->state);
7570
7571 linkwatch_init_dev(dev);
7572
7573 dev_init_scheduler(dev);
7574 dev_hold(dev);
7575 list_netdevice(dev);
7576 add_device_randomness(dev->dev_addr, dev->addr_len);
7577
7578
7579
7580
7581
7582 if (dev->addr_assign_type == NET_ADDR_PERM)
7583 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7584
7585
7586 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7587 ret = notifier_to_errno(ret);
7588 if (ret) {
7589 rollback_registered(dev);
7590 dev->reg_state = NETREG_UNREGISTERED;
7591 }
7592
7593
7594
7595
7596 if (!dev->rtnl_link_ops ||
7597 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7598 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7599
7600out:
7601 return ret;
7602
7603err_uninit:
7604 if (dev->netdev_ops->ndo_uninit)
7605 dev->netdev_ops->ndo_uninit(dev);
7606 if (dev->priv_destructor)
7607 dev->priv_destructor(dev);
7608 goto out;
7609}
7610EXPORT_SYMBOL(register_netdevice);
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622int init_dummy_netdev(struct net_device *dev)
7623{
7624
7625
7626
7627
7628
7629 memset(dev, 0, sizeof(struct net_device));
7630
7631
7632
7633
7634 dev->reg_state = NETREG_DUMMY;
7635
7636
7637 INIT_LIST_HEAD(&dev->napi_list);
7638
7639
7640 set_bit(__LINK_STATE_PRESENT, &dev->state);
7641 set_bit(__LINK_STATE_START, &dev->state);
7642
7643
7644
7645
7646
7647
7648 return 0;
7649}
7650EXPORT_SYMBOL_GPL(init_dummy_netdev);
7651
7652
7653
7654
7655
7656
7657
7658
7659
7660
7661
7662
7663
7664
7665
7666int register_netdev(struct net_device *dev)
7667{
7668 int err;
7669
7670 rtnl_lock();
7671 err = register_netdevice(dev);
7672 rtnl_unlock();
7673 return err;
7674}
7675EXPORT_SYMBOL(register_netdev);
7676
7677int netdev_refcnt_read(const struct net_device *dev)
7678{
7679 int i, refcnt = 0;
7680
7681 for_each_possible_cpu(i)
7682 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7683 return refcnt;
7684}
7685EXPORT_SYMBOL(netdev_refcnt_read);
7686
7687
7688
7689
7690
7691
7692
7693
7694
7695
7696
7697
7698
7699static void netdev_wait_allrefs(struct net_device *dev)
7700{
7701 unsigned long rebroadcast_time, warning_time;
7702 int refcnt;
7703
7704 linkwatch_forget_dev(dev);
7705
7706 rebroadcast_time = warning_time = jiffies;
7707 refcnt = netdev_refcnt_read(dev);
7708
7709 while (refcnt != 0) {
7710 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7711 rtnl_lock();
7712
7713
7714 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7715
7716 __rtnl_unlock();
7717 rcu_barrier();
7718 rtnl_lock();
7719
7720 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7721 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7722 &dev->state)) {
7723
7724
7725
7726
7727
7728
7729 linkwatch_run_queue();
7730 }
7731
7732 __rtnl_unlock();
7733
7734 rebroadcast_time = jiffies;
7735 }
7736
7737 msleep(250);
7738
7739 refcnt = netdev_refcnt_read(dev);
7740
7741 if (time_after(jiffies, warning_time + 10 * HZ)) {
7742 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7743 dev->name, refcnt);
7744 warning_time = jiffies;
7745 }
7746 }
7747}
7748
7749
7750
7751
7752
7753
7754
7755
7756
7757
7758
7759
7760
7761
7762
7763
7764
7765
7766
7767
7768
7769
7770
7771
7772
7773void netdev_run_todo(void)
7774{
7775 struct list_head list;
7776
7777
7778 list_replace_init(&net_todo_list, &list);
7779
7780 __rtnl_unlock();
7781
7782
7783
7784 if (!list_empty(&list))
7785 rcu_barrier();
7786
7787 while (!list_empty(&list)) {
7788 struct net_device *dev
7789 = list_first_entry(&list, struct net_device, todo_list);
7790 list_del(&dev->todo_list);
7791
7792 rtnl_lock();
7793 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7794 __rtnl_unlock();
7795
7796 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7797 pr_err("network todo '%s' but state %d\n",
7798 dev->name, dev->reg_state);
7799 dump_stack();
7800 continue;
7801 }
7802
7803 dev->reg_state = NETREG_UNREGISTERED;
7804
7805 netdev_wait_allrefs(dev);
7806
7807
7808 BUG_ON(netdev_refcnt_read(dev));
7809 BUG_ON(!list_empty(&dev->ptype_all));
7810 BUG_ON(!list_empty(&dev->ptype_specific));
7811 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7812 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7813 WARN_ON(dev->dn_ptr);
7814
7815 if (dev->priv_destructor)
7816 dev->priv_destructor(dev);
7817 if (dev->needs_free_netdev)
7818 free_netdev(dev);
7819
7820
7821 rtnl_lock();
7822 dev_net(dev)->dev_unreg_count--;
7823 __rtnl_unlock();
7824 wake_up(&netdev_unregistering_wq);
7825
7826
7827 kobject_put(&dev->dev.kobj);
7828 }
7829}
7830
7831
7832
7833
7834
7835
7836void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7837 const struct net_device_stats *netdev_stats)
7838{
7839#if BITS_PER_LONG == 64
7840 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7841 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
7842
7843 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7844 sizeof(*stats64) - sizeof(*netdev_stats));
7845#else
7846 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7847 const unsigned long *src = (const unsigned long *)netdev_stats;
7848 u64 *dst = (u64 *)stats64;
7849
7850 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7851 for (i = 0; i < n; i++)
7852 dst[i] = src[i];
7853
7854 memset((char *)stats64 + n * sizeof(u64), 0,
7855 sizeof(*stats64) - n * sizeof(u64));
7856#endif
7857}
7858EXPORT_SYMBOL(netdev_stats_to_stats64);
7859
7860
7861
7862
7863
7864
7865
7866
7867
7868
7869
7870struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7871 struct rtnl_link_stats64 *storage)
7872{
7873 const struct net_device_ops *ops = dev->netdev_ops;
7874
7875 if (ops->ndo_get_stats64) {
7876 memset(storage, 0, sizeof(*storage));
7877 ops->ndo_get_stats64(dev, storage);
7878 } else if (ops->ndo_get_stats) {
7879 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7880 } else {
7881 netdev_stats_to_stats64(storage, &dev->stats);
7882 }
7883 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
7884 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
7885 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
7886 return storage;
7887}
7888EXPORT_SYMBOL(dev_get_stats);
7889
7890struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7891{
7892 struct netdev_queue *queue = dev_ingress_queue(dev);
7893
7894#ifdef CONFIG_NET_CLS_ACT
7895 if (queue)
7896 return queue;
7897 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7898 if (!queue)
7899 return NULL;
7900 netdev_init_one_queue(dev, queue, NULL);
7901 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7902 queue->qdisc_sleeping = &noop_qdisc;
7903 rcu_assign_pointer(dev->ingress_queue, queue);
7904#endif
7905 return queue;
7906}
7907
7908static const struct ethtool_ops default_ethtool_ops;
7909
7910void netdev_set_default_ethtool_ops(struct net_device *dev,
7911 const struct ethtool_ops *ops)
7912{
7913 if (dev->ethtool_ops == &default_ethtool_ops)
7914 dev->ethtool_ops = ops;
7915}
7916EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7917
7918void netdev_freemem(struct net_device *dev)
7919{
7920 char *addr = (char *)dev - dev->padded;
7921
7922 kvfree(addr);
7923}
7924
7925
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7939 unsigned char name_assign_type,
7940 void (*setup)(struct net_device *),
7941 unsigned int txqs, unsigned int rxqs)
7942{
7943 struct net_device *dev;
7944 size_t alloc_size;
7945 struct net_device *p;
7946
7947 BUG_ON(strlen(name) >= sizeof(dev->name));
7948
7949 if (txqs < 1) {
7950 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7951 return NULL;
7952 }
7953
7954#ifdef CONFIG_SYSFS
7955 if (rxqs < 1) {
7956 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7957 return NULL;
7958 }
7959#endif
7960
7961 alloc_size = sizeof(struct net_device);
7962 if (sizeof_priv) {
7963
7964 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7965 alloc_size += sizeof_priv;
7966 }
7967
7968 alloc_size += NETDEV_ALIGN - 1;
7969
7970 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7971 if (!p)
7972 return NULL;
7973
7974 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7975 dev->padded = (char *)dev - (char *)p;
7976
7977 dev->pcpu_refcnt = alloc_percpu(int);
7978 if (!dev->pcpu_refcnt)
7979 goto free_dev;
7980
7981 if (dev_addr_init(dev))
7982 goto free_pcpu;
7983
7984 dev_mc_init(dev);
7985 dev_uc_init(dev);
7986
7987 dev_net_set(dev, &init_net);
7988
7989 dev->gso_max_size = GSO_MAX_SIZE;
7990 dev->gso_max_segs = GSO_MAX_SEGS;
7991
7992 INIT_LIST_HEAD(&dev->napi_list);
7993 INIT_LIST_HEAD(&dev->unreg_list);
7994 INIT_LIST_HEAD(&dev->close_list);
7995 INIT_LIST_HEAD(&dev->link_watch_list);
7996 INIT_LIST_HEAD(&dev->adj_list.upper);
7997 INIT_LIST_HEAD(&dev->adj_list.lower);
7998 INIT_LIST_HEAD(&dev->ptype_all);
7999 INIT_LIST_HEAD(&dev->ptype_specific);
8000#ifdef CONFIG_NET_SCHED
8001 hash_init(dev->qdisc_hash);
8002#endif
8003 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
8004 setup(dev);
8005
8006 if (!dev->tx_queue_len) {
8007 dev->priv_flags |= IFF_NO_QUEUE;
8008 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
8009 }
8010
8011 dev->num_tx_queues = txqs;
8012 dev->real_num_tx_queues = txqs;
8013 if (netif_alloc_netdev_queues(dev))
8014 goto free_all;
8015
8016#ifdef CONFIG_SYSFS
8017 dev->num_rx_queues = rxqs;
8018 dev->real_num_rx_queues = rxqs;
8019 if (netif_alloc_rx_queues(dev))
8020 goto free_all;
8021#endif
8022
8023 strcpy(dev->name, name);
8024 dev->name_assign_type = name_assign_type;
8025 dev->group = INIT_NETDEV_GROUP;
8026 if (!dev->ethtool_ops)
8027 dev->ethtool_ops = &default_ethtool_ops;
8028
8029 nf_hook_ingress_init(dev);
8030
8031 return dev;
8032
8033free_all:
8034 free_netdev(dev);
8035 return NULL;
8036
8037free_pcpu:
8038 free_percpu(dev->pcpu_refcnt);
8039free_dev:
8040 netdev_freemem(dev);
8041 return NULL;
8042}
8043EXPORT_SYMBOL(alloc_netdev_mqs);
8044
8045
8046
8047
8048
8049
8050
8051
8052
8053
8054void free_netdev(struct net_device *dev)
8055{
8056 struct napi_struct *p, *n;
8057 struct bpf_prog *prog;
8058
8059 might_sleep();
8060 netif_free_tx_queues(dev);
8061#ifdef CONFIG_SYSFS
8062 kvfree(dev->_rx);
8063#endif
8064
8065 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
8066
8067
8068 dev_addr_flush(dev);
8069
8070 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
8071 netif_napi_del(p);
8072
8073 free_percpu(dev->pcpu_refcnt);
8074 dev->pcpu_refcnt = NULL;
8075
8076 prog = rcu_dereference_protected(dev->xdp_prog, 1);
8077 if (prog) {
8078 bpf_prog_put(prog);
8079 static_key_slow_dec(&generic_xdp_needed);
8080 }
8081
8082
8083 if (dev->reg_state == NETREG_UNINITIALIZED) {
8084 netdev_freemem(dev);
8085 return;
8086 }
8087
8088 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
8089 dev->reg_state = NETREG_RELEASED;
8090
8091
8092 put_device(&dev->dev);
8093}
8094EXPORT_SYMBOL(free_netdev);
8095
8096
8097
8098
8099
8100
8101
8102void synchronize_net(void)
8103{
8104 might_sleep();
8105 if (rtnl_is_locked())
8106 synchronize_rcu_expedited();
8107 else
8108 synchronize_rcu();
8109}
8110EXPORT_SYMBOL(synchronize_net);
8111
8112
8113
8114
8115
8116
8117
8118
8119
8120
8121
8122
8123
8124
8125void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
8126{
8127 ASSERT_RTNL();
8128
8129 if (head) {
8130 list_move_tail(&dev->unreg_list, head);
8131 } else {
8132 rollback_registered(dev);
8133
8134 net_set_todo(dev);
8135 }
8136}
8137EXPORT_SYMBOL(unregister_netdevice_queue);
8138
8139
8140
8141
8142
8143
8144
8145
8146void unregister_netdevice_many(struct list_head *head)
8147{
8148 struct net_device *dev;
8149
8150 if (!list_empty(head)) {
8151 rollback_registered_many(head);
8152 list_for_each_entry(dev, head, unreg_list)
8153 net_set_todo(dev);
8154 list_del(head);
8155 }
8156}
8157EXPORT_SYMBOL(unregister_netdevice_many);
8158
8159
8160
8161
8162
8163
8164
8165
8166
8167
8168
8169
8170void unregister_netdev(struct net_device *dev)
8171{
8172 rtnl_lock();
8173 unregister_netdevice(dev);
8174 rtnl_unlock();
8175}
8176EXPORT_SYMBOL(unregister_netdev);
8177
8178
8179
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
8193{
8194 int err;
8195
8196 ASSERT_RTNL();
8197
8198
8199 err = -EINVAL;
8200 if (dev->features & NETIF_F_NETNS_LOCAL)
8201 goto out;
8202
8203
8204 if (dev->reg_state != NETREG_REGISTERED)
8205 goto out;
8206
8207
8208 err = 0;
8209 if (net_eq(dev_net(dev), net))
8210 goto out;
8211
8212
8213
8214
8215 err = -EEXIST;
8216 if (__dev_get_by_name(net, dev->name)) {
8217
8218 if (!pat)
8219 goto out;
8220 if (dev_get_valid_name(net, dev, pat) < 0)
8221 goto out;
8222 }
8223
8224
8225
8226
8227
8228
8229 dev_close(dev);
8230
8231
8232 err = -ENODEV;
8233 unlist_netdevice(dev);
8234
8235 synchronize_net();
8236
8237
8238 dev_shutdown(dev);
8239
8240
8241
8242
8243
8244
8245
8246
8247 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8248 rcu_barrier();
8249 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
8250 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
8251
8252
8253
8254
8255 dev_uc_flush(dev);
8256 dev_mc_flush(dev);
8257
8258
8259 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
8260 netdev_adjacent_del_links(dev);
8261
8262
8263 dev_net_set(dev, net);
8264
8265
8266 if (__dev_get_by_index(net, dev->ifindex))
8267 dev->ifindex = dev_new_index(net);
8268
8269
8270 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
8271 netdev_adjacent_add_links(dev);
8272
8273
8274 err = device_rename(&dev->dev, dev->name);
8275 WARN_ON(err);
8276
8277
8278 list_netdevice(dev);
8279
8280
8281 call_netdevice_notifiers(NETDEV_REGISTER, dev);
8282
8283
8284
8285
8286
8287 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8288
8289 synchronize_net();
8290 err = 0;
8291out:
8292 return err;
8293}
8294EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8295
8296static int dev_cpu_dead(unsigned int oldcpu)
8297{
8298 struct sk_buff **list_skb;
8299 struct sk_buff *skb;
8300 unsigned int cpu;
8301 struct softnet_data *sd, *oldsd, *remsd = NULL;
8302
8303 local_irq_disable();
8304 cpu = smp_processor_id();
8305 sd = &per_cpu(softnet_data, cpu);
8306 oldsd = &per_cpu(softnet_data, oldcpu);
8307
8308
8309 list_skb = &sd->completion_queue;
8310 while (*list_skb)
8311 list_skb = &(*list_skb)->next;
8312
8313 *list_skb = oldsd->completion_queue;
8314 oldsd->completion_queue = NULL;
8315
8316
8317 if (oldsd->output_queue) {
8318 *sd->output_queue_tailp = oldsd->output_queue;
8319 sd->output_queue_tailp = oldsd->output_queue_tailp;
8320 oldsd->output_queue = NULL;
8321 oldsd->output_queue_tailp = &oldsd->output_queue;
8322 }
8323
8324
8325
8326
8327 while (!list_empty(&oldsd->poll_list)) {
8328 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8329 struct napi_struct,
8330 poll_list);
8331
8332 list_del_init(&napi->poll_list);
8333 if (napi->poll == process_backlog)
8334 napi->state = 0;
8335 else
8336 ____napi_schedule(sd, napi);
8337 }
8338
8339 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8340 local_irq_enable();
8341
8342#ifdef CONFIG_RPS
8343 remsd = oldsd->rps_ipi_list;
8344 oldsd->rps_ipi_list = NULL;
8345#endif
8346
8347 net_rps_send_ipi(remsd);
8348
8349
8350 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8351 netif_rx_ni(skb);
8352 input_queue_head_incr(oldsd);
8353 }
8354 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8355 netif_rx_ni(skb);
8356 input_queue_head_incr(oldsd);
8357 }
8358
8359 return 0;
8360}
8361
8362
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372netdev_features_t netdev_increment_features(netdev_features_t all,
8373 netdev_features_t one, netdev_features_t mask)
8374{
8375 if (mask & NETIF_F_HW_CSUM)
8376 mask |= NETIF_F_CSUM_MASK;
8377 mask |= NETIF_F_VLAN_CHALLENGED;
8378
8379 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8380 all &= one | ~NETIF_F_ALL_FOR_ALL;
8381
8382
8383 if (all & NETIF_F_HW_CSUM)
8384 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8385
8386 return all;
8387}
8388EXPORT_SYMBOL(netdev_increment_features);
8389
8390static struct hlist_head * __net_init netdev_create_hash(void)
8391{
8392 int i;
8393 struct hlist_head *hash;
8394
8395 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8396 if (hash != NULL)
8397 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8398 INIT_HLIST_HEAD(&hash[i]);
8399
8400 return hash;
8401}
8402
8403
8404static int __net_init netdev_init(struct net *net)
8405{
8406 if (net != &init_net)
8407 INIT_LIST_HEAD(&net->dev_base_head);
8408
8409 net->dev_name_head = netdev_create_hash();
8410 if (net->dev_name_head == NULL)
8411 goto err_name;
8412
8413 net->dev_index_head = netdev_create_hash();
8414 if (net->dev_index_head == NULL)
8415 goto err_idx;
8416
8417 return 0;
8418
8419err_idx:
8420 kfree(net->dev_name_head);
8421err_name:
8422 return -ENOMEM;
8423}
8424
8425
8426
8427
8428
8429
8430
8431const char *netdev_drivername(const struct net_device *dev)
8432{
8433 const struct device_driver *driver;
8434 const struct device *parent;
8435 const char *empty = "";
8436
8437 parent = dev->dev.parent;
8438 if (!parent)
8439 return empty;
8440
8441 driver = parent->driver;
8442 if (driver && driver->name)
8443 return driver->name;
8444 return empty;
8445}
8446
8447static void __netdev_printk(const char *level, const struct net_device *dev,
8448 struct va_format *vaf)
8449{
8450 if (dev && dev->dev.parent) {
8451 dev_printk_emit(level[1] - '0',
8452 dev->dev.parent,
8453 "%s %s %s%s: %pV",
8454 dev_driver_string(dev->dev.parent),
8455 dev_name(dev->dev.parent),
8456 netdev_name(dev), netdev_reg_state(dev),
8457 vaf);
8458 } else if (dev) {
8459 printk("%s%s%s: %pV",
8460 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8461 } else {
8462 printk("%s(NULL net_device): %pV", level, vaf);
8463 }
8464}
8465
8466void netdev_printk(const char *level, const struct net_device *dev,
8467 const char *format, ...)
8468{
8469 struct va_format vaf;
8470 va_list args;
8471
8472 va_start(args, format);
8473
8474 vaf.fmt = format;
8475 vaf.va = &args;
8476
8477 __netdev_printk(level, dev, &vaf);
8478
8479 va_end(args);
8480}
8481EXPORT_SYMBOL(netdev_printk);
8482
8483#define define_netdev_printk_level(func, level) \
8484void func(const struct net_device *dev, const char *fmt, ...) \
8485{ \
8486 struct va_format vaf; \
8487 va_list args; \
8488 \
8489 va_start(args, fmt); \
8490 \
8491 vaf.fmt = fmt; \
8492 vaf.va = &args; \
8493 \
8494 __netdev_printk(level, dev, &vaf); \
8495 \
8496 va_end(args); \
8497} \
8498EXPORT_SYMBOL(func);
8499
8500define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8501define_netdev_printk_level(netdev_alert, KERN_ALERT);
8502define_netdev_printk_level(netdev_crit, KERN_CRIT);
8503define_netdev_printk_level(netdev_err, KERN_ERR);
8504define_netdev_printk_level(netdev_warn, KERN_WARNING);
8505define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8506define_netdev_printk_level(netdev_info, KERN_INFO);
8507
8508static void __net_exit netdev_exit(struct net *net)
8509{
8510 kfree(net->dev_name_head);
8511 kfree(net->dev_index_head);
8512}
8513
8514static struct pernet_operations __net_initdata netdev_net_ops = {
8515 .init = netdev_init,
8516 .exit = netdev_exit,
8517};
8518
8519static void __net_exit default_device_exit(struct net *net)
8520{
8521 struct net_device *dev, *aux;
8522
8523
8524
8525
8526 rtnl_lock();
8527 for_each_netdev_safe(net, dev, aux) {
8528 int err;
8529 char fb_name[IFNAMSIZ];
8530
8531
8532 if (dev->features & NETIF_F_NETNS_LOCAL)
8533 continue;
8534
8535
8536 if (dev->rtnl_link_ops)
8537 continue;
8538
8539
8540 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8541 err = dev_change_net_namespace(dev, &init_net, fb_name);
8542 if (err) {
8543 pr_emerg("%s: failed to move %s to init_net: %d\n",
8544 __func__, dev->name, err);
8545 BUG();
8546 }
8547 }
8548 rtnl_unlock();
8549}
8550
8551static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8552{
8553
8554
8555
8556 struct net *net;
8557 bool unregistering;
8558 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8559
8560 add_wait_queue(&netdev_unregistering_wq, &wait);
8561 for (;;) {
8562 unregistering = false;
8563 rtnl_lock();
8564 list_for_each_entry(net, net_list, exit_list) {
8565 if (net->dev_unreg_count > 0) {
8566 unregistering = true;
8567 break;
8568 }
8569 }
8570 if (!unregistering)
8571 break;
8572 __rtnl_unlock();
8573
8574 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8575 }
8576 remove_wait_queue(&netdev_unregistering_wq, &wait);
8577}
8578
8579static void __net_exit default_device_exit_batch(struct list_head *net_list)
8580{
8581
8582
8583
8584
8585
8586 struct net_device *dev;
8587 struct net *net;
8588 LIST_HEAD(dev_kill_list);
8589
8590
8591
8592
8593
8594
8595
8596
8597
8598
8599
8600
8601 rtnl_lock_unregistering(net_list);
8602 list_for_each_entry(net, net_list, exit_list) {
8603 for_each_netdev_reverse(net, dev) {
8604 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8605 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8606 else
8607 unregister_netdevice_queue(dev, &dev_kill_list);
8608 }
8609 }
8610 unregister_netdevice_many(&dev_kill_list);
8611 rtnl_unlock();
8612}
8613
8614static struct pernet_operations __net_initdata default_device_ops = {
8615 .exit = default_device_exit,
8616 .exit_batch = default_device_exit_batch,
8617};
8618
8619
8620
8621
8622
8623
8624
8625
8626
8627
8628
8629
8630static int __init net_dev_init(void)
8631{
8632 int i, rc = -ENOMEM;
8633
8634 BUG_ON(!dev_boot_phase);
8635
8636 if (dev_proc_init())
8637 goto out;
8638
8639 if (netdev_kobject_init())
8640 goto out;
8641
8642 INIT_LIST_HEAD(&ptype_all);
8643 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8644 INIT_LIST_HEAD(&ptype_base[i]);
8645
8646 INIT_LIST_HEAD(&offload_base);
8647
8648 if (register_pernet_subsys(&netdev_net_ops))
8649 goto out;
8650
8651
8652
8653
8654
8655 for_each_possible_cpu(i) {
8656 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8657 struct softnet_data *sd = &per_cpu(softnet_data, i);
8658
8659 INIT_WORK(flush, flush_backlog);
8660
8661 skb_queue_head_init(&sd->input_pkt_queue);
8662 skb_queue_head_init(&sd->process_queue);
8663 INIT_LIST_HEAD(&sd->poll_list);
8664 sd->output_queue_tailp = &sd->output_queue;
8665#ifdef CONFIG_RPS
8666 sd->csd.func = rps_trigger_softirq;
8667 sd->csd.info = sd;
8668 sd->cpu = i;
8669#endif
8670
8671 sd->backlog.poll = process_backlog;
8672 sd->backlog.weight = weight_p;
8673 }
8674
8675 dev_boot_phase = 0;
8676
8677
8678
8679
8680
8681
8682
8683
8684
8685
8686 if (register_pernet_device(&loopback_net_ops))
8687 goto out;
8688
8689 if (register_pernet_device(&default_device_ops))
8690 goto out;
8691
8692 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8693 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8694
8695 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8696 NULL, dev_cpu_dead);
8697 WARN_ON(rc < 0);
8698 rc = 0;
8699out:
8700 return rc;
8701}
8702
8703subsys_initcall(net_dev_init);
8704