]> git.ipfire.org Git - thirdparty/linux.git/blob - net/core/dev.c
Merge tag 'io_uring-5.7-2020-05-22' of git://git.kernel.dk/linux-block
[thirdparty/linux.git] / net / core / dev.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NET3 Protocol independent device support routines.
4 *
5 * Derived from the non IP parts of dev.c 1.0.19
6 * Authors: Ross Biro
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
9 *
10 * Additional Authors:
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
17 *
18 * Changes:
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
30 * drivers
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
40 * call a packet.
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
46 * changes.
47 * Rudi Cilibrasi : Pass the right thing to
48 * set_mac_address()
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
54 * 1 device.
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
62 * the backlog queue.
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
69 */
70
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/string.h>
83 #include <linux/mm.h>
84 #include <linux/socket.h>
85 #include <linux/sockios.h>
86 #include <linux/errno.h>
87 #include <linux/interrupt.h>
88 #include <linux/if_ether.h>
89 #include <linux/netdevice.h>
90 #include <linux/etherdevice.h>
91 #include <linux/ethtool.h>
92 #include <linux/skbuff.h>
93 #include <linux/bpf.h>
94 #include <linux/bpf_trace.h>
95 #include <net/net_namespace.h>
96 #include <net/sock.h>
97 #include <net/busy_poll.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/stat.h>
100 #include <net/dst.h>
101 #include <net/dst_metadata.h>
102 #include <net/pkt_sched.h>
103 #include <net/pkt_cls.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
121 #include <net/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
139 #include <linux/crash_dump.h>
140 #include <linux/sctp.h>
141 #include <net/udp_tunnel.h>
142 #include <linux/net_namespace.h>
143 #include <linux/indirect_call_wrapper.h>
144 #include <net/devlink.h>
145
146 #include "net-sysfs.h"
147
148 #define MAX_GRO_SKBS 8
149
150 /* This should be increased if a protocol with a bigger head is added. */
151 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152
153 static DEFINE_SPINLOCK(ptype_lock);
154 static DEFINE_SPINLOCK(offload_lock);
155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
156 struct list_head ptype_all __read_mostly; /* Taps */
157 static struct list_head offload_base __read_mostly;
158
159 static int netif_rx_internal(struct sk_buff *skb);
160 static int call_netdevice_notifiers_info(unsigned long val,
161 struct netdev_notifier_info *info);
162 static int call_netdevice_notifiers_extack(unsigned long val,
163 struct net_device *dev,
164 struct netlink_ext_ack *extack);
165 static struct napi_struct *napi_by_id(unsigned int napi_id);
166
167 /*
168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
169 * semaphore.
170 *
171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
172 *
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base_head list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
177 *
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
181 *
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
184 * semaphore held.
185 */
186 DEFINE_RWLOCK(dev_base_lock);
187 EXPORT_SYMBOL(dev_base_lock);
188
189 static DEFINE_MUTEX(ifalias_mutex);
190
191 /* protects napi_hash addition/deletion and napi_gen_id */
192 static DEFINE_SPINLOCK(napi_hash_lock);
193
194 static unsigned int napi_gen_id = NR_CPUS;
195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
196
197 static seqcount_t devnet_rename_seq;
198
199 static inline void dev_base_seq_inc(struct net *net)
200 {
201 while (++net->dev_base_seq == 0)
202 ;
203 }
204
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
206 {
207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
208
209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
210 }
211
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
213 {
214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
215 }
216
217 static inline void rps_lock(struct softnet_data *sd)
218 {
219 #ifdef CONFIG_RPS
220 spin_lock(&sd->input_pkt_queue.lock);
221 #endif
222 }
223
224 static inline void rps_unlock(struct softnet_data *sd)
225 {
226 #ifdef CONFIG_RPS
227 spin_unlock(&sd->input_pkt_queue.lock);
228 #endif
229 }
230
231 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
232 const char *name)
233 {
234 struct netdev_name_node *name_node;
235
236 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
237 if (!name_node)
238 return NULL;
239 INIT_HLIST_NODE(&name_node->hlist);
240 name_node->dev = dev;
241 name_node->name = name;
242 return name_node;
243 }
244
245 static struct netdev_name_node *
246 netdev_name_node_head_alloc(struct net_device *dev)
247 {
248 struct netdev_name_node *name_node;
249
250 name_node = netdev_name_node_alloc(dev, dev->name);
251 if (!name_node)
252 return NULL;
253 INIT_LIST_HEAD(&name_node->list);
254 return name_node;
255 }
256
257 static void netdev_name_node_free(struct netdev_name_node *name_node)
258 {
259 kfree(name_node);
260 }
261
262 static void netdev_name_node_add(struct net *net,
263 struct netdev_name_node *name_node)
264 {
265 hlist_add_head_rcu(&name_node->hlist,
266 dev_name_hash(net, name_node->name));
267 }
268
269 static void netdev_name_node_del(struct netdev_name_node *name_node)
270 {
271 hlist_del_rcu(&name_node->hlist);
272 }
273
274 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
275 const char *name)
276 {
277 struct hlist_head *head = dev_name_hash(net, name);
278 struct netdev_name_node *name_node;
279
280 hlist_for_each_entry(name_node, head, hlist)
281 if (!strcmp(name_node->name, name))
282 return name_node;
283 return NULL;
284 }
285
286 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
287 const char *name)
288 {
289 struct hlist_head *head = dev_name_hash(net, name);
290 struct netdev_name_node *name_node;
291
292 hlist_for_each_entry_rcu(name_node, head, hlist)
293 if (!strcmp(name_node->name, name))
294 return name_node;
295 return NULL;
296 }
297
298 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
299 {
300 struct netdev_name_node *name_node;
301 struct net *net = dev_net(dev);
302
303 name_node = netdev_name_node_lookup(net, name);
304 if (name_node)
305 return -EEXIST;
306 name_node = netdev_name_node_alloc(dev, name);
307 if (!name_node)
308 return -ENOMEM;
309 netdev_name_node_add(net, name_node);
310 /* The node that holds dev->name acts as a head of per-device list. */
311 list_add_tail(&name_node->list, &dev->name_node->list);
312
313 return 0;
314 }
315 EXPORT_SYMBOL(netdev_name_node_alt_create);
316
317 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
318 {
319 list_del(&name_node->list);
320 netdev_name_node_del(name_node);
321 kfree(name_node->name);
322 netdev_name_node_free(name_node);
323 }
324
325 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
326 {
327 struct netdev_name_node *name_node;
328 struct net *net = dev_net(dev);
329
330 name_node = netdev_name_node_lookup(net, name);
331 if (!name_node)
332 return -ENOENT;
333 /* lookup might have found our primary name or a name belonging
334 * to another device.
335 */
336 if (name_node == dev->name_node || name_node->dev != dev)
337 return -EINVAL;
338
339 __netdev_name_node_alt_destroy(name_node);
340
341 return 0;
342 }
343 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
344
345 static void netdev_name_node_alt_flush(struct net_device *dev)
346 {
347 struct netdev_name_node *name_node, *tmp;
348
349 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
350 __netdev_name_node_alt_destroy(name_node);
351 }
352
353 /* Device list insertion */
354 static void list_netdevice(struct net_device *dev)
355 {
356 struct net *net = dev_net(dev);
357
358 ASSERT_RTNL();
359
360 write_lock_bh(&dev_base_lock);
361 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
362 netdev_name_node_add(net, dev->name_node);
363 hlist_add_head_rcu(&dev->index_hlist,
364 dev_index_hash(net, dev->ifindex));
365 write_unlock_bh(&dev_base_lock);
366
367 dev_base_seq_inc(net);
368 }
369
370 /* Device list removal
371 * caller must respect a RCU grace period before freeing/reusing dev
372 */
373 static void unlist_netdevice(struct net_device *dev)
374 {
375 ASSERT_RTNL();
376
377 /* Unlink dev from the device chain */
378 write_lock_bh(&dev_base_lock);
379 list_del_rcu(&dev->dev_list);
380 netdev_name_node_del(dev->name_node);
381 hlist_del_rcu(&dev->index_hlist);
382 write_unlock_bh(&dev_base_lock);
383
384 dev_base_seq_inc(dev_net(dev));
385 }
386
387 /*
388 * Our notifier list
389 */
390
391 static RAW_NOTIFIER_HEAD(netdev_chain);
392
393 /*
394 * Device drivers call our routines to queue packets here. We empty the
395 * queue in the local softnet handler.
396 */
397
398 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
399 EXPORT_PER_CPU_SYMBOL(softnet_data);
400
401 /*******************************************************************************
402 *
403 * Protocol management and registration routines
404 *
405 *******************************************************************************/
406
407
408 /*
409 * Add a protocol ID to the list. Now that the input handler is
410 * smarter we can dispense with all the messy stuff that used to be
411 * here.
412 *
413 * BEWARE!!! Protocol handlers, mangling input packets,
414 * MUST BE last in hash buckets and checking protocol handlers
415 * MUST start from promiscuous ptype_all chain in net_bh.
416 * It is true now, do not change it.
417 * Explanation follows: if protocol handler, mangling packet, will
418 * be the first on list, it is not able to sense, that packet
419 * is cloned and should be copied-on-write, so that it will
420 * change it and subsequent readers will get broken packet.
421 * --ANK (980803)
422 */
423
424 static inline struct list_head *ptype_head(const struct packet_type *pt)
425 {
426 if (pt->type == htons(ETH_P_ALL))
427 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
428 else
429 return pt->dev ? &pt->dev->ptype_specific :
430 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
431 }
432
433 /**
434 * dev_add_pack - add packet handler
435 * @pt: packet type declaration
436 *
437 * Add a protocol handler to the networking stack. The passed &packet_type
438 * is linked into kernel lists and may not be freed until it has been
439 * removed from the kernel lists.
440 *
441 * This call does not sleep therefore it can not
442 * guarantee all CPU's that are in middle of receiving packets
443 * will see the new packet type (until the next received packet).
444 */
445
446 void dev_add_pack(struct packet_type *pt)
447 {
448 struct list_head *head = ptype_head(pt);
449
450 spin_lock(&ptype_lock);
451 list_add_rcu(&pt->list, head);
452 spin_unlock(&ptype_lock);
453 }
454 EXPORT_SYMBOL(dev_add_pack);
455
456 /**
457 * __dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
459 *
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
463 * returns.
464 *
465 * The packet type might still be in use by receivers
466 * and must not be freed until after all the CPU's have gone
467 * through a quiescent state.
468 */
469 void __dev_remove_pack(struct packet_type *pt)
470 {
471 struct list_head *head = ptype_head(pt);
472 struct packet_type *pt1;
473
474 spin_lock(&ptype_lock);
475
476 list_for_each_entry(pt1, head, list) {
477 if (pt == pt1) {
478 list_del_rcu(&pt->list);
479 goto out;
480 }
481 }
482
483 pr_warn("dev_remove_pack: %p not found\n", pt);
484 out:
485 spin_unlock(&ptype_lock);
486 }
487 EXPORT_SYMBOL(__dev_remove_pack);
488
489 /**
490 * dev_remove_pack - remove packet handler
491 * @pt: packet type declaration
492 *
493 * Remove a protocol handler that was previously added to the kernel
494 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
495 * from the kernel lists and can be freed or reused once this function
496 * returns.
497 *
498 * This call sleeps to guarantee that no CPU is looking at the packet
499 * type after return.
500 */
501 void dev_remove_pack(struct packet_type *pt)
502 {
503 __dev_remove_pack(pt);
504
505 synchronize_net();
506 }
507 EXPORT_SYMBOL(dev_remove_pack);
508
509
510 /**
511 * dev_add_offload - register offload handlers
512 * @po: protocol offload declaration
513 *
514 * Add protocol offload handlers to the networking stack. The passed
515 * &proto_offload is linked into kernel lists and may not be freed until
516 * it has been removed from the kernel lists.
517 *
518 * This call does not sleep therefore it can not
519 * guarantee all CPU's that are in middle of receiving packets
520 * will see the new offload handlers (until the next received packet).
521 */
522 void dev_add_offload(struct packet_offload *po)
523 {
524 struct packet_offload *elem;
525
526 spin_lock(&offload_lock);
527 list_for_each_entry(elem, &offload_base, list) {
528 if (po->priority < elem->priority)
529 break;
530 }
531 list_add_rcu(&po->list, elem->list.prev);
532 spin_unlock(&offload_lock);
533 }
534 EXPORT_SYMBOL(dev_add_offload);
535
536 /**
537 * __dev_remove_offload - remove offload handler
538 * @po: packet offload declaration
539 *
540 * Remove a protocol offload handler that was previously added to the
541 * kernel offload handlers by dev_add_offload(). The passed &offload_type
542 * is removed from the kernel lists and can be freed or reused once this
543 * function returns.
544 *
545 * The packet type might still be in use by receivers
546 * and must not be freed until after all the CPU's have gone
547 * through a quiescent state.
548 */
549 static void __dev_remove_offload(struct packet_offload *po)
550 {
551 struct list_head *head = &offload_base;
552 struct packet_offload *po1;
553
554 spin_lock(&offload_lock);
555
556 list_for_each_entry(po1, head, list) {
557 if (po == po1) {
558 list_del_rcu(&po->list);
559 goto out;
560 }
561 }
562
563 pr_warn("dev_remove_offload: %p not found\n", po);
564 out:
565 spin_unlock(&offload_lock);
566 }
567
568 /**
569 * dev_remove_offload - remove packet offload handler
570 * @po: packet offload declaration
571 *
572 * Remove a packet offload handler that was previously added to the kernel
573 * offload handlers by dev_add_offload(). The passed &offload_type is
574 * removed from the kernel lists and can be freed or reused once this
575 * function returns.
576 *
577 * This call sleeps to guarantee that no CPU is looking at the packet
578 * type after return.
579 */
580 void dev_remove_offload(struct packet_offload *po)
581 {
582 __dev_remove_offload(po);
583
584 synchronize_net();
585 }
586 EXPORT_SYMBOL(dev_remove_offload);
587
588 /******************************************************************************
589 *
590 * Device Boot-time Settings Routines
591 *
592 ******************************************************************************/
593
594 /* Boot time configuration table */
595 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
596
597 /**
598 * netdev_boot_setup_add - add new setup entry
599 * @name: name of the device
600 * @map: configured settings for the device
601 *
602 * Adds new setup entry to the dev_boot_setup list. The function
603 * returns 0 on error and 1 on success. This is a generic routine to
604 * all netdevices.
605 */
606 static int netdev_boot_setup_add(char *name, struct ifmap *map)
607 {
608 struct netdev_boot_setup *s;
609 int i;
610
611 s = dev_boot_setup;
612 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
613 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
614 memset(s[i].name, 0, sizeof(s[i].name));
615 strlcpy(s[i].name, name, IFNAMSIZ);
616 memcpy(&s[i].map, map, sizeof(s[i].map));
617 break;
618 }
619 }
620
621 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
622 }
623
624 /**
625 * netdev_boot_setup_check - check boot time settings
626 * @dev: the netdevice
627 *
628 * Check boot time settings for the device.
629 * The found settings are set for the device to be used
630 * later in the device probing.
631 * Returns 0 if no settings found, 1 if they are.
632 */
633 int netdev_boot_setup_check(struct net_device *dev)
634 {
635 struct netdev_boot_setup *s = dev_boot_setup;
636 int i;
637
638 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
639 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
640 !strcmp(dev->name, s[i].name)) {
641 dev->irq = s[i].map.irq;
642 dev->base_addr = s[i].map.base_addr;
643 dev->mem_start = s[i].map.mem_start;
644 dev->mem_end = s[i].map.mem_end;
645 return 1;
646 }
647 }
648 return 0;
649 }
650 EXPORT_SYMBOL(netdev_boot_setup_check);
651
652
653 /**
654 * netdev_boot_base - get address from boot time settings
655 * @prefix: prefix for network device
656 * @unit: id for network device
657 *
658 * Check boot time settings for the base address of device.
659 * The found settings are set for the device to be used
660 * later in the device probing.
661 * Returns 0 if no settings found.
662 */
663 unsigned long netdev_boot_base(const char *prefix, int unit)
664 {
665 const struct netdev_boot_setup *s = dev_boot_setup;
666 char name[IFNAMSIZ];
667 int i;
668
669 sprintf(name, "%s%d", prefix, unit);
670
671 /*
672 * If device already registered then return base of 1
673 * to indicate not to probe for this interface
674 */
675 if (__dev_get_by_name(&init_net, name))
676 return 1;
677
678 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
679 if (!strcmp(name, s[i].name))
680 return s[i].map.base_addr;
681 return 0;
682 }
683
684 /*
685 * Saves at boot time configured settings for any netdevice.
686 */
687 int __init netdev_boot_setup(char *str)
688 {
689 int ints[5];
690 struct ifmap map;
691
692 str = get_options(str, ARRAY_SIZE(ints), ints);
693 if (!str || !*str)
694 return 0;
695
696 /* Save settings */
697 memset(&map, 0, sizeof(map));
698 if (ints[0] > 0)
699 map.irq = ints[1];
700 if (ints[0] > 1)
701 map.base_addr = ints[2];
702 if (ints[0] > 2)
703 map.mem_start = ints[3];
704 if (ints[0] > 3)
705 map.mem_end = ints[4];
706
707 /* Add new entry to the list */
708 return netdev_boot_setup_add(str, &map);
709 }
710
711 __setup("netdev=", netdev_boot_setup);
712
713 /*******************************************************************************
714 *
715 * Device Interface Subroutines
716 *
717 *******************************************************************************/
718
719 /**
720 * dev_get_iflink - get 'iflink' value of a interface
721 * @dev: targeted interface
722 *
723 * Indicates the ifindex the interface is linked to.
724 * Physical interfaces have the same 'ifindex' and 'iflink' values.
725 */
726
727 int dev_get_iflink(const struct net_device *dev)
728 {
729 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
730 return dev->netdev_ops->ndo_get_iflink(dev);
731
732 return dev->ifindex;
733 }
734 EXPORT_SYMBOL(dev_get_iflink);
735
736 /**
737 * dev_fill_metadata_dst - Retrieve tunnel egress information.
738 * @dev: targeted interface
739 * @skb: The packet.
740 *
741 * For better visibility of tunnel traffic OVS needs to retrieve
742 * egress tunnel information for a packet. Following API allows
743 * user to get this info.
744 */
745 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
746 {
747 struct ip_tunnel_info *info;
748
749 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
750 return -EINVAL;
751
752 info = skb_tunnel_info_unclone(skb);
753 if (!info)
754 return -ENOMEM;
755 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
756 return -EINVAL;
757
758 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
759 }
760 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
761
762 /**
763 * __dev_get_by_name - find a device by its name
764 * @net: the applicable net namespace
765 * @name: name to find
766 *
767 * Find an interface by name. Must be called under RTNL semaphore
768 * or @dev_base_lock. If the name is found a pointer to the device
769 * is returned. If the name is not found then %NULL is returned. The
770 * reference counters are not incremented so the caller must be
771 * careful with locks.
772 */
773
774 struct net_device *__dev_get_by_name(struct net *net, const char *name)
775 {
776 struct netdev_name_node *node_name;
777
778 node_name = netdev_name_node_lookup(net, name);
779 return node_name ? node_name->dev : NULL;
780 }
781 EXPORT_SYMBOL(__dev_get_by_name);
782
783 /**
784 * dev_get_by_name_rcu - find a device by its name
785 * @net: the applicable net namespace
786 * @name: name to find
787 *
788 * Find an interface by name.
789 * If the name is found a pointer to the device is returned.
790 * If the name is not found then %NULL is returned.
791 * The reference counters are not incremented so the caller must be
792 * careful with locks. The caller must hold RCU lock.
793 */
794
795 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
796 {
797 struct netdev_name_node *node_name;
798
799 node_name = netdev_name_node_lookup_rcu(net, name);
800 return node_name ? node_name->dev : NULL;
801 }
802 EXPORT_SYMBOL(dev_get_by_name_rcu);
803
804 /**
805 * dev_get_by_name - find a device by its name
806 * @net: the applicable net namespace
807 * @name: name to find
808 *
809 * Find an interface by name. This can be called from any
810 * context and does its own locking. The returned handle has
811 * the usage count incremented and the caller must use dev_put() to
812 * release it when it is no longer needed. %NULL is returned if no
813 * matching device is found.
814 */
815
816 struct net_device *dev_get_by_name(struct net *net, const char *name)
817 {
818 struct net_device *dev;
819
820 rcu_read_lock();
821 dev = dev_get_by_name_rcu(net, name);
822 if (dev)
823 dev_hold(dev);
824 rcu_read_unlock();
825 return dev;
826 }
827 EXPORT_SYMBOL(dev_get_by_name);
828
829 /**
830 * __dev_get_by_index - find a device by its ifindex
831 * @net: the applicable net namespace
832 * @ifindex: index of device
833 *
834 * Search for an interface by index. Returns %NULL if the device
835 * is not found or a pointer to the device. The device has not
836 * had its reference counter increased so the caller must be careful
837 * about locking. The caller must hold either the RTNL semaphore
838 * or @dev_base_lock.
839 */
840
841 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
842 {
843 struct net_device *dev;
844 struct hlist_head *head = dev_index_hash(net, ifindex);
845
846 hlist_for_each_entry(dev, head, index_hlist)
847 if (dev->ifindex == ifindex)
848 return dev;
849
850 return NULL;
851 }
852 EXPORT_SYMBOL(__dev_get_by_index);
853
854 /**
855 * dev_get_by_index_rcu - find a device by its ifindex
856 * @net: the applicable net namespace
857 * @ifindex: index of device
858 *
859 * Search for an interface by index. Returns %NULL if the device
860 * is not found or a pointer to the device. The device has not
861 * had its reference counter increased so the caller must be careful
862 * about locking. The caller must hold RCU lock.
863 */
864
865 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
866 {
867 struct net_device *dev;
868 struct hlist_head *head = dev_index_hash(net, ifindex);
869
870 hlist_for_each_entry_rcu(dev, head, index_hlist)
871 if (dev->ifindex == ifindex)
872 return dev;
873
874 return NULL;
875 }
876 EXPORT_SYMBOL(dev_get_by_index_rcu);
877
878
879 /**
880 * dev_get_by_index - find a device by its ifindex
881 * @net: the applicable net namespace
882 * @ifindex: index of device
883 *
884 * Search for an interface by index. Returns NULL if the device
885 * is not found or a pointer to the device. The device returned has
886 * had a reference added and the pointer is safe until the user calls
887 * dev_put to indicate they have finished with it.
888 */
889
890 struct net_device *dev_get_by_index(struct net *net, int ifindex)
891 {
892 struct net_device *dev;
893
894 rcu_read_lock();
895 dev = dev_get_by_index_rcu(net, ifindex);
896 if (dev)
897 dev_hold(dev);
898 rcu_read_unlock();
899 return dev;
900 }
901 EXPORT_SYMBOL(dev_get_by_index);
902
903 /**
904 * dev_get_by_napi_id - find a device by napi_id
905 * @napi_id: ID of the NAPI struct
906 *
907 * Search for an interface by NAPI ID. Returns %NULL if the device
908 * is not found or a pointer to the device. The device has not had
909 * its reference counter increased so the caller must be careful
910 * about locking. The caller must hold RCU lock.
911 */
912
913 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
914 {
915 struct napi_struct *napi;
916
917 WARN_ON_ONCE(!rcu_read_lock_held());
918
919 if (napi_id < MIN_NAPI_ID)
920 return NULL;
921
922 napi = napi_by_id(napi_id);
923
924 return napi ? napi->dev : NULL;
925 }
926 EXPORT_SYMBOL(dev_get_by_napi_id);
927
928 /**
929 * netdev_get_name - get a netdevice name, knowing its ifindex.
930 * @net: network namespace
931 * @name: a pointer to the buffer where the name will be stored.
932 * @ifindex: the ifindex of the interface to get the name from.
933 *
934 * The use of raw_seqcount_begin() and cond_resched() before
935 * retrying is required as we want to give the writers a chance
936 * to complete when CONFIG_PREEMPTION is not set.
937 */
938 int netdev_get_name(struct net *net, char *name, int ifindex)
939 {
940 struct net_device *dev;
941 unsigned int seq;
942
943 retry:
944 seq = raw_seqcount_begin(&devnet_rename_seq);
945 rcu_read_lock();
946 dev = dev_get_by_index_rcu(net, ifindex);
947 if (!dev) {
948 rcu_read_unlock();
949 return -ENODEV;
950 }
951
952 strcpy(name, dev->name);
953 rcu_read_unlock();
954 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
955 cond_resched();
956 goto retry;
957 }
958
959 return 0;
960 }
961
962 /**
963 * dev_getbyhwaddr_rcu - find a device by its hardware address
964 * @net: the applicable net namespace
965 * @type: media type of device
966 * @ha: hardware address
967 *
968 * Search for an interface by MAC address. Returns NULL if the device
969 * is not found or a pointer to the device.
970 * The caller must hold RCU or RTNL.
971 * The returned device has not had its ref count increased
972 * and the caller must therefore be careful about locking
973 *
974 */
975
976 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
977 const char *ha)
978 {
979 struct net_device *dev;
980
981 for_each_netdev_rcu(net, dev)
982 if (dev->type == type &&
983 !memcmp(dev->dev_addr, ha, dev->addr_len))
984 return dev;
985
986 return NULL;
987 }
988 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
989
990 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
991 {
992 struct net_device *dev;
993
994 ASSERT_RTNL();
995 for_each_netdev(net, dev)
996 if (dev->type == type)
997 return dev;
998
999 return NULL;
1000 }
1001 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
1002
1003 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1004 {
1005 struct net_device *dev, *ret = NULL;
1006
1007 rcu_read_lock();
1008 for_each_netdev_rcu(net, dev)
1009 if (dev->type == type) {
1010 dev_hold(dev);
1011 ret = dev;
1012 break;
1013 }
1014 rcu_read_unlock();
1015 return ret;
1016 }
1017 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1018
1019 /**
1020 * __dev_get_by_flags - find any device with given flags
1021 * @net: the applicable net namespace
1022 * @if_flags: IFF_* values
1023 * @mask: bitmask of bits in if_flags to check
1024 *
1025 * Search for any interface with the given flags. Returns NULL if a device
1026 * is not found or a pointer to the device. Must be called inside
1027 * rtnl_lock(), and result refcount is unchanged.
1028 */
1029
1030 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1031 unsigned short mask)
1032 {
1033 struct net_device *dev, *ret;
1034
1035 ASSERT_RTNL();
1036
1037 ret = NULL;
1038 for_each_netdev(net, dev) {
1039 if (((dev->flags ^ if_flags) & mask) == 0) {
1040 ret = dev;
1041 break;
1042 }
1043 }
1044 return ret;
1045 }
1046 EXPORT_SYMBOL(__dev_get_by_flags);
1047
1048 /**
1049 * dev_valid_name - check if name is okay for network device
1050 * @name: name string
1051 *
1052 * Network device names need to be valid file names to
1053 * to allow sysfs to work. We also disallow any kind of
1054 * whitespace.
1055 */
1056 bool dev_valid_name(const char *name)
1057 {
1058 if (*name == '\0')
1059 return false;
1060 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1061 return false;
1062 if (!strcmp(name, ".") || !strcmp(name, ".."))
1063 return false;
1064
1065 while (*name) {
1066 if (*name == '/' || *name == ':' || isspace(*name))
1067 return false;
1068 name++;
1069 }
1070 return true;
1071 }
1072 EXPORT_SYMBOL(dev_valid_name);
1073
1074 /**
1075 * __dev_alloc_name - allocate a name for a device
1076 * @net: network namespace to allocate the device name in
1077 * @name: name format string
1078 * @buf: scratch buffer and result name string
1079 *
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1084 * duplicates.
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1087 */
1088
1089 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1090 {
1091 int i = 0;
1092 const char *p;
1093 const int max_netdevices = 8*PAGE_SIZE;
1094 unsigned long *inuse;
1095 struct net_device *d;
1096
1097 if (!dev_valid_name(name))
1098 return -EINVAL;
1099
1100 p = strchr(name, '%');
1101 if (p) {
1102 /*
1103 * Verify the string as this thing may have come from
1104 * the user. There must be either one "%d" and no other "%"
1105 * characters.
1106 */
1107 if (p[1] != 'd' || strchr(p + 2, '%'))
1108 return -EINVAL;
1109
1110 /* Use one page as a bit array of possible slots */
1111 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1112 if (!inuse)
1113 return -ENOMEM;
1114
1115 for_each_netdev(net, d) {
1116 if (!sscanf(d->name, name, &i))
1117 continue;
1118 if (i < 0 || i >= max_netdevices)
1119 continue;
1120
1121 /* avoid cases where sscanf is not exact inverse of printf */
1122 snprintf(buf, IFNAMSIZ, name, i);
1123 if (!strncmp(buf, d->name, IFNAMSIZ))
1124 set_bit(i, inuse);
1125 }
1126
1127 i = find_first_zero_bit(inuse, max_netdevices);
1128 free_page((unsigned long) inuse);
1129 }
1130
1131 snprintf(buf, IFNAMSIZ, name, i);
1132 if (!__dev_get_by_name(net, buf))
1133 return i;
1134
1135 /* It is possible to run out of possible slots
1136 * when the name is long and there isn't enough space left
1137 * for the digits, or if all bits are used.
1138 */
1139 return -ENFILE;
1140 }
1141
1142 static int dev_alloc_name_ns(struct net *net,
1143 struct net_device *dev,
1144 const char *name)
1145 {
1146 char buf[IFNAMSIZ];
1147 int ret;
1148
1149 BUG_ON(!net);
1150 ret = __dev_alloc_name(net, name, buf);
1151 if (ret >= 0)
1152 strlcpy(dev->name, buf, IFNAMSIZ);
1153 return ret;
1154 }
1155
1156 /**
1157 * dev_alloc_name - allocate a name for a device
1158 * @dev: device
1159 * @name: name format string
1160 *
1161 * Passed a format string - eg "lt%d" it will try and find a suitable
1162 * id. It scans list of devices to build up a free map, then chooses
1163 * the first empty slot. The caller must hold the dev_base or rtnl lock
1164 * while allocating the name and adding the device in order to avoid
1165 * duplicates.
1166 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1167 * Returns the number of the unit assigned or a negative errno code.
1168 */
1169
1170 int dev_alloc_name(struct net_device *dev, const char *name)
1171 {
1172 return dev_alloc_name_ns(dev_net(dev), dev, name);
1173 }
1174 EXPORT_SYMBOL(dev_alloc_name);
1175
1176 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1177 const char *name)
1178 {
1179 BUG_ON(!net);
1180
1181 if (!dev_valid_name(name))
1182 return -EINVAL;
1183
1184 if (strchr(name, '%'))
1185 return dev_alloc_name_ns(net, dev, name);
1186 else if (__dev_get_by_name(net, name))
1187 return -EEXIST;
1188 else if (dev->name != name)
1189 strlcpy(dev->name, name, IFNAMSIZ);
1190
1191 return 0;
1192 }
1193
1194 /**
1195 * dev_change_name - change name of a device
1196 * @dev: device
1197 * @newname: name (or format string) must be at least IFNAMSIZ
1198 *
1199 * Change name of a device, can pass format strings "eth%d".
1200 * for wildcarding.
1201 */
1202 int dev_change_name(struct net_device *dev, const char *newname)
1203 {
1204 unsigned char old_assign_type;
1205 char oldname[IFNAMSIZ];
1206 int err = 0;
1207 int ret;
1208 struct net *net;
1209
1210 ASSERT_RTNL();
1211 BUG_ON(!dev_net(dev));
1212
1213 net = dev_net(dev);
1214
1215 /* Some auto-enslaved devices e.g. failover slaves are
1216 * special, as userspace might rename the device after
1217 * the interface had been brought up and running since
1218 * the point kernel initiated auto-enslavement. Allow
1219 * live name change even when these slave devices are
1220 * up and running.
1221 *
1222 * Typically, users of these auto-enslaving devices
1223 * don't actually care about slave name change, as
1224 * they are supposed to operate on master interface
1225 * directly.
1226 */
1227 if (dev->flags & IFF_UP &&
1228 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1229 return -EBUSY;
1230
1231 write_seqcount_begin(&devnet_rename_seq);
1232
1233 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1234 write_seqcount_end(&devnet_rename_seq);
1235 return 0;
1236 }
1237
1238 memcpy(oldname, dev->name, IFNAMSIZ);
1239
1240 err = dev_get_valid_name(net, dev, newname);
1241 if (err < 0) {
1242 write_seqcount_end(&devnet_rename_seq);
1243 return err;
1244 }
1245
1246 if (oldname[0] && !strchr(oldname, '%'))
1247 netdev_info(dev, "renamed from %s\n", oldname);
1248
1249 old_assign_type = dev->name_assign_type;
1250 dev->name_assign_type = NET_NAME_RENAMED;
1251
1252 rollback:
1253 ret = device_rename(&dev->dev, dev->name);
1254 if (ret) {
1255 memcpy(dev->name, oldname, IFNAMSIZ);
1256 dev->name_assign_type = old_assign_type;
1257 write_seqcount_end(&devnet_rename_seq);
1258 return ret;
1259 }
1260
1261 write_seqcount_end(&devnet_rename_seq);
1262
1263 netdev_adjacent_rename_links(dev, oldname);
1264
1265 write_lock_bh(&dev_base_lock);
1266 netdev_name_node_del(dev->name_node);
1267 write_unlock_bh(&dev_base_lock);
1268
1269 synchronize_rcu();
1270
1271 write_lock_bh(&dev_base_lock);
1272 netdev_name_node_add(net, dev->name_node);
1273 write_unlock_bh(&dev_base_lock);
1274
1275 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1276 ret = notifier_to_errno(ret);
1277
1278 if (ret) {
1279 /* err >= 0 after dev_alloc_name() or stores the first errno */
1280 if (err >= 0) {
1281 err = ret;
1282 write_seqcount_begin(&devnet_rename_seq);
1283 memcpy(dev->name, oldname, IFNAMSIZ);
1284 memcpy(oldname, newname, IFNAMSIZ);
1285 dev->name_assign_type = old_assign_type;
1286 old_assign_type = NET_NAME_RENAMED;
1287 goto rollback;
1288 } else {
1289 pr_err("%s: name change rollback failed: %d\n",
1290 dev->name, ret);
1291 }
1292 }
1293
1294 return err;
1295 }
1296
1297 /**
1298 * dev_set_alias - change ifalias of a device
1299 * @dev: device
1300 * @alias: name up to IFALIASZ
1301 * @len: limit of bytes to copy from info
1302 *
1303 * Set ifalias for a device,
1304 */
1305 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1306 {
1307 struct dev_ifalias *new_alias = NULL;
1308
1309 if (len >= IFALIASZ)
1310 return -EINVAL;
1311
1312 if (len) {
1313 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1314 if (!new_alias)
1315 return -ENOMEM;
1316
1317 memcpy(new_alias->ifalias, alias, len);
1318 new_alias->ifalias[len] = 0;
1319 }
1320
1321 mutex_lock(&ifalias_mutex);
1322 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1323 mutex_is_locked(&ifalias_mutex));
1324 mutex_unlock(&ifalias_mutex);
1325
1326 if (new_alias)
1327 kfree_rcu(new_alias, rcuhead);
1328
1329 return len;
1330 }
1331 EXPORT_SYMBOL(dev_set_alias);
1332
1333 /**
1334 * dev_get_alias - get ifalias of a device
1335 * @dev: device
1336 * @name: buffer to store name of ifalias
1337 * @len: size of buffer
1338 *
1339 * get ifalias for a device. Caller must make sure dev cannot go
1340 * away, e.g. rcu read lock or own a reference count to device.
1341 */
1342 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1343 {
1344 const struct dev_ifalias *alias;
1345 int ret = 0;
1346
1347 rcu_read_lock();
1348 alias = rcu_dereference(dev->ifalias);
1349 if (alias)
1350 ret = snprintf(name, len, "%s", alias->ifalias);
1351 rcu_read_unlock();
1352
1353 return ret;
1354 }
1355
1356 /**
1357 * netdev_features_change - device changes features
1358 * @dev: device to cause notification
1359 *
1360 * Called to indicate a device has changed features.
1361 */
1362 void netdev_features_change(struct net_device *dev)
1363 {
1364 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1365 }
1366 EXPORT_SYMBOL(netdev_features_change);
1367
1368 /**
1369 * netdev_state_change - device changes state
1370 * @dev: device to cause notification
1371 *
1372 * Called to indicate a device has changed state. This function calls
1373 * the notifier chains for netdev_chain and sends a NEWLINK message
1374 * to the routing socket.
1375 */
1376 void netdev_state_change(struct net_device *dev)
1377 {
1378 if (dev->flags & IFF_UP) {
1379 struct netdev_notifier_change_info change_info = {
1380 .info.dev = dev,
1381 };
1382
1383 call_netdevice_notifiers_info(NETDEV_CHANGE,
1384 &change_info.info);
1385 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1386 }
1387 }
1388 EXPORT_SYMBOL(netdev_state_change);
1389
1390 /**
1391 * netdev_notify_peers - notify network peers about existence of @dev
1392 * @dev: network device
1393 *
1394 * Generate traffic such that interested network peers are aware of
1395 * @dev, such as by generating a gratuitous ARP. This may be used when
1396 * a device wants to inform the rest of the network about some sort of
1397 * reconfiguration such as a failover event or virtual machine
1398 * migration.
1399 */
1400 void netdev_notify_peers(struct net_device *dev)
1401 {
1402 rtnl_lock();
1403 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1404 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1405 rtnl_unlock();
1406 }
1407 EXPORT_SYMBOL(netdev_notify_peers);
1408
1409 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1410 {
1411 const struct net_device_ops *ops = dev->netdev_ops;
1412 int ret;
1413
1414 ASSERT_RTNL();
1415
1416 if (!netif_device_present(dev))
1417 return -ENODEV;
1418
1419 /* Block netpoll from trying to do any rx path servicing.
1420 * If we don't do this there is a chance ndo_poll_controller
1421 * or ndo_poll may be running while we open the device
1422 */
1423 netpoll_poll_disable(dev);
1424
1425 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1426 ret = notifier_to_errno(ret);
1427 if (ret)
1428 return ret;
1429
1430 set_bit(__LINK_STATE_START, &dev->state);
1431
1432 if (ops->ndo_validate_addr)
1433 ret = ops->ndo_validate_addr(dev);
1434
1435 if (!ret && ops->ndo_open)
1436 ret = ops->ndo_open(dev);
1437
1438 netpoll_poll_enable(dev);
1439
1440 if (ret)
1441 clear_bit(__LINK_STATE_START, &dev->state);
1442 else {
1443 dev->flags |= IFF_UP;
1444 dev_set_rx_mode(dev);
1445 dev_activate(dev);
1446 add_device_randomness(dev->dev_addr, dev->addr_len);
1447 }
1448
1449 return ret;
1450 }
1451
1452 /**
1453 * dev_open - prepare an interface for use.
1454 * @dev: device to open
1455 * @extack: netlink extended ack
1456 *
1457 * Takes a device from down to up state. The device's private open
1458 * function is invoked and then the multicast lists are loaded. Finally
1459 * the device is moved into the up state and a %NETDEV_UP message is
1460 * sent to the netdev notifier chain.
1461 *
1462 * Calling this function on an active interface is a nop. On a failure
1463 * a negative errno code is returned.
1464 */
1465 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1466 {
1467 int ret;
1468
1469 if (dev->flags & IFF_UP)
1470 return 0;
1471
1472 ret = __dev_open(dev, extack);
1473 if (ret < 0)
1474 return ret;
1475
1476 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1477 call_netdevice_notifiers(NETDEV_UP, dev);
1478
1479 return ret;
1480 }
1481 EXPORT_SYMBOL(dev_open);
1482
1483 static void __dev_close_many(struct list_head *head)
1484 {
1485 struct net_device *dev;
1486
1487 ASSERT_RTNL();
1488 might_sleep();
1489
1490 list_for_each_entry(dev, head, close_list) {
1491 /* Temporarily disable netpoll until the interface is down */
1492 netpoll_poll_disable(dev);
1493
1494 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1495
1496 clear_bit(__LINK_STATE_START, &dev->state);
1497
1498 /* Synchronize to scheduled poll. We cannot touch poll list, it
1499 * can be even on different cpu. So just clear netif_running().
1500 *
1501 * dev->stop() will invoke napi_disable() on all of it's
1502 * napi_struct instances on this device.
1503 */
1504 smp_mb__after_atomic(); /* Commit netif_running(). */
1505 }
1506
1507 dev_deactivate_many(head);
1508
1509 list_for_each_entry(dev, head, close_list) {
1510 const struct net_device_ops *ops = dev->netdev_ops;
1511
1512 /*
1513 * Call the device specific close. This cannot fail.
1514 * Only if device is UP
1515 *
1516 * We allow it to be called even after a DETACH hot-plug
1517 * event.
1518 */
1519 if (ops->ndo_stop)
1520 ops->ndo_stop(dev);
1521
1522 dev->flags &= ~IFF_UP;
1523 netpoll_poll_enable(dev);
1524 }
1525 }
1526
1527 static void __dev_close(struct net_device *dev)
1528 {
1529 LIST_HEAD(single);
1530
1531 list_add(&dev->close_list, &single);
1532 __dev_close_many(&single);
1533 list_del(&single);
1534 }
1535
1536 void dev_close_many(struct list_head *head, bool unlink)
1537 {
1538 struct net_device *dev, *tmp;
1539
1540 /* Remove the devices that don't need to be closed */
1541 list_for_each_entry_safe(dev, tmp, head, close_list)
1542 if (!(dev->flags & IFF_UP))
1543 list_del_init(&dev->close_list);
1544
1545 __dev_close_many(head);
1546
1547 list_for_each_entry_safe(dev, tmp, head, close_list) {
1548 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1549 call_netdevice_notifiers(NETDEV_DOWN, dev);
1550 if (unlink)
1551 list_del_init(&dev->close_list);
1552 }
1553 }
1554 EXPORT_SYMBOL(dev_close_many);
1555
1556 /**
1557 * dev_close - shutdown an interface.
1558 * @dev: device to shutdown
1559 *
1560 * This function moves an active device into down state. A
1561 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1562 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1563 * chain.
1564 */
1565 void dev_close(struct net_device *dev)
1566 {
1567 if (dev->flags & IFF_UP) {
1568 LIST_HEAD(single);
1569
1570 list_add(&dev->close_list, &single);
1571 dev_close_many(&single, true);
1572 list_del(&single);
1573 }
1574 }
1575 EXPORT_SYMBOL(dev_close);
1576
1577
1578 /**
1579 * dev_disable_lro - disable Large Receive Offload on a device
1580 * @dev: device
1581 *
1582 * Disable Large Receive Offload (LRO) on a net device. Must be
1583 * called under RTNL. This is needed if received packets may be
1584 * forwarded to another interface.
1585 */
1586 void dev_disable_lro(struct net_device *dev)
1587 {
1588 struct net_device *lower_dev;
1589 struct list_head *iter;
1590
1591 dev->wanted_features &= ~NETIF_F_LRO;
1592 netdev_update_features(dev);
1593
1594 if (unlikely(dev->features & NETIF_F_LRO))
1595 netdev_WARN(dev, "failed to disable LRO!\n");
1596
1597 netdev_for_each_lower_dev(dev, lower_dev, iter)
1598 dev_disable_lro(lower_dev);
1599 }
1600 EXPORT_SYMBOL(dev_disable_lro);
1601
1602 /**
1603 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1604 * @dev: device
1605 *
1606 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1607 * called under RTNL. This is needed if Generic XDP is installed on
1608 * the device.
1609 */
1610 static void dev_disable_gro_hw(struct net_device *dev)
1611 {
1612 dev->wanted_features &= ~NETIF_F_GRO_HW;
1613 netdev_update_features(dev);
1614
1615 if (unlikely(dev->features & NETIF_F_GRO_HW))
1616 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1617 }
1618
1619 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1620 {
1621 #define N(val) \
1622 case NETDEV_##val: \
1623 return "NETDEV_" __stringify(val);
1624 switch (cmd) {
1625 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1626 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1627 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1628 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1629 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1630 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1631 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1632 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1633 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1634 N(PRE_CHANGEADDR)
1635 }
1636 #undef N
1637 return "UNKNOWN_NETDEV_EVENT";
1638 }
1639 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1640
1641 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1642 struct net_device *dev)
1643 {
1644 struct netdev_notifier_info info = {
1645 .dev = dev,
1646 };
1647
1648 return nb->notifier_call(nb, val, &info);
1649 }
1650
1651 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1652 struct net_device *dev)
1653 {
1654 int err;
1655
1656 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1657 err = notifier_to_errno(err);
1658 if (err)
1659 return err;
1660
1661 if (!(dev->flags & IFF_UP))
1662 return 0;
1663
1664 call_netdevice_notifier(nb, NETDEV_UP, dev);
1665 return 0;
1666 }
1667
1668 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1669 struct net_device *dev)
1670 {
1671 if (dev->flags & IFF_UP) {
1672 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1673 dev);
1674 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1675 }
1676 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1677 }
1678
1679 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1680 struct net *net)
1681 {
1682 struct net_device *dev;
1683 int err;
1684
1685 for_each_netdev(net, dev) {
1686 err = call_netdevice_register_notifiers(nb, dev);
1687 if (err)
1688 goto rollback;
1689 }
1690 return 0;
1691
1692 rollback:
1693 for_each_netdev_continue_reverse(net, dev)
1694 call_netdevice_unregister_notifiers(nb, dev);
1695 return err;
1696 }
1697
1698 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1699 struct net *net)
1700 {
1701 struct net_device *dev;
1702
1703 for_each_netdev(net, dev)
1704 call_netdevice_unregister_notifiers(nb, dev);
1705 }
1706
1707 static int dev_boot_phase = 1;
1708
1709 /**
1710 * register_netdevice_notifier - register a network notifier block
1711 * @nb: notifier
1712 *
1713 * Register a notifier to be called when network device events occur.
1714 * The notifier passed is linked into the kernel structures and must
1715 * not be reused until it has been unregistered. A negative errno code
1716 * is returned on a failure.
1717 *
1718 * When registered all registration and up events are replayed
1719 * to the new notifier to allow device to have a race free
1720 * view of the network device list.
1721 */
1722
1723 int register_netdevice_notifier(struct notifier_block *nb)
1724 {
1725 struct net *net;
1726 int err;
1727
1728 /* Close race with setup_net() and cleanup_net() */
1729 down_write(&pernet_ops_rwsem);
1730 rtnl_lock();
1731 err = raw_notifier_chain_register(&netdev_chain, nb);
1732 if (err)
1733 goto unlock;
1734 if (dev_boot_phase)
1735 goto unlock;
1736 for_each_net(net) {
1737 err = call_netdevice_register_net_notifiers(nb, net);
1738 if (err)
1739 goto rollback;
1740 }
1741
1742 unlock:
1743 rtnl_unlock();
1744 up_write(&pernet_ops_rwsem);
1745 return err;
1746
1747 rollback:
1748 for_each_net_continue_reverse(net)
1749 call_netdevice_unregister_net_notifiers(nb, net);
1750
1751 raw_notifier_chain_unregister(&netdev_chain, nb);
1752 goto unlock;
1753 }
1754 EXPORT_SYMBOL(register_netdevice_notifier);
1755
1756 /**
1757 * unregister_netdevice_notifier - unregister a network notifier block
1758 * @nb: notifier
1759 *
1760 * Unregister a notifier previously registered by
1761 * register_netdevice_notifier(). The notifier is unlinked into the
1762 * kernel structures and may then be reused. A negative errno code
1763 * is returned on a failure.
1764 *
1765 * After unregistering unregister and down device events are synthesized
1766 * for all devices on the device list to the removed notifier to remove
1767 * the need for special case cleanup code.
1768 */
1769
1770 int unregister_netdevice_notifier(struct notifier_block *nb)
1771 {
1772 struct net *net;
1773 int err;
1774
1775 /* Close race with setup_net() and cleanup_net() */
1776 down_write(&pernet_ops_rwsem);
1777 rtnl_lock();
1778 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1779 if (err)
1780 goto unlock;
1781
1782 for_each_net(net)
1783 call_netdevice_unregister_net_notifiers(nb, net);
1784
1785 unlock:
1786 rtnl_unlock();
1787 up_write(&pernet_ops_rwsem);
1788 return err;
1789 }
1790 EXPORT_SYMBOL(unregister_netdevice_notifier);
1791
1792 static int __register_netdevice_notifier_net(struct net *net,
1793 struct notifier_block *nb,
1794 bool ignore_call_fail)
1795 {
1796 int err;
1797
1798 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1799 if (err)
1800 return err;
1801 if (dev_boot_phase)
1802 return 0;
1803
1804 err = call_netdevice_register_net_notifiers(nb, net);
1805 if (err && !ignore_call_fail)
1806 goto chain_unregister;
1807
1808 return 0;
1809
1810 chain_unregister:
1811 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1812 return err;
1813 }
1814
1815 static int __unregister_netdevice_notifier_net(struct net *net,
1816 struct notifier_block *nb)
1817 {
1818 int err;
1819
1820 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1821 if (err)
1822 return err;
1823
1824 call_netdevice_unregister_net_notifiers(nb, net);
1825 return 0;
1826 }
1827
1828 /**
1829 * register_netdevice_notifier_net - register a per-netns network notifier block
1830 * @net: network namespace
1831 * @nb: notifier
1832 *
1833 * Register a notifier to be called when network device events occur.
1834 * The notifier passed is linked into the kernel structures and must
1835 * not be reused until it has been unregistered. A negative errno code
1836 * is returned on a failure.
1837 *
1838 * When registered all registration and up events are replayed
1839 * to the new notifier to allow device to have a race free
1840 * view of the network device list.
1841 */
1842
1843 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1844 {
1845 int err;
1846
1847 rtnl_lock();
1848 err = __register_netdevice_notifier_net(net, nb, false);
1849 rtnl_unlock();
1850 return err;
1851 }
1852 EXPORT_SYMBOL(register_netdevice_notifier_net);
1853
1854 /**
1855 * unregister_netdevice_notifier_net - unregister a per-netns
1856 * network notifier block
1857 * @net: network namespace
1858 * @nb: notifier
1859 *
1860 * Unregister a notifier previously registered by
1861 * register_netdevice_notifier(). The notifier is unlinked into the
1862 * kernel structures and may then be reused. A negative errno code
1863 * is returned on a failure.
1864 *
1865 * After unregistering unregister and down device events are synthesized
1866 * for all devices on the device list to the removed notifier to remove
1867 * the need for special case cleanup code.
1868 */
1869
1870 int unregister_netdevice_notifier_net(struct net *net,
1871 struct notifier_block *nb)
1872 {
1873 int err;
1874
1875 rtnl_lock();
1876 err = __unregister_netdevice_notifier_net(net, nb);
1877 rtnl_unlock();
1878 return err;
1879 }
1880 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1881
1882 int register_netdevice_notifier_dev_net(struct net_device *dev,
1883 struct notifier_block *nb,
1884 struct netdev_net_notifier *nn)
1885 {
1886 int err;
1887
1888 rtnl_lock();
1889 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1890 if (!err) {
1891 nn->nb = nb;
1892 list_add(&nn->list, &dev->net_notifier_list);
1893 }
1894 rtnl_unlock();
1895 return err;
1896 }
1897 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1898
1899 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1900 struct notifier_block *nb,
1901 struct netdev_net_notifier *nn)
1902 {
1903 int err;
1904
1905 rtnl_lock();
1906 list_del(&nn->list);
1907 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1908 rtnl_unlock();
1909 return err;
1910 }
1911 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1912
1913 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1914 struct net *net)
1915 {
1916 struct netdev_net_notifier *nn;
1917
1918 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1919 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1920 __register_netdevice_notifier_net(net, nn->nb, true);
1921 }
1922 }
1923
1924 /**
1925 * call_netdevice_notifiers_info - call all network notifier blocks
1926 * @val: value passed unmodified to notifier function
1927 * @info: notifier information data
1928 *
1929 * Call all network notifier blocks. Parameters and return value
1930 * are as for raw_notifier_call_chain().
1931 */
1932
1933 static int call_netdevice_notifiers_info(unsigned long val,
1934 struct netdev_notifier_info *info)
1935 {
1936 struct net *net = dev_net(info->dev);
1937 int ret;
1938
1939 ASSERT_RTNL();
1940
1941 /* Run per-netns notifier block chain first, then run the global one.
1942 * Hopefully, one day, the global one is going to be removed after
1943 * all notifier block registrators get converted to be per-netns.
1944 */
1945 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1946 if (ret & NOTIFY_STOP_MASK)
1947 return ret;
1948 return raw_notifier_call_chain(&netdev_chain, val, info);
1949 }
1950
1951 static int call_netdevice_notifiers_extack(unsigned long val,
1952 struct net_device *dev,
1953 struct netlink_ext_ack *extack)
1954 {
1955 struct netdev_notifier_info info = {
1956 .dev = dev,
1957 .extack = extack,
1958 };
1959
1960 return call_netdevice_notifiers_info(val, &info);
1961 }
1962
1963 /**
1964 * call_netdevice_notifiers - call all network notifier blocks
1965 * @val: value passed unmodified to notifier function
1966 * @dev: net_device pointer passed unmodified to notifier function
1967 *
1968 * Call all network notifier blocks. Parameters and return value
1969 * are as for raw_notifier_call_chain().
1970 */
1971
1972 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1973 {
1974 return call_netdevice_notifiers_extack(val, dev, NULL);
1975 }
1976 EXPORT_SYMBOL(call_netdevice_notifiers);
1977
1978 /**
1979 * call_netdevice_notifiers_mtu - call all network notifier blocks
1980 * @val: value passed unmodified to notifier function
1981 * @dev: net_device pointer passed unmodified to notifier function
1982 * @arg: additional u32 argument passed to the notifier function
1983 *
1984 * Call all network notifier blocks. Parameters and return value
1985 * are as for raw_notifier_call_chain().
1986 */
1987 static int call_netdevice_notifiers_mtu(unsigned long val,
1988 struct net_device *dev, u32 arg)
1989 {
1990 struct netdev_notifier_info_ext info = {
1991 .info.dev = dev,
1992 .ext.mtu = arg,
1993 };
1994
1995 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1996
1997 return call_netdevice_notifiers_info(val, &info.info);
1998 }
1999
2000 #ifdef CONFIG_NET_INGRESS
2001 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2002
2003 void net_inc_ingress_queue(void)
2004 {
2005 static_branch_inc(&ingress_needed_key);
2006 }
2007 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2008
2009 void net_dec_ingress_queue(void)
2010 {
2011 static_branch_dec(&ingress_needed_key);
2012 }
2013 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2014 #endif
2015
2016 #ifdef CONFIG_NET_EGRESS
2017 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2018
2019 void net_inc_egress_queue(void)
2020 {
2021 static_branch_inc(&egress_needed_key);
2022 }
2023 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2024
2025 void net_dec_egress_queue(void)
2026 {
2027 static_branch_dec(&egress_needed_key);
2028 }
2029 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2030 #endif
2031
2032 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2033 #ifdef CONFIG_JUMP_LABEL
2034 static atomic_t netstamp_needed_deferred;
2035 static atomic_t netstamp_wanted;
2036 static void netstamp_clear(struct work_struct *work)
2037 {
2038 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2039 int wanted;
2040
2041 wanted = atomic_add_return(deferred, &netstamp_wanted);
2042 if (wanted > 0)
2043 static_branch_enable(&netstamp_needed_key);
2044 else
2045 static_branch_disable(&netstamp_needed_key);
2046 }
2047 static DECLARE_WORK(netstamp_work, netstamp_clear);
2048 #endif
2049
2050 void net_enable_timestamp(void)
2051 {
2052 #ifdef CONFIG_JUMP_LABEL
2053 int wanted;
2054
2055 while (1) {
2056 wanted = atomic_read(&netstamp_wanted);
2057 if (wanted <= 0)
2058 break;
2059 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2060 return;
2061 }
2062 atomic_inc(&netstamp_needed_deferred);
2063 schedule_work(&netstamp_work);
2064 #else
2065 static_branch_inc(&netstamp_needed_key);
2066 #endif
2067 }
2068 EXPORT_SYMBOL(net_enable_timestamp);
2069
2070 void net_disable_timestamp(void)
2071 {
2072 #ifdef CONFIG_JUMP_LABEL
2073 int wanted;
2074
2075 while (1) {
2076 wanted = atomic_read(&netstamp_wanted);
2077 if (wanted <= 1)
2078 break;
2079 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2080 return;
2081 }
2082 atomic_dec(&netstamp_needed_deferred);
2083 schedule_work(&netstamp_work);
2084 #else
2085 static_branch_dec(&netstamp_needed_key);
2086 #endif
2087 }
2088 EXPORT_SYMBOL(net_disable_timestamp);
2089
2090 static inline void net_timestamp_set(struct sk_buff *skb)
2091 {
2092 skb->tstamp = 0;
2093 if (static_branch_unlikely(&netstamp_needed_key))
2094 __net_timestamp(skb);
2095 }
2096
2097 #define net_timestamp_check(COND, SKB) \
2098 if (static_branch_unlikely(&netstamp_needed_key)) { \
2099 if ((COND) && !(SKB)->tstamp) \
2100 __net_timestamp(SKB); \
2101 } \
2102
2103 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2104 {
2105 unsigned int len;
2106
2107 if (!(dev->flags & IFF_UP))
2108 return false;
2109
2110 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2111 if (skb->len <= len)
2112 return true;
2113
2114 /* if TSO is enabled, we don't care about the length as the packet
2115 * could be forwarded without being segmented before
2116 */
2117 if (skb_is_gso(skb))
2118 return true;
2119
2120 return false;
2121 }
2122 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2123
2124 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2125 {
2126 int ret = ____dev_forward_skb(dev, skb);
2127
2128 if (likely(!ret)) {
2129 skb->protocol = eth_type_trans(skb, dev);
2130 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2131 }
2132
2133 return ret;
2134 }
2135 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2136
2137 /**
2138 * dev_forward_skb - loopback an skb to another netif
2139 *
2140 * @dev: destination network device
2141 * @skb: buffer to forward
2142 *
2143 * return values:
2144 * NET_RX_SUCCESS (no congestion)
2145 * NET_RX_DROP (packet was dropped, but freed)
2146 *
2147 * dev_forward_skb can be used for injecting an skb from the
2148 * start_xmit function of one device into the receive queue
2149 * of another device.
2150 *
2151 * The receiving device may be in another namespace, so
2152 * we have to clear all information in the skb that could
2153 * impact namespace isolation.
2154 */
2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2156 {
2157 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2158 }
2159 EXPORT_SYMBOL_GPL(dev_forward_skb);
2160
2161 static inline int deliver_skb(struct sk_buff *skb,
2162 struct packet_type *pt_prev,
2163 struct net_device *orig_dev)
2164 {
2165 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2166 return -ENOMEM;
2167 refcount_inc(&skb->users);
2168 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2169 }
2170
2171 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2172 struct packet_type **pt,
2173 struct net_device *orig_dev,
2174 __be16 type,
2175 struct list_head *ptype_list)
2176 {
2177 struct packet_type *ptype, *pt_prev = *pt;
2178
2179 list_for_each_entry_rcu(ptype, ptype_list, list) {
2180 if (ptype->type != type)
2181 continue;
2182 if (pt_prev)
2183 deliver_skb(skb, pt_prev, orig_dev);
2184 pt_prev = ptype;
2185 }
2186 *pt = pt_prev;
2187 }
2188
2189 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2190 {
2191 if (!ptype->af_packet_priv || !skb->sk)
2192 return false;
2193
2194 if (ptype->id_match)
2195 return ptype->id_match(ptype, skb->sk);
2196 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2197 return true;
2198
2199 return false;
2200 }
2201
2202 /**
2203 * dev_nit_active - return true if any network interface taps are in use
2204 *
2205 * @dev: network device to check for the presence of taps
2206 */
2207 bool dev_nit_active(struct net_device *dev)
2208 {
2209 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2210 }
2211 EXPORT_SYMBOL_GPL(dev_nit_active);
2212
2213 /*
2214 * Support routine. Sends outgoing frames to any network
2215 * taps currently in use.
2216 */
2217
2218 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2219 {
2220 struct packet_type *ptype;
2221 struct sk_buff *skb2 = NULL;
2222 struct packet_type *pt_prev = NULL;
2223 struct list_head *ptype_list = &ptype_all;
2224
2225 rcu_read_lock();
2226 again:
2227 list_for_each_entry_rcu(ptype, ptype_list, list) {
2228 if (ptype->ignore_outgoing)
2229 continue;
2230
2231 /* Never send packets back to the socket
2232 * they originated from - MvS (miquels@drinkel.ow.org)
2233 */
2234 if (skb_loop_sk(ptype, skb))
2235 continue;
2236
2237 if (pt_prev) {
2238 deliver_skb(skb2, pt_prev, skb->dev);
2239 pt_prev = ptype;
2240 continue;
2241 }
2242
2243 /* need to clone skb, done only once */
2244 skb2 = skb_clone(skb, GFP_ATOMIC);
2245 if (!skb2)
2246 goto out_unlock;
2247
2248 net_timestamp_set(skb2);
2249
2250 /* skb->nh should be correctly
2251 * set by sender, so that the second statement is
2252 * just protection against buggy protocols.
2253 */
2254 skb_reset_mac_header(skb2);
2255
2256 if (skb_network_header(skb2) < skb2->data ||
2257 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2258 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2259 ntohs(skb2->protocol),
2260 dev->name);
2261 skb_reset_network_header(skb2);
2262 }
2263
2264 skb2->transport_header = skb2->network_header;
2265 skb2->pkt_type = PACKET_OUTGOING;
2266 pt_prev = ptype;
2267 }
2268
2269 if (ptype_list == &ptype_all) {
2270 ptype_list = &dev->ptype_all;
2271 goto again;
2272 }
2273 out_unlock:
2274 if (pt_prev) {
2275 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2276 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2277 else
2278 kfree_skb(skb2);
2279 }
2280 rcu_read_unlock();
2281 }
2282 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2283
2284 /**
2285 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2286 * @dev: Network device
2287 * @txq: number of queues available
2288 *
2289 * If real_num_tx_queues is changed the tc mappings may no longer be
2290 * valid. To resolve this verify the tc mapping remains valid and if
2291 * not NULL the mapping. With no priorities mapping to this
2292 * offset/count pair it will no longer be used. In the worst case TC0
2293 * is invalid nothing can be done so disable priority mappings. If is
2294 * expected that drivers will fix this mapping if they can before
2295 * calling netif_set_real_num_tx_queues.
2296 */
2297 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2298 {
2299 int i;
2300 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2301
2302 /* If TC0 is invalidated disable TC mapping */
2303 if (tc->offset + tc->count > txq) {
2304 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2305 dev->num_tc = 0;
2306 return;
2307 }
2308
2309 /* Invalidated prio to tc mappings set to TC0 */
2310 for (i = 1; i < TC_BITMASK + 1; i++) {
2311 int q = netdev_get_prio_tc_map(dev, i);
2312
2313 tc = &dev->tc_to_txq[q];
2314 if (tc->offset + tc->count > txq) {
2315 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2316 i, q);
2317 netdev_set_prio_tc_map(dev, i, 0);
2318 }
2319 }
2320 }
2321
2322 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2323 {
2324 if (dev->num_tc) {
2325 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2326 int i;
2327
2328 /* walk through the TCs and see if it falls into any of them */
2329 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2330 if ((txq - tc->offset) < tc->count)
2331 return i;
2332 }
2333
2334 /* didn't find it, just return -1 to indicate no match */
2335 return -1;
2336 }
2337
2338 return 0;
2339 }
2340 EXPORT_SYMBOL(netdev_txq_to_tc);
2341
2342 #ifdef CONFIG_XPS
2343 struct static_key xps_needed __read_mostly;
2344 EXPORT_SYMBOL(xps_needed);
2345 struct static_key xps_rxqs_needed __read_mostly;
2346 EXPORT_SYMBOL(xps_rxqs_needed);
2347 static DEFINE_MUTEX(xps_map_mutex);
2348 #define xmap_dereference(P) \
2349 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2350
2351 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2352 int tci, u16 index)
2353 {
2354 struct xps_map *map = NULL;
2355 int pos;
2356
2357 if (dev_maps)
2358 map = xmap_dereference(dev_maps->attr_map[tci]);
2359 if (!map)
2360 return false;
2361
2362 for (pos = map->len; pos--;) {
2363 if (map->queues[pos] != index)
2364 continue;
2365
2366 if (map->len > 1) {
2367 map->queues[pos] = map->queues[--map->len];
2368 break;
2369 }
2370
2371 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2372 kfree_rcu(map, rcu);
2373 return false;
2374 }
2375
2376 return true;
2377 }
2378
2379 static bool remove_xps_queue_cpu(struct net_device *dev,
2380 struct xps_dev_maps *dev_maps,
2381 int cpu, u16 offset, u16 count)
2382 {
2383 int num_tc = dev->num_tc ? : 1;
2384 bool active = false;
2385 int tci;
2386
2387 for (tci = cpu * num_tc; num_tc--; tci++) {
2388 int i, j;
2389
2390 for (i = count, j = offset; i--; j++) {
2391 if (!remove_xps_queue(dev_maps, tci, j))
2392 break;
2393 }
2394
2395 active |= i < 0;
2396 }
2397
2398 return active;
2399 }
2400
2401 static void reset_xps_maps(struct net_device *dev,
2402 struct xps_dev_maps *dev_maps,
2403 bool is_rxqs_map)
2404 {
2405 if (is_rxqs_map) {
2406 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2407 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2408 } else {
2409 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2410 }
2411 static_key_slow_dec_cpuslocked(&xps_needed);
2412 kfree_rcu(dev_maps, rcu);
2413 }
2414
2415 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2416 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2417 u16 offset, u16 count, bool is_rxqs_map)
2418 {
2419 bool active = false;
2420 int i, j;
2421
2422 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2423 j < nr_ids;)
2424 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2425 count);
2426 if (!active)
2427 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2428
2429 if (!is_rxqs_map) {
2430 for (i = offset + (count - 1); count--; i--) {
2431 netdev_queue_numa_node_write(
2432 netdev_get_tx_queue(dev, i),
2433 NUMA_NO_NODE);
2434 }
2435 }
2436 }
2437
2438 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2439 u16 count)
2440 {
2441 const unsigned long *possible_mask = NULL;
2442 struct xps_dev_maps *dev_maps;
2443 unsigned int nr_ids;
2444
2445 if (!static_key_false(&xps_needed))
2446 return;
2447
2448 cpus_read_lock();
2449 mutex_lock(&xps_map_mutex);
2450
2451 if (static_key_false(&xps_rxqs_needed)) {
2452 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2453 if (dev_maps) {
2454 nr_ids = dev->num_rx_queues;
2455 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2456 offset, count, true);
2457 }
2458 }
2459
2460 dev_maps = xmap_dereference(dev->xps_cpus_map);
2461 if (!dev_maps)
2462 goto out_no_maps;
2463
2464 if (num_possible_cpus() > 1)
2465 possible_mask = cpumask_bits(cpu_possible_mask);
2466 nr_ids = nr_cpu_ids;
2467 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2468 false);
2469
2470 out_no_maps:
2471 mutex_unlock(&xps_map_mutex);
2472 cpus_read_unlock();
2473 }
2474
2475 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2476 {
2477 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2478 }
2479
2480 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2481 u16 index, bool is_rxqs_map)
2482 {
2483 struct xps_map *new_map;
2484 int alloc_len = XPS_MIN_MAP_ALLOC;
2485 int i, pos;
2486
2487 for (pos = 0; map && pos < map->len; pos++) {
2488 if (map->queues[pos] != index)
2489 continue;
2490 return map;
2491 }
2492
2493 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2494 if (map) {
2495 if (pos < map->alloc_len)
2496 return map;
2497
2498 alloc_len = map->alloc_len * 2;
2499 }
2500
2501 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2502 * map
2503 */
2504 if (is_rxqs_map)
2505 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2506 else
2507 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2508 cpu_to_node(attr_index));
2509 if (!new_map)
2510 return NULL;
2511
2512 for (i = 0; i < pos; i++)
2513 new_map->queues[i] = map->queues[i];
2514 new_map->alloc_len = alloc_len;
2515 new_map->len = pos;
2516
2517 return new_map;
2518 }
2519
2520 /* Must be called under cpus_read_lock */
2521 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2522 u16 index, bool is_rxqs_map)
2523 {
2524 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2525 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2526 int i, j, tci, numa_node_id = -2;
2527 int maps_sz, num_tc = 1, tc = 0;
2528 struct xps_map *map, *new_map;
2529 bool active = false;
2530 unsigned int nr_ids;
2531
2532 if (dev->num_tc) {
2533 /* Do not allow XPS on subordinate device directly */
2534 num_tc = dev->num_tc;
2535 if (num_tc < 0)
2536 return -EINVAL;
2537
2538 /* If queue belongs to subordinate dev use its map */
2539 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2540
2541 tc = netdev_txq_to_tc(dev, index);
2542 if (tc < 0)
2543 return -EINVAL;
2544 }
2545
2546 mutex_lock(&xps_map_mutex);
2547 if (is_rxqs_map) {
2548 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2549 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2550 nr_ids = dev->num_rx_queues;
2551 } else {
2552 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2553 if (num_possible_cpus() > 1) {
2554 online_mask = cpumask_bits(cpu_online_mask);
2555 possible_mask = cpumask_bits(cpu_possible_mask);
2556 }
2557 dev_maps = xmap_dereference(dev->xps_cpus_map);
2558 nr_ids = nr_cpu_ids;
2559 }
2560
2561 if (maps_sz < L1_CACHE_BYTES)
2562 maps_sz = L1_CACHE_BYTES;
2563
2564 /* allocate memory for queue storage */
2565 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2566 j < nr_ids;) {
2567 if (!new_dev_maps)
2568 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2569 if (!new_dev_maps) {
2570 mutex_unlock(&xps_map_mutex);
2571 return -ENOMEM;
2572 }
2573
2574 tci = j * num_tc + tc;
2575 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2576 NULL;
2577
2578 map = expand_xps_map(map, j, index, is_rxqs_map);
2579 if (!map)
2580 goto error;
2581
2582 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2583 }
2584
2585 if (!new_dev_maps)
2586 goto out_no_new_maps;
2587
2588 if (!dev_maps) {
2589 /* Increment static keys at most once per type */
2590 static_key_slow_inc_cpuslocked(&xps_needed);
2591 if (is_rxqs_map)
2592 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2593 }
2594
2595 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2596 j < nr_ids;) {
2597 /* copy maps belonging to foreign traffic classes */
2598 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2599 /* fill in the new device map from the old device map */
2600 map = xmap_dereference(dev_maps->attr_map[tci]);
2601 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2602 }
2603
2604 /* We need to explicitly update tci as prevous loop
2605 * could break out early if dev_maps is NULL.
2606 */
2607 tci = j * num_tc + tc;
2608
2609 if (netif_attr_test_mask(j, mask, nr_ids) &&
2610 netif_attr_test_online(j, online_mask, nr_ids)) {
2611 /* add tx-queue to CPU/rx-queue maps */
2612 int pos = 0;
2613
2614 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2615 while ((pos < map->len) && (map->queues[pos] != index))
2616 pos++;
2617
2618 if (pos == map->len)
2619 map->queues[map->len++] = index;
2620 #ifdef CONFIG_NUMA
2621 if (!is_rxqs_map) {
2622 if (numa_node_id == -2)
2623 numa_node_id = cpu_to_node(j);
2624 else if (numa_node_id != cpu_to_node(j))
2625 numa_node_id = -1;
2626 }
2627 #endif
2628 } else if (dev_maps) {
2629 /* fill in the new device map from the old device map */
2630 map = xmap_dereference(dev_maps->attr_map[tci]);
2631 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2632 }
2633
2634 /* copy maps belonging to foreign traffic classes */
2635 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2636 /* fill in the new device map from the old device map */
2637 map = xmap_dereference(dev_maps->attr_map[tci]);
2638 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2639 }
2640 }
2641
2642 if (is_rxqs_map)
2643 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2644 else
2645 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2646
2647 /* Cleanup old maps */
2648 if (!dev_maps)
2649 goto out_no_old_maps;
2650
2651 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2652 j < nr_ids;) {
2653 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2654 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2655 map = xmap_dereference(dev_maps->attr_map[tci]);
2656 if (map && map != new_map)
2657 kfree_rcu(map, rcu);
2658 }
2659 }
2660
2661 kfree_rcu(dev_maps, rcu);
2662
2663 out_no_old_maps:
2664 dev_maps = new_dev_maps;
2665 active = true;
2666
2667 out_no_new_maps:
2668 if (!is_rxqs_map) {
2669 /* update Tx queue numa node */
2670 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2671 (numa_node_id >= 0) ?
2672 numa_node_id : NUMA_NO_NODE);
2673 }
2674
2675 if (!dev_maps)
2676 goto out_no_maps;
2677
2678 /* removes tx-queue from unused CPUs/rx-queues */
2679 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2680 j < nr_ids;) {
2681 for (i = tc, tci = j * num_tc; i--; tci++)
2682 active |= remove_xps_queue(dev_maps, tci, index);
2683 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2684 !netif_attr_test_online(j, online_mask, nr_ids))
2685 active |= remove_xps_queue(dev_maps, tci, index);
2686 for (i = num_tc - tc, tci++; --i; tci++)
2687 active |= remove_xps_queue(dev_maps, tci, index);
2688 }
2689
2690 /* free map if not active */
2691 if (!active)
2692 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2693
2694 out_no_maps:
2695 mutex_unlock(&xps_map_mutex);
2696
2697 return 0;
2698 error:
2699 /* remove any maps that we added */
2700 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2701 j < nr_ids;) {
2702 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2703 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2704 map = dev_maps ?
2705 xmap_dereference(dev_maps->attr_map[tci]) :
2706 NULL;
2707 if (new_map && new_map != map)
2708 kfree(new_map);
2709 }
2710 }
2711
2712 mutex_unlock(&xps_map_mutex);
2713
2714 kfree(new_dev_maps);
2715 return -ENOMEM;
2716 }
2717 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2718
2719 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2720 u16 index)
2721 {
2722 int ret;
2723
2724 cpus_read_lock();
2725 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2726 cpus_read_unlock();
2727
2728 return ret;
2729 }
2730 EXPORT_SYMBOL(netif_set_xps_queue);
2731
2732 #endif
2733 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2734 {
2735 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2736
2737 /* Unbind any subordinate channels */
2738 while (txq-- != &dev->_tx[0]) {
2739 if (txq->sb_dev)
2740 netdev_unbind_sb_channel(dev, txq->sb_dev);
2741 }
2742 }
2743
2744 void netdev_reset_tc(struct net_device *dev)
2745 {
2746 #ifdef CONFIG_XPS
2747 netif_reset_xps_queues_gt(dev, 0);
2748 #endif
2749 netdev_unbind_all_sb_channels(dev);
2750
2751 /* Reset TC configuration of device */
2752 dev->num_tc = 0;
2753 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2754 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2755 }
2756 EXPORT_SYMBOL(netdev_reset_tc);
2757
2758 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2759 {
2760 if (tc >= dev->num_tc)
2761 return -EINVAL;
2762
2763 #ifdef CONFIG_XPS
2764 netif_reset_xps_queues(dev, offset, count);
2765 #endif
2766 dev->tc_to_txq[tc].count = count;
2767 dev->tc_to_txq[tc].offset = offset;
2768 return 0;
2769 }
2770 EXPORT_SYMBOL(netdev_set_tc_queue);
2771
2772 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2773 {
2774 if (num_tc > TC_MAX_QUEUE)
2775 return -EINVAL;
2776
2777 #ifdef CONFIG_XPS
2778 netif_reset_xps_queues_gt(dev, 0);
2779 #endif
2780 netdev_unbind_all_sb_channels(dev);
2781
2782 dev->num_tc = num_tc;
2783 return 0;
2784 }
2785 EXPORT_SYMBOL(netdev_set_num_tc);
2786
2787 void netdev_unbind_sb_channel(struct net_device *dev,
2788 struct net_device *sb_dev)
2789 {
2790 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2791
2792 #ifdef CONFIG_XPS
2793 netif_reset_xps_queues_gt(sb_dev, 0);
2794 #endif
2795 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2796 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2797
2798 while (txq-- != &dev->_tx[0]) {
2799 if (txq->sb_dev == sb_dev)
2800 txq->sb_dev = NULL;
2801 }
2802 }
2803 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2804
2805 int netdev_bind_sb_channel_queue(struct net_device *dev,
2806 struct net_device *sb_dev,
2807 u8 tc, u16 count, u16 offset)
2808 {
2809 /* Make certain the sb_dev and dev are already configured */
2810 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2811 return -EINVAL;
2812
2813 /* We cannot hand out queues we don't have */
2814 if ((offset + count) > dev->real_num_tx_queues)
2815 return -EINVAL;
2816
2817 /* Record the mapping */
2818 sb_dev->tc_to_txq[tc].count = count;
2819 sb_dev->tc_to_txq[tc].offset = offset;
2820
2821 /* Provide a way for Tx queue to find the tc_to_txq map or
2822 * XPS map for itself.
2823 */
2824 while (count--)
2825 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2826
2827 return 0;
2828 }
2829 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2830
2831 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2832 {
2833 /* Do not use a multiqueue device to represent a subordinate channel */
2834 if (netif_is_multiqueue(dev))
2835 return -ENODEV;
2836
2837 /* We allow channels 1 - 32767 to be used for subordinate channels.
2838 * Channel 0 is meant to be "native" mode and used only to represent
2839 * the main root device. We allow writing 0 to reset the device back
2840 * to normal mode after being used as a subordinate channel.
2841 */
2842 if (channel > S16_MAX)
2843 return -EINVAL;
2844
2845 dev->num_tc = -channel;
2846
2847 return 0;
2848 }
2849 EXPORT_SYMBOL(netdev_set_sb_channel);
2850
2851 /*
2852 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2853 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2854 */
2855 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2856 {
2857 bool disabling;
2858 int rc;
2859
2860 disabling = txq < dev->real_num_tx_queues;
2861
2862 if (txq < 1 || txq > dev->num_tx_queues)
2863 return -EINVAL;
2864
2865 if (dev->reg_state == NETREG_REGISTERED ||
2866 dev->reg_state == NETREG_UNREGISTERING) {
2867 ASSERT_RTNL();
2868
2869 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2870 txq);
2871 if (rc)
2872 return rc;
2873
2874 if (dev->num_tc)
2875 netif_setup_tc(dev, txq);
2876
2877 dev->real_num_tx_queues = txq;
2878
2879 if (disabling) {
2880 synchronize_net();
2881 qdisc_reset_all_tx_gt(dev, txq);
2882 #ifdef CONFIG_XPS
2883 netif_reset_xps_queues_gt(dev, txq);
2884 #endif
2885 }
2886 } else {
2887 dev->real_num_tx_queues = txq;
2888 }
2889
2890 return 0;
2891 }
2892 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2893
2894 #ifdef CONFIG_SYSFS
2895 /**
2896 * netif_set_real_num_rx_queues - set actual number of RX queues used
2897 * @dev: Network device
2898 * @rxq: Actual number of RX queues
2899 *
2900 * This must be called either with the rtnl_lock held or before
2901 * registration of the net device. Returns 0 on success, or a
2902 * negative error code. If called before registration, it always
2903 * succeeds.
2904 */
2905 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2906 {
2907 int rc;
2908
2909 if (rxq < 1 || rxq > dev->num_rx_queues)
2910 return -EINVAL;
2911
2912 if (dev->reg_state == NETREG_REGISTERED) {
2913 ASSERT_RTNL();
2914
2915 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2916 rxq);
2917 if (rc)
2918 return rc;
2919 }
2920
2921 dev->real_num_rx_queues = rxq;
2922 return 0;
2923 }
2924 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2925 #endif
2926
2927 /**
2928 * netif_get_num_default_rss_queues - default number of RSS queues
2929 *
2930 * This routine should set an upper limit on the number of RSS queues
2931 * used by default by multiqueue devices.
2932 */
2933 int netif_get_num_default_rss_queues(void)
2934 {
2935 return is_kdump_kernel() ?
2936 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2937 }
2938 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2939
2940 static void __netif_reschedule(struct Qdisc *q)
2941 {
2942 struct softnet_data *sd;
2943 unsigned long flags;
2944
2945 local_irq_save(flags);
2946 sd = this_cpu_ptr(&softnet_data);
2947 q->next_sched = NULL;
2948 *sd->output_queue_tailp = q;
2949 sd->output_queue_tailp = &q->next_sched;
2950 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2951 local_irq_restore(flags);
2952 }
2953
2954 void __netif_schedule(struct Qdisc *q)
2955 {
2956 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2957 __netif_reschedule(q);
2958 }
2959 EXPORT_SYMBOL(__netif_schedule);
2960
2961 struct dev_kfree_skb_cb {
2962 enum skb_free_reason reason;
2963 };
2964
2965 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2966 {
2967 return (struct dev_kfree_skb_cb *)skb->cb;
2968 }
2969
2970 void netif_schedule_queue(struct netdev_queue *txq)
2971 {
2972 rcu_read_lock();
2973 if (!netif_xmit_stopped(txq)) {
2974 struct Qdisc *q = rcu_dereference(txq->qdisc);
2975
2976 __netif_schedule(q);
2977 }
2978 rcu_read_unlock();
2979 }
2980 EXPORT_SYMBOL(netif_schedule_queue);
2981
2982 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2983 {
2984 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2985 struct Qdisc *q;
2986
2987 rcu_read_lock();
2988 q = rcu_dereference(dev_queue->qdisc);
2989 __netif_schedule(q);
2990 rcu_read_unlock();
2991 }
2992 }
2993 EXPORT_SYMBOL(netif_tx_wake_queue);
2994
2995 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2996 {
2997 unsigned long flags;
2998
2999 if (unlikely(!skb))
3000 return;
3001
3002 if (likely(refcount_read(&skb->users) == 1)) {
3003 smp_rmb();
3004 refcount_set(&skb->users, 0);
3005 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3006 return;
3007 }
3008 get_kfree_skb_cb(skb)->reason = reason;
3009 local_irq_save(flags);
3010 skb->next = __this_cpu_read(softnet_data.completion_queue);
3011 __this_cpu_write(softnet_data.completion_queue, skb);
3012 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3013 local_irq_restore(flags);
3014 }
3015 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3016
3017 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3018 {
3019 if (in_irq() || irqs_disabled())
3020 __dev_kfree_skb_irq(skb, reason);
3021 else
3022 dev_kfree_skb(skb);
3023 }
3024 EXPORT_SYMBOL(__dev_kfree_skb_any);
3025
3026
3027 /**
3028 * netif_device_detach - mark device as removed
3029 * @dev: network device
3030 *
3031 * Mark device as removed from system and therefore no longer available.
3032 */
3033 void netif_device_detach(struct net_device *dev)
3034 {
3035 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3036 netif_running(dev)) {
3037 netif_tx_stop_all_queues(dev);
3038 }
3039 }
3040 EXPORT_SYMBOL(netif_device_detach);
3041
3042 /**
3043 * netif_device_attach - mark device as attached
3044 * @dev: network device
3045 *
3046 * Mark device as attached from system and restart if needed.
3047 */
3048 void netif_device_attach(struct net_device *dev)
3049 {
3050 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3051 netif_running(dev)) {
3052 netif_tx_wake_all_queues(dev);
3053 __netdev_watchdog_up(dev);
3054 }
3055 }
3056 EXPORT_SYMBOL(netif_device_attach);
3057
3058 /*
3059 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3060 * to be used as a distribution range.
3061 */
3062 static u16 skb_tx_hash(const struct net_device *dev,
3063 const struct net_device *sb_dev,
3064 struct sk_buff *skb)
3065 {
3066 u32 hash;
3067 u16 qoffset = 0;
3068 u16 qcount = dev->real_num_tx_queues;
3069
3070 if (dev->num_tc) {
3071 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3072
3073 qoffset = sb_dev->tc_to_txq[tc].offset;
3074 qcount = sb_dev->tc_to_txq[tc].count;
3075 }
3076
3077 if (skb_rx_queue_recorded(skb)) {
3078 hash = skb_get_rx_queue(skb);
3079 if (hash >= qoffset)
3080 hash -= qoffset;
3081 while (unlikely(hash >= qcount))
3082 hash -= qcount;
3083 return hash + qoffset;
3084 }
3085
3086 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3087 }
3088
3089 static void skb_warn_bad_offload(const struct sk_buff *skb)
3090 {
3091 static const netdev_features_t null_features;
3092 struct net_device *dev = skb->dev;
3093 const char *name = "";
3094
3095 if (!net_ratelimit())
3096 return;
3097
3098 if (dev) {
3099 if (dev->dev.parent)
3100 name = dev_driver_string(dev->dev.parent);
3101 else
3102 name = netdev_name(dev);
3103 }
3104 skb_dump(KERN_WARNING, skb, false);
3105 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3106 name, dev ? &dev->features : &null_features,
3107 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3108 }
3109
3110 /*
3111 * Invalidate hardware checksum when packet is to be mangled, and
3112 * complete checksum manually on outgoing path.
3113 */
3114 int skb_checksum_help(struct sk_buff *skb)
3115 {
3116 __wsum csum;
3117 int ret = 0, offset;
3118
3119 if (skb->ip_summed == CHECKSUM_COMPLETE)
3120 goto out_set_summed;
3121
3122 if (unlikely(skb_shinfo(skb)->gso_size)) {
3123 skb_warn_bad_offload(skb);
3124 return -EINVAL;
3125 }
3126
3127 /* Before computing a checksum, we should make sure no frag could
3128 * be modified by an external entity : checksum could be wrong.
3129 */
3130 if (skb_has_shared_frag(skb)) {
3131 ret = __skb_linearize(skb);
3132 if (ret)
3133 goto out;
3134 }
3135
3136 offset = skb_checksum_start_offset(skb);
3137 BUG_ON(offset >= skb_headlen(skb));
3138 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3139
3140 offset += skb->csum_offset;
3141 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3142
3143 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3144 if (ret)
3145 goto out;
3146
3147 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3148 out_set_summed:
3149 skb->ip_summed = CHECKSUM_NONE;
3150 out:
3151 return ret;
3152 }
3153 EXPORT_SYMBOL(skb_checksum_help);
3154
3155 int skb_crc32c_csum_help(struct sk_buff *skb)
3156 {
3157 __le32 crc32c_csum;
3158 int ret = 0, offset, start;
3159
3160 if (skb->ip_summed != CHECKSUM_PARTIAL)
3161 goto out;
3162
3163 if (unlikely(skb_is_gso(skb)))
3164 goto out;
3165
3166 /* Before computing a checksum, we should make sure no frag could
3167 * be modified by an external entity : checksum could be wrong.
3168 */
3169 if (unlikely(skb_has_shared_frag(skb))) {
3170 ret = __skb_linearize(skb);
3171 if (ret)
3172 goto out;
3173 }
3174 start = skb_checksum_start_offset(skb);
3175 offset = start + offsetof(struct sctphdr, checksum);
3176 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3177 ret = -EINVAL;
3178 goto out;
3179 }
3180
3181 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3182 if (ret)
3183 goto out;
3184
3185 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3186 skb->len - start, ~(__u32)0,
3187 crc32c_csum_stub));
3188 *(__le32 *)(skb->data + offset) = crc32c_csum;
3189 skb->ip_summed = CHECKSUM_NONE;
3190 skb->csum_not_inet = 0;
3191 out:
3192 return ret;
3193 }
3194
3195 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3196 {
3197 __be16 type = skb->protocol;
3198
3199 /* Tunnel gso handlers can set protocol to ethernet. */
3200 if (type == htons(ETH_P_TEB)) {
3201 struct ethhdr *eth;
3202
3203 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3204 return 0;
3205
3206 eth = (struct ethhdr *)skb->data;
3207 type = eth->h_proto;
3208 }
3209
3210 return __vlan_get_protocol(skb, type, depth);
3211 }
3212
3213 /**
3214 * skb_mac_gso_segment - mac layer segmentation handler.
3215 * @skb: buffer to segment
3216 * @features: features for the output path (see dev->features)
3217 */
3218 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3219 netdev_features_t features)
3220 {
3221 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3222 struct packet_offload *ptype;
3223 int vlan_depth = skb->mac_len;
3224 __be16 type = skb_network_protocol(skb, &vlan_depth);
3225
3226 if (unlikely(!type))
3227 return ERR_PTR(-EINVAL);
3228
3229 __skb_pull(skb, vlan_depth);
3230
3231 rcu_read_lock();
3232 list_for_each_entry_rcu(ptype, &offload_base, list) {
3233 if (ptype->type == type && ptype->callbacks.gso_segment) {
3234 segs = ptype->callbacks.gso_segment(skb, features);
3235 break;
3236 }
3237 }
3238 rcu_read_unlock();
3239
3240 __skb_push(skb, skb->data - skb_mac_header(skb));
3241
3242 return segs;
3243 }
3244 EXPORT_SYMBOL(skb_mac_gso_segment);
3245
3246
3247 /* openvswitch calls this on rx path, so we need a different check.
3248 */
3249 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3250 {
3251 if (tx_path)
3252 return skb->ip_summed != CHECKSUM_PARTIAL &&
3253 skb->ip_summed != CHECKSUM_UNNECESSARY;
3254
3255 return skb->ip_summed == CHECKSUM_NONE;
3256 }
3257
3258 /**
3259 * __skb_gso_segment - Perform segmentation on skb.
3260 * @skb: buffer to segment
3261 * @features: features for the output path (see dev->features)
3262 * @tx_path: whether it is called in TX path
3263 *
3264 * This function segments the given skb and returns a list of segments.
3265 *
3266 * It may return NULL if the skb requires no segmentation. This is
3267 * only possible when GSO is used for verifying header integrity.
3268 *
3269 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3270 */
3271 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3272 netdev_features_t features, bool tx_path)
3273 {
3274 struct sk_buff *segs;
3275
3276 if (unlikely(skb_needs_check(skb, tx_path))) {
3277 int err;
3278
3279 /* We're going to init ->check field in TCP or UDP header */
3280 err = skb_cow_head(skb, 0);
3281 if (err < 0)
3282 return ERR_PTR(err);
3283 }
3284
3285 /* Only report GSO partial support if it will enable us to
3286 * support segmentation on this frame without needing additional
3287 * work.
3288 */
3289 if (features & NETIF_F_GSO_PARTIAL) {
3290 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3291 struct net_device *dev = skb->dev;
3292
3293 partial_features |= dev->features & dev->gso_partial_features;
3294 if (!skb_gso_ok(skb, features | partial_features))
3295 features &= ~NETIF_F_GSO_PARTIAL;
3296 }
3297
3298 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3299 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3300
3301 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3302 SKB_GSO_CB(skb)->encap_level = 0;
3303
3304 skb_reset_mac_header(skb);
3305 skb_reset_mac_len(skb);
3306
3307 segs = skb_mac_gso_segment(skb, features);
3308
3309 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3310 skb_warn_bad_offload(skb);
3311
3312 return segs;
3313 }
3314 EXPORT_SYMBOL(__skb_gso_segment);
3315
3316 /* Take action when hardware reception checksum errors are detected. */
3317 #ifdef CONFIG_BUG
3318 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3319 {
3320 if (net_ratelimit()) {
3321 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3322 skb_dump(KERN_ERR, skb, true);
3323 dump_stack();
3324 }
3325 }
3326 EXPORT_SYMBOL(netdev_rx_csum_fault);
3327 #endif
3328
3329 /* XXX: check that highmem exists at all on the given machine. */
3330 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3331 {
3332 #ifdef CONFIG_HIGHMEM
3333 int i;
3334
3335 if (!(dev->features & NETIF_F_HIGHDMA)) {
3336 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3337 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3338
3339 if (PageHighMem(skb_frag_page(frag)))
3340 return 1;
3341 }
3342 }
3343 #endif
3344 return 0;
3345 }
3346
3347 /* If MPLS offload request, verify we are testing hardware MPLS features
3348 * instead of standard features for the netdev.
3349 */
3350 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3351 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3352 netdev_features_t features,
3353 __be16 type)
3354 {
3355 if (eth_p_mpls(type))
3356 features &= skb->dev->mpls_features;
3357
3358 return features;
3359 }
3360 #else
3361 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3362 netdev_features_t features,
3363 __be16 type)
3364 {
3365 return features;
3366 }
3367 #endif
3368
3369 static netdev_features_t harmonize_features(struct sk_buff *skb,
3370 netdev_features_t features)
3371 {
3372 int tmp;
3373 __be16 type;
3374
3375 type = skb_network_protocol(skb, &tmp);
3376 features = net_mpls_features(skb, features, type);
3377
3378 if (skb->ip_summed != CHECKSUM_NONE &&
3379 !can_checksum_protocol(features, type)) {
3380 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3381 }
3382 if (illegal_highdma(skb->dev, skb))
3383 features &= ~NETIF_F_SG;
3384
3385 return features;
3386 }
3387
3388 netdev_features_t passthru_features_check(struct sk_buff *skb,
3389 struct net_device *dev,
3390 netdev_features_t features)
3391 {
3392 return features;
3393 }
3394 EXPORT_SYMBOL(passthru_features_check);
3395
3396 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3397 struct net_device *dev,
3398 netdev_features_t features)
3399 {
3400 return vlan_features_check(skb, features);
3401 }
3402
3403 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3404 struct net_device *dev,
3405 netdev_features_t features)
3406 {
3407 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3408
3409 if (gso_segs > dev->gso_max_segs)
3410 return features & ~NETIF_F_GSO_MASK;
3411
3412 /* Support for GSO partial features requires software
3413 * intervention before we can actually process the packets
3414 * so we need to strip support for any partial features now
3415 * and we can pull them back in after we have partially
3416 * segmented the frame.
3417 */
3418 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3419 features &= ~dev->gso_partial_features;
3420
3421 /* Make sure to clear the IPv4 ID mangling feature if the
3422 * IPv4 header has the potential to be fragmented.
3423 */
3424 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3425 struct iphdr *iph = skb->encapsulation ?
3426 inner_ip_hdr(skb) : ip_hdr(skb);
3427
3428 if (!(iph->frag_off & htons(IP_DF)))
3429 features &= ~NETIF_F_TSO_MANGLEID;
3430 }
3431
3432 return features;
3433 }
3434
3435 netdev_features_t netif_skb_features(struct sk_buff *skb)
3436 {
3437 struct net_device *dev = skb->dev;
3438 netdev_features_t features = dev->features;
3439
3440 if (skb_is_gso(skb))
3441 features = gso_features_check(skb, dev, features);
3442
3443 /* If encapsulation offload request, verify we are testing
3444 * hardware encapsulation features instead of standard
3445 * features for the netdev
3446 */
3447 if (skb->encapsulation)
3448 features &= dev->hw_enc_features;
3449
3450 if (skb_vlan_tagged(skb))
3451 features = netdev_intersect_features(features,
3452 dev->vlan_features |
3453 NETIF_F_HW_VLAN_CTAG_TX |
3454 NETIF_F_HW_VLAN_STAG_TX);
3455
3456 if (dev->netdev_ops->ndo_features_check)
3457 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3458 features);
3459 else
3460 features &= dflt_features_check(skb, dev, features);
3461
3462 return harmonize_features(skb, features);
3463 }
3464 EXPORT_SYMBOL(netif_skb_features);
3465
3466 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3467 struct netdev_queue *txq, bool more)
3468 {
3469 unsigned int len;
3470 int rc;
3471
3472 if (dev_nit_active(dev))
3473 dev_queue_xmit_nit(skb, dev);
3474
3475 len = skb->len;
3476 trace_net_dev_start_xmit(skb, dev);
3477 rc = netdev_start_xmit(skb, dev, txq, more);
3478 trace_net_dev_xmit(skb, rc, dev, len);
3479
3480 return rc;
3481 }
3482
3483 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3484 struct netdev_queue *txq, int *ret)
3485 {
3486 struct sk_buff *skb = first;
3487 int rc = NETDEV_TX_OK;
3488
3489 while (skb) {
3490 struct sk_buff *next = skb->next;
3491
3492 skb_mark_not_on_list(skb);
3493 rc = xmit_one(skb, dev, txq, next != NULL);
3494 if (unlikely(!dev_xmit_complete(rc))) {
3495 skb->next = next;
3496 goto out;
3497 }
3498
3499 skb = next;
3500 if (netif_tx_queue_stopped(txq) && skb) {
3501 rc = NETDEV_TX_BUSY;
3502 break;
3503 }
3504 }
3505
3506 out:
3507 *ret = rc;
3508 return skb;
3509 }
3510
3511 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3512 netdev_features_t features)
3513 {
3514 if (skb_vlan_tag_present(skb) &&
3515 !vlan_hw_offload_capable(features, skb->vlan_proto))
3516 skb = __vlan_hwaccel_push_inside(skb);
3517 return skb;
3518 }
3519
3520 int skb_csum_hwoffload_help(struct sk_buff *skb,
3521 const netdev_features_t features)
3522 {
3523 if (unlikely(skb->csum_not_inet))
3524 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3525 skb_crc32c_csum_help(skb);
3526
3527 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3528 }
3529 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3530
3531 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3532 {
3533 netdev_features_t features;
3534
3535 features = netif_skb_features(skb);
3536 skb = validate_xmit_vlan(skb, features);
3537 if (unlikely(!skb))
3538 goto out_null;
3539
3540 skb = sk_validate_xmit_skb(skb, dev);
3541 if (unlikely(!skb))
3542 goto out_null;
3543
3544 if (netif_needs_gso(skb, features)) {
3545 struct sk_buff *segs;
3546
3547 segs = skb_gso_segment(skb, features);
3548 if (IS_ERR(segs)) {
3549 goto out_kfree_skb;
3550 } else if (segs) {
3551 consume_skb(skb);
3552 skb = segs;
3553 }
3554 } else {
3555 if (skb_needs_linearize(skb, features) &&
3556 __skb_linearize(skb))
3557 goto out_kfree_skb;
3558
3559 /* If packet is not checksummed and device does not
3560 * support checksumming for this protocol, complete
3561 * checksumming here.
3562 */
3563 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3564 if (skb->encapsulation)
3565 skb_set_inner_transport_header(skb,
3566 skb_checksum_start_offset(skb));
3567 else
3568 skb_set_transport_header(skb,
3569 skb_checksum_start_offset(skb));
3570 if (skb_csum_hwoffload_help(skb, features))
3571 goto out_kfree_skb;
3572 }
3573 }
3574
3575 skb = validate_xmit_xfrm(skb, features, again);
3576
3577 return skb;
3578
3579 out_kfree_skb:
3580 kfree_skb(skb);
3581 out_null:
3582 atomic_long_inc(&dev->tx_dropped);
3583 return NULL;
3584 }
3585
3586 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3587 {
3588 struct sk_buff *next, *head = NULL, *tail;
3589
3590 for (; skb != NULL; skb = next) {
3591 next = skb->next;
3592 skb_mark_not_on_list(skb);
3593
3594 /* in case skb wont be segmented, point to itself */
3595 skb->prev = skb;
3596
3597 skb = validate_xmit_skb(skb, dev, again);
3598 if (!skb)
3599 continue;
3600
3601 if (!head)
3602 head = skb;
3603 else
3604 tail->next = skb;
3605 /* If skb was segmented, skb->prev points to
3606 * the last segment. If not, it still contains skb.
3607 */
3608 tail = skb->prev;
3609 }
3610 return head;
3611 }
3612 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3613
3614 static void qdisc_pkt_len_init(struct sk_buff *skb)
3615 {
3616 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3617
3618 qdisc_skb_cb(skb)->pkt_len = skb->len;
3619
3620 /* To get more precise estimation of bytes sent on wire,
3621 * we add to pkt_len the headers size of all segments
3622 */
3623 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3624 unsigned int hdr_len;
3625 u16 gso_segs = shinfo->gso_segs;
3626
3627 /* mac layer + network layer */
3628 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3629
3630 /* + transport layer */
3631 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3632 const struct tcphdr *th;
3633 struct tcphdr _tcphdr;
3634
3635 th = skb_header_pointer(skb, skb_transport_offset(skb),
3636 sizeof(_tcphdr), &_tcphdr);
3637 if (likely(th))
3638 hdr_len += __tcp_hdrlen(th);
3639 } else {
3640 struct udphdr _udphdr;
3641
3642 if (skb_header_pointer(skb, skb_transport_offset(skb),
3643 sizeof(_udphdr), &_udphdr))
3644 hdr_len += sizeof(struct udphdr);
3645 }
3646
3647 if (shinfo->gso_type & SKB_GSO_DODGY)
3648 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3649 shinfo->gso_size);
3650
3651 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3652 }
3653 }
3654
3655 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3656 struct net_device *dev,
3657 struct netdev_queue *txq)
3658 {
3659 spinlock_t *root_lock = qdisc_lock(q);
3660 struct sk_buff *to_free = NULL;
3661 bool contended;
3662 int rc;
3663
3664 qdisc_calculate_pkt_len(skb, q);
3665
3666 if (q->flags & TCQ_F_NOLOCK) {
3667 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3668 qdisc_run(q);
3669
3670 if (unlikely(to_free))
3671 kfree_skb_list(to_free);
3672 return rc;
3673 }
3674
3675 /*
3676 * Heuristic to force contended enqueues to serialize on a
3677 * separate lock before trying to get qdisc main lock.
3678 * This permits qdisc->running owner to get the lock more
3679 * often and dequeue packets faster.
3680 */
3681 contended = qdisc_is_running(q);
3682 if (unlikely(contended))
3683 spin_lock(&q->busylock);
3684
3685 spin_lock(root_lock);
3686 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3687 __qdisc_drop(skb, &to_free);
3688 rc = NET_XMIT_DROP;
3689 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3690 qdisc_run_begin(q)) {
3691 /*
3692 * This is a work-conserving queue; there are no old skbs
3693 * waiting to be sent out; and the qdisc is not running -
3694 * xmit the skb directly.
3695 */
3696
3697 qdisc_bstats_update(q, skb);
3698
3699 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3700 if (unlikely(contended)) {
3701 spin_unlock(&q->busylock);
3702 contended = false;
3703 }
3704 __qdisc_run(q);
3705 }
3706
3707 qdisc_run_end(q);
3708 rc = NET_XMIT_SUCCESS;
3709 } else {
3710 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3711 if (qdisc_run_begin(q)) {
3712 if (unlikely(contended)) {
3713 spin_unlock(&q->busylock);
3714 contended = false;
3715 }
3716 __qdisc_run(q);
3717 qdisc_run_end(q);
3718 }
3719 }
3720 spin_unlock(root_lock);
3721 if (unlikely(to_free))
3722 kfree_skb_list(to_free);
3723 if (unlikely(contended))
3724 spin_unlock(&q->busylock);
3725 return rc;
3726 }
3727
3728 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3729 static void skb_update_prio(struct sk_buff *skb)
3730 {
3731 const struct netprio_map *map;
3732 const struct sock *sk;
3733 unsigned int prioidx;
3734
3735 if (skb->priority)
3736 return;
3737 map = rcu_dereference_bh(skb->dev->priomap);
3738 if (!map)
3739 return;
3740 sk = skb_to_full_sk(skb);
3741 if (!sk)
3742 return;
3743
3744 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3745
3746 if (prioidx < map->priomap_len)
3747 skb->priority = map->priomap[prioidx];
3748 }
3749 #else
3750 #define skb_update_prio(skb)
3751 #endif
3752
3753 /**
3754 * dev_loopback_xmit - loop back @skb
3755 * @net: network namespace this loopback is happening in
3756 * @sk: sk needed to be a netfilter okfn
3757 * @skb: buffer to transmit
3758 */
3759 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3760 {
3761 skb_reset_mac_header(skb);
3762 __skb_pull(skb, skb_network_offset(skb));
3763 skb->pkt_type = PACKET_LOOPBACK;
3764 skb->ip_summed = CHECKSUM_UNNECESSARY;
3765 WARN_ON(!skb_dst(skb));
3766 skb_dst_force(skb);
3767 netif_rx_ni(skb);
3768 return 0;
3769 }
3770 EXPORT_SYMBOL(dev_loopback_xmit);
3771
3772 #ifdef CONFIG_NET_EGRESS
3773 static struct sk_buff *
3774 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3775 {
3776 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3777 struct tcf_result cl_res;
3778
3779 if (!miniq)
3780 return skb;
3781
3782 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3783 mini_qdisc_bstats_cpu_update(miniq, skb);
3784
3785 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3786 case TC_ACT_OK:
3787 case TC_ACT_RECLASSIFY:
3788 skb->tc_index = TC_H_MIN(cl_res.classid);
3789 break;
3790 case TC_ACT_SHOT:
3791 mini_qdisc_qstats_cpu_drop(miniq);
3792 *ret = NET_XMIT_DROP;
3793 kfree_skb(skb);
3794 return NULL;
3795 case TC_ACT_STOLEN:
3796 case TC_ACT_QUEUED:
3797 case TC_ACT_TRAP:
3798 *ret = NET_XMIT_SUCCESS;
3799 consume_skb(skb);
3800 return NULL;
3801 case TC_ACT_REDIRECT:
3802 /* No need to push/pop skb's mac_header here on egress! */
3803 skb_do_redirect(skb);
3804 *ret = NET_XMIT_SUCCESS;
3805 return NULL;
3806 default:
3807 break;
3808 }
3809
3810 return skb;
3811 }
3812 #endif /* CONFIG_NET_EGRESS */
3813
3814 #ifdef CONFIG_XPS
3815 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3816 struct xps_dev_maps *dev_maps, unsigned int tci)
3817 {
3818 struct xps_map *map;
3819 int queue_index = -1;
3820
3821 if (dev->num_tc) {
3822 tci *= dev->num_tc;
3823 tci += netdev_get_prio_tc_map(dev, skb->priority);
3824 }
3825
3826 map = rcu_dereference(dev_maps->attr_map[tci]);
3827 if (map) {
3828 if (map->len == 1)
3829 queue_index = map->queues[0];
3830 else
3831 queue_index = map->queues[reciprocal_scale(
3832 skb_get_hash(skb), map->len)];
3833 if (unlikely(queue_index >= dev->real_num_tx_queues))
3834 queue_index = -1;
3835 }
3836 return queue_index;
3837 }
3838 #endif
3839
3840 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3841 struct sk_buff *skb)
3842 {
3843 #ifdef CONFIG_XPS
3844 struct xps_dev_maps *dev_maps;
3845 struct sock *sk = skb->sk;
3846 int queue_index = -1;
3847
3848 if (!static_key_false(&xps_needed))
3849 return -1;
3850
3851 rcu_read_lock();
3852 if (!static_key_false(&xps_rxqs_needed))
3853 goto get_cpus_map;
3854
3855 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3856 if (dev_maps) {
3857 int tci = sk_rx_queue_get(sk);
3858
3859 if (tci >= 0 && tci < dev->num_rx_queues)
3860 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3861 tci);
3862 }
3863
3864 get_cpus_map:
3865 if (queue_index < 0) {
3866 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3867 if (dev_maps) {
3868 unsigned int tci = skb->sender_cpu - 1;
3869
3870 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3871 tci);
3872 }
3873 }
3874 rcu_read_unlock();
3875
3876 return queue_index;
3877 #else
3878 return -1;
3879 #endif
3880 }
3881
3882 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3883 struct net_device *sb_dev)
3884 {
3885 return 0;
3886 }
3887 EXPORT_SYMBOL(dev_pick_tx_zero);
3888
3889 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3890 struct net_device *sb_dev)
3891 {
3892 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3893 }
3894 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3895
3896 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3897 struct net_device *sb_dev)
3898 {
3899 struct sock *sk = skb->sk;
3900 int queue_index = sk_tx_queue_get(sk);
3901
3902 sb_dev = sb_dev ? : dev;
3903
3904 if (queue_index < 0 || skb->ooo_okay ||
3905 queue_index >= dev->real_num_tx_queues) {
3906 int new_index = get_xps_queue(dev, sb_dev, skb);
3907
3908 if (new_index < 0)
3909 new_index = skb_tx_hash(dev, sb_dev, skb);
3910
3911 if (queue_index != new_index && sk &&
3912 sk_fullsock(sk) &&
3913 rcu_access_pointer(sk->sk_dst_cache))
3914 sk_tx_queue_set(sk, new_index);
3915
3916 queue_index = new_index;
3917 }
3918
3919 return queue_index;
3920 }
3921 EXPORT_SYMBOL(netdev_pick_tx);
3922
3923 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3924 struct sk_buff *skb,
3925 struct net_device *sb_dev)
3926 {
3927 int queue_index = 0;
3928
3929 #ifdef CONFIG_XPS
3930 u32 sender_cpu = skb->sender_cpu - 1;
3931
3932 if (sender_cpu >= (u32)NR_CPUS)
3933 skb->sender_cpu = raw_smp_processor_id() + 1;
3934 #endif
3935
3936 if (dev->real_num_tx_queues != 1) {
3937 const struct net_device_ops *ops = dev->netdev_ops;
3938
3939 if (ops->ndo_select_queue)
3940 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3941 else
3942 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3943
3944 queue_index = netdev_cap_txqueue(dev, queue_index);
3945 }
3946
3947 skb_set_queue_mapping(skb, queue_index);
3948 return netdev_get_tx_queue(dev, queue_index);
3949 }
3950
3951 /**
3952 * __dev_queue_xmit - transmit a buffer
3953 * @skb: buffer to transmit
3954 * @sb_dev: suboordinate device used for L2 forwarding offload
3955 *
3956 * Queue a buffer for transmission to a network device. The caller must
3957 * have set the device and priority and built the buffer before calling
3958 * this function. The function can be called from an interrupt.
3959 *
3960 * A negative errno code is returned on a failure. A success does not
3961 * guarantee the frame will be transmitted as it may be dropped due
3962 * to congestion or traffic shaping.
3963 *
3964 * -----------------------------------------------------------------------------------
3965 * I notice this method can also return errors from the queue disciplines,
3966 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3967 * be positive.
3968 *
3969 * Regardless of the return value, the skb is consumed, so it is currently
3970 * difficult to retry a send to this method. (You can bump the ref count
3971 * before sending to hold a reference for retry if you are careful.)
3972 *
3973 * When calling this method, interrupts MUST be enabled. This is because
3974 * the BH enable code must have IRQs enabled so that it will not deadlock.
3975 * --BLG
3976 */
3977 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3978 {
3979 struct net_device *dev = skb->dev;
3980 struct netdev_queue *txq;
3981 struct Qdisc *q;
3982 int rc = -ENOMEM;
3983 bool again = false;
3984
3985 skb_reset_mac_header(skb);
3986
3987 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3988 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3989
3990 /* Disable soft irqs for various locks below. Also
3991 * stops preemption for RCU.
3992 */
3993 rcu_read_lock_bh();
3994
3995 skb_update_prio(skb);
3996
3997 qdisc_pkt_len_init(skb);
3998 #ifdef CONFIG_NET_CLS_ACT
3999 skb->tc_at_ingress = 0;
4000 # ifdef CONFIG_NET_EGRESS
4001 if (static_branch_unlikely(&egress_needed_key)) {
4002 skb = sch_handle_egress(skb, &rc, dev);
4003 if (!skb)
4004 goto out;
4005 }
4006 # endif
4007 #endif
4008 /* If device/qdisc don't need skb->dst, release it right now while
4009 * its hot in this cpu cache.
4010 */
4011 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4012 skb_dst_drop(skb);
4013 else
4014 skb_dst_force(skb);
4015
4016 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4017 q = rcu_dereference_bh(txq->qdisc);
4018
4019 trace_net_dev_queue(skb);
4020 if (q->enqueue) {
4021 rc = __dev_xmit_skb(skb, q, dev, txq);
4022 goto out;
4023 }
4024
4025 /* The device has no queue. Common case for software devices:
4026 * loopback, all the sorts of tunnels...
4027
4028 * Really, it is unlikely that netif_tx_lock protection is necessary
4029 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4030 * counters.)
4031 * However, it is possible, that they rely on protection
4032 * made by us here.
4033
4034 * Check this and shot the lock. It is not prone from deadlocks.
4035 *Either shot noqueue qdisc, it is even simpler 8)
4036 */
4037 if (dev->flags & IFF_UP) {
4038 int cpu = smp_processor_id(); /* ok because BHs are off */
4039
4040 if (txq->xmit_lock_owner != cpu) {
4041 if (dev_xmit_recursion())
4042 goto recursion_alert;
4043
4044 skb = validate_xmit_skb(skb, dev, &again);
4045 if (!skb)
4046 goto out;
4047
4048 HARD_TX_LOCK(dev, txq, cpu);
4049
4050 if (!netif_xmit_stopped(txq)) {
4051 dev_xmit_recursion_inc();
4052 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4053 dev_xmit_recursion_dec();
4054 if (dev_xmit_complete(rc)) {
4055 HARD_TX_UNLOCK(dev, txq);
4056 goto out;
4057 }
4058 }
4059 HARD_TX_UNLOCK(dev, txq);
4060 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4061 dev->name);
4062 } else {
4063 /* Recursion is detected! It is possible,
4064 * unfortunately
4065 */
4066 recursion_alert:
4067 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4068 dev->name);
4069 }
4070 }
4071
4072 rc = -ENETDOWN;
4073 rcu_read_unlock_bh();
4074
4075 atomic_long_inc(&dev->tx_dropped);
4076 kfree_skb_list(skb);
4077 return rc;
4078 out:
4079 rcu_read_unlock_bh();
4080 return rc;
4081 }
4082
4083 int dev_queue_xmit(struct sk_buff *skb)
4084 {
4085 return __dev_queue_xmit(skb, NULL);
4086 }
4087 EXPORT_SYMBOL(dev_queue_xmit);
4088
4089 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4090 {
4091 return __dev_queue_xmit(skb, sb_dev);
4092 }
4093 EXPORT_SYMBOL(dev_queue_xmit_accel);
4094
4095 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4096 {
4097 struct net_device *dev = skb->dev;
4098 struct sk_buff *orig_skb = skb;
4099 struct netdev_queue *txq;
4100 int ret = NETDEV_TX_BUSY;
4101 bool again = false;
4102
4103 if (unlikely(!netif_running(dev) ||
4104 !netif_carrier_ok(dev)))
4105 goto drop;
4106
4107 skb = validate_xmit_skb_list(skb, dev, &again);
4108 if (skb != orig_skb)
4109 goto drop;
4110
4111 skb_set_queue_mapping(skb, queue_id);
4112 txq = skb_get_tx_queue(dev, skb);
4113
4114 local_bh_disable();
4115
4116 HARD_TX_LOCK(dev, txq, smp_processor_id());
4117 if (!netif_xmit_frozen_or_drv_stopped(txq))
4118 ret = netdev_start_xmit(skb, dev, txq, false);
4119 HARD_TX_UNLOCK(dev, txq);
4120
4121 local_bh_enable();
4122
4123 if (!dev_xmit_complete(ret))
4124 kfree_skb(skb);
4125
4126 return ret;
4127 drop:
4128 atomic_long_inc(&dev->tx_dropped);
4129 kfree_skb_list(skb);
4130 return NET_XMIT_DROP;
4131 }
4132 EXPORT_SYMBOL(dev_direct_xmit);
4133
4134 /*************************************************************************
4135 * Receiver routines
4136 *************************************************************************/
4137
4138 int netdev_max_backlog __read_mostly = 1000;
4139 EXPORT_SYMBOL(netdev_max_backlog);
4140
4141 int netdev_tstamp_prequeue __read_mostly = 1;
4142 int netdev_budget __read_mostly = 300;
4143 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4144 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4145 int weight_p __read_mostly = 64; /* old backlog weight */
4146 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4147 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4148 int dev_rx_weight __read_mostly = 64;
4149 int dev_tx_weight __read_mostly = 64;
4150 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4151 int gro_normal_batch __read_mostly = 8;
4152
4153 /* Called with irq disabled */
4154 static inline void ____napi_schedule(struct softnet_data *sd,
4155 struct napi_struct *napi)
4156 {
4157 list_add_tail(&napi->poll_list, &sd->poll_list);
4158 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4159 }
4160
4161 #ifdef CONFIG_RPS
4162
4163 /* One global table that all flow-based protocols share. */
4164 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4165 EXPORT_SYMBOL(rps_sock_flow_table);
4166 u32 rps_cpu_mask __read_mostly;
4167 EXPORT_SYMBOL(rps_cpu_mask);
4168
4169 struct static_key_false rps_needed __read_mostly;
4170 EXPORT_SYMBOL(rps_needed);
4171 struct static_key_false rfs_needed __read_mostly;
4172 EXPORT_SYMBOL(rfs_needed);
4173
4174 static struct rps_dev_flow *
4175 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4176 struct rps_dev_flow *rflow, u16 next_cpu)
4177 {
4178 if (next_cpu < nr_cpu_ids) {
4179 #ifdef CONFIG_RFS_ACCEL
4180 struct netdev_rx_queue *rxqueue;
4181 struct rps_dev_flow_table *flow_table;
4182 struct rps_dev_flow *old_rflow;
4183 u32 flow_id;
4184 u16 rxq_index;
4185 int rc;
4186
4187 /* Should we steer this flow to a different hardware queue? */
4188 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4189 !(dev->features & NETIF_F_NTUPLE))
4190 goto out;
4191 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4192 if (rxq_index == skb_get_rx_queue(skb))
4193 goto out;
4194
4195 rxqueue = dev->_rx + rxq_index;
4196 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4197 if (!flow_table)
4198 goto out;
4199 flow_id = skb_get_hash(skb) & flow_table->mask;
4200 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4201 rxq_index, flow_id);
4202 if (rc < 0)
4203 goto out;
4204 old_rflow = rflow;
4205 rflow = &flow_table->flows[flow_id];
4206 rflow->filter = rc;
4207 if (old_rflow->filter == rflow->filter)
4208 old_rflow->filter = RPS_NO_FILTER;
4209 out:
4210 #endif
4211 rflow->last_qtail =
4212 per_cpu(softnet_data, next_cpu).input_queue_head;
4213 }
4214
4215 rflow->cpu = next_cpu;
4216 return rflow;
4217 }
4218
4219 /*
4220 * get_rps_cpu is called from netif_receive_skb and returns the target
4221 * CPU from the RPS map of the receiving queue for a given skb.
4222 * rcu_read_lock must be held on entry.
4223 */
4224 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4225 struct rps_dev_flow **rflowp)
4226 {
4227 const struct rps_sock_flow_table *sock_flow_table;
4228 struct netdev_rx_queue *rxqueue = dev->_rx;
4229 struct rps_dev_flow_table *flow_table;
4230 struct rps_map *map;
4231 int cpu = -1;
4232 u32 tcpu;
4233 u32 hash;
4234
4235 if (skb_rx_queue_recorded(skb)) {
4236 u16 index = skb_get_rx_queue(skb);
4237
4238 if (unlikely(index >= dev->real_num_rx_queues)) {
4239 WARN_ONCE(dev->real_num_rx_queues > 1,
4240 "%s received packet on queue %u, but number "
4241 "of RX queues is %u\n",
4242 dev->name, index, dev->real_num_rx_queues);
4243 goto done;
4244 }
4245 rxqueue += index;
4246 }
4247
4248 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4249
4250 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4251 map = rcu_dereference(rxqueue->rps_map);
4252 if (!flow_table && !map)
4253 goto done;
4254
4255 skb_reset_network_header(skb);
4256 hash = skb_get_hash(skb);
4257 if (!hash)
4258 goto done;
4259
4260 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4261 if (flow_table && sock_flow_table) {
4262 struct rps_dev_flow *rflow;
4263 u32 next_cpu;
4264 u32 ident;
4265
4266 /* First check into global flow table if there is a match */
4267 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4268 if ((ident ^ hash) & ~rps_cpu_mask)
4269 goto try_rps;
4270
4271 next_cpu = ident & rps_cpu_mask;
4272
4273 /* OK, now we know there is a match,
4274 * we can look at the local (per receive queue) flow table
4275 */
4276 rflow = &flow_table->flows[hash & flow_table->mask];
4277 tcpu = rflow->cpu;
4278
4279 /*
4280 * If the desired CPU (where last recvmsg was done) is
4281 * different from current CPU (one in the rx-queue flow
4282 * table entry), switch if one of the following holds:
4283 * - Current CPU is unset (>= nr_cpu_ids).
4284 * - Current CPU is offline.
4285 * - The current CPU's queue tail has advanced beyond the
4286 * last packet that was enqueued using this table entry.
4287 * This guarantees that all previous packets for the flow
4288 * have been dequeued, thus preserving in order delivery.
4289 */
4290 if (unlikely(tcpu != next_cpu) &&
4291 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4292 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4293 rflow->last_qtail)) >= 0)) {
4294 tcpu = next_cpu;
4295 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4296 }
4297
4298 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4299 *rflowp = rflow;
4300 cpu = tcpu;
4301 goto done;
4302 }
4303 }
4304
4305 try_rps:
4306
4307 if (map) {
4308 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4309 if (cpu_online(tcpu)) {
4310 cpu = tcpu;
4311 goto done;
4312 }
4313 }
4314
4315 done:
4316 return cpu;
4317 }
4318
4319 #ifdef CONFIG_RFS_ACCEL
4320
4321 /**
4322 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4323 * @dev: Device on which the filter was set
4324 * @rxq_index: RX queue index
4325 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4326 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4327 *
4328 * Drivers that implement ndo_rx_flow_steer() should periodically call
4329 * this function for each installed filter and remove the filters for
4330 * which it returns %true.
4331 */
4332 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4333 u32 flow_id, u16 filter_id)
4334 {
4335 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4336 struct rps_dev_flow_table *flow_table;
4337 struct rps_dev_flow *rflow;
4338 bool expire = true;
4339 unsigned int cpu;
4340
4341 rcu_read_lock();
4342 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4343 if (flow_table && flow_id <= flow_table->mask) {
4344 rflow = &flow_table->flows[flow_id];
4345 cpu = READ_ONCE(rflow->cpu);
4346 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4347 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4348 rflow->last_qtail) <
4349 (int)(10 * flow_table->mask)))
4350 expire = false;
4351 }
4352 rcu_read_unlock();
4353 return expire;
4354 }
4355 EXPORT_SYMBOL(rps_may_expire_flow);
4356
4357 #endif /* CONFIG_RFS_ACCEL */
4358
4359 /* Called from hardirq (IPI) context */
4360 static void rps_trigger_softirq(void *data)
4361 {
4362 struct softnet_data *sd = data;
4363
4364 ____napi_schedule(sd, &sd->backlog);
4365 sd->received_rps++;
4366 }
4367
4368 #endif /* CONFIG_RPS */
4369
4370 /*
4371 * Check if this softnet_data structure is another cpu one
4372 * If yes, queue it to our IPI list and return 1
4373 * If no, return 0
4374 */
4375 static int rps_ipi_queued(struct softnet_data *sd)
4376 {
4377 #ifdef CONFIG_RPS
4378 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4379
4380 if (sd != mysd) {
4381 sd->rps_ipi_next = mysd->rps_ipi_list;
4382 mysd->rps_ipi_list = sd;
4383
4384 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4385 return 1;
4386 }
4387 #endif /* CONFIG_RPS */
4388 return 0;
4389 }
4390
4391 #ifdef CONFIG_NET_FLOW_LIMIT
4392 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4393 #endif
4394
4395 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4396 {
4397 #ifdef CONFIG_NET_FLOW_LIMIT
4398 struct sd_flow_limit *fl;
4399 struct softnet_data *sd;
4400 unsigned int old_flow, new_flow;
4401
4402 if (qlen < (netdev_max_backlog >> 1))
4403 return false;
4404
4405 sd = this_cpu_ptr(&softnet_data);
4406
4407 rcu_read_lock();
4408 fl = rcu_dereference(sd->flow_limit);
4409 if (fl) {
4410 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4411 old_flow = fl->history[fl->history_head];
4412 fl->history[fl->history_head] = new_flow;
4413
4414 fl->history_head++;
4415 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4416
4417 if (likely(fl->buckets[old_flow]))
4418 fl->buckets[old_flow]--;
4419
4420 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4421 fl->count++;
4422 rcu_read_unlock();
4423 return true;
4424 }
4425 }
4426 rcu_read_unlock();
4427 #endif
4428 return false;
4429 }
4430
4431 /*
4432 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4433 * queue (may be a remote CPU queue).
4434 */
4435 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4436 unsigned int *qtail)
4437 {
4438 struct softnet_data *sd;
4439 unsigned long flags;
4440 unsigned int qlen;
4441
4442 sd = &per_cpu(softnet_data, cpu);
4443
4444 local_irq_save(flags);
4445
4446 rps_lock(sd);
4447 if (!netif_running(skb->dev))
4448 goto drop;
4449 qlen = skb_queue_len(&sd->input_pkt_queue);
4450 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4451 if (qlen) {
4452 enqueue:
4453 __skb_queue_tail(&sd->input_pkt_queue, skb);
4454 input_queue_tail_incr_save(sd, qtail);
4455 rps_unlock(sd);
4456 local_irq_restore(flags);
4457 return NET_RX_SUCCESS;
4458 }
4459
4460 /* Schedule NAPI for backlog device
4461 * We can use non atomic operation since we own the queue lock
4462 */
4463 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4464 if (!rps_ipi_queued(sd))
4465 ____napi_schedule(sd, &sd->backlog);
4466 }
4467 goto enqueue;
4468 }
4469
4470 drop:
4471 sd->dropped++;
4472 rps_unlock(sd);
4473
4474 local_irq_restore(flags);
4475
4476 atomic_long_inc(&skb->dev->rx_dropped);
4477 kfree_skb(skb);
4478 return NET_RX_DROP;
4479 }
4480
4481 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4482 {
4483 struct net_device *dev = skb->dev;
4484 struct netdev_rx_queue *rxqueue;
4485
4486 rxqueue = dev->_rx;
4487
4488 if (skb_rx_queue_recorded(skb)) {
4489 u16 index = skb_get_rx_queue(skb);
4490
4491 if (unlikely(index >= dev->real_num_rx_queues)) {
4492 WARN_ONCE(dev->real_num_rx_queues > 1,
4493 "%s received packet on queue %u, but number "
4494 "of RX queues is %u\n",
4495 dev->name, index, dev->real_num_rx_queues);
4496
4497 return rxqueue; /* Return first rxqueue */
4498 }
4499 rxqueue += index;
4500 }
4501 return rxqueue;
4502 }
4503
4504 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4505 struct xdp_buff *xdp,
4506 struct bpf_prog *xdp_prog)
4507 {
4508 struct netdev_rx_queue *rxqueue;
4509 void *orig_data, *orig_data_end;
4510 u32 metalen, act = XDP_DROP;
4511 __be16 orig_eth_type;
4512 struct ethhdr *eth;
4513 bool orig_bcast;
4514 int hlen, off;
4515 u32 mac_len;
4516
4517 /* Reinjected packets coming from act_mirred or similar should
4518 * not get XDP generic processing.
4519 */
4520 if (skb_is_redirected(skb))
4521 return XDP_PASS;
4522
4523 /* XDP packets must be linear and must have sufficient headroom
4524 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4525 * native XDP provides, thus we need to do it here as well.
4526 */
4527 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4528 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4529 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4530 int troom = skb->tail + skb->data_len - skb->end;
4531
4532 /* In case we have to go down the path and also linearize,
4533 * then lets do the pskb_expand_head() work just once here.
4534 */
4535 if (pskb_expand_head(skb,
4536 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4537 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4538 goto do_drop;
4539 if (skb_linearize(skb))
4540 goto do_drop;
4541 }
4542
4543 /* The XDP program wants to see the packet starting at the MAC
4544 * header.
4545 */
4546 mac_len = skb->data - skb_mac_header(skb);
4547 hlen = skb_headlen(skb) + mac_len;
4548 xdp->data = skb->data - mac_len;
4549 xdp->data_meta = xdp->data;
4550 xdp->data_end = xdp->data + hlen;
4551 xdp->data_hard_start = skb->data - skb_headroom(skb);
4552 orig_data_end = xdp->data_end;
4553 orig_data = xdp->data;
4554 eth = (struct ethhdr *)xdp->data;
4555 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4556 orig_eth_type = eth->h_proto;
4557
4558 rxqueue = netif_get_rxqueue(skb);
4559 xdp->rxq = &rxqueue->xdp_rxq;
4560
4561 act = bpf_prog_run_xdp(xdp_prog, xdp);
4562
4563 /* check if bpf_xdp_adjust_head was used */
4564 off = xdp->data - orig_data;
4565 if (off) {
4566 if (off > 0)
4567 __skb_pull(skb, off);
4568 else if (off < 0)
4569 __skb_push(skb, -off);
4570
4571 skb->mac_header += off;
4572 skb_reset_network_header(skb);
4573 }
4574
4575 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4576 * pckt.
4577 */
4578 off = orig_data_end - xdp->data_end;
4579 if (off != 0) {
4580 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4581 skb->len -= off;
4582
4583 }
4584
4585 /* check if XDP changed eth hdr such SKB needs update */
4586 eth = (struct ethhdr *)xdp->data;
4587 if ((orig_eth_type != eth->h_proto) ||
4588 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4589 __skb_push(skb, ETH_HLEN);
4590 skb->protocol = eth_type_trans(skb, skb->dev);
4591 }
4592
4593 switch (act) {
4594 case XDP_REDIRECT:
4595 case XDP_TX:
4596 __skb_push(skb, mac_len);
4597 break;
4598 case XDP_PASS:
4599 metalen = xdp->data - xdp->data_meta;
4600 if (metalen)
4601 skb_metadata_set(skb, metalen);
4602 break;
4603 default:
4604 bpf_warn_invalid_xdp_action(act);
4605 /* fall through */
4606 case XDP_ABORTED:
4607 trace_xdp_exception(skb->dev, xdp_prog, act);
4608 /* fall through */
4609 case XDP_DROP:
4610 do_drop:
4611 kfree_skb(skb);
4612 break;
4613 }
4614
4615 return act;
4616 }
4617
4618 /* When doing generic XDP we have to bypass the qdisc layer and the
4619 * network taps in order to match in-driver-XDP behavior.
4620 */
4621 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4622 {
4623 struct net_device *dev = skb->dev;
4624 struct netdev_queue *txq;
4625 bool free_skb = true;
4626 int cpu, rc;
4627
4628 txq = netdev_core_pick_tx(dev, skb, NULL);
4629 cpu = smp_processor_id();
4630 HARD_TX_LOCK(dev, txq, cpu);
4631 if (!netif_xmit_stopped(txq)) {
4632 rc = netdev_start_xmit(skb, dev, txq, 0);
4633 if (dev_xmit_complete(rc))
4634 free_skb = false;
4635 }
4636 HARD_TX_UNLOCK(dev, txq);
4637 if (free_skb) {
4638 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4639 kfree_skb(skb);
4640 }
4641 }
4642
4643 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4644
4645 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4646 {
4647 if (xdp_prog) {
4648 struct xdp_buff xdp;
4649 u32 act;
4650 int err;
4651
4652 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4653 if (act != XDP_PASS) {
4654 switch (act) {
4655 case XDP_REDIRECT:
4656 err = xdp_do_generic_redirect(skb->dev, skb,
4657 &xdp, xdp_prog);
4658 if (err)
4659 goto out_redir;
4660 break;
4661 case XDP_TX:
4662 generic_xdp_tx(skb, xdp_prog);
4663 break;
4664 }
4665 return XDP_DROP;
4666 }
4667 }
4668 return XDP_PASS;
4669 out_redir:
4670 kfree_skb(skb);
4671 return XDP_DROP;
4672 }
4673 EXPORT_SYMBOL_GPL(do_xdp_generic);
4674
4675 static int netif_rx_internal(struct sk_buff *skb)
4676 {
4677 int ret;
4678
4679 net_timestamp_check(netdev_tstamp_prequeue, skb);
4680
4681 trace_netif_rx(skb);
4682
4683 #ifdef CONFIG_RPS
4684 if (static_branch_unlikely(&rps_needed)) {
4685 struct rps_dev_flow voidflow, *rflow = &voidflow;
4686 int cpu;
4687
4688 preempt_disable();
4689 rcu_read_lock();
4690
4691 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4692 if (cpu < 0)
4693 cpu = smp_processor_id();
4694
4695 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4696
4697 rcu_read_unlock();
4698 preempt_enable();
4699 } else
4700 #endif
4701 {
4702 unsigned int qtail;
4703
4704 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4705 put_cpu();
4706 }
4707 return ret;
4708 }
4709
4710 /**
4711 * netif_rx - post buffer to the network code
4712 * @skb: buffer to post
4713 *
4714 * This function receives a packet from a device driver and queues it for
4715 * the upper (protocol) levels to process. It always succeeds. The buffer
4716 * may be dropped during processing for congestion control or by the
4717 * protocol layers.
4718 *
4719 * return values:
4720 * NET_RX_SUCCESS (no congestion)
4721 * NET_RX_DROP (packet was dropped)
4722 *
4723 */
4724
4725 int netif_rx(struct sk_buff *skb)
4726 {
4727 int ret;
4728
4729 trace_netif_rx_entry(skb);
4730
4731 ret = netif_rx_internal(skb);
4732 trace_netif_rx_exit(ret);
4733
4734 return ret;
4735 }
4736 EXPORT_SYMBOL(netif_rx);
4737
4738 int netif_rx_ni(struct sk_buff *skb)
4739 {
4740 int err;
4741
4742 trace_netif_rx_ni_entry(skb);
4743
4744 preempt_disable();
4745 err = netif_rx_internal(skb);
4746 if (local_softirq_pending())
4747 do_softirq();
4748 preempt_enable();
4749 trace_netif_rx_ni_exit(err);
4750
4751 return err;
4752 }
4753 EXPORT_SYMBOL(netif_rx_ni);
4754
4755 static __latent_entropy void net_tx_action(struct softirq_action *h)
4756 {
4757 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4758
4759 if (sd->completion_queue) {
4760 struct sk_buff *clist;
4761
4762 local_irq_disable();
4763 clist = sd->completion_queue;
4764 sd->completion_queue = NULL;
4765 local_irq_enable();
4766
4767 while (clist) {
4768 struct sk_buff *skb = clist;
4769
4770 clist = clist->next;
4771
4772 WARN_ON(refcount_read(&skb->users));
4773 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4774 trace_consume_skb(skb);
4775 else
4776 trace_kfree_skb(skb, net_tx_action);
4777
4778 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4779 __kfree_skb(skb);
4780 else
4781 __kfree_skb_defer(skb);
4782 }
4783
4784 __kfree_skb_flush();
4785 }
4786
4787 if (sd->output_queue) {
4788 struct Qdisc *head;
4789
4790 local_irq_disable();
4791 head = sd->output_queue;
4792 sd->output_queue = NULL;
4793 sd->output_queue_tailp = &sd->output_queue;
4794 local_irq_enable();
4795
4796 while (head) {
4797 struct Qdisc *q = head;
4798 spinlock_t *root_lock = NULL;
4799
4800 head = head->next_sched;
4801
4802 if (!(q->flags & TCQ_F_NOLOCK)) {
4803 root_lock = qdisc_lock(q);
4804 spin_lock(root_lock);
4805 }
4806 /* We need to make sure head->next_sched is read
4807 * before clearing __QDISC_STATE_SCHED
4808 */
4809 smp_mb__before_atomic();
4810 clear_bit(__QDISC_STATE_SCHED, &q->state);
4811 qdisc_run(q);
4812 if (root_lock)
4813 spin_unlock(root_lock);
4814 }
4815 }
4816
4817 xfrm_dev_backlog(sd);
4818 }
4819
4820 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4821 /* This hook is defined here for ATM LANE */
4822 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4823 unsigned char *addr) __read_mostly;
4824 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4825 #endif
4826
4827 static inline struct sk_buff *
4828 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4829 struct net_device *orig_dev)
4830 {
4831 #ifdef CONFIG_NET_CLS_ACT
4832 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4833 struct tcf_result cl_res;
4834
4835 /* If there's at least one ingress present somewhere (so
4836 * we get here via enabled static key), remaining devices
4837 * that are not configured with an ingress qdisc will bail
4838 * out here.
4839 */
4840 if (!miniq)
4841 return skb;
4842
4843 if (*pt_prev) {
4844 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4845 *pt_prev = NULL;
4846 }
4847
4848 qdisc_skb_cb(skb)->pkt_len = skb->len;
4849 skb->tc_at_ingress = 1;
4850 mini_qdisc_bstats_cpu_update(miniq, skb);
4851
4852 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
4853 &cl_res, false)) {
4854 case TC_ACT_OK:
4855 case TC_ACT_RECLASSIFY:
4856 skb->tc_index = TC_H_MIN(cl_res.classid);
4857 break;
4858 case TC_ACT_SHOT:
4859 mini_qdisc_qstats_cpu_drop(miniq);
4860 kfree_skb(skb);
4861 return NULL;
4862 case TC_ACT_STOLEN:
4863 case TC_ACT_QUEUED:
4864 case TC_ACT_TRAP:
4865 consume_skb(skb);
4866 return NULL;
4867 case TC_ACT_REDIRECT:
4868 /* skb_mac_header check was done by cls/act_bpf, so
4869 * we can safely push the L2 header back before
4870 * redirecting to another netdev
4871 */
4872 __skb_push(skb, skb->mac_len);
4873 skb_do_redirect(skb);
4874 return NULL;
4875 case TC_ACT_CONSUMED:
4876 return NULL;
4877 default:
4878 break;
4879 }
4880 #endif /* CONFIG_NET_CLS_ACT */
4881 return skb;
4882 }
4883
4884 /**
4885 * netdev_is_rx_handler_busy - check if receive handler is registered
4886 * @dev: device to check
4887 *
4888 * Check if a receive handler is already registered for a given device.
4889 * Return true if there one.
4890 *
4891 * The caller must hold the rtnl_mutex.
4892 */
4893 bool netdev_is_rx_handler_busy(struct net_device *dev)
4894 {
4895 ASSERT_RTNL();
4896 return dev && rtnl_dereference(dev->rx_handler);
4897 }
4898 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4899
4900 /**
4901 * netdev_rx_handler_register - register receive handler
4902 * @dev: device to register a handler for
4903 * @rx_handler: receive handler to register
4904 * @rx_handler_data: data pointer that is used by rx handler
4905 *
4906 * Register a receive handler for a device. This handler will then be
4907 * called from __netif_receive_skb. A negative errno code is returned
4908 * on a failure.
4909 *
4910 * The caller must hold the rtnl_mutex.
4911 *
4912 * For a general description of rx_handler, see enum rx_handler_result.
4913 */
4914 int netdev_rx_handler_register(struct net_device *dev,
4915 rx_handler_func_t *rx_handler,
4916 void *rx_handler_data)
4917 {
4918 if (netdev_is_rx_handler_busy(dev))
4919 return -EBUSY;
4920
4921 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4922 return -EINVAL;
4923
4924 /* Note: rx_handler_data must be set before rx_handler */
4925 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4926 rcu_assign_pointer(dev->rx_handler, rx_handler);
4927
4928 return 0;
4929 }
4930 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4931
4932 /**
4933 * netdev_rx_handler_unregister - unregister receive handler
4934 * @dev: device to unregister a handler from
4935 *
4936 * Unregister a receive handler from a device.
4937 *
4938 * The caller must hold the rtnl_mutex.
4939 */
4940 void netdev_rx_handler_unregister(struct net_device *dev)
4941 {
4942
4943 ASSERT_RTNL();
4944 RCU_INIT_POINTER(dev->rx_handler, NULL);
4945 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4946 * section has a guarantee to see a non NULL rx_handler_data
4947 * as well.
4948 */
4949 synchronize_net();
4950 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4951 }
4952 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4953
4954 /*
4955 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4956 * the special handling of PFMEMALLOC skbs.
4957 */
4958 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4959 {
4960 switch (skb->protocol) {
4961 case htons(ETH_P_ARP):
4962 case htons(ETH_P_IP):
4963 case htons(ETH_P_IPV6):
4964 case htons(ETH_P_8021Q):
4965 case htons(ETH_P_8021AD):
4966 return true;
4967 default:
4968 return false;
4969 }
4970 }
4971
4972 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4973 int *ret, struct net_device *orig_dev)
4974 {
4975 if (nf_hook_ingress_active(skb)) {
4976 int ingress_retval;
4977
4978 if (*pt_prev) {
4979 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4980 *pt_prev = NULL;
4981 }
4982
4983 rcu_read_lock();
4984 ingress_retval = nf_hook_ingress(skb);
4985 rcu_read_unlock();
4986 return ingress_retval;
4987 }
4988 return 0;
4989 }
4990
4991 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4992 struct packet_type **ppt_prev)
4993 {
4994 struct packet_type *ptype, *pt_prev;
4995 rx_handler_func_t *rx_handler;
4996 struct net_device *orig_dev;
4997 bool deliver_exact = false;
4998 int ret = NET_RX_DROP;
4999 __be16 type;
5000
5001 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5002
5003 trace_netif_receive_skb(skb);
5004
5005 orig_dev = skb->dev;
5006
5007 skb_reset_network_header(skb);
5008 if (!skb_transport_header_was_set(skb))
5009 skb_reset_transport_header(skb);
5010 skb_reset_mac_len(skb);
5011
5012 pt_prev = NULL;
5013
5014 another_round:
5015 skb->skb_iif = skb->dev->ifindex;
5016
5017 __this_cpu_inc(softnet_data.processed);
5018
5019 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5020 int ret2;
5021
5022 preempt_disable();
5023 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5024 preempt_enable();
5025
5026 if (ret2 != XDP_PASS)
5027 return NET_RX_DROP;
5028 skb_reset_mac_len(skb);
5029 }
5030
5031 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5032 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5033 skb = skb_vlan_untag(skb);
5034 if (unlikely(!skb))
5035 goto out;
5036 }
5037
5038 if (skb_skip_tc_classify(skb))
5039 goto skip_classify;
5040
5041 if (pfmemalloc)
5042 goto skip_taps;
5043
5044 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5045 if (pt_prev)
5046 ret = deliver_skb(skb, pt_prev, orig_dev);
5047 pt_prev = ptype;
5048 }
5049
5050 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5051 if (pt_prev)
5052 ret = deliver_skb(skb, pt_prev, orig_dev);
5053 pt_prev = ptype;
5054 }
5055
5056 skip_taps:
5057 #ifdef CONFIG_NET_INGRESS
5058 if (static_branch_unlikely(&ingress_needed_key)) {
5059 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5060 if (!skb)
5061 goto out;
5062
5063 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5064 goto out;
5065 }
5066 #endif
5067 skb_reset_redirect(skb);
5068 skip_classify:
5069 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5070 goto drop;
5071
5072 if (skb_vlan_tag_present(skb)) {
5073 if (pt_prev) {
5074 ret = deliver_skb(skb, pt_prev, orig_dev);
5075 pt_prev = NULL;
5076 }
5077 if (vlan_do_receive(&skb))
5078 goto another_round;
5079 else if (unlikely(!skb))
5080 goto out;
5081 }
5082
5083 rx_handler = rcu_dereference(skb->dev->rx_handler);
5084 if (rx_handler) {
5085 if (pt_prev) {
5086 ret = deliver_skb(skb, pt_prev, orig_dev);
5087 pt_prev = NULL;
5088 }
5089 switch (rx_handler(&skb)) {
5090 case RX_HANDLER_CONSUMED:
5091 ret = NET_RX_SUCCESS;
5092 goto out;
5093 case RX_HANDLER_ANOTHER:
5094 goto another_round;
5095 case RX_HANDLER_EXACT:
5096 deliver_exact = true;
5097 case RX_HANDLER_PASS:
5098 break;
5099 default:
5100 BUG();
5101 }
5102 }
5103
5104 if (unlikely(skb_vlan_tag_present(skb))) {
5105 check_vlan_id:
5106 if (skb_vlan_tag_get_id(skb)) {
5107 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5108 * find vlan device.
5109 */
5110 skb->pkt_type = PACKET_OTHERHOST;
5111 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5112 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5113 /* Outer header is 802.1P with vlan 0, inner header is
5114 * 802.1Q or 802.1AD and vlan_do_receive() above could
5115 * not find vlan dev for vlan id 0.
5116 */
5117 __vlan_hwaccel_clear_tag(skb);
5118 skb = skb_vlan_untag(skb);
5119 if (unlikely(!skb))
5120 goto out;
5121 if (vlan_do_receive(&skb))
5122 /* After stripping off 802.1P header with vlan 0
5123 * vlan dev is found for inner header.
5124 */
5125 goto another_round;
5126 else if (unlikely(!skb))
5127 goto out;
5128 else
5129 /* We have stripped outer 802.1P vlan 0 header.
5130 * But could not find vlan dev.
5131 * check again for vlan id to set OTHERHOST.
5132 */
5133 goto check_vlan_id;
5134 }
5135 /* Note: we might in the future use prio bits
5136 * and set skb->priority like in vlan_do_receive()
5137 * For the time being, just ignore Priority Code Point
5138 */
5139 __vlan_hwaccel_clear_tag(skb);
5140 }
5141
5142 type = skb->protocol;
5143
5144 /* deliver only exact match when indicated */
5145 if (likely(!deliver_exact)) {
5146 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5147 &ptype_base[ntohs(type) &
5148 PTYPE_HASH_MASK]);
5149 }
5150
5151 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5152 &orig_dev->ptype_specific);
5153
5154 if (unlikely(skb->dev != orig_dev)) {
5155 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5156 &skb->dev->ptype_specific);
5157 }
5158
5159 if (pt_prev) {
5160 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5161 goto drop;
5162 *ppt_prev = pt_prev;
5163 } else {
5164 drop:
5165 if (!deliver_exact)
5166 atomic_long_inc(&skb->dev->rx_dropped);
5167 else
5168 atomic_long_inc(&skb->dev->rx_nohandler);
5169 kfree_skb(skb);
5170 /* Jamal, now you will not able to escape explaining
5171 * me how you were going to use this. :-)
5172 */
5173 ret = NET_RX_DROP;
5174 }
5175
5176 out:
5177 return ret;
5178 }
5179
5180 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5181 {
5182 struct net_device *orig_dev = skb->dev;
5183 struct packet_type *pt_prev = NULL;
5184 int ret;
5185
5186 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5187 if (pt_prev)
5188 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5189 skb->dev, pt_prev, orig_dev);
5190 return ret;
5191 }
5192
5193 /**
5194 * netif_receive_skb_core - special purpose version of netif_receive_skb
5195 * @skb: buffer to process
5196 *
5197 * More direct receive version of netif_receive_skb(). It should
5198 * only be used by callers that have a need to skip RPS and Generic XDP.
5199 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5200 *
5201 * This function may only be called from softirq context and interrupts
5202 * should be enabled.
5203 *
5204 * Return values (usually ignored):
5205 * NET_RX_SUCCESS: no congestion
5206 * NET_RX_DROP: packet was dropped
5207 */
5208 int netif_receive_skb_core(struct sk_buff *skb)
5209 {
5210 int ret;
5211
5212 rcu_read_lock();
5213 ret = __netif_receive_skb_one_core(skb, false);
5214 rcu_read_unlock();
5215
5216 return ret;
5217 }
5218 EXPORT_SYMBOL(netif_receive_skb_core);
5219
5220 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5221 struct packet_type *pt_prev,
5222 struct net_device *orig_dev)
5223 {
5224 struct sk_buff *skb, *next;
5225
5226 if (!pt_prev)
5227 return;
5228 if (list_empty(head))
5229 return;
5230 if (pt_prev->list_func != NULL)
5231 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5232 ip_list_rcv, head, pt_prev, orig_dev);
5233 else
5234 list_for_each_entry_safe(skb, next, head, list) {
5235 skb_list_del_init(skb);
5236 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5237 }
5238 }
5239
5240 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5241 {
5242 /* Fast-path assumptions:
5243 * - There is no RX handler.
5244 * - Only one packet_type matches.
5245 * If either of these fails, we will end up doing some per-packet
5246 * processing in-line, then handling the 'last ptype' for the whole
5247 * sublist. This can't cause out-of-order delivery to any single ptype,
5248 * because the 'last ptype' must be constant across the sublist, and all
5249 * other ptypes are handled per-packet.
5250 */
5251 /* Current (common) ptype of sublist */
5252 struct packet_type *pt_curr = NULL;
5253 /* Current (common) orig_dev of sublist */
5254 struct net_device *od_curr = NULL;
5255 struct list_head sublist;
5256 struct sk_buff *skb, *next;
5257
5258 INIT_LIST_HEAD(&sublist);
5259 list_for_each_entry_safe(skb, next, head, list) {
5260 struct net_device *orig_dev = skb->dev;
5261 struct packet_type *pt_prev = NULL;
5262
5263 skb_list_del_init(skb);
5264 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5265 if (!pt_prev)
5266 continue;
5267 if (pt_curr != pt_prev || od_curr != orig_dev) {
5268 /* dispatch old sublist */
5269 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5270 /* start new sublist */
5271 INIT_LIST_HEAD(&sublist);
5272 pt_curr = pt_prev;
5273 od_curr = orig_dev;
5274 }
5275 list_add_tail(&skb->list, &sublist);
5276 }
5277
5278 /* dispatch final sublist */
5279 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5280 }
5281
5282 static int __netif_receive_skb(struct sk_buff *skb)
5283 {
5284 int ret;
5285
5286 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5287 unsigned int noreclaim_flag;
5288
5289 /*
5290 * PFMEMALLOC skbs are special, they should
5291 * - be delivered to SOCK_MEMALLOC sockets only
5292 * - stay away from userspace
5293 * - have bounded memory usage
5294 *
5295 * Use PF_MEMALLOC as this saves us from propagating the allocation
5296 * context down to all allocation sites.
5297 */
5298 noreclaim_flag = memalloc_noreclaim_save();
5299 ret = __netif_receive_skb_one_core(skb, true);
5300 memalloc_noreclaim_restore(noreclaim_flag);
5301 } else
5302 ret = __netif_receive_skb_one_core(skb, false);
5303
5304 return ret;
5305 }
5306
5307 static void __netif_receive_skb_list(struct list_head *head)
5308 {
5309 unsigned long noreclaim_flag = 0;
5310 struct sk_buff *skb, *next;
5311 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5312
5313 list_for_each_entry_safe(skb, next, head, list) {
5314 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5315 struct list_head sublist;
5316
5317 /* Handle the previous sublist */
5318 list_cut_before(&sublist, head, &skb->list);
5319 if (!list_empty(&sublist))
5320 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5321 pfmemalloc = !pfmemalloc;
5322 /* See comments in __netif_receive_skb */
5323 if (pfmemalloc)
5324 noreclaim_flag = memalloc_noreclaim_save();
5325 else
5326 memalloc_noreclaim_restore(noreclaim_flag);
5327 }
5328 }
5329 /* Handle the remaining sublist */
5330 if (!list_empty(head))
5331 __netif_receive_skb_list_core(head, pfmemalloc);
5332 /* Restore pflags */
5333 if (pfmemalloc)
5334 memalloc_noreclaim_restore(noreclaim_flag);
5335 }
5336
5337 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5338 {
5339 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5340 struct bpf_prog *new = xdp->prog;
5341 int ret = 0;
5342
5343 switch (xdp->command) {
5344 case XDP_SETUP_PROG:
5345 rcu_assign_pointer(dev->xdp_prog, new);
5346 if (old)
5347 bpf_prog_put(old);
5348
5349 if (old && !new) {
5350 static_branch_dec(&generic_xdp_needed_key);
5351 } else if (new && !old) {
5352 static_branch_inc(&generic_xdp_needed_key);
5353 dev_disable_lro(dev);
5354 dev_disable_gro_hw(dev);
5355 }
5356 break;
5357
5358 case XDP_QUERY_PROG:
5359 xdp->prog_id = old ? old->aux->id : 0;
5360 break;
5361
5362 default:
5363 ret = -EINVAL;
5364 break;
5365 }
5366
5367 return ret;
5368 }
5369
5370 static int netif_receive_skb_internal(struct sk_buff *skb)
5371 {
5372 int ret;
5373
5374 net_timestamp_check(netdev_tstamp_prequeue, skb);
5375
5376 if (skb_defer_rx_timestamp(skb))
5377 return NET_RX_SUCCESS;
5378
5379 rcu_read_lock();
5380 #ifdef CONFIG_RPS
5381 if (static_branch_unlikely(&rps_needed)) {
5382 struct rps_dev_flow voidflow, *rflow = &voidflow;
5383 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5384
5385 if (cpu >= 0) {
5386 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5387 rcu_read_unlock();
5388 return ret;
5389 }
5390 }
5391 #endif
5392 ret = __netif_receive_skb(skb);
5393 rcu_read_unlock();
5394 return ret;
5395 }
5396
5397 static void netif_receive_skb_list_internal(struct list_head *head)
5398 {
5399 struct sk_buff *skb, *next;
5400 struct list_head sublist;
5401
5402 INIT_LIST_HEAD(&sublist);
5403 list_for_each_entry_safe(skb, next, head, list) {
5404 net_timestamp_check(netdev_tstamp_prequeue, skb);
5405 skb_list_del_init(skb);
5406 if (!skb_defer_rx_timestamp(skb))
5407 list_add_tail(&skb->list, &sublist);
5408 }
5409 list_splice_init(&sublist, head);
5410
5411 rcu_read_lock();
5412 #ifdef CONFIG_RPS
5413 if (static_branch_unlikely(&rps_needed)) {
5414 list_for_each_entry_safe(skb, next, head, list) {
5415 struct rps_dev_flow voidflow, *rflow = &voidflow;
5416 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5417
5418 if (cpu >= 0) {
5419 /* Will be handled, remove from list */
5420 skb_list_del_init(skb);
5421 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5422 }
5423 }
5424 }
5425 #endif
5426 __netif_receive_skb_list(head);
5427 rcu_read_unlock();
5428 }
5429
5430 /**
5431 * netif_receive_skb - process receive buffer from network
5432 * @skb: buffer to process
5433 *
5434 * netif_receive_skb() is the main receive data processing function.
5435 * It always succeeds. The buffer may be dropped during processing
5436 * for congestion control or by the protocol layers.
5437 *
5438 * This function may only be called from softirq context and interrupts
5439 * should be enabled.
5440 *
5441 * Return values (usually ignored):
5442 * NET_RX_SUCCESS: no congestion
5443 * NET_RX_DROP: packet was dropped
5444 */
5445 int netif_receive_skb(struct sk_buff *skb)
5446 {
5447 int ret;
5448
5449 trace_netif_receive_skb_entry(skb);
5450
5451 ret = netif_receive_skb_internal(skb);
5452 trace_netif_receive_skb_exit(ret);
5453
5454 return ret;
5455 }
5456 EXPORT_SYMBOL(netif_receive_skb);
5457
5458 /**
5459 * netif_receive_skb_list - process many receive buffers from network
5460 * @head: list of skbs to process.
5461 *
5462 * Since return value of netif_receive_skb() is normally ignored, and
5463 * wouldn't be meaningful for a list, this function returns void.
5464 *
5465 * This function may only be called from softirq context and interrupts
5466 * should be enabled.
5467 */
5468 void netif_receive_skb_list(struct list_head *head)
5469 {
5470 struct sk_buff *skb;
5471
5472 if (list_empty(head))
5473 return;
5474 if (trace_netif_receive_skb_list_entry_enabled()) {
5475 list_for_each_entry(skb, head, list)
5476 trace_netif_receive_skb_list_entry(skb);
5477 }
5478 netif_receive_skb_list_internal(head);
5479 trace_netif_receive_skb_list_exit(0);
5480 }
5481 EXPORT_SYMBOL(netif_receive_skb_list);
5482
5483 DEFINE_PER_CPU(struct work_struct, flush_works);
5484
5485 /* Network device is going away, flush any packets still pending */
5486 static void flush_backlog(struct work_struct *work)
5487 {
5488 struct sk_buff *skb, *tmp;
5489 struct softnet_data *sd;
5490
5491 local_bh_disable();
5492 sd = this_cpu_ptr(&softnet_data);
5493
5494 local_irq_disable();
5495 rps_lock(sd);
5496 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5497 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5498 __skb_unlink(skb, &sd->input_pkt_queue);
5499 kfree_skb(skb);
5500 input_queue_head_incr(sd);
5501 }
5502 }
5503 rps_unlock(sd);
5504 local_irq_enable();
5505
5506 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5507 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5508 __skb_unlink(skb, &sd->process_queue);
5509 kfree_skb(skb);
5510 input_queue_head_incr(sd);
5511 }
5512 }
5513 local_bh_enable();
5514 }
5515
5516 static void flush_all_backlogs(void)
5517 {
5518 unsigned int cpu;
5519
5520 get_online_cpus();
5521
5522 for_each_online_cpu(cpu)
5523 queue_work_on(cpu, system_highpri_wq,
5524 per_cpu_ptr(&flush_works, cpu));
5525
5526 for_each_online_cpu(cpu)
5527 flush_work(per_cpu_ptr(&flush_works, cpu));
5528
5529 put_online_cpus();
5530 }
5531
5532 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5533 static void gro_normal_list(struct napi_struct *napi)
5534 {
5535 if (!napi->rx_count)
5536 return;
5537 netif_receive_skb_list_internal(&napi->rx_list);
5538 INIT_LIST_HEAD(&napi->rx_list);
5539 napi->rx_count = 0;
5540 }
5541
5542 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5543 * pass the whole batch up to the stack.
5544 */
5545 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5546 {
5547 list_add_tail(&skb->list, &napi->rx_list);
5548 if (++napi->rx_count >= gro_normal_batch)
5549 gro_normal_list(napi);
5550 }
5551
5552 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5553 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5554 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5555 {
5556 struct packet_offload *ptype;
5557 __be16 type = skb->protocol;
5558 struct list_head *head = &offload_base;
5559 int err = -ENOENT;
5560
5561 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5562
5563 if (NAPI_GRO_CB(skb)->count == 1) {
5564 skb_shinfo(skb)->gso_size = 0;
5565 goto out;
5566 }
5567
5568 rcu_read_lock();
5569 list_for_each_entry_rcu(ptype, head, list) {
5570 if (ptype->type != type || !ptype->callbacks.gro_complete)
5571 continue;
5572
5573 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5574 ipv6_gro_complete, inet_gro_complete,
5575 skb, 0);
5576 break;
5577 }
5578 rcu_read_unlock();
5579
5580 if (err) {
5581 WARN_ON(&ptype->list == head);
5582 kfree_skb(skb);
5583 return NET_RX_SUCCESS;
5584 }
5585
5586 out:
5587 gro_normal_one(napi, skb);
5588 return NET_RX_SUCCESS;
5589 }
5590
5591 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5592 bool flush_old)
5593 {
5594 struct list_head *head = &napi->gro_hash[index].list;
5595 struct sk_buff *skb, *p;
5596
5597 list_for_each_entry_safe_reverse(skb, p, head, list) {
5598 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5599 return;
5600 skb_list_del_init(skb);
5601 napi_gro_complete(napi, skb);
5602 napi->gro_hash[index].count--;
5603 }
5604
5605 if (!napi->gro_hash[index].count)
5606 __clear_bit(index, &napi->gro_bitmask);
5607 }
5608
5609 /* napi->gro_hash[].list contains packets ordered by age.
5610 * youngest packets at the head of it.
5611 * Complete skbs in reverse order to reduce latencies.
5612 */
5613 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5614 {
5615 unsigned long bitmask = napi->gro_bitmask;
5616 unsigned int i, base = ~0U;
5617
5618 while ((i = ffs(bitmask)) != 0) {
5619 bitmask >>= i;
5620 base += i;
5621 __napi_gro_flush_chain(napi, base, flush_old);
5622 }
5623 }
5624 EXPORT_SYMBOL(napi_gro_flush);
5625
5626 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5627 struct sk_buff *skb)
5628 {
5629 unsigned int maclen = skb->dev->hard_header_len;
5630 u32 hash = skb_get_hash_raw(skb);
5631 struct list_head *head;
5632 struct sk_buff *p;
5633
5634 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5635 list_for_each_entry(p, head, list) {
5636 unsigned long diffs;
5637
5638 NAPI_GRO_CB(p)->flush = 0;
5639
5640 if (hash != skb_get_hash_raw(p)) {
5641 NAPI_GRO_CB(p)->same_flow = 0;
5642 continue;
5643 }
5644
5645 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5646 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5647 if (skb_vlan_tag_present(p))
5648 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5649 diffs |= skb_metadata_dst_cmp(p, skb);
5650 diffs |= skb_metadata_differs(p, skb);
5651 if (maclen == ETH_HLEN)
5652 diffs |= compare_ether_header(skb_mac_header(p),
5653 skb_mac_header(skb));
5654 else if (!diffs)
5655 diffs = memcmp(skb_mac_header(p),
5656 skb_mac_header(skb),
5657 maclen);
5658 NAPI_GRO_CB(p)->same_flow = !diffs;
5659 }
5660
5661 return head;
5662 }
5663
5664 static void skb_gro_reset_offset(struct sk_buff *skb)
5665 {
5666 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5667 const skb_frag_t *frag0 = &pinfo->frags[0];
5668
5669 NAPI_GRO_CB(skb)->data_offset = 0;
5670 NAPI_GRO_CB(skb)->frag0 = NULL;
5671 NAPI_GRO_CB(skb)->frag0_len = 0;
5672
5673 if (!skb_headlen(skb) && pinfo->nr_frags &&
5674 !PageHighMem(skb_frag_page(frag0))) {
5675 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5676 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5677 skb_frag_size(frag0),
5678 skb->end - skb->tail);
5679 }
5680 }
5681
5682 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5683 {
5684 struct skb_shared_info *pinfo = skb_shinfo(skb);
5685
5686 BUG_ON(skb->end - skb->tail < grow);
5687
5688 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5689
5690 skb->data_len -= grow;
5691 skb->tail += grow;
5692
5693 skb_frag_off_add(&pinfo->frags[0], grow);
5694 skb_frag_size_sub(&pinfo->frags[0], grow);
5695
5696 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5697 skb_frag_unref(skb, 0);
5698 memmove(pinfo->frags, pinfo->frags + 1,
5699 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5700 }
5701 }
5702
5703 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5704 {
5705 struct sk_buff *oldest;
5706
5707 oldest = list_last_entry(head, struct sk_buff, list);
5708
5709 /* We are called with head length >= MAX_GRO_SKBS, so this is
5710 * impossible.
5711 */
5712 if (WARN_ON_ONCE(!oldest))
5713 return;
5714
5715 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5716 * SKB to the chain.
5717 */
5718 skb_list_del_init(oldest);
5719 napi_gro_complete(napi, oldest);
5720 }
5721
5722 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5723 struct sk_buff *));
5724 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5725 struct sk_buff *));
5726 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5727 {
5728 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5729 struct list_head *head = &offload_base;
5730 struct packet_offload *ptype;
5731 __be16 type = skb->protocol;
5732 struct list_head *gro_head;
5733 struct sk_buff *pp = NULL;
5734 enum gro_result ret;
5735 int same_flow;
5736 int grow;
5737
5738 if (netif_elide_gro(skb->dev))
5739 goto normal;
5740
5741 gro_head = gro_list_prepare(napi, skb);
5742
5743 rcu_read_lock();
5744 list_for_each_entry_rcu(ptype, head, list) {
5745 if (ptype->type != type || !ptype->callbacks.gro_receive)
5746 continue;
5747
5748 skb_set_network_header(skb, skb_gro_offset(skb));
5749 skb_reset_mac_len(skb);
5750 NAPI_GRO_CB(skb)->same_flow = 0;
5751 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5752 NAPI_GRO_CB(skb)->free = 0;
5753 NAPI_GRO_CB(skb)->encap_mark = 0;
5754 NAPI_GRO_CB(skb)->recursion_counter = 0;
5755 NAPI_GRO_CB(skb)->is_fou = 0;
5756 NAPI_GRO_CB(skb)->is_atomic = 1;
5757 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5758
5759 /* Setup for GRO checksum validation */
5760 switch (skb->ip_summed) {
5761 case CHECKSUM_COMPLETE:
5762 NAPI_GRO_CB(skb)->csum = skb->csum;
5763 NAPI_GRO_CB(skb)->csum_valid = 1;
5764 NAPI_GRO_CB(skb)->csum_cnt = 0;
5765 break;
5766 case CHECKSUM_UNNECESSARY:
5767 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5768 NAPI_GRO_CB(skb)->csum_valid = 0;
5769 break;
5770 default:
5771 NAPI_GRO_CB(skb)->csum_cnt = 0;
5772 NAPI_GRO_CB(skb)->csum_valid = 0;
5773 }
5774
5775 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5776 ipv6_gro_receive, inet_gro_receive,
5777 gro_head, skb);
5778 break;
5779 }
5780 rcu_read_unlock();
5781
5782 if (&ptype->list == head)
5783 goto normal;
5784
5785 if (PTR_ERR(pp) == -EINPROGRESS) {
5786 ret = GRO_CONSUMED;
5787 goto ok;
5788 }
5789
5790 same_flow = NAPI_GRO_CB(skb)->same_flow;
5791 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5792
5793 if (pp) {
5794 skb_list_del_init(pp);
5795 napi_gro_complete(napi, pp);
5796 napi->gro_hash[hash].count--;
5797 }
5798
5799 if (same_flow)
5800 goto ok;
5801
5802 if (NAPI_GRO_CB(skb)->flush)
5803 goto normal;
5804
5805 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5806 gro_flush_oldest(napi, gro_head);
5807 } else {
5808 napi->gro_hash[hash].count++;
5809 }
5810 NAPI_GRO_CB(skb)->count = 1;
5811 NAPI_GRO_CB(skb)->age = jiffies;
5812 NAPI_GRO_CB(skb)->last = skb;
5813 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5814 list_add(&skb->list, gro_head);
5815 ret = GRO_HELD;
5816
5817 pull:
5818 grow = skb_gro_offset(skb) - skb_headlen(skb);
5819 if (grow > 0)
5820 gro_pull_from_frag0(skb, grow);
5821 ok:
5822 if (napi->gro_hash[hash].count) {
5823 if (!test_bit(hash, &napi->gro_bitmask))
5824 __set_bit(hash, &napi->gro_bitmask);
5825 } else if (test_bit(hash, &napi->gro_bitmask)) {
5826 __clear_bit(hash, &napi->gro_bitmask);
5827 }
5828
5829 return ret;
5830
5831 normal:
5832 ret = GRO_NORMAL;
5833 goto pull;
5834 }
5835
5836 struct packet_offload *gro_find_receive_by_type(__be16 type)
5837 {
5838 struct list_head *offload_head = &offload_base;
5839 struct packet_offload *ptype;
5840
5841 list_for_each_entry_rcu(ptype, offload_head, list) {
5842 if (ptype->type != type || !ptype->callbacks.gro_receive)
5843 continue;
5844 return ptype;
5845 }
5846 return NULL;
5847 }
5848 EXPORT_SYMBOL(gro_find_receive_by_type);
5849
5850 struct packet_offload *gro_find_complete_by_type(__be16 type)
5851 {
5852 struct list_head *offload_head = &offload_base;
5853 struct packet_offload *ptype;
5854
5855 list_for_each_entry_rcu(ptype, offload_head, list) {
5856 if (ptype->type != type || !ptype->callbacks.gro_complete)
5857 continue;
5858 return ptype;
5859 }
5860 return NULL;
5861 }
5862 EXPORT_SYMBOL(gro_find_complete_by_type);
5863
5864 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5865 {
5866 skb_dst_drop(skb);
5867 skb_ext_put(skb);
5868 kmem_cache_free(skbuff_head_cache, skb);
5869 }
5870
5871 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5872 struct sk_buff *skb,
5873 gro_result_t ret)
5874 {
5875 switch (ret) {
5876 case GRO_NORMAL:
5877 gro_normal_one(napi, skb);
5878 break;
5879
5880 case GRO_DROP:
5881 kfree_skb(skb);
5882 break;
5883
5884 case GRO_MERGED_FREE:
5885 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5886 napi_skb_free_stolen_head(skb);
5887 else
5888 __kfree_skb(skb);
5889 break;
5890
5891 case GRO_HELD:
5892 case GRO_MERGED:
5893 case GRO_CONSUMED:
5894 break;
5895 }
5896
5897 return ret;
5898 }
5899
5900 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5901 {
5902 gro_result_t ret;
5903
5904 skb_mark_napi_id(skb, napi);
5905 trace_napi_gro_receive_entry(skb);
5906
5907 skb_gro_reset_offset(skb);
5908
5909 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
5910 trace_napi_gro_receive_exit(ret);
5911
5912 return ret;
5913 }
5914 EXPORT_SYMBOL(napi_gro_receive);
5915
5916 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5917 {
5918 if (unlikely(skb->pfmemalloc)) {
5919 consume_skb(skb);
5920 return;
5921 }
5922 __skb_pull(skb, skb_headlen(skb));
5923 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5924 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5925 __vlan_hwaccel_clear_tag(skb);
5926 skb->dev = napi->dev;
5927 skb->skb_iif = 0;
5928
5929 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5930 skb->pkt_type = PACKET_HOST;
5931
5932 skb->encapsulation = 0;
5933 skb_shinfo(skb)->gso_type = 0;
5934 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5935 skb_ext_reset(skb);
5936
5937 napi->skb = skb;
5938 }
5939
5940 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5941 {
5942 struct sk_buff *skb = napi->skb;
5943
5944 if (!skb) {
5945 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5946 if (skb) {
5947 napi->skb = skb;
5948 skb_mark_napi_id(skb, napi);
5949 }
5950 }
5951 return skb;
5952 }
5953 EXPORT_SYMBOL(napi_get_frags);
5954
5955 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5956 struct sk_buff *skb,
5957 gro_result_t ret)
5958 {
5959 switch (ret) {
5960 case GRO_NORMAL:
5961 case GRO_HELD:
5962 __skb_push(skb, ETH_HLEN);
5963 skb->protocol = eth_type_trans(skb, skb->dev);
5964 if (ret == GRO_NORMAL)
5965 gro_normal_one(napi, skb);
5966 break;
5967
5968 case GRO_DROP:
5969 napi_reuse_skb(napi, skb);
5970 break;
5971
5972 case GRO_MERGED_FREE:
5973 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5974 napi_skb_free_stolen_head(skb);
5975 else
5976 napi_reuse_skb(napi, skb);
5977 break;
5978
5979 case GRO_MERGED:
5980 case GRO_CONSUMED:
5981 break;
5982 }
5983
5984 return ret;
5985 }
5986
5987 /* Upper GRO stack assumes network header starts at gro_offset=0
5988 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5989 * We copy ethernet header into skb->data to have a common layout.
5990 */
5991 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5992 {
5993 struct sk_buff *skb = napi->skb;
5994 const struct ethhdr *eth;
5995 unsigned int hlen = sizeof(*eth);
5996
5997 napi->skb = NULL;
5998
5999 skb_reset_mac_header(skb);
6000 skb_gro_reset_offset(skb);
6001
6002 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6003 eth = skb_gro_header_slow(skb, hlen, 0);
6004 if (unlikely(!eth)) {
6005 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6006 __func__, napi->dev->name);
6007 napi_reuse_skb(napi, skb);
6008 return NULL;
6009 }
6010 } else {
6011 eth = (const struct ethhdr *)skb->data;
6012 gro_pull_from_frag0(skb, hlen);
6013 NAPI_GRO_CB(skb)->frag0 += hlen;
6014 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6015 }
6016 __skb_pull(skb, hlen);
6017
6018 /*
6019 * This works because the only protocols we care about don't require
6020 * special handling.
6021 * We'll fix it up properly in napi_frags_finish()
6022 */
6023 skb->protocol = eth->h_proto;
6024
6025 return skb;
6026 }
6027
6028 gro_result_t napi_gro_frags(struct napi_struct *napi)
6029 {
6030 gro_result_t ret;
6031 struct sk_buff *skb = napi_frags_skb(napi);
6032
6033 if (!skb)
6034 return GRO_DROP;
6035
6036 trace_napi_gro_frags_entry(skb);
6037
6038 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6039 trace_napi_gro_frags_exit(ret);
6040
6041 return ret;
6042 }
6043 EXPORT_SYMBOL(napi_gro_frags);
6044
6045 /* Compute the checksum from gro_offset and return the folded value
6046 * after adding in any pseudo checksum.
6047 */
6048 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6049 {
6050 __wsum wsum;
6051 __sum16 sum;
6052
6053 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6054
6055 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6056 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6057 /* See comments in __skb_checksum_complete(). */
6058 if (likely(!sum)) {
6059 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6060 !skb->csum_complete_sw)
6061 netdev_rx_csum_fault(skb->dev, skb);
6062 }
6063
6064 NAPI_GRO_CB(skb)->csum = wsum;
6065 NAPI_GRO_CB(skb)->csum_valid = 1;
6066
6067 return sum;
6068 }
6069 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6070
6071 static void net_rps_send_ipi(struct softnet_data *remsd)
6072 {
6073 #ifdef CONFIG_RPS
6074 while (remsd) {
6075 struct softnet_data *next = remsd->rps_ipi_next;
6076
6077 if (cpu_online(remsd->cpu))
6078 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6079 remsd = next;
6080 }
6081 #endif
6082 }
6083
6084 /*
6085 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6086 * Note: called with local irq disabled, but exits with local irq enabled.
6087 */
6088 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6089 {
6090 #ifdef CONFIG_RPS
6091 struct softnet_data *remsd = sd->rps_ipi_list;
6092
6093 if (remsd) {
6094 sd->rps_ipi_list = NULL;
6095
6096 local_irq_enable();
6097
6098 /* Send pending IPI's to kick RPS processing on remote cpus. */
6099 net_rps_send_ipi(remsd);
6100 } else
6101 #endif
6102 local_irq_enable();
6103 }
6104
6105 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6106 {
6107 #ifdef CONFIG_RPS
6108 return sd->rps_ipi_list != NULL;
6109 #else
6110 return false;
6111 #endif
6112 }
6113
6114 static int process_backlog(struct napi_struct *napi, int quota)
6115 {
6116 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6117 bool again = true;
6118 int work = 0;
6119
6120 /* Check if we have pending ipi, its better to send them now,
6121 * not waiting net_rx_action() end.
6122 */
6123 if (sd_has_rps_ipi_waiting(sd)) {
6124 local_irq_disable();
6125 net_rps_action_and_irq_enable(sd);
6126 }
6127
6128 napi->weight = dev_rx_weight;
6129 while (again) {
6130 struct sk_buff *skb;
6131
6132 while ((skb = __skb_dequeue(&sd->process_queue))) {
6133 rcu_read_lock();
6134 __netif_receive_skb(skb);
6135 rcu_read_unlock();
6136 input_queue_head_incr(sd);
6137 if (++work >= quota)
6138 return work;
6139
6140 }
6141
6142 local_irq_disable();
6143 rps_lock(sd);
6144 if (skb_queue_empty(&sd->input_pkt_queue)) {
6145 /*
6146 * Inline a custom version of __napi_complete().
6147 * only current cpu owns and manipulates this napi,
6148 * and NAPI_STATE_SCHED is the only possible flag set
6149 * on backlog.
6150 * We can use a plain write instead of clear_bit(),
6151 * and we dont need an smp_mb() memory barrier.
6152 */
6153 napi->state = 0;
6154 again = false;
6155 } else {
6156 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6157 &sd->process_queue);
6158 }
6159 rps_unlock(sd);
6160 local_irq_enable();
6161 }
6162
6163 return work;
6164 }
6165
6166 /**
6167 * __napi_schedule - schedule for receive
6168 * @n: entry to schedule
6169 *
6170 * The entry's receive function will be scheduled to run.
6171 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6172 */
6173 void __napi_schedule(struct napi_struct *n)
6174 {
6175 unsigned long flags;
6176
6177 local_irq_save(flags);
6178 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6179 local_irq_restore(flags);
6180 }
6181 EXPORT_SYMBOL(__napi_schedule);
6182
6183 /**
6184 * napi_schedule_prep - check if napi can be scheduled
6185 * @n: napi context
6186 *
6187 * Test if NAPI routine is already running, and if not mark
6188 * it as running. This is used as a condition variable
6189 * insure only one NAPI poll instance runs. We also make
6190 * sure there is no pending NAPI disable.
6191 */
6192 bool napi_schedule_prep(struct napi_struct *n)
6193 {
6194 unsigned long val, new;
6195
6196 do {
6197 val = READ_ONCE(n->state);
6198 if (unlikely(val & NAPIF_STATE_DISABLE))
6199 return false;
6200 new = val | NAPIF_STATE_SCHED;
6201
6202 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6203 * This was suggested by Alexander Duyck, as compiler
6204 * emits better code than :
6205 * if (val & NAPIF_STATE_SCHED)
6206 * new |= NAPIF_STATE_MISSED;
6207 */
6208 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6209 NAPIF_STATE_MISSED;
6210 } while (cmpxchg(&n->state, val, new) != val);
6211
6212 return !(val & NAPIF_STATE_SCHED);
6213 }
6214 EXPORT_SYMBOL(napi_schedule_prep);
6215
6216 /**
6217 * __napi_schedule_irqoff - schedule for receive
6218 * @n: entry to schedule
6219 *
6220 * Variant of __napi_schedule() assuming hard irqs are masked
6221 */
6222 void __napi_schedule_irqoff(struct napi_struct *n)
6223 {
6224 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6225 }
6226 EXPORT_SYMBOL(__napi_schedule_irqoff);
6227
6228 bool napi_complete_done(struct napi_struct *n, int work_done)
6229 {
6230 unsigned long flags, val, new;
6231
6232 /*
6233 * 1) Don't let napi dequeue from the cpu poll list
6234 * just in case its running on a different cpu.
6235 * 2) If we are busy polling, do nothing here, we have
6236 * the guarantee we will be called later.
6237 */
6238 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6239 NAPIF_STATE_IN_BUSY_POLL)))
6240 return false;
6241
6242 if (n->gro_bitmask) {
6243 unsigned long timeout = 0;
6244
6245 if (work_done)
6246 timeout = n->dev->gro_flush_timeout;
6247
6248 /* When the NAPI instance uses a timeout and keeps postponing
6249 * it, we need to bound somehow the time packets are kept in
6250 * the GRO layer
6251 */
6252 napi_gro_flush(n, !!timeout);
6253 if (timeout)
6254 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6255 HRTIMER_MODE_REL_PINNED);
6256 }
6257
6258 gro_normal_list(n);
6259
6260 if (unlikely(!list_empty(&n->poll_list))) {
6261 /* If n->poll_list is not empty, we need to mask irqs */
6262 local_irq_save(flags);
6263 list_del_init(&n->poll_list);
6264 local_irq_restore(flags);
6265 }
6266
6267 do {
6268 val = READ_ONCE(n->state);
6269
6270 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6271
6272 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6273
6274 /* If STATE_MISSED was set, leave STATE_SCHED set,
6275 * because we will call napi->poll() one more time.
6276 * This C code was suggested by Alexander Duyck to help gcc.
6277 */
6278 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6279 NAPIF_STATE_SCHED;
6280 } while (cmpxchg(&n->state, val, new) != val);
6281
6282 if (unlikely(val & NAPIF_STATE_MISSED)) {
6283 __napi_schedule(n);
6284 return false;
6285 }
6286
6287 return true;
6288 }
6289 EXPORT_SYMBOL(napi_complete_done);
6290
6291 /* must be called under rcu_read_lock(), as we dont take a reference */
6292 static struct napi_struct *napi_by_id(unsigned int napi_id)
6293 {
6294 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6295 struct napi_struct *napi;
6296
6297 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6298 if (napi->napi_id == napi_id)
6299 return napi;
6300
6301 return NULL;
6302 }
6303
6304 #if defined(CONFIG_NET_RX_BUSY_POLL)
6305
6306 #define BUSY_POLL_BUDGET 8
6307
6308 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6309 {
6310 int rc;
6311
6312 /* Busy polling means there is a high chance device driver hard irq
6313 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6314 * set in napi_schedule_prep().
6315 * Since we are about to call napi->poll() once more, we can safely
6316 * clear NAPI_STATE_MISSED.
6317 *
6318 * Note: x86 could use a single "lock and ..." instruction
6319 * to perform these two clear_bit()
6320 */
6321 clear_bit(NAPI_STATE_MISSED, &napi->state);
6322 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6323
6324 local_bh_disable();
6325
6326 /* All we really want here is to re-enable device interrupts.
6327 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6328 */
6329 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6330 /* We can't gro_normal_list() here, because napi->poll() might have
6331 * rearmed the napi (napi_complete_done()) in which case it could
6332 * already be running on another CPU.
6333 */
6334 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6335 netpoll_poll_unlock(have_poll_lock);
6336 if (rc == BUSY_POLL_BUDGET) {
6337 /* As the whole budget was spent, we still own the napi so can
6338 * safely handle the rx_list.
6339 */
6340 gro_normal_list(napi);
6341 __napi_schedule(napi);
6342 }
6343 local_bh_enable();
6344 }
6345
6346 void napi_busy_loop(unsigned int napi_id,
6347 bool (*loop_end)(void *, unsigned long),
6348 void *loop_end_arg)
6349 {
6350 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6351 int (*napi_poll)(struct napi_struct *napi, int budget);
6352 void *have_poll_lock = NULL;
6353 struct napi_struct *napi;
6354
6355 restart:
6356 napi_poll = NULL;
6357
6358 rcu_read_lock();
6359
6360 napi = napi_by_id(napi_id);
6361 if (!napi)
6362 goto out;
6363
6364 preempt_disable();
6365 for (;;) {
6366 int work = 0;
6367
6368 local_bh_disable();
6369 if (!napi_poll) {
6370 unsigned long val = READ_ONCE(napi->state);
6371
6372 /* If multiple threads are competing for this napi,
6373 * we avoid dirtying napi->state as much as we can.
6374 */
6375 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6376 NAPIF_STATE_IN_BUSY_POLL))
6377 goto count;
6378 if (cmpxchg(&napi->state, val,
6379 val | NAPIF_STATE_IN_BUSY_POLL |
6380 NAPIF_STATE_SCHED) != val)
6381 goto count;
6382 have_poll_lock = netpoll_poll_lock(napi);
6383 napi_poll = napi->poll;
6384 }
6385 work = napi_poll(napi, BUSY_POLL_BUDGET);
6386 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6387 gro_normal_list(napi);
6388 count:
6389 if (work > 0)
6390 __NET_ADD_STATS(dev_net(napi->dev),
6391 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6392 local_bh_enable();
6393
6394 if (!loop_end || loop_end(loop_end_arg, start_time))
6395 break;
6396
6397 if (unlikely(need_resched())) {
6398 if (napi_poll)
6399 busy_poll_stop(napi, have_poll_lock);
6400 preempt_enable();
6401 rcu_read_unlock();
6402 cond_resched();
6403 if (loop_end(loop_end_arg, start_time))
6404 return;
6405 goto restart;
6406 }
6407 cpu_relax();
6408 }
6409 if (napi_poll)
6410 busy_poll_stop(napi, have_poll_lock);
6411 preempt_enable();
6412 out:
6413 rcu_read_unlock();
6414 }
6415 EXPORT_SYMBOL(napi_busy_loop);
6416
6417 #endif /* CONFIG_NET_RX_BUSY_POLL */
6418
6419 static void napi_hash_add(struct napi_struct *napi)
6420 {
6421 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6422 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6423 return;
6424
6425 spin_lock(&napi_hash_lock);
6426
6427 /* 0..NR_CPUS range is reserved for sender_cpu use */
6428 do {
6429 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6430 napi_gen_id = MIN_NAPI_ID;
6431 } while (napi_by_id(napi_gen_id));
6432 napi->napi_id = napi_gen_id;
6433
6434 hlist_add_head_rcu(&napi->napi_hash_node,
6435 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6436
6437 spin_unlock(&napi_hash_lock);
6438 }
6439
6440 /* Warning : caller is responsible to make sure rcu grace period
6441 * is respected before freeing memory containing @napi
6442 */
6443 bool napi_hash_del(struct napi_struct *napi)
6444 {
6445 bool rcu_sync_needed = false;
6446
6447 spin_lock(&napi_hash_lock);
6448
6449 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6450 rcu_sync_needed = true;
6451 hlist_del_rcu(&napi->napi_hash_node);
6452 }
6453 spin_unlock(&napi_hash_lock);
6454 return rcu_sync_needed;
6455 }
6456 EXPORT_SYMBOL_GPL(napi_hash_del);
6457
6458 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6459 {
6460 struct napi_struct *napi;
6461
6462 napi = container_of(timer, struct napi_struct, timer);
6463
6464 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6465 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6466 */
6467 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6468 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6469 __napi_schedule_irqoff(napi);
6470
6471 return HRTIMER_NORESTART;
6472 }
6473
6474 static void init_gro_hash(struct napi_struct *napi)
6475 {
6476 int i;
6477
6478 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6479 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6480 napi->gro_hash[i].count = 0;
6481 }
6482 napi->gro_bitmask = 0;
6483 }
6484
6485 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6486 int (*poll)(struct napi_struct *, int), int weight)
6487 {
6488 INIT_LIST_HEAD(&napi->poll_list);
6489 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6490 napi->timer.function = napi_watchdog;
6491 init_gro_hash(napi);
6492 napi->skb = NULL;
6493 INIT_LIST_HEAD(&napi->rx_list);
6494 napi->rx_count = 0;
6495 napi->poll = poll;
6496 if (weight > NAPI_POLL_WEIGHT)
6497 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6498 weight);
6499 napi->weight = weight;
6500 list_add(&napi->dev_list, &dev->napi_list);
6501 napi->dev = dev;
6502 #ifdef CONFIG_NETPOLL
6503 napi->poll_owner = -1;
6504 #endif
6505 set_bit(NAPI_STATE_SCHED, &napi->state);
6506 napi_hash_add(napi);
6507 }
6508 EXPORT_SYMBOL(netif_napi_add);
6509
6510 void napi_disable(struct napi_struct *n)
6511 {
6512 might_sleep();
6513 set_bit(NAPI_STATE_DISABLE, &n->state);
6514
6515 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6516 msleep(1);
6517 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6518 msleep(1);
6519
6520 hrtimer_cancel(&n->timer);
6521
6522 clear_bit(NAPI_STATE_DISABLE, &n->state);
6523 }
6524 EXPORT_SYMBOL(napi_disable);
6525
6526 static void flush_gro_hash(struct napi_struct *napi)
6527 {
6528 int i;
6529
6530 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6531 struct sk_buff *skb, *n;
6532
6533 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6534 kfree_skb(skb);
6535 napi->gro_hash[i].count = 0;
6536 }
6537 }
6538
6539 /* Must be called in process context */
6540 void netif_napi_del(struct napi_struct *napi)
6541 {
6542 might_sleep();
6543 if (napi_hash_del(napi))
6544 synchronize_net();
6545 list_del_init(&napi->dev_list);
6546 napi_free_frags(napi);
6547
6548 flush_gro_hash(napi);
6549 napi->gro_bitmask = 0;
6550 }
6551 EXPORT_SYMBOL(netif_napi_del);
6552
6553 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6554 {
6555 void *have;
6556 int work, weight;
6557
6558 list_del_init(&n->poll_list);
6559
6560 have = netpoll_poll_lock(n);
6561
6562 weight = n->weight;
6563
6564 /* This NAPI_STATE_SCHED test is for avoiding a race
6565 * with netpoll's poll_napi(). Only the entity which
6566 * obtains the lock and sees NAPI_STATE_SCHED set will
6567 * actually make the ->poll() call. Therefore we avoid
6568 * accidentally calling ->poll() when NAPI is not scheduled.
6569 */
6570 work = 0;
6571 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6572 work = n->poll(n, weight);
6573 trace_napi_poll(n, work, weight);
6574 }
6575
6576 WARN_ON_ONCE(work > weight);
6577
6578 if (likely(work < weight))
6579 goto out_unlock;
6580
6581 /* Drivers must not modify the NAPI state if they
6582 * consume the entire weight. In such cases this code
6583 * still "owns" the NAPI instance and therefore can
6584 * move the instance around on the list at-will.
6585 */
6586 if (unlikely(napi_disable_pending(n))) {
6587 napi_complete(n);
6588 goto out_unlock;
6589 }
6590
6591 if (n->gro_bitmask) {
6592 /* flush too old packets
6593 * If HZ < 1000, flush all packets.
6594 */
6595 napi_gro_flush(n, HZ >= 1000);
6596 }
6597
6598 gro_normal_list(n);
6599
6600 /* Some drivers may have called napi_schedule
6601 * prior to exhausting their budget.
6602 */
6603 if (unlikely(!list_empty(&n->poll_list))) {
6604 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6605 n->dev ? n->dev->name : "backlog");
6606 goto out_unlock;
6607 }
6608
6609 list_add_tail(&n->poll_list, repoll);
6610
6611 out_unlock:
6612 netpoll_poll_unlock(have);
6613
6614 return work;
6615 }
6616
6617 static __latent_entropy void net_rx_action(struct softirq_action *h)
6618 {
6619 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6620 unsigned long time_limit = jiffies +
6621 usecs_to_jiffies(netdev_budget_usecs);
6622 int budget = netdev_budget;
6623 LIST_HEAD(list);
6624 LIST_HEAD(repoll);
6625
6626 local_irq_disable();
6627 list_splice_init(&sd->poll_list, &list);
6628 local_irq_enable();
6629
6630 for (;;) {
6631 struct napi_struct *n;
6632
6633 if (list_empty(&list)) {
6634 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6635 goto out;
6636 break;
6637 }
6638
6639 n = list_first_entry(&list, struct napi_struct, poll_list);
6640 budget -= napi_poll(n, &repoll);
6641
6642 /* If softirq window is exhausted then punt.
6643 * Allow this to run for 2 jiffies since which will allow
6644 * an average latency of 1.5/HZ.
6645 */
6646 if (unlikely(budget <= 0 ||
6647 time_after_eq(jiffies, time_limit))) {
6648 sd->time_squeeze++;
6649 break;
6650 }
6651 }
6652
6653 local_irq_disable();
6654
6655 list_splice_tail_init(&sd->poll_list, &list);
6656 list_splice_tail(&repoll, &list);
6657 list_splice(&list, &sd->poll_list);
6658 if (!list_empty(&sd->poll_list))
6659 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6660
6661 net_rps_action_and_irq_enable(sd);
6662 out:
6663 __kfree_skb_flush();
6664 }
6665
6666 struct netdev_adjacent {
6667 struct net_device *dev;
6668
6669 /* upper master flag, there can only be one master device per list */
6670 bool master;
6671
6672 /* lookup ignore flag */
6673 bool ignore;
6674
6675 /* counter for the number of times this device was added to us */
6676 u16 ref_nr;
6677
6678 /* private field for the users */
6679 void *private;
6680
6681 struct list_head list;
6682 struct rcu_head rcu;
6683 };
6684
6685 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6686 struct list_head *adj_list)
6687 {
6688 struct netdev_adjacent *adj;
6689
6690 list_for_each_entry(adj, adj_list, list) {
6691 if (adj->dev == adj_dev)
6692 return adj;
6693 }
6694 return NULL;
6695 }
6696
6697 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6698 {
6699 struct net_device *dev = data;
6700
6701 return upper_dev == dev;
6702 }
6703
6704 /**
6705 * netdev_has_upper_dev - Check if device is linked to an upper device
6706 * @dev: device
6707 * @upper_dev: upper device to check
6708 *
6709 * Find out if a device is linked to specified upper device and return true
6710 * in case it is. Note that this checks only immediate upper device,
6711 * not through a complete stack of devices. The caller must hold the RTNL lock.
6712 */
6713 bool netdev_has_upper_dev(struct net_device *dev,
6714 struct net_device *upper_dev)
6715 {
6716 ASSERT_RTNL();
6717
6718 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6719 upper_dev);
6720 }
6721 EXPORT_SYMBOL(netdev_has_upper_dev);
6722
6723 /**
6724 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6725 * @dev: device
6726 * @upper_dev: upper device to check
6727 *
6728 * Find out if a device is linked to specified upper device and return true
6729 * in case it is. Note that this checks the entire upper device chain.
6730 * The caller must hold rcu lock.
6731 */
6732
6733 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6734 struct net_device *upper_dev)
6735 {
6736 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6737 upper_dev);
6738 }
6739 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6740
6741 /**
6742 * netdev_has_any_upper_dev - Check if device is linked to some device
6743 * @dev: device
6744 *
6745 * Find out if a device is linked to an upper device and return true in case
6746 * it is. The caller must hold the RTNL lock.
6747 */
6748 bool netdev_has_any_upper_dev(struct net_device *dev)
6749 {
6750 ASSERT_RTNL();
6751
6752 return !list_empty(&dev->adj_list.upper);
6753 }
6754 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6755
6756 /**
6757 * netdev_master_upper_dev_get - Get master upper device
6758 * @dev: device
6759 *
6760 * Find a master upper device and return pointer to it or NULL in case
6761 * it's not there. The caller must hold the RTNL lock.
6762 */
6763 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6764 {
6765 struct netdev_adjacent *upper;
6766
6767 ASSERT_RTNL();
6768
6769 if (list_empty(&dev->adj_list.upper))
6770 return NULL;
6771
6772 upper = list_first_entry(&dev->adj_list.upper,
6773 struct netdev_adjacent, list);
6774 if (likely(upper->master))
6775 return upper->dev;
6776 return NULL;
6777 }
6778 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6779
6780 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6781 {
6782 struct netdev_adjacent *upper;
6783
6784 ASSERT_RTNL();
6785
6786 if (list_empty(&dev->adj_list.upper))
6787 return NULL;
6788
6789 upper = list_first_entry(&dev->adj_list.upper,
6790 struct netdev_adjacent, list);
6791 if (likely(upper->master) && !upper->ignore)
6792 return upper->dev;
6793 return NULL;
6794 }
6795
6796 /**
6797 * netdev_has_any_lower_dev - Check if device is linked to some device
6798 * @dev: device
6799 *
6800 * Find out if a device is linked to a lower device and return true in case
6801 * it is. The caller must hold the RTNL lock.
6802 */
6803 static bool netdev_has_any_lower_dev(struct net_device *dev)
6804 {
6805 ASSERT_RTNL();
6806
6807 return !list_empty(&dev->adj_list.lower);
6808 }
6809
6810 void *netdev_adjacent_get_private(struct list_head *adj_list)
6811 {
6812 struct netdev_adjacent *adj;
6813
6814 adj = list_entry(adj_list, struct netdev_adjacent, list);
6815
6816 return adj->private;
6817 }
6818 EXPORT_SYMBOL(netdev_adjacent_get_private);
6819
6820 /**
6821 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6822 * @dev: device
6823 * @iter: list_head ** of the current position
6824 *
6825 * Gets the next device from the dev's upper list, starting from iter
6826 * position. The caller must hold RCU read lock.
6827 */
6828 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6829 struct list_head **iter)
6830 {
6831 struct netdev_adjacent *upper;
6832
6833 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6834
6835 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6836
6837 if (&upper->list == &dev->adj_list.upper)
6838 return NULL;
6839
6840 *iter = &upper->list;
6841
6842 return upper->dev;
6843 }
6844 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6845
6846 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6847 struct list_head **iter,
6848 bool *ignore)
6849 {
6850 struct netdev_adjacent *upper;
6851
6852 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6853
6854 if (&upper->list == &dev->adj_list.upper)
6855 return NULL;
6856
6857 *iter = &upper->list;
6858 *ignore = upper->ignore;
6859
6860 return upper->dev;
6861 }
6862
6863 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6864 struct list_head **iter)
6865 {
6866 struct netdev_adjacent *upper;
6867
6868 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6869
6870 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6871
6872 if (&upper->list == &dev->adj_list.upper)
6873 return NULL;
6874
6875 *iter = &upper->list;
6876
6877 return upper->dev;
6878 }
6879
6880 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6881 int (*fn)(struct net_device *dev,
6882 void *data),
6883 void *data)
6884 {
6885 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6886 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6887 int ret, cur = 0;
6888 bool ignore;
6889
6890 now = dev;
6891 iter = &dev->adj_list.upper;
6892
6893 while (1) {
6894 if (now != dev) {
6895 ret = fn(now, data);
6896 if (ret)
6897 return ret;
6898 }
6899
6900 next = NULL;
6901 while (1) {
6902 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6903 if (!udev)
6904 break;
6905 if (ignore)
6906 continue;
6907
6908 next = udev;
6909 niter = &udev->adj_list.upper;
6910 dev_stack[cur] = now;
6911 iter_stack[cur++] = iter;
6912 break;
6913 }
6914
6915 if (!next) {
6916 if (!cur)
6917 return 0;
6918 next = dev_stack[--cur];
6919 niter = iter_stack[cur];
6920 }
6921
6922 now = next;
6923 iter = niter;
6924 }
6925
6926 return 0;
6927 }
6928
6929 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6930 int (*fn)(struct net_device *dev,
6931 void *data),
6932 void *data)
6933 {
6934 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6935 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6936 int ret, cur = 0;
6937
6938 now = dev;
6939 iter = &dev->adj_list.upper;
6940
6941 while (1) {
6942 if (now != dev) {
6943 ret = fn(now, data);
6944 if (ret)
6945 return ret;
6946 }
6947
6948 next = NULL;
6949 while (1) {
6950 udev = netdev_next_upper_dev_rcu(now, &iter);
6951 if (!udev)
6952 break;
6953
6954 next = udev;
6955 niter = &udev->adj_list.upper;
6956 dev_stack[cur] = now;
6957 iter_stack[cur++] = iter;
6958 break;
6959 }
6960
6961 if (!next) {
6962 if (!cur)
6963 return 0;
6964 next = dev_stack[--cur];
6965 niter = iter_stack[cur];
6966 }
6967
6968 now = next;
6969 iter = niter;
6970 }
6971
6972 return 0;
6973 }
6974 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6975
6976 static bool __netdev_has_upper_dev(struct net_device *dev,
6977 struct net_device *upper_dev)
6978 {
6979 ASSERT_RTNL();
6980
6981 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6982 upper_dev);
6983 }
6984
6985 /**
6986 * netdev_lower_get_next_private - Get the next ->private from the
6987 * lower neighbour list
6988 * @dev: device
6989 * @iter: list_head ** of the current position
6990 *
6991 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6992 * list, starting from iter position. The caller must hold either hold the
6993 * RTNL lock or its own locking that guarantees that the neighbour lower
6994 * list will remain unchanged.
6995 */
6996 void *netdev_lower_get_next_private(struct net_device *dev,
6997 struct list_head **iter)
6998 {
6999 struct netdev_adjacent *lower;
7000
7001 lower = list_entry(*iter, struct netdev_adjacent, list);
7002
7003 if (&lower->list == &dev->adj_list.lower)
7004 return NULL;
7005
7006 *iter = lower->list.next;
7007
7008 return lower->private;
7009 }
7010 EXPORT_SYMBOL(netdev_lower_get_next_private);
7011
7012 /**
7013 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7014 * lower neighbour list, RCU
7015 * variant
7016 * @dev: device
7017 * @iter: list_head ** of the current position
7018 *
7019 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7020 * list, starting from iter position. The caller must hold RCU read lock.
7021 */
7022 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7023 struct list_head **iter)
7024 {
7025 struct netdev_adjacent *lower;
7026
7027 WARN_ON_ONCE(!rcu_read_lock_held());
7028
7029 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7030
7031 if (&lower->list == &dev->adj_list.lower)
7032 return NULL;
7033
7034 *iter = &lower->list;
7035
7036 return lower->private;
7037 }
7038 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7039
7040 /**
7041 * netdev_lower_get_next - Get the next device from the lower neighbour
7042 * list
7043 * @dev: device
7044 * @iter: list_head ** of the current position
7045 *
7046 * Gets the next netdev_adjacent from the dev's lower neighbour
7047 * list, starting from iter position. The caller must hold RTNL lock or
7048 * its own locking that guarantees that the neighbour lower
7049 * list will remain unchanged.
7050 */
7051 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7052 {
7053 struct netdev_adjacent *lower;
7054
7055 lower = list_entry(*iter, struct netdev_adjacent, list);
7056
7057 if (&lower->list == &dev->adj_list.lower)
7058 return NULL;
7059
7060 *iter = lower->list.next;
7061
7062 return lower->dev;
7063 }
7064 EXPORT_SYMBOL(netdev_lower_get_next);
7065
7066 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7067 struct list_head **iter)
7068 {
7069 struct netdev_adjacent *lower;
7070
7071 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7072
7073 if (&lower->list == &dev->adj_list.lower)
7074 return NULL;
7075
7076 *iter = &lower->list;
7077
7078 return lower->dev;
7079 }
7080
7081 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7082 struct list_head **iter,
7083 bool *ignore)
7084 {
7085 struct netdev_adjacent *lower;
7086
7087 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7088
7089 if (&lower->list == &dev->adj_list.lower)
7090 return NULL;
7091
7092 *iter = &lower->list;
7093 *ignore = lower->ignore;
7094
7095 return lower->dev;
7096 }
7097
7098 int netdev_walk_all_lower_dev(struct net_device *dev,
7099 int (*fn)(struct net_device *dev,
7100 void *data),
7101 void *data)
7102 {
7103 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7104 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7105 int ret, cur = 0;
7106
7107 now = dev;
7108 iter = &dev->adj_list.lower;
7109
7110 while (1) {
7111 if (now != dev) {
7112 ret = fn(now, data);
7113 if (ret)
7114 return ret;
7115 }
7116
7117 next = NULL;
7118 while (1) {
7119 ldev = netdev_next_lower_dev(now, &iter);
7120 if (!ldev)
7121 break;
7122
7123 next = ldev;
7124 niter = &ldev->adj_list.lower;
7125 dev_stack[cur] = now;
7126 iter_stack[cur++] = iter;
7127 break;
7128 }
7129
7130 if (!next) {
7131 if (!cur)
7132 return 0;
7133 next = dev_stack[--cur];
7134 niter = iter_stack[cur];
7135 }
7136
7137 now = next;
7138 iter = niter;
7139 }
7140
7141 return 0;
7142 }
7143 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7144
7145 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7146 int (*fn)(struct net_device *dev,
7147 void *data),
7148 void *data)
7149 {
7150 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7151 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7152 int ret, cur = 0;
7153 bool ignore;
7154
7155 now = dev;
7156 iter = &dev->adj_list.lower;
7157
7158 while (1) {
7159 if (now != dev) {
7160 ret = fn(now, data);
7161 if (ret)
7162 return ret;
7163 }
7164
7165 next = NULL;
7166 while (1) {
7167 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7168 if (!ldev)
7169 break;
7170 if (ignore)
7171 continue;
7172
7173 next = ldev;
7174 niter = &ldev->adj_list.lower;
7175 dev_stack[cur] = now;
7176 iter_stack[cur++] = iter;
7177 break;
7178 }
7179
7180 if (!next) {
7181 if (!cur)
7182 return 0;
7183 next = dev_stack[--cur];
7184 niter = iter_stack[cur];
7185 }
7186
7187 now = next;
7188 iter = niter;
7189 }
7190
7191 return 0;
7192 }
7193
7194 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7195 struct list_head **iter)
7196 {
7197 struct netdev_adjacent *lower;
7198
7199 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7200 if (&lower->list == &dev->adj_list.lower)
7201 return NULL;
7202
7203 *iter = &lower->list;
7204
7205 return lower->dev;
7206 }
7207 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7208
7209 static u8 __netdev_upper_depth(struct net_device *dev)
7210 {
7211 struct net_device *udev;
7212 struct list_head *iter;
7213 u8 max_depth = 0;
7214 bool ignore;
7215
7216 for (iter = &dev->adj_list.upper,
7217 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7218 udev;
7219 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7220 if (ignore)
7221 continue;
7222 if (max_depth < udev->upper_level)
7223 max_depth = udev->upper_level;
7224 }
7225
7226 return max_depth;
7227 }
7228
7229 static u8 __netdev_lower_depth(struct net_device *dev)
7230 {
7231 struct net_device *ldev;
7232 struct list_head *iter;
7233 u8 max_depth = 0;
7234 bool ignore;
7235
7236 for (iter = &dev->adj_list.lower,
7237 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7238 ldev;
7239 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7240 if (ignore)
7241 continue;
7242 if (max_depth < ldev->lower_level)
7243 max_depth = ldev->lower_level;
7244 }
7245
7246 return max_depth;
7247 }
7248
7249 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7250 {
7251 dev->upper_level = __netdev_upper_depth(dev) + 1;
7252 return 0;
7253 }
7254
7255 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7256 {
7257 dev->lower_level = __netdev_lower_depth(dev) + 1;
7258 return 0;
7259 }
7260
7261 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7262 int (*fn)(struct net_device *dev,
7263 void *data),
7264 void *data)
7265 {
7266 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7267 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7268 int ret, cur = 0;
7269
7270 now = dev;
7271 iter = &dev->adj_list.lower;
7272
7273 while (1) {
7274 if (now != dev) {
7275 ret = fn(now, data);
7276 if (ret)
7277 return ret;
7278 }
7279
7280 next = NULL;
7281 while (1) {
7282 ldev = netdev_next_lower_dev_rcu(now, &iter);
7283 if (!ldev)
7284 break;
7285
7286 next = ldev;
7287 niter = &ldev->adj_list.lower;
7288 dev_stack[cur] = now;
7289 iter_stack[cur++] = iter;
7290 break;
7291 }
7292
7293 if (!next) {
7294 if (!cur)
7295 return 0;
7296 next = dev_stack[--cur];
7297 niter = iter_stack[cur];
7298 }
7299
7300 now = next;
7301 iter = niter;
7302 }
7303
7304 return 0;
7305 }
7306 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7307
7308 /**
7309 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7310 * lower neighbour list, RCU
7311 * variant
7312 * @dev: device
7313 *
7314 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7315 * list. The caller must hold RCU read lock.
7316 */
7317 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7318 {
7319 struct netdev_adjacent *lower;
7320
7321 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7322 struct netdev_adjacent, list);
7323 if (lower)
7324 return lower->private;
7325 return NULL;
7326 }
7327 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7328
7329 /**
7330 * netdev_master_upper_dev_get_rcu - Get master upper device
7331 * @dev: device
7332 *
7333 * Find a master upper device and return pointer to it or NULL in case
7334 * it's not there. The caller must hold the RCU read lock.
7335 */
7336 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7337 {
7338 struct netdev_adjacent *upper;
7339
7340 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7341 struct netdev_adjacent, list);
7342 if (upper && likely(upper->master))
7343 return upper->dev;
7344 return NULL;
7345 }
7346 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7347
7348 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7349 struct net_device *adj_dev,
7350 struct list_head *dev_list)
7351 {
7352 char linkname[IFNAMSIZ+7];
7353
7354 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7355 "upper_%s" : "lower_%s", adj_dev->name);
7356 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7357 linkname);
7358 }
7359 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7360 char *name,
7361 struct list_head *dev_list)
7362 {
7363 char linkname[IFNAMSIZ+7];
7364
7365 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7366 "upper_%s" : "lower_%s", name);
7367 sysfs_remove_link(&(dev->dev.kobj), linkname);
7368 }
7369
7370 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7371 struct net_device *adj_dev,
7372 struct list_head *dev_list)
7373 {
7374 return (dev_list == &dev->adj_list.upper ||
7375 dev_list == &dev->adj_list.lower) &&
7376 net_eq(dev_net(dev), dev_net(adj_dev));
7377 }
7378
7379 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7380 struct net_device *adj_dev,
7381 struct list_head *dev_list,
7382 void *private, bool master)
7383 {
7384 struct netdev_adjacent *adj;
7385 int ret;
7386
7387 adj = __netdev_find_adj(adj_dev, dev_list);
7388
7389 if (adj) {
7390 adj->ref_nr += 1;
7391 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7392 dev->name, adj_dev->name, adj->ref_nr);
7393
7394 return 0;
7395 }
7396
7397 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7398 if (!adj)
7399 return -ENOMEM;
7400
7401 adj->dev = adj_dev;
7402 adj->master = master;
7403 adj->ref_nr = 1;
7404 adj->private = private;
7405 adj->ignore = false;
7406 dev_hold(adj_dev);
7407
7408 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7409 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7410
7411 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7412 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7413 if (ret)
7414 goto free_adj;
7415 }
7416
7417 /* Ensure that master link is always the first item in list. */
7418 if (master) {
7419 ret = sysfs_create_link(&(dev->dev.kobj),
7420 &(adj_dev->dev.kobj), "master");
7421 if (ret)
7422 goto remove_symlinks;
7423
7424 list_add_rcu(&adj->list, dev_list);
7425 } else {
7426 list_add_tail_rcu(&adj->list, dev_list);
7427 }
7428
7429 return 0;
7430
7431 remove_symlinks:
7432 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7433 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7434 free_adj:
7435 kfree(adj);
7436 dev_put(adj_dev);
7437
7438 return ret;
7439 }
7440
7441 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7442 struct net_device *adj_dev,
7443 u16 ref_nr,
7444 struct list_head *dev_list)
7445 {
7446 struct netdev_adjacent *adj;
7447
7448 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7449 dev->name, adj_dev->name, ref_nr);
7450
7451 adj = __netdev_find_adj(adj_dev, dev_list);
7452
7453 if (!adj) {
7454 pr_err("Adjacency does not exist for device %s from %s\n",
7455 dev->name, adj_dev->name);
7456 WARN_ON(1);
7457 return;
7458 }
7459
7460 if (adj->ref_nr > ref_nr) {
7461 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7462 dev->name, adj_dev->name, ref_nr,
7463 adj->ref_nr - ref_nr);
7464 adj->ref_nr -= ref_nr;
7465 return;
7466 }
7467
7468 if (adj->master)
7469 sysfs_remove_link(&(dev->dev.kobj), "master");
7470
7471 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7472 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7473
7474 list_del_rcu(&adj->list);
7475 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7476 adj_dev->name, dev->name, adj_dev->name);
7477 dev_put(adj_dev);
7478 kfree_rcu(adj, rcu);
7479 }
7480
7481 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7482 struct net_device *upper_dev,
7483 struct list_head *up_list,
7484 struct list_head *down_list,
7485 void *private, bool master)
7486 {
7487 int ret;
7488
7489 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7490 private, master);
7491 if (ret)
7492 return ret;
7493
7494 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7495 private, false);
7496 if (ret) {
7497 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7498 return ret;
7499 }
7500
7501 return 0;
7502 }
7503
7504 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7505 struct net_device *upper_dev,
7506 u16 ref_nr,
7507 struct list_head *up_list,
7508 struct list_head *down_list)
7509 {
7510 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7511 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7512 }
7513
7514 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7515 struct net_device *upper_dev,
7516 void *private, bool master)
7517 {
7518 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7519 &dev->adj_list.upper,
7520 &upper_dev->adj_list.lower,
7521 private, master);
7522 }
7523
7524 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7525 struct net_device *upper_dev)
7526 {
7527 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7528 &dev->adj_list.upper,
7529 &upper_dev->adj_list.lower);
7530 }
7531
7532 static int __netdev_upper_dev_link(struct net_device *dev,
7533 struct net_device *upper_dev, bool master,
7534 void *upper_priv, void *upper_info,
7535 struct netlink_ext_ack *extack)
7536 {
7537 struct netdev_notifier_changeupper_info changeupper_info = {
7538 .info = {
7539 .dev = dev,
7540 .extack = extack,
7541 },
7542 .upper_dev = upper_dev,
7543 .master = master,
7544 .linking = true,
7545 .upper_info = upper_info,
7546 };
7547 struct net_device *master_dev;
7548 int ret = 0;
7549
7550 ASSERT_RTNL();
7551
7552 if (dev == upper_dev)
7553 return -EBUSY;
7554
7555 /* To prevent loops, check if dev is not upper device to upper_dev. */
7556 if (__netdev_has_upper_dev(upper_dev, dev))
7557 return -EBUSY;
7558
7559 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7560 return -EMLINK;
7561
7562 if (!master) {
7563 if (__netdev_has_upper_dev(dev, upper_dev))
7564 return -EEXIST;
7565 } else {
7566 master_dev = __netdev_master_upper_dev_get(dev);
7567 if (master_dev)
7568 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7569 }
7570
7571 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7572 &changeupper_info.info);
7573 ret = notifier_to_errno(ret);
7574 if (ret)
7575 return ret;
7576
7577 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7578 master);
7579 if (ret)
7580 return ret;
7581
7582 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7583 &changeupper_info.info);
7584 ret = notifier_to_errno(ret);
7585 if (ret)
7586 goto rollback;
7587
7588 __netdev_update_upper_level(dev, NULL);
7589 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7590
7591 __netdev_update_lower_level(upper_dev, NULL);
7592 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7593 NULL);
7594
7595 return 0;
7596
7597 rollback:
7598 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7599
7600 return ret;
7601 }
7602
7603 /**
7604 * netdev_upper_dev_link - Add a link to the upper device
7605 * @dev: device
7606 * @upper_dev: new upper device
7607 * @extack: netlink extended ack
7608 *
7609 * Adds a link to device which is upper to this one. The caller must hold
7610 * the RTNL lock. On a failure a negative errno code is returned.
7611 * On success the reference counts are adjusted and the function
7612 * returns zero.
7613 */
7614 int netdev_upper_dev_link(struct net_device *dev,
7615 struct net_device *upper_dev,
7616 struct netlink_ext_ack *extack)
7617 {
7618 return __netdev_upper_dev_link(dev, upper_dev, false,
7619 NULL, NULL, extack);
7620 }
7621 EXPORT_SYMBOL(netdev_upper_dev_link);
7622
7623 /**
7624 * netdev_master_upper_dev_link - Add a master link to the upper device
7625 * @dev: device
7626 * @upper_dev: new upper device
7627 * @upper_priv: upper device private
7628 * @upper_info: upper info to be passed down via notifier
7629 * @extack: netlink extended ack
7630 *
7631 * Adds a link to device which is upper to this one. In this case, only
7632 * one master upper device can be linked, although other non-master devices
7633 * might be linked as well. The caller must hold the RTNL lock.
7634 * On a failure a negative errno code is returned. On success the reference
7635 * counts are adjusted and the function returns zero.
7636 */
7637 int netdev_master_upper_dev_link(struct net_device *dev,
7638 struct net_device *upper_dev,
7639 void *upper_priv, void *upper_info,
7640 struct netlink_ext_ack *extack)
7641 {
7642 return __netdev_upper_dev_link(dev, upper_dev, true,
7643 upper_priv, upper_info, extack);
7644 }
7645 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7646
7647 /**
7648 * netdev_upper_dev_unlink - Removes a link to upper device
7649 * @dev: device
7650 * @upper_dev: new upper device
7651 *
7652 * Removes a link to device which is upper to this one. The caller must hold
7653 * the RTNL lock.
7654 */
7655 void netdev_upper_dev_unlink(struct net_device *dev,
7656 struct net_device *upper_dev)
7657 {
7658 struct netdev_notifier_changeupper_info changeupper_info = {
7659 .info = {
7660 .dev = dev,
7661 },
7662 .upper_dev = upper_dev,
7663 .linking = false,
7664 };
7665
7666 ASSERT_RTNL();
7667
7668 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7669
7670 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7671 &changeupper_info.info);
7672
7673 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7674
7675 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7676 &changeupper_info.info);
7677
7678 __netdev_update_upper_level(dev, NULL);
7679 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7680
7681 __netdev_update_lower_level(upper_dev, NULL);
7682 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7683 NULL);
7684 }
7685 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7686
7687 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7688 struct net_device *lower_dev,
7689 bool val)
7690 {
7691 struct netdev_adjacent *adj;
7692
7693 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7694 if (adj)
7695 adj->ignore = val;
7696
7697 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7698 if (adj)
7699 adj->ignore = val;
7700 }
7701
7702 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7703 struct net_device *lower_dev)
7704 {
7705 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7706 }
7707
7708 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7709 struct net_device *lower_dev)
7710 {
7711 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7712 }
7713
7714 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7715 struct net_device *new_dev,
7716 struct net_device *dev,
7717 struct netlink_ext_ack *extack)
7718 {
7719 int err;
7720
7721 if (!new_dev)
7722 return 0;
7723
7724 if (old_dev && new_dev != old_dev)
7725 netdev_adjacent_dev_disable(dev, old_dev);
7726
7727 err = netdev_upper_dev_link(new_dev, dev, extack);
7728 if (err) {
7729 if (old_dev && new_dev != old_dev)
7730 netdev_adjacent_dev_enable(dev, old_dev);
7731 return err;
7732 }
7733
7734 return 0;
7735 }
7736 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7737
7738 void netdev_adjacent_change_commit(struct net_device *old_dev,
7739 struct net_device *new_dev,
7740 struct net_device *dev)
7741 {
7742 if (!new_dev || !old_dev)
7743 return;
7744
7745 if (new_dev == old_dev)
7746 return;
7747
7748 netdev_adjacent_dev_enable(dev, old_dev);
7749 netdev_upper_dev_unlink(old_dev, dev);
7750 }
7751 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7752
7753 void netdev_adjacent_change_abort(struct net_device *old_dev,
7754 struct net_device *new_dev,
7755 struct net_device *dev)
7756 {
7757 if (!new_dev)
7758 return;
7759
7760 if (old_dev && new_dev != old_dev)
7761 netdev_adjacent_dev_enable(dev, old_dev);
7762
7763 netdev_upper_dev_unlink(new_dev, dev);
7764 }
7765 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7766
7767 /**
7768 * netdev_bonding_info_change - Dispatch event about slave change
7769 * @dev: device
7770 * @bonding_info: info to dispatch
7771 *
7772 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7773 * The caller must hold the RTNL lock.
7774 */
7775 void netdev_bonding_info_change(struct net_device *dev,
7776 struct netdev_bonding_info *bonding_info)
7777 {
7778 struct netdev_notifier_bonding_info info = {
7779 .info.dev = dev,
7780 };
7781
7782 memcpy(&info.bonding_info, bonding_info,
7783 sizeof(struct netdev_bonding_info));
7784 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7785 &info.info);
7786 }
7787 EXPORT_SYMBOL(netdev_bonding_info_change);
7788
7789 static void netdev_adjacent_add_links(struct net_device *dev)
7790 {
7791 struct netdev_adjacent *iter;
7792
7793 struct net *net = dev_net(dev);
7794
7795 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7796 if (!net_eq(net, dev_net(iter->dev)))
7797 continue;
7798 netdev_adjacent_sysfs_add(iter->dev, dev,
7799 &iter->dev->adj_list.lower);
7800 netdev_adjacent_sysfs_add(dev, iter->dev,
7801 &dev->adj_list.upper);
7802 }
7803
7804 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7805 if (!net_eq(net, dev_net(iter->dev)))
7806 continue;
7807 netdev_adjacent_sysfs_add(iter->dev, dev,
7808 &iter->dev->adj_list.upper);
7809 netdev_adjacent_sysfs_add(dev, iter->dev,
7810 &dev->adj_list.lower);
7811 }
7812 }
7813
7814 static void netdev_adjacent_del_links(struct net_device *dev)
7815 {
7816 struct netdev_adjacent *iter;
7817
7818 struct net *net = dev_net(dev);
7819
7820 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7821 if (!net_eq(net, dev_net(iter->dev)))
7822 continue;
7823 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7824 &iter->dev->adj_list.lower);
7825 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7826 &dev->adj_list.upper);
7827 }
7828
7829 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7830 if (!net_eq(net, dev_net(iter->dev)))
7831 continue;
7832 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7833 &iter->dev->adj_list.upper);
7834 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7835 &dev->adj_list.lower);
7836 }
7837 }
7838
7839 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7840 {
7841 struct netdev_adjacent *iter;
7842
7843 struct net *net = dev_net(dev);
7844
7845 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7846 if (!net_eq(net, dev_net(iter->dev)))
7847 continue;
7848 netdev_adjacent_sysfs_del(iter->dev, oldname,
7849 &iter->dev->adj_list.lower);
7850 netdev_adjacent_sysfs_add(iter->dev, dev,
7851 &iter->dev->adj_list.lower);
7852 }
7853
7854 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7855 if (!net_eq(net, dev_net(iter->dev)))
7856 continue;
7857 netdev_adjacent_sysfs_del(iter->dev, oldname,
7858 &iter->dev->adj_list.upper);
7859 netdev_adjacent_sysfs_add(iter->dev, dev,
7860 &iter->dev->adj_list.upper);
7861 }
7862 }
7863
7864 void *netdev_lower_dev_get_private(struct net_device *dev,
7865 struct net_device *lower_dev)
7866 {
7867 struct netdev_adjacent *lower;
7868
7869 if (!lower_dev)
7870 return NULL;
7871 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7872 if (!lower)
7873 return NULL;
7874
7875 return lower->private;
7876 }
7877 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7878
7879
7880 /**
7881 * netdev_lower_change - Dispatch event about lower device state change
7882 * @lower_dev: device
7883 * @lower_state_info: state to dispatch
7884 *
7885 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7886 * The caller must hold the RTNL lock.
7887 */
7888 void netdev_lower_state_changed(struct net_device *lower_dev,
7889 void *lower_state_info)
7890 {
7891 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7892 .info.dev = lower_dev,
7893 };
7894
7895 ASSERT_RTNL();
7896 changelowerstate_info.lower_state_info = lower_state_info;
7897 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7898 &changelowerstate_info.info);
7899 }
7900 EXPORT_SYMBOL(netdev_lower_state_changed);
7901
7902 static void dev_change_rx_flags(struct net_device *dev, int flags)
7903 {
7904 const struct net_device_ops *ops = dev->netdev_ops;
7905
7906 if (ops->ndo_change_rx_flags)
7907 ops->ndo_change_rx_flags(dev, flags);
7908 }
7909
7910 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7911 {
7912 unsigned int old_flags = dev->flags;
7913 kuid_t uid;
7914 kgid_t gid;
7915
7916 ASSERT_RTNL();
7917
7918 dev->flags |= IFF_PROMISC;
7919 dev->promiscuity += inc;
7920 if (dev->promiscuity == 0) {
7921 /*
7922 * Avoid overflow.
7923 * If inc causes overflow, untouch promisc and return error.
7924 */
7925 if (inc < 0)
7926 dev->flags &= ~IFF_PROMISC;
7927 else {
7928 dev->promiscuity -= inc;
7929 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7930 dev->name);
7931 return -EOVERFLOW;
7932 }
7933 }
7934 if (dev->flags != old_flags) {
7935 pr_info("device %s %s promiscuous mode\n",
7936 dev->name,
7937 dev->flags & IFF_PROMISC ? "entered" : "left");
7938 if (audit_enabled) {
7939 current_uid_gid(&uid, &gid);
7940 audit_log(audit_context(), GFP_ATOMIC,
7941 AUDIT_ANOM_PROMISCUOUS,
7942 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7943 dev->name, (dev->flags & IFF_PROMISC),
7944 (old_flags & IFF_PROMISC),
7945 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7946 from_kuid(&init_user_ns, uid),
7947 from_kgid(&init_user_ns, gid),
7948 audit_get_sessionid(current));
7949 }
7950
7951 dev_change_rx_flags(dev, IFF_PROMISC);
7952 }
7953 if (notify)
7954 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7955 return 0;
7956 }
7957
7958 /**
7959 * dev_set_promiscuity - update promiscuity count on a device
7960 * @dev: device
7961 * @inc: modifier
7962 *
7963 * Add or remove promiscuity from a device. While the count in the device
7964 * remains above zero the interface remains promiscuous. Once it hits zero
7965 * the device reverts back to normal filtering operation. A negative inc
7966 * value is used to drop promiscuity on the device.
7967 * Return 0 if successful or a negative errno code on error.
7968 */
7969 int dev_set_promiscuity(struct net_device *dev, int inc)
7970 {
7971 unsigned int old_flags = dev->flags;
7972 int err;
7973
7974 err = __dev_set_promiscuity(dev, inc, true);
7975 if (err < 0)
7976 return err;
7977 if (dev->flags != old_flags)
7978 dev_set_rx_mode(dev);
7979 return err;
7980 }
7981 EXPORT_SYMBOL(dev_set_promiscuity);
7982
7983 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7984 {
7985 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7986
7987 ASSERT_RTNL();
7988
7989 dev->flags |= IFF_ALLMULTI;
7990 dev->allmulti += inc;
7991 if (dev->allmulti == 0) {
7992 /*
7993 * Avoid overflow.
7994 * If inc causes overflow, untouch allmulti and return error.
7995 */
7996 if (inc < 0)
7997 dev->flags &= ~IFF_ALLMULTI;
7998 else {
7999 dev->allmulti -= inc;
8000 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8001 dev->name);
8002 return -EOVERFLOW;
8003 }
8004 }
8005 if (dev->flags ^ old_flags) {
8006 dev_change_rx_flags(dev, IFF_ALLMULTI);
8007 dev_set_rx_mode(dev);
8008 if (notify)
8009 __dev_notify_flags(dev, old_flags,
8010 dev->gflags ^ old_gflags);
8011 }
8012 return 0;
8013 }
8014
8015 /**
8016 * dev_set_allmulti - update allmulti count on a device
8017 * @dev: device
8018 * @inc: modifier
8019 *
8020 * Add or remove reception of all multicast frames to a device. While the
8021 * count in the device remains above zero the interface remains listening
8022 * to all interfaces. Once it hits zero the device reverts back to normal
8023 * filtering operation. A negative @inc value is used to drop the counter
8024 * when releasing a resource needing all multicasts.
8025 * Return 0 if successful or a negative errno code on error.
8026 */
8027
8028 int dev_set_allmulti(struct net_device *dev, int inc)
8029 {
8030 return __dev_set_allmulti(dev, inc, true);
8031 }
8032 EXPORT_SYMBOL(dev_set_allmulti);
8033
8034 /*
8035 * Upload unicast and multicast address lists to device and
8036 * configure RX filtering. When the device doesn't support unicast
8037 * filtering it is put in promiscuous mode while unicast addresses
8038 * are present.
8039 */
8040 void __dev_set_rx_mode(struct net_device *dev)
8041 {
8042 const struct net_device_ops *ops = dev->netdev_ops;
8043
8044 /* dev_open will call this function so the list will stay sane. */
8045 if (!(dev->flags&IFF_UP))
8046 return;
8047
8048 if (!netif_device_present(dev))
8049 return;
8050
8051 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8052 /* Unicast addresses changes may only happen under the rtnl,
8053 * therefore calling __dev_set_promiscuity here is safe.
8054 */
8055 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8056 __dev_set_promiscuity(dev, 1, false);
8057 dev->uc_promisc = true;
8058 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8059 __dev_set_promiscuity(dev, -1, false);
8060 dev->uc_promisc = false;
8061 }
8062 }
8063
8064 if (ops->ndo_set_rx_mode)
8065 ops->ndo_set_rx_mode(dev);
8066 }
8067
8068 void dev_set_rx_mode(struct net_device *dev)
8069 {
8070 netif_addr_lock_bh(dev);
8071 __dev_set_rx_mode(dev);
8072 netif_addr_unlock_bh(dev);
8073 }
8074
8075 /**
8076 * dev_get_flags - get flags reported to userspace
8077 * @dev: device
8078 *
8079 * Get the combination of flag bits exported through APIs to userspace.
8080 */
8081 unsigned int dev_get_flags(const struct net_device *dev)
8082 {
8083 unsigned int flags;
8084
8085 flags = (dev->flags & ~(IFF_PROMISC |
8086 IFF_ALLMULTI |
8087 IFF_RUNNING |
8088 IFF_LOWER_UP |
8089 IFF_DORMANT)) |
8090 (dev->gflags & (IFF_PROMISC |
8091 IFF_ALLMULTI));
8092
8093 if (netif_running(dev)) {
8094 if (netif_oper_up(dev))
8095 flags |= IFF_RUNNING;
8096 if (netif_carrier_ok(dev))
8097 flags |= IFF_LOWER_UP;
8098 if (netif_dormant(dev))
8099 flags |= IFF_DORMANT;
8100 }
8101
8102 return flags;
8103 }
8104 EXPORT_SYMBOL(dev_get_flags);
8105
8106 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8107 struct netlink_ext_ack *extack)
8108 {
8109 unsigned int old_flags = dev->flags;
8110 int ret;
8111
8112 ASSERT_RTNL();
8113
8114 /*
8115 * Set the flags on our device.
8116 */
8117
8118 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8119 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8120 IFF_AUTOMEDIA)) |
8121 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8122 IFF_ALLMULTI));
8123
8124 /*
8125 * Load in the correct multicast list now the flags have changed.
8126 */
8127
8128 if ((old_flags ^ flags) & IFF_MULTICAST)
8129 dev_change_rx_flags(dev, IFF_MULTICAST);
8130
8131 dev_set_rx_mode(dev);
8132
8133 /*
8134 * Have we downed the interface. We handle IFF_UP ourselves
8135 * according to user attempts to set it, rather than blindly
8136 * setting it.
8137 */
8138
8139 ret = 0;
8140 if ((old_flags ^ flags) & IFF_UP) {
8141 if (old_flags & IFF_UP)
8142 __dev_close(dev);
8143 else
8144 ret = __dev_open(dev, extack);
8145 }
8146
8147 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8148 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8149 unsigned int old_flags = dev->flags;
8150
8151 dev->gflags ^= IFF_PROMISC;
8152
8153 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8154 if (dev->flags != old_flags)
8155 dev_set_rx_mode(dev);
8156 }
8157
8158 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8159 * is important. Some (broken) drivers set IFF_PROMISC, when
8160 * IFF_ALLMULTI is requested not asking us and not reporting.
8161 */
8162 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8163 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8164
8165 dev->gflags ^= IFF_ALLMULTI;
8166 __dev_set_allmulti(dev, inc, false);
8167 }
8168
8169 return ret;
8170 }
8171
8172 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8173 unsigned int gchanges)
8174 {
8175 unsigned int changes = dev->flags ^ old_flags;
8176
8177 if (gchanges)
8178 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8179
8180 if (changes & IFF_UP) {
8181 if (dev->flags & IFF_UP)
8182 call_netdevice_notifiers(NETDEV_UP, dev);
8183 else
8184 call_netdevice_notifiers(NETDEV_DOWN, dev);
8185 }
8186
8187 if (dev->flags & IFF_UP &&
8188 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8189 struct netdev_notifier_change_info change_info = {
8190 .info = {
8191 .dev = dev,
8192 },
8193 .flags_changed = changes,
8194 };
8195
8196 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8197 }
8198 }
8199
8200 /**
8201 * dev_change_flags - change device settings
8202 * @dev: device
8203 * @flags: device state flags
8204 * @extack: netlink extended ack
8205 *
8206 * Change settings on device based state flags. The flags are
8207 * in the userspace exported format.
8208 */
8209 int dev_change_flags(struct net_device *dev, unsigned int flags,
8210 struct netlink_ext_ack *extack)
8211 {
8212 int ret;
8213 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8214
8215 ret = __dev_change_flags(dev, flags, extack);
8216 if (ret < 0)
8217 return ret;
8218
8219 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8220 __dev_notify_flags(dev, old_flags, changes);
8221 return ret;
8222 }
8223 EXPORT_SYMBOL(dev_change_flags);
8224
8225 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8226 {
8227 const struct net_device_ops *ops = dev->netdev_ops;
8228
8229 if (ops->ndo_change_mtu)
8230 return ops->ndo_change_mtu(dev, new_mtu);
8231
8232 /* Pairs with all the lockless reads of dev->mtu in the stack */
8233 WRITE_ONCE(dev->mtu, new_mtu);
8234 return 0;
8235 }
8236 EXPORT_SYMBOL(__dev_set_mtu);
8237
8238 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8239 struct netlink_ext_ack *extack)
8240 {
8241 /* MTU must be positive, and in range */
8242 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8243 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8244 return -EINVAL;
8245 }
8246
8247 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8248 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8249 return -EINVAL;
8250 }
8251 return 0;
8252 }
8253
8254 /**
8255 * dev_set_mtu_ext - Change maximum transfer unit
8256 * @dev: device
8257 * @new_mtu: new transfer unit
8258 * @extack: netlink extended ack
8259 *
8260 * Change the maximum transfer size of the network device.
8261 */
8262 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8263 struct netlink_ext_ack *extack)
8264 {
8265 int err, orig_mtu;
8266
8267 if (new_mtu == dev->mtu)
8268 return 0;
8269
8270 err = dev_validate_mtu(dev, new_mtu, extack);
8271 if (err)
8272 return err;
8273
8274 if (!netif_device_present(dev))
8275 return -ENODEV;
8276
8277 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8278 err = notifier_to_errno(err);
8279 if (err)
8280 return err;
8281
8282 orig_mtu = dev->mtu;
8283 err = __dev_set_mtu(dev, new_mtu);
8284
8285 if (!err) {
8286 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8287 orig_mtu);
8288 err = notifier_to_errno(err);
8289 if (err) {
8290 /* setting mtu back and notifying everyone again,
8291 * so that they have a chance to revert changes.
8292 */
8293 __dev_set_mtu(dev, orig_mtu);
8294 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8295 new_mtu);
8296 }
8297 }
8298 return err;
8299 }
8300
8301 int dev_set_mtu(struct net_device *dev, int new_mtu)
8302 {
8303 struct netlink_ext_ack extack;
8304 int err;
8305
8306 memset(&extack, 0, sizeof(extack));
8307 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8308 if (err && extack._msg)
8309 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8310 return err;
8311 }
8312 EXPORT_SYMBOL(dev_set_mtu);
8313
8314 /**
8315 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8316 * @dev: device
8317 * @new_len: new tx queue length
8318 */
8319 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8320 {
8321 unsigned int orig_len = dev->tx_queue_len;
8322 int res;
8323
8324 if (new_len != (unsigned int)new_len)
8325 return -ERANGE;
8326
8327 if (new_len != orig_len) {
8328 dev->tx_queue_len = new_len;
8329 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8330 res = notifier_to_errno(res);
8331 if (res)
8332 goto err_rollback;
8333 res = dev_qdisc_change_tx_queue_len(dev);
8334 if (res)
8335 goto err_rollback;
8336 }
8337
8338 return 0;
8339
8340 err_rollback:
8341 netdev_err(dev, "refused to change device tx_queue_len\n");
8342 dev->tx_queue_len = orig_len;
8343 return res;
8344 }
8345
8346 /**
8347 * dev_set_group - Change group this device belongs to
8348 * @dev: device
8349 * @new_group: group this device should belong to
8350 */
8351 void dev_set_group(struct net_device *dev, int new_group)
8352 {
8353 dev->group = new_group;
8354 }
8355 EXPORT_SYMBOL(dev_set_group);
8356
8357 /**
8358 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8359 * @dev: device
8360 * @addr: new address
8361 * @extack: netlink extended ack
8362 */
8363 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8364 struct netlink_ext_ack *extack)
8365 {
8366 struct netdev_notifier_pre_changeaddr_info info = {
8367 .info.dev = dev,
8368 .info.extack = extack,
8369 .dev_addr = addr,
8370 };
8371 int rc;
8372
8373 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8374 return notifier_to_errno(rc);
8375 }
8376 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8377
8378 /**
8379 * dev_set_mac_address - Change Media Access Control Address
8380 * @dev: device
8381 * @sa: new address
8382 * @extack: netlink extended ack
8383 *
8384 * Change the hardware (MAC) address of the device
8385 */
8386 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8387 struct netlink_ext_ack *extack)
8388 {
8389 const struct net_device_ops *ops = dev->netdev_ops;
8390 int err;
8391
8392 if (!ops->ndo_set_mac_address)
8393 return -EOPNOTSUPP;
8394 if (sa->sa_family != dev->type)
8395 return -EINVAL;
8396 if (!netif_device_present(dev))
8397 return -ENODEV;
8398 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8399 if (err)
8400 return err;
8401 err = ops->ndo_set_mac_address(dev, sa);
8402 if (err)
8403 return err;
8404 dev->addr_assign_type = NET_ADDR_SET;
8405 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8406 add_device_randomness(dev->dev_addr, dev->addr_len);
8407 return 0;
8408 }
8409 EXPORT_SYMBOL(dev_set_mac_address);
8410
8411 /**
8412 * dev_change_carrier - Change device carrier
8413 * @dev: device
8414 * @new_carrier: new value
8415 *
8416 * Change device carrier
8417 */
8418 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8419 {
8420 const struct net_device_ops *ops = dev->netdev_ops;
8421
8422 if (!ops->ndo_change_carrier)
8423 return -EOPNOTSUPP;
8424 if (!netif_device_present(dev))
8425 return -ENODEV;
8426 return ops->ndo_change_carrier(dev, new_carrier);
8427 }
8428 EXPORT_SYMBOL(dev_change_carrier);
8429
8430 /**
8431 * dev_get_phys_port_id - Get device physical port ID
8432 * @dev: device
8433 * @ppid: port ID
8434 *
8435 * Get device physical port ID
8436 */
8437 int dev_get_phys_port_id(struct net_device *dev,
8438 struct netdev_phys_item_id *ppid)
8439 {
8440 const struct net_device_ops *ops = dev->netdev_ops;
8441
8442 if (!ops->ndo_get_phys_port_id)
8443 return -EOPNOTSUPP;
8444 return ops->ndo_get_phys_port_id(dev, ppid);
8445 }
8446 EXPORT_SYMBOL(dev_get_phys_port_id);
8447
8448 /**
8449 * dev_get_phys_port_name - Get device physical port name
8450 * @dev: device
8451 * @name: port name
8452 * @len: limit of bytes to copy to name
8453 *
8454 * Get device physical port name
8455 */
8456 int dev_get_phys_port_name(struct net_device *dev,
8457 char *name, size_t len)
8458 {
8459 const struct net_device_ops *ops = dev->netdev_ops;
8460 int err;
8461
8462 if (ops->ndo_get_phys_port_name) {
8463 err = ops->ndo_get_phys_port_name(dev, name, len);
8464 if (err != -EOPNOTSUPP)
8465 return err;
8466 }
8467 return devlink_compat_phys_port_name_get(dev, name, len);
8468 }
8469 EXPORT_SYMBOL(dev_get_phys_port_name);
8470
8471 /**
8472 * dev_get_port_parent_id - Get the device's port parent identifier
8473 * @dev: network device
8474 * @ppid: pointer to a storage for the port's parent identifier
8475 * @recurse: allow/disallow recursion to lower devices
8476 *
8477 * Get the devices's port parent identifier
8478 */
8479 int dev_get_port_parent_id(struct net_device *dev,
8480 struct netdev_phys_item_id *ppid,
8481 bool recurse)
8482 {
8483 const struct net_device_ops *ops = dev->netdev_ops;
8484 struct netdev_phys_item_id first = { };
8485 struct net_device *lower_dev;
8486 struct list_head *iter;
8487 int err;
8488
8489 if (ops->ndo_get_port_parent_id) {
8490 err = ops->ndo_get_port_parent_id(dev, ppid);
8491 if (err != -EOPNOTSUPP)
8492 return err;
8493 }
8494
8495 err = devlink_compat_switch_id_get(dev, ppid);
8496 if (!err || err != -EOPNOTSUPP)
8497 return err;
8498
8499 if (!recurse)
8500 return -EOPNOTSUPP;
8501
8502 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8503 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8504 if (err)
8505 break;
8506 if (!first.id_len)
8507 first = *ppid;
8508 else if (memcmp(&first, ppid, sizeof(*ppid)))
8509 return -ENODATA;
8510 }
8511
8512 return err;
8513 }
8514 EXPORT_SYMBOL(dev_get_port_parent_id);
8515
8516 /**
8517 * netdev_port_same_parent_id - Indicate if two network devices have
8518 * the same port parent identifier
8519 * @a: first network device
8520 * @b: second network device
8521 */
8522 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8523 {
8524 struct netdev_phys_item_id a_id = { };
8525 struct netdev_phys_item_id b_id = { };
8526
8527 if (dev_get_port_parent_id(a, &a_id, true) ||
8528 dev_get_port_parent_id(b, &b_id, true))
8529 return false;
8530
8531 return netdev_phys_item_id_same(&a_id, &b_id);
8532 }
8533 EXPORT_SYMBOL(netdev_port_same_parent_id);
8534
8535 /**
8536 * dev_change_proto_down - update protocol port state information
8537 * @dev: device
8538 * @proto_down: new value
8539 *
8540 * This info can be used by switch drivers to set the phys state of the
8541 * port.
8542 */
8543 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8544 {
8545 const struct net_device_ops *ops = dev->netdev_ops;
8546
8547 if (!ops->ndo_change_proto_down)
8548 return -EOPNOTSUPP;
8549 if (!netif_device_present(dev))
8550 return -ENODEV;
8551 return ops->ndo_change_proto_down(dev, proto_down);
8552 }
8553 EXPORT_SYMBOL(dev_change_proto_down);
8554
8555 /**
8556 * dev_change_proto_down_generic - generic implementation for
8557 * ndo_change_proto_down that sets carrier according to
8558 * proto_down.
8559 *
8560 * @dev: device
8561 * @proto_down: new value
8562 */
8563 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8564 {
8565 if (proto_down)
8566 netif_carrier_off(dev);
8567 else
8568 netif_carrier_on(dev);
8569 dev->proto_down = proto_down;
8570 return 0;
8571 }
8572 EXPORT_SYMBOL(dev_change_proto_down_generic);
8573
8574 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8575 enum bpf_netdev_command cmd)
8576 {
8577 struct netdev_bpf xdp;
8578
8579 if (!bpf_op)
8580 return 0;
8581
8582 memset(&xdp, 0, sizeof(xdp));
8583 xdp.command = cmd;
8584
8585 /* Query must always succeed. */
8586 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8587
8588 return xdp.prog_id;
8589 }
8590
8591 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8592 struct netlink_ext_ack *extack, u32 flags,
8593 struct bpf_prog *prog)
8594 {
8595 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8596 struct bpf_prog *prev_prog = NULL;
8597 struct netdev_bpf xdp;
8598 int err;
8599
8600 if (non_hw) {
8601 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8602 XDP_QUERY_PROG));
8603 if (IS_ERR(prev_prog))
8604 prev_prog = NULL;
8605 }
8606
8607 memset(&xdp, 0, sizeof(xdp));
8608 if (flags & XDP_FLAGS_HW_MODE)
8609 xdp.command = XDP_SETUP_PROG_HW;
8610 else
8611 xdp.command = XDP_SETUP_PROG;
8612 xdp.extack = extack;
8613 xdp.flags = flags;
8614 xdp.prog = prog;
8615
8616 err = bpf_op(dev, &xdp);
8617 if (!err && non_hw)
8618 bpf_prog_change_xdp(prev_prog, prog);
8619
8620 if (prev_prog)
8621 bpf_prog_put(prev_prog);
8622
8623 return err;
8624 }
8625
8626 static void dev_xdp_uninstall(struct net_device *dev)
8627 {
8628 struct netdev_bpf xdp;
8629 bpf_op_t ndo_bpf;
8630
8631 /* Remove generic XDP */
8632 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8633
8634 /* Remove from the driver */
8635 ndo_bpf = dev->netdev_ops->ndo_bpf;
8636 if (!ndo_bpf)
8637 return;
8638
8639 memset(&xdp, 0, sizeof(xdp));
8640 xdp.command = XDP_QUERY_PROG;
8641 WARN_ON(ndo_bpf(dev, &xdp));
8642 if (xdp.prog_id)
8643 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8644 NULL));
8645
8646 /* Remove HW offload */
8647 memset(&xdp, 0, sizeof(xdp));
8648 xdp.command = XDP_QUERY_PROG_HW;
8649 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8650 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8651 NULL));
8652 }
8653
8654 /**
8655 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8656 * @dev: device
8657 * @extack: netlink extended ack
8658 * @fd: new program fd or negative value to clear
8659 * @expected_fd: old program fd that userspace expects to replace or clear
8660 * @flags: xdp-related flags
8661 *
8662 * Set or clear a bpf program for a device
8663 */
8664 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8665 int fd, int expected_fd, u32 flags)
8666 {
8667 const struct net_device_ops *ops = dev->netdev_ops;
8668 enum bpf_netdev_command query;
8669 u32 prog_id, expected_id = 0;
8670 bpf_op_t bpf_op, bpf_chk;
8671 struct bpf_prog *prog;
8672 bool offload;
8673 int err;
8674
8675 ASSERT_RTNL();
8676
8677 offload = flags & XDP_FLAGS_HW_MODE;
8678 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8679
8680 bpf_op = bpf_chk = ops->ndo_bpf;
8681 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8682 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8683 return -EOPNOTSUPP;
8684 }
8685 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8686 bpf_op = generic_xdp_install;
8687 if (bpf_op == bpf_chk)
8688 bpf_chk = generic_xdp_install;
8689
8690 prog_id = __dev_xdp_query(dev, bpf_op, query);
8691 if (flags & XDP_FLAGS_REPLACE) {
8692 if (expected_fd >= 0) {
8693 prog = bpf_prog_get_type_dev(expected_fd,
8694 BPF_PROG_TYPE_XDP,
8695 bpf_op == ops->ndo_bpf);
8696 if (IS_ERR(prog))
8697 return PTR_ERR(prog);
8698 expected_id = prog->aux->id;
8699 bpf_prog_put(prog);
8700 }
8701
8702 if (prog_id != expected_id) {
8703 NL_SET_ERR_MSG(extack, "Active program does not match expected");
8704 return -EEXIST;
8705 }
8706 }
8707 if (fd >= 0) {
8708 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8709 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8710 return -EEXIST;
8711 }
8712
8713 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8714 NL_SET_ERR_MSG(extack, "XDP program already attached");
8715 return -EBUSY;
8716 }
8717
8718 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8719 bpf_op == ops->ndo_bpf);
8720 if (IS_ERR(prog))
8721 return PTR_ERR(prog);
8722
8723 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8724 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8725 bpf_prog_put(prog);
8726 return -EINVAL;
8727 }
8728
8729 /* prog->aux->id may be 0 for orphaned device-bound progs */
8730 if (prog->aux->id && prog->aux->id == prog_id) {
8731 bpf_prog_put(prog);
8732 return 0;
8733 }
8734 } else {
8735 if (!prog_id)
8736 return 0;
8737 prog = NULL;
8738 }
8739
8740 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8741 if (err < 0 && prog)
8742 bpf_prog_put(prog);
8743
8744 return err;
8745 }
8746
8747 /**
8748 * dev_new_index - allocate an ifindex
8749 * @net: the applicable net namespace
8750 *
8751 * Returns a suitable unique value for a new device interface
8752 * number. The caller must hold the rtnl semaphore or the
8753 * dev_base_lock to be sure it remains unique.
8754 */
8755 static int dev_new_index(struct net *net)
8756 {
8757 int ifindex = net->ifindex;
8758
8759 for (;;) {
8760 if (++ifindex <= 0)
8761 ifindex = 1;
8762 if (!__dev_get_by_index(net, ifindex))
8763 return net->ifindex = ifindex;
8764 }
8765 }
8766
8767 /* Delayed registration/unregisteration */
8768 static LIST_HEAD(net_todo_list);
8769 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8770
8771 static void net_set_todo(struct net_device *dev)
8772 {
8773 list_add_tail(&dev->todo_list, &net_todo_list);
8774 dev_net(dev)->dev_unreg_count++;
8775 }
8776
8777 static void rollback_registered_many(struct list_head *head)
8778 {
8779 struct net_device *dev, *tmp;
8780 LIST_HEAD(close_head);
8781
8782 BUG_ON(dev_boot_phase);
8783 ASSERT_RTNL();
8784
8785 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8786 /* Some devices call without registering
8787 * for initialization unwind. Remove those
8788 * devices and proceed with the remaining.
8789 */
8790 if (dev->reg_state == NETREG_UNINITIALIZED) {
8791 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8792 dev->name, dev);
8793
8794 WARN_ON(1);
8795 list_del(&dev->unreg_list);
8796 continue;
8797 }
8798 dev->dismantle = true;
8799 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8800 }
8801
8802 /* If device is running, close it first. */
8803 list_for_each_entry(dev, head, unreg_list)
8804 list_add_tail(&dev->close_list, &close_head);
8805 dev_close_many(&close_head, true);
8806
8807 list_for_each_entry(dev, head, unreg_list) {
8808 /* And unlink it from device chain. */
8809 unlist_netdevice(dev);
8810
8811 dev->reg_state = NETREG_UNREGISTERING;
8812 }
8813 flush_all_backlogs();
8814
8815 synchronize_net();
8816
8817 list_for_each_entry(dev, head, unreg_list) {
8818 struct sk_buff *skb = NULL;
8819
8820 /* Shutdown queueing discipline. */
8821 dev_shutdown(dev);
8822
8823 dev_xdp_uninstall(dev);
8824
8825 /* Notify protocols, that we are about to destroy
8826 * this device. They should clean all the things.
8827 */
8828 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8829
8830 if (!dev->rtnl_link_ops ||
8831 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8832 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8833 GFP_KERNEL, NULL, 0);
8834
8835 /*
8836 * Flush the unicast and multicast chains
8837 */
8838 dev_uc_flush(dev);
8839 dev_mc_flush(dev);
8840
8841 netdev_name_node_alt_flush(dev);
8842 netdev_name_node_free(dev->name_node);
8843
8844 if (dev->netdev_ops->ndo_uninit)
8845 dev->netdev_ops->ndo_uninit(dev);
8846
8847 if (skb)
8848 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8849
8850 /* Notifier chain MUST detach us all upper devices. */
8851 WARN_ON(netdev_has_any_upper_dev(dev));
8852 WARN_ON(netdev_has_any_lower_dev(dev));
8853
8854 /* Remove entries from kobject tree */
8855 netdev_unregister_kobject(dev);
8856 #ifdef CONFIG_XPS
8857 /* Remove XPS queueing entries */
8858 netif_reset_xps_queues_gt(dev, 0);
8859 #endif
8860 }
8861
8862 synchronize_net();
8863
8864 list_for_each_entry(dev, head, unreg_list)
8865 dev_put(dev);
8866 }
8867
8868 static void rollback_registered(struct net_device *dev)
8869 {
8870 LIST_HEAD(single);
8871
8872 list_add(&dev->unreg_list, &single);
8873 rollback_registered_many(&single);
8874 list_del(&single);
8875 }
8876
8877 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8878 struct net_device *upper, netdev_features_t features)
8879 {
8880 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8881 netdev_features_t feature;
8882 int feature_bit;
8883
8884 for_each_netdev_feature(upper_disables, feature_bit) {
8885 feature = __NETIF_F_BIT(feature_bit);
8886 if (!(upper->wanted_features & feature)
8887 && (features & feature)) {
8888 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8889 &feature, upper->name);
8890 features &= ~feature;
8891 }
8892 }
8893
8894 return features;
8895 }
8896
8897 static void netdev_sync_lower_features(struct net_device *upper,
8898 struct net_device *lower, netdev_features_t features)
8899 {
8900 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8901 netdev_features_t feature;
8902 int feature_bit;
8903
8904 for_each_netdev_feature(upper_disables, feature_bit) {
8905 feature = __NETIF_F_BIT(feature_bit);
8906 if (!(features & feature) && (lower->features & feature)) {
8907 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8908 &feature, lower->name);
8909 lower->wanted_features &= ~feature;
8910 __netdev_update_features(lower);
8911
8912 if (unlikely(lower->features & feature))
8913 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8914 &feature, lower->name);
8915 else
8916 netdev_features_change(lower);
8917 }
8918 }
8919 }
8920
8921 static netdev_features_t netdev_fix_features(struct net_device *dev,
8922 netdev_features_t features)
8923 {
8924 /* Fix illegal checksum combinations */
8925 if ((features & NETIF_F_HW_CSUM) &&
8926 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8927 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8928 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8929 }
8930
8931 /* TSO requires that SG is present as well. */
8932 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8933 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8934 features &= ~NETIF_F_ALL_TSO;
8935 }
8936
8937 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8938 !(features & NETIF_F_IP_CSUM)) {
8939 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8940 features &= ~NETIF_F_TSO;
8941 features &= ~NETIF_F_TSO_ECN;
8942 }
8943
8944 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8945 !(features & NETIF_F_IPV6_CSUM)) {
8946 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8947 features &= ~NETIF_F_TSO6;
8948 }
8949
8950 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8951 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8952 features &= ~NETIF_F_TSO_MANGLEID;
8953
8954 /* TSO ECN requires that TSO is present as well. */
8955 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8956 features &= ~NETIF_F_TSO_ECN;
8957
8958 /* Software GSO depends on SG. */
8959 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8960 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8961 features &= ~NETIF_F_GSO;
8962 }
8963
8964 /* GSO partial features require GSO partial be set */
8965 if ((features & dev->gso_partial_features) &&
8966 !(features & NETIF_F_GSO_PARTIAL)) {
8967 netdev_dbg(dev,
8968 "Dropping partially supported GSO features since no GSO partial.\n");
8969 features &= ~dev->gso_partial_features;
8970 }
8971
8972 if (!(features & NETIF_F_RXCSUM)) {
8973 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8974 * successfully merged by hardware must also have the
8975 * checksum verified by hardware. If the user does not
8976 * want to enable RXCSUM, logically, we should disable GRO_HW.
8977 */
8978 if (features & NETIF_F_GRO_HW) {
8979 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8980 features &= ~NETIF_F_GRO_HW;
8981 }
8982 }
8983
8984 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8985 if (features & NETIF_F_RXFCS) {
8986 if (features & NETIF_F_LRO) {
8987 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8988 features &= ~NETIF_F_LRO;
8989 }
8990
8991 if (features & NETIF_F_GRO_HW) {
8992 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8993 features &= ~NETIF_F_GRO_HW;
8994 }
8995 }
8996
8997 return features;
8998 }
8999
9000 int __netdev_update_features(struct net_device *dev)
9001 {
9002 struct net_device *upper, *lower;
9003 netdev_features_t features;
9004 struct list_head *iter;
9005 int err = -1;
9006
9007 ASSERT_RTNL();
9008
9009 features = netdev_get_wanted_features(dev);
9010
9011 if (dev->netdev_ops->ndo_fix_features)
9012 features = dev->netdev_ops->ndo_fix_features(dev, features);
9013
9014 /* driver might be less strict about feature dependencies */
9015 features = netdev_fix_features(dev, features);
9016
9017 /* some features can't be enabled if they're off an an upper device */
9018 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9019 features = netdev_sync_upper_features(dev, upper, features);
9020
9021 if (dev->features == features)
9022 goto sync_lower;
9023
9024 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9025 &dev->features, &features);
9026
9027 if (dev->netdev_ops->ndo_set_features)
9028 err = dev->netdev_ops->ndo_set_features(dev, features);
9029 else
9030 err = 0;
9031
9032 if (unlikely(err < 0)) {
9033 netdev_err(dev,
9034 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9035 err, &features, &dev->features);
9036 /* return non-0 since some features might have changed and
9037 * it's better to fire a spurious notification than miss it
9038 */
9039 return -1;
9040 }
9041
9042 sync_lower:
9043 /* some features must be disabled on lower devices when disabled
9044 * on an upper device (think: bonding master or bridge)
9045 */
9046 netdev_for_each_lower_dev(dev, lower, iter)
9047 netdev_sync_lower_features(dev, lower, features);
9048
9049 if (!err) {
9050 netdev_features_t diff = features ^ dev->features;
9051
9052 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9053 /* udp_tunnel_{get,drop}_rx_info both need
9054 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9055 * device, or they won't do anything.
9056 * Thus we need to update dev->features
9057 * *before* calling udp_tunnel_get_rx_info,
9058 * but *after* calling udp_tunnel_drop_rx_info.
9059 */
9060 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9061 dev->features = features;
9062 udp_tunnel_get_rx_info(dev);
9063 } else {
9064 udp_tunnel_drop_rx_info(dev);
9065 }
9066 }
9067
9068 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9069 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9070 dev->features = features;
9071 err |= vlan_get_rx_ctag_filter_info(dev);
9072 } else {
9073 vlan_drop_rx_ctag_filter_info(dev);
9074 }
9075 }
9076
9077 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9078 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9079 dev->features = features;
9080 err |= vlan_get_rx_stag_filter_info(dev);
9081 } else {
9082 vlan_drop_rx_stag_filter_info(dev);
9083 }
9084 }
9085
9086 dev->features = features;
9087 }
9088
9089 return err < 0 ? 0 : 1;
9090 }
9091
9092 /**
9093 * netdev_update_features - recalculate device features
9094 * @dev: the device to check
9095 *
9096 * Recalculate dev->features set and send notifications if it
9097 * has changed. Should be called after driver or hardware dependent
9098 * conditions might have changed that influence the features.
9099 */
9100 void netdev_update_features(struct net_device *dev)
9101 {
9102 if (__netdev_update_features(dev))
9103 netdev_features_change(dev);
9104 }
9105 EXPORT_SYMBOL(netdev_update_features);
9106
9107 /**
9108 * netdev_change_features - recalculate device features
9109 * @dev: the device to check
9110 *
9111 * Recalculate dev->features set and send notifications even
9112 * if they have not changed. Should be called instead of
9113 * netdev_update_features() if also dev->vlan_features might
9114 * have changed to allow the changes to be propagated to stacked
9115 * VLAN devices.
9116 */
9117 void netdev_change_features(struct net_device *dev)
9118 {
9119 __netdev_update_features(dev);
9120 netdev_features_change(dev);
9121 }
9122 EXPORT_SYMBOL(netdev_change_features);
9123
9124 /**
9125 * netif_stacked_transfer_operstate - transfer operstate
9126 * @rootdev: the root or lower level device to transfer state from
9127 * @dev: the device to transfer operstate to
9128 *
9129 * Transfer operational state from root to device. This is normally
9130 * called when a stacking relationship exists between the root
9131 * device and the device(a leaf device).
9132 */
9133 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9134 struct net_device *dev)
9135 {
9136 if (rootdev->operstate == IF_OPER_DORMANT)
9137 netif_dormant_on(dev);
9138 else
9139 netif_dormant_off(dev);
9140
9141 if (netif_carrier_ok(rootdev))
9142 netif_carrier_on(dev);
9143 else
9144 netif_carrier_off(dev);
9145 }
9146 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9147
9148 static int netif_alloc_rx_queues(struct net_device *dev)
9149 {
9150 unsigned int i, count = dev->num_rx_queues;
9151 struct netdev_rx_queue *rx;
9152 size_t sz = count * sizeof(*rx);
9153 int err = 0;
9154
9155 BUG_ON(count < 1);
9156
9157 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9158 if (!rx)
9159 return -ENOMEM;
9160
9161 dev->_rx = rx;
9162
9163 for (i = 0; i < count; i++) {
9164 rx[i].dev = dev;
9165
9166 /* XDP RX-queue setup */
9167 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9168 if (err < 0)
9169 goto err_rxq_info;
9170 }
9171 return 0;
9172
9173 err_rxq_info:
9174 /* Rollback successful reg's and free other resources */
9175 while (i--)
9176 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9177 kvfree(dev->_rx);
9178 dev->_rx = NULL;
9179 return err;
9180 }
9181
9182 static void netif_free_rx_queues(struct net_device *dev)
9183 {
9184 unsigned int i, count = dev->num_rx_queues;
9185
9186 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9187 if (!dev->_rx)
9188 return;
9189
9190 for (i = 0; i < count; i++)
9191 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9192
9193 kvfree(dev->_rx);
9194 }
9195
9196 static void netdev_init_one_queue(struct net_device *dev,
9197 struct netdev_queue *queue, void *_unused)
9198 {
9199 /* Initialize queue lock */
9200 spin_lock_init(&queue->_xmit_lock);
9201 lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key);
9202 queue->xmit_lock_owner = -1;
9203 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9204 queue->dev = dev;
9205 #ifdef CONFIG_BQL
9206 dql_init(&queue->dql, HZ);
9207 #endif
9208 }
9209
9210 static void netif_free_tx_queues(struct net_device *dev)
9211 {
9212 kvfree(dev->_tx);
9213 }
9214
9215 static int netif_alloc_netdev_queues(struct net_device *dev)
9216 {
9217 unsigned int count = dev->num_tx_queues;
9218 struct netdev_queue *tx;
9219 size_t sz = count * sizeof(*tx);
9220
9221 if (count < 1 || count > 0xffff)
9222 return -EINVAL;
9223
9224 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9225 if (!tx)
9226 return -ENOMEM;
9227
9228 dev->_tx = tx;
9229
9230 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9231 spin_lock_init(&dev->tx_global_lock);
9232
9233 return 0;
9234 }
9235
9236 void netif_tx_stop_all_queues(struct net_device *dev)
9237 {
9238 unsigned int i;
9239
9240 for (i = 0; i < dev->num_tx_queues; i++) {
9241 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9242
9243 netif_tx_stop_queue(txq);
9244 }
9245 }
9246 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9247
9248 static void netdev_register_lockdep_key(struct net_device *dev)
9249 {
9250 lockdep_register_key(&dev->qdisc_tx_busylock_key);
9251 lockdep_register_key(&dev->qdisc_running_key);
9252 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9253 lockdep_register_key(&dev->addr_list_lock_key);
9254 }
9255
9256 static void netdev_unregister_lockdep_key(struct net_device *dev)
9257 {
9258 lockdep_unregister_key(&dev->qdisc_tx_busylock_key);
9259 lockdep_unregister_key(&dev->qdisc_running_key);
9260 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9261 lockdep_unregister_key(&dev->addr_list_lock_key);
9262 }
9263
9264 void netdev_update_lockdep_key(struct net_device *dev)
9265 {
9266 lockdep_unregister_key(&dev->addr_list_lock_key);
9267 lockdep_register_key(&dev->addr_list_lock_key);
9268
9269 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9270 }
9271 EXPORT_SYMBOL(netdev_update_lockdep_key);
9272
9273 /**
9274 * register_netdevice - register a network device
9275 * @dev: device to register
9276 *
9277 * Take a completed network device structure and add it to the kernel
9278 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9279 * chain. 0 is returned on success. A negative errno code is returned
9280 * on a failure to set up the device, or if the name is a duplicate.
9281 *
9282 * Callers must hold the rtnl semaphore. You may want
9283 * register_netdev() instead of this.
9284 *
9285 * BUGS:
9286 * The locking appears insufficient to guarantee two parallel registers
9287 * will not get the same name.
9288 */
9289
9290 int register_netdevice(struct net_device *dev)
9291 {
9292 int ret;
9293 struct net *net = dev_net(dev);
9294
9295 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9296 NETDEV_FEATURE_COUNT);
9297 BUG_ON(dev_boot_phase);
9298 ASSERT_RTNL();
9299
9300 might_sleep();
9301
9302 /* When net_device's are persistent, this will be fatal. */
9303 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9304 BUG_ON(!net);
9305
9306 ret = ethtool_check_ops(dev->ethtool_ops);
9307 if (ret)
9308 return ret;
9309
9310 spin_lock_init(&dev->addr_list_lock);
9311 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9312
9313 ret = dev_get_valid_name(net, dev, dev->name);
9314 if (ret < 0)
9315 goto out;
9316
9317 ret = -ENOMEM;
9318 dev->name_node = netdev_name_node_head_alloc(dev);
9319 if (!dev->name_node)
9320 goto out;
9321
9322 /* Init, if this function is available */
9323 if (dev->netdev_ops->ndo_init) {
9324 ret = dev->netdev_ops->ndo_init(dev);
9325 if (ret) {
9326 if (ret > 0)
9327 ret = -EIO;
9328 goto err_free_name;
9329 }
9330 }
9331
9332 if (((dev->hw_features | dev->features) &
9333 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9334 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9335 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9336 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9337 ret = -EINVAL;
9338 goto err_uninit;
9339 }
9340
9341 ret = -EBUSY;
9342 if (!dev->ifindex)
9343 dev->ifindex = dev_new_index(net);
9344 else if (__dev_get_by_index(net, dev->ifindex))
9345 goto err_uninit;
9346
9347 /* Transfer changeable features to wanted_features and enable
9348 * software offloads (GSO and GRO).
9349 */
9350 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9351 dev->features |= NETIF_F_SOFT_FEATURES;
9352
9353 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9354 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9355 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9356 }
9357
9358 dev->wanted_features = dev->features & dev->hw_features;
9359
9360 if (!(dev->flags & IFF_LOOPBACK))
9361 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9362
9363 /* If IPv4 TCP segmentation offload is supported we should also
9364 * allow the device to enable segmenting the frame with the option
9365 * of ignoring a static IP ID value. This doesn't enable the
9366 * feature itself but allows the user to enable it later.
9367 */
9368 if (dev->hw_features & NETIF_F_TSO)
9369 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9370 if (dev->vlan_features & NETIF_F_TSO)
9371 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9372 if (dev->mpls_features & NETIF_F_TSO)
9373 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9374 if (dev->hw_enc_features & NETIF_F_TSO)
9375 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9376
9377 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9378 */
9379 dev->vlan_features |= NETIF_F_HIGHDMA;
9380
9381 /* Make NETIF_F_SG inheritable to tunnel devices.
9382 */
9383 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9384
9385 /* Make NETIF_F_SG inheritable to MPLS.
9386 */
9387 dev->mpls_features |= NETIF_F_SG;
9388
9389 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9390 ret = notifier_to_errno(ret);
9391 if (ret)
9392 goto err_uninit;
9393
9394 ret = netdev_register_kobject(dev);
9395 if (ret) {
9396 dev->reg_state = NETREG_UNREGISTERED;
9397 goto err_uninit;
9398 }
9399 dev->reg_state = NETREG_REGISTERED;
9400
9401 __netdev_update_features(dev);
9402
9403 /*
9404 * Default initial state at registry is that the
9405 * device is present.
9406 */
9407
9408 set_bit(__LINK_STATE_PRESENT, &dev->state);
9409
9410 linkwatch_init_dev(dev);
9411
9412 dev_init_scheduler(dev);
9413 dev_hold(dev);
9414 list_netdevice(dev);
9415 add_device_randomness(dev->dev_addr, dev->addr_len);
9416
9417 /* If the device has permanent device address, driver should
9418 * set dev_addr and also addr_assign_type should be set to
9419 * NET_ADDR_PERM (default value).
9420 */
9421 if (dev->addr_assign_type == NET_ADDR_PERM)
9422 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9423
9424 /* Notify protocols, that a new device appeared. */
9425 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9426 ret = notifier_to_errno(ret);
9427 if (ret) {
9428 rollback_registered(dev);
9429 rcu_barrier();
9430
9431 dev->reg_state = NETREG_UNREGISTERED;
9432 }
9433 /*
9434 * Prevent userspace races by waiting until the network
9435 * device is fully setup before sending notifications.
9436 */
9437 if (!dev->rtnl_link_ops ||
9438 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9439 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9440
9441 out:
9442 return ret;
9443
9444 err_uninit:
9445 if (dev->netdev_ops->ndo_uninit)
9446 dev->netdev_ops->ndo_uninit(dev);
9447 if (dev->priv_destructor)
9448 dev->priv_destructor(dev);
9449 err_free_name:
9450 netdev_name_node_free(dev->name_node);
9451 goto out;
9452 }
9453 EXPORT_SYMBOL(register_netdevice);
9454
9455 /**
9456 * init_dummy_netdev - init a dummy network device for NAPI
9457 * @dev: device to init
9458 *
9459 * This takes a network device structure and initialize the minimum
9460 * amount of fields so it can be used to schedule NAPI polls without
9461 * registering a full blown interface. This is to be used by drivers
9462 * that need to tie several hardware interfaces to a single NAPI
9463 * poll scheduler due to HW limitations.
9464 */
9465 int init_dummy_netdev(struct net_device *dev)
9466 {
9467 /* Clear everything. Note we don't initialize spinlocks
9468 * are they aren't supposed to be taken by any of the
9469 * NAPI code and this dummy netdev is supposed to be
9470 * only ever used for NAPI polls
9471 */
9472 memset(dev, 0, sizeof(struct net_device));
9473
9474 /* make sure we BUG if trying to hit standard
9475 * register/unregister code path
9476 */
9477 dev->reg_state = NETREG_DUMMY;
9478
9479 /* NAPI wants this */
9480 INIT_LIST_HEAD(&dev->napi_list);
9481
9482 /* a dummy interface is started by default */
9483 set_bit(__LINK_STATE_PRESENT, &dev->state);
9484 set_bit(__LINK_STATE_START, &dev->state);
9485
9486 /* napi_busy_loop stats accounting wants this */
9487 dev_net_set(dev, &init_net);
9488
9489 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9490 * because users of this 'device' dont need to change
9491 * its refcount.
9492 */
9493
9494 return 0;
9495 }
9496 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9497
9498
9499 /**
9500 * register_netdev - register a network device
9501 * @dev: device to register
9502 *
9503 * Take a completed network device structure and add it to the kernel
9504 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9505 * chain. 0 is returned on success. A negative errno code is returned
9506 * on a failure to set up the device, or if the name is a duplicate.
9507 *
9508 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9509 * and expands the device name if you passed a format string to
9510 * alloc_netdev.
9511 */
9512 int register_netdev(struct net_device *dev)
9513 {
9514 int err;
9515
9516 if (rtnl_lock_killable())
9517 return -EINTR;
9518 err = register_netdevice(dev);
9519 rtnl_unlock();
9520 return err;
9521 }
9522 EXPORT_SYMBOL(register_netdev);
9523
9524 int netdev_refcnt_read(const struct net_device *dev)
9525 {
9526 int i, refcnt = 0;
9527
9528 for_each_possible_cpu(i)
9529 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9530 return refcnt;
9531 }
9532 EXPORT_SYMBOL(netdev_refcnt_read);
9533
9534 /**
9535 * netdev_wait_allrefs - wait until all references are gone.
9536 * @dev: target net_device
9537 *
9538 * This is called when unregistering network devices.
9539 *
9540 * Any protocol or device that holds a reference should register
9541 * for netdevice notification, and cleanup and put back the
9542 * reference if they receive an UNREGISTER event.
9543 * We can get stuck here if buggy protocols don't correctly
9544 * call dev_put.
9545 */
9546 static void netdev_wait_allrefs(struct net_device *dev)
9547 {
9548 unsigned long rebroadcast_time, warning_time;
9549 int refcnt;
9550
9551 linkwatch_forget_dev(dev);
9552
9553 rebroadcast_time = warning_time = jiffies;
9554 refcnt = netdev_refcnt_read(dev);
9555
9556 while (refcnt != 0) {
9557 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9558 rtnl_lock();
9559
9560 /* Rebroadcast unregister notification */
9561 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9562
9563 __rtnl_unlock();
9564 rcu_barrier();
9565 rtnl_lock();
9566
9567 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9568 &dev->state)) {
9569 /* We must not have linkwatch events
9570 * pending on unregister. If this
9571 * happens, we simply run the queue
9572 * unscheduled, resulting in a noop
9573 * for this device.
9574 */
9575 linkwatch_run_queue();
9576 }
9577
9578 __rtnl_unlock();
9579
9580 rebroadcast_time = jiffies;
9581 }
9582
9583 msleep(250);
9584
9585 refcnt = netdev_refcnt_read(dev);
9586
9587 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9588 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9589 dev->name, refcnt);
9590 warning_time = jiffies;
9591 }
9592 }
9593 }
9594
9595 /* The sequence is:
9596 *
9597 * rtnl_lock();
9598 * ...
9599 * register_netdevice(x1);
9600 * register_netdevice(x2);
9601 * ...
9602 * unregister_netdevice(y1);
9603 * unregister_netdevice(y2);
9604 * ...
9605 * rtnl_unlock();
9606 * free_netdev(y1);
9607 * free_netdev(y2);
9608 *
9609 * We are invoked by rtnl_unlock().
9610 * This allows us to deal with problems:
9611 * 1) We can delete sysfs objects which invoke hotplug
9612 * without deadlocking with linkwatch via keventd.
9613 * 2) Since we run with the RTNL semaphore not held, we can sleep
9614 * safely in order to wait for the netdev refcnt to drop to zero.
9615 *
9616 * We must not return until all unregister events added during
9617 * the interval the lock was held have been completed.
9618 */
9619 void netdev_run_todo(void)
9620 {
9621 struct list_head list;
9622
9623 /* Snapshot list, allow later requests */
9624 list_replace_init(&net_todo_list, &list);
9625
9626 __rtnl_unlock();
9627
9628
9629 /* Wait for rcu callbacks to finish before next phase */
9630 if (!list_empty(&list))
9631 rcu_barrier();
9632
9633 while (!list_empty(&list)) {
9634 struct net_device *dev
9635 = list_first_entry(&list, struct net_device, todo_list);
9636 list_del(&dev->todo_list);
9637
9638 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9639 pr_err("network todo '%s' but state %d\n",
9640 dev->name, dev->reg_state);
9641 dump_stack();
9642 continue;
9643 }
9644
9645 dev->reg_state = NETREG_UNREGISTERED;
9646
9647 netdev_wait_allrefs(dev);
9648
9649 /* paranoia */
9650 BUG_ON(netdev_refcnt_read(dev));
9651 BUG_ON(!list_empty(&dev->ptype_all));
9652 BUG_ON(!list_empty(&dev->ptype_specific));
9653 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9654 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9655 #if IS_ENABLED(CONFIG_DECNET)
9656 WARN_ON(dev->dn_ptr);
9657 #endif
9658 if (dev->priv_destructor)
9659 dev->priv_destructor(dev);
9660 if (dev->needs_free_netdev)
9661 free_netdev(dev);
9662
9663 /* Report a network device has been unregistered */
9664 rtnl_lock();
9665 dev_net(dev)->dev_unreg_count--;
9666 __rtnl_unlock();
9667 wake_up(&netdev_unregistering_wq);
9668
9669 /* Free network device */
9670 kobject_put(&dev->dev.kobj);
9671 }
9672 }
9673
9674 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9675 * all the same fields in the same order as net_device_stats, with only
9676 * the type differing, but rtnl_link_stats64 may have additional fields
9677 * at the end for newer counters.
9678 */
9679 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9680 const struct net_device_stats *netdev_stats)
9681 {
9682 #if BITS_PER_LONG == 64
9683 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9684 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9685 /* zero out counters that only exist in rtnl_link_stats64 */
9686 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9687 sizeof(*stats64) - sizeof(*netdev_stats));
9688 #else
9689 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9690 const unsigned long *src = (const unsigned long *)netdev_stats;
9691 u64 *dst = (u64 *)stats64;
9692
9693 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9694 for (i = 0; i < n; i++)
9695 dst[i] = src[i];
9696 /* zero out counters that only exist in rtnl_link_stats64 */
9697 memset((char *)stats64 + n * sizeof(u64), 0,
9698 sizeof(*stats64) - n * sizeof(u64));
9699 #endif
9700 }
9701 EXPORT_SYMBOL(netdev_stats_to_stats64);
9702
9703 /**
9704 * dev_get_stats - get network device statistics
9705 * @dev: device to get statistics from
9706 * @storage: place to store stats
9707 *
9708 * Get network statistics from device. Return @storage.
9709 * The device driver may provide its own method by setting
9710 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9711 * otherwise the internal statistics structure is used.
9712 */
9713 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9714 struct rtnl_link_stats64 *storage)
9715 {
9716 const struct net_device_ops *ops = dev->netdev_ops;
9717
9718 if (ops->ndo_get_stats64) {
9719 memset(storage, 0, sizeof(*storage));
9720 ops->ndo_get_stats64(dev, storage);
9721 } else if (ops->ndo_get_stats) {
9722 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9723 } else {
9724 netdev_stats_to_stats64(storage, &dev->stats);
9725 }
9726 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9727 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9728 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9729 return storage;
9730 }
9731 EXPORT_SYMBOL(dev_get_stats);
9732
9733 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9734 {
9735 struct netdev_queue *queue = dev_ingress_queue(dev);
9736
9737 #ifdef CONFIG_NET_CLS_ACT
9738 if (queue)
9739 return queue;
9740 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9741 if (!queue)
9742 return NULL;
9743 netdev_init_one_queue(dev, queue, NULL);
9744 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9745 queue->qdisc_sleeping = &noop_qdisc;
9746 rcu_assign_pointer(dev->ingress_queue, queue);
9747 #endif
9748 return queue;
9749 }
9750
9751 static const struct ethtool_ops default_ethtool_ops;
9752
9753 void netdev_set_default_ethtool_ops(struct net_device *dev,
9754 const struct ethtool_ops *ops)
9755 {
9756 if (dev->ethtool_ops == &default_ethtool_ops)
9757 dev->ethtool_ops = ops;
9758 }
9759 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9760
9761 void netdev_freemem(struct net_device *dev)
9762 {
9763 char *addr = (char *)dev - dev->padded;
9764
9765 kvfree(addr);
9766 }
9767
9768 /**
9769 * alloc_netdev_mqs - allocate network device
9770 * @sizeof_priv: size of private data to allocate space for
9771 * @name: device name format string
9772 * @name_assign_type: origin of device name
9773 * @setup: callback to initialize device
9774 * @txqs: the number of TX subqueues to allocate
9775 * @rxqs: the number of RX subqueues to allocate
9776 *
9777 * Allocates a struct net_device with private data area for driver use
9778 * and performs basic initialization. Also allocates subqueue structs
9779 * for each queue on the device.
9780 */
9781 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9782 unsigned char name_assign_type,
9783 void (*setup)(struct net_device *),
9784 unsigned int txqs, unsigned int rxqs)
9785 {
9786 struct net_device *dev;
9787 unsigned int alloc_size;
9788 struct net_device *p;
9789
9790 BUG_ON(strlen(name) >= sizeof(dev->name));
9791
9792 if (txqs < 1) {
9793 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9794 return NULL;
9795 }
9796
9797 if (rxqs < 1) {
9798 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9799 return NULL;
9800 }
9801
9802 alloc_size = sizeof(struct net_device);
9803 if (sizeof_priv) {
9804 /* ensure 32-byte alignment of private area */
9805 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9806 alloc_size += sizeof_priv;
9807 }
9808 /* ensure 32-byte alignment of whole construct */
9809 alloc_size += NETDEV_ALIGN - 1;
9810
9811 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9812 if (!p)
9813 return NULL;
9814
9815 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9816 dev->padded = (char *)dev - (char *)p;
9817
9818 dev->pcpu_refcnt = alloc_percpu(int);
9819 if (!dev->pcpu_refcnt)
9820 goto free_dev;
9821
9822 if (dev_addr_init(dev))
9823 goto free_pcpu;
9824
9825 dev_mc_init(dev);
9826 dev_uc_init(dev);
9827
9828 dev_net_set(dev, &init_net);
9829
9830 netdev_register_lockdep_key(dev);
9831
9832 dev->gso_max_size = GSO_MAX_SIZE;
9833 dev->gso_max_segs = GSO_MAX_SEGS;
9834 dev->upper_level = 1;
9835 dev->lower_level = 1;
9836
9837 INIT_LIST_HEAD(&dev->napi_list);
9838 INIT_LIST_HEAD(&dev->unreg_list);
9839 INIT_LIST_HEAD(&dev->close_list);
9840 INIT_LIST_HEAD(&dev->link_watch_list);
9841 INIT_LIST_HEAD(&dev->adj_list.upper);
9842 INIT_LIST_HEAD(&dev->adj_list.lower);
9843 INIT_LIST_HEAD(&dev->ptype_all);
9844 INIT_LIST_HEAD(&dev->ptype_specific);
9845 INIT_LIST_HEAD(&dev->net_notifier_list);
9846 #ifdef CONFIG_NET_SCHED
9847 hash_init(dev->qdisc_hash);
9848 #endif
9849 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9850 setup(dev);
9851
9852 if (!dev->tx_queue_len) {
9853 dev->priv_flags |= IFF_NO_QUEUE;
9854 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9855 }
9856
9857 dev->num_tx_queues = txqs;
9858 dev->real_num_tx_queues = txqs;
9859 if (netif_alloc_netdev_queues(dev))
9860 goto free_all;
9861
9862 dev->num_rx_queues = rxqs;
9863 dev->real_num_rx_queues = rxqs;
9864 if (netif_alloc_rx_queues(dev))
9865 goto free_all;
9866
9867 strcpy(dev->name, name);
9868 dev->name_assign_type = name_assign_type;
9869 dev->group = INIT_NETDEV_GROUP;
9870 if (!dev->ethtool_ops)
9871 dev->ethtool_ops = &default_ethtool_ops;
9872
9873 nf_hook_ingress_init(dev);
9874
9875 return dev;
9876
9877 free_all:
9878 free_netdev(dev);
9879 return NULL;
9880
9881 free_pcpu:
9882 free_percpu(dev->pcpu_refcnt);
9883 free_dev:
9884 netdev_freemem(dev);
9885 return NULL;
9886 }
9887 EXPORT_SYMBOL(alloc_netdev_mqs);
9888
9889 /**
9890 * free_netdev - free network device
9891 * @dev: device
9892 *
9893 * This function does the last stage of destroying an allocated device
9894 * interface. The reference to the device object is released. If this
9895 * is the last reference then it will be freed.Must be called in process
9896 * context.
9897 */
9898 void free_netdev(struct net_device *dev)
9899 {
9900 struct napi_struct *p, *n;
9901
9902 might_sleep();
9903 netif_free_tx_queues(dev);
9904 netif_free_rx_queues(dev);
9905
9906 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9907
9908 /* Flush device addresses */
9909 dev_addr_flush(dev);
9910
9911 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9912 netif_napi_del(p);
9913
9914 free_percpu(dev->pcpu_refcnt);
9915 dev->pcpu_refcnt = NULL;
9916 free_percpu(dev->xdp_bulkq);
9917 dev->xdp_bulkq = NULL;
9918
9919 netdev_unregister_lockdep_key(dev);
9920
9921 /* Compatibility with error handling in drivers */
9922 if (dev->reg_state == NETREG_UNINITIALIZED) {
9923 netdev_freemem(dev);
9924 return;
9925 }
9926
9927 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9928 dev->reg_state = NETREG_RELEASED;
9929
9930 /* will free via device release */
9931 put_device(&dev->dev);
9932 }
9933 EXPORT_SYMBOL(free_netdev);
9934
9935 /**
9936 * synchronize_net - Synchronize with packet receive processing
9937 *
9938 * Wait for packets currently being received to be done.
9939 * Does not block later packets from starting.
9940 */
9941 void synchronize_net(void)
9942 {
9943 might_sleep();
9944 if (rtnl_is_locked())
9945 synchronize_rcu_expedited();
9946 else
9947 synchronize_rcu();
9948 }
9949 EXPORT_SYMBOL(synchronize_net);
9950
9951 /**
9952 * unregister_netdevice_queue - remove device from the kernel
9953 * @dev: device
9954 * @head: list
9955 *
9956 * This function shuts down a device interface and removes it
9957 * from the kernel tables.
9958 * If head not NULL, device is queued to be unregistered later.
9959 *
9960 * Callers must hold the rtnl semaphore. You may want
9961 * unregister_netdev() instead of this.
9962 */
9963
9964 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9965 {
9966 ASSERT_RTNL();
9967
9968 if (head) {
9969 list_move_tail(&dev->unreg_list, head);
9970 } else {
9971 rollback_registered(dev);
9972 /* Finish processing unregister after unlock */
9973 net_set_todo(dev);
9974 }
9975 }
9976 EXPORT_SYMBOL(unregister_netdevice_queue);
9977
9978 /**
9979 * unregister_netdevice_many - unregister many devices
9980 * @head: list of devices
9981 *
9982 * Note: As most callers use a stack allocated list_head,
9983 * we force a list_del() to make sure stack wont be corrupted later.
9984 */
9985 void unregister_netdevice_many(struct list_head *head)
9986 {
9987 struct net_device *dev;
9988
9989 if (!list_empty(head)) {
9990 rollback_registered_many(head);
9991 list_for_each_entry(dev, head, unreg_list)
9992 net_set_todo(dev);
9993 list_del(head);
9994 }
9995 }
9996 EXPORT_SYMBOL(unregister_netdevice_many);
9997
9998 /**
9999 * unregister_netdev - remove device from the kernel
10000 * @dev: device
10001 *
10002 * This function shuts down a device interface and removes it
10003 * from the kernel tables.
10004 *
10005 * This is just a wrapper for unregister_netdevice that takes
10006 * the rtnl semaphore. In general you want to use this and not
10007 * unregister_netdevice.
10008 */
10009 void unregister_netdev(struct net_device *dev)
10010 {
10011 rtnl_lock();
10012 unregister_netdevice(dev);
10013 rtnl_unlock();
10014 }
10015 EXPORT_SYMBOL(unregister_netdev);
10016
10017 /**
10018 * dev_change_net_namespace - move device to different nethost namespace
10019 * @dev: device
10020 * @net: network namespace
10021 * @pat: If not NULL name pattern to try if the current device name
10022 * is already taken in the destination network namespace.
10023 *
10024 * This function shuts down a device interface and moves it
10025 * to a new network namespace. On success 0 is returned, on
10026 * a failure a netagive errno code is returned.
10027 *
10028 * Callers must hold the rtnl semaphore.
10029 */
10030
10031 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
10032 {
10033 struct net *net_old = dev_net(dev);
10034 int err, new_nsid, new_ifindex;
10035
10036 ASSERT_RTNL();
10037
10038 /* Don't allow namespace local devices to be moved. */
10039 err = -EINVAL;
10040 if (dev->features & NETIF_F_NETNS_LOCAL)
10041 goto out;
10042
10043 /* Ensure the device has been registrered */
10044 if (dev->reg_state != NETREG_REGISTERED)
10045 goto out;
10046
10047 /* Get out if there is nothing todo */
10048 err = 0;
10049 if (net_eq(net_old, net))
10050 goto out;
10051
10052 /* Pick the destination device name, and ensure
10053 * we can use it in the destination network namespace.
10054 */
10055 err = -EEXIST;
10056 if (__dev_get_by_name(net, dev->name)) {
10057 /* We get here if we can't use the current device name */
10058 if (!pat)
10059 goto out;
10060 err = dev_get_valid_name(net, dev, pat);
10061 if (err < 0)
10062 goto out;
10063 }
10064
10065 /*
10066 * And now a mini version of register_netdevice unregister_netdevice.
10067 */
10068
10069 /* If device is running close it first. */
10070 dev_close(dev);
10071
10072 /* And unlink it from device chain */
10073 unlist_netdevice(dev);
10074
10075 synchronize_net();
10076
10077 /* Shutdown queueing discipline. */
10078 dev_shutdown(dev);
10079
10080 /* Notify protocols, that we are about to destroy
10081 * this device. They should clean all the things.
10082 *
10083 * Note that dev->reg_state stays at NETREG_REGISTERED.
10084 * This is wanted because this way 8021q and macvlan know
10085 * the device is just moving and can keep their slaves up.
10086 */
10087 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10088 rcu_barrier();
10089
10090 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10091 /* If there is an ifindex conflict assign a new one */
10092 if (__dev_get_by_index(net, dev->ifindex))
10093 new_ifindex = dev_new_index(net);
10094 else
10095 new_ifindex = dev->ifindex;
10096
10097 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10098 new_ifindex);
10099
10100 /*
10101 * Flush the unicast and multicast chains
10102 */
10103 dev_uc_flush(dev);
10104 dev_mc_flush(dev);
10105
10106 /* Send a netdev-removed uevent to the old namespace */
10107 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10108 netdev_adjacent_del_links(dev);
10109
10110 /* Move per-net netdevice notifiers that are following the netdevice */
10111 move_netdevice_notifiers_dev_net(dev, net);
10112
10113 /* Actually switch the network namespace */
10114 dev_net_set(dev, net);
10115 dev->ifindex = new_ifindex;
10116
10117 /* Send a netdev-add uevent to the new namespace */
10118 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10119 netdev_adjacent_add_links(dev);
10120
10121 /* Fixup kobjects */
10122 err = device_rename(&dev->dev, dev->name);
10123 WARN_ON(err);
10124
10125 /* Adapt owner in case owning user namespace of target network
10126 * namespace is different from the original one.
10127 */
10128 err = netdev_change_owner(dev, net_old, net);
10129 WARN_ON(err);
10130
10131 /* Add the device back in the hashes */
10132 list_netdevice(dev);
10133
10134 /* Notify protocols, that a new device appeared. */
10135 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10136
10137 /*
10138 * Prevent userspace races by waiting until the network
10139 * device is fully setup before sending notifications.
10140 */
10141 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10142
10143 synchronize_net();
10144 err = 0;
10145 out:
10146 return err;
10147 }
10148 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10149
10150 static int dev_cpu_dead(unsigned int oldcpu)
10151 {
10152 struct sk_buff **list_skb;
10153 struct sk_buff *skb;
10154 unsigned int cpu;
10155 struct softnet_data *sd, *oldsd, *remsd = NULL;
10156
10157 local_irq_disable();
10158 cpu = smp_processor_id();
10159 sd = &per_cpu(softnet_data, cpu);
10160 oldsd = &per_cpu(softnet_data, oldcpu);
10161
10162 /* Find end of our completion_queue. */
10163 list_skb = &sd->completion_queue;
10164 while (*list_skb)
10165 list_skb = &(*list_skb)->next;
10166 /* Append completion queue from offline CPU. */
10167 *list_skb = oldsd->completion_queue;
10168 oldsd->completion_queue = NULL;
10169
10170 /* Append output queue from offline CPU. */
10171 if (oldsd->output_queue) {
10172 *sd->output_queue_tailp = oldsd->output_queue;
10173 sd->output_queue_tailp = oldsd->output_queue_tailp;
10174 oldsd->output_queue = NULL;
10175 oldsd->output_queue_tailp = &oldsd->output_queue;
10176 }
10177 /* Append NAPI poll list from offline CPU, with one exception :
10178 * process_backlog() must be called by cpu owning percpu backlog.
10179 * We properly handle process_queue & input_pkt_queue later.
10180 */
10181 while (!list_empty(&oldsd->poll_list)) {
10182 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10183 struct napi_struct,
10184 poll_list);
10185
10186 list_del_init(&napi->poll_list);
10187 if (napi->poll == process_backlog)
10188 napi->state = 0;
10189 else
10190 ____napi_schedule(sd, napi);
10191 }
10192
10193 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10194 local_irq_enable();
10195
10196 #ifdef CONFIG_RPS
10197 remsd = oldsd->rps_ipi_list;
10198 oldsd->rps_ipi_list = NULL;
10199 #endif
10200 /* send out pending IPI's on offline CPU */
10201 net_rps_send_ipi(remsd);
10202
10203 /* Process offline CPU's input_pkt_queue */
10204 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10205 netif_rx_ni(skb);
10206 input_queue_head_incr(oldsd);
10207 }
10208 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10209 netif_rx_ni(skb);
10210 input_queue_head_incr(oldsd);
10211 }
10212
10213 return 0;
10214 }
10215
10216 /**
10217 * netdev_increment_features - increment feature set by one
10218 * @all: current feature set
10219 * @one: new feature set
10220 * @mask: mask feature set
10221 *
10222 * Computes a new feature set after adding a device with feature set
10223 * @one to the master device with current feature set @all. Will not
10224 * enable anything that is off in @mask. Returns the new feature set.
10225 */
10226 netdev_features_t netdev_increment_features(netdev_features_t all,
10227 netdev_features_t one, netdev_features_t mask)
10228 {
10229 if (mask & NETIF_F_HW_CSUM)
10230 mask |= NETIF_F_CSUM_MASK;
10231 mask |= NETIF_F_VLAN_CHALLENGED;
10232
10233 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10234 all &= one | ~NETIF_F_ALL_FOR_ALL;
10235
10236 /* If one device supports hw checksumming, set for all. */
10237 if (all & NETIF_F_HW_CSUM)
10238 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10239
10240 return all;
10241 }
10242 EXPORT_SYMBOL(netdev_increment_features);
10243
10244 static struct hlist_head * __net_init netdev_create_hash(void)
10245 {
10246 int i;
10247 struct hlist_head *hash;
10248
10249 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10250 if (hash != NULL)
10251 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10252 INIT_HLIST_HEAD(&hash[i]);
10253
10254 return hash;
10255 }
10256
10257 /* Initialize per network namespace state */
10258 static int __net_init netdev_init(struct net *net)
10259 {
10260 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10261 8 * sizeof_field(struct napi_struct, gro_bitmask));
10262
10263 if (net != &init_net)
10264 INIT_LIST_HEAD(&net->dev_base_head);
10265
10266 net->dev_name_head = netdev_create_hash();
10267 if (net->dev_name_head == NULL)
10268 goto err_name;
10269
10270 net->dev_index_head = netdev_create_hash();
10271 if (net->dev_index_head == NULL)
10272 goto err_idx;
10273
10274 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10275
10276 return 0;
10277
10278 err_idx:
10279 kfree(net->dev_name_head);
10280 err_name:
10281 return -ENOMEM;
10282 }
10283
10284 /**
10285 * netdev_drivername - network driver for the device
10286 * @dev: network device
10287 *
10288 * Determine network driver for device.
10289 */
10290 const char *netdev_drivername(const struct net_device *dev)
10291 {
10292 const struct device_driver *driver;
10293 const struct device *parent;
10294 const char *empty = "";
10295
10296 parent = dev->dev.parent;
10297 if (!parent)
10298 return empty;
10299
10300 driver = parent->driver;
10301 if (driver && driver->name)
10302 return driver->name;
10303 return empty;
10304 }
10305
10306 static void __netdev_printk(const char *level, const struct net_device *dev,
10307 struct va_format *vaf)
10308 {
10309 if (dev && dev->dev.parent) {
10310 dev_printk_emit(level[1] - '0',
10311 dev->dev.parent,
10312 "%s %s %s%s: %pV",
10313 dev_driver_string(dev->dev.parent),
10314 dev_name(dev->dev.parent),
10315 netdev_name(dev), netdev_reg_state(dev),
10316 vaf);
10317 } else if (dev) {
10318 printk("%s%s%s: %pV",
10319 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10320 } else {
10321 printk("%s(NULL net_device): %pV", level, vaf);
10322 }
10323 }
10324
10325 void netdev_printk(const char *level, const struct net_device *dev,
10326 const char *format, ...)
10327 {
10328 struct va_format vaf;
10329 va_list args;
10330
10331 va_start(args, format);
10332
10333 vaf.fmt = format;
10334 vaf.va = &args;
10335
10336 __netdev_printk(level, dev, &vaf);
10337
10338 va_end(args);
10339 }
10340 EXPORT_SYMBOL(netdev_printk);
10341
10342 #define define_netdev_printk_level(func, level) \
10343 void func(const struct net_device *dev, const char *fmt, ...) \
10344 { \
10345 struct va_format vaf; \
10346 va_list args; \
10347 \
10348 va_start(args, fmt); \
10349 \
10350 vaf.fmt = fmt; \
10351 vaf.va = &args; \
10352 \
10353 __netdev_printk(level, dev, &vaf); \
10354 \
10355 va_end(args); \
10356 } \
10357 EXPORT_SYMBOL(func);
10358
10359 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10360 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10361 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10362 define_netdev_printk_level(netdev_err, KERN_ERR);
10363 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10364 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10365 define_netdev_printk_level(netdev_info, KERN_INFO);
10366
10367 static void __net_exit netdev_exit(struct net *net)
10368 {
10369 kfree(net->dev_name_head);
10370 kfree(net->dev_index_head);
10371 if (net != &init_net)
10372 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10373 }
10374
10375 static struct pernet_operations __net_initdata netdev_net_ops = {
10376 .init = netdev_init,
10377 .exit = netdev_exit,
10378 };
10379
10380 static void __net_exit default_device_exit(struct net *net)
10381 {
10382 struct net_device *dev, *aux;
10383 /*
10384 * Push all migratable network devices back to the
10385 * initial network namespace
10386 */
10387 rtnl_lock();
10388 for_each_netdev_safe(net, dev, aux) {
10389 int err;
10390 char fb_name[IFNAMSIZ];
10391
10392 /* Ignore unmoveable devices (i.e. loopback) */
10393 if (dev->features & NETIF_F_NETNS_LOCAL)
10394 continue;
10395
10396 /* Leave virtual devices for the generic cleanup */
10397 if (dev->rtnl_link_ops)
10398 continue;
10399
10400 /* Push remaining network devices to init_net */
10401 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10402 if (__dev_get_by_name(&init_net, fb_name))
10403 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10404 err = dev_change_net_namespace(dev, &init_net, fb_name);
10405 if (err) {
10406 pr_emerg("%s: failed to move %s to init_net: %d\n",
10407 __func__, dev->name, err);
10408 BUG();
10409 }
10410 }
10411 rtnl_unlock();
10412 }
10413
10414 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10415 {
10416 /* Return with the rtnl_lock held when there are no network
10417 * devices unregistering in any network namespace in net_list.
10418 */
10419 struct net *net;
10420 bool unregistering;
10421 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10422
10423 add_wait_queue(&netdev_unregistering_wq, &wait);
10424 for (;;) {
10425 unregistering = false;
10426 rtnl_lock();
10427 list_for_each_entry(net, net_list, exit_list) {
10428 if (net->dev_unreg_count > 0) {
10429 unregistering = true;
10430 break;
10431 }
10432 }
10433 if (!unregistering)
10434 break;
10435 __rtnl_unlock();
10436
10437 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10438 }
10439 remove_wait_queue(&netdev_unregistering_wq, &wait);
10440 }
10441
10442 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10443 {
10444 /* At exit all network devices most be removed from a network
10445 * namespace. Do this in the reverse order of registration.
10446 * Do this across as many network namespaces as possible to
10447 * improve batching efficiency.
10448 */
10449 struct net_device *dev;
10450 struct net *net;
10451 LIST_HEAD(dev_kill_list);
10452
10453 /* To prevent network device cleanup code from dereferencing
10454 * loopback devices or network devices that have been freed
10455 * wait here for all pending unregistrations to complete,
10456 * before unregistring the loopback device and allowing the
10457 * network namespace be freed.
10458 *
10459 * The netdev todo list containing all network devices
10460 * unregistrations that happen in default_device_exit_batch
10461 * will run in the rtnl_unlock() at the end of
10462 * default_device_exit_batch.
10463 */
10464 rtnl_lock_unregistering(net_list);
10465 list_for_each_entry(net, net_list, exit_list) {
10466 for_each_netdev_reverse(net, dev) {
10467 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10468 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10469 else
10470 unregister_netdevice_queue(dev, &dev_kill_list);
10471 }
10472 }
10473 unregister_netdevice_many(&dev_kill_list);
10474 rtnl_unlock();
10475 }
10476
10477 static struct pernet_operations __net_initdata default_device_ops = {
10478 .exit = default_device_exit,
10479 .exit_batch = default_device_exit_batch,
10480 };
10481
10482 /*
10483 * Initialize the DEV module. At boot time this walks the device list and
10484 * unhooks any devices that fail to initialise (normally hardware not
10485 * present) and leaves us with a valid list of present and active devices.
10486 *
10487 */
10488
10489 /*
10490 * This is called single threaded during boot, so no need
10491 * to take the rtnl semaphore.
10492 */
10493 static int __init net_dev_init(void)
10494 {
10495 int i, rc = -ENOMEM;
10496
10497 BUG_ON(!dev_boot_phase);
10498
10499 if (dev_proc_init())
10500 goto out;
10501
10502 if (netdev_kobject_init())
10503 goto out;
10504
10505 INIT_LIST_HEAD(&ptype_all);
10506 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10507 INIT_LIST_HEAD(&ptype_base[i]);
10508
10509 INIT_LIST_HEAD(&offload_base);
10510
10511 if (register_pernet_subsys(&netdev_net_ops))
10512 goto out;
10513
10514 /*
10515 * Initialise the packet receive queues.
10516 */
10517
10518 for_each_possible_cpu(i) {
10519 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10520 struct softnet_data *sd = &per_cpu(softnet_data, i);
10521
10522 INIT_WORK(flush, flush_backlog);
10523
10524 skb_queue_head_init(&sd->input_pkt_queue);
10525 skb_queue_head_init(&sd->process_queue);
10526 #ifdef CONFIG_XFRM_OFFLOAD
10527 skb_queue_head_init(&sd->xfrm_backlog);
10528 #endif
10529 INIT_LIST_HEAD(&sd->poll_list);
10530 sd->output_queue_tailp = &sd->output_queue;
10531 #ifdef CONFIG_RPS
10532 sd->csd.func = rps_trigger_softirq;
10533 sd->csd.info = sd;
10534 sd->cpu = i;
10535 #endif
10536
10537 init_gro_hash(&sd->backlog);
10538 sd->backlog.poll = process_backlog;
10539 sd->backlog.weight = weight_p;
10540 }
10541
10542 dev_boot_phase = 0;
10543
10544 /* The loopback device is special if any other network devices
10545 * is present in a network namespace the loopback device must
10546 * be present. Since we now dynamically allocate and free the
10547 * loopback device ensure this invariant is maintained by
10548 * keeping the loopback device as the first device on the
10549 * list of network devices. Ensuring the loopback devices
10550 * is the first device that appears and the last network device
10551 * that disappears.
10552 */
10553 if (register_pernet_device(&loopback_net_ops))
10554 goto out;
10555
10556 if (register_pernet_device(&default_device_ops))
10557 goto out;
10558
10559 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10560 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10561
10562 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10563 NULL, dev_cpu_dead);
10564 WARN_ON(rc < 0);
10565 rc = 0;
10566 out:
10567 return rc;
10568 }
10569
10570 subsys_initcall(net_dev_init);