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/bitmap.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/isolation.h>
81 #include <linux/sched/mm.h>
82 #include <linux/smpboot.h>
83 #include <linux/mutex.h>
84 #include <linux/rwsem.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/skbuff.h>
96 #include <linux/kthread.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
104 #include <net/dsa.h>
105 #include <net/dst.h>
106 #include <net/dst_metadata.h>
107 #include <net/gro.h>
108 #include <net/pkt_sched.h>
109 #include <net/pkt_cls.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <net/tcx.h>
113 #include <linux/highmem.h>
114 #include <linux/init.h>
115 #include <linux/module.h>
116 #include <linux/netpoll.h>
117 #include <linux/rcupdate.h>
118 #include <linux/delay.h>
119 #include <net/iw_handler.h>
120 #include <asm/current.h>
121 #include <linux/audit.h>
122 #include <linux/dmaengine.h>
123 #include <linux/err.h>
124 #include <linux/ctype.h>
125 #include <linux/if_arp.h>
126 #include <linux/if_vlan.h>
127 #include <linux/ip.h>
128 #include <net/ip.h>
129 #include <net/mpls.h>
130 #include <linux/ipv6.h>
131 #include <linux/in.h>
132 #include <linux/jhash.h>
133 #include <linux/random.h>
134 #include <trace/events/napi.h>
135 #include <trace/events/net.h>
136 #include <trace/events/skb.h>
137 #include <trace/events/qdisc.h>
138 #include <trace/events/xdp.h>
139 #include <linux/inetdevice.h>
140 #include <linux/cpu_rmap.h>
141 #include <linux/static_key.h>
142 #include <linux/hashtable.h>
143 #include <linux/vmalloc.h>
144 #include <linux/if_macvlan.h>
145 #include <linux/errqueue.h>
146 #include <linux/hrtimer.h>
147 #include <linux/netfilter_netdev.h>
148 #include <linux/crash_dump.h>
149 #include <linux/sctp.h>
150 #include <net/udp_tunnel.h>
151 #include <linux/net_namespace.h>
152 #include <linux/indirect_call_wrapper.h>
153 #include <net/devlink.h>
154 #include <linux/pm_runtime.h>
155 #include <linux/prandom.h>
156 #include <linux/once_lite.h>
157 #include <net/netdev_rx_queue.h>
158 #include <net/page_pool/types.h>
159 #include <net/page_pool/helpers.h>
160 #include <net/rps.h>
161
162 #include "dev.h"
163 #include "net-sysfs.h"
164
165 static DEFINE_SPINLOCK(ptype_lock);
166 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
167
168 static int netif_rx_internal(struct sk_buff *skb);
169 static int call_netdevice_notifiers_extack(unsigned long val,
170 struct net_device *dev,
171 struct netlink_ext_ack *extack);
172
173 static DEFINE_MUTEX(ifalias_mutex);
174
175 /* protects napi_hash addition/deletion and napi_gen_id */
176 static DEFINE_SPINLOCK(napi_hash_lock);
177
178 static unsigned int napi_gen_id = NR_CPUS;
179 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
180
181 static DECLARE_RWSEM(devnet_rename_sem);
182
183 static inline void dev_base_seq_inc(struct net *net)
184 {
185 unsigned int val = net->dev_base_seq + 1;
186
187 WRITE_ONCE(net->dev_base_seq, val ?: 1);
188 }
189
190 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
191 {
192 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
193
194 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
195 }
196
197 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
198 {
199 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
200 }
201
202 #ifndef CONFIG_PREEMPT_RT
203
204 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key);
205
206 static int __init setup_backlog_napi_threads(char *arg)
207 {
208 static_branch_enable(&use_backlog_threads_key);
209 return 0;
210 }
211 early_param("thread_backlog_napi", setup_backlog_napi_threads);
212
213 static bool use_backlog_threads(void)
214 {
215 return static_branch_unlikely(&use_backlog_threads_key);
216 }
217
218 #else
219
220 static bool use_backlog_threads(void)
221 {
222 return true;
223 }
224
225 #endif
226
227 static inline void backlog_lock_irq_save(struct softnet_data *sd,
228 unsigned long *flags)
229 {
230 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
231 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
232 else
233 local_irq_save(*flags);
234 }
235
236 static inline void backlog_lock_irq_disable(struct softnet_data *sd)
237 {
238 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
239 spin_lock_irq(&sd->input_pkt_queue.lock);
240 else
241 local_irq_disable();
242 }
243
244 static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
245 unsigned long *flags)
246 {
247 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
248 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
249 else
250 local_irq_restore(*flags);
251 }
252
253 static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
254 {
255 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
256 spin_unlock_irq(&sd->input_pkt_queue.lock);
257 else
258 local_irq_enable();
259 }
260
261 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
262 const char *name)
263 {
264 struct netdev_name_node *name_node;
265
266 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
267 if (!name_node)
268 return NULL;
269 INIT_HLIST_NODE(&name_node->hlist);
270 name_node->dev = dev;
271 name_node->name = name;
272 return name_node;
273 }
274
275 static struct netdev_name_node *
276 netdev_name_node_head_alloc(struct net_device *dev)
277 {
278 struct netdev_name_node *name_node;
279
280 name_node = netdev_name_node_alloc(dev, dev->name);
281 if (!name_node)
282 return NULL;
283 INIT_LIST_HEAD(&name_node->list);
284 return name_node;
285 }
286
287 static void netdev_name_node_free(struct netdev_name_node *name_node)
288 {
289 kfree(name_node);
290 }
291
292 static void netdev_name_node_add(struct net *net,
293 struct netdev_name_node *name_node)
294 {
295 hlist_add_head_rcu(&name_node->hlist,
296 dev_name_hash(net, name_node->name));
297 }
298
299 static void netdev_name_node_del(struct netdev_name_node *name_node)
300 {
301 hlist_del_rcu(&name_node->hlist);
302 }
303
304 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
305 const char *name)
306 {
307 struct hlist_head *head = dev_name_hash(net, name);
308 struct netdev_name_node *name_node;
309
310 hlist_for_each_entry(name_node, head, hlist)
311 if (!strcmp(name_node->name, name))
312 return name_node;
313 return NULL;
314 }
315
316 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
317 const char *name)
318 {
319 struct hlist_head *head = dev_name_hash(net, name);
320 struct netdev_name_node *name_node;
321
322 hlist_for_each_entry_rcu(name_node, head, hlist)
323 if (!strcmp(name_node->name, name))
324 return name_node;
325 return NULL;
326 }
327
328 bool netdev_name_in_use(struct net *net, const char *name)
329 {
330 return netdev_name_node_lookup(net, name);
331 }
332 EXPORT_SYMBOL(netdev_name_in_use);
333
334 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
335 {
336 struct netdev_name_node *name_node;
337 struct net *net = dev_net(dev);
338
339 name_node = netdev_name_node_lookup(net, name);
340 if (name_node)
341 return -EEXIST;
342 name_node = netdev_name_node_alloc(dev, name);
343 if (!name_node)
344 return -ENOMEM;
345 netdev_name_node_add(net, name_node);
346 /* The node that holds dev->name acts as a head of per-device list. */
347 list_add_tail_rcu(&name_node->list, &dev->name_node->list);
348
349 return 0;
350 }
351
352 static void netdev_name_node_alt_free(struct rcu_head *head)
353 {
354 struct netdev_name_node *name_node =
355 container_of(head, struct netdev_name_node, rcu);
356
357 kfree(name_node->name);
358 netdev_name_node_free(name_node);
359 }
360
361 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
362 {
363 netdev_name_node_del(name_node);
364 list_del(&name_node->list);
365 call_rcu(&name_node->rcu, netdev_name_node_alt_free);
366 }
367
368 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
369 {
370 struct netdev_name_node *name_node;
371 struct net *net = dev_net(dev);
372
373 name_node = netdev_name_node_lookup(net, name);
374 if (!name_node)
375 return -ENOENT;
376 /* lookup might have found our primary name or a name belonging
377 * to another device.
378 */
379 if (name_node == dev->name_node || name_node->dev != dev)
380 return -EINVAL;
381
382 __netdev_name_node_alt_destroy(name_node);
383 return 0;
384 }
385
386 static void netdev_name_node_alt_flush(struct net_device *dev)
387 {
388 struct netdev_name_node *name_node, *tmp;
389
390 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) {
391 list_del(&name_node->list);
392 netdev_name_node_alt_free(&name_node->rcu);
393 }
394 }
395
396 /* Device list insertion */
397 static void list_netdevice(struct net_device *dev)
398 {
399 struct netdev_name_node *name_node;
400 struct net *net = dev_net(dev);
401
402 ASSERT_RTNL();
403
404 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
405 netdev_name_node_add(net, dev->name_node);
406 hlist_add_head_rcu(&dev->index_hlist,
407 dev_index_hash(net, dev->ifindex));
408
409 netdev_for_each_altname(dev, name_node)
410 netdev_name_node_add(net, name_node);
411
412 /* We reserved the ifindex, this can't fail */
413 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL));
414
415 dev_base_seq_inc(net);
416 }
417
418 /* Device list removal
419 * caller must respect a RCU grace period before freeing/reusing dev
420 */
421 static void unlist_netdevice(struct net_device *dev)
422 {
423 struct netdev_name_node *name_node;
424 struct net *net = dev_net(dev);
425
426 ASSERT_RTNL();
427
428 xa_erase(&net->dev_by_index, dev->ifindex);
429
430 netdev_for_each_altname(dev, name_node)
431 netdev_name_node_del(name_node);
432
433 /* Unlink dev from the device chain */
434 list_del_rcu(&dev->dev_list);
435 netdev_name_node_del(dev->name_node);
436 hlist_del_rcu(&dev->index_hlist);
437
438 dev_base_seq_inc(dev_net(dev));
439 }
440
441 /*
442 * Our notifier list
443 */
444
445 static RAW_NOTIFIER_HEAD(netdev_chain);
446
447 /*
448 * Device drivers call our routines to queue packets here. We empty the
449 * queue in the local softnet handler.
450 */
451
452 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = {
453 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock),
454 };
455 EXPORT_PER_CPU_SYMBOL(softnet_data);
456
457 /* Page_pool has a lockless array/stack to alloc/recycle pages.
458 * PP consumers must pay attention to run APIs in the appropriate context
459 * (e.g. NAPI context).
460 */
461 static DEFINE_PER_CPU(struct page_pool *, system_page_pool);
462
463 #ifdef CONFIG_LOCKDEP
464 /*
465 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
466 * according to dev->type
467 */
468 static const unsigned short netdev_lock_type[] = {
469 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
470 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
471 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
472 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
473 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
474 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
475 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
476 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
477 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
478 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
479 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
480 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
481 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
482 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
483 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
484
485 static const char *const netdev_lock_name[] = {
486 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
487 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
488 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
489 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
490 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
491 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
492 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
493 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
494 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
495 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
496 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
497 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
498 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
499 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
500 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
501
502 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
503 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
504
505 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
506 {
507 int i;
508
509 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
510 if (netdev_lock_type[i] == dev_type)
511 return i;
512 /* the last key is used by default */
513 return ARRAY_SIZE(netdev_lock_type) - 1;
514 }
515
516 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
517 unsigned short dev_type)
518 {
519 int i;
520
521 i = netdev_lock_pos(dev_type);
522 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
523 netdev_lock_name[i]);
524 }
525
526 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
527 {
528 int i;
529
530 i = netdev_lock_pos(dev->type);
531 lockdep_set_class_and_name(&dev->addr_list_lock,
532 &netdev_addr_lock_key[i],
533 netdev_lock_name[i]);
534 }
535 #else
536 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
537 unsigned short dev_type)
538 {
539 }
540
541 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
542 {
543 }
544 #endif
545
546 /*******************************************************************************
547 *
548 * Protocol management and registration routines
549 *
550 *******************************************************************************/
551
552
553 /*
554 * Add a protocol ID to the list. Now that the input handler is
555 * smarter we can dispense with all the messy stuff that used to be
556 * here.
557 *
558 * BEWARE!!! Protocol handlers, mangling input packets,
559 * MUST BE last in hash buckets and checking protocol handlers
560 * MUST start from promiscuous ptype_all chain in net_bh.
561 * It is true now, do not change it.
562 * Explanation follows: if protocol handler, mangling packet, will
563 * be the first on list, it is not able to sense, that packet
564 * is cloned and should be copied-on-write, so that it will
565 * change it and subsequent readers will get broken packet.
566 * --ANK (980803)
567 */
568
569 static inline struct list_head *ptype_head(const struct packet_type *pt)
570 {
571 if (pt->type == htons(ETH_P_ALL))
572 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all;
573 else
574 return pt->dev ? &pt->dev->ptype_specific :
575 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
576 }
577
578 /**
579 * dev_add_pack - add packet handler
580 * @pt: packet type declaration
581 *
582 * Add a protocol handler to the networking stack. The passed &packet_type
583 * is linked into kernel lists and may not be freed until it has been
584 * removed from the kernel lists.
585 *
586 * This call does not sleep therefore it can not
587 * guarantee all CPU's that are in middle of receiving packets
588 * will see the new packet type (until the next received packet).
589 */
590
591 void dev_add_pack(struct packet_type *pt)
592 {
593 struct list_head *head = ptype_head(pt);
594
595 spin_lock(&ptype_lock);
596 list_add_rcu(&pt->list, head);
597 spin_unlock(&ptype_lock);
598 }
599 EXPORT_SYMBOL(dev_add_pack);
600
601 /**
602 * __dev_remove_pack - remove packet handler
603 * @pt: packet type declaration
604 *
605 * Remove a protocol handler that was previously added to the kernel
606 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
607 * from the kernel lists and can be freed or reused once this function
608 * returns.
609 *
610 * The packet type might still be in use by receivers
611 * and must not be freed until after all the CPU's have gone
612 * through a quiescent state.
613 */
614 void __dev_remove_pack(struct packet_type *pt)
615 {
616 struct list_head *head = ptype_head(pt);
617 struct packet_type *pt1;
618
619 spin_lock(&ptype_lock);
620
621 list_for_each_entry(pt1, head, list) {
622 if (pt == pt1) {
623 list_del_rcu(&pt->list);
624 goto out;
625 }
626 }
627
628 pr_warn("dev_remove_pack: %p not found\n", pt);
629 out:
630 spin_unlock(&ptype_lock);
631 }
632 EXPORT_SYMBOL(__dev_remove_pack);
633
634 /**
635 * dev_remove_pack - remove packet handler
636 * @pt: packet type declaration
637 *
638 * Remove a protocol handler that was previously added to the kernel
639 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
640 * from the kernel lists and can be freed or reused once this function
641 * returns.
642 *
643 * This call sleeps to guarantee that no CPU is looking at the packet
644 * type after return.
645 */
646 void dev_remove_pack(struct packet_type *pt)
647 {
648 __dev_remove_pack(pt);
649
650 synchronize_net();
651 }
652 EXPORT_SYMBOL(dev_remove_pack);
653
654
655 /*******************************************************************************
656 *
657 * Device Interface Subroutines
658 *
659 *******************************************************************************/
660
661 /**
662 * dev_get_iflink - get 'iflink' value of a interface
663 * @dev: targeted interface
664 *
665 * Indicates the ifindex the interface is linked to.
666 * Physical interfaces have the same 'ifindex' and 'iflink' values.
667 */
668
669 int dev_get_iflink(const struct net_device *dev)
670 {
671 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
672 return dev->netdev_ops->ndo_get_iflink(dev);
673
674 return READ_ONCE(dev->ifindex);
675 }
676 EXPORT_SYMBOL(dev_get_iflink);
677
678 /**
679 * dev_fill_metadata_dst - Retrieve tunnel egress information.
680 * @dev: targeted interface
681 * @skb: The packet.
682 *
683 * For better visibility of tunnel traffic OVS needs to retrieve
684 * egress tunnel information for a packet. Following API allows
685 * user to get this info.
686 */
687 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
688 {
689 struct ip_tunnel_info *info;
690
691 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
692 return -EINVAL;
693
694 info = skb_tunnel_info_unclone(skb);
695 if (!info)
696 return -ENOMEM;
697 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
698 return -EINVAL;
699
700 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
701 }
702 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
703
704 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
705 {
706 int k = stack->num_paths++;
707
708 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
709 return NULL;
710
711 return &stack->path[k];
712 }
713
714 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
715 struct net_device_path_stack *stack)
716 {
717 const struct net_device *last_dev;
718 struct net_device_path_ctx ctx = {
719 .dev = dev,
720 };
721 struct net_device_path *path;
722 int ret = 0;
723
724 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
725 stack->num_paths = 0;
726 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
727 last_dev = ctx.dev;
728 path = dev_fwd_path(stack);
729 if (!path)
730 return -1;
731
732 memset(path, 0, sizeof(struct net_device_path));
733 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
734 if (ret < 0)
735 return -1;
736
737 if (WARN_ON_ONCE(last_dev == ctx.dev))
738 return -1;
739 }
740
741 if (!ctx.dev)
742 return ret;
743
744 path = dev_fwd_path(stack);
745 if (!path)
746 return -1;
747 path->type = DEV_PATH_ETHERNET;
748 path->dev = ctx.dev;
749
750 return ret;
751 }
752 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
753
754 /**
755 * __dev_get_by_name - find a device by its name
756 * @net: the applicable net namespace
757 * @name: name to find
758 *
759 * Find an interface by name. Must be called under RTNL semaphore.
760 * If the name is found a pointer to the device is returned.
761 * If the name is not found then %NULL is returned. The
762 * reference counters are not incremented so the caller must be
763 * careful with locks.
764 */
765
766 struct net_device *__dev_get_by_name(struct net *net, const char *name)
767 {
768 struct netdev_name_node *node_name;
769
770 node_name = netdev_name_node_lookup(net, name);
771 return node_name ? node_name->dev : NULL;
772 }
773 EXPORT_SYMBOL(__dev_get_by_name);
774
775 /**
776 * dev_get_by_name_rcu - find a device by its name
777 * @net: the applicable net namespace
778 * @name: name to find
779 *
780 * Find an interface by name.
781 * If the name is found a pointer to the device is returned.
782 * If the name is not found then %NULL is returned.
783 * The reference counters are not incremented so the caller must be
784 * careful with locks. The caller must hold RCU lock.
785 */
786
787 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
788 {
789 struct netdev_name_node *node_name;
790
791 node_name = netdev_name_node_lookup_rcu(net, name);
792 return node_name ? node_name->dev : NULL;
793 }
794 EXPORT_SYMBOL(dev_get_by_name_rcu);
795
796 /* Deprecated for new users, call netdev_get_by_name() instead */
797 struct net_device *dev_get_by_name(struct net *net, const char *name)
798 {
799 struct net_device *dev;
800
801 rcu_read_lock();
802 dev = dev_get_by_name_rcu(net, name);
803 dev_hold(dev);
804 rcu_read_unlock();
805 return dev;
806 }
807 EXPORT_SYMBOL(dev_get_by_name);
808
809 /**
810 * netdev_get_by_name() - find a device by its name
811 * @net: the applicable net namespace
812 * @name: name to find
813 * @tracker: tracking object for the acquired reference
814 * @gfp: allocation flags for the tracker
815 *
816 * Find an interface by name. This can be called from any
817 * context and does its own locking. The returned handle has
818 * the usage count incremented and the caller must use netdev_put() to
819 * release it when it is no longer needed. %NULL is returned if no
820 * matching device is found.
821 */
822 struct net_device *netdev_get_by_name(struct net *net, const char *name,
823 netdevice_tracker *tracker, gfp_t gfp)
824 {
825 struct net_device *dev;
826
827 dev = dev_get_by_name(net, name);
828 if (dev)
829 netdev_tracker_alloc(dev, tracker, gfp);
830 return dev;
831 }
832 EXPORT_SYMBOL(netdev_get_by_name);
833
834 /**
835 * __dev_get_by_index - find a device by its ifindex
836 * @net: the applicable net namespace
837 * @ifindex: index of device
838 *
839 * Search for an interface by index. Returns %NULL if the device
840 * is not found or a pointer to the device. The device has not
841 * had its reference counter increased so the caller must be careful
842 * about locking. The caller must hold the RTNL semaphore.
843 */
844
845 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
846 {
847 struct net_device *dev;
848 struct hlist_head *head = dev_index_hash(net, ifindex);
849
850 hlist_for_each_entry(dev, head, index_hlist)
851 if (dev->ifindex == ifindex)
852 return dev;
853
854 return NULL;
855 }
856 EXPORT_SYMBOL(__dev_get_by_index);
857
858 /**
859 * dev_get_by_index_rcu - find a device by its ifindex
860 * @net: the applicable net namespace
861 * @ifindex: index of device
862 *
863 * Search for an interface by index. Returns %NULL if the device
864 * is not found or a pointer to the device. The device has not
865 * had its reference counter increased so the caller must be careful
866 * about locking. The caller must hold RCU lock.
867 */
868
869 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
870 {
871 struct net_device *dev;
872 struct hlist_head *head = dev_index_hash(net, ifindex);
873
874 hlist_for_each_entry_rcu(dev, head, index_hlist)
875 if (dev->ifindex == ifindex)
876 return dev;
877
878 return NULL;
879 }
880 EXPORT_SYMBOL(dev_get_by_index_rcu);
881
882 /* Deprecated for new users, call netdev_get_by_index() instead */
883 struct net_device *dev_get_by_index(struct net *net, int ifindex)
884 {
885 struct net_device *dev;
886
887 rcu_read_lock();
888 dev = dev_get_by_index_rcu(net, ifindex);
889 dev_hold(dev);
890 rcu_read_unlock();
891 return dev;
892 }
893 EXPORT_SYMBOL(dev_get_by_index);
894
895 /**
896 * netdev_get_by_index() - find a device by its ifindex
897 * @net: the applicable net namespace
898 * @ifindex: index of device
899 * @tracker: tracking object for the acquired reference
900 * @gfp: allocation flags for the tracker
901 *
902 * Search for an interface by index. Returns NULL if the device
903 * is not found or a pointer to the device. The device returned has
904 * had a reference added and the pointer is safe until the user calls
905 * netdev_put() to indicate they have finished with it.
906 */
907 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
908 netdevice_tracker *tracker, gfp_t gfp)
909 {
910 struct net_device *dev;
911
912 dev = dev_get_by_index(net, ifindex);
913 if (dev)
914 netdev_tracker_alloc(dev, tracker, gfp);
915 return dev;
916 }
917 EXPORT_SYMBOL(netdev_get_by_index);
918
919 /**
920 * dev_get_by_napi_id - find a device by napi_id
921 * @napi_id: ID of the NAPI struct
922 *
923 * Search for an interface by NAPI ID. Returns %NULL if the device
924 * is not found or a pointer to the device. The device has not had
925 * its reference counter increased so the caller must be careful
926 * about locking. The caller must hold RCU lock.
927 */
928
929 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
930 {
931 struct napi_struct *napi;
932
933 WARN_ON_ONCE(!rcu_read_lock_held());
934
935 if (napi_id < MIN_NAPI_ID)
936 return NULL;
937
938 napi = napi_by_id(napi_id);
939
940 return napi ? napi->dev : NULL;
941 }
942 EXPORT_SYMBOL(dev_get_by_napi_id);
943
944 static DEFINE_SEQLOCK(netdev_rename_lock);
945
946 void netdev_copy_name(struct net_device *dev, char *name)
947 {
948 unsigned int seq;
949
950 do {
951 seq = read_seqbegin(&netdev_rename_lock);
952 strscpy(name, dev->name, IFNAMSIZ);
953 } while (read_seqretry(&netdev_rename_lock, seq));
954 }
955
956 /**
957 * netdev_get_name - get a netdevice name, knowing its ifindex.
958 * @net: network namespace
959 * @name: a pointer to the buffer where the name will be stored.
960 * @ifindex: the ifindex of the interface to get the name from.
961 */
962 int netdev_get_name(struct net *net, char *name, int ifindex)
963 {
964 struct net_device *dev;
965 int ret;
966
967 rcu_read_lock();
968
969 dev = dev_get_by_index_rcu(net, ifindex);
970 if (!dev) {
971 ret = -ENODEV;
972 goto out;
973 }
974
975 netdev_copy_name(dev, name);
976
977 ret = 0;
978 out:
979 rcu_read_unlock();
980 return ret;
981 }
982
983 /**
984 * dev_getbyhwaddr_rcu - find a device by its hardware address
985 * @net: the applicable net namespace
986 * @type: media type of device
987 * @ha: hardware address
988 *
989 * Search for an interface by MAC address. Returns NULL if the device
990 * is not found or a pointer to the device.
991 * The caller must hold RCU or RTNL.
992 * The returned device has not had its ref count increased
993 * and the caller must therefore be careful about locking
994 *
995 */
996
997 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
998 const char *ha)
999 {
1000 struct net_device *dev;
1001
1002 for_each_netdev_rcu(net, dev)
1003 if (dev->type == type &&
1004 !memcmp(dev->dev_addr, ha, dev->addr_len))
1005 return dev;
1006
1007 return NULL;
1008 }
1009 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1010
1011 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1012 {
1013 struct net_device *dev, *ret = NULL;
1014
1015 rcu_read_lock();
1016 for_each_netdev_rcu(net, dev)
1017 if (dev->type == type) {
1018 dev_hold(dev);
1019 ret = dev;
1020 break;
1021 }
1022 rcu_read_unlock();
1023 return ret;
1024 }
1025 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1026
1027 /**
1028 * __dev_get_by_flags - find any device with given flags
1029 * @net: the applicable net namespace
1030 * @if_flags: IFF_* values
1031 * @mask: bitmask of bits in if_flags to check
1032 *
1033 * Search for any interface with the given flags. Returns NULL if a device
1034 * is not found or a pointer to the device. Must be called inside
1035 * rtnl_lock(), and result refcount is unchanged.
1036 */
1037
1038 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1039 unsigned short mask)
1040 {
1041 struct net_device *dev, *ret;
1042
1043 ASSERT_RTNL();
1044
1045 ret = NULL;
1046 for_each_netdev(net, dev) {
1047 if (((dev->flags ^ if_flags) & mask) == 0) {
1048 ret = dev;
1049 break;
1050 }
1051 }
1052 return ret;
1053 }
1054 EXPORT_SYMBOL(__dev_get_by_flags);
1055
1056 /**
1057 * dev_valid_name - check if name is okay for network device
1058 * @name: name string
1059 *
1060 * Network device names need to be valid file names to
1061 * allow sysfs to work. We also disallow any kind of
1062 * whitespace.
1063 */
1064 bool dev_valid_name(const char *name)
1065 {
1066 if (*name == '\0')
1067 return false;
1068 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1069 return false;
1070 if (!strcmp(name, ".") || !strcmp(name, ".."))
1071 return false;
1072
1073 while (*name) {
1074 if (*name == '/' || *name == ':' || isspace(*name))
1075 return false;
1076 name++;
1077 }
1078 return true;
1079 }
1080 EXPORT_SYMBOL(dev_valid_name);
1081
1082 /**
1083 * __dev_alloc_name - allocate a name for a device
1084 * @net: network namespace to allocate the device name in
1085 * @name: name format string
1086 * @res: result name string
1087 *
1088 * Passed a format string - eg "lt%d" it will try and find a suitable
1089 * id. It scans list of devices to build up a free map, then chooses
1090 * the first empty slot. The caller must hold the dev_base or rtnl lock
1091 * while allocating the name and adding the device in order to avoid
1092 * duplicates.
1093 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1094 * Returns the number of the unit assigned or a negative errno code.
1095 */
1096
1097 static int __dev_alloc_name(struct net *net, const char *name, char *res)
1098 {
1099 int i = 0;
1100 const char *p;
1101 const int max_netdevices = 8*PAGE_SIZE;
1102 unsigned long *inuse;
1103 struct net_device *d;
1104 char buf[IFNAMSIZ];
1105
1106 /* Verify the string as this thing may have come from the user.
1107 * There must be one "%d" and no other "%" characters.
1108 */
1109 p = strchr(name, '%');
1110 if (!p || p[1] != 'd' || strchr(p + 2, '%'))
1111 return -EINVAL;
1112
1113 /* Use one page as a bit array of possible slots */
1114 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC);
1115 if (!inuse)
1116 return -ENOMEM;
1117
1118 for_each_netdev(net, d) {
1119 struct netdev_name_node *name_node;
1120
1121 netdev_for_each_altname(d, name_node) {
1122 if (!sscanf(name_node->name, name, &i))
1123 continue;
1124 if (i < 0 || i >= max_netdevices)
1125 continue;
1126
1127 /* avoid cases where sscanf is not exact inverse of printf */
1128 snprintf(buf, IFNAMSIZ, name, i);
1129 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1130 __set_bit(i, inuse);
1131 }
1132 if (!sscanf(d->name, name, &i))
1133 continue;
1134 if (i < 0 || i >= max_netdevices)
1135 continue;
1136
1137 /* avoid cases where sscanf is not exact inverse of printf */
1138 snprintf(buf, IFNAMSIZ, name, i);
1139 if (!strncmp(buf, d->name, IFNAMSIZ))
1140 __set_bit(i, inuse);
1141 }
1142
1143 i = find_first_zero_bit(inuse, max_netdevices);
1144 bitmap_free(inuse);
1145 if (i == max_netdevices)
1146 return -ENFILE;
1147
1148 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */
1149 strscpy(buf, name, IFNAMSIZ);
1150 snprintf(res, IFNAMSIZ, buf, i);
1151 return i;
1152 }
1153
1154 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */
1155 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1156 const char *want_name, char *out_name,
1157 int dup_errno)
1158 {
1159 if (!dev_valid_name(want_name))
1160 return -EINVAL;
1161
1162 if (strchr(want_name, '%'))
1163 return __dev_alloc_name(net, want_name, out_name);
1164
1165 if (netdev_name_in_use(net, want_name))
1166 return -dup_errno;
1167 if (out_name != want_name)
1168 strscpy(out_name, want_name, IFNAMSIZ);
1169 return 0;
1170 }
1171
1172 /**
1173 * dev_alloc_name - allocate a name for a device
1174 * @dev: device
1175 * @name: name format string
1176 *
1177 * Passed a format string - eg "lt%d" it will try and find a suitable
1178 * id. It scans list of devices to build up a free map, then chooses
1179 * the first empty slot. The caller must hold the dev_base or rtnl lock
1180 * while allocating the name and adding the device in order to avoid
1181 * duplicates.
1182 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1183 * Returns the number of the unit assigned or a negative errno code.
1184 */
1185
1186 int dev_alloc_name(struct net_device *dev, const char *name)
1187 {
1188 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE);
1189 }
1190 EXPORT_SYMBOL(dev_alloc_name);
1191
1192 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1193 const char *name)
1194 {
1195 int ret;
1196
1197 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST);
1198 return ret < 0 ? ret : 0;
1199 }
1200
1201 /**
1202 * dev_change_name - change name of a device
1203 * @dev: device
1204 * @newname: name (or format string) must be at least IFNAMSIZ
1205 *
1206 * Change name of a device, can pass format strings "eth%d".
1207 * for wildcarding.
1208 */
1209 int dev_change_name(struct net_device *dev, const char *newname)
1210 {
1211 unsigned char old_assign_type;
1212 char oldname[IFNAMSIZ];
1213 int err = 0;
1214 int ret;
1215 struct net *net;
1216
1217 ASSERT_RTNL();
1218 BUG_ON(!dev_net(dev));
1219
1220 net = dev_net(dev);
1221
1222 down_write(&devnet_rename_sem);
1223
1224 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1225 up_write(&devnet_rename_sem);
1226 return 0;
1227 }
1228
1229 memcpy(oldname, dev->name, IFNAMSIZ);
1230
1231 write_seqlock_bh(&netdev_rename_lock);
1232 err = dev_get_valid_name(net, dev, newname);
1233 write_sequnlock_bh(&netdev_rename_lock);
1234
1235 if (err < 0) {
1236 up_write(&devnet_rename_sem);
1237 return err;
1238 }
1239
1240 if (oldname[0] && !strchr(oldname, '%'))
1241 netdev_info(dev, "renamed from %s%s\n", oldname,
1242 dev->flags & IFF_UP ? " (while UP)" : "");
1243
1244 old_assign_type = dev->name_assign_type;
1245 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED);
1246
1247 rollback:
1248 ret = device_rename(&dev->dev, dev->name);
1249 if (ret) {
1250 memcpy(dev->name, oldname, IFNAMSIZ);
1251 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1252 up_write(&devnet_rename_sem);
1253 return ret;
1254 }
1255
1256 up_write(&devnet_rename_sem);
1257
1258 netdev_adjacent_rename_links(dev, oldname);
1259
1260 netdev_name_node_del(dev->name_node);
1261
1262 synchronize_net();
1263
1264 netdev_name_node_add(net, dev->name_node);
1265
1266 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1267 ret = notifier_to_errno(ret);
1268
1269 if (ret) {
1270 /* err >= 0 after dev_alloc_name() or stores the first errno */
1271 if (err >= 0) {
1272 err = ret;
1273 down_write(&devnet_rename_sem);
1274 write_seqlock_bh(&netdev_rename_lock);
1275 memcpy(dev->name, oldname, IFNAMSIZ);
1276 write_sequnlock_bh(&netdev_rename_lock);
1277 memcpy(oldname, newname, IFNAMSIZ);
1278 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1279 old_assign_type = NET_NAME_RENAMED;
1280 goto rollback;
1281 } else {
1282 netdev_err(dev, "name change rollback failed: %d\n",
1283 ret);
1284 }
1285 }
1286
1287 return err;
1288 }
1289
1290 /**
1291 * dev_set_alias - change ifalias of a device
1292 * @dev: device
1293 * @alias: name up to IFALIASZ
1294 * @len: limit of bytes to copy from info
1295 *
1296 * Set ifalias for a device,
1297 */
1298 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1299 {
1300 struct dev_ifalias *new_alias = NULL;
1301
1302 if (len >= IFALIASZ)
1303 return -EINVAL;
1304
1305 if (len) {
1306 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1307 if (!new_alias)
1308 return -ENOMEM;
1309
1310 memcpy(new_alias->ifalias, alias, len);
1311 new_alias->ifalias[len] = 0;
1312 }
1313
1314 mutex_lock(&ifalias_mutex);
1315 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1316 mutex_is_locked(&ifalias_mutex));
1317 mutex_unlock(&ifalias_mutex);
1318
1319 if (new_alias)
1320 kfree_rcu(new_alias, rcuhead);
1321
1322 return len;
1323 }
1324 EXPORT_SYMBOL(dev_set_alias);
1325
1326 /**
1327 * dev_get_alias - get ifalias of a device
1328 * @dev: device
1329 * @name: buffer to store name of ifalias
1330 * @len: size of buffer
1331 *
1332 * get ifalias for a device. Caller must make sure dev cannot go
1333 * away, e.g. rcu read lock or own a reference count to device.
1334 */
1335 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1336 {
1337 const struct dev_ifalias *alias;
1338 int ret = 0;
1339
1340 rcu_read_lock();
1341 alias = rcu_dereference(dev->ifalias);
1342 if (alias)
1343 ret = snprintf(name, len, "%s", alias->ifalias);
1344 rcu_read_unlock();
1345
1346 return ret;
1347 }
1348
1349 /**
1350 * netdev_features_change - device changes features
1351 * @dev: device to cause notification
1352 *
1353 * Called to indicate a device has changed features.
1354 */
1355 void netdev_features_change(struct net_device *dev)
1356 {
1357 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1358 }
1359 EXPORT_SYMBOL(netdev_features_change);
1360
1361 /**
1362 * netdev_state_change - device changes state
1363 * @dev: device to cause notification
1364 *
1365 * Called to indicate a device has changed state. This function calls
1366 * the notifier chains for netdev_chain and sends a NEWLINK message
1367 * to the routing socket.
1368 */
1369 void netdev_state_change(struct net_device *dev)
1370 {
1371 if (dev->flags & IFF_UP) {
1372 struct netdev_notifier_change_info change_info = {
1373 .info.dev = dev,
1374 };
1375
1376 call_netdevice_notifiers_info(NETDEV_CHANGE,
1377 &change_info.info);
1378 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1379 }
1380 }
1381 EXPORT_SYMBOL(netdev_state_change);
1382
1383 /**
1384 * __netdev_notify_peers - notify network peers about existence of @dev,
1385 * to be called when rtnl lock is already held.
1386 * @dev: network device
1387 *
1388 * Generate traffic such that interested network peers are aware of
1389 * @dev, such as by generating a gratuitous ARP. This may be used when
1390 * a device wants to inform the rest of the network about some sort of
1391 * reconfiguration such as a failover event or virtual machine
1392 * migration.
1393 */
1394 void __netdev_notify_peers(struct net_device *dev)
1395 {
1396 ASSERT_RTNL();
1397 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1398 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1399 }
1400 EXPORT_SYMBOL(__netdev_notify_peers);
1401
1402 /**
1403 * netdev_notify_peers - notify network peers about existence of @dev
1404 * @dev: network device
1405 *
1406 * Generate traffic such that interested network peers are aware of
1407 * @dev, such as by generating a gratuitous ARP. This may be used when
1408 * a device wants to inform the rest of the network about some sort of
1409 * reconfiguration such as a failover event or virtual machine
1410 * migration.
1411 */
1412 void netdev_notify_peers(struct net_device *dev)
1413 {
1414 rtnl_lock();
1415 __netdev_notify_peers(dev);
1416 rtnl_unlock();
1417 }
1418 EXPORT_SYMBOL(netdev_notify_peers);
1419
1420 static int napi_threaded_poll(void *data);
1421
1422 static int napi_kthread_create(struct napi_struct *n)
1423 {
1424 int err = 0;
1425
1426 /* Create and wake up the kthread once to put it in
1427 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1428 * warning and work with loadavg.
1429 */
1430 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1431 n->dev->name, n->napi_id);
1432 if (IS_ERR(n->thread)) {
1433 err = PTR_ERR(n->thread);
1434 pr_err("kthread_run failed with err %d\n", err);
1435 n->thread = NULL;
1436 }
1437
1438 return err;
1439 }
1440
1441 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1442 {
1443 const struct net_device_ops *ops = dev->netdev_ops;
1444 int ret;
1445
1446 ASSERT_RTNL();
1447 dev_addr_check(dev);
1448
1449 if (!netif_device_present(dev)) {
1450 /* may be detached because parent is runtime-suspended */
1451 if (dev->dev.parent)
1452 pm_runtime_resume(dev->dev.parent);
1453 if (!netif_device_present(dev))
1454 return -ENODEV;
1455 }
1456
1457 /* Block netpoll from trying to do any rx path servicing.
1458 * If we don't do this there is a chance ndo_poll_controller
1459 * or ndo_poll may be running while we open the device
1460 */
1461 netpoll_poll_disable(dev);
1462
1463 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1464 ret = notifier_to_errno(ret);
1465 if (ret)
1466 return ret;
1467
1468 set_bit(__LINK_STATE_START, &dev->state);
1469
1470 if (ops->ndo_validate_addr)
1471 ret = ops->ndo_validate_addr(dev);
1472
1473 if (!ret && ops->ndo_open)
1474 ret = ops->ndo_open(dev);
1475
1476 netpoll_poll_enable(dev);
1477
1478 if (ret)
1479 clear_bit(__LINK_STATE_START, &dev->state);
1480 else {
1481 dev->flags |= IFF_UP;
1482 dev_set_rx_mode(dev);
1483 dev_activate(dev);
1484 add_device_randomness(dev->dev_addr, dev->addr_len);
1485 }
1486
1487 return ret;
1488 }
1489
1490 /**
1491 * dev_open - prepare an interface for use.
1492 * @dev: device to open
1493 * @extack: netlink extended ack
1494 *
1495 * Takes a device from down to up state. The device's private open
1496 * function is invoked and then the multicast lists are loaded. Finally
1497 * the device is moved into the up state and a %NETDEV_UP message is
1498 * sent to the netdev notifier chain.
1499 *
1500 * Calling this function on an active interface is a nop. On a failure
1501 * a negative errno code is returned.
1502 */
1503 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1504 {
1505 int ret;
1506
1507 if (dev->flags & IFF_UP)
1508 return 0;
1509
1510 ret = __dev_open(dev, extack);
1511 if (ret < 0)
1512 return ret;
1513
1514 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1515 call_netdevice_notifiers(NETDEV_UP, dev);
1516
1517 return ret;
1518 }
1519 EXPORT_SYMBOL(dev_open);
1520
1521 static void __dev_close_many(struct list_head *head)
1522 {
1523 struct net_device *dev;
1524
1525 ASSERT_RTNL();
1526 might_sleep();
1527
1528 list_for_each_entry(dev, head, close_list) {
1529 /* Temporarily disable netpoll until the interface is down */
1530 netpoll_poll_disable(dev);
1531
1532 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1533
1534 clear_bit(__LINK_STATE_START, &dev->state);
1535
1536 /* Synchronize to scheduled poll. We cannot touch poll list, it
1537 * can be even on different cpu. So just clear netif_running().
1538 *
1539 * dev->stop() will invoke napi_disable() on all of it's
1540 * napi_struct instances on this device.
1541 */
1542 smp_mb__after_atomic(); /* Commit netif_running(). */
1543 }
1544
1545 dev_deactivate_many(head);
1546
1547 list_for_each_entry(dev, head, close_list) {
1548 const struct net_device_ops *ops = dev->netdev_ops;
1549
1550 /*
1551 * Call the device specific close. This cannot fail.
1552 * Only if device is UP
1553 *
1554 * We allow it to be called even after a DETACH hot-plug
1555 * event.
1556 */
1557 if (ops->ndo_stop)
1558 ops->ndo_stop(dev);
1559
1560 dev->flags &= ~IFF_UP;
1561 netpoll_poll_enable(dev);
1562 }
1563 }
1564
1565 static void __dev_close(struct net_device *dev)
1566 {
1567 LIST_HEAD(single);
1568
1569 list_add(&dev->close_list, &single);
1570 __dev_close_many(&single);
1571 list_del(&single);
1572 }
1573
1574 void dev_close_many(struct list_head *head, bool unlink)
1575 {
1576 struct net_device *dev, *tmp;
1577
1578 /* Remove the devices that don't need to be closed */
1579 list_for_each_entry_safe(dev, tmp, head, close_list)
1580 if (!(dev->flags & IFF_UP))
1581 list_del_init(&dev->close_list);
1582
1583 __dev_close_many(head);
1584
1585 list_for_each_entry_safe(dev, tmp, head, close_list) {
1586 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1587 call_netdevice_notifiers(NETDEV_DOWN, dev);
1588 if (unlink)
1589 list_del_init(&dev->close_list);
1590 }
1591 }
1592 EXPORT_SYMBOL(dev_close_many);
1593
1594 /**
1595 * dev_close - shutdown an interface.
1596 * @dev: device to shutdown
1597 *
1598 * This function moves an active device into down state. A
1599 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1600 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1601 * chain.
1602 */
1603 void dev_close(struct net_device *dev)
1604 {
1605 if (dev->flags & IFF_UP) {
1606 LIST_HEAD(single);
1607
1608 list_add(&dev->close_list, &single);
1609 dev_close_many(&single, true);
1610 list_del(&single);
1611 }
1612 }
1613 EXPORT_SYMBOL(dev_close);
1614
1615
1616 /**
1617 * dev_disable_lro - disable Large Receive Offload on a device
1618 * @dev: device
1619 *
1620 * Disable Large Receive Offload (LRO) on a net device. Must be
1621 * called under RTNL. This is needed if received packets may be
1622 * forwarded to another interface.
1623 */
1624 void dev_disable_lro(struct net_device *dev)
1625 {
1626 struct net_device *lower_dev;
1627 struct list_head *iter;
1628
1629 dev->wanted_features &= ~NETIF_F_LRO;
1630 netdev_update_features(dev);
1631
1632 if (unlikely(dev->features & NETIF_F_LRO))
1633 netdev_WARN(dev, "failed to disable LRO!\n");
1634
1635 netdev_for_each_lower_dev(dev, lower_dev, iter)
1636 dev_disable_lro(lower_dev);
1637 }
1638 EXPORT_SYMBOL(dev_disable_lro);
1639
1640 /**
1641 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1642 * @dev: device
1643 *
1644 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1645 * called under RTNL. This is needed if Generic XDP is installed on
1646 * the device.
1647 */
1648 static void dev_disable_gro_hw(struct net_device *dev)
1649 {
1650 dev->wanted_features &= ~NETIF_F_GRO_HW;
1651 netdev_update_features(dev);
1652
1653 if (unlikely(dev->features & NETIF_F_GRO_HW))
1654 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1655 }
1656
1657 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1658 {
1659 #define N(val) \
1660 case NETDEV_##val: \
1661 return "NETDEV_" __stringify(val);
1662 switch (cmd) {
1663 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1664 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1665 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1666 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1667 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1668 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1669 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1670 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1671 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1672 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1673 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1674 N(XDP_FEAT_CHANGE)
1675 }
1676 #undef N
1677 return "UNKNOWN_NETDEV_EVENT";
1678 }
1679 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1680
1681 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1682 struct net_device *dev)
1683 {
1684 struct netdev_notifier_info info = {
1685 .dev = dev,
1686 };
1687
1688 return nb->notifier_call(nb, val, &info);
1689 }
1690
1691 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1692 struct net_device *dev)
1693 {
1694 int err;
1695
1696 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1697 err = notifier_to_errno(err);
1698 if (err)
1699 return err;
1700
1701 if (!(dev->flags & IFF_UP))
1702 return 0;
1703
1704 call_netdevice_notifier(nb, NETDEV_UP, dev);
1705 return 0;
1706 }
1707
1708 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1709 struct net_device *dev)
1710 {
1711 if (dev->flags & IFF_UP) {
1712 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1713 dev);
1714 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1715 }
1716 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1717 }
1718
1719 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1720 struct net *net)
1721 {
1722 struct net_device *dev;
1723 int err;
1724
1725 for_each_netdev(net, dev) {
1726 err = call_netdevice_register_notifiers(nb, dev);
1727 if (err)
1728 goto rollback;
1729 }
1730 return 0;
1731
1732 rollback:
1733 for_each_netdev_continue_reverse(net, dev)
1734 call_netdevice_unregister_notifiers(nb, dev);
1735 return err;
1736 }
1737
1738 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1739 struct net *net)
1740 {
1741 struct net_device *dev;
1742
1743 for_each_netdev(net, dev)
1744 call_netdevice_unregister_notifiers(nb, dev);
1745 }
1746
1747 static int dev_boot_phase = 1;
1748
1749 /**
1750 * register_netdevice_notifier - register a network notifier block
1751 * @nb: notifier
1752 *
1753 * Register a notifier to be called when network device events occur.
1754 * The notifier passed is linked into the kernel structures and must
1755 * not be reused until it has been unregistered. A negative errno code
1756 * is returned on a failure.
1757 *
1758 * When registered all registration and up events are replayed
1759 * to the new notifier to allow device to have a race free
1760 * view of the network device list.
1761 */
1762
1763 int register_netdevice_notifier(struct notifier_block *nb)
1764 {
1765 struct net *net;
1766 int err;
1767
1768 /* Close race with setup_net() and cleanup_net() */
1769 down_write(&pernet_ops_rwsem);
1770 rtnl_lock();
1771 err = raw_notifier_chain_register(&netdev_chain, nb);
1772 if (err)
1773 goto unlock;
1774 if (dev_boot_phase)
1775 goto unlock;
1776 for_each_net(net) {
1777 err = call_netdevice_register_net_notifiers(nb, net);
1778 if (err)
1779 goto rollback;
1780 }
1781
1782 unlock:
1783 rtnl_unlock();
1784 up_write(&pernet_ops_rwsem);
1785 return err;
1786
1787 rollback:
1788 for_each_net_continue_reverse(net)
1789 call_netdevice_unregister_net_notifiers(nb, net);
1790
1791 raw_notifier_chain_unregister(&netdev_chain, nb);
1792 goto unlock;
1793 }
1794 EXPORT_SYMBOL(register_netdevice_notifier);
1795
1796 /**
1797 * unregister_netdevice_notifier - unregister a network notifier block
1798 * @nb: notifier
1799 *
1800 * Unregister a notifier previously registered by
1801 * register_netdevice_notifier(). The notifier is unlinked into the
1802 * kernel structures and may then be reused. A negative errno code
1803 * is returned on a failure.
1804 *
1805 * After unregistering unregister and down device events are synthesized
1806 * for all devices on the device list to the removed notifier to remove
1807 * the need for special case cleanup code.
1808 */
1809
1810 int unregister_netdevice_notifier(struct notifier_block *nb)
1811 {
1812 struct net *net;
1813 int err;
1814
1815 /* Close race with setup_net() and cleanup_net() */
1816 down_write(&pernet_ops_rwsem);
1817 rtnl_lock();
1818 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1819 if (err)
1820 goto unlock;
1821
1822 for_each_net(net)
1823 call_netdevice_unregister_net_notifiers(nb, net);
1824
1825 unlock:
1826 rtnl_unlock();
1827 up_write(&pernet_ops_rwsem);
1828 return err;
1829 }
1830 EXPORT_SYMBOL(unregister_netdevice_notifier);
1831
1832 static int __register_netdevice_notifier_net(struct net *net,
1833 struct notifier_block *nb,
1834 bool ignore_call_fail)
1835 {
1836 int err;
1837
1838 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1839 if (err)
1840 return err;
1841 if (dev_boot_phase)
1842 return 0;
1843
1844 err = call_netdevice_register_net_notifiers(nb, net);
1845 if (err && !ignore_call_fail)
1846 goto chain_unregister;
1847
1848 return 0;
1849
1850 chain_unregister:
1851 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1852 return err;
1853 }
1854
1855 static int __unregister_netdevice_notifier_net(struct net *net,
1856 struct notifier_block *nb)
1857 {
1858 int err;
1859
1860 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1861 if (err)
1862 return err;
1863
1864 call_netdevice_unregister_net_notifiers(nb, net);
1865 return 0;
1866 }
1867
1868 /**
1869 * register_netdevice_notifier_net - register a per-netns network notifier block
1870 * @net: network namespace
1871 * @nb: notifier
1872 *
1873 * Register a notifier to be called when network device events occur.
1874 * The notifier passed is linked into the kernel structures and must
1875 * not be reused until it has been unregistered. A negative errno code
1876 * is returned on a failure.
1877 *
1878 * When registered all registration and up events are replayed
1879 * to the new notifier to allow device to have a race free
1880 * view of the network device list.
1881 */
1882
1883 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1884 {
1885 int err;
1886
1887 rtnl_lock();
1888 err = __register_netdevice_notifier_net(net, nb, false);
1889 rtnl_unlock();
1890 return err;
1891 }
1892 EXPORT_SYMBOL(register_netdevice_notifier_net);
1893
1894 /**
1895 * unregister_netdevice_notifier_net - unregister a per-netns
1896 * network notifier block
1897 * @net: network namespace
1898 * @nb: notifier
1899 *
1900 * Unregister a notifier previously registered by
1901 * register_netdevice_notifier_net(). The notifier is unlinked from the
1902 * kernel structures and may then be reused. A negative errno code
1903 * is returned on a failure.
1904 *
1905 * After unregistering unregister and down device events are synthesized
1906 * for all devices on the device list to the removed notifier to remove
1907 * the need for special case cleanup code.
1908 */
1909
1910 int unregister_netdevice_notifier_net(struct net *net,
1911 struct notifier_block *nb)
1912 {
1913 int err;
1914
1915 rtnl_lock();
1916 err = __unregister_netdevice_notifier_net(net, nb);
1917 rtnl_unlock();
1918 return err;
1919 }
1920 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1921
1922 static void __move_netdevice_notifier_net(struct net *src_net,
1923 struct net *dst_net,
1924 struct notifier_block *nb)
1925 {
1926 __unregister_netdevice_notifier_net(src_net, nb);
1927 __register_netdevice_notifier_net(dst_net, nb, true);
1928 }
1929
1930 int register_netdevice_notifier_dev_net(struct net_device *dev,
1931 struct notifier_block *nb,
1932 struct netdev_net_notifier *nn)
1933 {
1934 int err;
1935
1936 rtnl_lock();
1937 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1938 if (!err) {
1939 nn->nb = nb;
1940 list_add(&nn->list, &dev->net_notifier_list);
1941 }
1942 rtnl_unlock();
1943 return err;
1944 }
1945 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1946
1947 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1948 struct notifier_block *nb,
1949 struct netdev_net_notifier *nn)
1950 {
1951 int err;
1952
1953 rtnl_lock();
1954 list_del(&nn->list);
1955 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1956 rtnl_unlock();
1957 return err;
1958 }
1959 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1960
1961 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1962 struct net *net)
1963 {
1964 struct netdev_net_notifier *nn;
1965
1966 list_for_each_entry(nn, &dev->net_notifier_list, list)
1967 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1968 }
1969
1970 /**
1971 * call_netdevice_notifiers_info - call all network notifier blocks
1972 * @val: value passed unmodified to notifier function
1973 * @info: notifier information data
1974 *
1975 * Call all network notifier blocks. Parameters and return value
1976 * are as for raw_notifier_call_chain().
1977 */
1978
1979 int call_netdevice_notifiers_info(unsigned long val,
1980 struct netdev_notifier_info *info)
1981 {
1982 struct net *net = dev_net(info->dev);
1983 int ret;
1984
1985 ASSERT_RTNL();
1986
1987 /* Run per-netns notifier block chain first, then run the global one.
1988 * Hopefully, one day, the global one is going to be removed after
1989 * all notifier block registrators get converted to be per-netns.
1990 */
1991 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1992 if (ret & NOTIFY_STOP_MASK)
1993 return ret;
1994 return raw_notifier_call_chain(&netdev_chain, val, info);
1995 }
1996
1997 /**
1998 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1999 * for and rollback on error
2000 * @val_up: value passed unmodified to notifier function
2001 * @val_down: value passed unmodified to the notifier function when
2002 * recovering from an error on @val_up
2003 * @info: notifier information data
2004 *
2005 * Call all per-netns network notifier blocks, but not notifier blocks on
2006 * the global notifier chain. Parameters and return value are as for
2007 * raw_notifier_call_chain_robust().
2008 */
2009
2010 static int
2011 call_netdevice_notifiers_info_robust(unsigned long val_up,
2012 unsigned long val_down,
2013 struct netdev_notifier_info *info)
2014 {
2015 struct net *net = dev_net(info->dev);
2016
2017 ASSERT_RTNL();
2018
2019 return raw_notifier_call_chain_robust(&net->netdev_chain,
2020 val_up, val_down, info);
2021 }
2022
2023 static int call_netdevice_notifiers_extack(unsigned long val,
2024 struct net_device *dev,
2025 struct netlink_ext_ack *extack)
2026 {
2027 struct netdev_notifier_info info = {
2028 .dev = dev,
2029 .extack = extack,
2030 };
2031
2032 return call_netdevice_notifiers_info(val, &info);
2033 }
2034
2035 /**
2036 * call_netdevice_notifiers - call all network notifier blocks
2037 * @val: value passed unmodified to notifier function
2038 * @dev: net_device pointer passed unmodified to notifier function
2039 *
2040 * Call all network notifier blocks. Parameters and return value
2041 * are as for raw_notifier_call_chain().
2042 */
2043
2044 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2045 {
2046 return call_netdevice_notifiers_extack(val, dev, NULL);
2047 }
2048 EXPORT_SYMBOL(call_netdevice_notifiers);
2049
2050 /**
2051 * call_netdevice_notifiers_mtu - call all network notifier blocks
2052 * @val: value passed unmodified to notifier function
2053 * @dev: net_device pointer passed unmodified to notifier function
2054 * @arg: additional u32 argument passed to the notifier function
2055 *
2056 * Call all network notifier blocks. Parameters and return value
2057 * are as for raw_notifier_call_chain().
2058 */
2059 static int call_netdevice_notifiers_mtu(unsigned long val,
2060 struct net_device *dev, u32 arg)
2061 {
2062 struct netdev_notifier_info_ext info = {
2063 .info.dev = dev,
2064 .ext.mtu = arg,
2065 };
2066
2067 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2068
2069 return call_netdevice_notifiers_info(val, &info.info);
2070 }
2071
2072 #ifdef CONFIG_NET_INGRESS
2073 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2074
2075 void net_inc_ingress_queue(void)
2076 {
2077 static_branch_inc(&ingress_needed_key);
2078 }
2079 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2080
2081 void net_dec_ingress_queue(void)
2082 {
2083 static_branch_dec(&ingress_needed_key);
2084 }
2085 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2086 #endif
2087
2088 #ifdef CONFIG_NET_EGRESS
2089 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2090
2091 void net_inc_egress_queue(void)
2092 {
2093 static_branch_inc(&egress_needed_key);
2094 }
2095 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2096
2097 void net_dec_egress_queue(void)
2098 {
2099 static_branch_dec(&egress_needed_key);
2100 }
2101 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2102 #endif
2103
2104 #ifdef CONFIG_NET_CLS_ACT
2105 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key);
2106 EXPORT_SYMBOL(tcf_bypass_check_needed_key);
2107 #endif
2108
2109 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2110 EXPORT_SYMBOL(netstamp_needed_key);
2111 #ifdef CONFIG_JUMP_LABEL
2112 static atomic_t netstamp_needed_deferred;
2113 static atomic_t netstamp_wanted;
2114 static void netstamp_clear(struct work_struct *work)
2115 {
2116 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2117 int wanted;
2118
2119 wanted = atomic_add_return(deferred, &netstamp_wanted);
2120 if (wanted > 0)
2121 static_branch_enable(&netstamp_needed_key);
2122 else
2123 static_branch_disable(&netstamp_needed_key);
2124 }
2125 static DECLARE_WORK(netstamp_work, netstamp_clear);
2126 #endif
2127
2128 void net_enable_timestamp(void)
2129 {
2130 #ifdef CONFIG_JUMP_LABEL
2131 int wanted = atomic_read(&netstamp_wanted);
2132
2133 while (wanted > 0) {
2134 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2135 return;
2136 }
2137 atomic_inc(&netstamp_needed_deferred);
2138 schedule_work(&netstamp_work);
2139 #else
2140 static_branch_inc(&netstamp_needed_key);
2141 #endif
2142 }
2143 EXPORT_SYMBOL(net_enable_timestamp);
2144
2145 void net_disable_timestamp(void)
2146 {
2147 #ifdef CONFIG_JUMP_LABEL
2148 int wanted = atomic_read(&netstamp_wanted);
2149
2150 while (wanted > 1) {
2151 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2152 return;
2153 }
2154 atomic_dec(&netstamp_needed_deferred);
2155 schedule_work(&netstamp_work);
2156 #else
2157 static_branch_dec(&netstamp_needed_key);
2158 #endif
2159 }
2160 EXPORT_SYMBOL(net_disable_timestamp);
2161
2162 static inline void net_timestamp_set(struct sk_buff *skb)
2163 {
2164 skb->tstamp = 0;
2165 skb->tstamp_type = SKB_CLOCK_REALTIME;
2166 if (static_branch_unlikely(&netstamp_needed_key))
2167 skb->tstamp = ktime_get_real();
2168 }
2169
2170 #define net_timestamp_check(COND, SKB) \
2171 if (static_branch_unlikely(&netstamp_needed_key)) { \
2172 if ((COND) && !(SKB)->tstamp) \
2173 (SKB)->tstamp = ktime_get_real(); \
2174 } \
2175
2176 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2177 {
2178 return __is_skb_forwardable(dev, skb, true);
2179 }
2180 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2181
2182 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2183 bool check_mtu)
2184 {
2185 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2186
2187 if (likely(!ret)) {
2188 skb->protocol = eth_type_trans(skb, dev);
2189 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2190 }
2191
2192 return ret;
2193 }
2194
2195 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2196 {
2197 return __dev_forward_skb2(dev, skb, true);
2198 }
2199 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2200
2201 /**
2202 * dev_forward_skb - loopback an skb to another netif
2203 *
2204 * @dev: destination network device
2205 * @skb: buffer to forward
2206 *
2207 * return values:
2208 * NET_RX_SUCCESS (no congestion)
2209 * NET_RX_DROP (packet was dropped, but freed)
2210 *
2211 * dev_forward_skb can be used for injecting an skb from the
2212 * start_xmit function of one device into the receive queue
2213 * of another device.
2214 *
2215 * The receiving device may be in another namespace, so
2216 * we have to clear all information in the skb that could
2217 * impact namespace isolation.
2218 */
2219 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2220 {
2221 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2222 }
2223 EXPORT_SYMBOL_GPL(dev_forward_skb);
2224
2225 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2226 {
2227 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2228 }
2229
2230 static inline int deliver_skb(struct sk_buff *skb,
2231 struct packet_type *pt_prev,
2232 struct net_device *orig_dev)
2233 {
2234 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2235 return -ENOMEM;
2236 refcount_inc(&skb->users);
2237 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2238 }
2239
2240 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2241 struct packet_type **pt,
2242 struct net_device *orig_dev,
2243 __be16 type,
2244 struct list_head *ptype_list)
2245 {
2246 struct packet_type *ptype, *pt_prev = *pt;
2247
2248 list_for_each_entry_rcu(ptype, ptype_list, list) {
2249 if (ptype->type != type)
2250 continue;
2251 if (pt_prev)
2252 deliver_skb(skb, pt_prev, orig_dev);
2253 pt_prev = ptype;
2254 }
2255 *pt = pt_prev;
2256 }
2257
2258 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2259 {
2260 if (!ptype->af_packet_priv || !skb->sk)
2261 return false;
2262
2263 if (ptype->id_match)
2264 return ptype->id_match(ptype, skb->sk);
2265 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2266 return true;
2267
2268 return false;
2269 }
2270
2271 /**
2272 * dev_nit_active - return true if any network interface taps are in use
2273 *
2274 * @dev: network device to check for the presence of taps
2275 */
2276 bool dev_nit_active(struct net_device *dev)
2277 {
2278 return !list_empty(&net_hotdata.ptype_all) ||
2279 !list_empty(&dev->ptype_all);
2280 }
2281 EXPORT_SYMBOL_GPL(dev_nit_active);
2282
2283 /*
2284 * Support routine. Sends outgoing frames to any network
2285 * taps currently in use.
2286 */
2287
2288 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2289 {
2290 struct list_head *ptype_list = &net_hotdata.ptype_all;
2291 struct packet_type *ptype, *pt_prev = NULL;
2292 struct sk_buff *skb2 = NULL;
2293
2294 rcu_read_lock();
2295 again:
2296 list_for_each_entry_rcu(ptype, ptype_list, list) {
2297 if (READ_ONCE(ptype->ignore_outgoing))
2298 continue;
2299
2300 /* Never send packets back to the socket
2301 * they originated from - MvS (miquels@drinkel.ow.org)
2302 */
2303 if (skb_loop_sk(ptype, skb))
2304 continue;
2305
2306 if (pt_prev) {
2307 deliver_skb(skb2, pt_prev, skb->dev);
2308 pt_prev = ptype;
2309 continue;
2310 }
2311
2312 /* need to clone skb, done only once */
2313 skb2 = skb_clone(skb, GFP_ATOMIC);
2314 if (!skb2)
2315 goto out_unlock;
2316
2317 net_timestamp_set(skb2);
2318
2319 /* skb->nh should be correctly
2320 * set by sender, so that the second statement is
2321 * just protection against buggy protocols.
2322 */
2323 skb_reset_mac_header(skb2);
2324
2325 if (skb_network_header(skb2) < skb2->data ||
2326 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2327 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2328 ntohs(skb2->protocol),
2329 dev->name);
2330 skb_reset_network_header(skb2);
2331 }
2332
2333 skb2->transport_header = skb2->network_header;
2334 skb2->pkt_type = PACKET_OUTGOING;
2335 pt_prev = ptype;
2336 }
2337
2338 if (ptype_list == &net_hotdata.ptype_all) {
2339 ptype_list = &dev->ptype_all;
2340 goto again;
2341 }
2342 out_unlock:
2343 if (pt_prev) {
2344 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2345 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2346 else
2347 kfree_skb(skb2);
2348 }
2349 rcu_read_unlock();
2350 }
2351 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2352
2353 /**
2354 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2355 * @dev: Network device
2356 * @txq: number of queues available
2357 *
2358 * If real_num_tx_queues is changed the tc mappings may no longer be
2359 * valid. To resolve this verify the tc mapping remains valid and if
2360 * not NULL the mapping. With no priorities mapping to this
2361 * offset/count pair it will no longer be used. In the worst case TC0
2362 * is invalid nothing can be done so disable priority mappings. If is
2363 * expected that drivers will fix this mapping if they can before
2364 * calling netif_set_real_num_tx_queues.
2365 */
2366 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2367 {
2368 int i;
2369 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2370
2371 /* If TC0 is invalidated disable TC mapping */
2372 if (tc->offset + tc->count > txq) {
2373 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2374 dev->num_tc = 0;
2375 return;
2376 }
2377
2378 /* Invalidated prio to tc mappings set to TC0 */
2379 for (i = 1; i < TC_BITMASK + 1; i++) {
2380 int q = netdev_get_prio_tc_map(dev, i);
2381
2382 tc = &dev->tc_to_txq[q];
2383 if (tc->offset + tc->count > txq) {
2384 netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2385 i, q);
2386 netdev_set_prio_tc_map(dev, i, 0);
2387 }
2388 }
2389 }
2390
2391 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2392 {
2393 if (dev->num_tc) {
2394 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2395 int i;
2396
2397 /* walk through the TCs and see if it falls into any of them */
2398 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2399 if ((txq - tc->offset) < tc->count)
2400 return i;
2401 }
2402
2403 /* didn't find it, just return -1 to indicate no match */
2404 return -1;
2405 }
2406
2407 return 0;
2408 }
2409 EXPORT_SYMBOL(netdev_txq_to_tc);
2410
2411 #ifdef CONFIG_XPS
2412 static struct static_key xps_needed __read_mostly;
2413 static struct static_key xps_rxqs_needed __read_mostly;
2414 static DEFINE_MUTEX(xps_map_mutex);
2415 #define xmap_dereference(P) \
2416 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2417
2418 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2419 struct xps_dev_maps *old_maps, int tci, u16 index)
2420 {
2421 struct xps_map *map = NULL;
2422 int pos;
2423
2424 map = xmap_dereference(dev_maps->attr_map[tci]);
2425 if (!map)
2426 return false;
2427
2428 for (pos = map->len; pos--;) {
2429 if (map->queues[pos] != index)
2430 continue;
2431
2432 if (map->len > 1) {
2433 map->queues[pos] = map->queues[--map->len];
2434 break;
2435 }
2436
2437 if (old_maps)
2438 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2439 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2440 kfree_rcu(map, rcu);
2441 return false;
2442 }
2443
2444 return true;
2445 }
2446
2447 static bool remove_xps_queue_cpu(struct net_device *dev,
2448 struct xps_dev_maps *dev_maps,
2449 int cpu, u16 offset, u16 count)
2450 {
2451 int num_tc = dev_maps->num_tc;
2452 bool active = false;
2453 int tci;
2454
2455 for (tci = cpu * num_tc; num_tc--; tci++) {
2456 int i, j;
2457
2458 for (i = count, j = offset; i--; j++) {
2459 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2460 break;
2461 }
2462
2463 active |= i < 0;
2464 }
2465
2466 return active;
2467 }
2468
2469 static void reset_xps_maps(struct net_device *dev,
2470 struct xps_dev_maps *dev_maps,
2471 enum xps_map_type type)
2472 {
2473 static_key_slow_dec_cpuslocked(&xps_needed);
2474 if (type == XPS_RXQS)
2475 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2476
2477 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2478
2479 kfree_rcu(dev_maps, rcu);
2480 }
2481
2482 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2483 u16 offset, u16 count)
2484 {
2485 struct xps_dev_maps *dev_maps;
2486 bool active = false;
2487 int i, j;
2488
2489 dev_maps = xmap_dereference(dev->xps_maps[type]);
2490 if (!dev_maps)
2491 return;
2492
2493 for (j = 0; j < dev_maps->nr_ids; j++)
2494 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2495 if (!active)
2496 reset_xps_maps(dev, dev_maps, type);
2497
2498 if (type == XPS_CPUS) {
2499 for (i = offset + (count - 1); count--; i--)
2500 netdev_queue_numa_node_write(
2501 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2502 }
2503 }
2504
2505 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2506 u16 count)
2507 {
2508 if (!static_key_false(&xps_needed))
2509 return;
2510
2511 cpus_read_lock();
2512 mutex_lock(&xps_map_mutex);
2513
2514 if (static_key_false(&xps_rxqs_needed))
2515 clean_xps_maps(dev, XPS_RXQS, offset, count);
2516
2517 clean_xps_maps(dev, XPS_CPUS, offset, count);
2518
2519 mutex_unlock(&xps_map_mutex);
2520 cpus_read_unlock();
2521 }
2522
2523 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2524 {
2525 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2526 }
2527
2528 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2529 u16 index, bool is_rxqs_map)
2530 {
2531 struct xps_map *new_map;
2532 int alloc_len = XPS_MIN_MAP_ALLOC;
2533 int i, pos;
2534
2535 for (pos = 0; map && pos < map->len; pos++) {
2536 if (map->queues[pos] != index)
2537 continue;
2538 return map;
2539 }
2540
2541 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2542 if (map) {
2543 if (pos < map->alloc_len)
2544 return map;
2545
2546 alloc_len = map->alloc_len * 2;
2547 }
2548
2549 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2550 * map
2551 */
2552 if (is_rxqs_map)
2553 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2554 else
2555 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2556 cpu_to_node(attr_index));
2557 if (!new_map)
2558 return NULL;
2559
2560 for (i = 0; i < pos; i++)
2561 new_map->queues[i] = map->queues[i];
2562 new_map->alloc_len = alloc_len;
2563 new_map->len = pos;
2564
2565 return new_map;
2566 }
2567
2568 /* Copy xps maps at a given index */
2569 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2570 struct xps_dev_maps *new_dev_maps, int index,
2571 int tc, bool skip_tc)
2572 {
2573 int i, tci = index * dev_maps->num_tc;
2574 struct xps_map *map;
2575
2576 /* copy maps belonging to foreign traffic classes */
2577 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2578 if (i == tc && skip_tc)
2579 continue;
2580
2581 /* fill in the new device map from the old device map */
2582 map = xmap_dereference(dev_maps->attr_map[tci]);
2583 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2584 }
2585 }
2586
2587 /* Must be called under cpus_read_lock */
2588 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2589 u16 index, enum xps_map_type type)
2590 {
2591 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2592 const unsigned long *online_mask = NULL;
2593 bool active = false, copy = false;
2594 int i, j, tci, numa_node_id = -2;
2595 int maps_sz, num_tc = 1, tc = 0;
2596 struct xps_map *map, *new_map;
2597 unsigned int nr_ids;
2598
2599 WARN_ON_ONCE(index >= dev->num_tx_queues);
2600
2601 if (dev->num_tc) {
2602 /* Do not allow XPS on subordinate device directly */
2603 num_tc = dev->num_tc;
2604 if (num_tc < 0)
2605 return -EINVAL;
2606
2607 /* If queue belongs to subordinate dev use its map */
2608 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2609
2610 tc = netdev_txq_to_tc(dev, index);
2611 if (tc < 0)
2612 return -EINVAL;
2613 }
2614
2615 mutex_lock(&xps_map_mutex);
2616
2617 dev_maps = xmap_dereference(dev->xps_maps[type]);
2618 if (type == XPS_RXQS) {
2619 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2620 nr_ids = dev->num_rx_queues;
2621 } else {
2622 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2623 if (num_possible_cpus() > 1)
2624 online_mask = cpumask_bits(cpu_online_mask);
2625 nr_ids = nr_cpu_ids;
2626 }
2627
2628 if (maps_sz < L1_CACHE_BYTES)
2629 maps_sz = L1_CACHE_BYTES;
2630
2631 /* The old dev_maps could be larger or smaller than the one we're
2632 * setting up now, as dev->num_tc or nr_ids could have been updated in
2633 * between. We could try to be smart, but let's be safe instead and only
2634 * copy foreign traffic classes if the two map sizes match.
2635 */
2636 if (dev_maps &&
2637 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2638 copy = true;
2639
2640 /* allocate memory for queue storage */
2641 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2642 j < nr_ids;) {
2643 if (!new_dev_maps) {
2644 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2645 if (!new_dev_maps) {
2646 mutex_unlock(&xps_map_mutex);
2647 return -ENOMEM;
2648 }
2649
2650 new_dev_maps->nr_ids = nr_ids;
2651 new_dev_maps->num_tc = num_tc;
2652 }
2653
2654 tci = j * num_tc + tc;
2655 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2656
2657 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2658 if (!map)
2659 goto error;
2660
2661 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2662 }
2663
2664 if (!new_dev_maps)
2665 goto out_no_new_maps;
2666
2667 if (!dev_maps) {
2668 /* Increment static keys at most once per type */
2669 static_key_slow_inc_cpuslocked(&xps_needed);
2670 if (type == XPS_RXQS)
2671 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2672 }
2673
2674 for (j = 0; j < nr_ids; j++) {
2675 bool skip_tc = false;
2676
2677 tci = j * num_tc + tc;
2678 if (netif_attr_test_mask(j, mask, nr_ids) &&
2679 netif_attr_test_online(j, online_mask, nr_ids)) {
2680 /* add tx-queue to CPU/rx-queue maps */
2681 int pos = 0;
2682
2683 skip_tc = true;
2684
2685 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2686 while ((pos < map->len) && (map->queues[pos] != index))
2687 pos++;
2688
2689 if (pos == map->len)
2690 map->queues[map->len++] = index;
2691 #ifdef CONFIG_NUMA
2692 if (type == XPS_CPUS) {
2693 if (numa_node_id == -2)
2694 numa_node_id = cpu_to_node(j);
2695 else if (numa_node_id != cpu_to_node(j))
2696 numa_node_id = -1;
2697 }
2698 #endif
2699 }
2700
2701 if (copy)
2702 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2703 skip_tc);
2704 }
2705
2706 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2707
2708 /* Cleanup old maps */
2709 if (!dev_maps)
2710 goto out_no_old_maps;
2711
2712 for (j = 0; j < dev_maps->nr_ids; j++) {
2713 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2714 map = xmap_dereference(dev_maps->attr_map[tci]);
2715 if (!map)
2716 continue;
2717
2718 if (copy) {
2719 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2720 if (map == new_map)
2721 continue;
2722 }
2723
2724 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2725 kfree_rcu(map, rcu);
2726 }
2727 }
2728
2729 old_dev_maps = dev_maps;
2730
2731 out_no_old_maps:
2732 dev_maps = new_dev_maps;
2733 active = true;
2734
2735 out_no_new_maps:
2736 if (type == XPS_CPUS)
2737 /* update Tx queue numa node */
2738 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2739 (numa_node_id >= 0) ?
2740 numa_node_id : NUMA_NO_NODE);
2741
2742 if (!dev_maps)
2743 goto out_no_maps;
2744
2745 /* removes tx-queue from unused CPUs/rx-queues */
2746 for (j = 0; j < dev_maps->nr_ids; j++) {
2747 tci = j * dev_maps->num_tc;
2748
2749 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2750 if (i == tc &&
2751 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2752 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2753 continue;
2754
2755 active |= remove_xps_queue(dev_maps,
2756 copy ? old_dev_maps : NULL,
2757 tci, index);
2758 }
2759 }
2760
2761 if (old_dev_maps)
2762 kfree_rcu(old_dev_maps, rcu);
2763
2764 /* free map if not active */
2765 if (!active)
2766 reset_xps_maps(dev, dev_maps, type);
2767
2768 out_no_maps:
2769 mutex_unlock(&xps_map_mutex);
2770
2771 return 0;
2772 error:
2773 /* remove any maps that we added */
2774 for (j = 0; j < nr_ids; j++) {
2775 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2776 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2777 map = copy ?
2778 xmap_dereference(dev_maps->attr_map[tci]) :
2779 NULL;
2780 if (new_map && new_map != map)
2781 kfree(new_map);
2782 }
2783 }
2784
2785 mutex_unlock(&xps_map_mutex);
2786
2787 kfree(new_dev_maps);
2788 return -ENOMEM;
2789 }
2790 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2791
2792 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2793 u16 index)
2794 {
2795 int ret;
2796
2797 cpus_read_lock();
2798 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2799 cpus_read_unlock();
2800
2801 return ret;
2802 }
2803 EXPORT_SYMBOL(netif_set_xps_queue);
2804
2805 #endif
2806 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2807 {
2808 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2809
2810 /* Unbind any subordinate channels */
2811 while (txq-- != &dev->_tx[0]) {
2812 if (txq->sb_dev)
2813 netdev_unbind_sb_channel(dev, txq->sb_dev);
2814 }
2815 }
2816
2817 void netdev_reset_tc(struct net_device *dev)
2818 {
2819 #ifdef CONFIG_XPS
2820 netif_reset_xps_queues_gt(dev, 0);
2821 #endif
2822 netdev_unbind_all_sb_channels(dev);
2823
2824 /* Reset TC configuration of device */
2825 dev->num_tc = 0;
2826 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2827 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2828 }
2829 EXPORT_SYMBOL(netdev_reset_tc);
2830
2831 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2832 {
2833 if (tc >= dev->num_tc)
2834 return -EINVAL;
2835
2836 #ifdef CONFIG_XPS
2837 netif_reset_xps_queues(dev, offset, count);
2838 #endif
2839 dev->tc_to_txq[tc].count = count;
2840 dev->tc_to_txq[tc].offset = offset;
2841 return 0;
2842 }
2843 EXPORT_SYMBOL(netdev_set_tc_queue);
2844
2845 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2846 {
2847 if (num_tc > TC_MAX_QUEUE)
2848 return -EINVAL;
2849
2850 #ifdef CONFIG_XPS
2851 netif_reset_xps_queues_gt(dev, 0);
2852 #endif
2853 netdev_unbind_all_sb_channels(dev);
2854
2855 dev->num_tc = num_tc;
2856 return 0;
2857 }
2858 EXPORT_SYMBOL(netdev_set_num_tc);
2859
2860 void netdev_unbind_sb_channel(struct net_device *dev,
2861 struct net_device *sb_dev)
2862 {
2863 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2864
2865 #ifdef CONFIG_XPS
2866 netif_reset_xps_queues_gt(sb_dev, 0);
2867 #endif
2868 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2869 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2870
2871 while (txq-- != &dev->_tx[0]) {
2872 if (txq->sb_dev == sb_dev)
2873 txq->sb_dev = NULL;
2874 }
2875 }
2876 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2877
2878 int netdev_bind_sb_channel_queue(struct net_device *dev,
2879 struct net_device *sb_dev,
2880 u8 tc, u16 count, u16 offset)
2881 {
2882 /* Make certain the sb_dev and dev are already configured */
2883 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2884 return -EINVAL;
2885
2886 /* We cannot hand out queues we don't have */
2887 if ((offset + count) > dev->real_num_tx_queues)
2888 return -EINVAL;
2889
2890 /* Record the mapping */
2891 sb_dev->tc_to_txq[tc].count = count;
2892 sb_dev->tc_to_txq[tc].offset = offset;
2893
2894 /* Provide a way for Tx queue to find the tc_to_txq map or
2895 * XPS map for itself.
2896 */
2897 while (count--)
2898 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2899
2900 return 0;
2901 }
2902 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2903
2904 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2905 {
2906 /* Do not use a multiqueue device to represent a subordinate channel */
2907 if (netif_is_multiqueue(dev))
2908 return -ENODEV;
2909
2910 /* We allow channels 1 - 32767 to be used for subordinate channels.
2911 * Channel 0 is meant to be "native" mode and used only to represent
2912 * the main root device. We allow writing 0 to reset the device back
2913 * to normal mode after being used as a subordinate channel.
2914 */
2915 if (channel > S16_MAX)
2916 return -EINVAL;
2917
2918 dev->num_tc = -channel;
2919
2920 return 0;
2921 }
2922 EXPORT_SYMBOL(netdev_set_sb_channel);
2923
2924 /*
2925 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2926 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2927 */
2928 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2929 {
2930 bool disabling;
2931 int rc;
2932
2933 disabling = txq < dev->real_num_tx_queues;
2934
2935 if (txq < 1 || txq > dev->num_tx_queues)
2936 return -EINVAL;
2937
2938 if (dev->reg_state == NETREG_REGISTERED ||
2939 dev->reg_state == NETREG_UNREGISTERING) {
2940 ASSERT_RTNL();
2941
2942 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2943 txq);
2944 if (rc)
2945 return rc;
2946
2947 if (dev->num_tc)
2948 netif_setup_tc(dev, txq);
2949
2950 dev_qdisc_change_real_num_tx(dev, txq);
2951
2952 dev->real_num_tx_queues = txq;
2953
2954 if (disabling) {
2955 synchronize_net();
2956 qdisc_reset_all_tx_gt(dev, txq);
2957 #ifdef CONFIG_XPS
2958 netif_reset_xps_queues_gt(dev, txq);
2959 #endif
2960 }
2961 } else {
2962 dev->real_num_tx_queues = txq;
2963 }
2964
2965 return 0;
2966 }
2967 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2968
2969 #ifdef CONFIG_SYSFS
2970 /**
2971 * netif_set_real_num_rx_queues - set actual number of RX queues used
2972 * @dev: Network device
2973 * @rxq: Actual number of RX queues
2974 *
2975 * This must be called either with the rtnl_lock held or before
2976 * registration of the net device. Returns 0 on success, or a
2977 * negative error code. If called before registration, it always
2978 * succeeds.
2979 */
2980 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2981 {
2982 int rc;
2983
2984 if (rxq < 1 || rxq > dev->num_rx_queues)
2985 return -EINVAL;
2986
2987 if (dev->reg_state == NETREG_REGISTERED) {
2988 ASSERT_RTNL();
2989
2990 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2991 rxq);
2992 if (rc)
2993 return rc;
2994 }
2995
2996 dev->real_num_rx_queues = rxq;
2997 return 0;
2998 }
2999 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3000 #endif
3001
3002 /**
3003 * netif_set_real_num_queues - set actual number of RX and TX queues used
3004 * @dev: Network device
3005 * @txq: Actual number of TX queues
3006 * @rxq: Actual number of RX queues
3007 *
3008 * Set the real number of both TX and RX queues.
3009 * Does nothing if the number of queues is already correct.
3010 */
3011 int netif_set_real_num_queues(struct net_device *dev,
3012 unsigned int txq, unsigned int rxq)
3013 {
3014 unsigned int old_rxq = dev->real_num_rx_queues;
3015 int err;
3016
3017 if (txq < 1 || txq > dev->num_tx_queues ||
3018 rxq < 1 || rxq > dev->num_rx_queues)
3019 return -EINVAL;
3020
3021 /* Start from increases, so the error path only does decreases -
3022 * decreases can't fail.
3023 */
3024 if (rxq > dev->real_num_rx_queues) {
3025 err = netif_set_real_num_rx_queues(dev, rxq);
3026 if (err)
3027 return err;
3028 }
3029 if (txq > dev->real_num_tx_queues) {
3030 err = netif_set_real_num_tx_queues(dev, txq);
3031 if (err)
3032 goto undo_rx;
3033 }
3034 if (rxq < dev->real_num_rx_queues)
3035 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3036 if (txq < dev->real_num_tx_queues)
3037 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3038
3039 return 0;
3040 undo_rx:
3041 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3042 return err;
3043 }
3044 EXPORT_SYMBOL(netif_set_real_num_queues);
3045
3046 /**
3047 * netif_set_tso_max_size() - set the max size of TSO frames supported
3048 * @dev: netdev to update
3049 * @size: max skb->len of a TSO frame
3050 *
3051 * Set the limit on the size of TSO super-frames the device can handle.
3052 * Unless explicitly set the stack will assume the value of
3053 * %GSO_LEGACY_MAX_SIZE.
3054 */
3055 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3056 {
3057 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3058 if (size < READ_ONCE(dev->gso_max_size))
3059 netif_set_gso_max_size(dev, size);
3060 if (size < READ_ONCE(dev->gso_ipv4_max_size))
3061 netif_set_gso_ipv4_max_size(dev, size);
3062 }
3063 EXPORT_SYMBOL(netif_set_tso_max_size);
3064
3065 /**
3066 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3067 * @dev: netdev to update
3068 * @segs: max number of TCP segments
3069 *
3070 * Set the limit on the number of TCP segments the device can generate from
3071 * a single TSO super-frame.
3072 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3073 */
3074 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3075 {
3076 dev->tso_max_segs = segs;
3077 if (segs < READ_ONCE(dev->gso_max_segs))
3078 netif_set_gso_max_segs(dev, segs);
3079 }
3080 EXPORT_SYMBOL(netif_set_tso_max_segs);
3081
3082 /**
3083 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3084 * @to: netdev to update
3085 * @from: netdev from which to copy the limits
3086 */
3087 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3088 {
3089 netif_set_tso_max_size(to, from->tso_max_size);
3090 netif_set_tso_max_segs(to, from->tso_max_segs);
3091 }
3092 EXPORT_SYMBOL(netif_inherit_tso_max);
3093
3094 /**
3095 * netif_get_num_default_rss_queues - default number of RSS queues
3096 *
3097 * Default value is the number of physical cores if there are only 1 or 2, or
3098 * divided by 2 if there are more.
3099 */
3100 int netif_get_num_default_rss_queues(void)
3101 {
3102 cpumask_var_t cpus;
3103 int cpu, count = 0;
3104
3105 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3106 return 1;
3107
3108 cpumask_copy(cpus, cpu_online_mask);
3109 for_each_cpu(cpu, cpus) {
3110 ++count;
3111 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3112 }
3113 free_cpumask_var(cpus);
3114
3115 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3116 }
3117 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3118
3119 static void __netif_reschedule(struct Qdisc *q)
3120 {
3121 struct softnet_data *sd;
3122 unsigned long flags;
3123
3124 local_irq_save(flags);
3125 sd = this_cpu_ptr(&softnet_data);
3126 q->next_sched = NULL;
3127 *sd->output_queue_tailp = q;
3128 sd->output_queue_tailp = &q->next_sched;
3129 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3130 local_irq_restore(flags);
3131 }
3132
3133 void __netif_schedule(struct Qdisc *q)
3134 {
3135 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3136 __netif_reschedule(q);
3137 }
3138 EXPORT_SYMBOL(__netif_schedule);
3139
3140 struct dev_kfree_skb_cb {
3141 enum skb_drop_reason reason;
3142 };
3143
3144 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3145 {
3146 return (struct dev_kfree_skb_cb *)skb->cb;
3147 }
3148
3149 void netif_schedule_queue(struct netdev_queue *txq)
3150 {
3151 rcu_read_lock();
3152 if (!netif_xmit_stopped(txq)) {
3153 struct Qdisc *q = rcu_dereference(txq->qdisc);
3154
3155 __netif_schedule(q);
3156 }
3157 rcu_read_unlock();
3158 }
3159 EXPORT_SYMBOL(netif_schedule_queue);
3160
3161 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3162 {
3163 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3164 struct Qdisc *q;
3165
3166 rcu_read_lock();
3167 q = rcu_dereference(dev_queue->qdisc);
3168 __netif_schedule(q);
3169 rcu_read_unlock();
3170 }
3171 }
3172 EXPORT_SYMBOL(netif_tx_wake_queue);
3173
3174 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3175 {
3176 unsigned long flags;
3177
3178 if (unlikely(!skb))
3179 return;
3180
3181 if (likely(refcount_read(&skb->users) == 1)) {
3182 smp_rmb();
3183 refcount_set(&skb->users, 0);
3184 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3185 return;
3186 }
3187 get_kfree_skb_cb(skb)->reason = reason;
3188 local_irq_save(flags);
3189 skb->next = __this_cpu_read(softnet_data.completion_queue);
3190 __this_cpu_write(softnet_data.completion_queue, skb);
3191 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3192 local_irq_restore(flags);
3193 }
3194 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3195
3196 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3197 {
3198 if (in_hardirq() || irqs_disabled())
3199 dev_kfree_skb_irq_reason(skb, reason);
3200 else
3201 kfree_skb_reason(skb, reason);
3202 }
3203 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3204
3205
3206 /**
3207 * netif_device_detach - mark device as removed
3208 * @dev: network device
3209 *
3210 * Mark device as removed from system and therefore no longer available.
3211 */
3212 void netif_device_detach(struct net_device *dev)
3213 {
3214 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3215 netif_running(dev)) {
3216 netif_tx_stop_all_queues(dev);
3217 }
3218 }
3219 EXPORT_SYMBOL(netif_device_detach);
3220
3221 /**
3222 * netif_device_attach - mark device as attached
3223 * @dev: network device
3224 *
3225 * Mark device as attached from system and restart if needed.
3226 */
3227 void netif_device_attach(struct net_device *dev)
3228 {
3229 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3230 netif_running(dev)) {
3231 netif_tx_wake_all_queues(dev);
3232 __netdev_watchdog_up(dev);
3233 }
3234 }
3235 EXPORT_SYMBOL(netif_device_attach);
3236
3237 /*
3238 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3239 * to be used as a distribution range.
3240 */
3241 static u16 skb_tx_hash(const struct net_device *dev,
3242 const struct net_device *sb_dev,
3243 struct sk_buff *skb)
3244 {
3245 u32 hash;
3246 u16 qoffset = 0;
3247 u16 qcount = dev->real_num_tx_queues;
3248
3249 if (dev->num_tc) {
3250 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3251
3252 qoffset = sb_dev->tc_to_txq[tc].offset;
3253 qcount = sb_dev->tc_to_txq[tc].count;
3254 if (unlikely(!qcount)) {
3255 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3256 sb_dev->name, qoffset, tc);
3257 qoffset = 0;
3258 qcount = dev->real_num_tx_queues;
3259 }
3260 }
3261
3262 if (skb_rx_queue_recorded(skb)) {
3263 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3264 hash = skb_get_rx_queue(skb);
3265 if (hash >= qoffset)
3266 hash -= qoffset;
3267 while (unlikely(hash >= qcount))
3268 hash -= qcount;
3269 return hash + qoffset;
3270 }
3271
3272 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3273 }
3274
3275 void skb_warn_bad_offload(const struct sk_buff *skb)
3276 {
3277 static const netdev_features_t null_features;
3278 struct net_device *dev = skb->dev;
3279 const char *name = "";
3280
3281 if (!net_ratelimit())
3282 return;
3283
3284 if (dev) {
3285 if (dev->dev.parent)
3286 name = dev_driver_string(dev->dev.parent);
3287 else
3288 name = netdev_name(dev);
3289 }
3290 skb_dump(KERN_WARNING, skb, false);
3291 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3292 name, dev ? &dev->features : &null_features,
3293 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3294 }
3295
3296 /*
3297 * Invalidate hardware checksum when packet is to be mangled, and
3298 * complete checksum manually on outgoing path.
3299 */
3300 int skb_checksum_help(struct sk_buff *skb)
3301 {
3302 __wsum csum;
3303 int ret = 0, offset;
3304
3305 if (skb->ip_summed == CHECKSUM_COMPLETE)
3306 goto out_set_summed;
3307
3308 if (unlikely(skb_is_gso(skb))) {
3309 skb_warn_bad_offload(skb);
3310 return -EINVAL;
3311 }
3312
3313 /* Before computing a checksum, we should make sure no frag could
3314 * be modified by an external entity : checksum could be wrong.
3315 */
3316 if (skb_has_shared_frag(skb)) {
3317 ret = __skb_linearize(skb);
3318 if (ret)
3319 goto out;
3320 }
3321
3322 offset = skb_checksum_start_offset(skb);
3323 ret = -EINVAL;
3324 if (unlikely(offset >= skb_headlen(skb))) {
3325 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3326 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3327 offset, skb_headlen(skb));
3328 goto out;
3329 }
3330 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3331
3332 offset += skb->csum_offset;
3333 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3334 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3335 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3336 offset + sizeof(__sum16), skb_headlen(skb));
3337 goto out;
3338 }
3339 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3340 if (ret)
3341 goto out;
3342
3343 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3344 out_set_summed:
3345 skb->ip_summed = CHECKSUM_NONE;
3346 out:
3347 return ret;
3348 }
3349 EXPORT_SYMBOL(skb_checksum_help);
3350
3351 int skb_crc32c_csum_help(struct sk_buff *skb)
3352 {
3353 __le32 crc32c_csum;
3354 int ret = 0, offset, start;
3355
3356 if (skb->ip_summed != CHECKSUM_PARTIAL)
3357 goto out;
3358
3359 if (unlikely(skb_is_gso(skb)))
3360 goto out;
3361
3362 /* Before computing a checksum, we should make sure no frag could
3363 * be modified by an external entity : checksum could be wrong.
3364 */
3365 if (unlikely(skb_has_shared_frag(skb))) {
3366 ret = __skb_linearize(skb);
3367 if (ret)
3368 goto out;
3369 }
3370 start = skb_checksum_start_offset(skb);
3371 offset = start + offsetof(struct sctphdr, checksum);
3372 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3373 ret = -EINVAL;
3374 goto out;
3375 }
3376
3377 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3378 if (ret)
3379 goto out;
3380
3381 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3382 skb->len - start, ~(__u32)0,
3383 crc32c_csum_stub));
3384 *(__le32 *)(skb->data + offset) = crc32c_csum;
3385 skb_reset_csum_not_inet(skb);
3386 out:
3387 return ret;
3388 }
3389
3390 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3391 {
3392 __be16 type = skb->protocol;
3393
3394 /* Tunnel gso handlers can set protocol to ethernet. */
3395 if (type == htons(ETH_P_TEB)) {
3396 struct ethhdr *eth;
3397
3398 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3399 return 0;
3400
3401 eth = (struct ethhdr *)skb->data;
3402 type = eth->h_proto;
3403 }
3404
3405 return vlan_get_protocol_and_depth(skb, type, depth);
3406 }
3407
3408
3409 /* Take action when hardware reception checksum errors are detected. */
3410 #ifdef CONFIG_BUG
3411 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3412 {
3413 netdev_err(dev, "hw csum failure\n");
3414 skb_dump(KERN_ERR, skb, true);
3415 dump_stack();
3416 }
3417
3418 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3419 {
3420 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3421 }
3422 EXPORT_SYMBOL(netdev_rx_csum_fault);
3423 #endif
3424
3425 /* XXX: check that highmem exists at all on the given machine. */
3426 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3427 {
3428 #ifdef CONFIG_HIGHMEM
3429 int i;
3430
3431 if (!(dev->features & NETIF_F_HIGHDMA)) {
3432 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3433 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3434
3435 if (PageHighMem(skb_frag_page(frag)))
3436 return 1;
3437 }
3438 }
3439 #endif
3440 return 0;
3441 }
3442
3443 /* If MPLS offload request, verify we are testing hardware MPLS features
3444 * instead of standard features for the netdev.
3445 */
3446 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3447 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3448 netdev_features_t features,
3449 __be16 type)
3450 {
3451 if (eth_p_mpls(type))
3452 features &= skb->dev->mpls_features;
3453
3454 return features;
3455 }
3456 #else
3457 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3458 netdev_features_t features,
3459 __be16 type)
3460 {
3461 return features;
3462 }
3463 #endif
3464
3465 static netdev_features_t harmonize_features(struct sk_buff *skb,
3466 netdev_features_t features)
3467 {
3468 __be16 type;
3469
3470 type = skb_network_protocol(skb, NULL);
3471 features = net_mpls_features(skb, features, type);
3472
3473 if (skb->ip_summed != CHECKSUM_NONE &&
3474 !can_checksum_protocol(features, type)) {
3475 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3476 }
3477 if (illegal_highdma(skb->dev, skb))
3478 features &= ~NETIF_F_SG;
3479
3480 return features;
3481 }
3482
3483 netdev_features_t passthru_features_check(struct sk_buff *skb,
3484 struct net_device *dev,
3485 netdev_features_t features)
3486 {
3487 return features;
3488 }
3489 EXPORT_SYMBOL(passthru_features_check);
3490
3491 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3492 struct net_device *dev,
3493 netdev_features_t features)
3494 {
3495 return vlan_features_check(skb, features);
3496 }
3497
3498 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3499 struct net_device *dev,
3500 netdev_features_t features)
3501 {
3502 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3503
3504 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3505 return features & ~NETIF_F_GSO_MASK;
3506
3507 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb)))
3508 return features & ~NETIF_F_GSO_MASK;
3509
3510 if (!skb_shinfo(skb)->gso_type) {
3511 skb_warn_bad_offload(skb);
3512 return features & ~NETIF_F_GSO_MASK;
3513 }
3514
3515 /* Support for GSO partial features requires software
3516 * intervention before we can actually process the packets
3517 * so we need to strip support for any partial features now
3518 * and we can pull them back in after we have partially
3519 * segmented the frame.
3520 */
3521 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3522 features &= ~dev->gso_partial_features;
3523
3524 /* Make sure to clear the IPv4 ID mangling feature if the
3525 * IPv4 header has the potential to be fragmented.
3526 */
3527 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3528 struct iphdr *iph = skb->encapsulation ?
3529 inner_ip_hdr(skb) : ip_hdr(skb);
3530
3531 if (!(iph->frag_off & htons(IP_DF)))
3532 features &= ~NETIF_F_TSO_MANGLEID;
3533 }
3534
3535 return features;
3536 }
3537
3538 netdev_features_t netif_skb_features(struct sk_buff *skb)
3539 {
3540 struct net_device *dev = skb->dev;
3541 netdev_features_t features = dev->features;
3542
3543 if (skb_is_gso(skb))
3544 features = gso_features_check(skb, dev, features);
3545
3546 /* If encapsulation offload request, verify we are testing
3547 * hardware encapsulation features instead of standard
3548 * features for the netdev
3549 */
3550 if (skb->encapsulation)
3551 features &= dev->hw_enc_features;
3552
3553 if (skb_vlan_tagged(skb))
3554 features = netdev_intersect_features(features,
3555 dev->vlan_features |
3556 NETIF_F_HW_VLAN_CTAG_TX |
3557 NETIF_F_HW_VLAN_STAG_TX);
3558
3559 if (dev->netdev_ops->ndo_features_check)
3560 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3561 features);
3562 else
3563 features &= dflt_features_check(skb, dev, features);
3564
3565 return harmonize_features(skb, features);
3566 }
3567 EXPORT_SYMBOL(netif_skb_features);
3568
3569 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3570 struct netdev_queue *txq, bool more)
3571 {
3572 unsigned int len;
3573 int rc;
3574
3575 if (dev_nit_active(dev))
3576 dev_queue_xmit_nit(skb, dev);
3577
3578 len = skb->len;
3579 trace_net_dev_start_xmit(skb, dev);
3580 rc = netdev_start_xmit(skb, dev, txq, more);
3581 trace_net_dev_xmit(skb, rc, dev, len);
3582
3583 return rc;
3584 }
3585
3586 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3587 struct netdev_queue *txq, int *ret)
3588 {
3589 struct sk_buff *skb = first;
3590 int rc = NETDEV_TX_OK;
3591
3592 while (skb) {
3593 struct sk_buff *next = skb->next;
3594
3595 skb_mark_not_on_list(skb);
3596 rc = xmit_one(skb, dev, txq, next != NULL);
3597 if (unlikely(!dev_xmit_complete(rc))) {
3598 skb->next = next;
3599 goto out;
3600 }
3601
3602 skb = next;
3603 if (netif_tx_queue_stopped(txq) && skb) {
3604 rc = NETDEV_TX_BUSY;
3605 break;
3606 }
3607 }
3608
3609 out:
3610 *ret = rc;
3611 return skb;
3612 }
3613
3614 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3615 netdev_features_t features)
3616 {
3617 if (skb_vlan_tag_present(skb) &&
3618 !vlan_hw_offload_capable(features, skb->vlan_proto))
3619 skb = __vlan_hwaccel_push_inside(skb);
3620 return skb;
3621 }
3622
3623 int skb_csum_hwoffload_help(struct sk_buff *skb,
3624 const netdev_features_t features)
3625 {
3626 if (unlikely(skb_csum_is_sctp(skb)))
3627 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3628 skb_crc32c_csum_help(skb);
3629
3630 if (features & NETIF_F_HW_CSUM)
3631 return 0;
3632
3633 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3634 switch (skb->csum_offset) {
3635 case offsetof(struct tcphdr, check):
3636 case offsetof(struct udphdr, check):
3637 return 0;
3638 }
3639 }
3640
3641 return skb_checksum_help(skb);
3642 }
3643 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3644
3645 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3646 {
3647 netdev_features_t features;
3648
3649 features = netif_skb_features(skb);
3650 skb = validate_xmit_vlan(skb, features);
3651 if (unlikely(!skb))
3652 goto out_null;
3653
3654 skb = sk_validate_xmit_skb(skb, dev);
3655 if (unlikely(!skb))
3656 goto out_null;
3657
3658 if (netif_needs_gso(skb, features)) {
3659 struct sk_buff *segs;
3660
3661 segs = skb_gso_segment(skb, features);
3662 if (IS_ERR(segs)) {
3663 goto out_kfree_skb;
3664 } else if (segs) {
3665 consume_skb(skb);
3666 skb = segs;
3667 }
3668 } else {
3669 if (skb_needs_linearize(skb, features) &&
3670 __skb_linearize(skb))
3671 goto out_kfree_skb;
3672
3673 /* If packet is not checksummed and device does not
3674 * support checksumming for this protocol, complete
3675 * checksumming here.
3676 */
3677 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3678 if (skb->encapsulation)
3679 skb_set_inner_transport_header(skb,
3680 skb_checksum_start_offset(skb));
3681 else
3682 skb_set_transport_header(skb,
3683 skb_checksum_start_offset(skb));
3684 if (skb_csum_hwoffload_help(skb, features))
3685 goto out_kfree_skb;
3686 }
3687 }
3688
3689 skb = validate_xmit_xfrm(skb, features, again);
3690
3691 return skb;
3692
3693 out_kfree_skb:
3694 kfree_skb(skb);
3695 out_null:
3696 dev_core_stats_tx_dropped_inc(dev);
3697 return NULL;
3698 }
3699
3700 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3701 {
3702 struct sk_buff *next, *head = NULL, *tail;
3703
3704 for (; skb != NULL; skb = next) {
3705 next = skb->next;
3706 skb_mark_not_on_list(skb);
3707
3708 /* in case skb wont be segmented, point to itself */
3709 skb->prev = skb;
3710
3711 skb = validate_xmit_skb(skb, dev, again);
3712 if (!skb)
3713 continue;
3714
3715 if (!head)
3716 head = skb;
3717 else
3718 tail->next = skb;
3719 /* If skb was segmented, skb->prev points to
3720 * the last segment. If not, it still contains skb.
3721 */
3722 tail = skb->prev;
3723 }
3724 return head;
3725 }
3726 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3727
3728 static void qdisc_pkt_len_init(struct sk_buff *skb)
3729 {
3730 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3731
3732 qdisc_skb_cb(skb)->pkt_len = skb->len;
3733
3734 /* To get more precise estimation of bytes sent on wire,
3735 * we add to pkt_len the headers size of all segments
3736 */
3737 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3738 u16 gso_segs = shinfo->gso_segs;
3739 unsigned int hdr_len;
3740
3741 /* mac layer + network layer */
3742 hdr_len = skb_transport_offset(skb);
3743
3744 /* + transport layer */
3745 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3746 const struct tcphdr *th;
3747 struct tcphdr _tcphdr;
3748
3749 th = skb_header_pointer(skb, hdr_len,
3750 sizeof(_tcphdr), &_tcphdr);
3751 if (likely(th))
3752 hdr_len += __tcp_hdrlen(th);
3753 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
3754 struct udphdr _udphdr;
3755
3756 if (skb_header_pointer(skb, hdr_len,
3757 sizeof(_udphdr), &_udphdr))
3758 hdr_len += sizeof(struct udphdr);
3759 }
3760
3761 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) {
3762 int payload = skb->len - hdr_len;
3763
3764 /* Malicious packet. */
3765 if (payload <= 0)
3766 return;
3767 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size);
3768 }
3769 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3770 }
3771 }
3772
3773 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3774 struct sk_buff **to_free,
3775 struct netdev_queue *txq)
3776 {
3777 int rc;
3778
3779 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3780 if (rc == NET_XMIT_SUCCESS)
3781 trace_qdisc_enqueue(q, txq, skb);
3782 return rc;
3783 }
3784
3785 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3786 struct net_device *dev,
3787 struct netdev_queue *txq)
3788 {
3789 spinlock_t *root_lock = qdisc_lock(q);
3790 struct sk_buff *to_free = NULL;
3791 bool contended;
3792 int rc;
3793
3794 qdisc_calculate_pkt_len(skb, q);
3795
3796 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP);
3797
3798 if (q->flags & TCQ_F_NOLOCK) {
3799 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3800 qdisc_run_begin(q)) {
3801 /* Retest nolock_qdisc_is_empty() within the protection
3802 * of q->seqlock to protect from racing with requeuing.
3803 */
3804 if (unlikely(!nolock_qdisc_is_empty(q))) {
3805 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3806 __qdisc_run(q);
3807 qdisc_run_end(q);
3808
3809 goto no_lock_out;
3810 }
3811
3812 qdisc_bstats_cpu_update(q, skb);
3813 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3814 !nolock_qdisc_is_empty(q))
3815 __qdisc_run(q);
3816
3817 qdisc_run_end(q);
3818 return NET_XMIT_SUCCESS;
3819 }
3820
3821 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3822 qdisc_run(q);
3823
3824 no_lock_out:
3825 if (unlikely(to_free))
3826 kfree_skb_list_reason(to_free,
3827 tcf_get_drop_reason(to_free));
3828 return rc;
3829 }
3830
3831 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) {
3832 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
3833 return NET_XMIT_DROP;
3834 }
3835 /*
3836 * Heuristic to force contended enqueues to serialize on a
3837 * separate lock before trying to get qdisc main lock.
3838 * This permits qdisc->running owner to get the lock more
3839 * often and dequeue packets faster.
3840 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3841 * and then other tasks will only enqueue packets. The packets will be
3842 * sent after the qdisc owner is scheduled again. To prevent this
3843 * scenario the task always serialize on the lock.
3844 */
3845 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3846 if (unlikely(contended))
3847 spin_lock(&q->busylock);
3848
3849 spin_lock(root_lock);
3850 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3851 __qdisc_drop(skb, &to_free);
3852 rc = NET_XMIT_DROP;
3853 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3854 qdisc_run_begin(q)) {
3855 /*
3856 * This is a work-conserving queue; there are no old skbs
3857 * waiting to be sent out; and the qdisc is not running -
3858 * xmit the skb directly.
3859 */
3860
3861 qdisc_bstats_update(q, skb);
3862
3863 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3864 if (unlikely(contended)) {
3865 spin_unlock(&q->busylock);
3866 contended = false;
3867 }
3868 __qdisc_run(q);
3869 }
3870
3871 qdisc_run_end(q);
3872 rc = NET_XMIT_SUCCESS;
3873 } else {
3874 WRITE_ONCE(q->owner, smp_processor_id());
3875 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3876 WRITE_ONCE(q->owner, -1);
3877 if (qdisc_run_begin(q)) {
3878 if (unlikely(contended)) {
3879 spin_unlock(&q->busylock);
3880 contended = false;
3881 }
3882 __qdisc_run(q);
3883 qdisc_run_end(q);
3884 }
3885 }
3886 spin_unlock(root_lock);
3887 if (unlikely(to_free))
3888 kfree_skb_list_reason(to_free,
3889 tcf_get_drop_reason(to_free));
3890 if (unlikely(contended))
3891 spin_unlock(&q->busylock);
3892 return rc;
3893 }
3894
3895 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3896 static void skb_update_prio(struct sk_buff *skb)
3897 {
3898 const struct netprio_map *map;
3899 const struct sock *sk;
3900 unsigned int prioidx;
3901
3902 if (skb->priority)
3903 return;
3904 map = rcu_dereference_bh(skb->dev->priomap);
3905 if (!map)
3906 return;
3907 sk = skb_to_full_sk(skb);
3908 if (!sk)
3909 return;
3910
3911 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3912
3913 if (prioidx < map->priomap_len)
3914 skb->priority = map->priomap[prioidx];
3915 }
3916 #else
3917 #define skb_update_prio(skb)
3918 #endif
3919
3920 /**
3921 * dev_loopback_xmit - loop back @skb
3922 * @net: network namespace this loopback is happening in
3923 * @sk: sk needed to be a netfilter okfn
3924 * @skb: buffer to transmit
3925 */
3926 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3927 {
3928 skb_reset_mac_header(skb);
3929 __skb_pull(skb, skb_network_offset(skb));
3930 skb->pkt_type = PACKET_LOOPBACK;
3931 if (skb->ip_summed == CHECKSUM_NONE)
3932 skb->ip_summed = CHECKSUM_UNNECESSARY;
3933 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3934 skb_dst_force(skb);
3935 netif_rx(skb);
3936 return 0;
3937 }
3938 EXPORT_SYMBOL(dev_loopback_xmit);
3939
3940 #ifdef CONFIG_NET_EGRESS
3941 static struct netdev_queue *
3942 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3943 {
3944 int qm = skb_get_queue_mapping(skb);
3945
3946 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3947 }
3948
3949 #ifndef CONFIG_PREEMPT_RT
3950 static bool netdev_xmit_txqueue_skipped(void)
3951 {
3952 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3953 }
3954
3955 void netdev_xmit_skip_txqueue(bool skip)
3956 {
3957 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3958 }
3959 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3960
3961 #else
3962 static bool netdev_xmit_txqueue_skipped(void)
3963 {
3964 return current->net_xmit.skip_txqueue;
3965 }
3966
3967 void netdev_xmit_skip_txqueue(bool skip)
3968 {
3969 current->net_xmit.skip_txqueue = skip;
3970 }
3971 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3972 #endif
3973 #endif /* CONFIG_NET_EGRESS */
3974
3975 #ifdef CONFIG_NET_XGRESS
3976 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb,
3977 enum skb_drop_reason *drop_reason)
3978 {
3979 int ret = TC_ACT_UNSPEC;
3980 #ifdef CONFIG_NET_CLS_ACT
3981 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
3982 struct tcf_result res;
3983
3984 if (!miniq)
3985 return ret;
3986
3987 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) {
3988 if (tcf_block_bypass_sw(miniq->block))
3989 return ret;
3990 }
3991
3992 tc_skb_cb(skb)->mru = 0;
3993 tc_skb_cb(skb)->post_ct = false;
3994 tcf_set_drop_reason(skb, *drop_reason);
3995
3996 mini_qdisc_bstats_cpu_update(miniq, skb);
3997 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
3998 /* Only tcf related quirks below. */
3999 switch (ret) {
4000 case TC_ACT_SHOT:
4001 *drop_reason = tcf_get_drop_reason(skb);
4002 mini_qdisc_qstats_cpu_drop(miniq);
4003 break;
4004 case TC_ACT_OK:
4005 case TC_ACT_RECLASSIFY:
4006 skb->tc_index = TC_H_MIN(res.classid);
4007 break;
4008 }
4009 #endif /* CONFIG_NET_CLS_ACT */
4010 return ret;
4011 }
4012
4013 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
4014
4015 void tcx_inc(void)
4016 {
4017 static_branch_inc(&tcx_needed_key);
4018 }
4019
4020 void tcx_dec(void)
4021 {
4022 static_branch_dec(&tcx_needed_key);
4023 }
4024
4025 static __always_inline enum tcx_action_base
4026 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
4027 const bool needs_mac)
4028 {
4029 const struct bpf_mprog_fp *fp;
4030 const struct bpf_prog *prog;
4031 int ret = TCX_NEXT;
4032
4033 if (needs_mac)
4034 __skb_push(skb, skb->mac_len);
4035 bpf_mprog_foreach_prog(entry, fp, prog) {
4036 bpf_compute_data_pointers(skb);
4037 ret = bpf_prog_run(prog, skb);
4038 if (ret != TCX_NEXT)
4039 break;
4040 }
4041 if (needs_mac)
4042 __skb_pull(skb, skb->mac_len);
4043 return tcx_action_code(skb, ret);
4044 }
4045
4046 static __always_inline struct sk_buff *
4047 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4048 struct net_device *orig_dev, bool *another)
4049 {
4050 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
4051 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
4052 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4053 int sch_ret;
4054
4055 if (!entry)
4056 return skb;
4057
4058 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4059 if (*pt_prev) {
4060 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4061 *pt_prev = NULL;
4062 }
4063
4064 qdisc_skb_cb(skb)->pkt_len = skb->len;
4065 tcx_set_ingress(skb, true);
4066
4067 if (static_branch_unlikely(&tcx_needed_key)) {
4068 sch_ret = tcx_run(entry, skb, true);
4069 if (sch_ret != TC_ACT_UNSPEC)
4070 goto ingress_verdict;
4071 }
4072 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4073 ingress_verdict:
4074 switch (sch_ret) {
4075 case TC_ACT_REDIRECT:
4076 /* skb_mac_header check was done by BPF, so we can safely
4077 * push the L2 header back before redirecting to another
4078 * netdev.
4079 */
4080 __skb_push(skb, skb->mac_len);
4081 if (skb_do_redirect(skb) == -EAGAIN) {
4082 __skb_pull(skb, skb->mac_len);
4083 *another = true;
4084 break;
4085 }
4086 *ret = NET_RX_SUCCESS;
4087 bpf_net_ctx_clear(bpf_net_ctx);
4088 return NULL;
4089 case TC_ACT_SHOT:
4090 kfree_skb_reason(skb, drop_reason);
4091 *ret = NET_RX_DROP;
4092 bpf_net_ctx_clear(bpf_net_ctx);
4093 return NULL;
4094 /* used by tc_run */
4095 case TC_ACT_STOLEN:
4096 case TC_ACT_QUEUED:
4097 case TC_ACT_TRAP:
4098 consume_skb(skb);
4099 fallthrough;
4100 case TC_ACT_CONSUMED:
4101 *ret = NET_RX_SUCCESS;
4102 bpf_net_ctx_clear(bpf_net_ctx);
4103 return NULL;
4104 }
4105 bpf_net_ctx_clear(bpf_net_ctx);
4106
4107 return skb;
4108 }
4109
4110 static __always_inline struct sk_buff *
4111 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4112 {
4113 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4114 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
4115 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4116 int sch_ret;
4117
4118 if (!entry)
4119 return skb;
4120
4121 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4122
4123 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4124 * already set by the caller.
4125 */
4126 if (static_branch_unlikely(&tcx_needed_key)) {
4127 sch_ret = tcx_run(entry, skb, false);
4128 if (sch_ret != TC_ACT_UNSPEC)
4129 goto egress_verdict;
4130 }
4131 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4132 egress_verdict:
4133 switch (sch_ret) {
4134 case TC_ACT_REDIRECT:
4135 /* No need to push/pop skb's mac_header here on egress! */
4136 skb_do_redirect(skb);
4137 *ret = NET_XMIT_SUCCESS;
4138 bpf_net_ctx_clear(bpf_net_ctx);
4139 return NULL;
4140 case TC_ACT_SHOT:
4141 kfree_skb_reason(skb, drop_reason);
4142 *ret = NET_XMIT_DROP;
4143 bpf_net_ctx_clear(bpf_net_ctx);
4144 return NULL;
4145 /* used by tc_run */
4146 case TC_ACT_STOLEN:
4147 case TC_ACT_QUEUED:
4148 case TC_ACT_TRAP:
4149 consume_skb(skb);
4150 fallthrough;
4151 case TC_ACT_CONSUMED:
4152 *ret = NET_XMIT_SUCCESS;
4153 bpf_net_ctx_clear(bpf_net_ctx);
4154 return NULL;
4155 }
4156 bpf_net_ctx_clear(bpf_net_ctx);
4157
4158 return skb;
4159 }
4160 #else
4161 static __always_inline struct sk_buff *
4162 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4163 struct net_device *orig_dev, bool *another)
4164 {
4165 return skb;
4166 }
4167
4168 static __always_inline struct sk_buff *
4169 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4170 {
4171 return skb;
4172 }
4173 #endif /* CONFIG_NET_XGRESS */
4174
4175 #ifdef CONFIG_XPS
4176 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4177 struct xps_dev_maps *dev_maps, unsigned int tci)
4178 {
4179 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4180 struct xps_map *map;
4181 int queue_index = -1;
4182
4183 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4184 return queue_index;
4185
4186 tci *= dev_maps->num_tc;
4187 tci += tc;
4188
4189 map = rcu_dereference(dev_maps->attr_map[tci]);
4190 if (map) {
4191 if (map->len == 1)
4192 queue_index = map->queues[0];
4193 else
4194 queue_index = map->queues[reciprocal_scale(
4195 skb_get_hash(skb), map->len)];
4196 if (unlikely(queue_index >= dev->real_num_tx_queues))
4197 queue_index = -1;
4198 }
4199 return queue_index;
4200 }
4201 #endif
4202
4203 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4204 struct sk_buff *skb)
4205 {
4206 #ifdef CONFIG_XPS
4207 struct xps_dev_maps *dev_maps;
4208 struct sock *sk = skb->sk;
4209 int queue_index = -1;
4210
4211 if (!static_key_false(&xps_needed))
4212 return -1;
4213
4214 rcu_read_lock();
4215 if (!static_key_false(&xps_rxqs_needed))
4216 goto get_cpus_map;
4217
4218 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4219 if (dev_maps) {
4220 int tci = sk_rx_queue_get(sk);
4221
4222 if (tci >= 0)
4223 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4224 tci);
4225 }
4226
4227 get_cpus_map:
4228 if (queue_index < 0) {
4229 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4230 if (dev_maps) {
4231 unsigned int tci = skb->sender_cpu - 1;
4232
4233 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4234 tci);
4235 }
4236 }
4237 rcu_read_unlock();
4238
4239 return queue_index;
4240 #else
4241 return -1;
4242 #endif
4243 }
4244
4245 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4246 struct net_device *sb_dev)
4247 {
4248 return 0;
4249 }
4250 EXPORT_SYMBOL(dev_pick_tx_zero);
4251
4252 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4253 struct net_device *sb_dev)
4254 {
4255 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4256 }
4257 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4258
4259 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4260 struct net_device *sb_dev)
4261 {
4262 struct sock *sk = skb->sk;
4263 int queue_index = sk_tx_queue_get(sk);
4264
4265 sb_dev = sb_dev ? : dev;
4266
4267 if (queue_index < 0 || skb->ooo_okay ||
4268 queue_index >= dev->real_num_tx_queues) {
4269 int new_index = get_xps_queue(dev, sb_dev, skb);
4270
4271 if (new_index < 0)
4272 new_index = skb_tx_hash(dev, sb_dev, skb);
4273
4274 if (queue_index != new_index && sk &&
4275 sk_fullsock(sk) &&
4276 rcu_access_pointer(sk->sk_dst_cache))
4277 sk_tx_queue_set(sk, new_index);
4278
4279 queue_index = new_index;
4280 }
4281
4282 return queue_index;
4283 }
4284 EXPORT_SYMBOL(netdev_pick_tx);
4285
4286 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4287 struct sk_buff *skb,
4288 struct net_device *sb_dev)
4289 {
4290 int queue_index = 0;
4291
4292 #ifdef CONFIG_XPS
4293 u32 sender_cpu = skb->sender_cpu - 1;
4294
4295 if (sender_cpu >= (u32)NR_CPUS)
4296 skb->sender_cpu = raw_smp_processor_id() + 1;
4297 #endif
4298
4299 if (dev->real_num_tx_queues != 1) {
4300 const struct net_device_ops *ops = dev->netdev_ops;
4301
4302 if (ops->ndo_select_queue)
4303 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4304 else
4305 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4306
4307 queue_index = netdev_cap_txqueue(dev, queue_index);
4308 }
4309
4310 skb_set_queue_mapping(skb, queue_index);
4311 return netdev_get_tx_queue(dev, queue_index);
4312 }
4313
4314 /**
4315 * __dev_queue_xmit() - transmit a buffer
4316 * @skb: buffer to transmit
4317 * @sb_dev: suboordinate device used for L2 forwarding offload
4318 *
4319 * Queue a buffer for transmission to a network device. The caller must
4320 * have set the device and priority and built the buffer before calling
4321 * this function. The function can be called from an interrupt.
4322 *
4323 * When calling this method, interrupts MUST be enabled. This is because
4324 * the BH enable code must have IRQs enabled so that it will not deadlock.
4325 *
4326 * Regardless of the return value, the skb is consumed, so it is currently
4327 * difficult to retry a send to this method. (You can bump the ref count
4328 * before sending to hold a reference for retry if you are careful.)
4329 *
4330 * Return:
4331 * * 0 - buffer successfully transmitted
4332 * * positive qdisc return code - NET_XMIT_DROP etc.
4333 * * negative errno - other errors
4334 */
4335 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4336 {
4337 struct net_device *dev = skb->dev;
4338 struct netdev_queue *txq = NULL;
4339 struct Qdisc *q;
4340 int rc = -ENOMEM;
4341 bool again = false;
4342
4343 skb_reset_mac_header(skb);
4344 skb_assert_len(skb);
4345
4346 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4347 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4348
4349 /* Disable soft irqs for various locks below. Also
4350 * stops preemption for RCU.
4351 */
4352 rcu_read_lock_bh();
4353
4354 skb_update_prio(skb);
4355
4356 qdisc_pkt_len_init(skb);
4357 tcx_set_ingress(skb, false);
4358 #ifdef CONFIG_NET_EGRESS
4359 if (static_branch_unlikely(&egress_needed_key)) {
4360 if (nf_hook_egress_active()) {
4361 skb = nf_hook_egress(skb, &rc, dev);
4362 if (!skb)
4363 goto out;
4364 }
4365
4366 netdev_xmit_skip_txqueue(false);
4367
4368 nf_skip_egress(skb, true);
4369 skb = sch_handle_egress(skb, &rc, dev);
4370 if (!skb)
4371 goto out;
4372 nf_skip_egress(skb, false);
4373
4374 if (netdev_xmit_txqueue_skipped())
4375 txq = netdev_tx_queue_mapping(dev, skb);
4376 }
4377 #endif
4378 /* If device/qdisc don't need skb->dst, release it right now while
4379 * its hot in this cpu cache.
4380 */
4381 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4382 skb_dst_drop(skb);
4383 else
4384 skb_dst_force(skb);
4385
4386 if (!txq)
4387 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4388
4389 q = rcu_dereference_bh(txq->qdisc);
4390
4391 trace_net_dev_queue(skb);
4392 if (q->enqueue) {
4393 rc = __dev_xmit_skb(skb, q, dev, txq);
4394 goto out;
4395 }
4396
4397 /* The device has no queue. Common case for software devices:
4398 * loopback, all the sorts of tunnels...
4399
4400 * Really, it is unlikely that netif_tx_lock protection is necessary
4401 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4402 * counters.)
4403 * However, it is possible, that they rely on protection
4404 * made by us here.
4405
4406 * Check this and shot the lock. It is not prone from deadlocks.
4407 *Either shot noqueue qdisc, it is even simpler 8)
4408 */
4409 if (dev->flags & IFF_UP) {
4410 int cpu = smp_processor_id(); /* ok because BHs are off */
4411
4412 /* Other cpus might concurrently change txq->xmit_lock_owner
4413 * to -1 or to their cpu id, but not to our id.
4414 */
4415 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4416 if (dev_xmit_recursion())
4417 goto recursion_alert;
4418
4419 skb = validate_xmit_skb(skb, dev, &again);
4420 if (!skb)
4421 goto out;
4422
4423 HARD_TX_LOCK(dev, txq, cpu);
4424
4425 if (!netif_xmit_stopped(txq)) {
4426 dev_xmit_recursion_inc();
4427 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4428 dev_xmit_recursion_dec();
4429 if (dev_xmit_complete(rc)) {
4430 HARD_TX_UNLOCK(dev, txq);
4431 goto out;
4432 }
4433 }
4434 HARD_TX_UNLOCK(dev, txq);
4435 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4436 dev->name);
4437 } else {
4438 /* Recursion is detected! It is possible,
4439 * unfortunately
4440 */
4441 recursion_alert:
4442 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4443 dev->name);
4444 }
4445 }
4446
4447 rc = -ENETDOWN;
4448 rcu_read_unlock_bh();
4449
4450 dev_core_stats_tx_dropped_inc(dev);
4451 kfree_skb_list(skb);
4452 return rc;
4453 out:
4454 rcu_read_unlock_bh();
4455 return rc;
4456 }
4457 EXPORT_SYMBOL(__dev_queue_xmit);
4458
4459 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4460 {
4461 struct net_device *dev = skb->dev;
4462 struct sk_buff *orig_skb = skb;
4463 struct netdev_queue *txq;
4464 int ret = NETDEV_TX_BUSY;
4465 bool again = false;
4466
4467 if (unlikely(!netif_running(dev) ||
4468 !netif_carrier_ok(dev)))
4469 goto drop;
4470
4471 skb = validate_xmit_skb_list(skb, dev, &again);
4472 if (skb != orig_skb)
4473 goto drop;
4474
4475 skb_set_queue_mapping(skb, queue_id);
4476 txq = skb_get_tx_queue(dev, skb);
4477
4478 local_bh_disable();
4479
4480 dev_xmit_recursion_inc();
4481 HARD_TX_LOCK(dev, txq, smp_processor_id());
4482 if (!netif_xmit_frozen_or_drv_stopped(txq))
4483 ret = netdev_start_xmit(skb, dev, txq, false);
4484 HARD_TX_UNLOCK(dev, txq);
4485 dev_xmit_recursion_dec();
4486
4487 local_bh_enable();
4488 return ret;
4489 drop:
4490 dev_core_stats_tx_dropped_inc(dev);
4491 kfree_skb_list(skb);
4492 return NET_XMIT_DROP;
4493 }
4494 EXPORT_SYMBOL(__dev_direct_xmit);
4495
4496 /*************************************************************************
4497 * Receiver routines
4498 *************************************************************************/
4499 static DEFINE_PER_CPU(struct task_struct *, backlog_napi);
4500
4501 int weight_p __read_mostly = 64; /* old backlog weight */
4502 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4503 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4504
4505 /* Called with irq disabled */
4506 static inline void ____napi_schedule(struct softnet_data *sd,
4507 struct napi_struct *napi)
4508 {
4509 struct task_struct *thread;
4510
4511 lockdep_assert_irqs_disabled();
4512
4513 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4514 /* Paired with smp_mb__before_atomic() in
4515 * napi_enable()/dev_set_threaded().
4516 * Use READ_ONCE() to guarantee a complete
4517 * read on napi->thread. Only call
4518 * wake_up_process() when it's not NULL.
4519 */
4520 thread = READ_ONCE(napi->thread);
4521 if (thread) {
4522 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi))
4523 goto use_local_napi;
4524
4525 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4526 wake_up_process(thread);
4527 return;
4528 }
4529 }
4530
4531 use_local_napi:
4532 list_add_tail(&napi->poll_list, &sd->poll_list);
4533 WRITE_ONCE(napi->list_owner, smp_processor_id());
4534 /* If not called from net_rx_action()
4535 * we have to raise NET_RX_SOFTIRQ.
4536 */
4537 if (!sd->in_net_rx_action)
4538 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4539 }
4540
4541 #ifdef CONFIG_RPS
4542
4543 struct static_key_false rps_needed __read_mostly;
4544 EXPORT_SYMBOL(rps_needed);
4545 struct static_key_false rfs_needed __read_mostly;
4546 EXPORT_SYMBOL(rfs_needed);
4547
4548 static struct rps_dev_flow *
4549 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4550 struct rps_dev_flow *rflow, u16 next_cpu)
4551 {
4552 if (next_cpu < nr_cpu_ids) {
4553 u32 head;
4554 #ifdef CONFIG_RFS_ACCEL
4555 struct netdev_rx_queue *rxqueue;
4556 struct rps_dev_flow_table *flow_table;
4557 struct rps_dev_flow *old_rflow;
4558 u16 rxq_index;
4559 u32 flow_id;
4560 int rc;
4561
4562 /* Should we steer this flow to a different hardware queue? */
4563 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4564 !(dev->features & NETIF_F_NTUPLE))
4565 goto out;
4566 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4567 if (rxq_index == skb_get_rx_queue(skb))
4568 goto out;
4569
4570 rxqueue = dev->_rx + rxq_index;
4571 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4572 if (!flow_table)
4573 goto out;
4574 flow_id = skb_get_hash(skb) & flow_table->mask;
4575 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4576 rxq_index, flow_id);
4577 if (rc < 0)
4578 goto out;
4579 old_rflow = rflow;
4580 rflow = &flow_table->flows[flow_id];
4581 WRITE_ONCE(rflow->filter, rc);
4582 if (old_rflow->filter == rc)
4583 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
4584 out:
4585 #endif
4586 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
4587 rps_input_queue_tail_save(&rflow->last_qtail, head);
4588 }
4589
4590 WRITE_ONCE(rflow->cpu, next_cpu);
4591 return rflow;
4592 }
4593
4594 /*
4595 * get_rps_cpu is called from netif_receive_skb and returns the target
4596 * CPU from the RPS map of the receiving queue for a given skb.
4597 * rcu_read_lock must be held on entry.
4598 */
4599 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4600 struct rps_dev_flow **rflowp)
4601 {
4602 const struct rps_sock_flow_table *sock_flow_table;
4603 struct netdev_rx_queue *rxqueue = dev->_rx;
4604 struct rps_dev_flow_table *flow_table;
4605 struct rps_map *map;
4606 int cpu = -1;
4607 u32 tcpu;
4608 u32 hash;
4609
4610 if (skb_rx_queue_recorded(skb)) {
4611 u16 index = skb_get_rx_queue(skb);
4612
4613 if (unlikely(index >= dev->real_num_rx_queues)) {
4614 WARN_ONCE(dev->real_num_rx_queues > 1,
4615 "%s received packet on queue %u, but number "
4616 "of RX queues is %u\n",
4617 dev->name, index, dev->real_num_rx_queues);
4618 goto done;
4619 }
4620 rxqueue += index;
4621 }
4622
4623 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4624
4625 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4626 map = rcu_dereference(rxqueue->rps_map);
4627 if (!flow_table && !map)
4628 goto done;
4629
4630 skb_reset_network_header(skb);
4631 hash = skb_get_hash(skb);
4632 if (!hash)
4633 goto done;
4634
4635 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
4636 if (flow_table && sock_flow_table) {
4637 struct rps_dev_flow *rflow;
4638 u32 next_cpu;
4639 u32 ident;
4640
4641 /* First check into global flow table if there is a match.
4642 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4643 */
4644 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4645 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
4646 goto try_rps;
4647
4648 next_cpu = ident & net_hotdata.rps_cpu_mask;
4649
4650 /* OK, now we know there is a match,
4651 * we can look at the local (per receive queue) flow table
4652 */
4653 rflow = &flow_table->flows[hash & flow_table->mask];
4654 tcpu = rflow->cpu;
4655
4656 /*
4657 * If the desired CPU (where last recvmsg was done) is
4658 * different from current CPU (one in the rx-queue flow
4659 * table entry), switch if one of the following holds:
4660 * - Current CPU is unset (>= nr_cpu_ids).
4661 * - Current CPU is offline.
4662 * - The current CPU's queue tail has advanced beyond the
4663 * last packet that was enqueued using this table entry.
4664 * This guarantees that all previous packets for the flow
4665 * have been dequeued, thus preserving in order delivery.
4666 */
4667 if (unlikely(tcpu != next_cpu) &&
4668 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4669 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
4670 rflow->last_qtail)) >= 0)) {
4671 tcpu = next_cpu;
4672 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4673 }
4674
4675 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4676 *rflowp = rflow;
4677 cpu = tcpu;
4678 goto done;
4679 }
4680 }
4681
4682 try_rps:
4683
4684 if (map) {
4685 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4686 if (cpu_online(tcpu)) {
4687 cpu = tcpu;
4688 goto done;
4689 }
4690 }
4691
4692 done:
4693 return cpu;
4694 }
4695
4696 #ifdef CONFIG_RFS_ACCEL
4697
4698 /**
4699 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4700 * @dev: Device on which the filter was set
4701 * @rxq_index: RX queue index
4702 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4703 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4704 *
4705 * Drivers that implement ndo_rx_flow_steer() should periodically call
4706 * this function for each installed filter and remove the filters for
4707 * which it returns %true.
4708 */
4709 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4710 u32 flow_id, u16 filter_id)
4711 {
4712 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4713 struct rps_dev_flow_table *flow_table;
4714 struct rps_dev_flow *rflow;
4715 bool expire = true;
4716 unsigned int cpu;
4717
4718 rcu_read_lock();
4719 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4720 if (flow_table && flow_id <= flow_table->mask) {
4721 rflow = &flow_table->flows[flow_id];
4722 cpu = READ_ONCE(rflow->cpu);
4723 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids &&
4724 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) -
4725 READ_ONCE(rflow->last_qtail)) <
4726 (int)(10 * flow_table->mask)))
4727 expire = false;
4728 }
4729 rcu_read_unlock();
4730 return expire;
4731 }
4732 EXPORT_SYMBOL(rps_may_expire_flow);
4733
4734 #endif /* CONFIG_RFS_ACCEL */
4735
4736 /* Called from hardirq (IPI) context */
4737 static void rps_trigger_softirq(void *data)
4738 {
4739 struct softnet_data *sd = data;
4740
4741 ____napi_schedule(sd, &sd->backlog);
4742 sd->received_rps++;
4743 }
4744
4745 #endif /* CONFIG_RPS */
4746
4747 /* Called from hardirq (IPI) context */
4748 static void trigger_rx_softirq(void *data)
4749 {
4750 struct softnet_data *sd = data;
4751
4752 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4753 smp_store_release(&sd->defer_ipi_scheduled, 0);
4754 }
4755
4756 /*
4757 * After we queued a packet into sd->input_pkt_queue,
4758 * we need to make sure this queue is serviced soon.
4759 *
4760 * - If this is another cpu queue, link it to our rps_ipi_list,
4761 * and make sure we will process rps_ipi_list from net_rx_action().
4762 *
4763 * - If this is our own queue, NAPI schedule our backlog.
4764 * Note that this also raises NET_RX_SOFTIRQ.
4765 */
4766 static void napi_schedule_rps(struct softnet_data *sd)
4767 {
4768 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4769
4770 #ifdef CONFIG_RPS
4771 if (sd != mysd) {
4772 if (use_backlog_threads()) {
4773 __napi_schedule_irqoff(&sd->backlog);
4774 return;
4775 }
4776
4777 sd->rps_ipi_next = mysd->rps_ipi_list;
4778 mysd->rps_ipi_list = sd;
4779
4780 /* If not called from net_rx_action() or napi_threaded_poll()
4781 * we have to raise NET_RX_SOFTIRQ.
4782 */
4783 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4784 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4785 return;
4786 }
4787 #endif /* CONFIG_RPS */
4788 __napi_schedule_irqoff(&mysd->backlog);
4789 }
4790
4791 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu)
4792 {
4793 unsigned long flags;
4794
4795 if (use_backlog_threads()) {
4796 backlog_lock_irq_save(sd, &flags);
4797
4798 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4799 __napi_schedule_irqoff(&sd->backlog);
4800
4801 backlog_unlock_irq_restore(sd, &flags);
4802
4803 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) {
4804 smp_call_function_single_async(cpu, &sd->defer_csd);
4805 }
4806 }
4807
4808 #ifdef CONFIG_NET_FLOW_LIMIT
4809 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4810 #endif
4811
4812 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4813 {
4814 #ifdef CONFIG_NET_FLOW_LIMIT
4815 struct sd_flow_limit *fl;
4816 struct softnet_data *sd;
4817 unsigned int old_flow, new_flow;
4818
4819 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
4820 return false;
4821
4822 sd = this_cpu_ptr(&softnet_data);
4823
4824 rcu_read_lock();
4825 fl = rcu_dereference(sd->flow_limit);
4826 if (fl) {
4827 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4828 old_flow = fl->history[fl->history_head];
4829 fl->history[fl->history_head] = new_flow;
4830
4831 fl->history_head++;
4832 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4833
4834 if (likely(fl->buckets[old_flow]))
4835 fl->buckets[old_flow]--;
4836
4837 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4838 fl->count++;
4839 rcu_read_unlock();
4840 return true;
4841 }
4842 }
4843 rcu_read_unlock();
4844 #endif
4845 return false;
4846 }
4847
4848 /*
4849 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4850 * queue (may be a remote CPU queue).
4851 */
4852 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4853 unsigned int *qtail)
4854 {
4855 enum skb_drop_reason reason;
4856 struct softnet_data *sd;
4857 unsigned long flags;
4858 unsigned int qlen;
4859 int max_backlog;
4860 u32 tail;
4861
4862 reason = SKB_DROP_REASON_DEV_READY;
4863 if (!netif_running(skb->dev))
4864 goto bad_dev;
4865
4866 reason = SKB_DROP_REASON_CPU_BACKLOG;
4867 sd = &per_cpu(softnet_data, cpu);
4868
4869 qlen = skb_queue_len_lockless(&sd->input_pkt_queue);
4870 max_backlog = READ_ONCE(net_hotdata.max_backlog);
4871 if (unlikely(qlen > max_backlog))
4872 goto cpu_backlog_drop;
4873 backlog_lock_irq_save(sd, &flags);
4874 qlen = skb_queue_len(&sd->input_pkt_queue);
4875 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) {
4876 if (!qlen) {
4877 /* Schedule NAPI for backlog device. We can use
4878 * non atomic operation as we own the queue lock.
4879 */
4880 if (!__test_and_set_bit(NAPI_STATE_SCHED,
4881 &sd->backlog.state))
4882 napi_schedule_rps(sd);
4883 }
4884 __skb_queue_tail(&sd->input_pkt_queue, skb);
4885 tail = rps_input_queue_tail_incr(sd);
4886 backlog_unlock_irq_restore(sd, &flags);
4887
4888 /* save the tail outside of the critical section */
4889 rps_input_queue_tail_save(qtail, tail);
4890 return NET_RX_SUCCESS;
4891 }
4892
4893 backlog_unlock_irq_restore(sd, &flags);
4894
4895 cpu_backlog_drop:
4896 atomic_inc(&sd->dropped);
4897 bad_dev:
4898 dev_core_stats_rx_dropped_inc(skb->dev);
4899 kfree_skb_reason(skb, reason);
4900 return NET_RX_DROP;
4901 }
4902
4903 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4904 {
4905 struct net_device *dev = skb->dev;
4906 struct netdev_rx_queue *rxqueue;
4907
4908 rxqueue = dev->_rx;
4909
4910 if (skb_rx_queue_recorded(skb)) {
4911 u16 index = skb_get_rx_queue(skb);
4912
4913 if (unlikely(index >= dev->real_num_rx_queues)) {
4914 WARN_ONCE(dev->real_num_rx_queues > 1,
4915 "%s received packet on queue %u, but number "
4916 "of RX queues is %u\n",
4917 dev->name, index, dev->real_num_rx_queues);
4918
4919 return rxqueue; /* Return first rxqueue */
4920 }
4921 rxqueue += index;
4922 }
4923 return rxqueue;
4924 }
4925
4926 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4927 struct bpf_prog *xdp_prog)
4928 {
4929 void *orig_data, *orig_data_end, *hard_start;
4930 struct netdev_rx_queue *rxqueue;
4931 bool orig_bcast, orig_host;
4932 u32 mac_len, frame_sz;
4933 __be16 orig_eth_type;
4934 struct ethhdr *eth;
4935 u32 metalen, act;
4936 int off;
4937
4938 /* The XDP program wants to see the packet starting at the MAC
4939 * header.
4940 */
4941 mac_len = skb->data - skb_mac_header(skb);
4942 hard_start = skb->data - skb_headroom(skb);
4943
4944 /* SKB "head" area always have tailroom for skb_shared_info */
4945 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4946 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4947
4948 rxqueue = netif_get_rxqueue(skb);
4949 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4950 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4951 skb_headlen(skb) + mac_len, true);
4952 if (skb_is_nonlinear(skb)) {
4953 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
4954 xdp_buff_set_frags_flag(xdp);
4955 } else {
4956 xdp_buff_clear_frags_flag(xdp);
4957 }
4958
4959 orig_data_end = xdp->data_end;
4960 orig_data = xdp->data;
4961 eth = (struct ethhdr *)xdp->data;
4962 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4963 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4964 orig_eth_type = eth->h_proto;
4965
4966 act = bpf_prog_run_xdp(xdp_prog, xdp);
4967
4968 /* check if bpf_xdp_adjust_head was used */
4969 off = xdp->data - orig_data;
4970 if (off) {
4971 if (off > 0)
4972 __skb_pull(skb, off);
4973 else if (off < 0)
4974 __skb_push(skb, -off);
4975
4976 skb->mac_header += off;
4977 skb_reset_network_header(skb);
4978 }
4979
4980 /* check if bpf_xdp_adjust_tail was used */
4981 off = xdp->data_end - orig_data_end;
4982 if (off != 0) {
4983 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4984 skb->len += off; /* positive on grow, negative on shrink */
4985 }
4986
4987 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
4988 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
4989 */
4990 if (xdp_buff_has_frags(xdp))
4991 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
4992 else
4993 skb->data_len = 0;
4994
4995 /* check if XDP changed eth hdr such SKB needs update */
4996 eth = (struct ethhdr *)xdp->data;
4997 if ((orig_eth_type != eth->h_proto) ||
4998 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4999 skb->dev->dev_addr)) ||
5000 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
5001 __skb_push(skb, ETH_HLEN);
5002 skb->pkt_type = PACKET_HOST;
5003 skb->protocol = eth_type_trans(skb, skb->dev);
5004 }
5005
5006 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
5007 * before calling us again on redirect path. We do not call do_redirect
5008 * as we leave that up to the caller.
5009 *
5010 * Caller is responsible for managing lifetime of skb (i.e. calling
5011 * kfree_skb in response to actions it cannot handle/XDP_DROP).
5012 */
5013 switch (act) {
5014 case XDP_REDIRECT:
5015 case XDP_TX:
5016 __skb_push(skb, mac_len);
5017 break;
5018 case XDP_PASS:
5019 metalen = xdp->data - xdp->data_meta;
5020 if (metalen)
5021 skb_metadata_set(skb, metalen);
5022 break;
5023 }
5024
5025 return act;
5026 }
5027
5028 static int
5029 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog)
5030 {
5031 struct sk_buff *skb = *pskb;
5032 int err, hroom, troom;
5033
5034 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog))
5035 return 0;
5036
5037 /* In case we have to go down the path and also linearize,
5038 * then lets do the pskb_expand_head() work just once here.
5039 */
5040 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
5041 troom = skb->tail + skb->data_len - skb->end;
5042 err = pskb_expand_head(skb,
5043 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
5044 troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
5045 if (err)
5046 return err;
5047
5048 return skb_linearize(skb);
5049 }
5050
5051 static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
5052 struct xdp_buff *xdp,
5053 struct bpf_prog *xdp_prog)
5054 {
5055 struct sk_buff *skb = *pskb;
5056 u32 mac_len, act = XDP_DROP;
5057
5058 /* Reinjected packets coming from act_mirred or similar should
5059 * not get XDP generic processing.
5060 */
5061 if (skb_is_redirected(skb))
5062 return XDP_PASS;
5063
5064 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
5065 * bytes. This is the guarantee that also native XDP provides,
5066 * thus we need to do it here as well.
5067 */
5068 mac_len = skb->data - skb_mac_header(skb);
5069 __skb_push(skb, mac_len);
5070
5071 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
5072 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
5073 if (netif_skb_check_for_xdp(pskb, xdp_prog))
5074 goto do_drop;
5075 }
5076
5077 __skb_pull(*pskb, mac_len);
5078
5079 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
5080 switch (act) {
5081 case XDP_REDIRECT:
5082 case XDP_TX:
5083 case XDP_PASS:
5084 break;
5085 default:
5086 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
5087 fallthrough;
5088 case XDP_ABORTED:
5089 trace_xdp_exception((*pskb)->dev, xdp_prog, act);
5090 fallthrough;
5091 case XDP_DROP:
5092 do_drop:
5093 kfree_skb(*pskb);
5094 break;
5095 }
5096
5097 return act;
5098 }
5099
5100 /* When doing generic XDP we have to bypass the qdisc layer and the
5101 * network taps in order to match in-driver-XDP behavior. This also means
5102 * that XDP packets are able to starve other packets going through a qdisc,
5103 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
5104 * queues, so they do not have this starvation issue.
5105 */
5106 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
5107 {
5108 struct net_device *dev = skb->dev;
5109 struct netdev_queue *txq;
5110 bool free_skb = true;
5111 int cpu, rc;
5112
5113 txq = netdev_core_pick_tx(dev, skb, NULL);
5114 cpu = smp_processor_id();
5115 HARD_TX_LOCK(dev, txq, cpu);
5116 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
5117 rc = netdev_start_xmit(skb, dev, txq, 0);
5118 if (dev_xmit_complete(rc))
5119 free_skb = false;
5120 }
5121 HARD_TX_UNLOCK(dev, txq);
5122 if (free_skb) {
5123 trace_xdp_exception(dev, xdp_prog, XDP_TX);
5124 dev_core_stats_tx_dropped_inc(dev);
5125 kfree_skb(skb);
5126 }
5127 }
5128
5129 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
5130
5131 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb)
5132 {
5133 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
5134
5135 if (xdp_prog) {
5136 struct xdp_buff xdp;
5137 u32 act;
5138 int err;
5139
5140 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
5141 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
5142 if (act != XDP_PASS) {
5143 switch (act) {
5144 case XDP_REDIRECT:
5145 err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
5146 &xdp, xdp_prog);
5147 if (err)
5148 goto out_redir;
5149 break;
5150 case XDP_TX:
5151 generic_xdp_tx(*pskb, xdp_prog);
5152 break;
5153 }
5154 bpf_net_ctx_clear(bpf_net_ctx);
5155 return XDP_DROP;
5156 }
5157 bpf_net_ctx_clear(bpf_net_ctx);
5158 }
5159 return XDP_PASS;
5160 out_redir:
5161 bpf_net_ctx_clear(bpf_net_ctx);
5162 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
5163 return XDP_DROP;
5164 }
5165 EXPORT_SYMBOL_GPL(do_xdp_generic);
5166
5167 static int netif_rx_internal(struct sk_buff *skb)
5168 {
5169 int ret;
5170
5171 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5172
5173 trace_netif_rx(skb);
5174
5175 #ifdef CONFIG_RPS
5176 if (static_branch_unlikely(&rps_needed)) {
5177 struct rps_dev_flow voidflow, *rflow = &voidflow;
5178 int cpu;
5179
5180 rcu_read_lock();
5181
5182 cpu = get_rps_cpu(skb->dev, skb, &rflow);
5183 if (cpu < 0)
5184 cpu = smp_processor_id();
5185
5186 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5187
5188 rcu_read_unlock();
5189 } else
5190 #endif
5191 {
5192 unsigned int qtail;
5193
5194 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5195 }
5196 return ret;
5197 }
5198
5199 /**
5200 * __netif_rx - Slightly optimized version of netif_rx
5201 * @skb: buffer to post
5202 *
5203 * This behaves as netif_rx except that it does not disable bottom halves.
5204 * As a result this function may only be invoked from the interrupt context
5205 * (either hard or soft interrupt).
5206 */
5207 int __netif_rx(struct sk_buff *skb)
5208 {
5209 int ret;
5210
5211 lockdep_assert_once(hardirq_count() | softirq_count());
5212
5213 trace_netif_rx_entry(skb);
5214 ret = netif_rx_internal(skb);
5215 trace_netif_rx_exit(ret);
5216 return ret;
5217 }
5218 EXPORT_SYMBOL(__netif_rx);
5219
5220 /**
5221 * netif_rx - post buffer to the network code
5222 * @skb: buffer to post
5223 *
5224 * This function receives a packet from a device driver and queues it for
5225 * the upper (protocol) levels to process via the backlog NAPI device. It
5226 * always succeeds. The buffer may be dropped during processing for
5227 * congestion control or by the protocol layers.
5228 * The network buffer is passed via the backlog NAPI device. Modern NIC
5229 * driver should use NAPI and GRO.
5230 * This function can used from interrupt and from process context. The
5231 * caller from process context must not disable interrupts before invoking
5232 * this function.
5233 *
5234 * return values:
5235 * NET_RX_SUCCESS (no congestion)
5236 * NET_RX_DROP (packet was dropped)
5237 *
5238 */
5239 int netif_rx(struct sk_buff *skb)
5240 {
5241 bool need_bh_off = !(hardirq_count() | softirq_count());
5242 int ret;
5243
5244 if (need_bh_off)
5245 local_bh_disable();
5246 trace_netif_rx_entry(skb);
5247 ret = netif_rx_internal(skb);
5248 trace_netif_rx_exit(ret);
5249 if (need_bh_off)
5250 local_bh_enable();
5251 return ret;
5252 }
5253 EXPORT_SYMBOL(netif_rx);
5254
5255 static __latent_entropy void net_tx_action(struct softirq_action *h)
5256 {
5257 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5258
5259 if (sd->completion_queue) {
5260 struct sk_buff *clist;
5261
5262 local_irq_disable();
5263 clist = sd->completion_queue;
5264 sd->completion_queue = NULL;
5265 local_irq_enable();
5266
5267 while (clist) {
5268 struct sk_buff *skb = clist;
5269
5270 clist = clist->next;
5271
5272 WARN_ON(refcount_read(&skb->users));
5273 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5274 trace_consume_skb(skb, net_tx_action);
5275 else
5276 trace_kfree_skb(skb, net_tx_action,
5277 get_kfree_skb_cb(skb)->reason, NULL);
5278
5279 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5280 __kfree_skb(skb);
5281 else
5282 __napi_kfree_skb(skb,
5283 get_kfree_skb_cb(skb)->reason);
5284 }
5285 }
5286
5287 if (sd->output_queue) {
5288 struct Qdisc *head;
5289
5290 local_irq_disable();
5291 head = sd->output_queue;
5292 sd->output_queue = NULL;
5293 sd->output_queue_tailp = &sd->output_queue;
5294 local_irq_enable();
5295
5296 rcu_read_lock();
5297
5298 while (head) {
5299 struct Qdisc *q = head;
5300 spinlock_t *root_lock = NULL;
5301
5302 head = head->next_sched;
5303
5304 /* We need to make sure head->next_sched is read
5305 * before clearing __QDISC_STATE_SCHED
5306 */
5307 smp_mb__before_atomic();
5308
5309 if (!(q->flags & TCQ_F_NOLOCK)) {
5310 root_lock = qdisc_lock(q);
5311 spin_lock(root_lock);
5312 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5313 &q->state))) {
5314 /* There is a synchronize_net() between
5315 * STATE_DEACTIVATED flag being set and
5316 * qdisc_reset()/some_qdisc_is_busy() in
5317 * dev_deactivate(), so we can safely bail out
5318 * early here to avoid data race between
5319 * qdisc_deactivate() and some_qdisc_is_busy()
5320 * for lockless qdisc.
5321 */
5322 clear_bit(__QDISC_STATE_SCHED, &q->state);
5323 continue;
5324 }
5325
5326 clear_bit(__QDISC_STATE_SCHED, &q->state);
5327 qdisc_run(q);
5328 if (root_lock)
5329 spin_unlock(root_lock);
5330 }
5331
5332 rcu_read_unlock();
5333 }
5334
5335 xfrm_dev_backlog(sd);
5336 }
5337
5338 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5339 /* This hook is defined here for ATM LANE */
5340 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5341 unsigned char *addr) __read_mostly;
5342 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5343 #endif
5344
5345 /**
5346 * netdev_is_rx_handler_busy - check if receive handler is registered
5347 * @dev: device to check
5348 *
5349 * Check if a receive handler is already registered for a given device.
5350 * Return true if there one.
5351 *
5352 * The caller must hold the rtnl_mutex.
5353 */
5354 bool netdev_is_rx_handler_busy(struct net_device *dev)
5355 {
5356 ASSERT_RTNL();
5357 return dev && rtnl_dereference(dev->rx_handler);
5358 }
5359 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5360
5361 /**
5362 * netdev_rx_handler_register - register receive handler
5363 * @dev: device to register a handler for
5364 * @rx_handler: receive handler to register
5365 * @rx_handler_data: data pointer that is used by rx handler
5366 *
5367 * Register a receive handler for a device. This handler will then be
5368 * called from __netif_receive_skb. A negative errno code is returned
5369 * on a failure.
5370 *
5371 * The caller must hold the rtnl_mutex.
5372 *
5373 * For a general description of rx_handler, see enum rx_handler_result.
5374 */
5375 int netdev_rx_handler_register(struct net_device *dev,
5376 rx_handler_func_t *rx_handler,
5377 void *rx_handler_data)
5378 {
5379 if (netdev_is_rx_handler_busy(dev))
5380 return -EBUSY;
5381
5382 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5383 return -EINVAL;
5384
5385 /* Note: rx_handler_data must be set before rx_handler */
5386 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5387 rcu_assign_pointer(dev->rx_handler, rx_handler);
5388
5389 return 0;
5390 }
5391 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5392
5393 /**
5394 * netdev_rx_handler_unregister - unregister receive handler
5395 * @dev: device to unregister a handler from
5396 *
5397 * Unregister a receive handler from a device.
5398 *
5399 * The caller must hold the rtnl_mutex.
5400 */
5401 void netdev_rx_handler_unregister(struct net_device *dev)
5402 {
5403
5404 ASSERT_RTNL();
5405 RCU_INIT_POINTER(dev->rx_handler, NULL);
5406 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5407 * section has a guarantee to see a non NULL rx_handler_data
5408 * as well.
5409 */
5410 synchronize_net();
5411 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5412 }
5413 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5414
5415 /*
5416 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5417 * the special handling of PFMEMALLOC skbs.
5418 */
5419 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5420 {
5421 switch (skb->protocol) {
5422 case htons(ETH_P_ARP):
5423 case htons(ETH_P_IP):
5424 case htons(ETH_P_IPV6):
5425 case htons(ETH_P_8021Q):
5426 case htons(ETH_P_8021AD):
5427 return true;
5428 default:
5429 return false;
5430 }
5431 }
5432
5433 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5434 int *ret, struct net_device *orig_dev)
5435 {
5436 if (nf_hook_ingress_active(skb)) {
5437 int ingress_retval;
5438
5439 if (*pt_prev) {
5440 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5441 *pt_prev = NULL;
5442 }
5443
5444 rcu_read_lock();
5445 ingress_retval = nf_hook_ingress(skb);
5446 rcu_read_unlock();
5447 return ingress_retval;
5448 }
5449 return 0;
5450 }
5451
5452 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5453 struct packet_type **ppt_prev)
5454 {
5455 struct packet_type *ptype, *pt_prev;
5456 rx_handler_func_t *rx_handler;
5457 struct sk_buff *skb = *pskb;
5458 struct net_device *orig_dev;
5459 bool deliver_exact = false;
5460 int ret = NET_RX_DROP;
5461 __be16 type;
5462
5463 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5464
5465 trace_netif_receive_skb(skb);
5466
5467 orig_dev = skb->dev;
5468
5469 skb_reset_network_header(skb);
5470 if (!skb_transport_header_was_set(skb))
5471 skb_reset_transport_header(skb);
5472 skb_reset_mac_len(skb);
5473
5474 pt_prev = NULL;
5475
5476 another_round:
5477 skb->skb_iif = skb->dev->ifindex;
5478
5479 __this_cpu_inc(softnet_data.processed);
5480
5481 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5482 int ret2;
5483
5484 migrate_disable();
5485 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
5486 &skb);
5487 migrate_enable();
5488
5489 if (ret2 != XDP_PASS) {
5490 ret = NET_RX_DROP;
5491 goto out;
5492 }
5493 }
5494
5495 if (eth_type_vlan(skb->protocol)) {
5496 skb = skb_vlan_untag(skb);
5497 if (unlikely(!skb))
5498 goto out;
5499 }
5500
5501 if (skb_skip_tc_classify(skb))
5502 goto skip_classify;
5503
5504 if (pfmemalloc)
5505 goto skip_taps;
5506
5507 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) {
5508 if (pt_prev)
5509 ret = deliver_skb(skb, pt_prev, orig_dev);
5510 pt_prev = ptype;
5511 }
5512
5513 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5514 if (pt_prev)
5515 ret = deliver_skb(skb, pt_prev, orig_dev);
5516 pt_prev = ptype;
5517 }
5518
5519 skip_taps:
5520 #ifdef CONFIG_NET_INGRESS
5521 if (static_branch_unlikely(&ingress_needed_key)) {
5522 bool another = false;
5523
5524 nf_skip_egress(skb, true);
5525 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5526 &another);
5527 if (another)
5528 goto another_round;
5529 if (!skb)
5530 goto out;
5531
5532 nf_skip_egress(skb, false);
5533 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5534 goto out;
5535 }
5536 #endif
5537 skb_reset_redirect(skb);
5538 skip_classify:
5539 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5540 goto drop;
5541
5542 if (skb_vlan_tag_present(skb)) {
5543 if (pt_prev) {
5544 ret = deliver_skb(skb, pt_prev, orig_dev);
5545 pt_prev = NULL;
5546 }
5547 if (vlan_do_receive(&skb))
5548 goto another_round;
5549 else if (unlikely(!skb))
5550 goto out;
5551 }
5552
5553 rx_handler = rcu_dereference(skb->dev->rx_handler);
5554 if (rx_handler) {
5555 if (pt_prev) {
5556 ret = deliver_skb(skb, pt_prev, orig_dev);
5557 pt_prev = NULL;
5558 }
5559 switch (rx_handler(&skb)) {
5560 case RX_HANDLER_CONSUMED:
5561 ret = NET_RX_SUCCESS;
5562 goto out;
5563 case RX_HANDLER_ANOTHER:
5564 goto another_round;
5565 case RX_HANDLER_EXACT:
5566 deliver_exact = true;
5567 break;
5568 case RX_HANDLER_PASS:
5569 break;
5570 default:
5571 BUG();
5572 }
5573 }
5574
5575 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5576 check_vlan_id:
5577 if (skb_vlan_tag_get_id(skb)) {
5578 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5579 * find vlan device.
5580 */
5581 skb->pkt_type = PACKET_OTHERHOST;
5582 } else if (eth_type_vlan(skb->protocol)) {
5583 /* Outer header is 802.1P with vlan 0, inner header is
5584 * 802.1Q or 802.1AD and vlan_do_receive() above could
5585 * not find vlan dev for vlan id 0.
5586 */
5587 __vlan_hwaccel_clear_tag(skb);
5588 skb = skb_vlan_untag(skb);
5589 if (unlikely(!skb))
5590 goto out;
5591 if (vlan_do_receive(&skb))
5592 /* After stripping off 802.1P header with vlan 0
5593 * vlan dev is found for inner header.
5594 */
5595 goto another_round;
5596 else if (unlikely(!skb))
5597 goto out;
5598 else
5599 /* We have stripped outer 802.1P vlan 0 header.
5600 * But could not find vlan dev.
5601 * check again for vlan id to set OTHERHOST.
5602 */
5603 goto check_vlan_id;
5604 }
5605 /* Note: we might in the future use prio bits
5606 * and set skb->priority like in vlan_do_receive()
5607 * For the time being, just ignore Priority Code Point
5608 */
5609 __vlan_hwaccel_clear_tag(skb);
5610 }
5611
5612 type = skb->protocol;
5613
5614 /* deliver only exact match when indicated */
5615 if (likely(!deliver_exact)) {
5616 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5617 &ptype_base[ntohs(type) &
5618 PTYPE_HASH_MASK]);
5619 }
5620
5621 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5622 &orig_dev->ptype_specific);
5623
5624 if (unlikely(skb->dev != orig_dev)) {
5625 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5626 &skb->dev->ptype_specific);
5627 }
5628
5629 if (pt_prev) {
5630 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5631 goto drop;
5632 *ppt_prev = pt_prev;
5633 } else {
5634 drop:
5635 if (!deliver_exact)
5636 dev_core_stats_rx_dropped_inc(skb->dev);
5637 else
5638 dev_core_stats_rx_nohandler_inc(skb->dev);
5639 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5640 /* Jamal, now you will not able to escape explaining
5641 * me how you were going to use this. :-)
5642 */
5643 ret = NET_RX_DROP;
5644 }
5645
5646 out:
5647 /* The invariant here is that if *ppt_prev is not NULL
5648 * then skb should also be non-NULL.
5649 *
5650 * Apparently *ppt_prev assignment above holds this invariant due to
5651 * skb dereferencing near it.
5652 */
5653 *pskb = skb;
5654 return ret;
5655 }
5656
5657 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5658 {
5659 struct net_device *orig_dev = skb->dev;
5660 struct packet_type *pt_prev = NULL;
5661 int ret;
5662
5663 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5664 if (pt_prev)
5665 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5666 skb->dev, pt_prev, orig_dev);
5667 return ret;
5668 }
5669
5670 /**
5671 * netif_receive_skb_core - special purpose version of netif_receive_skb
5672 * @skb: buffer to process
5673 *
5674 * More direct receive version of netif_receive_skb(). It should
5675 * only be used by callers that have a need to skip RPS and Generic XDP.
5676 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5677 *
5678 * This function may only be called from softirq context and interrupts
5679 * should be enabled.
5680 *
5681 * Return values (usually ignored):
5682 * NET_RX_SUCCESS: no congestion
5683 * NET_RX_DROP: packet was dropped
5684 */
5685 int netif_receive_skb_core(struct sk_buff *skb)
5686 {
5687 int ret;
5688
5689 rcu_read_lock();
5690 ret = __netif_receive_skb_one_core(skb, false);
5691 rcu_read_unlock();
5692
5693 return ret;
5694 }
5695 EXPORT_SYMBOL(netif_receive_skb_core);
5696
5697 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5698 struct packet_type *pt_prev,
5699 struct net_device *orig_dev)
5700 {
5701 struct sk_buff *skb, *next;
5702
5703 if (!pt_prev)
5704 return;
5705 if (list_empty(head))
5706 return;
5707 if (pt_prev->list_func != NULL)
5708 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5709 ip_list_rcv, head, pt_prev, orig_dev);
5710 else
5711 list_for_each_entry_safe(skb, next, head, list) {
5712 skb_list_del_init(skb);
5713 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5714 }
5715 }
5716
5717 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5718 {
5719 /* Fast-path assumptions:
5720 * - There is no RX handler.
5721 * - Only one packet_type matches.
5722 * If either of these fails, we will end up doing some per-packet
5723 * processing in-line, then handling the 'last ptype' for the whole
5724 * sublist. This can't cause out-of-order delivery to any single ptype,
5725 * because the 'last ptype' must be constant across the sublist, and all
5726 * other ptypes are handled per-packet.
5727 */
5728 /* Current (common) ptype of sublist */
5729 struct packet_type *pt_curr = NULL;
5730 /* Current (common) orig_dev of sublist */
5731 struct net_device *od_curr = NULL;
5732 struct list_head sublist;
5733 struct sk_buff *skb, *next;
5734
5735 INIT_LIST_HEAD(&sublist);
5736 list_for_each_entry_safe(skb, next, head, list) {
5737 struct net_device *orig_dev = skb->dev;
5738 struct packet_type *pt_prev = NULL;
5739
5740 skb_list_del_init(skb);
5741 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5742 if (!pt_prev)
5743 continue;
5744 if (pt_curr != pt_prev || od_curr != orig_dev) {
5745 /* dispatch old sublist */
5746 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5747 /* start new sublist */
5748 INIT_LIST_HEAD(&sublist);
5749 pt_curr = pt_prev;
5750 od_curr = orig_dev;
5751 }
5752 list_add_tail(&skb->list, &sublist);
5753 }
5754
5755 /* dispatch final sublist */
5756 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5757 }
5758
5759 static int __netif_receive_skb(struct sk_buff *skb)
5760 {
5761 int ret;
5762
5763 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5764 unsigned int noreclaim_flag;
5765
5766 /*
5767 * PFMEMALLOC skbs are special, they should
5768 * - be delivered to SOCK_MEMALLOC sockets only
5769 * - stay away from userspace
5770 * - have bounded memory usage
5771 *
5772 * Use PF_MEMALLOC as this saves us from propagating the allocation
5773 * context down to all allocation sites.
5774 */
5775 noreclaim_flag = memalloc_noreclaim_save();
5776 ret = __netif_receive_skb_one_core(skb, true);
5777 memalloc_noreclaim_restore(noreclaim_flag);
5778 } else
5779 ret = __netif_receive_skb_one_core(skb, false);
5780
5781 return ret;
5782 }
5783
5784 static void __netif_receive_skb_list(struct list_head *head)
5785 {
5786 unsigned long noreclaim_flag = 0;
5787 struct sk_buff *skb, *next;
5788 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5789
5790 list_for_each_entry_safe(skb, next, head, list) {
5791 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5792 struct list_head sublist;
5793
5794 /* Handle the previous sublist */
5795 list_cut_before(&sublist, head, &skb->list);
5796 if (!list_empty(&sublist))
5797 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5798 pfmemalloc = !pfmemalloc;
5799 /* See comments in __netif_receive_skb */
5800 if (pfmemalloc)
5801 noreclaim_flag = memalloc_noreclaim_save();
5802 else
5803 memalloc_noreclaim_restore(noreclaim_flag);
5804 }
5805 }
5806 /* Handle the remaining sublist */
5807 if (!list_empty(head))
5808 __netif_receive_skb_list_core(head, pfmemalloc);
5809 /* Restore pflags */
5810 if (pfmemalloc)
5811 memalloc_noreclaim_restore(noreclaim_flag);
5812 }
5813
5814 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5815 {
5816 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5817 struct bpf_prog *new = xdp->prog;
5818 int ret = 0;
5819
5820 switch (xdp->command) {
5821 case XDP_SETUP_PROG:
5822 rcu_assign_pointer(dev->xdp_prog, new);
5823 if (old)
5824 bpf_prog_put(old);
5825
5826 if (old && !new) {
5827 static_branch_dec(&generic_xdp_needed_key);
5828 } else if (new && !old) {
5829 static_branch_inc(&generic_xdp_needed_key);
5830 dev_disable_lro(dev);
5831 dev_disable_gro_hw(dev);
5832 }
5833 break;
5834
5835 default:
5836 ret = -EINVAL;
5837 break;
5838 }
5839
5840 return ret;
5841 }
5842
5843 static int netif_receive_skb_internal(struct sk_buff *skb)
5844 {
5845 int ret;
5846
5847 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5848
5849 if (skb_defer_rx_timestamp(skb))
5850 return NET_RX_SUCCESS;
5851
5852 rcu_read_lock();
5853 #ifdef CONFIG_RPS
5854 if (static_branch_unlikely(&rps_needed)) {
5855 struct rps_dev_flow voidflow, *rflow = &voidflow;
5856 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5857
5858 if (cpu >= 0) {
5859 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5860 rcu_read_unlock();
5861 return ret;
5862 }
5863 }
5864 #endif
5865 ret = __netif_receive_skb(skb);
5866 rcu_read_unlock();
5867 return ret;
5868 }
5869
5870 void netif_receive_skb_list_internal(struct list_head *head)
5871 {
5872 struct sk_buff *skb, *next;
5873 struct list_head sublist;
5874
5875 INIT_LIST_HEAD(&sublist);
5876 list_for_each_entry_safe(skb, next, head, list) {
5877 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
5878 skb);
5879 skb_list_del_init(skb);
5880 if (!skb_defer_rx_timestamp(skb))
5881 list_add_tail(&skb->list, &sublist);
5882 }
5883 list_splice_init(&sublist, head);
5884
5885 rcu_read_lock();
5886 #ifdef CONFIG_RPS
5887 if (static_branch_unlikely(&rps_needed)) {
5888 list_for_each_entry_safe(skb, next, head, list) {
5889 struct rps_dev_flow voidflow, *rflow = &voidflow;
5890 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5891
5892 if (cpu >= 0) {
5893 /* Will be handled, remove from list */
5894 skb_list_del_init(skb);
5895 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5896 }
5897 }
5898 }
5899 #endif
5900 __netif_receive_skb_list(head);
5901 rcu_read_unlock();
5902 }
5903
5904 /**
5905 * netif_receive_skb - process receive buffer from network
5906 * @skb: buffer to process
5907 *
5908 * netif_receive_skb() is the main receive data processing function.
5909 * It always succeeds. The buffer may be dropped during processing
5910 * for congestion control or by the protocol layers.
5911 *
5912 * This function may only be called from softirq context and interrupts
5913 * should be enabled.
5914 *
5915 * Return values (usually ignored):
5916 * NET_RX_SUCCESS: no congestion
5917 * NET_RX_DROP: packet was dropped
5918 */
5919 int netif_receive_skb(struct sk_buff *skb)
5920 {
5921 int ret;
5922
5923 trace_netif_receive_skb_entry(skb);
5924
5925 ret = netif_receive_skb_internal(skb);
5926 trace_netif_receive_skb_exit(ret);
5927
5928 return ret;
5929 }
5930 EXPORT_SYMBOL(netif_receive_skb);
5931
5932 /**
5933 * netif_receive_skb_list - process many receive buffers from network
5934 * @head: list of skbs to process.
5935 *
5936 * Since return value of netif_receive_skb() is normally ignored, and
5937 * wouldn't be meaningful for a list, this function returns void.
5938 *
5939 * This function may only be called from softirq context and interrupts
5940 * should be enabled.
5941 */
5942 void netif_receive_skb_list(struct list_head *head)
5943 {
5944 struct sk_buff *skb;
5945
5946 if (list_empty(head))
5947 return;
5948 if (trace_netif_receive_skb_list_entry_enabled()) {
5949 list_for_each_entry(skb, head, list)
5950 trace_netif_receive_skb_list_entry(skb);
5951 }
5952 netif_receive_skb_list_internal(head);
5953 trace_netif_receive_skb_list_exit(0);
5954 }
5955 EXPORT_SYMBOL(netif_receive_skb_list);
5956
5957 static DEFINE_PER_CPU(struct work_struct, flush_works);
5958
5959 /* Network device is going away, flush any packets still pending */
5960 static void flush_backlog(struct work_struct *work)
5961 {
5962 struct sk_buff *skb, *tmp;
5963 struct softnet_data *sd;
5964
5965 local_bh_disable();
5966 sd = this_cpu_ptr(&softnet_data);
5967
5968 backlog_lock_irq_disable(sd);
5969 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5970 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5971 __skb_unlink(skb, &sd->input_pkt_queue);
5972 dev_kfree_skb_irq(skb);
5973 rps_input_queue_head_incr(sd);
5974 }
5975 }
5976 backlog_unlock_irq_enable(sd);
5977
5978 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
5979 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5980 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5981 __skb_unlink(skb, &sd->process_queue);
5982 kfree_skb(skb);
5983 rps_input_queue_head_incr(sd);
5984 }
5985 }
5986 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
5987 local_bh_enable();
5988 }
5989
5990 static bool flush_required(int cpu)
5991 {
5992 #if IS_ENABLED(CONFIG_RPS)
5993 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5994 bool do_flush;
5995
5996 backlog_lock_irq_disable(sd);
5997
5998 /* as insertion into process_queue happens with the rps lock held,
5999 * process_queue access may race only with dequeue
6000 */
6001 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
6002 !skb_queue_empty_lockless(&sd->process_queue);
6003 backlog_unlock_irq_enable(sd);
6004
6005 return do_flush;
6006 #endif
6007 /* without RPS we can't safely check input_pkt_queue: during a
6008 * concurrent remote skb_queue_splice() we can detect as empty both
6009 * input_pkt_queue and process_queue even if the latter could end-up
6010 * containing a lot of packets.
6011 */
6012 return true;
6013 }
6014
6015 static void flush_all_backlogs(void)
6016 {
6017 static cpumask_t flush_cpus;
6018 unsigned int cpu;
6019
6020 /* since we are under rtnl lock protection we can use static data
6021 * for the cpumask and avoid allocating on stack the possibly
6022 * large mask
6023 */
6024 ASSERT_RTNL();
6025
6026 cpus_read_lock();
6027
6028 cpumask_clear(&flush_cpus);
6029 for_each_online_cpu(cpu) {
6030 if (flush_required(cpu)) {
6031 queue_work_on(cpu, system_highpri_wq,
6032 per_cpu_ptr(&flush_works, cpu));
6033 cpumask_set_cpu(cpu, &flush_cpus);
6034 }
6035 }
6036
6037 /* we can have in flight packet[s] on the cpus we are not flushing,
6038 * synchronize_net() in unregister_netdevice_many() will take care of
6039 * them
6040 */
6041 for_each_cpu(cpu, &flush_cpus)
6042 flush_work(per_cpu_ptr(&flush_works, cpu));
6043
6044 cpus_read_unlock();
6045 }
6046
6047 static void net_rps_send_ipi(struct softnet_data *remsd)
6048 {
6049 #ifdef CONFIG_RPS
6050 while (remsd) {
6051 struct softnet_data *next = remsd->rps_ipi_next;
6052
6053 if (cpu_online(remsd->cpu))
6054 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6055 remsd = next;
6056 }
6057 #endif
6058 }
6059
6060 /*
6061 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6062 * Note: called with local irq disabled, but exits with local irq enabled.
6063 */
6064 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6065 {
6066 #ifdef CONFIG_RPS
6067 struct softnet_data *remsd = sd->rps_ipi_list;
6068
6069 if (!use_backlog_threads() && remsd) {
6070 sd->rps_ipi_list = NULL;
6071
6072 local_irq_enable();
6073
6074 /* Send pending IPI's to kick RPS processing on remote cpus. */
6075 net_rps_send_ipi(remsd);
6076 } else
6077 #endif
6078 local_irq_enable();
6079 }
6080
6081 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6082 {
6083 #ifdef CONFIG_RPS
6084 return !use_backlog_threads() && sd->rps_ipi_list;
6085 #else
6086 return false;
6087 #endif
6088 }
6089
6090 static int process_backlog(struct napi_struct *napi, int quota)
6091 {
6092 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6093 bool again = true;
6094 int work = 0;
6095
6096 /* Check if we have pending ipi, its better to send them now,
6097 * not waiting net_rx_action() end.
6098 */
6099 if (sd_has_rps_ipi_waiting(sd)) {
6100 local_irq_disable();
6101 net_rps_action_and_irq_enable(sd);
6102 }
6103
6104 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
6105 while (again) {
6106 struct sk_buff *skb;
6107
6108 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6109 while ((skb = __skb_dequeue(&sd->process_queue))) {
6110 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6111 rcu_read_lock();
6112 __netif_receive_skb(skb);
6113 rcu_read_unlock();
6114 if (++work >= quota) {
6115 rps_input_queue_head_add(sd, work);
6116 return work;
6117 }
6118
6119 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6120 }
6121 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6122
6123 backlog_lock_irq_disable(sd);
6124 if (skb_queue_empty(&sd->input_pkt_queue)) {
6125 /*
6126 * Inline a custom version of __napi_complete().
6127 * only current cpu owns and manipulates this napi,
6128 * and NAPI_STATE_SCHED is the only possible flag set
6129 * on backlog.
6130 * We can use a plain write instead of clear_bit(),
6131 * and we dont need an smp_mb() memory barrier.
6132 */
6133 napi->state &= NAPIF_STATE_THREADED;
6134 again = false;
6135 } else {
6136 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6137 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6138 &sd->process_queue);
6139 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6140 }
6141 backlog_unlock_irq_enable(sd);
6142 }
6143
6144 if (work)
6145 rps_input_queue_head_add(sd, work);
6146 return work;
6147 }
6148
6149 /**
6150 * __napi_schedule - schedule for receive
6151 * @n: entry to schedule
6152 *
6153 * The entry's receive function will be scheduled to run.
6154 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6155 */
6156 void __napi_schedule(struct napi_struct *n)
6157 {
6158 unsigned long flags;
6159
6160 local_irq_save(flags);
6161 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6162 local_irq_restore(flags);
6163 }
6164 EXPORT_SYMBOL(__napi_schedule);
6165
6166 /**
6167 * napi_schedule_prep - check if napi can be scheduled
6168 * @n: napi context
6169 *
6170 * Test if NAPI routine is already running, and if not mark
6171 * it as running. This is used as a condition variable to
6172 * insure only one NAPI poll instance runs. We also make
6173 * sure there is no pending NAPI disable.
6174 */
6175 bool napi_schedule_prep(struct napi_struct *n)
6176 {
6177 unsigned long new, val = READ_ONCE(n->state);
6178
6179 do {
6180 if (unlikely(val & NAPIF_STATE_DISABLE))
6181 return false;
6182 new = val | NAPIF_STATE_SCHED;
6183
6184 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6185 * This was suggested by Alexander Duyck, as compiler
6186 * emits better code than :
6187 * if (val & NAPIF_STATE_SCHED)
6188 * new |= NAPIF_STATE_MISSED;
6189 */
6190 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6191 NAPIF_STATE_MISSED;
6192 } while (!try_cmpxchg(&n->state, &val, new));
6193
6194 return !(val & NAPIF_STATE_SCHED);
6195 }
6196 EXPORT_SYMBOL(napi_schedule_prep);
6197
6198 /**
6199 * __napi_schedule_irqoff - schedule for receive
6200 * @n: entry to schedule
6201 *
6202 * Variant of __napi_schedule() assuming hard irqs are masked.
6203 *
6204 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6205 * because the interrupt disabled assumption might not be true
6206 * due to force-threaded interrupts and spinlock substitution.
6207 */
6208 void __napi_schedule_irqoff(struct napi_struct *n)
6209 {
6210 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6211 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6212 else
6213 __napi_schedule(n);
6214 }
6215 EXPORT_SYMBOL(__napi_schedule_irqoff);
6216
6217 bool napi_complete_done(struct napi_struct *n, int work_done)
6218 {
6219 unsigned long flags, val, new, timeout = 0;
6220 bool ret = true;
6221
6222 /*
6223 * 1) Don't let napi dequeue from the cpu poll list
6224 * just in case its running on a different cpu.
6225 * 2) If we are busy polling, do nothing here, we have
6226 * the guarantee we will be called later.
6227 */
6228 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6229 NAPIF_STATE_IN_BUSY_POLL)))
6230 return false;
6231
6232 if (work_done) {
6233 if (n->gro_bitmask)
6234 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6235 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6236 }
6237 if (n->defer_hard_irqs_count > 0) {
6238 n->defer_hard_irqs_count--;
6239 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6240 if (timeout)
6241 ret = false;
6242 }
6243 if (n->gro_bitmask) {
6244 /* When the NAPI instance uses a timeout and keeps postponing
6245 * it, we need to bound somehow the time packets are kept in
6246 * the GRO layer
6247 */
6248 napi_gro_flush(n, !!timeout);
6249 }
6250
6251 gro_normal_list(n);
6252
6253 if (unlikely(!list_empty(&n->poll_list))) {
6254 /* If n->poll_list is not empty, we need to mask irqs */
6255 local_irq_save(flags);
6256 list_del_init(&n->poll_list);
6257 local_irq_restore(flags);
6258 }
6259 WRITE_ONCE(n->list_owner, -1);
6260
6261 val = READ_ONCE(n->state);
6262 do {
6263 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6264
6265 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6266 NAPIF_STATE_SCHED_THREADED |
6267 NAPIF_STATE_PREFER_BUSY_POLL);
6268
6269 /* If STATE_MISSED was set, leave STATE_SCHED set,
6270 * because we will call napi->poll() one more time.
6271 * This C code was suggested by Alexander Duyck to help gcc.
6272 */
6273 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6274 NAPIF_STATE_SCHED;
6275 } while (!try_cmpxchg(&n->state, &val, new));
6276
6277 if (unlikely(val & NAPIF_STATE_MISSED)) {
6278 __napi_schedule(n);
6279 return false;
6280 }
6281
6282 if (timeout)
6283 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6284 HRTIMER_MODE_REL_PINNED);
6285 return ret;
6286 }
6287 EXPORT_SYMBOL(napi_complete_done);
6288
6289 /* must be called under rcu_read_lock(), as we dont take a reference */
6290 struct napi_struct *napi_by_id(unsigned int napi_id)
6291 {
6292 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6293 struct napi_struct *napi;
6294
6295 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6296 if (napi->napi_id == napi_id)
6297 return napi;
6298
6299 return NULL;
6300 }
6301
6302 static void skb_defer_free_flush(struct softnet_data *sd)
6303 {
6304 struct sk_buff *skb, *next;
6305
6306 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6307 if (!READ_ONCE(sd->defer_list))
6308 return;
6309
6310 spin_lock(&sd->defer_lock);
6311 skb = sd->defer_list;
6312 sd->defer_list = NULL;
6313 sd->defer_count = 0;
6314 spin_unlock(&sd->defer_lock);
6315
6316 while (skb != NULL) {
6317 next = skb->next;
6318 napi_consume_skb(skb, 1);
6319 skb = next;
6320 }
6321 }
6322
6323 #if defined(CONFIG_NET_RX_BUSY_POLL)
6324
6325 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6326 {
6327 if (!skip_schedule) {
6328 gro_normal_list(napi);
6329 __napi_schedule(napi);
6330 return;
6331 }
6332
6333 if (napi->gro_bitmask) {
6334 /* flush too old packets
6335 * If HZ < 1000, flush all packets.
6336 */
6337 napi_gro_flush(napi, HZ >= 1000);
6338 }
6339
6340 gro_normal_list(napi);
6341 clear_bit(NAPI_STATE_SCHED, &napi->state);
6342 }
6343
6344 enum {
6345 NAPI_F_PREFER_BUSY_POLL = 1,
6346 NAPI_F_END_ON_RESCHED = 2,
6347 };
6348
6349 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock,
6350 unsigned flags, u16 budget)
6351 {
6352 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6353 bool skip_schedule = false;
6354 unsigned long timeout;
6355 int rc;
6356
6357 /* Busy polling means there is a high chance device driver hard irq
6358 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6359 * set in napi_schedule_prep().
6360 * Since we are about to call napi->poll() once more, we can safely
6361 * clear NAPI_STATE_MISSED.
6362 *
6363 * Note: x86 could use a single "lock and ..." instruction
6364 * to perform these two clear_bit()
6365 */
6366 clear_bit(NAPI_STATE_MISSED, &napi->state);
6367 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6368
6369 local_bh_disable();
6370 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6371
6372 if (flags & NAPI_F_PREFER_BUSY_POLL) {
6373 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6374 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6375 if (napi->defer_hard_irqs_count && timeout) {
6376 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6377 skip_schedule = true;
6378 }
6379 }
6380
6381 /* All we really want here is to re-enable device interrupts.
6382 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6383 */
6384 rc = napi->poll(napi, budget);
6385 /* We can't gro_normal_list() here, because napi->poll() might have
6386 * rearmed the napi (napi_complete_done()) in which case it could
6387 * already be running on another CPU.
6388 */
6389 trace_napi_poll(napi, rc, budget);
6390 netpoll_poll_unlock(have_poll_lock);
6391 if (rc == budget)
6392 __busy_poll_stop(napi, skip_schedule);
6393 bpf_net_ctx_clear(bpf_net_ctx);
6394 local_bh_enable();
6395 }
6396
6397 static void __napi_busy_loop(unsigned int napi_id,
6398 bool (*loop_end)(void *, unsigned long),
6399 void *loop_end_arg, unsigned flags, u16 budget)
6400 {
6401 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6402 int (*napi_poll)(struct napi_struct *napi, int budget);
6403 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6404 void *have_poll_lock = NULL;
6405 struct napi_struct *napi;
6406
6407 WARN_ON_ONCE(!rcu_read_lock_held());
6408
6409 restart:
6410 napi_poll = NULL;
6411
6412 napi = napi_by_id(napi_id);
6413 if (!napi)
6414 return;
6415
6416 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6417 preempt_disable();
6418 for (;;) {
6419 int work = 0;
6420
6421 local_bh_disable();
6422 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6423 if (!napi_poll) {
6424 unsigned long val = READ_ONCE(napi->state);
6425
6426 /* If multiple threads are competing for this napi,
6427 * we avoid dirtying napi->state as much as we can.
6428 */
6429 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6430 NAPIF_STATE_IN_BUSY_POLL)) {
6431 if (flags & NAPI_F_PREFER_BUSY_POLL)
6432 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6433 goto count;
6434 }
6435 if (cmpxchg(&napi->state, val,
6436 val | NAPIF_STATE_IN_BUSY_POLL |
6437 NAPIF_STATE_SCHED) != val) {
6438 if (flags & NAPI_F_PREFER_BUSY_POLL)
6439 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6440 goto count;
6441 }
6442 have_poll_lock = netpoll_poll_lock(napi);
6443 napi_poll = napi->poll;
6444 }
6445 work = napi_poll(napi, budget);
6446 trace_napi_poll(napi, work, budget);
6447 gro_normal_list(napi);
6448 count:
6449 if (work > 0)
6450 __NET_ADD_STATS(dev_net(napi->dev),
6451 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6452 skb_defer_free_flush(this_cpu_ptr(&softnet_data));
6453 bpf_net_ctx_clear(bpf_net_ctx);
6454 local_bh_enable();
6455
6456 if (!loop_end || loop_end(loop_end_arg, start_time))
6457 break;
6458
6459 if (unlikely(need_resched())) {
6460 if (flags & NAPI_F_END_ON_RESCHED)
6461 break;
6462 if (napi_poll)
6463 busy_poll_stop(napi, have_poll_lock, flags, budget);
6464 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6465 preempt_enable();
6466 rcu_read_unlock();
6467 cond_resched();
6468 rcu_read_lock();
6469 if (loop_end(loop_end_arg, start_time))
6470 return;
6471 goto restart;
6472 }
6473 cpu_relax();
6474 }
6475 if (napi_poll)
6476 busy_poll_stop(napi, have_poll_lock, flags, budget);
6477 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6478 preempt_enable();
6479 }
6480
6481 void napi_busy_loop_rcu(unsigned int napi_id,
6482 bool (*loop_end)(void *, unsigned long),
6483 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6484 {
6485 unsigned flags = NAPI_F_END_ON_RESCHED;
6486
6487 if (prefer_busy_poll)
6488 flags |= NAPI_F_PREFER_BUSY_POLL;
6489
6490 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6491 }
6492
6493 void napi_busy_loop(unsigned int napi_id,
6494 bool (*loop_end)(void *, unsigned long),
6495 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6496 {
6497 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0;
6498
6499 rcu_read_lock();
6500 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6501 rcu_read_unlock();
6502 }
6503 EXPORT_SYMBOL(napi_busy_loop);
6504
6505 #endif /* CONFIG_NET_RX_BUSY_POLL */
6506
6507 static void napi_hash_add(struct napi_struct *napi)
6508 {
6509 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6510 return;
6511
6512 spin_lock(&napi_hash_lock);
6513
6514 /* 0..NR_CPUS range is reserved for sender_cpu use */
6515 do {
6516 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6517 napi_gen_id = MIN_NAPI_ID;
6518 } while (napi_by_id(napi_gen_id));
6519 napi->napi_id = napi_gen_id;
6520
6521 hlist_add_head_rcu(&napi->napi_hash_node,
6522 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6523
6524 spin_unlock(&napi_hash_lock);
6525 }
6526
6527 /* Warning : caller is responsible to make sure rcu grace period
6528 * is respected before freeing memory containing @napi
6529 */
6530 static void napi_hash_del(struct napi_struct *napi)
6531 {
6532 spin_lock(&napi_hash_lock);
6533
6534 hlist_del_init_rcu(&napi->napi_hash_node);
6535
6536 spin_unlock(&napi_hash_lock);
6537 }
6538
6539 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6540 {
6541 struct napi_struct *napi;
6542
6543 napi = container_of(timer, struct napi_struct, timer);
6544
6545 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6546 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6547 */
6548 if (!napi_disable_pending(napi) &&
6549 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6550 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6551 __napi_schedule_irqoff(napi);
6552 }
6553
6554 return HRTIMER_NORESTART;
6555 }
6556
6557 static void init_gro_hash(struct napi_struct *napi)
6558 {
6559 int i;
6560
6561 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6562 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6563 napi->gro_hash[i].count = 0;
6564 }
6565 napi->gro_bitmask = 0;
6566 }
6567
6568 int dev_set_threaded(struct net_device *dev, bool threaded)
6569 {
6570 struct napi_struct *napi;
6571 int err = 0;
6572
6573 if (dev->threaded == threaded)
6574 return 0;
6575
6576 if (threaded) {
6577 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6578 if (!napi->thread) {
6579 err = napi_kthread_create(napi);
6580 if (err) {
6581 threaded = false;
6582 break;
6583 }
6584 }
6585 }
6586 }
6587
6588 WRITE_ONCE(dev->threaded, threaded);
6589
6590 /* Make sure kthread is created before THREADED bit
6591 * is set.
6592 */
6593 smp_mb__before_atomic();
6594
6595 /* Setting/unsetting threaded mode on a napi might not immediately
6596 * take effect, if the current napi instance is actively being
6597 * polled. In this case, the switch between threaded mode and
6598 * softirq mode will happen in the next round of napi_schedule().
6599 * This should not cause hiccups/stalls to the live traffic.
6600 */
6601 list_for_each_entry(napi, &dev->napi_list, dev_list)
6602 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6603
6604 return err;
6605 }
6606 EXPORT_SYMBOL(dev_set_threaded);
6607
6608 /**
6609 * netif_queue_set_napi - Associate queue with the napi
6610 * @dev: device to which NAPI and queue belong
6611 * @queue_index: Index of queue
6612 * @type: queue type as RX or TX
6613 * @napi: NAPI context, pass NULL to clear previously set NAPI
6614 *
6615 * Set queue with its corresponding napi context. This should be done after
6616 * registering the NAPI handler for the queue-vector and the queues have been
6617 * mapped to the corresponding interrupt vector.
6618 */
6619 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index,
6620 enum netdev_queue_type type, struct napi_struct *napi)
6621 {
6622 struct netdev_rx_queue *rxq;
6623 struct netdev_queue *txq;
6624
6625 if (WARN_ON_ONCE(napi && !napi->dev))
6626 return;
6627 if (dev->reg_state >= NETREG_REGISTERED)
6628 ASSERT_RTNL();
6629
6630 switch (type) {
6631 case NETDEV_QUEUE_TYPE_RX:
6632 rxq = __netif_get_rx_queue(dev, queue_index);
6633 rxq->napi = napi;
6634 return;
6635 case NETDEV_QUEUE_TYPE_TX:
6636 txq = netdev_get_tx_queue(dev, queue_index);
6637 txq->napi = napi;
6638 return;
6639 default:
6640 return;
6641 }
6642 }
6643 EXPORT_SYMBOL(netif_queue_set_napi);
6644
6645 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6646 int (*poll)(struct napi_struct *, int), int weight)
6647 {
6648 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6649 return;
6650
6651 INIT_LIST_HEAD(&napi->poll_list);
6652 INIT_HLIST_NODE(&napi->napi_hash_node);
6653 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6654 napi->timer.function = napi_watchdog;
6655 init_gro_hash(napi);
6656 napi->skb = NULL;
6657 INIT_LIST_HEAD(&napi->rx_list);
6658 napi->rx_count = 0;
6659 napi->poll = poll;
6660 if (weight > NAPI_POLL_WEIGHT)
6661 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6662 weight);
6663 napi->weight = weight;
6664 napi->dev = dev;
6665 #ifdef CONFIG_NETPOLL
6666 napi->poll_owner = -1;
6667 #endif
6668 napi->list_owner = -1;
6669 set_bit(NAPI_STATE_SCHED, &napi->state);
6670 set_bit(NAPI_STATE_NPSVC, &napi->state);
6671 list_add_rcu(&napi->dev_list, &dev->napi_list);
6672 napi_hash_add(napi);
6673 napi_get_frags_check(napi);
6674 /* Create kthread for this napi if dev->threaded is set.
6675 * Clear dev->threaded if kthread creation failed so that
6676 * threaded mode will not be enabled in napi_enable().
6677 */
6678 if (dev->threaded && napi_kthread_create(napi))
6679 dev->threaded = false;
6680 netif_napi_set_irq(napi, -1);
6681 }
6682 EXPORT_SYMBOL(netif_napi_add_weight);
6683
6684 void napi_disable(struct napi_struct *n)
6685 {
6686 unsigned long val, new;
6687
6688 might_sleep();
6689 set_bit(NAPI_STATE_DISABLE, &n->state);
6690
6691 val = READ_ONCE(n->state);
6692 do {
6693 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6694 usleep_range(20, 200);
6695 val = READ_ONCE(n->state);
6696 }
6697
6698 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6699 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6700 } while (!try_cmpxchg(&n->state, &val, new));
6701
6702 hrtimer_cancel(&n->timer);
6703
6704 clear_bit(NAPI_STATE_DISABLE, &n->state);
6705 }
6706 EXPORT_SYMBOL(napi_disable);
6707
6708 /**
6709 * napi_enable - enable NAPI scheduling
6710 * @n: NAPI context
6711 *
6712 * Resume NAPI from being scheduled on this context.
6713 * Must be paired with napi_disable.
6714 */
6715 void napi_enable(struct napi_struct *n)
6716 {
6717 unsigned long new, val = READ_ONCE(n->state);
6718
6719 do {
6720 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6721
6722 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6723 if (n->dev->threaded && n->thread)
6724 new |= NAPIF_STATE_THREADED;
6725 } while (!try_cmpxchg(&n->state, &val, new));
6726 }
6727 EXPORT_SYMBOL(napi_enable);
6728
6729 static void flush_gro_hash(struct napi_struct *napi)
6730 {
6731 int i;
6732
6733 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6734 struct sk_buff *skb, *n;
6735
6736 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6737 kfree_skb(skb);
6738 napi->gro_hash[i].count = 0;
6739 }
6740 }
6741
6742 /* Must be called in process context */
6743 void __netif_napi_del(struct napi_struct *napi)
6744 {
6745 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6746 return;
6747
6748 napi_hash_del(napi);
6749 list_del_rcu(&napi->dev_list);
6750 napi_free_frags(napi);
6751
6752 flush_gro_hash(napi);
6753 napi->gro_bitmask = 0;
6754
6755 if (napi->thread) {
6756 kthread_stop(napi->thread);
6757 napi->thread = NULL;
6758 }
6759 }
6760 EXPORT_SYMBOL(__netif_napi_del);
6761
6762 static int __napi_poll(struct napi_struct *n, bool *repoll)
6763 {
6764 int work, weight;
6765
6766 weight = n->weight;
6767
6768 /* This NAPI_STATE_SCHED test is for avoiding a race
6769 * with netpoll's poll_napi(). Only the entity which
6770 * obtains the lock and sees NAPI_STATE_SCHED set will
6771 * actually make the ->poll() call. Therefore we avoid
6772 * accidentally calling ->poll() when NAPI is not scheduled.
6773 */
6774 work = 0;
6775 if (napi_is_scheduled(n)) {
6776 work = n->poll(n, weight);
6777 trace_napi_poll(n, work, weight);
6778
6779 xdp_do_check_flushed(n);
6780 }
6781
6782 if (unlikely(work > weight))
6783 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6784 n->poll, work, weight);
6785
6786 if (likely(work < weight))
6787 return work;
6788
6789 /* Drivers must not modify the NAPI state if they
6790 * consume the entire weight. In such cases this code
6791 * still "owns" the NAPI instance and therefore can
6792 * move the instance around on the list at-will.
6793 */
6794 if (unlikely(napi_disable_pending(n))) {
6795 napi_complete(n);
6796 return work;
6797 }
6798
6799 /* The NAPI context has more processing work, but busy-polling
6800 * is preferred. Exit early.
6801 */
6802 if (napi_prefer_busy_poll(n)) {
6803 if (napi_complete_done(n, work)) {
6804 /* If timeout is not set, we need to make sure
6805 * that the NAPI is re-scheduled.
6806 */
6807 napi_schedule(n);
6808 }
6809 return work;
6810 }
6811
6812 if (n->gro_bitmask) {
6813 /* flush too old packets
6814 * If HZ < 1000, flush all packets.
6815 */
6816 napi_gro_flush(n, HZ >= 1000);
6817 }
6818
6819 gro_normal_list(n);
6820
6821 /* Some drivers may have called napi_schedule
6822 * prior to exhausting their budget.
6823 */
6824 if (unlikely(!list_empty(&n->poll_list))) {
6825 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6826 n->dev ? n->dev->name : "backlog");
6827 return work;
6828 }
6829
6830 *repoll = true;
6831
6832 return work;
6833 }
6834
6835 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6836 {
6837 bool do_repoll = false;
6838 void *have;
6839 int work;
6840
6841 list_del_init(&n->poll_list);
6842
6843 have = netpoll_poll_lock(n);
6844
6845 work = __napi_poll(n, &do_repoll);
6846
6847 if (do_repoll)
6848 list_add_tail(&n->poll_list, repoll);
6849
6850 netpoll_poll_unlock(have);
6851
6852 return work;
6853 }
6854
6855 static int napi_thread_wait(struct napi_struct *napi)
6856 {
6857 set_current_state(TASK_INTERRUPTIBLE);
6858
6859 while (!kthread_should_stop()) {
6860 /* Testing SCHED_THREADED bit here to make sure the current
6861 * kthread owns this napi and could poll on this napi.
6862 * Testing SCHED bit is not enough because SCHED bit might be
6863 * set by some other busy poll thread or by napi_disable().
6864 */
6865 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) {
6866 WARN_ON(!list_empty(&napi->poll_list));
6867 __set_current_state(TASK_RUNNING);
6868 return 0;
6869 }
6870
6871 schedule();
6872 set_current_state(TASK_INTERRUPTIBLE);
6873 }
6874 __set_current_state(TASK_RUNNING);
6875
6876 return -1;
6877 }
6878
6879 static void napi_threaded_poll_loop(struct napi_struct *napi)
6880 {
6881 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6882 struct softnet_data *sd;
6883 unsigned long last_qs = jiffies;
6884
6885 for (;;) {
6886 bool repoll = false;
6887 void *have;
6888
6889 local_bh_disable();
6890 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6891
6892 sd = this_cpu_ptr(&softnet_data);
6893 sd->in_napi_threaded_poll = true;
6894
6895 have = netpoll_poll_lock(napi);
6896 __napi_poll(napi, &repoll);
6897 netpoll_poll_unlock(have);
6898
6899 sd->in_napi_threaded_poll = false;
6900 barrier();
6901
6902 if (sd_has_rps_ipi_waiting(sd)) {
6903 local_irq_disable();
6904 net_rps_action_and_irq_enable(sd);
6905 }
6906 skb_defer_free_flush(sd);
6907 bpf_net_ctx_clear(bpf_net_ctx);
6908 local_bh_enable();
6909
6910 if (!repoll)
6911 break;
6912
6913 rcu_softirq_qs_periodic(last_qs);
6914 cond_resched();
6915 }
6916 }
6917
6918 static int napi_threaded_poll(void *data)
6919 {
6920 struct napi_struct *napi = data;
6921
6922 while (!napi_thread_wait(napi))
6923 napi_threaded_poll_loop(napi);
6924
6925 return 0;
6926 }
6927
6928 static __latent_entropy void net_rx_action(struct softirq_action *h)
6929 {
6930 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6931 unsigned long time_limit = jiffies +
6932 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
6933 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6934 int budget = READ_ONCE(net_hotdata.netdev_budget);
6935 LIST_HEAD(list);
6936 LIST_HEAD(repoll);
6937
6938 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6939 start:
6940 sd->in_net_rx_action = true;
6941 local_irq_disable();
6942 list_splice_init(&sd->poll_list, &list);
6943 local_irq_enable();
6944
6945 for (;;) {
6946 struct napi_struct *n;
6947
6948 skb_defer_free_flush(sd);
6949
6950 if (list_empty(&list)) {
6951 if (list_empty(&repoll)) {
6952 sd->in_net_rx_action = false;
6953 barrier();
6954 /* We need to check if ____napi_schedule()
6955 * had refilled poll_list while
6956 * sd->in_net_rx_action was true.
6957 */
6958 if (!list_empty(&sd->poll_list))
6959 goto start;
6960 if (!sd_has_rps_ipi_waiting(sd))
6961 goto end;
6962 }
6963 break;
6964 }
6965
6966 n = list_first_entry(&list, struct napi_struct, poll_list);
6967 budget -= napi_poll(n, &repoll);
6968
6969 /* If softirq window is exhausted then punt.
6970 * Allow this to run for 2 jiffies since which will allow
6971 * an average latency of 1.5/HZ.
6972 */
6973 if (unlikely(budget <= 0 ||
6974 time_after_eq(jiffies, time_limit))) {
6975 sd->time_squeeze++;
6976 break;
6977 }
6978 }
6979
6980 local_irq_disable();
6981
6982 list_splice_tail_init(&sd->poll_list, &list);
6983 list_splice_tail(&repoll, &list);
6984 list_splice(&list, &sd->poll_list);
6985 if (!list_empty(&sd->poll_list))
6986 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6987 else
6988 sd->in_net_rx_action = false;
6989
6990 net_rps_action_and_irq_enable(sd);
6991 end:
6992 bpf_net_ctx_clear(bpf_net_ctx);
6993 }
6994
6995 struct netdev_adjacent {
6996 struct net_device *dev;
6997 netdevice_tracker dev_tracker;
6998
6999 /* upper master flag, there can only be one master device per list */
7000 bool master;
7001
7002 /* lookup ignore flag */
7003 bool ignore;
7004
7005 /* counter for the number of times this device was added to us */
7006 u16 ref_nr;
7007
7008 /* private field for the users */
7009 void *private;
7010
7011 struct list_head list;
7012 struct rcu_head rcu;
7013 };
7014
7015 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7016 struct list_head *adj_list)
7017 {
7018 struct netdev_adjacent *adj;
7019
7020 list_for_each_entry(adj, adj_list, list) {
7021 if (adj->dev == adj_dev)
7022 return adj;
7023 }
7024 return NULL;
7025 }
7026
7027 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7028 struct netdev_nested_priv *priv)
7029 {
7030 struct net_device *dev = (struct net_device *)priv->data;
7031
7032 return upper_dev == dev;
7033 }
7034
7035 /**
7036 * netdev_has_upper_dev - Check if device is linked to an upper device
7037 * @dev: device
7038 * @upper_dev: upper device to check
7039 *
7040 * Find out if a device is linked to specified upper device and return true
7041 * in case it is. Note that this checks only immediate upper device,
7042 * not through a complete stack of devices. The caller must hold the RTNL lock.
7043 */
7044 bool netdev_has_upper_dev(struct net_device *dev,
7045 struct net_device *upper_dev)
7046 {
7047 struct netdev_nested_priv priv = {
7048 .data = (void *)upper_dev,
7049 };
7050
7051 ASSERT_RTNL();
7052
7053 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7054 &priv);
7055 }
7056 EXPORT_SYMBOL(netdev_has_upper_dev);
7057
7058 /**
7059 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7060 * @dev: device
7061 * @upper_dev: upper device to check
7062 *
7063 * Find out if a device is linked to specified upper device and return true
7064 * in case it is. Note that this checks the entire upper device chain.
7065 * The caller must hold rcu lock.
7066 */
7067
7068 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7069 struct net_device *upper_dev)
7070 {
7071 struct netdev_nested_priv priv = {
7072 .data = (void *)upper_dev,
7073 };
7074
7075 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7076 &priv);
7077 }
7078 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7079
7080 /**
7081 * netdev_has_any_upper_dev - Check if device is linked to some device
7082 * @dev: device
7083 *
7084 * Find out if a device is linked to an upper device and return true in case
7085 * it is. The caller must hold the RTNL lock.
7086 */
7087 bool netdev_has_any_upper_dev(struct net_device *dev)
7088 {
7089 ASSERT_RTNL();
7090
7091 return !list_empty(&dev->adj_list.upper);
7092 }
7093 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7094
7095 /**
7096 * netdev_master_upper_dev_get - Get master upper device
7097 * @dev: device
7098 *
7099 * Find a master upper device and return pointer to it or NULL in case
7100 * it's not there. The caller must hold the RTNL lock.
7101 */
7102 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7103 {
7104 struct netdev_adjacent *upper;
7105
7106 ASSERT_RTNL();
7107
7108 if (list_empty(&dev->adj_list.upper))
7109 return NULL;
7110
7111 upper = list_first_entry(&dev->adj_list.upper,
7112 struct netdev_adjacent, list);
7113 if (likely(upper->master))
7114 return upper->dev;
7115 return NULL;
7116 }
7117 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7118
7119 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7120 {
7121 struct netdev_adjacent *upper;
7122
7123 ASSERT_RTNL();
7124
7125 if (list_empty(&dev->adj_list.upper))
7126 return NULL;
7127
7128 upper = list_first_entry(&dev->adj_list.upper,
7129 struct netdev_adjacent, list);
7130 if (likely(upper->master) && !upper->ignore)
7131 return upper->dev;
7132 return NULL;
7133 }
7134
7135 /**
7136 * netdev_has_any_lower_dev - Check if device is linked to some device
7137 * @dev: device
7138 *
7139 * Find out if a device is linked to a lower device and return true in case
7140 * it is. The caller must hold the RTNL lock.
7141 */
7142 static bool netdev_has_any_lower_dev(struct net_device *dev)
7143 {
7144 ASSERT_RTNL();
7145
7146 return !list_empty(&dev->adj_list.lower);
7147 }
7148
7149 void *netdev_adjacent_get_private(struct list_head *adj_list)
7150 {
7151 struct netdev_adjacent *adj;
7152
7153 adj = list_entry(adj_list, struct netdev_adjacent, list);
7154
7155 return adj->private;
7156 }
7157 EXPORT_SYMBOL(netdev_adjacent_get_private);
7158
7159 /**
7160 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7161 * @dev: device
7162 * @iter: list_head ** of the current position
7163 *
7164 * Gets the next device from the dev's upper list, starting from iter
7165 * position. The caller must hold RCU read lock.
7166 */
7167 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7168 struct list_head **iter)
7169 {
7170 struct netdev_adjacent *upper;
7171
7172 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7173
7174 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7175
7176 if (&upper->list == &dev->adj_list.upper)
7177 return NULL;
7178
7179 *iter = &upper->list;
7180
7181 return upper->dev;
7182 }
7183 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7184
7185 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7186 struct list_head **iter,
7187 bool *ignore)
7188 {
7189 struct netdev_adjacent *upper;
7190
7191 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7192
7193 if (&upper->list == &dev->adj_list.upper)
7194 return NULL;
7195
7196 *iter = &upper->list;
7197 *ignore = upper->ignore;
7198
7199 return upper->dev;
7200 }
7201
7202 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7203 struct list_head **iter)
7204 {
7205 struct netdev_adjacent *upper;
7206
7207 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7208
7209 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7210
7211 if (&upper->list == &dev->adj_list.upper)
7212 return NULL;
7213
7214 *iter = &upper->list;
7215
7216 return upper->dev;
7217 }
7218
7219 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7220 int (*fn)(struct net_device *dev,
7221 struct netdev_nested_priv *priv),
7222 struct netdev_nested_priv *priv)
7223 {
7224 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7225 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7226 int ret, cur = 0;
7227 bool ignore;
7228
7229 now = dev;
7230 iter = &dev->adj_list.upper;
7231
7232 while (1) {
7233 if (now != dev) {
7234 ret = fn(now, priv);
7235 if (ret)
7236 return ret;
7237 }
7238
7239 next = NULL;
7240 while (1) {
7241 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7242 if (!udev)
7243 break;
7244 if (ignore)
7245 continue;
7246
7247 next = udev;
7248 niter = &udev->adj_list.upper;
7249 dev_stack[cur] = now;
7250 iter_stack[cur++] = iter;
7251 break;
7252 }
7253
7254 if (!next) {
7255 if (!cur)
7256 return 0;
7257 next = dev_stack[--cur];
7258 niter = iter_stack[cur];
7259 }
7260
7261 now = next;
7262 iter = niter;
7263 }
7264
7265 return 0;
7266 }
7267
7268 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7269 int (*fn)(struct net_device *dev,
7270 struct netdev_nested_priv *priv),
7271 struct netdev_nested_priv *priv)
7272 {
7273 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7274 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7275 int ret, cur = 0;
7276
7277 now = dev;
7278 iter = &dev->adj_list.upper;
7279
7280 while (1) {
7281 if (now != dev) {
7282 ret = fn(now, priv);
7283 if (ret)
7284 return ret;
7285 }
7286
7287 next = NULL;
7288 while (1) {
7289 udev = netdev_next_upper_dev_rcu(now, &iter);
7290 if (!udev)
7291 break;
7292
7293 next = udev;
7294 niter = &udev->adj_list.upper;
7295 dev_stack[cur] = now;
7296 iter_stack[cur++] = iter;
7297 break;
7298 }
7299
7300 if (!next) {
7301 if (!cur)
7302 return 0;
7303 next = dev_stack[--cur];
7304 niter = iter_stack[cur];
7305 }
7306
7307 now = next;
7308 iter = niter;
7309 }
7310
7311 return 0;
7312 }
7313 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7314
7315 static bool __netdev_has_upper_dev(struct net_device *dev,
7316 struct net_device *upper_dev)
7317 {
7318 struct netdev_nested_priv priv = {
7319 .flags = 0,
7320 .data = (void *)upper_dev,
7321 };
7322
7323 ASSERT_RTNL();
7324
7325 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7326 &priv);
7327 }
7328
7329 /**
7330 * netdev_lower_get_next_private - Get the next ->private from the
7331 * lower neighbour list
7332 * @dev: device
7333 * @iter: list_head ** of the current position
7334 *
7335 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7336 * list, starting from iter position. The caller must hold either hold the
7337 * RTNL lock or its own locking that guarantees that the neighbour lower
7338 * list will remain unchanged.
7339 */
7340 void *netdev_lower_get_next_private(struct net_device *dev,
7341 struct list_head **iter)
7342 {
7343 struct netdev_adjacent *lower;
7344
7345 lower = list_entry(*iter, struct netdev_adjacent, list);
7346
7347 if (&lower->list == &dev->adj_list.lower)
7348 return NULL;
7349
7350 *iter = lower->list.next;
7351
7352 return lower->private;
7353 }
7354 EXPORT_SYMBOL(netdev_lower_get_next_private);
7355
7356 /**
7357 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7358 * lower neighbour list, RCU
7359 * variant
7360 * @dev: device
7361 * @iter: list_head ** of the current position
7362 *
7363 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7364 * list, starting from iter position. The caller must hold RCU read lock.
7365 */
7366 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7367 struct list_head **iter)
7368 {
7369 struct netdev_adjacent *lower;
7370
7371 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7372
7373 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7374
7375 if (&lower->list == &dev->adj_list.lower)
7376 return NULL;
7377
7378 *iter = &lower->list;
7379
7380 return lower->private;
7381 }
7382 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7383
7384 /**
7385 * netdev_lower_get_next - Get the next device from the lower neighbour
7386 * list
7387 * @dev: device
7388 * @iter: list_head ** of the current position
7389 *
7390 * Gets the next netdev_adjacent from the dev's lower neighbour
7391 * list, starting from iter position. The caller must hold RTNL lock or
7392 * its own locking that guarantees that the neighbour lower
7393 * list will remain unchanged.
7394 */
7395 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7396 {
7397 struct netdev_adjacent *lower;
7398
7399 lower = list_entry(*iter, struct netdev_adjacent, list);
7400
7401 if (&lower->list == &dev->adj_list.lower)
7402 return NULL;
7403
7404 *iter = lower->list.next;
7405
7406 return lower->dev;
7407 }
7408 EXPORT_SYMBOL(netdev_lower_get_next);
7409
7410 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7411 struct list_head **iter)
7412 {
7413 struct netdev_adjacent *lower;
7414
7415 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7416
7417 if (&lower->list == &dev->adj_list.lower)
7418 return NULL;
7419
7420 *iter = &lower->list;
7421
7422 return lower->dev;
7423 }
7424
7425 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7426 struct list_head **iter,
7427 bool *ignore)
7428 {
7429 struct netdev_adjacent *lower;
7430
7431 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7432
7433 if (&lower->list == &dev->adj_list.lower)
7434 return NULL;
7435
7436 *iter = &lower->list;
7437 *ignore = lower->ignore;
7438
7439 return lower->dev;
7440 }
7441
7442 int netdev_walk_all_lower_dev(struct net_device *dev,
7443 int (*fn)(struct net_device *dev,
7444 struct netdev_nested_priv *priv),
7445 struct netdev_nested_priv *priv)
7446 {
7447 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7448 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7449 int ret, cur = 0;
7450
7451 now = dev;
7452 iter = &dev->adj_list.lower;
7453
7454 while (1) {
7455 if (now != dev) {
7456 ret = fn(now, priv);
7457 if (ret)
7458 return ret;
7459 }
7460
7461 next = NULL;
7462 while (1) {
7463 ldev = netdev_next_lower_dev(now, &iter);
7464 if (!ldev)
7465 break;
7466
7467 next = ldev;
7468 niter = &ldev->adj_list.lower;
7469 dev_stack[cur] = now;
7470 iter_stack[cur++] = iter;
7471 break;
7472 }
7473
7474 if (!next) {
7475 if (!cur)
7476 return 0;
7477 next = dev_stack[--cur];
7478 niter = iter_stack[cur];
7479 }
7480
7481 now = next;
7482 iter = niter;
7483 }
7484
7485 return 0;
7486 }
7487 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7488
7489 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7490 int (*fn)(struct net_device *dev,
7491 struct netdev_nested_priv *priv),
7492 struct netdev_nested_priv *priv)
7493 {
7494 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7495 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7496 int ret, cur = 0;
7497 bool ignore;
7498
7499 now = dev;
7500 iter = &dev->adj_list.lower;
7501
7502 while (1) {
7503 if (now != dev) {
7504 ret = fn(now, priv);
7505 if (ret)
7506 return ret;
7507 }
7508
7509 next = NULL;
7510 while (1) {
7511 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7512 if (!ldev)
7513 break;
7514 if (ignore)
7515 continue;
7516
7517 next = ldev;
7518 niter = &ldev->adj_list.lower;
7519 dev_stack[cur] = now;
7520 iter_stack[cur++] = iter;
7521 break;
7522 }
7523
7524 if (!next) {
7525 if (!cur)
7526 return 0;
7527 next = dev_stack[--cur];
7528 niter = iter_stack[cur];
7529 }
7530
7531 now = next;
7532 iter = niter;
7533 }
7534
7535 return 0;
7536 }
7537
7538 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7539 struct list_head **iter)
7540 {
7541 struct netdev_adjacent *lower;
7542
7543 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7544 if (&lower->list == &dev->adj_list.lower)
7545 return NULL;
7546
7547 *iter = &lower->list;
7548
7549 return lower->dev;
7550 }
7551 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7552
7553 static u8 __netdev_upper_depth(struct net_device *dev)
7554 {
7555 struct net_device *udev;
7556 struct list_head *iter;
7557 u8 max_depth = 0;
7558 bool ignore;
7559
7560 for (iter = &dev->adj_list.upper,
7561 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7562 udev;
7563 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7564 if (ignore)
7565 continue;
7566 if (max_depth < udev->upper_level)
7567 max_depth = udev->upper_level;
7568 }
7569
7570 return max_depth;
7571 }
7572
7573 static u8 __netdev_lower_depth(struct net_device *dev)
7574 {
7575 struct net_device *ldev;
7576 struct list_head *iter;
7577 u8 max_depth = 0;
7578 bool ignore;
7579
7580 for (iter = &dev->adj_list.lower,
7581 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7582 ldev;
7583 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7584 if (ignore)
7585 continue;
7586 if (max_depth < ldev->lower_level)
7587 max_depth = ldev->lower_level;
7588 }
7589
7590 return max_depth;
7591 }
7592
7593 static int __netdev_update_upper_level(struct net_device *dev,
7594 struct netdev_nested_priv *__unused)
7595 {
7596 dev->upper_level = __netdev_upper_depth(dev) + 1;
7597 return 0;
7598 }
7599
7600 #ifdef CONFIG_LOCKDEP
7601 static LIST_HEAD(net_unlink_list);
7602
7603 static void net_unlink_todo(struct net_device *dev)
7604 {
7605 if (list_empty(&dev->unlink_list))
7606 list_add_tail(&dev->unlink_list, &net_unlink_list);
7607 }
7608 #endif
7609
7610 static int __netdev_update_lower_level(struct net_device *dev,
7611 struct netdev_nested_priv *priv)
7612 {
7613 dev->lower_level = __netdev_lower_depth(dev) + 1;
7614
7615 #ifdef CONFIG_LOCKDEP
7616 if (!priv)
7617 return 0;
7618
7619 if (priv->flags & NESTED_SYNC_IMM)
7620 dev->nested_level = dev->lower_level - 1;
7621 if (priv->flags & NESTED_SYNC_TODO)
7622 net_unlink_todo(dev);
7623 #endif
7624 return 0;
7625 }
7626
7627 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7628 int (*fn)(struct net_device *dev,
7629 struct netdev_nested_priv *priv),
7630 struct netdev_nested_priv *priv)
7631 {
7632 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7633 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7634 int ret, cur = 0;
7635
7636 now = dev;
7637 iter = &dev->adj_list.lower;
7638
7639 while (1) {
7640 if (now != dev) {
7641 ret = fn(now, priv);
7642 if (ret)
7643 return ret;
7644 }
7645
7646 next = NULL;
7647 while (1) {
7648 ldev = netdev_next_lower_dev_rcu(now, &iter);
7649 if (!ldev)
7650 break;
7651
7652 next = ldev;
7653 niter = &ldev->adj_list.lower;
7654 dev_stack[cur] = now;
7655 iter_stack[cur++] = iter;
7656 break;
7657 }
7658
7659 if (!next) {
7660 if (!cur)
7661 return 0;
7662 next = dev_stack[--cur];
7663 niter = iter_stack[cur];
7664 }
7665
7666 now = next;
7667 iter = niter;
7668 }
7669
7670 return 0;
7671 }
7672 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7673
7674 /**
7675 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7676 * lower neighbour list, RCU
7677 * variant
7678 * @dev: device
7679 *
7680 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7681 * list. The caller must hold RCU read lock.
7682 */
7683 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7684 {
7685 struct netdev_adjacent *lower;
7686
7687 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7688 struct netdev_adjacent, list);
7689 if (lower)
7690 return lower->private;
7691 return NULL;
7692 }
7693 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7694
7695 /**
7696 * netdev_master_upper_dev_get_rcu - Get master upper device
7697 * @dev: device
7698 *
7699 * Find a master upper device and return pointer to it or NULL in case
7700 * it's not there. The caller must hold the RCU read lock.
7701 */
7702 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7703 {
7704 struct netdev_adjacent *upper;
7705
7706 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7707 struct netdev_adjacent, list);
7708 if (upper && likely(upper->master))
7709 return upper->dev;
7710 return NULL;
7711 }
7712 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7713
7714 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7715 struct net_device *adj_dev,
7716 struct list_head *dev_list)
7717 {
7718 char linkname[IFNAMSIZ+7];
7719
7720 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7721 "upper_%s" : "lower_%s", adj_dev->name);
7722 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7723 linkname);
7724 }
7725 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7726 char *name,
7727 struct list_head *dev_list)
7728 {
7729 char linkname[IFNAMSIZ+7];
7730
7731 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7732 "upper_%s" : "lower_%s", name);
7733 sysfs_remove_link(&(dev->dev.kobj), linkname);
7734 }
7735
7736 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7737 struct net_device *adj_dev,
7738 struct list_head *dev_list)
7739 {
7740 return (dev_list == &dev->adj_list.upper ||
7741 dev_list == &dev->adj_list.lower) &&
7742 net_eq(dev_net(dev), dev_net(adj_dev));
7743 }
7744
7745 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7746 struct net_device *adj_dev,
7747 struct list_head *dev_list,
7748 void *private, bool master)
7749 {
7750 struct netdev_adjacent *adj;
7751 int ret;
7752
7753 adj = __netdev_find_adj(adj_dev, dev_list);
7754
7755 if (adj) {
7756 adj->ref_nr += 1;
7757 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7758 dev->name, adj_dev->name, adj->ref_nr);
7759
7760 return 0;
7761 }
7762
7763 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7764 if (!adj)
7765 return -ENOMEM;
7766
7767 adj->dev = adj_dev;
7768 adj->master = master;
7769 adj->ref_nr = 1;
7770 adj->private = private;
7771 adj->ignore = false;
7772 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7773
7774 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7775 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7776
7777 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7778 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7779 if (ret)
7780 goto free_adj;
7781 }
7782
7783 /* Ensure that master link is always the first item in list. */
7784 if (master) {
7785 ret = sysfs_create_link(&(dev->dev.kobj),
7786 &(adj_dev->dev.kobj), "master");
7787 if (ret)
7788 goto remove_symlinks;
7789
7790 list_add_rcu(&adj->list, dev_list);
7791 } else {
7792 list_add_tail_rcu(&adj->list, dev_list);
7793 }
7794
7795 return 0;
7796
7797 remove_symlinks:
7798 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7799 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7800 free_adj:
7801 netdev_put(adj_dev, &adj->dev_tracker);
7802 kfree(adj);
7803
7804 return ret;
7805 }
7806
7807 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7808 struct net_device *adj_dev,
7809 u16 ref_nr,
7810 struct list_head *dev_list)
7811 {
7812 struct netdev_adjacent *adj;
7813
7814 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7815 dev->name, adj_dev->name, ref_nr);
7816
7817 adj = __netdev_find_adj(adj_dev, dev_list);
7818
7819 if (!adj) {
7820 pr_err("Adjacency does not exist for device %s from %s\n",
7821 dev->name, adj_dev->name);
7822 WARN_ON(1);
7823 return;
7824 }
7825
7826 if (adj->ref_nr > ref_nr) {
7827 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7828 dev->name, adj_dev->name, ref_nr,
7829 adj->ref_nr - ref_nr);
7830 adj->ref_nr -= ref_nr;
7831 return;
7832 }
7833
7834 if (adj->master)
7835 sysfs_remove_link(&(dev->dev.kobj), "master");
7836
7837 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7838 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7839
7840 list_del_rcu(&adj->list);
7841 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7842 adj_dev->name, dev->name, adj_dev->name);
7843 netdev_put(adj_dev, &adj->dev_tracker);
7844 kfree_rcu(adj, rcu);
7845 }
7846
7847 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7848 struct net_device *upper_dev,
7849 struct list_head *up_list,
7850 struct list_head *down_list,
7851 void *private, bool master)
7852 {
7853 int ret;
7854
7855 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7856 private, master);
7857 if (ret)
7858 return ret;
7859
7860 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7861 private, false);
7862 if (ret) {
7863 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7864 return ret;
7865 }
7866
7867 return 0;
7868 }
7869
7870 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7871 struct net_device *upper_dev,
7872 u16 ref_nr,
7873 struct list_head *up_list,
7874 struct list_head *down_list)
7875 {
7876 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7877 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7878 }
7879
7880 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7881 struct net_device *upper_dev,
7882 void *private, bool master)
7883 {
7884 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7885 &dev->adj_list.upper,
7886 &upper_dev->adj_list.lower,
7887 private, master);
7888 }
7889
7890 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7891 struct net_device *upper_dev)
7892 {
7893 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7894 &dev->adj_list.upper,
7895 &upper_dev->adj_list.lower);
7896 }
7897
7898 static int __netdev_upper_dev_link(struct net_device *dev,
7899 struct net_device *upper_dev, bool master,
7900 void *upper_priv, void *upper_info,
7901 struct netdev_nested_priv *priv,
7902 struct netlink_ext_ack *extack)
7903 {
7904 struct netdev_notifier_changeupper_info changeupper_info = {
7905 .info = {
7906 .dev = dev,
7907 .extack = extack,
7908 },
7909 .upper_dev = upper_dev,
7910 .master = master,
7911 .linking = true,
7912 .upper_info = upper_info,
7913 };
7914 struct net_device *master_dev;
7915 int ret = 0;
7916
7917 ASSERT_RTNL();
7918
7919 if (dev == upper_dev)
7920 return -EBUSY;
7921
7922 /* To prevent loops, check if dev is not upper device to upper_dev. */
7923 if (__netdev_has_upper_dev(upper_dev, dev))
7924 return -EBUSY;
7925
7926 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7927 return -EMLINK;
7928
7929 if (!master) {
7930 if (__netdev_has_upper_dev(dev, upper_dev))
7931 return -EEXIST;
7932 } else {
7933 master_dev = __netdev_master_upper_dev_get(dev);
7934 if (master_dev)
7935 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7936 }
7937
7938 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7939 &changeupper_info.info);
7940 ret = notifier_to_errno(ret);
7941 if (ret)
7942 return ret;
7943
7944 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7945 master);
7946 if (ret)
7947 return ret;
7948
7949 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7950 &changeupper_info.info);
7951 ret = notifier_to_errno(ret);
7952 if (ret)
7953 goto rollback;
7954
7955 __netdev_update_upper_level(dev, NULL);
7956 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7957
7958 __netdev_update_lower_level(upper_dev, priv);
7959 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7960 priv);
7961
7962 return 0;
7963
7964 rollback:
7965 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7966
7967 return ret;
7968 }
7969
7970 /**
7971 * netdev_upper_dev_link - Add a link to the upper device
7972 * @dev: device
7973 * @upper_dev: new upper device
7974 * @extack: netlink extended ack
7975 *
7976 * Adds a link to device which is upper to this one. The caller must hold
7977 * the RTNL lock. On a failure a negative errno code is returned.
7978 * On success the reference counts are adjusted and the function
7979 * returns zero.
7980 */
7981 int netdev_upper_dev_link(struct net_device *dev,
7982 struct net_device *upper_dev,
7983 struct netlink_ext_ack *extack)
7984 {
7985 struct netdev_nested_priv priv = {
7986 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7987 .data = NULL,
7988 };
7989
7990 return __netdev_upper_dev_link(dev, upper_dev, false,
7991 NULL, NULL, &priv, extack);
7992 }
7993 EXPORT_SYMBOL(netdev_upper_dev_link);
7994
7995 /**
7996 * netdev_master_upper_dev_link - Add a master link to the upper device
7997 * @dev: device
7998 * @upper_dev: new upper device
7999 * @upper_priv: upper device private
8000 * @upper_info: upper info to be passed down via notifier
8001 * @extack: netlink extended ack
8002 *
8003 * Adds a link to device which is upper to this one. In this case, only
8004 * one master upper device can be linked, although other non-master devices
8005 * might be linked as well. The caller must hold the RTNL lock.
8006 * On a failure a negative errno code is returned. On success the reference
8007 * counts are adjusted and the function returns zero.
8008 */
8009 int netdev_master_upper_dev_link(struct net_device *dev,
8010 struct net_device *upper_dev,
8011 void *upper_priv, void *upper_info,
8012 struct netlink_ext_ack *extack)
8013 {
8014 struct netdev_nested_priv priv = {
8015 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8016 .data = NULL,
8017 };
8018
8019 return __netdev_upper_dev_link(dev, upper_dev, true,
8020 upper_priv, upper_info, &priv, extack);
8021 }
8022 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8023
8024 static void __netdev_upper_dev_unlink(struct net_device *dev,
8025 struct net_device *upper_dev,
8026 struct netdev_nested_priv *priv)
8027 {
8028 struct netdev_notifier_changeupper_info changeupper_info = {
8029 .info = {
8030 .dev = dev,
8031 },
8032 .upper_dev = upper_dev,
8033 .linking = false,
8034 };
8035
8036 ASSERT_RTNL();
8037
8038 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8039
8040 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8041 &changeupper_info.info);
8042
8043 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8044
8045 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8046 &changeupper_info.info);
8047
8048 __netdev_update_upper_level(dev, NULL);
8049 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8050
8051 __netdev_update_lower_level(upper_dev, priv);
8052 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8053 priv);
8054 }
8055
8056 /**
8057 * netdev_upper_dev_unlink - Removes a link to upper device
8058 * @dev: device
8059 * @upper_dev: new upper device
8060 *
8061 * Removes a link to device which is upper to this one. The caller must hold
8062 * the RTNL lock.
8063 */
8064 void netdev_upper_dev_unlink(struct net_device *dev,
8065 struct net_device *upper_dev)
8066 {
8067 struct netdev_nested_priv priv = {
8068 .flags = NESTED_SYNC_TODO,
8069 .data = NULL,
8070 };
8071
8072 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8073 }
8074 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8075
8076 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8077 struct net_device *lower_dev,
8078 bool val)
8079 {
8080 struct netdev_adjacent *adj;
8081
8082 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8083 if (adj)
8084 adj->ignore = val;
8085
8086 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8087 if (adj)
8088 adj->ignore = val;
8089 }
8090
8091 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8092 struct net_device *lower_dev)
8093 {
8094 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8095 }
8096
8097 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8098 struct net_device *lower_dev)
8099 {
8100 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8101 }
8102
8103 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8104 struct net_device *new_dev,
8105 struct net_device *dev,
8106 struct netlink_ext_ack *extack)
8107 {
8108 struct netdev_nested_priv priv = {
8109 .flags = 0,
8110 .data = NULL,
8111 };
8112 int err;
8113
8114 if (!new_dev)
8115 return 0;
8116
8117 if (old_dev && new_dev != old_dev)
8118 netdev_adjacent_dev_disable(dev, old_dev);
8119 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8120 extack);
8121 if (err) {
8122 if (old_dev && new_dev != old_dev)
8123 netdev_adjacent_dev_enable(dev, old_dev);
8124 return err;
8125 }
8126
8127 return 0;
8128 }
8129 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8130
8131 void netdev_adjacent_change_commit(struct net_device *old_dev,
8132 struct net_device *new_dev,
8133 struct net_device *dev)
8134 {
8135 struct netdev_nested_priv priv = {
8136 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8137 .data = NULL,
8138 };
8139
8140 if (!new_dev || !old_dev)
8141 return;
8142
8143 if (new_dev == old_dev)
8144 return;
8145
8146 netdev_adjacent_dev_enable(dev, old_dev);
8147 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8148 }
8149 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8150
8151 void netdev_adjacent_change_abort(struct net_device *old_dev,
8152 struct net_device *new_dev,
8153 struct net_device *dev)
8154 {
8155 struct netdev_nested_priv priv = {
8156 .flags = 0,
8157 .data = NULL,
8158 };
8159
8160 if (!new_dev)
8161 return;
8162
8163 if (old_dev && new_dev != old_dev)
8164 netdev_adjacent_dev_enable(dev, old_dev);
8165
8166 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8167 }
8168 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8169
8170 /**
8171 * netdev_bonding_info_change - Dispatch event about slave change
8172 * @dev: device
8173 * @bonding_info: info to dispatch
8174 *
8175 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8176 * The caller must hold the RTNL lock.
8177 */
8178 void netdev_bonding_info_change(struct net_device *dev,
8179 struct netdev_bonding_info *bonding_info)
8180 {
8181 struct netdev_notifier_bonding_info info = {
8182 .info.dev = dev,
8183 };
8184
8185 memcpy(&info.bonding_info, bonding_info,
8186 sizeof(struct netdev_bonding_info));
8187 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8188 &info.info);
8189 }
8190 EXPORT_SYMBOL(netdev_bonding_info_change);
8191
8192 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
8193 struct netlink_ext_ack *extack)
8194 {
8195 struct netdev_notifier_offload_xstats_info info = {
8196 .info.dev = dev,
8197 .info.extack = extack,
8198 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8199 };
8200 int err;
8201 int rc;
8202
8203 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
8204 GFP_KERNEL);
8205 if (!dev->offload_xstats_l3)
8206 return -ENOMEM;
8207
8208 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
8209 NETDEV_OFFLOAD_XSTATS_DISABLE,
8210 &info.info);
8211 err = notifier_to_errno(rc);
8212 if (err)
8213 goto free_stats;
8214
8215 return 0;
8216
8217 free_stats:
8218 kfree(dev->offload_xstats_l3);
8219 dev->offload_xstats_l3 = NULL;
8220 return err;
8221 }
8222
8223 int netdev_offload_xstats_enable(struct net_device *dev,
8224 enum netdev_offload_xstats_type type,
8225 struct netlink_ext_ack *extack)
8226 {
8227 ASSERT_RTNL();
8228
8229 if (netdev_offload_xstats_enabled(dev, type))
8230 return -EALREADY;
8231
8232 switch (type) {
8233 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8234 return netdev_offload_xstats_enable_l3(dev, extack);
8235 }
8236
8237 WARN_ON(1);
8238 return -EINVAL;
8239 }
8240 EXPORT_SYMBOL(netdev_offload_xstats_enable);
8241
8242 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
8243 {
8244 struct netdev_notifier_offload_xstats_info info = {
8245 .info.dev = dev,
8246 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8247 };
8248
8249 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
8250 &info.info);
8251 kfree(dev->offload_xstats_l3);
8252 dev->offload_xstats_l3 = NULL;
8253 }
8254
8255 int netdev_offload_xstats_disable(struct net_device *dev,
8256 enum netdev_offload_xstats_type type)
8257 {
8258 ASSERT_RTNL();
8259
8260 if (!netdev_offload_xstats_enabled(dev, type))
8261 return -EALREADY;
8262
8263 switch (type) {
8264 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8265 netdev_offload_xstats_disable_l3(dev);
8266 return 0;
8267 }
8268
8269 WARN_ON(1);
8270 return -EINVAL;
8271 }
8272 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8273
8274 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8275 {
8276 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8277 }
8278
8279 static struct rtnl_hw_stats64 *
8280 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8281 enum netdev_offload_xstats_type type)
8282 {
8283 switch (type) {
8284 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8285 return dev->offload_xstats_l3;
8286 }
8287
8288 WARN_ON(1);
8289 return NULL;
8290 }
8291
8292 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8293 enum netdev_offload_xstats_type type)
8294 {
8295 ASSERT_RTNL();
8296
8297 return netdev_offload_xstats_get_ptr(dev, type);
8298 }
8299 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8300
8301 struct netdev_notifier_offload_xstats_ru {
8302 bool used;
8303 };
8304
8305 struct netdev_notifier_offload_xstats_rd {
8306 struct rtnl_hw_stats64 stats;
8307 bool used;
8308 };
8309
8310 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8311 const struct rtnl_hw_stats64 *src)
8312 {
8313 dest->rx_packets += src->rx_packets;
8314 dest->tx_packets += src->tx_packets;
8315 dest->rx_bytes += src->rx_bytes;
8316 dest->tx_bytes += src->tx_bytes;
8317 dest->rx_errors += src->rx_errors;
8318 dest->tx_errors += src->tx_errors;
8319 dest->rx_dropped += src->rx_dropped;
8320 dest->tx_dropped += src->tx_dropped;
8321 dest->multicast += src->multicast;
8322 }
8323
8324 static int netdev_offload_xstats_get_used(struct net_device *dev,
8325 enum netdev_offload_xstats_type type,
8326 bool *p_used,
8327 struct netlink_ext_ack *extack)
8328 {
8329 struct netdev_notifier_offload_xstats_ru report_used = {};
8330 struct netdev_notifier_offload_xstats_info info = {
8331 .info.dev = dev,
8332 .info.extack = extack,
8333 .type = type,
8334 .report_used = &report_used,
8335 };
8336 int rc;
8337
8338 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8339 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8340 &info.info);
8341 *p_used = report_used.used;
8342 return notifier_to_errno(rc);
8343 }
8344
8345 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8346 enum netdev_offload_xstats_type type,
8347 struct rtnl_hw_stats64 *p_stats,
8348 bool *p_used,
8349 struct netlink_ext_ack *extack)
8350 {
8351 struct netdev_notifier_offload_xstats_rd report_delta = {};
8352 struct netdev_notifier_offload_xstats_info info = {
8353 .info.dev = dev,
8354 .info.extack = extack,
8355 .type = type,
8356 .report_delta = &report_delta,
8357 };
8358 struct rtnl_hw_stats64 *stats;
8359 int rc;
8360
8361 stats = netdev_offload_xstats_get_ptr(dev, type);
8362 if (WARN_ON(!stats))
8363 return -EINVAL;
8364
8365 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8366 &info.info);
8367
8368 /* Cache whatever we got, even if there was an error, otherwise the
8369 * successful stats retrievals would get lost.
8370 */
8371 netdev_hw_stats64_add(stats, &report_delta.stats);
8372
8373 if (p_stats)
8374 *p_stats = *stats;
8375 *p_used = report_delta.used;
8376
8377 return notifier_to_errno(rc);
8378 }
8379
8380 int netdev_offload_xstats_get(struct net_device *dev,
8381 enum netdev_offload_xstats_type type,
8382 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8383 struct netlink_ext_ack *extack)
8384 {
8385 ASSERT_RTNL();
8386
8387 if (p_stats)
8388 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8389 p_used, extack);
8390 else
8391 return netdev_offload_xstats_get_used(dev, type, p_used,
8392 extack);
8393 }
8394 EXPORT_SYMBOL(netdev_offload_xstats_get);
8395
8396 void
8397 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8398 const struct rtnl_hw_stats64 *stats)
8399 {
8400 report_delta->used = true;
8401 netdev_hw_stats64_add(&report_delta->stats, stats);
8402 }
8403 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8404
8405 void
8406 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8407 {
8408 report_used->used = true;
8409 }
8410 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8411
8412 void netdev_offload_xstats_push_delta(struct net_device *dev,
8413 enum netdev_offload_xstats_type type,
8414 const struct rtnl_hw_stats64 *p_stats)
8415 {
8416 struct rtnl_hw_stats64 *stats;
8417
8418 ASSERT_RTNL();
8419
8420 stats = netdev_offload_xstats_get_ptr(dev, type);
8421 if (WARN_ON(!stats))
8422 return;
8423
8424 netdev_hw_stats64_add(stats, p_stats);
8425 }
8426 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8427
8428 /**
8429 * netdev_get_xmit_slave - Get the xmit slave of master device
8430 * @dev: device
8431 * @skb: The packet
8432 * @all_slaves: assume all the slaves are active
8433 *
8434 * The reference counters are not incremented so the caller must be
8435 * careful with locks. The caller must hold RCU lock.
8436 * %NULL is returned if no slave is found.
8437 */
8438
8439 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8440 struct sk_buff *skb,
8441 bool all_slaves)
8442 {
8443 const struct net_device_ops *ops = dev->netdev_ops;
8444
8445 if (!ops->ndo_get_xmit_slave)
8446 return NULL;
8447 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8448 }
8449 EXPORT_SYMBOL(netdev_get_xmit_slave);
8450
8451 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8452 struct sock *sk)
8453 {
8454 const struct net_device_ops *ops = dev->netdev_ops;
8455
8456 if (!ops->ndo_sk_get_lower_dev)
8457 return NULL;
8458 return ops->ndo_sk_get_lower_dev(dev, sk);
8459 }
8460
8461 /**
8462 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8463 * @dev: device
8464 * @sk: the socket
8465 *
8466 * %NULL is returned if no lower device is found.
8467 */
8468
8469 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8470 struct sock *sk)
8471 {
8472 struct net_device *lower;
8473
8474 lower = netdev_sk_get_lower_dev(dev, sk);
8475 while (lower) {
8476 dev = lower;
8477 lower = netdev_sk_get_lower_dev(dev, sk);
8478 }
8479
8480 return dev;
8481 }
8482 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8483
8484 static void netdev_adjacent_add_links(struct net_device *dev)
8485 {
8486 struct netdev_adjacent *iter;
8487
8488 struct net *net = dev_net(dev);
8489
8490 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8491 if (!net_eq(net, dev_net(iter->dev)))
8492 continue;
8493 netdev_adjacent_sysfs_add(iter->dev, dev,
8494 &iter->dev->adj_list.lower);
8495 netdev_adjacent_sysfs_add(dev, iter->dev,
8496 &dev->adj_list.upper);
8497 }
8498
8499 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8500 if (!net_eq(net, dev_net(iter->dev)))
8501 continue;
8502 netdev_adjacent_sysfs_add(iter->dev, dev,
8503 &iter->dev->adj_list.upper);
8504 netdev_adjacent_sysfs_add(dev, iter->dev,
8505 &dev->adj_list.lower);
8506 }
8507 }
8508
8509 static void netdev_adjacent_del_links(struct net_device *dev)
8510 {
8511 struct netdev_adjacent *iter;
8512
8513 struct net *net = dev_net(dev);
8514
8515 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8516 if (!net_eq(net, dev_net(iter->dev)))
8517 continue;
8518 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8519 &iter->dev->adj_list.lower);
8520 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8521 &dev->adj_list.upper);
8522 }
8523
8524 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8525 if (!net_eq(net, dev_net(iter->dev)))
8526 continue;
8527 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8528 &iter->dev->adj_list.upper);
8529 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8530 &dev->adj_list.lower);
8531 }
8532 }
8533
8534 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8535 {
8536 struct netdev_adjacent *iter;
8537
8538 struct net *net = dev_net(dev);
8539
8540 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8541 if (!net_eq(net, dev_net(iter->dev)))
8542 continue;
8543 netdev_adjacent_sysfs_del(iter->dev, oldname,
8544 &iter->dev->adj_list.lower);
8545 netdev_adjacent_sysfs_add(iter->dev, dev,
8546 &iter->dev->adj_list.lower);
8547 }
8548
8549 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8550 if (!net_eq(net, dev_net(iter->dev)))
8551 continue;
8552 netdev_adjacent_sysfs_del(iter->dev, oldname,
8553 &iter->dev->adj_list.upper);
8554 netdev_adjacent_sysfs_add(iter->dev, dev,
8555 &iter->dev->adj_list.upper);
8556 }
8557 }
8558
8559 void *netdev_lower_dev_get_private(struct net_device *dev,
8560 struct net_device *lower_dev)
8561 {
8562 struct netdev_adjacent *lower;
8563
8564 if (!lower_dev)
8565 return NULL;
8566 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8567 if (!lower)
8568 return NULL;
8569
8570 return lower->private;
8571 }
8572 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8573
8574
8575 /**
8576 * netdev_lower_state_changed - Dispatch event about lower device state change
8577 * @lower_dev: device
8578 * @lower_state_info: state to dispatch
8579 *
8580 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8581 * The caller must hold the RTNL lock.
8582 */
8583 void netdev_lower_state_changed(struct net_device *lower_dev,
8584 void *lower_state_info)
8585 {
8586 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8587 .info.dev = lower_dev,
8588 };
8589
8590 ASSERT_RTNL();
8591 changelowerstate_info.lower_state_info = lower_state_info;
8592 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8593 &changelowerstate_info.info);
8594 }
8595 EXPORT_SYMBOL(netdev_lower_state_changed);
8596
8597 static void dev_change_rx_flags(struct net_device *dev, int flags)
8598 {
8599 const struct net_device_ops *ops = dev->netdev_ops;
8600
8601 if (ops->ndo_change_rx_flags)
8602 ops->ndo_change_rx_flags(dev, flags);
8603 }
8604
8605 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8606 {
8607 unsigned int old_flags = dev->flags;
8608 unsigned int promiscuity, flags;
8609 kuid_t uid;
8610 kgid_t gid;
8611
8612 ASSERT_RTNL();
8613
8614 promiscuity = dev->promiscuity + inc;
8615 if (promiscuity == 0) {
8616 /*
8617 * Avoid overflow.
8618 * If inc causes overflow, untouch promisc and return error.
8619 */
8620 if (unlikely(inc > 0)) {
8621 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8622 return -EOVERFLOW;
8623 }
8624 flags = old_flags & ~IFF_PROMISC;
8625 } else {
8626 flags = old_flags | IFF_PROMISC;
8627 }
8628 WRITE_ONCE(dev->promiscuity, promiscuity);
8629 if (flags != old_flags) {
8630 WRITE_ONCE(dev->flags, flags);
8631 netdev_info(dev, "%s promiscuous mode\n",
8632 dev->flags & IFF_PROMISC ? "entered" : "left");
8633 if (audit_enabled) {
8634 current_uid_gid(&uid, &gid);
8635 audit_log(audit_context(), GFP_ATOMIC,
8636 AUDIT_ANOM_PROMISCUOUS,
8637 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8638 dev->name, (dev->flags & IFF_PROMISC),
8639 (old_flags & IFF_PROMISC),
8640 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8641 from_kuid(&init_user_ns, uid),
8642 from_kgid(&init_user_ns, gid),
8643 audit_get_sessionid(current));
8644 }
8645
8646 dev_change_rx_flags(dev, IFF_PROMISC);
8647 }
8648 if (notify)
8649 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8650 return 0;
8651 }
8652
8653 /**
8654 * dev_set_promiscuity - update promiscuity count on a device
8655 * @dev: device
8656 * @inc: modifier
8657 *
8658 * Add or remove promiscuity from a device. While the count in the device
8659 * remains above zero the interface remains promiscuous. Once it hits zero
8660 * the device reverts back to normal filtering operation. A negative inc
8661 * value is used to drop promiscuity on the device.
8662 * Return 0 if successful or a negative errno code on error.
8663 */
8664 int dev_set_promiscuity(struct net_device *dev, int inc)
8665 {
8666 unsigned int old_flags = dev->flags;
8667 int err;
8668
8669 err = __dev_set_promiscuity(dev, inc, true);
8670 if (err < 0)
8671 return err;
8672 if (dev->flags != old_flags)
8673 dev_set_rx_mode(dev);
8674 return err;
8675 }
8676 EXPORT_SYMBOL(dev_set_promiscuity);
8677
8678 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8679 {
8680 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8681 unsigned int allmulti, flags;
8682
8683 ASSERT_RTNL();
8684
8685 allmulti = dev->allmulti + inc;
8686 if (allmulti == 0) {
8687 /*
8688 * Avoid overflow.
8689 * If inc causes overflow, untouch allmulti and return error.
8690 */
8691 if (unlikely(inc > 0)) {
8692 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8693 return -EOVERFLOW;
8694 }
8695 flags = old_flags & ~IFF_ALLMULTI;
8696 } else {
8697 flags = old_flags | IFF_ALLMULTI;
8698 }
8699 WRITE_ONCE(dev->allmulti, allmulti);
8700 if (flags != old_flags) {
8701 WRITE_ONCE(dev->flags, flags);
8702 netdev_info(dev, "%s allmulticast mode\n",
8703 dev->flags & IFF_ALLMULTI ? "entered" : "left");
8704 dev_change_rx_flags(dev, IFF_ALLMULTI);
8705 dev_set_rx_mode(dev);
8706 if (notify)
8707 __dev_notify_flags(dev, old_flags,
8708 dev->gflags ^ old_gflags, 0, NULL);
8709 }
8710 return 0;
8711 }
8712
8713 /**
8714 * dev_set_allmulti - update allmulti count on a device
8715 * @dev: device
8716 * @inc: modifier
8717 *
8718 * Add or remove reception of all multicast frames to a device. While the
8719 * count in the device remains above zero the interface remains listening
8720 * to all interfaces. Once it hits zero the device reverts back to normal
8721 * filtering operation. A negative @inc value is used to drop the counter
8722 * when releasing a resource needing all multicasts.
8723 * Return 0 if successful or a negative errno code on error.
8724 */
8725
8726 int dev_set_allmulti(struct net_device *dev, int inc)
8727 {
8728 return __dev_set_allmulti(dev, inc, true);
8729 }
8730 EXPORT_SYMBOL(dev_set_allmulti);
8731
8732 /*
8733 * Upload unicast and multicast address lists to device and
8734 * configure RX filtering. When the device doesn't support unicast
8735 * filtering it is put in promiscuous mode while unicast addresses
8736 * are present.
8737 */
8738 void __dev_set_rx_mode(struct net_device *dev)
8739 {
8740 const struct net_device_ops *ops = dev->netdev_ops;
8741
8742 /* dev_open will call this function so the list will stay sane. */
8743 if (!(dev->flags&IFF_UP))
8744 return;
8745
8746 if (!netif_device_present(dev))
8747 return;
8748
8749 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8750 /* Unicast addresses changes may only happen under the rtnl,
8751 * therefore calling __dev_set_promiscuity here is safe.
8752 */
8753 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8754 __dev_set_promiscuity(dev, 1, false);
8755 dev->uc_promisc = true;
8756 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8757 __dev_set_promiscuity(dev, -1, false);
8758 dev->uc_promisc = false;
8759 }
8760 }
8761
8762 if (ops->ndo_set_rx_mode)
8763 ops->ndo_set_rx_mode(dev);
8764 }
8765
8766 void dev_set_rx_mode(struct net_device *dev)
8767 {
8768 netif_addr_lock_bh(dev);
8769 __dev_set_rx_mode(dev);
8770 netif_addr_unlock_bh(dev);
8771 }
8772
8773 /**
8774 * dev_get_flags - get flags reported to userspace
8775 * @dev: device
8776 *
8777 * Get the combination of flag bits exported through APIs to userspace.
8778 */
8779 unsigned int dev_get_flags(const struct net_device *dev)
8780 {
8781 unsigned int flags;
8782
8783 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC |
8784 IFF_ALLMULTI |
8785 IFF_RUNNING |
8786 IFF_LOWER_UP |
8787 IFF_DORMANT)) |
8788 (READ_ONCE(dev->gflags) & (IFF_PROMISC |
8789 IFF_ALLMULTI));
8790
8791 if (netif_running(dev)) {
8792 if (netif_oper_up(dev))
8793 flags |= IFF_RUNNING;
8794 if (netif_carrier_ok(dev))
8795 flags |= IFF_LOWER_UP;
8796 if (netif_dormant(dev))
8797 flags |= IFF_DORMANT;
8798 }
8799
8800 return flags;
8801 }
8802 EXPORT_SYMBOL(dev_get_flags);
8803
8804 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8805 struct netlink_ext_ack *extack)
8806 {
8807 unsigned int old_flags = dev->flags;
8808 int ret;
8809
8810 ASSERT_RTNL();
8811
8812 /*
8813 * Set the flags on our device.
8814 */
8815
8816 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8817 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8818 IFF_AUTOMEDIA)) |
8819 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8820 IFF_ALLMULTI));
8821
8822 /*
8823 * Load in the correct multicast list now the flags have changed.
8824 */
8825
8826 if ((old_flags ^ flags) & IFF_MULTICAST)
8827 dev_change_rx_flags(dev, IFF_MULTICAST);
8828
8829 dev_set_rx_mode(dev);
8830
8831 /*
8832 * Have we downed the interface. We handle IFF_UP ourselves
8833 * according to user attempts to set it, rather than blindly
8834 * setting it.
8835 */
8836
8837 ret = 0;
8838 if ((old_flags ^ flags) & IFF_UP) {
8839 if (old_flags & IFF_UP)
8840 __dev_close(dev);
8841 else
8842 ret = __dev_open(dev, extack);
8843 }
8844
8845 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8846 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8847 unsigned int old_flags = dev->flags;
8848
8849 dev->gflags ^= IFF_PROMISC;
8850
8851 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8852 if (dev->flags != old_flags)
8853 dev_set_rx_mode(dev);
8854 }
8855
8856 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8857 * is important. Some (broken) drivers set IFF_PROMISC, when
8858 * IFF_ALLMULTI is requested not asking us and not reporting.
8859 */
8860 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8861 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8862
8863 dev->gflags ^= IFF_ALLMULTI;
8864 __dev_set_allmulti(dev, inc, false);
8865 }
8866
8867 return ret;
8868 }
8869
8870 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8871 unsigned int gchanges, u32 portid,
8872 const struct nlmsghdr *nlh)
8873 {
8874 unsigned int changes = dev->flags ^ old_flags;
8875
8876 if (gchanges)
8877 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8878
8879 if (changes & IFF_UP) {
8880 if (dev->flags & IFF_UP)
8881 call_netdevice_notifiers(NETDEV_UP, dev);
8882 else
8883 call_netdevice_notifiers(NETDEV_DOWN, dev);
8884 }
8885
8886 if (dev->flags & IFF_UP &&
8887 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8888 struct netdev_notifier_change_info change_info = {
8889 .info = {
8890 .dev = dev,
8891 },
8892 .flags_changed = changes,
8893 };
8894
8895 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8896 }
8897 }
8898
8899 /**
8900 * dev_change_flags - change device settings
8901 * @dev: device
8902 * @flags: device state flags
8903 * @extack: netlink extended ack
8904 *
8905 * Change settings on device based state flags. The flags are
8906 * in the userspace exported format.
8907 */
8908 int dev_change_flags(struct net_device *dev, unsigned int flags,
8909 struct netlink_ext_ack *extack)
8910 {
8911 int ret;
8912 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8913
8914 ret = __dev_change_flags(dev, flags, extack);
8915 if (ret < 0)
8916 return ret;
8917
8918 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8919 __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8920 return ret;
8921 }
8922 EXPORT_SYMBOL(dev_change_flags);
8923
8924 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8925 {
8926 const struct net_device_ops *ops = dev->netdev_ops;
8927
8928 if (ops->ndo_change_mtu)
8929 return ops->ndo_change_mtu(dev, new_mtu);
8930
8931 /* Pairs with all the lockless reads of dev->mtu in the stack */
8932 WRITE_ONCE(dev->mtu, new_mtu);
8933 return 0;
8934 }
8935 EXPORT_SYMBOL(__dev_set_mtu);
8936
8937 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8938 struct netlink_ext_ack *extack)
8939 {
8940 /* MTU must be positive, and in range */
8941 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8942 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8943 return -EINVAL;
8944 }
8945
8946 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8947 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8948 return -EINVAL;
8949 }
8950 return 0;
8951 }
8952
8953 /**
8954 * dev_set_mtu_ext - Change maximum transfer unit
8955 * @dev: device
8956 * @new_mtu: new transfer unit
8957 * @extack: netlink extended ack
8958 *
8959 * Change the maximum transfer size of the network device.
8960 */
8961 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8962 struct netlink_ext_ack *extack)
8963 {
8964 int err, orig_mtu;
8965
8966 if (new_mtu == dev->mtu)
8967 return 0;
8968
8969 err = dev_validate_mtu(dev, new_mtu, extack);
8970 if (err)
8971 return err;
8972
8973 if (!netif_device_present(dev))
8974 return -ENODEV;
8975
8976 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8977 err = notifier_to_errno(err);
8978 if (err)
8979 return err;
8980
8981 orig_mtu = dev->mtu;
8982 err = __dev_set_mtu(dev, new_mtu);
8983
8984 if (!err) {
8985 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8986 orig_mtu);
8987 err = notifier_to_errno(err);
8988 if (err) {
8989 /* setting mtu back and notifying everyone again,
8990 * so that they have a chance to revert changes.
8991 */
8992 __dev_set_mtu(dev, orig_mtu);
8993 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8994 new_mtu);
8995 }
8996 }
8997 return err;
8998 }
8999
9000 int dev_set_mtu(struct net_device *dev, int new_mtu)
9001 {
9002 struct netlink_ext_ack extack;
9003 int err;
9004
9005 memset(&extack, 0, sizeof(extack));
9006 err = dev_set_mtu_ext(dev, new_mtu, &extack);
9007 if (err && extack._msg)
9008 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
9009 return err;
9010 }
9011 EXPORT_SYMBOL(dev_set_mtu);
9012
9013 /**
9014 * dev_change_tx_queue_len - Change TX queue length of a netdevice
9015 * @dev: device
9016 * @new_len: new tx queue length
9017 */
9018 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9019 {
9020 unsigned int orig_len = dev->tx_queue_len;
9021 int res;
9022
9023 if (new_len != (unsigned int)new_len)
9024 return -ERANGE;
9025
9026 if (new_len != orig_len) {
9027 WRITE_ONCE(dev->tx_queue_len, new_len);
9028 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9029 res = notifier_to_errno(res);
9030 if (res)
9031 goto err_rollback;
9032 res = dev_qdisc_change_tx_queue_len(dev);
9033 if (res)
9034 goto err_rollback;
9035 }
9036
9037 return 0;
9038
9039 err_rollback:
9040 netdev_err(dev, "refused to change device tx_queue_len\n");
9041 WRITE_ONCE(dev->tx_queue_len, orig_len);
9042 return res;
9043 }
9044
9045 /**
9046 * dev_set_group - Change group this device belongs to
9047 * @dev: device
9048 * @new_group: group this device should belong to
9049 */
9050 void dev_set_group(struct net_device *dev, int new_group)
9051 {
9052 dev->group = new_group;
9053 }
9054
9055 /**
9056 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9057 * @dev: device
9058 * @addr: new address
9059 * @extack: netlink extended ack
9060 */
9061 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9062 struct netlink_ext_ack *extack)
9063 {
9064 struct netdev_notifier_pre_changeaddr_info info = {
9065 .info.dev = dev,
9066 .info.extack = extack,
9067 .dev_addr = addr,
9068 };
9069 int rc;
9070
9071 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9072 return notifier_to_errno(rc);
9073 }
9074 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9075
9076 /**
9077 * dev_set_mac_address - Change Media Access Control Address
9078 * @dev: device
9079 * @sa: new address
9080 * @extack: netlink extended ack
9081 *
9082 * Change the hardware (MAC) address of the device
9083 */
9084 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9085 struct netlink_ext_ack *extack)
9086 {
9087 const struct net_device_ops *ops = dev->netdev_ops;
9088 int err;
9089
9090 if (!ops->ndo_set_mac_address)
9091 return -EOPNOTSUPP;
9092 if (sa->sa_family != dev->type)
9093 return -EINVAL;
9094 if (!netif_device_present(dev))
9095 return -ENODEV;
9096 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9097 if (err)
9098 return err;
9099 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) {
9100 err = ops->ndo_set_mac_address(dev, sa);
9101 if (err)
9102 return err;
9103 }
9104 dev->addr_assign_type = NET_ADDR_SET;
9105 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9106 add_device_randomness(dev->dev_addr, dev->addr_len);
9107 return 0;
9108 }
9109 EXPORT_SYMBOL(dev_set_mac_address);
9110
9111 DECLARE_RWSEM(dev_addr_sem);
9112
9113 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9114 struct netlink_ext_ack *extack)
9115 {
9116 int ret;
9117
9118 down_write(&dev_addr_sem);
9119 ret = dev_set_mac_address(dev, sa, extack);
9120 up_write(&dev_addr_sem);
9121 return ret;
9122 }
9123 EXPORT_SYMBOL(dev_set_mac_address_user);
9124
9125 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9126 {
9127 size_t size = sizeof(sa->sa_data_min);
9128 struct net_device *dev;
9129 int ret = 0;
9130
9131 down_read(&dev_addr_sem);
9132 rcu_read_lock();
9133
9134 dev = dev_get_by_name_rcu(net, dev_name);
9135 if (!dev) {
9136 ret = -ENODEV;
9137 goto unlock;
9138 }
9139 if (!dev->addr_len)
9140 memset(sa->sa_data, 0, size);
9141 else
9142 memcpy(sa->sa_data, dev->dev_addr,
9143 min_t(size_t, size, dev->addr_len));
9144 sa->sa_family = dev->type;
9145
9146 unlock:
9147 rcu_read_unlock();
9148 up_read(&dev_addr_sem);
9149 return ret;
9150 }
9151 EXPORT_SYMBOL(dev_get_mac_address);
9152
9153 /**
9154 * dev_change_carrier - Change device carrier
9155 * @dev: device
9156 * @new_carrier: new value
9157 *
9158 * Change device carrier
9159 */
9160 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9161 {
9162 const struct net_device_ops *ops = dev->netdev_ops;
9163
9164 if (!ops->ndo_change_carrier)
9165 return -EOPNOTSUPP;
9166 if (!netif_device_present(dev))
9167 return -ENODEV;
9168 return ops->ndo_change_carrier(dev, new_carrier);
9169 }
9170
9171 /**
9172 * dev_get_phys_port_id - Get device physical port ID
9173 * @dev: device
9174 * @ppid: port ID
9175 *
9176 * Get device physical port ID
9177 */
9178 int dev_get_phys_port_id(struct net_device *dev,
9179 struct netdev_phys_item_id *ppid)
9180 {
9181 const struct net_device_ops *ops = dev->netdev_ops;
9182
9183 if (!ops->ndo_get_phys_port_id)
9184 return -EOPNOTSUPP;
9185 return ops->ndo_get_phys_port_id(dev, ppid);
9186 }
9187
9188 /**
9189 * dev_get_phys_port_name - Get device physical port name
9190 * @dev: device
9191 * @name: port name
9192 * @len: limit of bytes to copy to name
9193 *
9194 * Get device physical port name
9195 */
9196 int dev_get_phys_port_name(struct net_device *dev,
9197 char *name, size_t len)
9198 {
9199 const struct net_device_ops *ops = dev->netdev_ops;
9200 int err;
9201
9202 if (ops->ndo_get_phys_port_name) {
9203 err = ops->ndo_get_phys_port_name(dev, name, len);
9204 if (err != -EOPNOTSUPP)
9205 return err;
9206 }
9207 return devlink_compat_phys_port_name_get(dev, name, len);
9208 }
9209
9210 /**
9211 * dev_get_port_parent_id - Get the device's port parent identifier
9212 * @dev: network device
9213 * @ppid: pointer to a storage for the port's parent identifier
9214 * @recurse: allow/disallow recursion to lower devices
9215 *
9216 * Get the devices's port parent identifier
9217 */
9218 int dev_get_port_parent_id(struct net_device *dev,
9219 struct netdev_phys_item_id *ppid,
9220 bool recurse)
9221 {
9222 const struct net_device_ops *ops = dev->netdev_ops;
9223 struct netdev_phys_item_id first = { };
9224 struct net_device *lower_dev;
9225 struct list_head *iter;
9226 int err;
9227
9228 if (ops->ndo_get_port_parent_id) {
9229 err = ops->ndo_get_port_parent_id(dev, ppid);
9230 if (err != -EOPNOTSUPP)
9231 return err;
9232 }
9233
9234 err = devlink_compat_switch_id_get(dev, ppid);
9235 if (!recurse || err != -EOPNOTSUPP)
9236 return err;
9237
9238 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9239 err = dev_get_port_parent_id(lower_dev, ppid, true);
9240 if (err)
9241 break;
9242 if (!first.id_len)
9243 first = *ppid;
9244 else if (memcmp(&first, ppid, sizeof(*ppid)))
9245 return -EOPNOTSUPP;
9246 }
9247
9248 return err;
9249 }
9250 EXPORT_SYMBOL(dev_get_port_parent_id);
9251
9252 /**
9253 * netdev_port_same_parent_id - Indicate if two network devices have
9254 * the same port parent identifier
9255 * @a: first network device
9256 * @b: second network device
9257 */
9258 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9259 {
9260 struct netdev_phys_item_id a_id = { };
9261 struct netdev_phys_item_id b_id = { };
9262
9263 if (dev_get_port_parent_id(a, &a_id, true) ||
9264 dev_get_port_parent_id(b, &b_id, true))
9265 return false;
9266
9267 return netdev_phys_item_id_same(&a_id, &b_id);
9268 }
9269 EXPORT_SYMBOL(netdev_port_same_parent_id);
9270
9271 /**
9272 * dev_change_proto_down - set carrier according to proto_down.
9273 *
9274 * @dev: device
9275 * @proto_down: new value
9276 */
9277 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9278 {
9279 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
9280 return -EOPNOTSUPP;
9281 if (!netif_device_present(dev))
9282 return -ENODEV;
9283 if (proto_down)
9284 netif_carrier_off(dev);
9285 else
9286 netif_carrier_on(dev);
9287 WRITE_ONCE(dev->proto_down, proto_down);
9288 return 0;
9289 }
9290
9291 /**
9292 * dev_change_proto_down_reason - proto down reason
9293 *
9294 * @dev: device
9295 * @mask: proto down mask
9296 * @value: proto down value
9297 */
9298 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9299 u32 value)
9300 {
9301 u32 proto_down_reason;
9302 int b;
9303
9304 if (!mask) {
9305 proto_down_reason = value;
9306 } else {
9307 proto_down_reason = dev->proto_down_reason;
9308 for_each_set_bit(b, &mask, 32) {
9309 if (value & (1 << b))
9310 proto_down_reason |= BIT(b);
9311 else
9312 proto_down_reason &= ~BIT(b);
9313 }
9314 }
9315 WRITE_ONCE(dev->proto_down_reason, proto_down_reason);
9316 }
9317
9318 struct bpf_xdp_link {
9319 struct bpf_link link;
9320 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9321 int flags;
9322 };
9323
9324 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9325 {
9326 if (flags & XDP_FLAGS_HW_MODE)
9327 return XDP_MODE_HW;
9328 if (flags & XDP_FLAGS_DRV_MODE)
9329 return XDP_MODE_DRV;
9330 if (flags & XDP_FLAGS_SKB_MODE)
9331 return XDP_MODE_SKB;
9332 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9333 }
9334
9335 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9336 {
9337 switch (mode) {
9338 case XDP_MODE_SKB:
9339 return generic_xdp_install;
9340 case XDP_MODE_DRV:
9341 case XDP_MODE_HW:
9342 return dev->netdev_ops->ndo_bpf;
9343 default:
9344 return NULL;
9345 }
9346 }
9347
9348 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9349 enum bpf_xdp_mode mode)
9350 {
9351 return dev->xdp_state[mode].link;
9352 }
9353
9354 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9355 enum bpf_xdp_mode mode)
9356 {
9357 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9358
9359 if (link)
9360 return link->link.prog;
9361 return dev->xdp_state[mode].prog;
9362 }
9363
9364 u8 dev_xdp_prog_count(struct net_device *dev)
9365 {
9366 u8 count = 0;
9367 int i;
9368
9369 for (i = 0; i < __MAX_XDP_MODE; i++)
9370 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9371 count++;
9372 return count;
9373 }
9374 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9375
9376 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9377 {
9378 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9379
9380 return prog ? prog->aux->id : 0;
9381 }
9382
9383 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9384 struct bpf_xdp_link *link)
9385 {
9386 dev->xdp_state[mode].link = link;
9387 dev->xdp_state[mode].prog = NULL;
9388 }
9389
9390 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9391 struct bpf_prog *prog)
9392 {
9393 dev->xdp_state[mode].link = NULL;
9394 dev->xdp_state[mode].prog = prog;
9395 }
9396
9397 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9398 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9399 u32 flags, struct bpf_prog *prog)
9400 {
9401 struct netdev_bpf xdp;
9402 int err;
9403
9404 memset(&xdp, 0, sizeof(xdp));
9405 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9406 xdp.extack = extack;
9407 xdp.flags = flags;
9408 xdp.prog = prog;
9409
9410 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9411 * "moved" into driver), so they don't increment it on their own, but
9412 * they do decrement refcnt when program is detached or replaced.
9413 * Given net_device also owns link/prog, we need to bump refcnt here
9414 * to prevent drivers from underflowing it.
9415 */
9416 if (prog)
9417 bpf_prog_inc(prog);
9418 err = bpf_op(dev, &xdp);
9419 if (err) {
9420 if (prog)
9421 bpf_prog_put(prog);
9422 return err;
9423 }
9424
9425 if (mode != XDP_MODE_HW)
9426 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9427
9428 return 0;
9429 }
9430
9431 static void dev_xdp_uninstall(struct net_device *dev)
9432 {
9433 struct bpf_xdp_link *link;
9434 struct bpf_prog *prog;
9435 enum bpf_xdp_mode mode;
9436 bpf_op_t bpf_op;
9437
9438 ASSERT_RTNL();
9439
9440 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9441 prog = dev_xdp_prog(dev, mode);
9442 if (!prog)
9443 continue;
9444
9445 bpf_op = dev_xdp_bpf_op(dev, mode);
9446 if (!bpf_op)
9447 continue;
9448
9449 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9450
9451 /* auto-detach link from net device */
9452 link = dev_xdp_link(dev, mode);
9453 if (link)
9454 link->dev = NULL;
9455 else
9456 bpf_prog_put(prog);
9457
9458 dev_xdp_set_link(dev, mode, NULL);
9459 }
9460 }
9461
9462 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9463 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9464 struct bpf_prog *old_prog, u32 flags)
9465 {
9466 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9467 struct bpf_prog *cur_prog;
9468 struct net_device *upper;
9469 struct list_head *iter;
9470 enum bpf_xdp_mode mode;
9471 bpf_op_t bpf_op;
9472 int err;
9473
9474 ASSERT_RTNL();
9475
9476 /* either link or prog attachment, never both */
9477 if (link && (new_prog || old_prog))
9478 return -EINVAL;
9479 /* link supports only XDP mode flags */
9480 if (link && (flags & ~XDP_FLAGS_MODES)) {
9481 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9482 return -EINVAL;
9483 }
9484 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9485 if (num_modes > 1) {
9486 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9487 return -EINVAL;
9488 }
9489 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9490 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9491 NL_SET_ERR_MSG(extack,
9492 "More than one program loaded, unset mode is ambiguous");
9493 return -EINVAL;
9494 }
9495 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9496 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9497 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9498 return -EINVAL;
9499 }
9500
9501 mode = dev_xdp_mode(dev, flags);
9502 /* can't replace attached link */
9503 if (dev_xdp_link(dev, mode)) {
9504 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9505 return -EBUSY;
9506 }
9507
9508 /* don't allow if an upper device already has a program */
9509 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9510 if (dev_xdp_prog_count(upper) > 0) {
9511 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9512 return -EEXIST;
9513 }
9514 }
9515
9516 cur_prog = dev_xdp_prog(dev, mode);
9517 /* can't replace attached prog with link */
9518 if (link && cur_prog) {
9519 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9520 return -EBUSY;
9521 }
9522 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9523 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9524 return -EEXIST;
9525 }
9526
9527 /* put effective new program into new_prog */
9528 if (link)
9529 new_prog = link->link.prog;
9530
9531 if (new_prog) {
9532 bool offload = mode == XDP_MODE_HW;
9533 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9534 ? XDP_MODE_DRV : XDP_MODE_SKB;
9535
9536 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9537 NL_SET_ERR_MSG(extack, "XDP program already attached");
9538 return -EBUSY;
9539 }
9540 if (!offload && dev_xdp_prog(dev, other_mode)) {
9541 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9542 return -EEXIST;
9543 }
9544 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9545 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9546 return -EINVAL;
9547 }
9548 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9549 NL_SET_ERR_MSG(extack, "Program bound to different device");
9550 return -EINVAL;
9551 }
9552 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9553 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9554 return -EINVAL;
9555 }
9556 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9557 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9558 return -EINVAL;
9559 }
9560 }
9561
9562 /* don't call drivers if the effective program didn't change */
9563 if (new_prog != cur_prog) {
9564 bpf_op = dev_xdp_bpf_op(dev, mode);
9565 if (!bpf_op) {
9566 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9567 return -EOPNOTSUPP;
9568 }
9569
9570 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9571 if (err)
9572 return err;
9573 }
9574
9575 if (link)
9576 dev_xdp_set_link(dev, mode, link);
9577 else
9578 dev_xdp_set_prog(dev, mode, new_prog);
9579 if (cur_prog)
9580 bpf_prog_put(cur_prog);
9581
9582 return 0;
9583 }
9584
9585 static int dev_xdp_attach_link(struct net_device *dev,
9586 struct netlink_ext_ack *extack,
9587 struct bpf_xdp_link *link)
9588 {
9589 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9590 }
9591
9592 static int dev_xdp_detach_link(struct net_device *dev,
9593 struct netlink_ext_ack *extack,
9594 struct bpf_xdp_link *link)
9595 {
9596 enum bpf_xdp_mode mode;
9597 bpf_op_t bpf_op;
9598
9599 ASSERT_RTNL();
9600
9601 mode = dev_xdp_mode(dev, link->flags);
9602 if (dev_xdp_link(dev, mode) != link)
9603 return -EINVAL;
9604
9605 bpf_op = dev_xdp_bpf_op(dev, mode);
9606 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9607 dev_xdp_set_link(dev, mode, NULL);
9608 return 0;
9609 }
9610
9611 static void bpf_xdp_link_release(struct bpf_link *link)
9612 {
9613 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9614
9615 rtnl_lock();
9616
9617 /* if racing with net_device's tear down, xdp_link->dev might be
9618 * already NULL, in which case link was already auto-detached
9619 */
9620 if (xdp_link->dev) {
9621 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9622 xdp_link->dev = NULL;
9623 }
9624
9625 rtnl_unlock();
9626 }
9627
9628 static int bpf_xdp_link_detach(struct bpf_link *link)
9629 {
9630 bpf_xdp_link_release(link);
9631 return 0;
9632 }
9633
9634 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9635 {
9636 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9637
9638 kfree(xdp_link);
9639 }
9640
9641 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9642 struct seq_file *seq)
9643 {
9644 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9645 u32 ifindex = 0;
9646
9647 rtnl_lock();
9648 if (xdp_link->dev)
9649 ifindex = xdp_link->dev->ifindex;
9650 rtnl_unlock();
9651
9652 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9653 }
9654
9655 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9656 struct bpf_link_info *info)
9657 {
9658 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9659 u32 ifindex = 0;
9660
9661 rtnl_lock();
9662 if (xdp_link->dev)
9663 ifindex = xdp_link->dev->ifindex;
9664 rtnl_unlock();
9665
9666 info->xdp.ifindex = ifindex;
9667 return 0;
9668 }
9669
9670 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9671 struct bpf_prog *old_prog)
9672 {
9673 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9674 enum bpf_xdp_mode mode;
9675 bpf_op_t bpf_op;
9676 int err = 0;
9677
9678 rtnl_lock();
9679
9680 /* link might have been auto-released already, so fail */
9681 if (!xdp_link->dev) {
9682 err = -ENOLINK;
9683 goto out_unlock;
9684 }
9685
9686 if (old_prog && link->prog != old_prog) {
9687 err = -EPERM;
9688 goto out_unlock;
9689 }
9690 old_prog = link->prog;
9691 if (old_prog->type != new_prog->type ||
9692 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9693 err = -EINVAL;
9694 goto out_unlock;
9695 }
9696
9697 if (old_prog == new_prog) {
9698 /* no-op, don't disturb drivers */
9699 bpf_prog_put(new_prog);
9700 goto out_unlock;
9701 }
9702
9703 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9704 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9705 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9706 xdp_link->flags, new_prog);
9707 if (err)
9708 goto out_unlock;
9709
9710 old_prog = xchg(&link->prog, new_prog);
9711 bpf_prog_put(old_prog);
9712
9713 out_unlock:
9714 rtnl_unlock();
9715 return err;
9716 }
9717
9718 static const struct bpf_link_ops bpf_xdp_link_lops = {
9719 .release = bpf_xdp_link_release,
9720 .dealloc = bpf_xdp_link_dealloc,
9721 .detach = bpf_xdp_link_detach,
9722 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9723 .fill_link_info = bpf_xdp_link_fill_link_info,
9724 .update_prog = bpf_xdp_link_update,
9725 };
9726
9727 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9728 {
9729 struct net *net = current->nsproxy->net_ns;
9730 struct bpf_link_primer link_primer;
9731 struct netlink_ext_ack extack = {};
9732 struct bpf_xdp_link *link;
9733 struct net_device *dev;
9734 int err, fd;
9735
9736 rtnl_lock();
9737 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9738 if (!dev) {
9739 rtnl_unlock();
9740 return -EINVAL;
9741 }
9742
9743 link = kzalloc(sizeof(*link), GFP_USER);
9744 if (!link) {
9745 err = -ENOMEM;
9746 goto unlock;
9747 }
9748
9749 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9750 link->dev = dev;
9751 link->flags = attr->link_create.flags;
9752
9753 err = bpf_link_prime(&link->link, &link_primer);
9754 if (err) {
9755 kfree(link);
9756 goto unlock;
9757 }
9758
9759 err = dev_xdp_attach_link(dev, &extack, link);
9760 rtnl_unlock();
9761
9762 if (err) {
9763 link->dev = NULL;
9764 bpf_link_cleanup(&link_primer);
9765 trace_bpf_xdp_link_attach_failed(extack._msg);
9766 goto out_put_dev;
9767 }
9768
9769 fd = bpf_link_settle(&link_primer);
9770 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9771 dev_put(dev);
9772 return fd;
9773
9774 unlock:
9775 rtnl_unlock();
9776
9777 out_put_dev:
9778 dev_put(dev);
9779 return err;
9780 }
9781
9782 /**
9783 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9784 * @dev: device
9785 * @extack: netlink extended ack
9786 * @fd: new program fd or negative value to clear
9787 * @expected_fd: old program fd that userspace expects to replace or clear
9788 * @flags: xdp-related flags
9789 *
9790 * Set or clear a bpf program for a device
9791 */
9792 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9793 int fd, int expected_fd, u32 flags)
9794 {
9795 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9796 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9797 int err;
9798
9799 ASSERT_RTNL();
9800
9801 if (fd >= 0) {
9802 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9803 mode != XDP_MODE_SKB);
9804 if (IS_ERR(new_prog))
9805 return PTR_ERR(new_prog);
9806 }
9807
9808 if (expected_fd >= 0) {
9809 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9810 mode != XDP_MODE_SKB);
9811 if (IS_ERR(old_prog)) {
9812 err = PTR_ERR(old_prog);
9813 old_prog = NULL;
9814 goto err_out;
9815 }
9816 }
9817
9818 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9819
9820 err_out:
9821 if (err && new_prog)
9822 bpf_prog_put(new_prog);
9823 if (old_prog)
9824 bpf_prog_put(old_prog);
9825 return err;
9826 }
9827
9828 /**
9829 * dev_index_reserve() - allocate an ifindex in a namespace
9830 * @net: the applicable net namespace
9831 * @ifindex: requested ifindex, pass %0 to get one allocated
9832 *
9833 * Allocate a ifindex for a new device. Caller must either use the ifindex
9834 * to store the device (via list_netdevice()) or call dev_index_release()
9835 * to give the index up.
9836 *
9837 * Return: a suitable unique value for a new device interface number or -errno.
9838 */
9839 static int dev_index_reserve(struct net *net, u32 ifindex)
9840 {
9841 int err;
9842
9843 if (ifindex > INT_MAX) {
9844 DEBUG_NET_WARN_ON_ONCE(1);
9845 return -EINVAL;
9846 }
9847
9848 if (!ifindex)
9849 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL,
9850 xa_limit_31b, &net->ifindex, GFP_KERNEL);
9851 else
9852 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL);
9853 if (err < 0)
9854 return err;
9855
9856 return ifindex;
9857 }
9858
9859 static void dev_index_release(struct net *net, int ifindex)
9860 {
9861 /* Expect only unused indexes, unlist_netdevice() removes the used */
9862 WARN_ON(xa_erase(&net->dev_by_index, ifindex));
9863 }
9864
9865 /* Delayed registration/unregisteration */
9866 LIST_HEAD(net_todo_list);
9867 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9868 atomic_t dev_unreg_count = ATOMIC_INIT(0);
9869
9870 static void net_set_todo(struct net_device *dev)
9871 {
9872 list_add_tail(&dev->todo_list, &net_todo_list);
9873 }
9874
9875 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9876 struct net_device *upper, netdev_features_t features)
9877 {
9878 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9879 netdev_features_t feature;
9880 int feature_bit;
9881
9882 for_each_netdev_feature(upper_disables, feature_bit) {
9883 feature = __NETIF_F_BIT(feature_bit);
9884 if (!(upper->wanted_features & feature)
9885 && (features & feature)) {
9886 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9887 &feature, upper->name);
9888 features &= ~feature;
9889 }
9890 }
9891
9892 return features;
9893 }
9894
9895 static void netdev_sync_lower_features(struct net_device *upper,
9896 struct net_device *lower, netdev_features_t features)
9897 {
9898 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9899 netdev_features_t feature;
9900 int feature_bit;
9901
9902 for_each_netdev_feature(upper_disables, feature_bit) {
9903 feature = __NETIF_F_BIT(feature_bit);
9904 if (!(features & feature) && (lower->features & feature)) {
9905 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9906 &feature, lower->name);
9907 lower->wanted_features &= ~feature;
9908 __netdev_update_features(lower);
9909
9910 if (unlikely(lower->features & feature))
9911 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9912 &feature, lower->name);
9913 else
9914 netdev_features_change(lower);
9915 }
9916 }
9917 }
9918
9919 static bool netdev_has_ip_or_hw_csum(netdev_features_t features)
9920 {
9921 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
9922 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask;
9923 bool hw_csum = features & NETIF_F_HW_CSUM;
9924
9925 return ip_csum || hw_csum;
9926 }
9927
9928 static netdev_features_t netdev_fix_features(struct net_device *dev,
9929 netdev_features_t features)
9930 {
9931 /* Fix illegal checksum combinations */
9932 if ((features & NETIF_F_HW_CSUM) &&
9933 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9934 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9935 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9936 }
9937
9938 /* TSO requires that SG is present as well. */
9939 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9940 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9941 features &= ~NETIF_F_ALL_TSO;
9942 }
9943
9944 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9945 !(features & NETIF_F_IP_CSUM)) {
9946 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9947 features &= ~NETIF_F_TSO;
9948 features &= ~NETIF_F_TSO_ECN;
9949 }
9950
9951 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9952 !(features & NETIF_F_IPV6_CSUM)) {
9953 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9954 features &= ~NETIF_F_TSO6;
9955 }
9956
9957 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9958 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9959 features &= ~NETIF_F_TSO_MANGLEID;
9960
9961 /* TSO ECN requires that TSO is present as well. */
9962 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9963 features &= ~NETIF_F_TSO_ECN;
9964
9965 /* Software GSO depends on SG. */
9966 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9967 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9968 features &= ~NETIF_F_GSO;
9969 }
9970
9971 /* GSO partial features require GSO partial be set */
9972 if ((features & dev->gso_partial_features) &&
9973 !(features & NETIF_F_GSO_PARTIAL)) {
9974 netdev_dbg(dev,
9975 "Dropping partially supported GSO features since no GSO partial.\n");
9976 features &= ~dev->gso_partial_features;
9977 }
9978
9979 if (!(features & NETIF_F_RXCSUM)) {
9980 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9981 * successfully merged by hardware must also have the
9982 * checksum verified by hardware. If the user does not
9983 * want to enable RXCSUM, logically, we should disable GRO_HW.
9984 */
9985 if (features & NETIF_F_GRO_HW) {
9986 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9987 features &= ~NETIF_F_GRO_HW;
9988 }
9989 }
9990
9991 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9992 if (features & NETIF_F_RXFCS) {
9993 if (features & NETIF_F_LRO) {
9994 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9995 features &= ~NETIF_F_LRO;
9996 }
9997
9998 if (features & NETIF_F_GRO_HW) {
9999 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
10000 features &= ~NETIF_F_GRO_HW;
10001 }
10002 }
10003
10004 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
10005 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
10006 features &= ~NETIF_F_LRO;
10007 }
10008
10009 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) {
10010 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
10011 features &= ~NETIF_F_HW_TLS_TX;
10012 }
10013
10014 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
10015 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
10016 features &= ~NETIF_F_HW_TLS_RX;
10017 }
10018
10019 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) {
10020 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n");
10021 features &= ~NETIF_F_GSO_UDP_L4;
10022 }
10023
10024 return features;
10025 }
10026
10027 int __netdev_update_features(struct net_device *dev)
10028 {
10029 struct net_device *upper, *lower;
10030 netdev_features_t features;
10031 struct list_head *iter;
10032 int err = -1;
10033
10034 ASSERT_RTNL();
10035
10036 features = netdev_get_wanted_features(dev);
10037
10038 if (dev->netdev_ops->ndo_fix_features)
10039 features = dev->netdev_ops->ndo_fix_features(dev, features);
10040
10041 /* driver might be less strict about feature dependencies */
10042 features = netdev_fix_features(dev, features);
10043
10044 /* some features can't be enabled if they're off on an upper device */
10045 netdev_for_each_upper_dev_rcu(dev, upper, iter)
10046 features = netdev_sync_upper_features(dev, upper, features);
10047
10048 if (dev->features == features)
10049 goto sync_lower;
10050
10051 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10052 &dev->features, &features);
10053
10054 if (dev->netdev_ops->ndo_set_features)
10055 err = dev->netdev_ops->ndo_set_features(dev, features);
10056 else
10057 err = 0;
10058
10059 if (unlikely(err < 0)) {
10060 netdev_err(dev,
10061 "set_features() failed (%d); wanted %pNF, left %pNF\n",
10062 err, &features, &dev->features);
10063 /* return non-0 since some features might have changed and
10064 * it's better to fire a spurious notification than miss it
10065 */
10066 return -1;
10067 }
10068
10069 sync_lower:
10070 /* some features must be disabled on lower devices when disabled
10071 * on an upper device (think: bonding master or bridge)
10072 */
10073 netdev_for_each_lower_dev(dev, lower, iter)
10074 netdev_sync_lower_features(dev, lower, features);
10075
10076 if (!err) {
10077 netdev_features_t diff = features ^ dev->features;
10078
10079 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10080 /* udp_tunnel_{get,drop}_rx_info both need
10081 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10082 * device, or they won't do anything.
10083 * Thus we need to update dev->features
10084 * *before* calling udp_tunnel_get_rx_info,
10085 * but *after* calling udp_tunnel_drop_rx_info.
10086 */
10087 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10088 dev->features = features;
10089 udp_tunnel_get_rx_info(dev);
10090 } else {
10091 udp_tunnel_drop_rx_info(dev);
10092 }
10093 }
10094
10095 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10096 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10097 dev->features = features;
10098 err |= vlan_get_rx_ctag_filter_info(dev);
10099 } else {
10100 vlan_drop_rx_ctag_filter_info(dev);
10101 }
10102 }
10103
10104 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10105 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10106 dev->features = features;
10107 err |= vlan_get_rx_stag_filter_info(dev);
10108 } else {
10109 vlan_drop_rx_stag_filter_info(dev);
10110 }
10111 }
10112
10113 dev->features = features;
10114 }
10115
10116 return err < 0 ? 0 : 1;
10117 }
10118
10119 /**
10120 * netdev_update_features - recalculate device features
10121 * @dev: the device to check
10122 *
10123 * Recalculate dev->features set and send notifications if it
10124 * has changed. Should be called after driver or hardware dependent
10125 * conditions might have changed that influence the features.
10126 */
10127 void netdev_update_features(struct net_device *dev)
10128 {
10129 if (__netdev_update_features(dev))
10130 netdev_features_change(dev);
10131 }
10132 EXPORT_SYMBOL(netdev_update_features);
10133
10134 /**
10135 * netdev_change_features - recalculate device features
10136 * @dev: the device to check
10137 *
10138 * Recalculate dev->features set and send notifications even
10139 * if they have not changed. Should be called instead of
10140 * netdev_update_features() if also dev->vlan_features might
10141 * have changed to allow the changes to be propagated to stacked
10142 * VLAN devices.
10143 */
10144 void netdev_change_features(struct net_device *dev)
10145 {
10146 __netdev_update_features(dev);
10147 netdev_features_change(dev);
10148 }
10149 EXPORT_SYMBOL(netdev_change_features);
10150
10151 /**
10152 * netif_stacked_transfer_operstate - transfer operstate
10153 * @rootdev: the root or lower level device to transfer state from
10154 * @dev: the device to transfer operstate to
10155 *
10156 * Transfer operational state from root to device. This is normally
10157 * called when a stacking relationship exists between the root
10158 * device and the device(a leaf device).
10159 */
10160 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10161 struct net_device *dev)
10162 {
10163 if (rootdev->operstate == IF_OPER_DORMANT)
10164 netif_dormant_on(dev);
10165 else
10166 netif_dormant_off(dev);
10167
10168 if (rootdev->operstate == IF_OPER_TESTING)
10169 netif_testing_on(dev);
10170 else
10171 netif_testing_off(dev);
10172
10173 if (netif_carrier_ok(rootdev))
10174 netif_carrier_on(dev);
10175 else
10176 netif_carrier_off(dev);
10177 }
10178 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10179
10180 static int netif_alloc_rx_queues(struct net_device *dev)
10181 {
10182 unsigned int i, count = dev->num_rx_queues;
10183 struct netdev_rx_queue *rx;
10184 size_t sz = count * sizeof(*rx);
10185 int err = 0;
10186
10187 BUG_ON(count < 1);
10188
10189 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10190 if (!rx)
10191 return -ENOMEM;
10192
10193 dev->_rx = rx;
10194
10195 for (i = 0; i < count; i++) {
10196 rx[i].dev = dev;
10197
10198 /* XDP RX-queue setup */
10199 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10200 if (err < 0)
10201 goto err_rxq_info;
10202 }
10203 return 0;
10204
10205 err_rxq_info:
10206 /* Rollback successful reg's and free other resources */
10207 while (i--)
10208 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10209 kvfree(dev->_rx);
10210 dev->_rx = NULL;
10211 return err;
10212 }
10213
10214 static void netif_free_rx_queues(struct net_device *dev)
10215 {
10216 unsigned int i, count = dev->num_rx_queues;
10217
10218 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10219 if (!dev->_rx)
10220 return;
10221
10222 for (i = 0; i < count; i++)
10223 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10224
10225 kvfree(dev->_rx);
10226 }
10227
10228 static void netdev_init_one_queue(struct net_device *dev,
10229 struct netdev_queue *queue, void *_unused)
10230 {
10231 /* Initialize queue lock */
10232 spin_lock_init(&queue->_xmit_lock);
10233 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10234 queue->xmit_lock_owner = -1;
10235 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10236 queue->dev = dev;
10237 #ifdef CONFIG_BQL
10238 dql_init(&queue->dql, HZ);
10239 #endif
10240 }
10241
10242 static void netif_free_tx_queues(struct net_device *dev)
10243 {
10244 kvfree(dev->_tx);
10245 }
10246
10247 static int netif_alloc_netdev_queues(struct net_device *dev)
10248 {
10249 unsigned int count = dev->num_tx_queues;
10250 struct netdev_queue *tx;
10251 size_t sz = count * sizeof(*tx);
10252
10253 if (count < 1 || count > 0xffff)
10254 return -EINVAL;
10255
10256 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10257 if (!tx)
10258 return -ENOMEM;
10259
10260 dev->_tx = tx;
10261
10262 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10263 spin_lock_init(&dev->tx_global_lock);
10264
10265 return 0;
10266 }
10267
10268 void netif_tx_stop_all_queues(struct net_device *dev)
10269 {
10270 unsigned int i;
10271
10272 for (i = 0; i < dev->num_tx_queues; i++) {
10273 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10274
10275 netif_tx_stop_queue(txq);
10276 }
10277 }
10278 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10279
10280 static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
10281 {
10282 void __percpu *v;
10283
10284 /* Drivers implementing ndo_get_peer_dev must support tstat
10285 * accounting, so that skb_do_redirect() can bump the dev's
10286 * RX stats upon network namespace switch.
10287 */
10288 if (dev->netdev_ops->ndo_get_peer_dev &&
10289 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
10290 return -EOPNOTSUPP;
10291
10292 switch (dev->pcpu_stat_type) {
10293 case NETDEV_PCPU_STAT_NONE:
10294 return 0;
10295 case NETDEV_PCPU_STAT_LSTATS:
10296 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
10297 break;
10298 case NETDEV_PCPU_STAT_TSTATS:
10299 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
10300 break;
10301 case NETDEV_PCPU_STAT_DSTATS:
10302 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
10303 break;
10304 default:
10305 return -EINVAL;
10306 }
10307
10308 return v ? 0 : -ENOMEM;
10309 }
10310
10311 static void netdev_do_free_pcpu_stats(struct net_device *dev)
10312 {
10313 switch (dev->pcpu_stat_type) {
10314 case NETDEV_PCPU_STAT_NONE:
10315 return;
10316 case NETDEV_PCPU_STAT_LSTATS:
10317 free_percpu(dev->lstats);
10318 break;
10319 case NETDEV_PCPU_STAT_TSTATS:
10320 free_percpu(dev->tstats);
10321 break;
10322 case NETDEV_PCPU_STAT_DSTATS:
10323 free_percpu(dev->dstats);
10324 break;
10325 }
10326 }
10327
10328 /**
10329 * register_netdevice() - register a network device
10330 * @dev: device to register
10331 *
10332 * Take a prepared network device structure and make it externally accessible.
10333 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
10334 * Callers must hold the rtnl lock - you may want register_netdev()
10335 * instead of this.
10336 */
10337 int register_netdevice(struct net_device *dev)
10338 {
10339 int ret;
10340 struct net *net = dev_net(dev);
10341
10342 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10343 NETDEV_FEATURE_COUNT);
10344 BUG_ON(dev_boot_phase);
10345 ASSERT_RTNL();
10346
10347 might_sleep();
10348
10349 /* When net_device's are persistent, this will be fatal. */
10350 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10351 BUG_ON(!net);
10352
10353 ret = ethtool_check_ops(dev->ethtool_ops);
10354 if (ret)
10355 return ret;
10356
10357 /* rss ctx ID 0 is reserved for the default context, start from 1 */
10358 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1);
10359 mutex_init(&dev->ethtool->rss_lock);
10360
10361 spin_lock_init(&dev->addr_list_lock);
10362 netdev_set_addr_lockdep_class(dev);
10363
10364 ret = dev_get_valid_name(net, dev, dev->name);
10365 if (ret < 0)
10366 goto out;
10367
10368 ret = -ENOMEM;
10369 dev->name_node = netdev_name_node_head_alloc(dev);
10370 if (!dev->name_node)
10371 goto out;
10372
10373 /* Init, if this function is available */
10374 if (dev->netdev_ops->ndo_init) {
10375 ret = dev->netdev_ops->ndo_init(dev);
10376 if (ret) {
10377 if (ret > 0)
10378 ret = -EIO;
10379 goto err_free_name;
10380 }
10381 }
10382
10383 if (((dev->hw_features | dev->features) &
10384 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10385 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10386 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10387 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10388 ret = -EINVAL;
10389 goto err_uninit;
10390 }
10391
10392 ret = netdev_do_alloc_pcpu_stats(dev);
10393 if (ret)
10394 goto err_uninit;
10395
10396 ret = dev_index_reserve(net, dev->ifindex);
10397 if (ret < 0)
10398 goto err_free_pcpu;
10399 dev->ifindex = ret;
10400
10401 /* Transfer changeable features to wanted_features and enable
10402 * software offloads (GSO and GRO).
10403 */
10404 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10405 dev->features |= NETIF_F_SOFT_FEATURES;
10406
10407 if (dev->udp_tunnel_nic_info) {
10408 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10409 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10410 }
10411
10412 dev->wanted_features = dev->features & dev->hw_features;
10413
10414 if (!(dev->flags & IFF_LOOPBACK))
10415 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10416
10417 /* If IPv4 TCP segmentation offload is supported we should also
10418 * allow the device to enable segmenting the frame with the option
10419 * of ignoring a static IP ID value. This doesn't enable the
10420 * feature itself but allows the user to enable it later.
10421 */
10422 if (dev->hw_features & NETIF_F_TSO)
10423 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10424 if (dev->vlan_features & NETIF_F_TSO)
10425 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10426 if (dev->mpls_features & NETIF_F_TSO)
10427 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10428 if (dev->hw_enc_features & NETIF_F_TSO)
10429 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10430
10431 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10432 */
10433 dev->vlan_features |= NETIF_F_HIGHDMA;
10434
10435 /* Make NETIF_F_SG inheritable to tunnel devices.
10436 */
10437 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10438
10439 /* Make NETIF_F_SG inheritable to MPLS.
10440 */
10441 dev->mpls_features |= NETIF_F_SG;
10442
10443 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10444 ret = notifier_to_errno(ret);
10445 if (ret)
10446 goto err_ifindex_release;
10447
10448 ret = netdev_register_kobject(dev);
10449
10450 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED);
10451
10452 if (ret)
10453 goto err_uninit_notify;
10454
10455 __netdev_update_features(dev);
10456
10457 /*
10458 * Default initial state at registry is that the
10459 * device is present.
10460 */
10461
10462 set_bit(__LINK_STATE_PRESENT, &dev->state);
10463
10464 linkwatch_init_dev(dev);
10465
10466 dev_init_scheduler(dev);
10467
10468 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10469 list_netdevice(dev);
10470
10471 add_device_randomness(dev->dev_addr, dev->addr_len);
10472
10473 /* If the device has permanent device address, driver should
10474 * set dev_addr and also addr_assign_type should be set to
10475 * NET_ADDR_PERM (default value).
10476 */
10477 if (dev->addr_assign_type == NET_ADDR_PERM)
10478 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10479
10480 /* Notify protocols, that a new device appeared. */
10481 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10482 ret = notifier_to_errno(ret);
10483 if (ret) {
10484 /* Expect explicit free_netdev() on failure */
10485 dev->needs_free_netdev = false;
10486 unregister_netdevice_queue(dev, NULL);
10487 goto out;
10488 }
10489 /*
10490 * Prevent userspace races by waiting until the network
10491 * device is fully setup before sending notifications.
10492 */
10493 if (!dev->rtnl_link_ops ||
10494 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10495 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10496
10497 out:
10498 return ret;
10499
10500 err_uninit_notify:
10501 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10502 err_ifindex_release:
10503 dev_index_release(net, dev->ifindex);
10504 err_free_pcpu:
10505 netdev_do_free_pcpu_stats(dev);
10506 err_uninit:
10507 if (dev->netdev_ops->ndo_uninit)
10508 dev->netdev_ops->ndo_uninit(dev);
10509 if (dev->priv_destructor)
10510 dev->priv_destructor(dev);
10511 err_free_name:
10512 netdev_name_node_free(dev->name_node);
10513 goto out;
10514 }
10515 EXPORT_SYMBOL(register_netdevice);
10516
10517 /* Initialize the core of a dummy net device.
10518 * This is useful if you are calling this function after alloc_netdev(),
10519 * since it does not memset the net_device fields.
10520 */
10521 static void init_dummy_netdev_core(struct net_device *dev)
10522 {
10523 /* make sure we BUG if trying to hit standard
10524 * register/unregister code path
10525 */
10526 dev->reg_state = NETREG_DUMMY;
10527
10528 /* NAPI wants this */
10529 INIT_LIST_HEAD(&dev->napi_list);
10530
10531 /* a dummy interface is started by default */
10532 set_bit(__LINK_STATE_PRESENT, &dev->state);
10533 set_bit(__LINK_STATE_START, &dev->state);
10534
10535 /* napi_busy_loop stats accounting wants this */
10536 dev_net_set(dev, &init_net);
10537
10538 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10539 * because users of this 'device' dont need to change
10540 * its refcount.
10541 */
10542 }
10543
10544 /**
10545 * init_dummy_netdev - init a dummy network device for NAPI
10546 * @dev: device to init
10547 *
10548 * This takes a network device structure and initializes the minimum
10549 * amount of fields so it can be used to schedule NAPI polls without
10550 * registering a full blown interface. This is to be used by drivers
10551 * that need to tie several hardware interfaces to a single NAPI
10552 * poll scheduler due to HW limitations.
10553 */
10554 void init_dummy_netdev(struct net_device *dev)
10555 {
10556 /* Clear everything. Note we don't initialize spinlocks
10557 * as they aren't supposed to be taken by any of the
10558 * NAPI code and this dummy netdev is supposed to be
10559 * only ever used for NAPI polls
10560 */
10561 memset(dev, 0, sizeof(struct net_device));
10562 init_dummy_netdev_core(dev);
10563 }
10564 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10565
10566 /**
10567 * register_netdev - register a network device
10568 * @dev: device to register
10569 *
10570 * Take a completed network device structure and add it to the kernel
10571 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10572 * chain. 0 is returned on success. A negative errno code is returned
10573 * on a failure to set up the device, or if the name is a duplicate.
10574 *
10575 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10576 * and expands the device name if you passed a format string to
10577 * alloc_netdev.
10578 */
10579 int register_netdev(struct net_device *dev)
10580 {
10581 int err;
10582
10583 if (rtnl_lock_killable())
10584 return -EINTR;
10585 err = register_netdevice(dev);
10586 rtnl_unlock();
10587 return err;
10588 }
10589 EXPORT_SYMBOL(register_netdev);
10590
10591 int netdev_refcnt_read(const struct net_device *dev)
10592 {
10593 #ifdef CONFIG_PCPU_DEV_REFCNT
10594 int i, refcnt = 0;
10595
10596 for_each_possible_cpu(i)
10597 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10598 return refcnt;
10599 #else
10600 return refcount_read(&dev->dev_refcnt);
10601 #endif
10602 }
10603 EXPORT_SYMBOL(netdev_refcnt_read);
10604
10605 int netdev_unregister_timeout_secs __read_mostly = 10;
10606
10607 #define WAIT_REFS_MIN_MSECS 1
10608 #define WAIT_REFS_MAX_MSECS 250
10609 /**
10610 * netdev_wait_allrefs_any - wait until all references are gone.
10611 * @list: list of net_devices to wait on
10612 *
10613 * This is called when unregistering network devices.
10614 *
10615 * Any protocol or device that holds a reference should register
10616 * for netdevice notification, and cleanup and put back the
10617 * reference if they receive an UNREGISTER event.
10618 * We can get stuck here if buggy protocols don't correctly
10619 * call dev_put.
10620 */
10621 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10622 {
10623 unsigned long rebroadcast_time, warning_time;
10624 struct net_device *dev;
10625 int wait = 0;
10626
10627 rebroadcast_time = warning_time = jiffies;
10628
10629 list_for_each_entry(dev, list, todo_list)
10630 if (netdev_refcnt_read(dev) == 1)
10631 return dev;
10632
10633 while (true) {
10634 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10635 rtnl_lock();
10636
10637 /* Rebroadcast unregister notification */
10638 list_for_each_entry(dev, list, todo_list)
10639 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10640
10641 __rtnl_unlock();
10642 rcu_barrier();
10643 rtnl_lock();
10644
10645 list_for_each_entry(dev, list, todo_list)
10646 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10647 &dev->state)) {
10648 /* We must not have linkwatch events
10649 * pending on unregister. If this
10650 * happens, we simply run the queue
10651 * unscheduled, resulting in a noop
10652 * for this device.
10653 */
10654 linkwatch_run_queue();
10655 break;
10656 }
10657
10658 __rtnl_unlock();
10659
10660 rebroadcast_time = jiffies;
10661 }
10662
10663 rcu_barrier();
10664
10665 if (!wait) {
10666 wait = WAIT_REFS_MIN_MSECS;
10667 } else {
10668 msleep(wait);
10669 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10670 }
10671
10672 list_for_each_entry(dev, list, todo_list)
10673 if (netdev_refcnt_read(dev) == 1)
10674 return dev;
10675
10676 if (time_after(jiffies, warning_time +
10677 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10678 list_for_each_entry(dev, list, todo_list) {
10679 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10680 dev->name, netdev_refcnt_read(dev));
10681 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10682 }
10683
10684 warning_time = jiffies;
10685 }
10686 }
10687 }
10688
10689 /* The sequence is:
10690 *
10691 * rtnl_lock();
10692 * ...
10693 * register_netdevice(x1);
10694 * register_netdevice(x2);
10695 * ...
10696 * unregister_netdevice(y1);
10697 * unregister_netdevice(y2);
10698 * ...
10699 * rtnl_unlock();
10700 * free_netdev(y1);
10701 * free_netdev(y2);
10702 *
10703 * We are invoked by rtnl_unlock().
10704 * This allows us to deal with problems:
10705 * 1) We can delete sysfs objects which invoke hotplug
10706 * without deadlocking with linkwatch via keventd.
10707 * 2) Since we run with the RTNL semaphore not held, we can sleep
10708 * safely in order to wait for the netdev refcnt to drop to zero.
10709 *
10710 * We must not return until all unregister events added during
10711 * the interval the lock was held have been completed.
10712 */
10713 void netdev_run_todo(void)
10714 {
10715 struct net_device *dev, *tmp;
10716 struct list_head list;
10717 int cnt;
10718 #ifdef CONFIG_LOCKDEP
10719 struct list_head unlink_list;
10720
10721 list_replace_init(&net_unlink_list, &unlink_list);
10722
10723 while (!list_empty(&unlink_list)) {
10724 struct net_device *dev = list_first_entry(&unlink_list,
10725 struct net_device,
10726 unlink_list);
10727 list_del_init(&dev->unlink_list);
10728 dev->nested_level = dev->lower_level - 1;
10729 }
10730 #endif
10731
10732 /* Snapshot list, allow later requests */
10733 list_replace_init(&net_todo_list, &list);
10734
10735 __rtnl_unlock();
10736
10737 /* Wait for rcu callbacks to finish before next phase */
10738 if (!list_empty(&list))
10739 rcu_barrier();
10740
10741 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10742 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10743 netdev_WARN(dev, "run_todo but not unregistering\n");
10744 list_del(&dev->todo_list);
10745 continue;
10746 }
10747
10748 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED);
10749 linkwatch_sync_dev(dev);
10750 }
10751
10752 cnt = 0;
10753 while (!list_empty(&list)) {
10754 dev = netdev_wait_allrefs_any(&list);
10755 list_del(&dev->todo_list);
10756
10757 /* paranoia */
10758 BUG_ON(netdev_refcnt_read(dev) != 1);
10759 BUG_ON(!list_empty(&dev->ptype_all));
10760 BUG_ON(!list_empty(&dev->ptype_specific));
10761 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10762 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10763
10764 netdev_do_free_pcpu_stats(dev);
10765 if (dev->priv_destructor)
10766 dev->priv_destructor(dev);
10767 if (dev->needs_free_netdev)
10768 free_netdev(dev);
10769
10770 cnt++;
10771
10772 /* Free network device */
10773 kobject_put(&dev->dev.kobj);
10774 }
10775 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count))
10776 wake_up(&netdev_unregistering_wq);
10777 }
10778
10779 /* Collate per-cpu network dstats statistics
10780 *
10781 * Read per-cpu network statistics from dev->dstats and populate the related
10782 * fields in @s.
10783 */
10784 static void dev_fetch_dstats(struct rtnl_link_stats64 *s,
10785 const struct pcpu_dstats __percpu *dstats)
10786 {
10787 int cpu;
10788
10789 for_each_possible_cpu(cpu) {
10790 u64 rx_packets, rx_bytes, rx_drops;
10791 u64 tx_packets, tx_bytes, tx_drops;
10792 const struct pcpu_dstats *stats;
10793 unsigned int start;
10794
10795 stats = per_cpu_ptr(dstats, cpu);
10796 do {
10797 start = u64_stats_fetch_begin(&stats->syncp);
10798 rx_packets = u64_stats_read(&stats->rx_packets);
10799 rx_bytes = u64_stats_read(&stats->rx_bytes);
10800 rx_drops = u64_stats_read(&stats->rx_drops);
10801 tx_packets = u64_stats_read(&stats->tx_packets);
10802 tx_bytes = u64_stats_read(&stats->tx_bytes);
10803 tx_drops = u64_stats_read(&stats->tx_drops);
10804 } while (u64_stats_fetch_retry(&stats->syncp, start));
10805
10806 s->rx_packets += rx_packets;
10807 s->rx_bytes += rx_bytes;
10808 s->rx_dropped += rx_drops;
10809 s->tx_packets += tx_packets;
10810 s->tx_bytes += tx_bytes;
10811 s->tx_dropped += tx_drops;
10812 }
10813 }
10814
10815 /* ndo_get_stats64 implementation for dtstats-based accounting.
10816 *
10817 * Populate @s from dev->stats and dev->dstats. This is used internally by the
10818 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection.
10819 */
10820 static void dev_get_dstats64(const struct net_device *dev,
10821 struct rtnl_link_stats64 *s)
10822 {
10823 netdev_stats_to_stats64(s, &dev->stats);
10824 dev_fetch_dstats(s, dev->dstats);
10825 }
10826
10827 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10828 * all the same fields in the same order as net_device_stats, with only
10829 * the type differing, but rtnl_link_stats64 may have additional fields
10830 * at the end for newer counters.
10831 */
10832 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10833 const struct net_device_stats *netdev_stats)
10834 {
10835 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10836 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10837 u64 *dst = (u64 *)stats64;
10838
10839 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10840 for (i = 0; i < n; i++)
10841 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10842 /* zero out counters that only exist in rtnl_link_stats64 */
10843 memset((char *)stats64 + n * sizeof(u64), 0,
10844 sizeof(*stats64) - n * sizeof(u64));
10845 }
10846 EXPORT_SYMBOL(netdev_stats_to_stats64);
10847
10848 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc(
10849 struct net_device *dev)
10850 {
10851 struct net_device_core_stats __percpu *p;
10852
10853 p = alloc_percpu_gfp(struct net_device_core_stats,
10854 GFP_ATOMIC | __GFP_NOWARN);
10855
10856 if (p && cmpxchg(&dev->core_stats, NULL, p))
10857 free_percpu(p);
10858
10859 /* This READ_ONCE() pairs with the cmpxchg() above */
10860 return READ_ONCE(dev->core_stats);
10861 }
10862
10863 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset)
10864 {
10865 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10866 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
10867 unsigned long __percpu *field;
10868
10869 if (unlikely(!p)) {
10870 p = netdev_core_stats_alloc(dev);
10871 if (!p)
10872 return;
10873 }
10874
10875 field = (__force unsigned long __percpu *)((__force void *)p + offset);
10876 this_cpu_inc(*field);
10877 }
10878 EXPORT_SYMBOL_GPL(netdev_core_stats_inc);
10879
10880 /**
10881 * dev_get_stats - get network device statistics
10882 * @dev: device to get statistics from
10883 * @storage: place to store stats
10884 *
10885 * Get network statistics from device. Return @storage.
10886 * The device driver may provide its own method by setting
10887 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10888 * otherwise the internal statistics structure is used.
10889 */
10890 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10891 struct rtnl_link_stats64 *storage)
10892 {
10893 const struct net_device_ops *ops = dev->netdev_ops;
10894 const struct net_device_core_stats __percpu *p;
10895
10896 if (ops->ndo_get_stats64) {
10897 memset(storage, 0, sizeof(*storage));
10898 ops->ndo_get_stats64(dev, storage);
10899 } else if (ops->ndo_get_stats) {
10900 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10901 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) {
10902 dev_get_tstats64(dev, storage);
10903 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) {
10904 dev_get_dstats64(dev, storage);
10905 } else {
10906 netdev_stats_to_stats64(storage, &dev->stats);
10907 }
10908
10909 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10910 p = READ_ONCE(dev->core_stats);
10911 if (p) {
10912 const struct net_device_core_stats *core_stats;
10913 int i;
10914
10915 for_each_possible_cpu(i) {
10916 core_stats = per_cpu_ptr(p, i);
10917 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10918 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10919 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10920 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10921 }
10922 }
10923 return storage;
10924 }
10925 EXPORT_SYMBOL(dev_get_stats);
10926
10927 /**
10928 * dev_fetch_sw_netstats - get per-cpu network device statistics
10929 * @s: place to store stats
10930 * @netstats: per-cpu network stats to read from
10931 *
10932 * Read per-cpu network statistics and populate the related fields in @s.
10933 */
10934 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10935 const struct pcpu_sw_netstats __percpu *netstats)
10936 {
10937 int cpu;
10938
10939 for_each_possible_cpu(cpu) {
10940 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10941 const struct pcpu_sw_netstats *stats;
10942 unsigned int start;
10943
10944 stats = per_cpu_ptr(netstats, cpu);
10945 do {
10946 start = u64_stats_fetch_begin(&stats->syncp);
10947 rx_packets = u64_stats_read(&stats->rx_packets);
10948 rx_bytes = u64_stats_read(&stats->rx_bytes);
10949 tx_packets = u64_stats_read(&stats->tx_packets);
10950 tx_bytes = u64_stats_read(&stats->tx_bytes);
10951 } while (u64_stats_fetch_retry(&stats->syncp, start));
10952
10953 s->rx_packets += rx_packets;
10954 s->rx_bytes += rx_bytes;
10955 s->tx_packets += tx_packets;
10956 s->tx_bytes += tx_bytes;
10957 }
10958 }
10959 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10960
10961 /**
10962 * dev_get_tstats64 - ndo_get_stats64 implementation
10963 * @dev: device to get statistics from
10964 * @s: place to store stats
10965 *
10966 * Populate @s from dev->stats and dev->tstats. Can be used as
10967 * ndo_get_stats64() callback.
10968 */
10969 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10970 {
10971 netdev_stats_to_stats64(s, &dev->stats);
10972 dev_fetch_sw_netstats(s, dev->tstats);
10973 }
10974 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10975
10976 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10977 {
10978 struct netdev_queue *queue = dev_ingress_queue(dev);
10979
10980 #ifdef CONFIG_NET_CLS_ACT
10981 if (queue)
10982 return queue;
10983 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10984 if (!queue)
10985 return NULL;
10986 netdev_init_one_queue(dev, queue, NULL);
10987 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10988 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10989 rcu_assign_pointer(dev->ingress_queue, queue);
10990 #endif
10991 return queue;
10992 }
10993
10994 static const struct ethtool_ops default_ethtool_ops;
10995
10996 void netdev_set_default_ethtool_ops(struct net_device *dev,
10997 const struct ethtool_ops *ops)
10998 {
10999 if (dev->ethtool_ops == &default_ethtool_ops)
11000 dev->ethtool_ops = ops;
11001 }
11002 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
11003
11004 /**
11005 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
11006 * @dev: netdev to enable the IRQ coalescing on
11007 *
11008 * Sets a conservative default for SW IRQ coalescing. Users can use
11009 * sysfs attributes to override the default values.
11010 */
11011 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
11012 {
11013 WARN_ON(dev->reg_state == NETREG_REGISTERED);
11014
11015 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
11016 dev->gro_flush_timeout = 20000;
11017 dev->napi_defer_hard_irqs = 1;
11018 }
11019 }
11020 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
11021
11022 /**
11023 * alloc_netdev_mqs - allocate network device
11024 * @sizeof_priv: size of private data to allocate space for
11025 * @name: device name format string
11026 * @name_assign_type: origin of device name
11027 * @setup: callback to initialize device
11028 * @txqs: the number of TX subqueues to allocate
11029 * @rxqs: the number of RX subqueues to allocate
11030 *
11031 * Allocates a struct net_device with private data area for driver use
11032 * and performs basic initialization. Also allocates subqueue structs
11033 * for each queue on the device.
11034 */
11035 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
11036 unsigned char name_assign_type,
11037 void (*setup)(struct net_device *),
11038 unsigned int txqs, unsigned int rxqs)
11039 {
11040 struct net_device *dev;
11041
11042 BUG_ON(strlen(name) >= sizeof(dev->name));
11043
11044 if (txqs < 1) {
11045 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
11046 return NULL;
11047 }
11048
11049 if (rxqs < 1) {
11050 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
11051 return NULL;
11052 }
11053
11054 dev = kvzalloc(struct_size(dev, priv, sizeof_priv),
11055 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11056 if (!dev)
11057 return NULL;
11058
11059 dev->priv_len = sizeof_priv;
11060
11061 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
11062 #ifdef CONFIG_PCPU_DEV_REFCNT
11063 dev->pcpu_refcnt = alloc_percpu(int);
11064 if (!dev->pcpu_refcnt)
11065 goto free_dev;
11066 __dev_hold(dev);
11067 #else
11068 refcount_set(&dev->dev_refcnt, 1);
11069 #endif
11070
11071 if (dev_addr_init(dev))
11072 goto free_pcpu;
11073
11074 dev_mc_init(dev);
11075 dev_uc_init(dev);
11076
11077 dev_net_set(dev, &init_net);
11078
11079 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
11080 dev->xdp_zc_max_segs = 1;
11081 dev->gso_max_segs = GSO_MAX_SEGS;
11082 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
11083 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
11084 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
11085 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
11086 dev->tso_max_segs = TSO_MAX_SEGS;
11087 dev->upper_level = 1;
11088 dev->lower_level = 1;
11089 #ifdef CONFIG_LOCKDEP
11090 dev->nested_level = 0;
11091 INIT_LIST_HEAD(&dev->unlink_list);
11092 #endif
11093
11094 INIT_LIST_HEAD(&dev->napi_list);
11095 INIT_LIST_HEAD(&dev->unreg_list);
11096 INIT_LIST_HEAD(&dev->close_list);
11097 INIT_LIST_HEAD(&dev->link_watch_list);
11098 INIT_LIST_HEAD(&dev->adj_list.upper);
11099 INIT_LIST_HEAD(&dev->adj_list.lower);
11100 INIT_LIST_HEAD(&dev->ptype_all);
11101 INIT_LIST_HEAD(&dev->ptype_specific);
11102 INIT_LIST_HEAD(&dev->net_notifier_list);
11103 #ifdef CONFIG_NET_SCHED
11104 hash_init(dev->qdisc_hash);
11105 #endif
11106 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
11107 setup(dev);
11108
11109 if (!dev->tx_queue_len) {
11110 dev->priv_flags |= IFF_NO_QUEUE;
11111 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
11112 }
11113
11114 dev->num_tx_queues = txqs;
11115 dev->real_num_tx_queues = txqs;
11116 if (netif_alloc_netdev_queues(dev))
11117 goto free_all;
11118
11119 dev->num_rx_queues = rxqs;
11120 dev->real_num_rx_queues = rxqs;
11121 if (netif_alloc_rx_queues(dev))
11122 goto free_all;
11123 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT);
11124 if (!dev->ethtool)
11125 goto free_all;
11126
11127 strcpy(dev->name, name);
11128 dev->name_assign_type = name_assign_type;
11129 dev->group = INIT_NETDEV_GROUP;
11130 if (!dev->ethtool_ops)
11131 dev->ethtool_ops = &default_ethtool_ops;
11132
11133 nf_hook_netdev_init(dev);
11134
11135 return dev;
11136
11137 free_all:
11138 free_netdev(dev);
11139 return NULL;
11140
11141 free_pcpu:
11142 #ifdef CONFIG_PCPU_DEV_REFCNT
11143 free_percpu(dev->pcpu_refcnt);
11144 free_dev:
11145 #endif
11146 kvfree(dev);
11147 return NULL;
11148 }
11149 EXPORT_SYMBOL(alloc_netdev_mqs);
11150
11151 /**
11152 * free_netdev - free network device
11153 * @dev: device
11154 *
11155 * This function does the last stage of destroying an allocated device
11156 * interface. The reference to the device object is released. If this
11157 * is the last reference then it will be freed.Must be called in process
11158 * context.
11159 */
11160 void free_netdev(struct net_device *dev)
11161 {
11162 struct napi_struct *p, *n;
11163
11164 might_sleep();
11165
11166 /* When called immediately after register_netdevice() failed the unwind
11167 * handling may still be dismantling the device. Handle that case by
11168 * deferring the free.
11169 */
11170 if (dev->reg_state == NETREG_UNREGISTERING) {
11171 ASSERT_RTNL();
11172 dev->needs_free_netdev = true;
11173 return;
11174 }
11175
11176 kfree(dev->ethtool);
11177 netif_free_tx_queues(dev);
11178 netif_free_rx_queues(dev);
11179
11180 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
11181
11182 /* Flush device addresses */
11183 dev_addr_flush(dev);
11184
11185 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
11186 netif_napi_del(p);
11187
11188 ref_tracker_dir_exit(&dev->refcnt_tracker);
11189 #ifdef CONFIG_PCPU_DEV_REFCNT
11190 free_percpu(dev->pcpu_refcnt);
11191 dev->pcpu_refcnt = NULL;
11192 #endif
11193 free_percpu(dev->core_stats);
11194 dev->core_stats = NULL;
11195 free_percpu(dev->xdp_bulkq);
11196 dev->xdp_bulkq = NULL;
11197
11198 /* Compatibility with error handling in drivers */
11199 if (dev->reg_state == NETREG_UNINITIALIZED ||
11200 dev->reg_state == NETREG_DUMMY) {
11201 kvfree(dev);
11202 return;
11203 }
11204
11205 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
11206 WRITE_ONCE(dev->reg_state, NETREG_RELEASED);
11207
11208 /* will free via device release */
11209 put_device(&dev->dev);
11210 }
11211 EXPORT_SYMBOL(free_netdev);
11212
11213 /**
11214 * alloc_netdev_dummy - Allocate and initialize a dummy net device.
11215 * @sizeof_priv: size of private data to allocate space for
11216 *
11217 * Return: the allocated net_device on success, NULL otherwise
11218 */
11219 struct net_device *alloc_netdev_dummy(int sizeof_priv)
11220 {
11221 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN,
11222 init_dummy_netdev_core);
11223 }
11224 EXPORT_SYMBOL_GPL(alloc_netdev_dummy);
11225
11226 /**
11227 * synchronize_net - Synchronize with packet receive processing
11228 *
11229 * Wait for packets currently being received to be done.
11230 * Does not block later packets from starting.
11231 */
11232 void synchronize_net(void)
11233 {
11234 might_sleep();
11235 if (rtnl_is_locked())
11236 synchronize_rcu_expedited();
11237 else
11238 synchronize_rcu();
11239 }
11240 EXPORT_SYMBOL(synchronize_net);
11241
11242 static void netdev_rss_contexts_free(struct net_device *dev)
11243 {
11244 struct ethtool_rxfh_context *ctx;
11245 unsigned long context;
11246
11247 mutex_lock(&dev->ethtool->rss_lock);
11248 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) {
11249 struct ethtool_rxfh_param rxfh;
11250
11251 rxfh.indir = ethtool_rxfh_context_indir(ctx);
11252 rxfh.key = ethtool_rxfh_context_key(ctx);
11253 rxfh.hfunc = ctx->hfunc;
11254 rxfh.input_xfrm = ctx->input_xfrm;
11255 rxfh.rss_context = context;
11256 rxfh.rss_delete = true;
11257
11258 xa_erase(&dev->ethtool->rss_ctx, context);
11259 if (dev->ethtool_ops->create_rxfh_context)
11260 dev->ethtool_ops->remove_rxfh_context(dev, ctx,
11261 context, NULL);
11262 else
11263 dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL);
11264 kfree(ctx);
11265 }
11266 xa_destroy(&dev->ethtool->rss_ctx);
11267 mutex_unlock(&dev->ethtool->rss_lock);
11268 }
11269
11270 /**
11271 * unregister_netdevice_queue - remove device from the kernel
11272 * @dev: device
11273 * @head: list
11274 *
11275 * This function shuts down a device interface and removes it
11276 * from the kernel tables.
11277 * If head not NULL, device is queued to be unregistered later.
11278 *
11279 * Callers must hold the rtnl semaphore. You may want
11280 * unregister_netdev() instead of this.
11281 */
11282
11283 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11284 {
11285 ASSERT_RTNL();
11286
11287 if (head) {
11288 list_move_tail(&dev->unreg_list, head);
11289 } else {
11290 LIST_HEAD(single);
11291
11292 list_add(&dev->unreg_list, &single);
11293 unregister_netdevice_many(&single);
11294 }
11295 }
11296 EXPORT_SYMBOL(unregister_netdevice_queue);
11297
11298 void unregister_netdevice_many_notify(struct list_head *head,
11299 u32 portid, const struct nlmsghdr *nlh)
11300 {
11301 struct net_device *dev, *tmp;
11302 LIST_HEAD(close_head);
11303 int cnt = 0;
11304
11305 BUG_ON(dev_boot_phase);
11306 ASSERT_RTNL();
11307
11308 if (list_empty(head))
11309 return;
11310
11311 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11312 /* Some devices call without registering
11313 * for initialization unwind. Remove those
11314 * devices and proceed with the remaining.
11315 */
11316 if (dev->reg_state == NETREG_UNINITIALIZED) {
11317 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11318 dev->name, dev);
11319
11320 WARN_ON(1);
11321 list_del(&dev->unreg_list);
11322 continue;
11323 }
11324 dev->dismantle = true;
11325 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11326 }
11327
11328 /* If device is running, close it first. */
11329 list_for_each_entry(dev, head, unreg_list)
11330 list_add_tail(&dev->close_list, &close_head);
11331 dev_close_many(&close_head, true);
11332
11333 list_for_each_entry(dev, head, unreg_list) {
11334 /* And unlink it from device chain. */
11335 unlist_netdevice(dev);
11336 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING);
11337 }
11338 flush_all_backlogs();
11339
11340 synchronize_net();
11341
11342 list_for_each_entry(dev, head, unreg_list) {
11343 struct sk_buff *skb = NULL;
11344
11345 /* Shutdown queueing discipline. */
11346 dev_shutdown(dev);
11347 dev_tcx_uninstall(dev);
11348 dev_xdp_uninstall(dev);
11349 bpf_dev_bound_netdev_unregister(dev);
11350
11351 netdev_offload_xstats_disable_all(dev);
11352
11353 /* Notify protocols, that we are about to destroy
11354 * this device. They should clean all the things.
11355 */
11356 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11357
11358 if (!dev->rtnl_link_ops ||
11359 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11360 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11361 GFP_KERNEL, NULL, 0,
11362 portid, nlh);
11363
11364 /*
11365 * Flush the unicast and multicast chains
11366 */
11367 dev_uc_flush(dev);
11368 dev_mc_flush(dev);
11369
11370 netdev_name_node_alt_flush(dev);
11371 netdev_name_node_free(dev->name_node);
11372
11373 netdev_rss_contexts_free(dev);
11374
11375 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
11376
11377 if (dev->netdev_ops->ndo_uninit)
11378 dev->netdev_ops->ndo_uninit(dev);
11379
11380 mutex_destroy(&dev->ethtool->rss_lock);
11381
11382 if (skb)
11383 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
11384
11385 /* Notifier chain MUST detach us all upper devices. */
11386 WARN_ON(netdev_has_any_upper_dev(dev));
11387 WARN_ON(netdev_has_any_lower_dev(dev));
11388
11389 /* Remove entries from kobject tree */
11390 netdev_unregister_kobject(dev);
11391 #ifdef CONFIG_XPS
11392 /* Remove XPS queueing entries */
11393 netif_reset_xps_queues_gt(dev, 0);
11394 #endif
11395 }
11396
11397 synchronize_net();
11398
11399 list_for_each_entry(dev, head, unreg_list) {
11400 netdev_put(dev, &dev->dev_registered_tracker);
11401 net_set_todo(dev);
11402 cnt++;
11403 }
11404 atomic_add(cnt, &dev_unreg_count);
11405
11406 list_del(head);
11407 }
11408
11409 /**
11410 * unregister_netdevice_many - unregister many devices
11411 * @head: list of devices
11412 *
11413 * Note: As most callers use a stack allocated list_head,
11414 * we force a list_del() to make sure stack wont be corrupted later.
11415 */
11416 void unregister_netdevice_many(struct list_head *head)
11417 {
11418 unregister_netdevice_many_notify(head, 0, NULL);
11419 }
11420 EXPORT_SYMBOL(unregister_netdevice_many);
11421
11422 /**
11423 * unregister_netdev - remove device from the kernel
11424 * @dev: device
11425 *
11426 * This function shuts down a device interface and removes it
11427 * from the kernel tables.
11428 *
11429 * This is just a wrapper for unregister_netdevice that takes
11430 * the rtnl semaphore. In general you want to use this and not
11431 * unregister_netdevice.
11432 */
11433 void unregister_netdev(struct net_device *dev)
11434 {
11435 rtnl_lock();
11436 unregister_netdevice(dev);
11437 rtnl_unlock();
11438 }
11439 EXPORT_SYMBOL(unregister_netdev);
11440
11441 /**
11442 * __dev_change_net_namespace - move device to different nethost namespace
11443 * @dev: device
11444 * @net: network namespace
11445 * @pat: If not NULL name pattern to try if the current device name
11446 * is already taken in the destination network namespace.
11447 * @new_ifindex: If not zero, specifies device index in the target
11448 * namespace.
11449 *
11450 * This function shuts down a device interface and moves it
11451 * to a new network namespace. On success 0 is returned, on
11452 * a failure a netagive errno code is returned.
11453 *
11454 * Callers must hold the rtnl semaphore.
11455 */
11456
11457 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11458 const char *pat, int new_ifindex)
11459 {
11460 struct netdev_name_node *name_node;
11461 struct net *net_old = dev_net(dev);
11462 char new_name[IFNAMSIZ] = {};
11463 int err, new_nsid;
11464
11465 ASSERT_RTNL();
11466
11467 /* Don't allow namespace local devices to be moved. */
11468 err = -EINVAL;
11469 if (dev->features & NETIF_F_NETNS_LOCAL)
11470 goto out;
11471
11472 /* Ensure the device has been registrered */
11473 if (dev->reg_state != NETREG_REGISTERED)
11474 goto out;
11475
11476 /* Get out if there is nothing todo */
11477 err = 0;
11478 if (net_eq(net_old, net))
11479 goto out;
11480
11481 /* Pick the destination device name, and ensure
11482 * we can use it in the destination network namespace.
11483 */
11484 err = -EEXIST;
11485 if (netdev_name_in_use(net, dev->name)) {
11486 /* We get here if we can't use the current device name */
11487 if (!pat)
11488 goto out;
11489 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST);
11490 if (err < 0)
11491 goto out;
11492 }
11493 /* Check that none of the altnames conflicts. */
11494 err = -EEXIST;
11495 netdev_for_each_altname(dev, name_node)
11496 if (netdev_name_in_use(net, name_node->name))
11497 goto out;
11498
11499 /* Check that new_ifindex isn't used yet. */
11500 if (new_ifindex) {
11501 err = dev_index_reserve(net, new_ifindex);
11502 if (err < 0)
11503 goto out;
11504 } else {
11505 /* If there is an ifindex conflict assign a new one */
11506 err = dev_index_reserve(net, dev->ifindex);
11507 if (err == -EBUSY)
11508 err = dev_index_reserve(net, 0);
11509 if (err < 0)
11510 goto out;
11511 new_ifindex = err;
11512 }
11513
11514 /*
11515 * And now a mini version of register_netdevice unregister_netdevice.
11516 */
11517
11518 /* If device is running close it first. */
11519 dev_close(dev);
11520
11521 /* And unlink it from device chain */
11522 unlist_netdevice(dev);
11523
11524 synchronize_net();
11525
11526 /* Shutdown queueing discipline. */
11527 dev_shutdown(dev);
11528
11529 /* Notify protocols, that we are about to destroy
11530 * this device. They should clean all the things.
11531 *
11532 * Note that dev->reg_state stays at NETREG_REGISTERED.
11533 * This is wanted because this way 8021q and macvlan know
11534 * the device is just moving and can keep their slaves up.
11535 */
11536 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11537 rcu_barrier();
11538
11539 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11540
11541 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11542 new_ifindex);
11543
11544 /*
11545 * Flush the unicast and multicast chains
11546 */
11547 dev_uc_flush(dev);
11548 dev_mc_flush(dev);
11549
11550 /* Send a netdev-removed uevent to the old namespace */
11551 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11552 netdev_adjacent_del_links(dev);
11553
11554 /* Move per-net netdevice notifiers that are following the netdevice */
11555 move_netdevice_notifiers_dev_net(dev, net);
11556
11557 /* Actually switch the network namespace */
11558 dev_net_set(dev, net);
11559 dev->ifindex = new_ifindex;
11560
11561 if (new_name[0]) {
11562 /* Rename the netdev to prepared name */
11563 write_seqlock_bh(&netdev_rename_lock);
11564 strscpy(dev->name, new_name, IFNAMSIZ);
11565 write_sequnlock_bh(&netdev_rename_lock);
11566 }
11567
11568 /* Fixup kobjects */
11569 dev_set_uevent_suppress(&dev->dev, 1);
11570 err = device_rename(&dev->dev, dev->name);
11571 dev_set_uevent_suppress(&dev->dev, 0);
11572 WARN_ON(err);
11573
11574 /* Send a netdev-add uevent to the new namespace */
11575 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11576 netdev_adjacent_add_links(dev);
11577
11578 /* Adapt owner in case owning user namespace of target network
11579 * namespace is different from the original one.
11580 */
11581 err = netdev_change_owner(dev, net_old, net);
11582 WARN_ON(err);
11583
11584 /* Add the device back in the hashes */
11585 list_netdevice(dev);
11586
11587 /* Notify protocols, that a new device appeared. */
11588 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11589
11590 /*
11591 * Prevent userspace races by waiting until the network
11592 * device is fully setup before sending notifications.
11593 */
11594 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11595
11596 synchronize_net();
11597 err = 0;
11598 out:
11599 return err;
11600 }
11601 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11602
11603 static int dev_cpu_dead(unsigned int oldcpu)
11604 {
11605 struct sk_buff **list_skb;
11606 struct sk_buff *skb;
11607 unsigned int cpu;
11608 struct softnet_data *sd, *oldsd, *remsd = NULL;
11609
11610 local_irq_disable();
11611 cpu = smp_processor_id();
11612 sd = &per_cpu(softnet_data, cpu);
11613 oldsd = &per_cpu(softnet_data, oldcpu);
11614
11615 /* Find end of our completion_queue. */
11616 list_skb = &sd->completion_queue;
11617 while (*list_skb)
11618 list_skb = &(*list_skb)->next;
11619 /* Append completion queue from offline CPU. */
11620 *list_skb = oldsd->completion_queue;
11621 oldsd->completion_queue = NULL;
11622
11623 /* Append output queue from offline CPU. */
11624 if (oldsd->output_queue) {
11625 *sd->output_queue_tailp = oldsd->output_queue;
11626 sd->output_queue_tailp = oldsd->output_queue_tailp;
11627 oldsd->output_queue = NULL;
11628 oldsd->output_queue_tailp = &oldsd->output_queue;
11629 }
11630 /* Append NAPI poll list from offline CPU, with one exception :
11631 * process_backlog() must be called by cpu owning percpu backlog.
11632 * We properly handle process_queue & input_pkt_queue later.
11633 */
11634 while (!list_empty(&oldsd->poll_list)) {
11635 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11636 struct napi_struct,
11637 poll_list);
11638
11639 list_del_init(&napi->poll_list);
11640 if (napi->poll == process_backlog)
11641 napi->state &= NAPIF_STATE_THREADED;
11642 else
11643 ____napi_schedule(sd, napi);
11644 }
11645
11646 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11647 local_irq_enable();
11648
11649 if (!use_backlog_threads()) {
11650 #ifdef CONFIG_RPS
11651 remsd = oldsd->rps_ipi_list;
11652 oldsd->rps_ipi_list = NULL;
11653 #endif
11654 /* send out pending IPI's on offline CPU */
11655 net_rps_send_ipi(remsd);
11656 }
11657
11658 /* Process offline CPU's input_pkt_queue */
11659 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11660 netif_rx(skb);
11661 rps_input_queue_head_incr(oldsd);
11662 }
11663 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11664 netif_rx(skb);
11665 rps_input_queue_head_incr(oldsd);
11666 }
11667
11668 return 0;
11669 }
11670
11671 /**
11672 * netdev_increment_features - increment feature set by one
11673 * @all: current feature set
11674 * @one: new feature set
11675 * @mask: mask feature set
11676 *
11677 * Computes a new feature set after adding a device with feature set
11678 * @one to the master device with current feature set @all. Will not
11679 * enable anything that is off in @mask. Returns the new feature set.
11680 */
11681 netdev_features_t netdev_increment_features(netdev_features_t all,
11682 netdev_features_t one, netdev_features_t mask)
11683 {
11684 if (mask & NETIF_F_HW_CSUM)
11685 mask |= NETIF_F_CSUM_MASK;
11686 mask |= NETIF_F_VLAN_CHALLENGED;
11687
11688 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11689 all &= one | ~NETIF_F_ALL_FOR_ALL;
11690
11691 /* If one device supports hw checksumming, set for all. */
11692 if (all & NETIF_F_HW_CSUM)
11693 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11694
11695 return all;
11696 }
11697 EXPORT_SYMBOL(netdev_increment_features);
11698
11699 static struct hlist_head * __net_init netdev_create_hash(void)
11700 {
11701 int i;
11702 struct hlist_head *hash;
11703
11704 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11705 if (hash != NULL)
11706 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11707 INIT_HLIST_HEAD(&hash[i]);
11708
11709 return hash;
11710 }
11711
11712 /* Initialize per network namespace state */
11713 static int __net_init netdev_init(struct net *net)
11714 {
11715 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11716 8 * sizeof_field(struct napi_struct, gro_bitmask));
11717
11718 INIT_LIST_HEAD(&net->dev_base_head);
11719
11720 net->dev_name_head = netdev_create_hash();
11721 if (net->dev_name_head == NULL)
11722 goto err_name;
11723
11724 net->dev_index_head = netdev_create_hash();
11725 if (net->dev_index_head == NULL)
11726 goto err_idx;
11727
11728 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1);
11729
11730 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11731
11732 return 0;
11733
11734 err_idx:
11735 kfree(net->dev_name_head);
11736 err_name:
11737 return -ENOMEM;
11738 }
11739
11740 /**
11741 * netdev_drivername - network driver for the device
11742 * @dev: network device
11743 *
11744 * Determine network driver for device.
11745 */
11746 const char *netdev_drivername(const struct net_device *dev)
11747 {
11748 const struct device_driver *driver;
11749 const struct device *parent;
11750 const char *empty = "";
11751
11752 parent = dev->dev.parent;
11753 if (!parent)
11754 return empty;
11755
11756 driver = parent->driver;
11757 if (driver && driver->name)
11758 return driver->name;
11759 return empty;
11760 }
11761
11762 static void __netdev_printk(const char *level, const struct net_device *dev,
11763 struct va_format *vaf)
11764 {
11765 if (dev && dev->dev.parent) {
11766 dev_printk_emit(level[1] - '',
11767 dev->dev.parent,
11768 "%s %s %s%s: %pV",
11769 dev_driver_string(dev->dev.parent),
11770 dev_name(dev->dev.parent),
11771 netdev_name(dev), netdev_reg_state(dev),
11772 vaf);
11773 } else if (dev) {
11774 printk("%s%s%s: %pV",
11775 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11776 } else {
11777 printk("%s(NULL net_device): %pV", level, vaf);
11778 }
11779 }
11780
11781 void netdev_printk(const char *level, const struct net_device *dev,
11782 const char *format, ...)
11783 {
11784 struct va_format vaf;
11785 va_list args;
11786
11787 va_start(args, format);
11788
11789 vaf.fmt = format;
11790 vaf.va = &args;
11791
11792 __netdev_printk(level, dev, &vaf);
11793
11794 va_end(args);
11795 }
11796 EXPORT_SYMBOL(netdev_printk);
11797
11798 #define define_netdev_printk_level(func, level) \
11799 void func(const struct net_device *dev, const char *fmt, ...) \
11800 { \
11801 struct va_format vaf; \
11802 va_list args; \
11803 \
11804 va_start(args, fmt); \
11805 \
11806 vaf.fmt = fmt; \
11807 vaf.va = &args; \
11808 \
11809 __netdev_printk(level, dev, &vaf); \
11810 \
11811 va_end(args); \
11812 } \
11813 EXPORT_SYMBOL(func);
11814
11815 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11816 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11817 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11818 define_netdev_printk_level(netdev_err, KERN_ERR);
11819 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11820 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11821 define_netdev_printk_level(netdev_info, KERN_INFO);
11822
11823 static void __net_exit netdev_exit(struct net *net)
11824 {
11825 kfree(net->dev_name_head);
11826 kfree(net->dev_index_head);
11827 xa_destroy(&net->dev_by_index);
11828 if (net != &init_net)
11829 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11830 }
11831
11832 static struct pernet_operations __net_initdata netdev_net_ops = {
11833 .init = netdev_init,
11834 .exit = netdev_exit,
11835 };
11836
11837 static void __net_exit default_device_exit_net(struct net *net)
11838 {
11839 struct netdev_name_node *name_node, *tmp;
11840 struct net_device *dev, *aux;
11841 /*
11842 * Push all migratable network devices back to the
11843 * initial network namespace
11844 */
11845 ASSERT_RTNL();
11846 for_each_netdev_safe(net, dev, aux) {
11847 int err;
11848 char fb_name[IFNAMSIZ];
11849
11850 /* Ignore unmoveable devices (i.e. loopback) */
11851 if (dev->features & NETIF_F_NETNS_LOCAL)
11852 continue;
11853
11854 /* Leave virtual devices for the generic cleanup */
11855 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11856 continue;
11857
11858 /* Push remaining network devices to init_net */
11859 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11860 if (netdev_name_in_use(&init_net, fb_name))
11861 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11862
11863 netdev_for_each_altname_safe(dev, name_node, tmp)
11864 if (netdev_name_in_use(&init_net, name_node->name))
11865 __netdev_name_node_alt_destroy(name_node);
11866
11867 err = dev_change_net_namespace(dev, &init_net, fb_name);
11868 if (err) {
11869 pr_emerg("%s: failed to move %s to init_net: %d\n",
11870 __func__, dev->name, err);
11871 BUG();
11872 }
11873 }
11874 }
11875
11876 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11877 {
11878 /* At exit all network devices most be removed from a network
11879 * namespace. Do this in the reverse order of registration.
11880 * Do this across as many network namespaces as possible to
11881 * improve batching efficiency.
11882 */
11883 struct net_device *dev;
11884 struct net *net;
11885 LIST_HEAD(dev_kill_list);
11886
11887 rtnl_lock();
11888 list_for_each_entry(net, net_list, exit_list) {
11889 default_device_exit_net(net);
11890 cond_resched();
11891 }
11892
11893 list_for_each_entry(net, net_list, exit_list) {
11894 for_each_netdev_reverse(net, dev) {
11895 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11896 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11897 else
11898 unregister_netdevice_queue(dev, &dev_kill_list);
11899 }
11900 }
11901 unregister_netdevice_many(&dev_kill_list);
11902 rtnl_unlock();
11903 }
11904
11905 static struct pernet_operations __net_initdata default_device_ops = {
11906 .exit_batch = default_device_exit_batch,
11907 };
11908
11909 static void __init net_dev_struct_check(void)
11910 {
11911 /* TX read-mostly hotpath */
11912 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags);
11913 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops);
11914 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops);
11915 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx);
11916 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues);
11917 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size);
11918 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size);
11919 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs);
11920 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features);
11921 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc);
11922 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu);
11923 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom);
11924 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq);
11925 #ifdef CONFIG_XPS
11926 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps);
11927 #endif
11928 #ifdef CONFIG_NETFILTER_EGRESS
11929 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress);
11930 #endif
11931 #ifdef CONFIG_NET_XGRESS
11932 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress);
11933 #endif
11934 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160);
11935
11936 /* TXRX read-mostly hotpath */
11937 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats);
11938 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state);
11939 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags);
11940 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len);
11941 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features);
11942 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr);
11943 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46);
11944
11945 /* RX read-mostly hotpath */
11946 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific);
11947 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex);
11948 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues);
11949 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx);
11950 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout);
11951 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs);
11952 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size);
11953 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size);
11954 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler);
11955 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data);
11956 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net);
11957 #ifdef CONFIG_NETPOLL
11958 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo);
11959 #endif
11960 #ifdef CONFIG_NET_XGRESS
11961 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress);
11962 #endif
11963 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104);
11964 }
11965
11966 /*
11967 * Initialize the DEV module. At boot time this walks the device list and
11968 * unhooks any devices that fail to initialise (normally hardware not
11969 * present) and leaves us with a valid list of present and active devices.
11970 *
11971 */
11972
11973 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
11974 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE)
11975
11976 static int net_page_pool_create(int cpuid)
11977 {
11978 #if IS_ENABLED(CONFIG_PAGE_POOL)
11979 struct page_pool_params page_pool_params = {
11980 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
11981 .flags = PP_FLAG_SYSTEM_POOL,
11982 .nid = cpu_to_mem(cpuid),
11983 };
11984 struct page_pool *pp_ptr;
11985
11986 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
11987 if (IS_ERR(pp_ptr))
11988 return -ENOMEM;
11989
11990 per_cpu(system_page_pool, cpuid) = pp_ptr;
11991 #endif
11992 return 0;
11993 }
11994
11995 static int backlog_napi_should_run(unsigned int cpu)
11996 {
11997 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11998 struct napi_struct *napi = &sd->backlog;
11999
12000 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
12001 }
12002
12003 static void run_backlog_napi(unsigned int cpu)
12004 {
12005 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12006
12007 napi_threaded_poll_loop(&sd->backlog);
12008 }
12009
12010 static void backlog_napi_setup(unsigned int cpu)
12011 {
12012 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12013 struct napi_struct *napi = &sd->backlog;
12014
12015 napi->thread = this_cpu_read(backlog_napi);
12016 set_bit(NAPI_STATE_THREADED, &napi->state);
12017 }
12018
12019 static struct smp_hotplug_thread backlog_threads = {
12020 .store = &backlog_napi,
12021 .thread_should_run = backlog_napi_should_run,
12022 .thread_fn = run_backlog_napi,
12023 .thread_comm = "backlog_napi/%u",
12024 .setup = backlog_napi_setup,
12025 };
12026
12027 /*
12028 * This is called single threaded during boot, so no need
12029 * to take the rtnl semaphore.
12030 */
12031 static int __init net_dev_init(void)
12032 {
12033 int i, rc = -ENOMEM;
12034
12035 BUG_ON(!dev_boot_phase);
12036
12037 net_dev_struct_check();
12038
12039 if (dev_proc_init())
12040 goto out;
12041
12042 if (netdev_kobject_init())
12043 goto out;
12044
12045 for (i = 0; i < PTYPE_HASH_SIZE; i++)
12046 INIT_LIST_HEAD(&ptype_base[i]);
12047
12048 if (register_pernet_subsys(&netdev_net_ops))
12049 goto out;
12050
12051 /*
12052 * Initialise the packet receive queues.
12053 */
12054
12055 for_each_possible_cpu(i) {
12056 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
12057 struct softnet_data *sd = &per_cpu(softnet_data, i);
12058
12059 INIT_WORK(flush, flush_backlog);
12060
12061 skb_queue_head_init(&sd->input_pkt_queue);
12062 skb_queue_head_init(&sd->process_queue);
12063 #ifdef CONFIG_XFRM_OFFLOAD
12064 skb_queue_head_init(&sd->xfrm_backlog);
12065 #endif
12066 INIT_LIST_HEAD(&sd->poll_list);
12067 sd->output_queue_tailp = &sd->output_queue;
12068 #ifdef CONFIG_RPS
12069 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
12070 sd->cpu = i;
12071 #endif
12072 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
12073 spin_lock_init(&sd->defer_lock);
12074
12075 init_gro_hash(&sd->backlog);
12076 sd->backlog.poll = process_backlog;
12077 sd->backlog.weight = weight_p;
12078 INIT_LIST_HEAD(&sd->backlog.poll_list);
12079
12080 if (net_page_pool_create(i))
12081 goto out;
12082 }
12083 if (use_backlog_threads())
12084 smpboot_register_percpu_thread(&backlog_threads);
12085
12086 dev_boot_phase = 0;
12087
12088 /* The loopback device is special if any other network devices
12089 * is present in a network namespace the loopback device must
12090 * be present. Since we now dynamically allocate and free the
12091 * loopback device ensure this invariant is maintained by
12092 * keeping the loopback device as the first device on the
12093 * list of network devices. Ensuring the loopback devices
12094 * is the first device that appears and the last network device
12095 * that disappears.
12096 */
12097 if (register_pernet_device(&loopback_net_ops))
12098 goto out;
12099
12100 if (register_pernet_device(&default_device_ops))
12101 goto out;
12102
12103 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
12104 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
12105
12106 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
12107 NULL, dev_cpu_dead);
12108 WARN_ON(rc < 0);
12109 rc = 0;
12110
12111 /* avoid static key IPIs to isolated CPUs */
12112 if (housekeeping_enabled(HK_TYPE_MISC))
12113 net_enable_timestamp();
12114 out:
12115 if (rc < 0) {
12116 for_each_possible_cpu(i) {
12117 struct page_pool *pp_ptr;
12118
12119 pp_ptr = per_cpu(system_page_pool, i);
12120 if (!pp_ptr)
12121 continue;
12122
12123 page_pool_destroy(pp_ptr);
12124 per_cpu(system_page_pool, i) = NULL;
12125 }
12126 }
12127
12128 return rc;
12129 }
12130
12131 subsys_initcall(net_dev_init);
12132
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.