1/*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
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#include <net/udp_tunnel.h>
148#include <linux/net_namespace.h>
149
150#include "net-sysfs.h"
151
152/* Instead of increasing this, you should create a hash table. */
153#define MAX_GRO_SKBS 8
154
155/* This should be increased if a protocol with a bigger head is added. */
156#define GRO_MAX_HEAD (MAX_HEADER + 128)
157
158static DEFINE_SPINLOCK(ptype_lock);
159static DEFINE_SPINLOCK(offload_lock);
160struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
161struct list_head ptype_all __read_mostly; /* Taps */
162static struct list_head offload_base __read_mostly;
163
164static int netif_rx_internal(struct sk_buff *skb);
165static int call_netdevice_notifiers_info(unsigned long val,
166 struct netdev_notifier_info *info);
167static struct napi_struct *napi_by_id(unsigned int napi_id);
168
169/*
170 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171 * semaphore.
172 *
173 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
174 *
175 * Writers must hold the rtnl semaphore while they loop through the
176 * dev_base_head list, and hold dev_base_lock for writing when they do the
177 * actual updates. This allows pure readers to access the list even
178 * while a writer is preparing to update it.
179 *
180 * To put it another way, dev_base_lock is held for writing only to
181 * protect against pure readers; the rtnl semaphore provides the
182 * protection against other writers.
183 *
184 * See, for example usages, register_netdevice() and
185 * unregister_netdevice(), which must be called with the rtnl
186 * semaphore held.
187 */
188DEFINE_RWLOCK(dev_base_lock);
189EXPORT_SYMBOL(dev_base_lock);
190
191static DEFINE_MUTEX(ifalias_mutex);
192
193/* protects napi_hash addition/deletion and napi_gen_id */
194static DEFINE_SPINLOCK(napi_hash_lock);
195
196static unsigned int napi_gen_id = NR_CPUS;
197static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
198
199static seqcount_t devnet_rename_seq;
200
201static inline void dev_base_seq_inc(struct net *net)
202{
203 while (++net->dev_base_seq == 0)
204 ;
205}
206
207static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208{
209 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
210
211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212}
213
214static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215{
216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217}
218
219static inline void rps_lock(struct softnet_data *sd)
220{
221#ifdef CONFIG_RPS
222 spin_lock(&sd->input_pkt_queue.lock);
223#endif
224}
225
226static inline void rps_unlock(struct softnet_data *sd)
227{
228#ifdef CONFIG_RPS
229 spin_unlock(&sd->input_pkt_queue.lock);
230#endif
231}
232
233/* Device list insertion */
234static void list_netdevice(struct net_device *dev)
235{
236 struct net *net = dev_net(dev);
237
238 ASSERT_RTNL();
239
240 write_lock_bh(&dev_base_lock);
241 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
243 hlist_add_head_rcu(&dev->index_hlist,
244 dev_index_hash(net, dev->ifindex));
245 write_unlock_bh(&dev_base_lock);
246
247 dev_base_seq_inc(net);
248}
249
250/* Device list removal
251 * caller must respect a RCU grace period before freeing/reusing dev
252 */
253static void unlist_netdevice(struct net_device *dev)
254{
255 ASSERT_RTNL();
256
257 /* Unlink dev from the device chain */
258 write_lock_bh(&dev_base_lock);
259 list_del_rcu(&dev->dev_list);
260 hlist_del_rcu(&dev->name_hlist);
261 hlist_del_rcu(&dev->index_hlist);
262 write_unlock_bh(&dev_base_lock);
263
264 dev_base_seq_inc(dev_net(dev));
265}
266
267/*
268 * Our notifier list
269 */
270
271static RAW_NOTIFIER_HEAD(netdev_chain);
272
273/*
274 * Device drivers call our routines to queue packets here. We empty the
275 * queue in the local softnet handler.
276 */
277
278DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
279EXPORT_PER_CPU_SYMBOL(softnet_data);
280
281#ifdef CONFIG_LOCKDEP
282/*
283 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
284 * according to dev->type
285 */
286static const unsigned short netdev_lock_type[] = {
287 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
288 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
289 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
290 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
291 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
292 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
293 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
294 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
295 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
296 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
297 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
298 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
299 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
300 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
301 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
302
303static const char *const netdev_lock_name[] = {
304 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
305 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
306 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
307 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
308 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
309 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
310 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
311 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
312 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
313 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
314 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
315 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
316 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
317 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
318 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
319
320static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
321static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
322
323static inline unsigned short netdev_lock_pos(unsigned short dev_type)
324{
325 int i;
326
327 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
328 if (netdev_lock_type[i] == dev_type)
329 return i;
330 /* the last key is used by default */
331 return ARRAY_SIZE(netdev_lock_type) - 1;
332}
333
334static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
336{
337 int i;
338
339 i = netdev_lock_pos(dev_type);
340 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
341 netdev_lock_name[i]);
342}
343
344static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
345{
346 int i;
347
348 i = netdev_lock_pos(dev->type);
349 lockdep_set_class_and_name(&dev->addr_list_lock,
350 &netdev_addr_lock_key[i],
351 netdev_lock_name[i]);
352}
353#else
354static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
355 unsigned short dev_type)
356{
357}
358static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
359{
360}
361#endif
362
363/*******************************************************************************
364 *
365 * Protocol management and registration routines
366 *
367 *******************************************************************************/
368
369
370/*
371 * Add a protocol ID to the list. Now that the input handler is
372 * smarter we can dispense with all the messy stuff that used to be
373 * here.
374 *
375 * BEWARE!!! Protocol handlers, mangling input packets,
376 * MUST BE last in hash buckets and checking protocol handlers
377 * MUST start from promiscuous ptype_all chain in net_bh.
378 * It is true now, do not change it.
379 * Explanation follows: if protocol handler, mangling packet, will
380 * be the first on list, it is not able to sense, that packet
381 * is cloned and should be copied-on-write, so that it will
382 * change it and subsequent readers will get broken packet.
383 * --ANK (980803)
384 */
385
386static inline struct list_head *ptype_head(const struct packet_type *pt)
387{
388 if (pt->type == htons(ETH_P_ALL))
389 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
390 else
391 return pt->dev ? &pt->dev->ptype_specific :
392 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
393}
394
395/**
396 * dev_add_pack - add packet handler
397 * @pt: packet type declaration
398 *
399 * Add a protocol handler to the networking stack. The passed &packet_type
400 * is linked into kernel lists and may not be freed until it has been
401 * removed from the kernel lists.
402 *
403 * This call does not sleep therefore it can not
404 * guarantee all CPU's that are in middle of receiving packets
405 * will see the new packet type (until the next received packet).
406 */
407
408void dev_add_pack(struct packet_type *pt)
409{
410 struct list_head *head = ptype_head(pt);
411
412 spin_lock(&ptype_lock);
413 list_add_rcu(&pt->list, head);
414 spin_unlock(&ptype_lock);
415}
416EXPORT_SYMBOL(dev_add_pack);
417
418/**
419 * __dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
421 *
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
425 * returns.
426 *
427 * The packet type might still be in use by receivers
428 * and must not be freed until after all the CPU's have gone
429 * through a quiescent state.
430 */
431void __dev_remove_pack(struct packet_type *pt)
432{
433 struct list_head *head = ptype_head(pt);
434 struct packet_type *pt1;
435
436 spin_lock(&ptype_lock);
437
438 list_for_each_entry(pt1, head, list) {
439 if (pt == pt1) {
440 list_del_rcu(&pt->list);
441 goto out;
442 }
443 }
444
445 pr_warn("dev_remove_pack: %p not found\n", pt);
446out:
447 spin_unlock(&ptype_lock);
448}
449EXPORT_SYMBOL(__dev_remove_pack);
450
451/**
452 * dev_remove_pack - remove packet handler
453 * @pt: packet type declaration
454 *
455 * Remove a protocol handler that was previously added to the kernel
456 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
457 * from the kernel lists and can be freed or reused once this function
458 * returns.
459 *
460 * This call sleeps to guarantee that no CPU is looking at the packet
461 * type after return.
462 */
463void dev_remove_pack(struct packet_type *pt)
464{
465 __dev_remove_pack(pt);
466
467 synchronize_net();
468}
469EXPORT_SYMBOL(dev_remove_pack);
470
471
472/**
473 * dev_add_offload - register offload handlers
474 * @po: protocol offload declaration
475 *
476 * Add protocol offload handlers to the networking stack. The passed
477 * &proto_offload is linked into kernel lists and may not be freed until
478 * it has been removed from the kernel lists.
479 *
480 * This call does not sleep therefore it can not
481 * guarantee all CPU's that are in middle of receiving packets
482 * will see the new offload handlers (until the next received packet).
483 */
484void dev_add_offload(struct packet_offload *po)
485{
486 struct packet_offload *elem;
487
488 spin_lock(&offload_lock);
489 list_for_each_entry(elem, &offload_base, list) {
490 if (po->priority < elem->priority)
491 break;
492 }
493 list_add_rcu(&po->list, elem->list.prev);
494 spin_unlock(&offload_lock);
495}
496EXPORT_SYMBOL(dev_add_offload);
497
498/**
499 * __dev_remove_offload - remove offload handler
500 * @po: packet offload declaration
501 *
502 * Remove a protocol offload handler that was previously added to the
503 * kernel offload handlers by dev_add_offload(). The passed &offload_type
504 * is removed from the kernel lists and can be freed or reused once this
505 * function returns.
506 *
507 * The packet type might still be in use by receivers
508 * and must not be freed until after all the CPU's have gone
509 * through a quiescent state.
510 */
511static void __dev_remove_offload(struct packet_offload *po)
512{
513 struct list_head *head = &offload_base;
514 struct packet_offload *po1;
515
516 spin_lock(&offload_lock);
517
518 list_for_each_entry(po1, head, list) {
519 if (po == po1) {
520 list_del_rcu(&po->list);
521 goto out;
522 }
523 }
524
525 pr_warn("dev_remove_offload: %p not found\n", po);
526out:
527 spin_unlock(&offload_lock);
528}
529
530/**
531 * dev_remove_offload - remove packet offload handler
532 * @po: packet offload declaration
533 *
534 * Remove a packet offload handler that was previously added to the kernel
535 * offload handlers by dev_add_offload(). The passed &offload_type is
536 * removed from the kernel lists and can be freed or reused once this
537 * function returns.
538 *
539 * This call sleeps to guarantee that no CPU is looking at the packet
540 * type after return.
541 */
542void dev_remove_offload(struct packet_offload *po)
543{
544 __dev_remove_offload(po);
545
546 synchronize_net();
547}
548EXPORT_SYMBOL(dev_remove_offload);
549
550/******************************************************************************
551 *
552 * Device Boot-time Settings Routines
553 *
554 ******************************************************************************/
555
556/* Boot time configuration table */
557static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
558
559/**
560 * netdev_boot_setup_add - add new setup entry
561 * @name: name of the device
562 * @map: configured settings for the device
563 *
564 * Adds new setup entry to the dev_boot_setup list. The function
565 * returns 0 on error and 1 on success. This is a generic routine to
566 * all netdevices.
567 */
568static int netdev_boot_setup_add(char *name, struct ifmap *map)
569{
570 struct netdev_boot_setup *s;
571 int i;
572
573 s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
576 memset(s[i].name, 0, sizeof(s[i].name));
577 strlcpy(s[i].name, name, IFNAMSIZ);
578 memcpy(&s[i].map, map, sizeof(s[i].map));
579 break;
580 }
581 }
582
583 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
584}
585
586/**
587 * netdev_boot_setup_check - check boot time settings
588 * @dev: the netdevice
589 *
590 * Check boot time settings for the device.
591 * The found settings are set for the device to be used
592 * later in the device probing.
593 * Returns 0 if no settings found, 1 if they are.
594 */
595int netdev_boot_setup_check(struct net_device *dev)
596{
597 struct netdev_boot_setup *s = dev_boot_setup;
598 int i;
599
600 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
601 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
602 !strcmp(dev->name, s[i].name)) {
603 dev->irq = s[i].map.irq;
604 dev->base_addr = s[i].map.base_addr;
605 dev->mem_start = s[i].map.mem_start;
606 dev->mem_end = s[i].map.mem_end;
607 return 1;
608 }
609 }
610 return 0;
611}
612EXPORT_SYMBOL(netdev_boot_setup_check);
613
614
615/**
616 * netdev_boot_base - get address from boot time settings
617 * @prefix: prefix for network device
618 * @unit: id for network device
619 *
620 * Check boot time settings for the base address of device.
621 * The found settings are set for the device to be used
622 * later in the device probing.
623 * Returns 0 if no settings found.
624 */
625unsigned long netdev_boot_base(const char *prefix, int unit)
626{
627 const struct netdev_boot_setup *s = dev_boot_setup;
628 char name[IFNAMSIZ];
629 int i;
630
631 sprintf(name, "%s%d", prefix, unit);
632
633 /*
634 * If device already registered then return base of 1
635 * to indicate not to probe for this interface
636 */
637 if (__dev_get_by_name(&init_net, name))
638 return 1;
639
640 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
641 if (!strcmp(name, s[i].name))
642 return s[i].map.base_addr;
643 return 0;
644}
645
646/*
647 * Saves at boot time configured settings for any netdevice.
648 */
649int __init netdev_boot_setup(char *str)
650{
651 int ints[5];
652 struct ifmap map;
653
654 str = get_options(str, ARRAY_SIZE(ints), ints);
655 if (!str || !*str)
656 return 0;
657
658 /* Save settings */
659 memset(&map, 0, sizeof(map));
660 if (ints[0] > 0)
661 map.irq = ints[1];
662 if (ints[0] > 1)
663 map.base_addr = ints[2];
664 if (ints[0] > 2)
665 map.mem_start = ints[3];
666 if (ints[0] > 3)
667 map.mem_end = ints[4];
668
669 /* Add new entry to the list */
670 return netdev_boot_setup_add(str, &map);
671}
672
673__setup("netdev=", netdev_boot_setup);
674
675/*******************************************************************************
676 *
677 * Device Interface Subroutines
678 *
679 *******************************************************************************/
680
681/**
682 * dev_get_iflink - get 'iflink' value of a interface
683 * @dev: targeted interface
684 *
685 * Indicates the ifindex the interface is linked to.
686 * Physical interfaces have the same 'ifindex' and 'iflink' values.
687 */
688
689int dev_get_iflink(const struct net_device *dev)
690{
691 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
692 return dev->netdev_ops->ndo_get_iflink(dev);
693
694 return dev->ifindex;
695}
696EXPORT_SYMBOL(dev_get_iflink);
697
698/**
699 * dev_fill_metadata_dst - Retrieve tunnel egress information.
700 * @dev: targeted interface
701 * @skb: The packet.
702 *
703 * For better visibility of tunnel traffic OVS needs to retrieve
704 * egress tunnel information for a packet. Following API allows
705 * user to get this info.
706 */
707int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
708{
709 struct ip_tunnel_info *info;
710
711 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
712 return -EINVAL;
713
714 info = skb_tunnel_info_unclone(skb);
715 if (!info)
716 return -ENOMEM;
717 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
718 return -EINVAL;
719
720 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
721}
722EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
723
724/**
725 * __dev_get_by_name - find a device by its name
726 * @net: the applicable net namespace
727 * @name: name to find
728 *
729 * Find an interface by name. Must be called under RTNL semaphore
730 * or @dev_base_lock. If the name is found a pointer to the device
731 * is returned. If the name is not found then %NULL is returned. The
732 * reference counters are not incremented so the caller must be
733 * careful with locks.
734 */
735
736struct net_device *__dev_get_by_name(struct net *net, const char *name)
737{
738 struct net_device *dev;
739 struct hlist_head *head = dev_name_hash(net, name);
740
741 hlist_for_each_entry(dev, head, name_hlist)
742 if (!strncmp(dev->name, name, IFNAMSIZ))
743 return dev;
744
745 return NULL;
746}
747EXPORT_SYMBOL(__dev_get_by_name);
748
749/**
750 * dev_get_by_name_rcu - find a device by its name
751 * @net: the applicable net namespace
752 * @name: name to find
753 *
754 * Find an interface by name.
755 * If the name is found a pointer to the device is returned.
756 * If the name is not found then %NULL is returned.
757 * The reference counters are not incremented so the caller must be
758 * careful with locks. The caller must hold RCU lock.
759 */
760
761struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
762{
763 struct net_device *dev;
764 struct hlist_head *head = dev_name_hash(net, name);
765
766 hlist_for_each_entry_rcu(dev, head, name_hlist)
767 if (!strncmp(dev->name, name, IFNAMSIZ))
768 return dev;
769
770 return NULL;
771}
772EXPORT_SYMBOL(dev_get_by_name_rcu);
773
774/**
775 * dev_get_by_name - find a device by its name
776 * @net: the applicable net namespace
777 * @name: name to find
778 *
779 * Find an interface by name. This can be called from any
780 * context and does its own locking. The returned handle has
781 * the usage count incremented and the caller must use dev_put() to
782 * release it when it is no longer needed. %NULL is returned if no
783 * matching device is found.
784 */
785
786struct net_device *dev_get_by_name(struct net *net, const char *name)
787{
788 struct net_device *dev;
789
790 rcu_read_lock();
791 dev = dev_get_by_name_rcu(net, name);
792 if (dev)
793 dev_hold(dev);
794 rcu_read_unlock();
795 return dev;
796}
797EXPORT_SYMBOL(dev_get_by_name);
798
799/**
800 * __dev_get_by_index - find a device by its ifindex
801 * @net: the applicable net namespace
802 * @ifindex: index of device
803 *
804 * Search for an interface by index. Returns %NULL if the device
805 * is not found or a pointer to the device. The device has not
806 * had its reference counter increased so the caller must be careful
807 * about locking. The caller must hold either the RTNL semaphore
808 * or @dev_base_lock.
809 */
810
811struct net_device *__dev_get_by_index(struct net *net, int ifindex)
812{
813 struct net_device *dev;
814 struct hlist_head *head = dev_index_hash(net, ifindex);
815
816 hlist_for_each_entry(dev, head, index_hlist)
817 if (dev->ifindex == ifindex)
818 return dev;
819
820 return NULL;
821}
822EXPORT_SYMBOL(__dev_get_by_index);
823
824/**
825 * dev_get_by_index_rcu - find a device by its ifindex
826 * @net: the applicable net namespace
827 * @ifindex: index of device
828 *
829 * Search for an interface by index. Returns %NULL if the device
830 * is not found or a pointer to the device. The device has not
831 * had its reference counter increased so the caller must be careful
832 * about locking. The caller must hold RCU lock.
833 */
834
835struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
836{
837 struct net_device *dev;
838 struct hlist_head *head = dev_index_hash(net, ifindex);
839
840 hlist_for_each_entry_rcu(dev, head, index_hlist)
841 if (dev->ifindex == ifindex)
842 return dev;
843
844 return NULL;
845}
846EXPORT_SYMBOL(dev_get_by_index_rcu);
847
848
849/**
850 * dev_get_by_index - find a device by its ifindex
851 * @net: the applicable net namespace
852 * @ifindex: index of device
853 *
854 * Search for an interface by index. Returns NULL if the device
855 * is not found or a pointer to the device. The device returned has
856 * had a reference added and the pointer is safe until the user calls
857 * dev_put to indicate they have finished with it.
858 */
859
860struct net_device *dev_get_by_index(struct net *net, int ifindex)
861{
862 struct net_device *dev;
863
864 rcu_read_lock();
865 dev = dev_get_by_index_rcu(net, ifindex);
866 if (dev)
867 dev_hold(dev);
868 rcu_read_unlock();
869 return dev;
870}
871EXPORT_SYMBOL(dev_get_by_index);
872
873/**
874 * dev_get_by_napi_id - find a device by napi_id
875 * @napi_id: ID of the NAPI struct
876 *
877 * Search for an interface by NAPI ID. Returns %NULL if the device
878 * is not found or a pointer to the device. The device has not had
879 * its reference counter increased so the caller must be careful
880 * about locking. The caller must hold RCU lock.
881 */
882
883struct net_device *dev_get_by_napi_id(unsigned int napi_id)
884{
885 struct napi_struct *napi;
886
887 WARN_ON_ONCE(!rcu_read_lock_held());
888
889 if (napi_id < MIN_NAPI_ID)
890 return NULL;
891
892 napi = napi_by_id(napi_id);
893
894 return napi ? napi->dev : NULL;
895}
896EXPORT_SYMBOL(dev_get_by_napi_id);
897
898/**
899 * netdev_get_name - get a netdevice name, knowing its ifindex.
900 * @net: network namespace
901 * @name: a pointer to the buffer where the name will be stored.
902 * @ifindex: the ifindex of the interface to get the name from.
903 *
904 * The use of raw_seqcount_begin() and cond_resched() before
905 * retrying is required as we want to give the writers a chance
906 * to complete when CONFIG_PREEMPT is not set.
907 */
908int netdev_get_name(struct net *net, char *name, int ifindex)
909{
910 struct net_device *dev;
911 unsigned int seq;
912
913retry:
914 seq = raw_seqcount_begin(&devnet_rename_seq);
915 rcu_read_lock();
916 dev = dev_get_by_index_rcu(net, ifindex);
917 if (!dev) {
918 rcu_read_unlock();
919 return -ENODEV;
920 }
921
922 strcpy(name, dev->name);
923 rcu_read_unlock();
924 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
925 cond_resched();
926 goto retry;
927 }
928
929 return 0;
930}
931
932/**
933 * dev_getbyhwaddr_rcu - find a device by its hardware address
934 * @net: the applicable net namespace
935 * @type: media type of device
936 * @ha: hardware address
937 *
938 * Search for an interface by MAC address. Returns NULL if the device
939 * is not found or a pointer to the device.
940 * The caller must hold RCU or RTNL.
941 * The returned device has not had its ref count increased
942 * and the caller must therefore be careful about locking
943 *
944 */
945
946struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
947 const char *ha)
948{
949 struct net_device *dev;
950
951 for_each_netdev_rcu(net, dev)
952 if (dev->type == type &&
953 !memcmp(dev->dev_addr, ha, dev->addr_len))
954 return dev;
955
956 return NULL;
957}
958EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
959
960struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
961{
962 struct net_device *dev;
963
964 ASSERT_RTNL();
965 for_each_netdev(net, dev)
966 if (dev->type == type)
967 return dev;
968
969 return NULL;
970}
971EXPORT_SYMBOL(__dev_getfirstbyhwtype);
972
973struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
974{
975 struct net_device *dev, *ret = NULL;
976
977 rcu_read_lock();
978 for_each_netdev_rcu(net, dev)
979 if (dev->type == type) {
980 dev_hold(dev);
981 ret = dev;
982 break;
983 }
984 rcu_read_unlock();
985 return ret;
986}
987EXPORT_SYMBOL(dev_getfirstbyhwtype);
988
989/**
990 * __dev_get_by_flags - find any device with given flags
991 * @net: the applicable net namespace
992 * @if_flags: IFF_* values
993 * @mask: bitmask of bits in if_flags to check
994 *
995 * Search for any interface with the given flags. Returns NULL if a device
996 * is not found or a pointer to the device. Must be called inside
997 * rtnl_lock(), and result refcount is unchanged.
998 */
999
1000struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1001 unsigned short mask)
1002{
1003 struct net_device *dev, *ret;
1004
1005 ASSERT_RTNL();
1006
1007 ret = NULL;
1008 for_each_netdev(net, dev) {
1009 if (((dev->flags ^ if_flags) & mask) == 0) {
1010 ret = dev;
1011 break;
1012 }
1013 }
1014 return ret;
1015}
1016EXPORT_SYMBOL(__dev_get_by_flags);
1017
1018/**
1019 * dev_valid_name - check if name is okay for network device
1020 * @name: name string
1021 *
1022 * Network device names need to be valid file names to
1023 * to allow sysfs to work. We also disallow any kind of
1024 * whitespace.
1025 */
1026bool dev_valid_name(const char *name)
1027{
1028 if (*name == '\0')
1029 return false;
1030 if (strlen(name) >= IFNAMSIZ)
1031 return false;
1032 if (!strcmp(name, ".") || !strcmp(name, ".."))
1033 return false;
1034
1035 while (*name) {
1036 if (*name == '/' || *name == ':' || isspace(*name))
1037 return false;
1038 name++;
1039 }
1040 return true;
1041}
1042EXPORT_SYMBOL(dev_valid_name);
1043
1044/**
1045 * __dev_alloc_name - allocate a name for a device
1046 * @net: network namespace to allocate the device name in
1047 * @name: name format string
1048 * @buf: scratch buffer and result name string
1049 *
1050 * Passed a format string - eg "lt%d" it will try and find a suitable
1051 * id. It scans list of devices to build up a free map, then chooses
1052 * the first empty slot. The caller must hold the dev_base or rtnl lock
1053 * while allocating the name and adding the device in order to avoid
1054 * duplicates.
1055 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1056 * Returns the number of the unit assigned or a negative errno code.
1057 */
1058
1059static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1060{
1061 int i = 0;
1062 const char *p;
1063 const int max_netdevices = 8*PAGE_SIZE;
1064 unsigned long *inuse;
1065 struct net_device *d;
1066
1067 if (!dev_valid_name(name))
1068 return -EINVAL;
1069
1070 p = strchr(name, '%');
1071 if (p) {
1072 /*
1073 * Verify the string as this thing may have come from
1074 * the user. There must be either one "%d" and no other "%"
1075 * characters.
1076 */
1077 if (p[1] != 'd' || strchr(p + 2, '%'))
1078 return -EINVAL;
1079
1080 /* Use one page as a bit array of possible slots */
1081 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1082 if (!inuse)
1083 return -ENOMEM;
1084
1085 for_each_netdev(net, d) {
1086 if (!sscanf(d->name, name, &i))
1087 continue;
1088 if (i < 0 || i >= max_netdevices)
1089 continue;
1090
1091 /* avoid cases where sscanf is not exact inverse of printf */
1092 snprintf(buf, IFNAMSIZ, name, i);
1093 if (!strncmp(buf, d->name, IFNAMSIZ))
1094 set_bit(i, inuse);
1095 }
1096
1097 i = find_first_zero_bit(inuse, max_netdevices);
1098 free_page((unsigned long) inuse);
1099 }
1100
1101 snprintf(buf, IFNAMSIZ, name, i);
1102 if (!__dev_get_by_name(net, buf))
1103 return i;
1104
1105 /* It is possible to run out of possible slots
1106 * when the name is long and there isn't enough space left
1107 * for the digits, or if all bits are used.
1108 */
1109 return -ENFILE;
1110}
1111
1112static int dev_alloc_name_ns(struct net *net,
1113 struct net_device *dev,
1114 const char *name)
1115{
1116 char buf[IFNAMSIZ];
1117 int ret;
1118
1119 BUG_ON(!net);
1120 ret = __dev_alloc_name(net, name, buf);
1121 if (ret >= 0)
1122 strlcpy(dev->name, buf, IFNAMSIZ);
1123 return ret;
1124}
1125
1126/**
1127 * dev_alloc_name - allocate a name for a device
1128 * @dev: device
1129 * @name: name format string
1130 *
1131 * Passed a format string - eg "lt%d" it will try and find a suitable
1132 * id. It scans list of devices to build up a free map, then chooses
1133 * the first empty slot. The caller must hold the dev_base or rtnl lock
1134 * while allocating the name and adding the device in order to avoid
1135 * duplicates.
1136 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1137 * Returns the number of the unit assigned or a negative errno code.
1138 */
1139
1140int dev_alloc_name(struct net_device *dev, const char *name)
1141{
1142 return dev_alloc_name_ns(dev_net(dev), dev, name);
1143}
1144EXPORT_SYMBOL(dev_alloc_name);
1145
1146int dev_get_valid_name(struct net *net, struct net_device *dev,
1147 const char *name)
1148{
1149 BUG_ON(!net);
1150
1151 if (!dev_valid_name(name))
1152 return -EINVAL;
1153
1154 if (strchr(name, '%'))
1155 return dev_alloc_name_ns(net, dev, name);
1156 else if (__dev_get_by_name(net, name))
1157 return -EEXIST;
1158 else if (dev->name != name)
1159 strlcpy(dev->name, name, IFNAMSIZ);
1160
1161 return 0;
1162}
1163EXPORT_SYMBOL(dev_get_valid_name);
1164
1165/**
1166 * dev_change_name - change name of a device
1167 * @dev: device
1168 * @newname: name (or format string) must be at least IFNAMSIZ
1169 *
1170 * Change name of a device, can pass format strings "eth%d".
1171 * for wildcarding.
1172 */
1173int dev_change_name(struct net_device *dev, const char *newname)
1174{
1175 unsigned char old_assign_type;
1176 char oldname[IFNAMSIZ];
1177 int err = 0;
1178 int ret;
1179 struct net *net;
1180
1181 ASSERT_RTNL();
1182 BUG_ON(!dev_net(dev));
1183
1184 net = dev_net(dev);
1185 if (dev->flags & IFF_UP)
1186 return -EBUSY;
1187
1188 write_seqcount_begin(&devnet_rename_seq);
1189
1190 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1191 write_seqcount_end(&devnet_rename_seq);
1192 return 0;
1193 }
1194
1195 memcpy(oldname, dev->name, IFNAMSIZ);
1196
1197 err = dev_get_valid_name(net, dev, newname);
1198 if (err < 0) {
1199 write_seqcount_end(&devnet_rename_seq);
1200 return err;
1201 }
1202
1203 if (oldname[0] && !strchr(oldname, '%'))
1204 netdev_info(dev, "renamed from %s\n", oldname);
1205
1206 old_assign_type = dev->name_assign_type;
1207 dev->name_assign_type = NET_NAME_RENAMED;
1208
1209rollback:
1210 ret = device_rename(&dev->dev, dev->name);
1211 if (ret) {
1212 memcpy(dev->name, oldname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 write_seqcount_end(&devnet_rename_seq);
1215 return ret;
1216 }
1217
1218 write_seqcount_end(&devnet_rename_seq);
1219
1220 netdev_adjacent_rename_links(dev, oldname);
1221
1222 write_lock_bh(&dev_base_lock);
1223 hlist_del_rcu(&dev->name_hlist);
1224 write_unlock_bh(&dev_base_lock);
1225
1226 synchronize_rcu();
1227
1228 write_lock_bh(&dev_base_lock);
1229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1230 write_unlock_bh(&dev_base_lock);
1231
1232 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1233 ret = notifier_to_errno(ret);
1234
1235 if (ret) {
1236 /* err >= 0 after dev_alloc_name() or stores the first errno */
1237 if (err >= 0) {
1238 err = ret;
1239 write_seqcount_begin(&devnet_rename_seq);
1240 memcpy(dev->name, oldname, IFNAMSIZ);
1241 memcpy(oldname, newname, IFNAMSIZ);
1242 dev->name_assign_type = old_assign_type;
1243 old_assign_type = NET_NAME_RENAMED;
1244 goto rollback;
1245 } else {
1246 pr_err("%s: name change rollback failed: %d\n",
1247 dev->name, ret);
1248 }
1249 }
1250
1251 return err;
1252}
1253
1254/**
1255 * dev_set_alias - change ifalias of a device
1256 * @dev: device
1257 * @alias: name up to IFALIASZ
1258 * @len: limit of bytes to copy from info
1259 *
1260 * Set ifalias for a device,
1261 */
1262int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1263{
1264 struct dev_ifalias *new_alias = NULL;
1265
1266 if (len >= IFALIASZ)
1267 return -EINVAL;
1268
1269 if (len) {
1270 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1271 if (!new_alias)
1272 return -ENOMEM;
1273
1274 memcpy(new_alias->ifalias, alias, len);
1275 new_alias->ifalias[len] = 0;
1276 }
1277
1278 mutex_lock(&ifalias_mutex);
1279 rcu_swap_protected(dev->ifalias, new_alias,
1280 mutex_is_locked(&ifalias_mutex));
1281 mutex_unlock(&ifalias_mutex);
1282
1283 if (new_alias)
1284 kfree_rcu(new_alias, rcuhead);
1285
1286 return len;
1287}
1288
1289/**
1290 * dev_get_alias - get ifalias of a device
1291 * @dev: device
1292 * @name: buffer to store name of ifalias
1293 * @len: size of buffer
1294 *
1295 * get ifalias for a device. Caller must make sure dev cannot go
1296 * away, e.g. rcu read lock or own a reference count to device.
1297 */
1298int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1299{
1300 const struct dev_ifalias *alias;
1301 int ret = 0;
1302
1303 rcu_read_lock();
1304 alias = rcu_dereference(dev->ifalias);
1305 if (alias)
1306 ret = snprintf(name, len, "%s", alias->ifalias);
1307 rcu_read_unlock();
1308
1309 return ret;
1310}
1311
1312/**
1313 * netdev_features_change - device changes features
1314 * @dev: device to cause notification
1315 *
1316 * Called to indicate a device has changed features.
1317 */
1318void netdev_features_change(struct net_device *dev)
1319{
1320 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1321}
1322EXPORT_SYMBOL(netdev_features_change);
1323
1324/**
1325 * netdev_state_change - device changes state
1326 * @dev: device to cause notification
1327 *
1328 * Called to indicate a device has changed state. This function calls
1329 * the notifier chains for netdev_chain and sends a NEWLINK message
1330 * to the routing socket.
1331 */
1332void netdev_state_change(struct net_device *dev)
1333{
1334 if (dev->flags & IFF_UP) {
1335 struct netdev_notifier_change_info change_info = {
1336 .info.dev = dev,
1337 };
1338
1339 call_netdevice_notifiers_info(NETDEV_CHANGE,
1340 &change_info.info);
1341 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1342 }
1343}
1344EXPORT_SYMBOL(netdev_state_change);
1345
1346/**
1347 * netdev_notify_peers - notify network peers about existence of @dev
1348 * @dev: network device
1349 *
1350 * Generate traffic such that interested network peers are aware of
1351 * @dev, such as by generating a gratuitous ARP. This may be used when
1352 * a device wants to inform the rest of the network about some sort of
1353 * reconfiguration such as a failover event or virtual machine
1354 * migration.
1355 */
1356void netdev_notify_peers(struct net_device *dev)
1357{
1358 rtnl_lock();
1359 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1360 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1361 rtnl_unlock();
1362}
1363EXPORT_SYMBOL(netdev_notify_peers);
1364
1365static int __dev_open(struct net_device *dev)
1366{
1367 const struct net_device_ops *ops = dev->netdev_ops;
1368 int ret;
1369
1370 ASSERT_RTNL();
1371
1372 if (!netif_device_present(dev))
1373 return -ENODEV;
1374
1375 /* Block netpoll from trying to do any rx path servicing.
1376 * If we don't do this there is a chance ndo_poll_controller
1377 * or ndo_poll may be running while we open the device
1378 */
1379 netpoll_poll_disable(dev);
1380
1381 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1382 ret = notifier_to_errno(ret);
1383 if (ret)
1384 return ret;
1385
1386 set_bit(__LINK_STATE_START, &dev->state);
1387
1388 if (ops->ndo_validate_addr)
1389 ret = ops->ndo_validate_addr(dev);
1390
1391 if (!ret && ops->ndo_open)
1392 ret = ops->ndo_open(dev);
1393
1394 netpoll_poll_enable(dev);
1395
1396 if (ret)
1397 clear_bit(__LINK_STATE_START, &dev->state);
1398 else {
1399 dev->flags |= IFF_UP;
1400 dev_set_rx_mode(dev);
1401 dev_activate(dev);
1402 add_device_randomness(dev->dev_addr, dev->addr_len);
1403 }
1404
1405 return ret;
1406}
1407
1408/**
1409 * dev_open - prepare an interface for use.
1410 * @dev: device to open
1411 *
1412 * Takes a device from down to up state. The device's private open
1413 * function is invoked and then the multicast lists are loaded. Finally
1414 * the device is moved into the up state and a %NETDEV_UP message is
1415 * sent to the netdev notifier chain.
1416 *
1417 * Calling this function on an active interface is a nop. On a failure
1418 * a negative errno code is returned.
1419 */
1420int dev_open(struct net_device *dev)
1421{
1422 int ret;
1423
1424 if (dev->flags & IFF_UP)
1425 return 0;
1426
1427 ret = __dev_open(dev);
1428 if (ret < 0)
1429 return ret;
1430
1431 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1432 call_netdevice_notifiers(NETDEV_UP, dev);
1433
1434 return ret;
1435}
1436EXPORT_SYMBOL(dev_open);
1437
1438static void __dev_close_many(struct list_head *head)
1439{
1440 struct net_device *dev;
1441
1442 ASSERT_RTNL();
1443 might_sleep();
1444
1445 list_for_each_entry(dev, head, close_list) {
1446 /* Temporarily disable netpoll until the interface is down */
1447 netpoll_poll_disable(dev);
1448
1449 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1450
1451 clear_bit(__LINK_STATE_START, &dev->state);
1452
1453 /* Synchronize to scheduled poll. We cannot touch poll list, it
1454 * can be even on different cpu. So just clear netif_running().
1455 *
1456 * dev->stop() will invoke napi_disable() on all of it's
1457 * napi_struct instances on this device.
1458 */
1459 smp_mb__after_atomic(); /* Commit netif_running(). */
1460 }
1461
1462 dev_deactivate_many(head);
1463
1464 list_for_each_entry(dev, head, close_list) {
1465 const struct net_device_ops *ops = dev->netdev_ops;
1466
1467 /*
1468 * Call the device specific close. This cannot fail.
1469 * Only if device is UP
1470 *
1471 * We allow it to be called even after a DETACH hot-plug
1472 * event.
1473 */
1474 if (ops->ndo_stop)
1475 ops->ndo_stop(dev);
1476
1477 dev->flags &= ~IFF_UP;
1478 netpoll_poll_enable(dev);
1479 }
1480}
1481
1482static void __dev_close(struct net_device *dev)
1483{
1484 LIST_HEAD(single);
1485
1486 list_add(&dev->close_list, &single);
1487 __dev_close_many(&single);
1488 list_del(&single);
1489}
1490
1491void dev_close_many(struct list_head *head, bool unlink)
1492{
1493 struct net_device *dev, *tmp;
1494
1495 /* Remove the devices that don't need to be closed */
1496 list_for_each_entry_safe(dev, tmp, head, close_list)
1497 if (!(dev->flags & IFF_UP))
1498 list_del_init(&dev->close_list);
1499
1500 __dev_close_many(head);
1501
1502 list_for_each_entry_safe(dev, tmp, head, close_list) {
1503 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1504 call_netdevice_notifiers(NETDEV_DOWN, dev);
1505 if (unlink)
1506 list_del_init(&dev->close_list);
1507 }
1508}
1509EXPORT_SYMBOL(dev_close_many);
1510
1511/**
1512 * dev_close - shutdown an interface.
1513 * @dev: device to shutdown
1514 *
1515 * This function moves an active device into down state. A
1516 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1517 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1518 * chain.
1519 */
1520void dev_close(struct net_device *dev)
1521{
1522 if (dev->flags & IFF_UP) {
1523 LIST_HEAD(single);
1524
1525 list_add(&dev->close_list, &single);
1526 dev_close_many(&single, true);
1527 list_del(&single);
1528 }
1529}
1530EXPORT_SYMBOL(dev_close);
1531
1532
1533/**
1534 * dev_disable_lro - disable Large Receive Offload on a device
1535 * @dev: device
1536 *
1537 * Disable Large Receive Offload (LRO) on a net device. Must be
1538 * called under RTNL. This is needed if received packets may be
1539 * forwarded to another interface.
1540 */
1541void dev_disable_lro(struct net_device *dev)
1542{
1543 struct net_device *lower_dev;
1544 struct list_head *iter;
1545
1546 dev->wanted_features &= ~NETIF_F_LRO;
1547 netdev_update_features(dev);
1548
1549 if (unlikely(dev->features & NETIF_F_LRO))
1550 netdev_WARN(dev, "failed to disable LRO!\n");
1551
1552 netdev_for_each_lower_dev(dev, lower_dev, iter)
1553 dev_disable_lro(lower_dev);
1554}
1555EXPORT_SYMBOL(dev_disable_lro);
1556
1557/**
1558 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1559 * @dev: device
1560 *
1561 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1562 * called under RTNL. This is needed if Generic XDP is installed on
1563 * the device.
1564 */
1565static void dev_disable_gro_hw(struct net_device *dev)
1566{
1567 dev->wanted_features &= ~NETIF_F_GRO_HW;
1568 netdev_update_features(dev);
1569
1570 if (unlikely(dev->features & NETIF_F_GRO_HW))
1571 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1572}
1573
1574static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1575 struct net_device *dev)
1576{
1577 struct netdev_notifier_info info = {
1578 .dev = dev,
1579 };
1580
1581 return nb->notifier_call(nb, val, &info);
1582}
1583
1584static int dev_boot_phase = 1;
1585
1586/**
1587 * register_netdevice_notifier - register a network notifier block
1588 * @nb: notifier
1589 *
1590 * Register a notifier to be called when network device events occur.
1591 * The notifier passed is linked into the kernel structures and must
1592 * not be reused until it has been unregistered. A negative errno code
1593 * is returned on a failure.
1594 *
1595 * When registered all registration and up events are replayed
1596 * to the new notifier to allow device to have a race free
1597 * view of the network device list.
1598 */
1599
1600int register_netdevice_notifier(struct notifier_block *nb)
1601{
1602 struct net_device *dev;
1603 struct net_device *last;
1604 struct net *net;
1605 int err;
1606
1607 rtnl_lock();
1608 err = raw_notifier_chain_register(&netdev_chain, nb);
1609 if (err)
1610 goto unlock;
1611 if (dev_boot_phase)
1612 goto unlock;
1613 for_each_net(net) {
1614 for_each_netdev(net, dev) {
1615 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1616 err = notifier_to_errno(err);
1617 if (err)
1618 goto rollback;
1619
1620 if (!(dev->flags & IFF_UP))
1621 continue;
1622
1623 call_netdevice_notifier(nb, NETDEV_UP, dev);
1624 }
1625 }
1626
1627unlock:
1628 rtnl_unlock();
1629 return err;
1630
1631rollback:
1632 last = dev;
1633 for_each_net(net) {
1634 for_each_netdev(net, dev) {
1635 if (dev == last)
1636 goto outroll;
1637
1638 if (dev->flags & IFF_UP) {
1639 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1640 dev);
1641 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1642 }
1643 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1644 }
1645 }
1646
1647outroll:
1648 raw_notifier_chain_unregister(&netdev_chain, nb);
1649 goto unlock;
1650}
1651EXPORT_SYMBOL(register_netdevice_notifier);
1652
1653/**
1654 * unregister_netdevice_notifier - unregister a network notifier block
1655 * @nb: notifier
1656 *
1657 * Unregister a notifier previously registered by
1658 * register_netdevice_notifier(). The notifier is unlinked into the
1659 * kernel structures and may then be reused. A negative errno code
1660 * is returned on a failure.
1661 *
1662 * After unregistering unregister and down device events are synthesized
1663 * for all devices on the device list to the removed notifier to remove
1664 * the need for special case cleanup code.
1665 */
1666
1667int unregister_netdevice_notifier(struct notifier_block *nb)
1668{
1669 struct net_device *dev;
1670 struct net *net;
1671 int err;
1672
1673 rtnl_lock();
1674 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1675 if (err)
1676 goto unlock;
1677
1678 for_each_net(net) {
1679 for_each_netdev(net, dev) {
1680 if (dev->flags & IFF_UP) {
1681 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1682 dev);
1683 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1684 }
1685 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1686 }
1687 }
1688unlock:
1689 rtnl_unlock();
1690 return err;
1691}
1692EXPORT_SYMBOL(unregister_netdevice_notifier);
1693
1694/**
1695 * call_netdevice_notifiers_info - call all network notifier blocks
1696 * @val: value passed unmodified to notifier function
1697 * @info: notifier information data
1698 *
1699 * Call all network notifier blocks. Parameters and return value
1700 * are as for raw_notifier_call_chain().
1701 */
1702
1703static int call_netdevice_notifiers_info(unsigned long val,
1704 struct netdev_notifier_info *info)
1705{
1706 ASSERT_RTNL();
1707 return raw_notifier_call_chain(&netdev_chain, val, info);
1708}
1709
1710/**
1711 * call_netdevice_notifiers - call all network notifier blocks
1712 * @val: value passed unmodified to notifier function
1713 * @dev: net_device pointer passed unmodified to notifier function
1714 *
1715 * Call all network notifier blocks. Parameters and return value
1716 * are as for raw_notifier_call_chain().
1717 */
1718
1719int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1720{
1721 struct netdev_notifier_info info = {
1722 .dev = dev,
1723 };
1724
1725 return call_netdevice_notifiers_info(val, &info);
1726}
1727EXPORT_SYMBOL(call_netdevice_notifiers);
1728
1729#ifdef CONFIG_NET_INGRESS
1730static struct static_key ingress_needed __read_mostly;
1731
1732void net_inc_ingress_queue(void)
1733{
1734 static_key_slow_inc(&ingress_needed);
1735}
1736EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1737
1738void net_dec_ingress_queue(void)
1739{
1740 static_key_slow_dec(&ingress_needed);
1741}
1742EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1743#endif
1744
1745#ifdef CONFIG_NET_EGRESS
1746static struct static_key egress_needed __read_mostly;
1747
1748void net_inc_egress_queue(void)
1749{
1750 static_key_slow_inc(&egress_needed);
1751}
1752EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1753
1754void net_dec_egress_queue(void)
1755{
1756 static_key_slow_dec(&egress_needed);
1757}
1758EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1759#endif
1760
1761static struct static_key netstamp_needed __read_mostly;
1762#ifdef HAVE_JUMP_LABEL
1763static atomic_t netstamp_needed_deferred;
1764static atomic_t netstamp_wanted;
1765static void netstamp_clear(struct work_struct *work)
1766{
1767 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1768 int wanted;
1769
1770 wanted = atomic_add_return(deferred, &netstamp_wanted);
1771 if (wanted > 0)
1772 static_key_enable(&netstamp_needed);
1773 else
1774 static_key_disable(&netstamp_needed);
1775}
1776static DECLARE_WORK(netstamp_work, netstamp_clear);
1777#endif
1778
1779void net_enable_timestamp(void)
1780{
1781#ifdef HAVE_JUMP_LABEL
1782 int wanted;
1783
1784 while (1) {
1785 wanted = atomic_read(&netstamp_wanted);
1786 if (wanted <= 0)
1787 break;
1788 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1789 return;
1790 }
1791 atomic_inc(&netstamp_needed_deferred);
1792 schedule_work(&netstamp_work);
1793#else
1794 static_key_slow_inc(&netstamp_needed);
1795#endif
1796}
1797EXPORT_SYMBOL(net_enable_timestamp);
1798
1799void net_disable_timestamp(void)
1800{
1801#ifdef HAVE_JUMP_LABEL
1802 int wanted;
1803
1804 while (1) {
1805 wanted = atomic_read(&netstamp_wanted);
1806 if (wanted <= 1)
1807 break;
1808 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1809 return;
1810 }
1811 atomic_dec(&netstamp_needed_deferred);
1812 schedule_work(&netstamp_work);
1813#else
1814 static_key_slow_dec(&netstamp_needed);
1815#endif
1816}
1817EXPORT_SYMBOL(net_disable_timestamp);
1818
1819static inline void net_timestamp_set(struct sk_buff *skb)
1820{
1821 skb->tstamp = 0;
1822 if (static_key_false(&netstamp_needed))
1823 __net_timestamp(skb);
1824}
1825
1826#define net_timestamp_check(COND, SKB) \
1827 if (static_key_false(&netstamp_needed)) { \
1828 if ((COND) && !(SKB)->tstamp) \
1829 __net_timestamp(SKB); \
1830 } \
1831
1832bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1833{
1834 unsigned int len;
1835
1836 if (!(dev->flags & IFF_UP))
1837 return false;
1838
1839 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1840 if (skb->len <= len)
1841 return true;
1842
1843 /* if TSO is enabled, we don't care about the length as the packet
1844 * could be forwarded without being segmented before
1845 */
1846 if (skb_is_gso(skb))
1847 return true;
1848
1849 return false;
1850}
1851EXPORT_SYMBOL_GPL(is_skb_forwardable);
1852
1853int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1854{
1855 int ret = ____dev_forward_skb(dev, skb);
1856
1857 if (likely(!ret)) {
1858 skb->protocol = eth_type_trans(skb, dev);
1859 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1860 }
1861
1862 return ret;
1863}
1864EXPORT_SYMBOL_GPL(__dev_forward_skb);
1865
1866/**
1867 * dev_forward_skb - loopback an skb to another netif
1868 *
1869 * @dev: destination network device
1870 * @skb: buffer to forward
1871 *
1872 * return values:
1873 * NET_RX_SUCCESS (no congestion)
1874 * NET_RX_DROP (packet was dropped, but freed)
1875 *
1876 * dev_forward_skb can be used for injecting an skb from the
1877 * start_xmit function of one device into the receive queue
1878 * of another device.
1879 *
1880 * The receiving device may be in another namespace, so
1881 * we have to clear all information in the skb that could
1882 * impact namespace isolation.
1883 */
1884int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1885{
1886 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1887}
1888EXPORT_SYMBOL_GPL(dev_forward_skb);
1889
1890static inline int deliver_skb(struct sk_buff *skb,
1891 struct packet_type *pt_prev,
1892 struct net_device *orig_dev)
1893{
1894 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1895 return -ENOMEM;
1896 refcount_inc(&skb->users);
1897 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1898}
1899
1900static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1901 struct packet_type **pt,
1902 struct net_device *orig_dev,
1903 __be16 type,
1904 struct list_head *ptype_list)
1905{
1906 struct packet_type *ptype, *pt_prev = *pt;
1907
1908 list_for_each_entry_rcu(ptype, ptype_list, list) {
1909 if (ptype->type != type)
1910 continue;
1911 if (pt_prev)
1912 deliver_skb(skb, pt_prev, orig_dev);
1913 pt_prev = ptype;
1914 }
1915 *pt = pt_prev;
1916}
1917
1918static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1919{
1920 if (!ptype->af_packet_priv || !skb->sk)
1921 return false;
1922
1923 if (ptype->id_match)
1924 return ptype->id_match(ptype, skb->sk);
1925 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1926 return true;
1927
1928 return false;
1929}
1930
1931/*
1932 * Support routine. Sends outgoing frames to any network
1933 * taps currently in use.
1934 */
1935
1936void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1937{
1938 struct packet_type *ptype;
1939 struct sk_buff *skb2 = NULL;
1940 struct packet_type *pt_prev = NULL;
1941 struct list_head *ptype_list = &ptype_all;
1942
1943 rcu_read_lock();
1944again:
1945 list_for_each_entry_rcu(ptype, ptype_list, list) {
1946 /* Never send packets back to the socket
1947 * they originated from - MvS (miquels@drinkel.ow.org)
1948 */
1949 if (skb_loop_sk(ptype, skb))
1950 continue;
1951
1952 if (pt_prev) {
1953 deliver_skb(skb2, pt_prev, skb->dev);
1954 pt_prev = ptype;
1955 continue;
1956 }
1957
1958 /* need to clone skb, done only once */
1959 skb2 = skb_clone(skb, GFP_ATOMIC);
1960 if (!skb2)
1961 goto out_unlock;
1962
1963 net_timestamp_set(skb2);
1964
1965 /* skb->nh should be correctly
1966 * set by sender, so that the second statement is
1967 * just protection against buggy protocols.
1968 */
1969 skb_reset_mac_header(skb2);
1970
1971 if (skb_network_header(skb2) < skb2->data ||
1972 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1973 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1974 ntohs(skb2->protocol),
1975 dev->name);
1976 skb_reset_network_header(skb2);
1977 }
1978
1979 skb2->transport_header = skb2->network_header;
1980 skb2->pkt_type = PACKET_OUTGOING;
1981 pt_prev = ptype;
1982 }
1983
1984 if (ptype_list == &ptype_all) {
1985 ptype_list = &dev->ptype_all;
1986 goto again;
1987 }
1988out_unlock:
1989 if (pt_prev) {
1990 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
1991 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1992 else
1993 kfree_skb(skb2);
1994 }
1995 rcu_read_unlock();
1996}
1997EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1998
1999/**
2000 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2001 * @dev: Network device
2002 * @txq: number of queues available
2003 *
2004 * If real_num_tx_queues is changed the tc mappings may no longer be
2005 * valid. To resolve this verify the tc mapping remains valid and if
2006 * not NULL the mapping. With no priorities mapping to this
2007 * offset/count pair it will no longer be used. In the worst case TC0
2008 * is invalid nothing can be done so disable priority mappings. If is
2009 * expected that drivers will fix this mapping if they can before
2010 * calling netif_set_real_num_tx_queues.
2011 */
2012static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2013{
2014 int i;
2015 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2016
2017 /* If TC0 is invalidated disable TC mapping */
2018 if (tc->offset + tc->count > txq) {
2019 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2020 dev->num_tc = 0;
2021 return;
2022 }
2023
2024 /* Invalidated prio to tc mappings set to TC0 */
2025 for (i = 1; i < TC_BITMASK + 1; i++) {
2026 int q = netdev_get_prio_tc_map(dev, i);
2027
2028 tc = &dev->tc_to_txq[q];
2029 if (tc->offset + tc->count > txq) {
2030 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2031 i, q);
2032 netdev_set_prio_tc_map(dev, i, 0);
2033 }
2034 }
2035}
2036
2037int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2038{
2039 if (dev->num_tc) {
2040 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2041 int i;
2042
2043 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2044 if ((txq - tc->offset) < tc->count)
2045 return i;
2046 }
2047
2048 return -1;
2049 }
2050
2051 return 0;
2052}
2053EXPORT_SYMBOL(netdev_txq_to_tc);
2054
2055#ifdef CONFIG_XPS
2056static DEFINE_MUTEX(xps_map_mutex);
2057#define xmap_dereference(P) \
2058 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2059
2060static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2061 int tci, u16 index)
2062{
2063 struct xps_map *map = NULL;
2064 int pos;
2065
2066 if (dev_maps)
2067 map = xmap_dereference(dev_maps->cpu_map[tci]);
2068 if (!map)
2069 return false;
2070
2071 for (pos = map->len; pos--;) {
2072 if (map->queues[pos] != index)
2073 continue;
2074
2075 if (map->len > 1) {
2076 map->queues[pos] = map->queues[--map->len];
2077 break;
2078 }
2079
2080 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
2081 kfree_rcu(map, rcu);
2082 return false;
2083 }
2084
2085 return true;
2086}
2087
2088static bool remove_xps_queue_cpu(struct net_device *dev,
2089 struct xps_dev_maps *dev_maps,
2090 int cpu, u16 offset, u16 count)
2091{
2092 int num_tc = dev->num_tc ? : 1;
2093 bool active = false;
2094 int tci;
2095
2096 for (tci = cpu * num_tc; num_tc--; tci++) {
2097 int i, j;
2098
2099 for (i = count, j = offset; i--; j++) {
2100 if (!remove_xps_queue(dev_maps, cpu, j))
2101 break;
2102 }
2103
2104 active |= i < 0;
2105 }
2106
2107 return active;
2108}
2109
2110static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2111 u16 count)
2112{
2113 struct xps_dev_maps *dev_maps;
2114 int cpu, i;
2115 bool active = false;
2116
2117 mutex_lock(&xps_map_mutex);
2118 dev_maps = xmap_dereference(dev->xps_maps);
2119
2120 if (!dev_maps)
2121 goto out_no_maps;
2122
2123 for_each_possible_cpu(cpu)
2124 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2125 offset, count);
2126
2127 if (!active) {
2128 RCU_INIT_POINTER(dev->xps_maps, NULL);
2129 kfree_rcu(dev_maps, rcu);
2130 }
2131
2132 for (i = offset + (count - 1); count--; i--)
2133 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2134 NUMA_NO_NODE);
2135
2136out_no_maps:
2137 mutex_unlock(&xps_map_mutex);
2138}
2139
2140static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2141{
2142 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2143}
2144
2145static struct xps_map *expand_xps_map(struct xps_map *map,
2146 int cpu, u16 index)
2147{
2148 struct xps_map *new_map;
2149 int alloc_len = XPS_MIN_MAP_ALLOC;
2150 int i, pos;
2151
2152 for (pos = 0; map && pos < map->len; pos++) {
2153 if (map->queues[pos] != index)
2154 continue;
2155 return map;
2156 }
2157
2158 /* Need to add queue to this CPU's existing map */
2159 if (map) {
2160 if (pos < map->alloc_len)
2161 return map;
2162
2163 alloc_len = map->alloc_len * 2;
2164 }
2165
2166 /* Need to allocate new map to store queue on this CPU's map */
2167 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2168 cpu_to_node(cpu));
2169 if (!new_map)
2170 return NULL;
2171
2172 for (i = 0; i < pos; i++)
2173 new_map->queues[i] = map->queues[i];
2174 new_map->alloc_len = alloc_len;
2175 new_map->len = pos;
2176
2177 return new_map;
2178}
2179
2180int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2181 u16 index)
2182{
2183 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2184 int i, cpu, tci, numa_node_id = -2;
2185 int maps_sz, num_tc = 1, tc = 0;
2186 struct xps_map *map, *new_map;
2187 bool active = false;
2188
2189 if (dev->num_tc) {
2190 num_tc = dev->num_tc;
2191 tc = netdev_txq_to_tc(dev, index);
2192 if (tc < 0)
2193 return -EINVAL;
2194 }
2195
2196 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2197 if (maps_sz < L1_CACHE_BYTES)
2198 maps_sz = L1_CACHE_BYTES;
2199
2200 mutex_lock(&xps_map_mutex);
2201
2202 dev_maps = xmap_dereference(dev->xps_maps);
2203
2204 /* allocate memory for queue storage */
2205 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2206 if (!new_dev_maps)
2207 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2208 if (!new_dev_maps) {
2209 mutex_unlock(&xps_map_mutex);
2210 return -ENOMEM;
2211 }
2212
2213 tci = cpu * num_tc + tc;
2214 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2215 NULL;
2216
2217 map = expand_xps_map(map, cpu, index);
2218 if (!map)
2219 goto error;
2220
2221 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2222 }
2223
2224 if (!new_dev_maps)
2225 goto out_no_new_maps;
2226
2227 for_each_possible_cpu(cpu) {
2228 /* copy maps belonging to foreign traffic classes */
2229 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2230 /* fill in the new device map from the old device map */
2231 map = xmap_dereference(dev_maps->cpu_map[tci]);
2232 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2233 }
2234
2235 /* We need to explicitly update tci as prevous loop
2236 * could break out early if dev_maps is NULL.
2237 */
2238 tci = cpu * num_tc + tc;
2239
2240 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2241 /* add queue to CPU maps */
2242 int pos = 0;
2243
2244 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2245 while ((pos < map->len) && (map->queues[pos] != index))
2246 pos++;
2247
2248 if (pos == map->len)
2249 map->queues[map->len++] = index;
2250#ifdef CONFIG_NUMA
2251 if (numa_node_id == -2)
2252 numa_node_id = cpu_to_node(cpu);
2253 else if (numa_node_id != cpu_to_node(cpu))
2254 numa_node_id = -1;
2255#endif
2256 } else if (dev_maps) {
2257 /* fill in the new device map from the old device map */
2258 map = xmap_dereference(dev_maps->cpu_map[tci]);
2259 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2260 }
2261
2262 /* copy maps belonging to foreign traffic classes */
2263 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2264 /* fill in the new device map from the old device map */
2265 map = xmap_dereference(dev_maps->cpu_map[tci]);
2266 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2267 }
2268 }
2269
2270 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2271
2272 /* Cleanup old maps */
2273 if (!dev_maps)
2274 goto out_no_old_maps;
2275
2276 for_each_possible_cpu(cpu) {
2277 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2278 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2279 map = xmap_dereference(dev_maps->cpu_map[tci]);
2280 if (map && map != new_map)
2281 kfree_rcu(map, rcu);
2282 }
2283 }
2284
2285 kfree_rcu(dev_maps, rcu);
2286
2287out_no_old_maps:
2288 dev_maps = new_dev_maps;
2289 active = true;
2290
2291out_no_new_maps:
2292 /* update Tx queue numa node */
2293 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2294 (numa_node_id >= 0) ? numa_node_id :
2295 NUMA_NO_NODE);
2296
2297 if (!dev_maps)
2298 goto out_no_maps;
2299
2300 /* removes queue from unused CPUs */
2301 for_each_possible_cpu(cpu) {
2302 for (i = tc, tci = cpu * num_tc; i--; tci++)
2303 active |= remove_xps_queue(dev_maps, tci, index);
2304 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2305 active |= remove_xps_queue(dev_maps, tci, index);
2306 for (i = num_tc - tc, tci++; --i; tci++)
2307 active |= remove_xps_queue(dev_maps, tci, index);
2308 }
2309
2310 /* free map if not active */
2311 if (!active) {
2312 RCU_INIT_POINTER(dev->xps_maps, NULL);
2313 kfree_rcu(dev_maps, rcu);
2314 }
2315
2316out_no_maps:
2317 mutex_unlock(&xps_map_mutex);
2318
2319 return 0;
2320error:
2321 /* remove any maps that we added */
2322 for_each_possible_cpu(cpu) {
2323 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2324 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2325 map = dev_maps ?
2326 xmap_dereference(dev_maps->cpu_map[tci]) :
2327 NULL;
2328 if (new_map && new_map != map)
2329 kfree(new_map);
2330 }
2331 }
2332
2333 mutex_unlock(&xps_map_mutex);
2334
2335 kfree(new_dev_maps);
2336 return -ENOMEM;
2337}
2338EXPORT_SYMBOL(netif_set_xps_queue);
2339
2340#endif
2341void netdev_reset_tc(struct net_device *dev)
2342{
2343#ifdef CONFIG_XPS
2344 netif_reset_xps_queues_gt(dev, 0);
2345#endif
2346 dev->num_tc = 0;
2347 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2348 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2349}
2350EXPORT_SYMBOL(netdev_reset_tc);
2351
2352int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2353{
2354 if (tc >= dev->num_tc)
2355 return -EINVAL;
2356
2357#ifdef CONFIG_XPS
2358 netif_reset_xps_queues(dev, offset, count);
2359#endif
2360 dev->tc_to_txq[tc].count = count;
2361 dev->tc_to_txq[tc].offset = offset;
2362 return 0;
2363}
2364EXPORT_SYMBOL(netdev_set_tc_queue);
2365
2366int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2367{
2368 if (num_tc > TC_MAX_QUEUE)
2369 return -EINVAL;
2370
2371#ifdef CONFIG_XPS
2372 netif_reset_xps_queues_gt(dev, 0);
2373#endif
2374 dev->num_tc = num_tc;
2375 return 0;
2376}
2377EXPORT_SYMBOL(netdev_set_num_tc);
2378
2379/*
2380 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2381 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2382 */
2383int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2384{
2385 bool disabling;
2386 int rc;
2387
2388 disabling = txq < dev->real_num_tx_queues;
2389
2390 if (txq < 1 || txq > dev->num_tx_queues)
2391 return -EINVAL;
2392
2393 if (dev->reg_state == NETREG_REGISTERED ||
2394 dev->reg_state == NETREG_UNREGISTERING) {
2395 ASSERT_RTNL();
2396
2397 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2398 txq);
2399 if (rc)
2400 return rc;
2401
2402 if (dev->num_tc)
2403 netif_setup_tc(dev, txq);
2404
2405 dev->real_num_tx_queues = txq;
2406
2407 if (disabling) {
2408 synchronize_net();
2409 qdisc_reset_all_tx_gt(dev, txq);
2410#ifdef CONFIG_XPS
2411 netif_reset_xps_queues_gt(dev, txq);
2412#endif
2413 }
2414 } else {
2415 dev->real_num_tx_queues = txq;
2416 }
2417
2418 return 0;
2419}
2420EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2421
2422#ifdef CONFIG_SYSFS
2423/**
2424 * netif_set_real_num_rx_queues - set actual number of RX queues used
2425 * @dev: Network device
2426 * @rxq: Actual number of RX queues
2427 *
2428 * This must be called either with the rtnl_lock held or before
2429 * registration of the net device. Returns 0 on success, or a
2430 * negative error code. If called before registration, it always
2431 * succeeds.
2432 */
2433int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2434{
2435 int rc;
2436
2437 if (rxq < 1 || rxq > dev->num_rx_queues)
2438 return -EINVAL;
2439
2440 if (dev->reg_state == NETREG_REGISTERED) {
2441 ASSERT_RTNL();
2442
2443 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2444 rxq);
2445 if (rc)
2446 return rc;
2447 }
2448
2449 dev->real_num_rx_queues = rxq;
2450 return 0;
2451}
2452EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2453#endif
2454
2455/**
2456 * netif_get_num_default_rss_queues - default number of RSS queues
2457 *
2458 * This routine should set an upper limit on the number of RSS queues
2459 * used by default by multiqueue devices.
2460 */
2461int netif_get_num_default_rss_queues(void)
2462{
2463 return is_kdump_kernel() ?
2464 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2465}
2466EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2467
2468static void __netif_reschedule(struct Qdisc *q)
2469{
2470 struct softnet_data *sd;
2471 unsigned long flags;
2472
2473 local_irq_save(flags);
2474 sd = this_cpu_ptr(&softnet_data);
2475 q->next_sched = NULL;
2476 *sd->output_queue_tailp = q;
2477 sd->output_queue_tailp = &q->next_sched;
2478 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2479 local_irq_restore(flags);
2480}
2481
2482void __netif_schedule(struct Qdisc *q)
2483{
2484 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2485 __netif_reschedule(q);
2486}
2487EXPORT_SYMBOL(__netif_schedule);
2488
2489struct dev_kfree_skb_cb {
2490 enum skb_free_reason reason;
2491};
2492
2493static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2494{
2495 return (struct dev_kfree_skb_cb *)skb->cb;
2496}
2497
2498void netif_schedule_queue(struct netdev_queue *txq)
2499{
2500 rcu_read_lock();
2501 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2502 struct Qdisc *q = rcu_dereference(txq->qdisc);
2503
2504 __netif_schedule(q);
2505 }
2506 rcu_read_unlock();
2507}
2508EXPORT_SYMBOL(netif_schedule_queue);
2509
2510void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2511{
2512 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2513 struct Qdisc *q;
2514
2515 rcu_read_lock();
2516 q = rcu_dereference(dev_queue->qdisc);
2517 __netif_schedule(q);
2518 rcu_read_unlock();
2519 }
2520}
2521EXPORT_SYMBOL(netif_tx_wake_queue);
2522
2523void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2524{
2525 unsigned long flags;
2526
2527 if (unlikely(!skb))
2528 return;
2529
2530 if (likely(refcount_read(&skb->users) == 1)) {
2531 smp_rmb();
2532 refcount_set(&skb->users, 0);
2533 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2534 return;
2535 }
2536 get_kfree_skb_cb(skb)->reason = reason;
2537 local_irq_save(flags);
2538 skb->next = __this_cpu_read(softnet_data.completion_queue);
2539 __this_cpu_write(softnet_data.completion_queue, skb);
2540 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2541 local_irq_restore(flags);
2542}
2543EXPORT_SYMBOL(__dev_kfree_skb_irq);
2544
2545void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2546{
2547 if (in_irq() || irqs_disabled())
2548 __dev_kfree_skb_irq(skb, reason);
2549 else
2550 dev_kfree_skb(skb);
2551}
2552EXPORT_SYMBOL(__dev_kfree_skb_any);
2553
2554
2555/**
2556 * netif_device_detach - mark device as removed
2557 * @dev: network device
2558 *
2559 * Mark device as removed from system and therefore no longer available.
2560 */
2561void netif_device_detach(struct net_device *dev)
2562{
2563 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2564 netif_running(dev)) {
2565 netif_tx_stop_all_queues(dev);
2566 }
2567}
2568EXPORT_SYMBOL(netif_device_detach);
2569
2570/**
2571 * netif_device_attach - mark device as attached
2572 * @dev: network device
2573 *
2574 * Mark device as attached from system and restart if needed.
2575 */
2576void netif_device_attach(struct net_device *dev)
2577{
2578 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2579 netif_running(dev)) {
2580 netif_tx_wake_all_queues(dev);
2581 __netdev_watchdog_up(dev);
2582 }
2583}
2584EXPORT_SYMBOL(netif_device_attach);
2585
2586/*
2587 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2588 * to be used as a distribution range.
2589 */
2590u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2591 unsigned int num_tx_queues)
2592{
2593 u32 hash;
2594 u16 qoffset = 0;
2595 u16 qcount = num_tx_queues;
2596
2597 if (skb_rx_queue_recorded(skb)) {
2598 hash = skb_get_rx_queue(skb);
2599 while (unlikely(hash >= num_tx_queues))
2600 hash -= num_tx_queues;
2601 return hash;
2602 }
2603
2604 if (dev->num_tc) {
2605 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2606
2607 qoffset = dev->tc_to_txq[tc].offset;
2608 qcount = dev->tc_to_txq[tc].count;
2609 }
2610
2611 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2612}
2613EXPORT_SYMBOL(__skb_tx_hash);
2614
2615static void skb_warn_bad_offload(const struct sk_buff *skb)
2616{
2617 static const netdev_features_t null_features;
2618 struct net_device *dev = skb->dev;
2619 const char *name = "";
2620
2621 if (!net_ratelimit())
2622 return;
2623
2624 if (dev) {
2625 if (dev->dev.parent)
2626 name = dev_driver_string(dev->dev.parent);
2627 else
2628 name = netdev_name(dev);
2629 }
2630 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2631 "gso_type=%d ip_summed=%d\n",
2632 name, dev ? &dev->features : &null_features,
2633 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2634 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2635 skb_shinfo(skb)->gso_type, skb->ip_summed);
2636}
2637
2638/*
2639 * Invalidate hardware checksum when packet is to be mangled, and
2640 * complete checksum manually on outgoing path.
2641 */
2642int skb_checksum_help(struct sk_buff *skb)
2643{
2644 __wsum csum;
2645 int ret = 0, offset;
2646
2647 if (skb->ip_summed == CHECKSUM_COMPLETE)
2648 goto out_set_summed;
2649
2650 if (unlikely(skb_shinfo(skb)->gso_size)) {
2651 skb_warn_bad_offload(skb);
2652 return -EINVAL;
2653 }
2654
2655 /* Before computing a checksum, we should make sure no frag could
2656 * be modified by an external entity : checksum could be wrong.
2657 */
2658 if (skb_has_shared_frag(skb)) {
2659 ret = __skb_linearize(skb);
2660 if (ret)
2661 goto out;
2662 }
2663
2664 offset = skb_checksum_start_offset(skb);
2665 BUG_ON(offset >= skb_headlen(skb));
2666 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2667
2668 offset += skb->csum_offset;
2669 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2670
2671 if (skb_cloned(skb) &&
2672 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2673 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2674 if (ret)
2675 goto out;
2676 }
2677
2678 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2679out_set_summed:
2680 skb->ip_summed = CHECKSUM_NONE;
2681out:
2682 return ret;
2683}
2684EXPORT_SYMBOL(skb_checksum_help);
2685
2686int skb_crc32c_csum_help(struct sk_buff *skb)
2687{
2688 __le32 crc32c_csum;
2689 int ret = 0, offset, start;
2690
2691 if (skb->ip_summed != CHECKSUM_PARTIAL)
2692 goto out;
2693
2694 if (unlikely(skb_is_gso(skb)))
2695 goto out;
2696
2697 /* Before computing a checksum, we should make sure no frag could
2698 * be modified by an external entity : checksum could be wrong.
2699 */
2700 if (unlikely(skb_has_shared_frag(skb))) {
2701 ret = __skb_linearize(skb);
2702 if (ret)
2703 goto out;
2704 }
2705 start = skb_checksum_start_offset(skb);
2706 offset = start + offsetof(struct sctphdr, checksum);
2707 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2708 ret = -EINVAL;
2709 goto out;
2710 }
2711 if (skb_cloned(skb) &&
2712 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2713 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2714 if (ret)
2715 goto out;
2716 }
2717 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2718 skb->len - start, ~(__u32)0,
2719 crc32c_csum_stub));
2720 *(__le32 *)(skb->data + offset) = crc32c_csum;
2721 skb->ip_summed = CHECKSUM_NONE;
2722 skb->csum_not_inet = 0;
2723out:
2724 return ret;
2725}
2726
2727__be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2728{
2729 __be16 type = skb->protocol;
2730
2731 /* Tunnel gso handlers can set protocol to ethernet. */
2732 if (type == htons(ETH_P_TEB)) {
2733 struct ethhdr *eth;
2734
2735 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2736 return 0;
2737
2738 eth = (struct ethhdr *)skb_mac_header(skb);
2739 type = eth->h_proto;
2740 }
2741
2742 return __vlan_get_protocol(skb, type, depth);
2743}
2744
2745/**
2746 * skb_mac_gso_segment - mac layer segmentation handler.
2747 * @skb: buffer to segment
2748 * @features: features for the output path (see dev->features)
2749 */
2750struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2751 netdev_features_t features)
2752{
2753 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2754 struct packet_offload *ptype;
2755 int vlan_depth = skb->mac_len;
2756 __be16 type = skb_network_protocol(skb, &vlan_depth);
2757
2758 if (unlikely(!type))
2759 return ERR_PTR(-EINVAL);
2760
2761 __skb_pull(skb, vlan_depth);
2762
2763 rcu_read_lock();
2764 list_for_each_entry_rcu(ptype, &offload_base, list) {
2765 if (ptype->type == type && ptype->callbacks.gso_segment) {
2766 segs = ptype->callbacks.gso_segment(skb, features);
2767 break;
2768 }
2769 }
2770 rcu_read_unlock();
2771
2772 __skb_push(skb, skb->data - skb_mac_header(skb));
2773
2774 return segs;
2775}
2776EXPORT_SYMBOL(skb_mac_gso_segment);
2777
2778
2779/* openvswitch calls this on rx path, so we need a different check.
2780 */
2781static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2782{
2783 if (tx_path)
2784 return skb->ip_summed != CHECKSUM_PARTIAL &&
2785 skb->ip_summed != CHECKSUM_UNNECESSARY;
2786
2787 return skb->ip_summed == CHECKSUM_NONE;
2788}
2789
2790/**
2791 * __skb_gso_segment - Perform segmentation on skb.
2792 * @skb: buffer to segment
2793 * @features: features for the output path (see dev->features)
2794 * @tx_path: whether it is called in TX path
2795 *
2796 * This function segments the given skb and returns a list of segments.
2797 *
2798 * It may return NULL if the skb requires no segmentation. This is
2799 * only possible when GSO is used for verifying header integrity.
2800 *
2801 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2802 */
2803struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2804 netdev_features_t features, bool tx_path)
2805{
2806 struct sk_buff *segs;
2807
2808 if (unlikely(skb_needs_check(skb, tx_path))) {
2809 int err;
2810
2811 /* We're going to init ->check field in TCP or UDP header */
2812 err = skb_cow_head(skb, 0);
2813 if (err < 0)
2814 return ERR_PTR(err);
2815 }
2816
2817 /* Only report GSO partial support if it will enable us to
2818 * support segmentation on this frame without needing additional
2819 * work.
2820 */
2821 if (features & NETIF_F_GSO_PARTIAL) {
2822 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2823 struct net_device *dev = skb->dev;
2824
2825 partial_features |= dev->features & dev->gso_partial_features;
2826 if (!skb_gso_ok(skb, features | partial_features))
2827 features &= ~NETIF_F_GSO_PARTIAL;
2828 }
2829
2830 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2831 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2832
2833 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2834 SKB_GSO_CB(skb)->encap_level = 0;
2835
2836 skb_reset_mac_header(skb);
2837 skb_reset_mac_len(skb);
2838
2839 segs = skb_mac_gso_segment(skb, features);
2840
2841 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
2842 skb_warn_bad_offload(skb);
2843
2844 return segs;
2845}
2846EXPORT_SYMBOL(__skb_gso_segment);
2847
2848/* Take action when hardware reception checksum errors are detected. */
2849#ifdef CONFIG_BUG
2850void netdev_rx_csum_fault(struct net_device *dev)
2851{
2852 if (net_ratelimit()) {
2853 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2854 dump_stack();
2855 }
2856}
2857EXPORT_SYMBOL(netdev_rx_csum_fault);
2858#endif
2859
2860/* Actually, we should eliminate this check as soon as we know, that:
2861 * 1. IOMMU is present and allows to map all the memory.
2862 * 2. No high memory really exists on this machine.
2863 */
2864
2865static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2866{
2867#ifdef CONFIG_HIGHMEM
2868 int i;
2869
2870 if (!(dev->features & NETIF_F_HIGHDMA)) {
2871 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2872 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2873
2874 if (PageHighMem(skb_frag_page(frag)))
2875 return 1;
2876 }
2877 }
2878
2879 if (PCI_DMA_BUS_IS_PHYS) {
2880 struct device *pdev = dev->dev.parent;
2881
2882 if (!pdev)
2883 return 0;
2884 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2885 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2886 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2887
2888 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2889 return 1;
2890 }
2891 }
2892#endif
2893 return 0;
2894}
2895
2896/* If MPLS offload request, verify we are testing hardware MPLS features
2897 * instead of standard features for the netdev.
2898 */
2899#if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2900static netdev_features_t net_mpls_features(struct sk_buff *skb,
2901 netdev_features_t features,
2902 __be16 type)
2903{
2904 if (eth_p_mpls(type))
2905 features &= skb->dev->mpls_features;
2906
2907 return features;
2908}
2909#else
2910static netdev_features_t net_mpls_features(struct sk_buff *skb,
2911 netdev_features_t features,
2912 __be16 type)
2913{
2914 return features;
2915}
2916#endif
2917
2918static netdev_features_t harmonize_features(struct sk_buff *skb,
2919 netdev_features_t features)
2920{
2921 int tmp;
2922 __be16 type;
2923
2924 type = skb_network_protocol(skb, &tmp);
2925 features = net_mpls_features(skb, features, type);
2926
2927 if (skb->ip_summed != CHECKSUM_NONE &&
2928 !can_checksum_protocol(features, type)) {
2929 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2930 }
2931 if (illegal_highdma(skb->dev, skb))
2932 features &= ~NETIF_F_SG;
2933
2934 return features;
2935}
2936
2937netdev_features_t passthru_features_check(struct sk_buff *skb,
2938 struct net_device *dev,
2939 netdev_features_t features)
2940{
2941 return features;
2942}
2943EXPORT_SYMBOL(passthru_features_check);
2944
2945static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2946 struct net_device *dev,
2947 netdev_features_t features)
2948{
2949 return vlan_features_check(skb, features);
2950}
2951
2952static netdev_features_t gso_features_check(const struct sk_buff *skb,
2953 struct net_device *dev,
2954 netdev_features_t features)
2955{
2956 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2957
2958 if (gso_segs > dev->gso_max_segs)
2959 return features & ~NETIF_F_GSO_MASK;
2960
2961 /* Support for GSO partial features requires software
2962 * intervention before we can actually process the packets
2963 * so we need to strip support for any partial features now
2964 * and we can pull them back in after we have partially
2965 * segmented the frame.
2966 */
2967 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2968 features &= ~dev->gso_partial_features;
2969
2970 /* Make sure to clear the IPv4 ID mangling feature if the
2971 * IPv4 header has the potential to be fragmented.
2972 */
2973 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2974 struct iphdr *iph = skb->encapsulation ?
2975 inner_ip_hdr(skb) : ip_hdr(skb);
2976
2977 if (!(iph->frag_off & htons(IP_DF)))
2978 features &= ~NETIF_F_TSO_MANGLEID;
2979 }
2980
2981 return features;
2982}
2983
2984netdev_features_t netif_skb_features(struct sk_buff *skb)
2985{
2986 struct net_device *dev = skb->dev;
2987 netdev_features_t features = dev->features;
2988
2989 if (skb_is_gso(skb))
2990 features = gso_features_check(skb, dev, features);
2991
2992 /* If encapsulation offload request, verify we are testing
2993 * hardware encapsulation features instead of standard
2994 * features for the netdev
2995 */
2996 if (skb->encapsulation)
2997 features &= dev->hw_enc_features;
2998
2999 if (skb_vlan_tagged(skb))
3000 features = netdev_intersect_features(features,
3001 dev->vlan_features |
3002 NETIF_F_HW_VLAN_CTAG_TX |
3003 NETIF_F_HW_VLAN_STAG_TX);
3004
3005 if (dev->netdev_ops->ndo_features_check)
3006 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3007 features);
3008 else
3009 features &= dflt_features_check(skb, dev, features);
3010
3011 return harmonize_features(skb, features);
3012}
3013EXPORT_SYMBOL(netif_skb_features);
3014
3015static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3016 struct netdev_queue *txq, bool more)
3017{
3018 unsigned int len;
3019 int rc;
3020
3021 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3022 dev_queue_xmit_nit(skb, dev);
3023
3024 len = skb->len;
3025 trace_net_dev_start_xmit(skb, dev);
3026 rc = netdev_start_xmit(skb, dev, txq, more);
3027 trace_net_dev_xmit(skb, rc, dev, len);
3028
3029 return rc;
3030}
3031
3032struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3033 struct netdev_queue *txq, int *ret)
3034{
3035 struct sk_buff *skb = first;
3036 int rc = NETDEV_TX_OK;
3037
3038 while (skb) {
3039 struct sk_buff *next = skb->next;
3040
3041 skb->next = NULL;
3042 rc = xmit_one(skb, dev, txq, next != NULL);
3043 if (unlikely(!dev_xmit_complete(rc))) {
3044 skb->next = next;
3045 goto out;
3046 }
3047
3048 skb = next;
3049 if (netif_xmit_stopped(txq) && skb) {
3050 rc = NETDEV_TX_BUSY;
3051 break;
3052 }
3053 }
3054
3055out:
3056 *ret = rc;
3057 return skb;
3058}
3059
3060static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3061 netdev_features_t features)
3062{
3063 if (skb_vlan_tag_present(skb) &&
3064 !vlan_hw_offload_capable(features, skb->vlan_proto))
3065 skb = __vlan_hwaccel_push_inside(skb);
3066 return skb;
3067}
3068
3069int skb_csum_hwoffload_help(struct sk_buff *skb,
3070 const netdev_features_t features)
3071{
3072 if (unlikely(skb->csum_not_inet))
3073 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3074 skb_crc32c_csum_help(skb);
3075
3076 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3077}
3078EXPORT_SYMBOL(skb_csum_hwoffload_help);
3079
3080static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3081{
3082 netdev_features_t features;
3083
3084 features = netif_skb_features(skb);
3085 skb = validate_xmit_vlan(skb, features);
3086 if (unlikely(!skb))
3087 goto out_null;
3088
3089 if (netif_needs_gso(skb, features)) {
3090 struct sk_buff *segs;
3091
3092 segs = skb_gso_segment(skb, features);
3093 if (IS_ERR(segs)) {
3094 goto out_kfree_skb;
3095 } else if (segs) {
3096 consume_skb(skb);
3097 skb = segs;
3098 }
3099 } else {
3100 if (skb_needs_linearize(skb, features) &&
3101 __skb_linearize(skb))
3102 goto out_kfree_skb;
3103
3104 /* If packet is not checksummed and device does not
3105 * support checksumming for this protocol, complete
3106 * checksumming here.
3107 */
3108 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3109 if (skb->encapsulation)
3110 skb_set_inner_transport_header(skb,
3111 skb_checksum_start_offset(skb));
3112 else
3113 skb_set_transport_header(skb,
3114 skb_checksum_start_offset(skb));
3115 if (skb_csum_hwoffload_help(skb, features))
3116 goto out_kfree_skb;
3117 }
3118 }
3119
3120 skb = validate_xmit_xfrm(skb, features, again);
3121
3122 return skb;
3123
3124out_kfree_skb:
3125 kfree_skb(skb);
3126out_null:
3127 atomic_long_inc(&dev->tx_dropped);
3128 return NULL;
3129}
3130
3131struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3132{
3133 struct sk_buff *next, *head = NULL, *tail;
3134
3135 for (; skb != NULL; skb = next) {
3136 next = skb->next;
3137 skb->next = NULL;
3138
3139 /* in case skb wont be segmented, point to itself */
3140 skb->prev = skb;
3141
3142 skb = validate_xmit_skb(skb, dev, again);
3143 if (!skb)
3144 continue;
3145
3146 if (!head)
3147 head = skb;
3148 else
3149 tail->next = skb;
3150 /* If skb was segmented, skb->prev points to
3151 * the last segment. If not, it still contains skb.
3152 */
3153 tail = skb->prev;
3154 }
3155 return head;
3156}
3157EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3158
3159static void qdisc_pkt_len_init(struct sk_buff *skb)
3160{
3161 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3162
3163 qdisc_skb_cb(skb)->pkt_len = skb->len;
3164
3165 /* To get more precise estimation of bytes sent on wire,
3166 * we add to pkt_len the headers size of all segments
3167 */
3168 if (shinfo->gso_size) {
3169 unsigned int hdr_len;
3170 u16 gso_segs = shinfo->gso_segs;
3171
3172 /* mac layer + network layer */
3173 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3174
3175 /* + transport layer */
3176 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3177 const struct tcphdr *th;
3178 struct tcphdr _tcphdr;
3179
3180 th = skb_header_pointer(skb, skb_transport_offset(skb),
3181 sizeof(_tcphdr), &_tcphdr);
3182 if (likely(th))
3183 hdr_len += __tcp_hdrlen(th);
3184 } else {
3185 struct udphdr _udphdr;
3186
3187 if (skb_header_pointer(skb, skb_transport_offset(skb),
3188 sizeof(_udphdr), &_udphdr))
3189 hdr_len += sizeof(struct udphdr);
3190 }
3191
3192 if (shinfo->gso_type & SKB_GSO_DODGY)
3193 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3194 shinfo->gso_size);
3195
3196 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3197 }
3198}
3199
3200static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3201 struct net_device *dev,
3202 struct netdev_queue *txq)
3203{
3204 spinlock_t *root_lock = qdisc_lock(q);
3205 struct sk_buff *to_free = NULL;
3206 bool contended;
3207 int rc;
3208
3209 qdisc_calculate_pkt_len(skb, q);
3210
3211 if (q->flags & TCQ_F_NOLOCK) {
3212 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3213 __qdisc_drop(skb, &to_free);
3214 rc = NET_XMIT_DROP;
3215 } else {
3216 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3217 __qdisc_run(q);
3218 }
3219
3220 if (unlikely(to_free))
3221 kfree_skb_list(to_free);
3222 return rc;
3223 }
3224
3225 /*
3226 * Heuristic to force contended enqueues to serialize on a
3227 * separate lock before trying to get qdisc main lock.
3228 * This permits qdisc->running owner to get the lock more
3229 * often and dequeue packets faster.
3230 */
3231 contended = qdisc_is_running(q);
3232 if (unlikely(contended))
3233 spin_lock(&q->busylock);
3234
3235 spin_lock(root_lock);
3236 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3237 __qdisc_drop(skb, &to_free);
3238 rc = NET_XMIT_DROP;
3239 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3240 qdisc_run_begin(q)) {
3241 /*
3242 * This is a work-conserving queue; there are no old skbs
3243 * waiting to be sent out; and the qdisc is not running -
3244 * xmit the skb directly.
3245 */
3246
3247 qdisc_bstats_update(q, skb);
3248
3249 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3250 if (unlikely(contended)) {
3251 spin_unlock(&q->busylock);
3252 contended = false;
3253 }
3254 __qdisc_run(q);
3255 }
3256
3257 qdisc_run_end(q);
3258 rc = NET_XMIT_SUCCESS;
3259 } else {
3260 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3261 if (qdisc_run_begin(q)) {
3262 if (unlikely(contended)) {
3263 spin_unlock(&q->busylock);
3264 contended = false;
3265 }
3266 __qdisc_run(q);
3267 qdisc_run_end(q);
3268 }
3269 }
3270 spin_unlock(root_lock);
3271 if (unlikely(to_free))
3272 kfree_skb_list(to_free);
3273 if (unlikely(contended))
3274 spin_unlock(&q->busylock);
3275 return rc;
3276}
3277
3278#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3279static void skb_update_prio(struct sk_buff *skb)
3280{
3281 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3282
3283 if (!skb->priority && skb->sk && map) {
3284 unsigned int prioidx =
3285 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3286
3287 if (prioidx < map->priomap_len)
3288 skb->priority = map->priomap[prioidx];
3289 }
3290}
3291#else
3292#define skb_update_prio(skb)
3293#endif
3294
3295DEFINE_PER_CPU(int, xmit_recursion);
3296EXPORT_SYMBOL(xmit_recursion);
3297
3298/**
3299 * dev_loopback_xmit - loop back @skb
3300 * @net: network namespace this loopback is happening in
3301 * @sk: sk needed to be a netfilter okfn
3302 * @skb: buffer to transmit
3303 */
3304int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3305{
3306 skb_reset_mac_header(skb);
3307 __skb_pull(skb, skb_network_offset(skb));
3308 skb->pkt_type = PACKET_LOOPBACK;
3309 skb->ip_summed = CHECKSUM_UNNECESSARY;
3310 WARN_ON(!skb_dst(skb));
3311 skb_dst_force(skb);
3312 netif_rx_ni(skb);
3313 return 0;
3314}
3315EXPORT_SYMBOL(dev_loopback_xmit);
3316
3317#ifdef CONFIG_NET_EGRESS
3318static struct sk_buff *
3319sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3320{
3321 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3322 struct tcf_result cl_res;
3323
3324 if (!miniq)
3325 return skb;
3326
3327 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3328 mini_qdisc_bstats_cpu_update(miniq, skb);
3329
3330 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3331 case TC_ACT_OK:
3332 case TC_ACT_RECLASSIFY:
3333 skb->tc_index = TC_H_MIN(cl_res.classid);
3334 break;
3335 case TC_ACT_SHOT:
3336 mini_qdisc_qstats_cpu_drop(miniq);
3337 *ret = NET_XMIT_DROP;
3338 kfree_skb(skb);
3339 return NULL;
3340 case TC_ACT_STOLEN:
3341 case TC_ACT_QUEUED:
3342 case TC_ACT_TRAP:
3343 *ret = NET_XMIT_SUCCESS;
3344 consume_skb(skb);
3345 return NULL;
3346 case TC_ACT_REDIRECT:
3347 /* No need to push/pop skb's mac_header here on egress! */
3348 skb_do_redirect(skb);
3349 *ret = NET_XMIT_SUCCESS;
3350 return NULL;
3351 default:
3352 break;
3353 }
3354
3355 return skb;
3356}
3357#endif /* CONFIG_NET_EGRESS */
3358
3359static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3360{
3361#ifdef CONFIG_XPS
3362 struct xps_dev_maps *dev_maps;
3363 struct xps_map *map;
3364 int queue_index = -1;
3365
3366 rcu_read_lock();
3367 dev_maps = rcu_dereference(dev->xps_maps);
3368 if (dev_maps) {
3369 unsigned int tci = skb->sender_cpu - 1;
3370
3371 if (dev->num_tc) {
3372 tci *= dev->num_tc;
3373 tci += netdev_get_prio_tc_map(dev, skb->priority);
3374 }
3375
3376 map = rcu_dereference(dev_maps->cpu_map[tci]);
3377 if (map) {
3378 if (map->len == 1)
3379 queue_index = map->queues[0];
3380 else
3381 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3382 map->len)];
3383 if (unlikely(queue_index >= dev->real_num_tx_queues))
3384 queue_index = -1;
3385 }
3386 }
3387 rcu_read_unlock();
3388
3389 return queue_index;
3390#else
3391 return -1;
3392#endif
3393}
3394
3395static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3396{
3397 struct sock *sk = skb->sk;
3398 int queue_index = sk_tx_queue_get(sk);
3399
3400 if (queue_index < 0 || skb->ooo_okay ||
3401 queue_index >= dev->real_num_tx_queues) {
3402 int new_index = get_xps_queue(dev, skb);
3403
3404 if (new_index < 0)
3405 new_index = skb_tx_hash(dev, skb);
3406
3407 if (queue_index != new_index && sk &&
3408 sk_fullsock(sk) &&
3409 rcu_access_pointer(sk->sk_dst_cache))
3410 sk_tx_queue_set(sk, new_index);
3411
3412 queue_index = new_index;
3413 }
3414
3415 return queue_index;
3416}
3417
3418struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3419 struct sk_buff *skb,
3420 void *accel_priv)
3421{
3422 int queue_index = 0;
3423
3424#ifdef CONFIG_XPS
3425 u32 sender_cpu = skb->sender_cpu - 1;
3426
3427 if (sender_cpu >= (u32)NR_CPUS)
3428 skb->sender_cpu = raw_smp_processor_id() + 1;
3429#endif
3430
3431 if (dev->real_num_tx_queues != 1) {
3432 const struct net_device_ops *ops = dev->netdev_ops;
3433
3434 if (ops->ndo_select_queue)
3435 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3436 __netdev_pick_tx);
3437 else
3438 queue_index = __netdev_pick_tx(dev, skb);
3439
3440 queue_index = netdev_cap_txqueue(dev, queue_index);
3441 }
3442
3443 skb_set_queue_mapping(skb, queue_index);
3444 return netdev_get_tx_queue(dev, queue_index);
3445}
3446
3447/**
3448 * __dev_queue_xmit - transmit a buffer
3449 * @skb: buffer to transmit
3450 * @accel_priv: private data used for L2 forwarding offload
3451 *
3452 * Queue a buffer for transmission to a network device. The caller must
3453 * have set the device and priority and built the buffer before calling
3454 * this function. The function can be called from an interrupt.
3455 *
3456 * A negative errno code is returned on a failure. A success does not
3457 * guarantee the frame will be transmitted as it may be dropped due
3458 * to congestion or traffic shaping.
3459 *
3460 * -----------------------------------------------------------------------------------
3461 * I notice this method can also return errors from the queue disciplines,
3462 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3463 * be positive.
3464 *
3465 * Regardless of the return value, the skb is consumed, so it is currently
3466 * difficult to retry a send to this method. (You can bump the ref count
3467 * before sending to hold a reference for retry if you are careful.)
3468 *
3469 * When calling this method, interrupts MUST be enabled. This is because
3470 * the BH enable code must have IRQs enabled so that it will not deadlock.
3471 * --BLG
3472 */
3473static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3474{
3475 struct net_device *dev = skb->dev;
3476 struct netdev_queue *txq;
3477 struct Qdisc *q;
3478 int rc = -ENOMEM;
3479 bool again = false;
3480
3481 skb_reset_mac_header(skb);
3482
3483 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3484 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3485
3486 /* Disable soft irqs for various locks below. Also
3487 * stops preemption for RCU.
3488 */
3489 rcu_read_lock_bh();
3490
3491 skb_update_prio(skb);
3492
3493 qdisc_pkt_len_init(skb);
3494#ifdef CONFIG_NET_CLS_ACT
3495 skb->tc_at_ingress = 0;
3496# ifdef CONFIG_NET_EGRESS
3497 if (static_key_false(&egress_needed)) {
3498 skb = sch_handle_egress(skb, &rc, dev);
3499 if (!skb)
3500 goto out;
3501 }
3502# endif
3503#endif
3504 /* If device/qdisc don't need skb->dst, release it right now while
3505 * its hot in this cpu cache.
3506 */
3507 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3508 skb_dst_drop(skb);
3509 else
3510 skb_dst_force(skb);
3511
3512 txq = netdev_pick_tx(dev, skb, accel_priv);
3513 q = rcu_dereference_bh(txq->qdisc);
3514
3515 trace_net_dev_queue(skb);
3516 if (q->enqueue) {
3517 rc = __dev_xmit_skb(skb, q, dev, txq);
3518 goto out;
3519 }
3520
3521 /* The device has no queue. Common case for software devices:
3522 * loopback, all the sorts of tunnels...
3523
3524 * Really, it is unlikely that netif_tx_lock protection is necessary
3525 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3526 * counters.)
3527 * However, it is possible, that they rely on protection
3528 * made by us here.
3529
3530 * Check this and shot the lock. It is not prone from deadlocks.
3531 *Either shot noqueue qdisc, it is even simpler 8)
3532 */
3533 if (dev->flags & IFF_UP) {
3534 int cpu = smp_processor_id(); /* ok because BHs are off */
3535
3536 if (txq->xmit_lock_owner != cpu) {
3537 if (unlikely(__this_cpu_read(xmit_recursion) >
3538 XMIT_RECURSION_LIMIT))
3539 goto recursion_alert;
3540
3541 skb = validate_xmit_skb(skb, dev, &again);
3542 if (!skb)
3543 goto out;
3544
3545 HARD_TX_LOCK(dev, txq, cpu);
3546
3547 if (!netif_xmit_stopped(txq)) {
3548 __this_cpu_inc(xmit_recursion);
3549 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3550 __this_cpu_dec(xmit_recursion);
3551 if (dev_xmit_complete(rc)) {
3552 HARD_TX_UNLOCK(dev, txq);
3553 goto out;
3554 }
3555 }
3556 HARD_TX_UNLOCK(dev, txq);
3557 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3558 dev->name);
3559 } else {
3560 /* Recursion is detected! It is possible,
3561 * unfortunately
3562 */
3563recursion_alert:
3564 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3565 dev->name);
3566 }
3567 }
3568
3569 rc = -ENETDOWN;
3570 rcu_read_unlock_bh();
3571
3572 atomic_long_inc(&dev->tx_dropped);
3573 kfree_skb_list(skb);
3574 return rc;
3575out:
3576 rcu_read_unlock_bh();
3577 return rc;
3578}
3579
3580int dev_queue_xmit(struct sk_buff *skb)
3581{
3582 return __dev_queue_xmit(skb, NULL);
3583}
3584EXPORT_SYMBOL(dev_queue_xmit);
3585
3586int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3587{
3588 return __dev_queue_xmit(skb, accel_priv);
3589}
3590EXPORT_SYMBOL(dev_queue_xmit_accel);
3591
3592
3593/*************************************************************************
3594 * Receiver routines
3595 *************************************************************************/
3596
3597int netdev_max_backlog __read_mostly = 1000;
3598EXPORT_SYMBOL(netdev_max_backlog);
3599
3600int netdev_tstamp_prequeue __read_mostly = 1;
3601int netdev_budget __read_mostly = 300;
3602unsigned int __read_mostly netdev_budget_usecs = 2000;
3603int weight_p __read_mostly = 64; /* old backlog weight */
3604int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3605int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3606int dev_rx_weight __read_mostly = 64;
3607int dev_tx_weight __read_mostly = 64;
3608
3609/* Called with irq disabled */
3610static inline void ____napi_schedule(struct softnet_data *sd,
3611 struct napi_struct *napi)
3612{
3613 list_add_tail(&napi->poll_list, &sd->poll_list);
3614 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3615}
3616
3617#ifdef CONFIG_RPS
3618
3619/* One global table that all flow-based protocols share. */
3620struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3621EXPORT_SYMBOL(rps_sock_flow_table);
3622u32 rps_cpu_mask __read_mostly;
3623EXPORT_SYMBOL(rps_cpu_mask);
3624
3625struct static_key rps_needed __read_mostly;
3626EXPORT_SYMBOL(rps_needed);
3627struct static_key rfs_needed __read_mostly;
3628EXPORT_SYMBOL(rfs_needed);
3629
3630static struct rps_dev_flow *
3631set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3632 struct rps_dev_flow *rflow, u16 next_cpu)
3633{
3634 if (next_cpu < nr_cpu_ids) {
3635#ifdef CONFIG_RFS_ACCEL
3636 struct netdev_rx_queue *rxqueue;
3637 struct rps_dev_flow_table *flow_table;
3638 struct rps_dev_flow *old_rflow;
3639 u32 flow_id;
3640 u16 rxq_index;
3641 int rc;
3642
3643 /* Should we steer this flow to a different hardware queue? */
3644 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3645 !(dev->features & NETIF_F_NTUPLE))
3646 goto out;
3647 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3648 if (rxq_index == skb_get_rx_queue(skb))
3649 goto out;
3650
3651 rxqueue = dev->_rx + rxq_index;
3652 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3653 if (!flow_table)
3654 goto out;
3655 flow_id = skb_get_hash(skb) & flow_table->mask;
3656 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3657 rxq_index, flow_id);
3658 if (rc < 0)
3659 goto out;
3660 old_rflow = rflow;
3661 rflow = &flow_table->flows[flow_id];
3662 rflow->filter = rc;
3663 if (old_rflow->filter == rflow->filter)
3664 old_rflow->filter = RPS_NO_FILTER;
3665 out:
3666#endif
3667 rflow->last_qtail =
3668 per_cpu(softnet_data, next_cpu).input_queue_head;
3669 }
3670
3671 rflow->cpu = next_cpu;
3672 return rflow;
3673}
3674
3675/*
3676 * get_rps_cpu is called from netif_receive_skb and returns the target
3677 * CPU from the RPS map of the receiving queue for a given skb.
3678 * rcu_read_lock must be held on entry.
3679 */
3680static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3681 struct rps_dev_flow **rflowp)
3682{
3683 const struct rps_sock_flow_table *sock_flow_table;
3684 struct netdev_rx_queue *rxqueue = dev->_rx;
3685 struct rps_dev_flow_table *flow_table;
3686 struct rps_map *map;
3687 int cpu = -1;
3688 u32 tcpu;
3689 u32 hash;
3690
3691 if (skb_rx_queue_recorded(skb)) {
3692 u16 index = skb_get_rx_queue(skb);
3693
3694 if (unlikely(index >= dev->real_num_rx_queues)) {
3695 WARN_ONCE(dev->real_num_rx_queues > 1,
3696 "%s received packet on queue %u, but number "
3697 "of RX queues is %u\n",
3698 dev->name, index, dev->real_num_rx_queues);
3699 goto done;
3700 }
3701 rxqueue += index;
3702 }
3703
3704 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3705
3706 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3707 map = rcu_dereference(rxqueue->rps_map);
3708 if (!flow_table && !map)
3709 goto done;
3710
3711 skb_reset_network_header(skb);
3712 hash = skb_get_hash(skb);
3713 if (!hash)
3714 goto done;
3715
3716 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3717 if (flow_table && sock_flow_table) {
3718 struct rps_dev_flow *rflow;
3719 u32 next_cpu;
3720 u32 ident;
3721
3722 /* First check into global flow table if there is a match */
3723 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3724 if ((ident ^ hash) & ~rps_cpu_mask)
3725 goto try_rps;
3726
3727 next_cpu = ident & rps_cpu_mask;
3728
3729 /* OK, now we know there is a match,
3730 * we can look at the local (per receive queue) flow table
3731 */
3732 rflow = &flow_table->flows[hash & flow_table->mask];
3733 tcpu = rflow->cpu;
3734
3735 /*
3736 * If the desired CPU (where last recvmsg was done) is
3737 * different from current CPU (one in the rx-queue flow
3738 * table entry), switch if one of the following holds:
3739 * - Current CPU is unset (>= nr_cpu_ids).
3740 * - Current CPU is offline.
3741 * - The current CPU's queue tail has advanced beyond the
3742 * last packet that was enqueued using this table entry.
3743 * This guarantees that all previous packets for the flow
3744 * have been dequeued, thus preserving in order delivery.
3745 */
3746 if (unlikely(tcpu != next_cpu) &&
3747 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3748 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3749 rflow->last_qtail)) >= 0)) {
3750 tcpu = next_cpu;
3751 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3752 }
3753
3754 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3755 *rflowp = rflow;
3756 cpu = tcpu;
3757 goto done;
3758 }
3759 }
3760
3761try_rps:
3762
3763 if (map) {
3764 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3765 if (cpu_online(tcpu)) {
3766 cpu = tcpu;
3767 goto done;
3768 }
3769 }
3770
3771done:
3772 return cpu;
3773}
3774
3775#ifdef CONFIG_RFS_ACCEL
3776
3777/**
3778 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3779 * @dev: Device on which the filter was set
3780 * @rxq_index: RX queue index
3781 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3782 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3783 *
3784 * Drivers that implement ndo_rx_flow_steer() should periodically call
3785 * this function for each installed filter and remove the filters for
3786 * which it returns %true.
3787 */
3788bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3789 u32 flow_id, u16 filter_id)
3790{
3791 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3792 struct rps_dev_flow_table *flow_table;
3793 struct rps_dev_flow *rflow;
3794 bool expire = true;
3795 unsigned int cpu;
3796
3797 rcu_read_lock();
3798 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3799 if (flow_table && flow_id <= flow_table->mask) {
3800 rflow = &flow_table->flows[flow_id];
3801 cpu = READ_ONCE(rflow->cpu);
3802 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3803 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3804 rflow->last_qtail) <
3805 (int)(10 * flow_table->mask)))
3806 expire = false;
3807 }
3808 rcu_read_unlock();
3809 return expire;
3810}
3811EXPORT_SYMBOL(rps_may_expire_flow);
3812
3813#endif /* CONFIG_RFS_ACCEL */
3814
3815/* Called from hardirq (IPI) context */
3816static void rps_trigger_softirq(void *data)
3817{
3818 struct softnet_data *sd = data;
3819
3820 ____napi_schedule(sd, &sd->backlog);
3821 sd->received_rps++;
3822}
3823
3824#endif /* CONFIG_RPS */
3825
3826/*
3827 * Check if this softnet_data structure is another cpu one
3828 * If yes, queue it to our IPI list and return 1
3829 * If no, return 0
3830 */
3831static int rps_ipi_queued(struct softnet_data *sd)
3832{
3833#ifdef CONFIG_RPS
3834 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3835
3836 if (sd != mysd) {
3837 sd->rps_ipi_next = mysd->rps_ipi_list;
3838 mysd->