개념 설명 전체
net/core/dev.c · v6.18.37
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, <[email protected]>
8 * Mark Evans, <[email protected]>
9 *
10 * Additional Authors:
11 * Florian la Roche <[email protected]>
12 * Alan Cox <[email protected]>
13 * David Hinds <[email protected]>
14 * Alexey Kuznetsov <[email protected]>
15 * Adam Sulmicki <[email protected]>
16 * Pekka Riikonen <[email protected]>
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/ethtool_netlink.h>
96 #include <linux/skbuff.h>
97 #include <linux/kthread.h>
98 #include <linux/bpf.h>
99 #include <linux/bpf_trace.h>
100 #include <net/net_namespace.h>
101 #include <net/sock.h>
102 #include <net/busy_poll.h>
103 #include <linux/rtnetlink.h>
104 #include <linux/stat.h>
105 #include <net/dsa.h>
106 #include <net/dst.h>
107 #include <net/dst_metadata.h>
108 #include <net/gro.h>
109 #include <net/netdev_queues.h>
110 #include <net/pkt_sched.h>
111 #include <net/pkt_cls.h>
112 #include <net/checksum.h>
113 #include <net/xfrm.h>
114 #include <net/tcx.h>
115 #include <linux/highmem.h>
116 #include <linux/init.h>
117 #include <linux/module.h>
118 #include <linux/netpoll.h>
119 #include <linux/rcupdate.h>
120 #include <linux/delay.h>
121 #include <net/iw_handler.h>
122 #include <asm/current.h>
123 #include <linux/audit.h>
124 #include <linux/dmaengine.h>
125 #include <linux/err.h>
126 #include <linux/ctype.h>
127 #include <linux/if_arp.h>
128 #include <linux/if_vlan.h>
129 #include <linux/ip.h>
130 #include <net/ip.h>
131 #include <net/mpls.h>
132 #include <linux/ipv6.h>
133 #include <linux/in.h>
134 #include <linux/jhash.h>
135 #include <linux/random.h>
136 #include <trace/events/napi.h>
137 #include <trace/events/net.h>
138 #include <trace/events/skb.h>
139 #include <trace/events/qdisc.h>
140 #include <trace/events/xdp.h>
141 #include <linux/inetdevice.h>
142 #include <linux/cpu_rmap.h>
143 #include <linux/static_key.h>
144 #include <linux/hashtable.h>
145 #include <linux/vmalloc.h>
146 #include <linux/if_macvlan.h>
147 #include <linux/errqueue.h>
148 #include <linux/hrtimer.h>
149 #include <linux/netfilter_netdev.h>
150 #include <linux/crash_dump.h>
151 #include <linux/sctp.h>
152 #include <net/udp_tunnel.h>
153 #include <linux/net_namespace.h>
154 #include <linux/indirect_call_wrapper.h>
155 #include <net/devlink.h>
156 #include <linux/pm_runtime.h>
157 #include <linux/prandom.h>
158 #include <linux/once_lite.h>
159 #include <net/netdev_lock.h>
160 #include <net/netdev_rx_queue.h>
161 #include <net/page_pool/types.h>
162 #include <net/page_pool/helpers.h>
163 #include <net/page_pool/memory_provider.h>
164 #include <net/rps.h>
165 #include <linux/phy_link_topology.h>
166
167 #include "dev.h"
168 #include "devmem.h"
169 #include "net-sysfs.h"
170
171 static DEFINE_SPINLOCK(ptype_lock);
172 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
173
174 static int netif_rx_internal(struct sk_buff *skb);
175 static int call_netdevice_notifiers_extack(unsigned long val,
176 struct net_device *dev,
177 struct netlink_ext_ack *extack);
178
179 static DEFINE_MUTEX(ifalias_mutex);
180
181 /* protects napi_hash addition/deletion and napi_gen_id */
182 static DEFINE_SPINLOCK(napi_hash_lock);
183
184 static unsigned int napi_gen_id = NR_CPUS;
185 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
186
187 static inline void dev_base_seq_inc(struct net *net)
188 {
189 unsigned int val = net->dev_base_seq + 1;
190
191 WRITE_ONCE(net->dev_base_seq, val ?: 1);
192 }
193
194 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
195 {
196 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
197
198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
199 }
200
201 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
202 {
203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
204 }
205
206 #ifndef CONFIG_PREEMPT_RT
207
208 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key);
209
210 static int __init setup_backlog_napi_threads(char *arg)
211 {
212 static_branch_enable(&use_backlog_threads_key);
213 return 0;
214 }
215 early_param("thread_backlog_napi", setup_backlog_napi_threads);
216
217 static bool use_backlog_threads(void)
218 {
219 return static_branch_unlikely(&use_backlog_threads_key);
220 }
221
222 #else
223
224 static bool use_backlog_threads(void)
225 {
226 return true;
227 }
228
229 #endif
230
231 static inline void backlog_lock_irq_save(struct softnet_data *sd,
232 unsigned long *flags)
233 {
234 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
235 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
236 else
237 local_irq_save(*flags);
238 }
239
240 static inline void backlog_lock_irq_disable(struct softnet_data *sd)
241 {
242 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
243 spin_lock_irq(&sd->input_pkt_queue.lock);
244 else
245 local_irq_disable();
246 }
247
248 static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
249 unsigned long *flags)
250 {
251 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
252 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
253 else
254 local_irq_restore(*flags);
255 }
256
257 static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
258 {
259 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
260 spin_unlock_irq(&sd->input_pkt_queue.lock);
261 else
262 local_irq_enable();
263 }
264
265 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
266 const char *name)
267 {
268 struct netdev_name_node *name_node;
269
270 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
271 if (!name_node)
272 return NULL;
273 INIT_HLIST_NODE(&name_node->hlist);
274 name_node->dev = dev;
275 name_node->name = name;
276 return name_node;
277 }
278
279 static struct netdev_name_node *
280 netdev_name_node_head_alloc(struct net_device *dev)
281 {
282 struct netdev_name_node *name_node;
283
284 name_node = netdev_name_node_alloc(dev, dev->name);
285 if (!name_node)
286 return NULL;
287 INIT_LIST_HEAD(&name_node->list);
288 return name_node;
289 }
290
291 static void netdev_name_node_free(struct netdev_name_node *name_node)
292 {
293 kfree(name_node);
294 }
295
296 static void netdev_name_node_add(struct net *net,
297 struct netdev_name_node *name_node)
298 {
299 hlist_add_head_rcu(&name_node->hlist,
300 dev_name_hash(net, name_node->name));
301 }
302
303 static void netdev_name_node_del(struct netdev_name_node *name_node)
304 {
305 hlist_del_rcu(&name_node->hlist);
306 }
307
308 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
309 const char *name)
310 {
311 struct hlist_head *head = dev_name_hash(net, name);
312 struct netdev_name_node *name_node;
313
314 hlist_for_each_entry(name_node, head, hlist)
315 if (!strcmp(name_node->name, name))
316 return name_node;
317 return NULL;
318 }
319
320 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
321 const char *name)
322 {
323 struct hlist_head *head = dev_name_hash(net, name);
324 struct netdev_name_node *name_node;
325
326 hlist_for_each_entry_rcu(name_node, head, hlist)
327 if (!strcmp(name_node->name, name))
328 return name_node;
329 return NULL;
330 }
331
332 bool netdev_name_in_use(struct net *net, const char *name)
333 {
334 return netdev_name_node_lookup(net, name);
335 }
336 EXPORT_SYMBOL(netdev_name_in_use);
337
338 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
339 {
340 struct netdev_name_node *name_node;
341 struct net *net = dev_net(dev);
342
343 name_node = netdev_name_node_lookup(net, name);
344 if (name_node)
345 return -EEXIST;
346 name_node = netdev_name_node_alloc(dev, name);
347 if (!name_node)
348 return -ENOMEM;
349 netdev_name_node_add(net, name_node);
350 /* The node that holds dev->name acts as a head of per-device list. */
351 list_add_tail_rcu(&name_node->list, &dev->name_node->list);
352
353 return 0;
354 }
355
356 static void netdev_name_node_alt_free(struct rcu_head *head)
357 {
358 struct netdev_name_node *name_node =
359 container_of(head, struct netdev_name_node, rcu);
360
361 kfree(name_node->name);
362 netdev_name_node_free(name_node);
363 }
364
365 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
366 {
367 netdev_name_node_del(name_node);
368 list_del(&name_node->list);
369 call_rcu(&name_node->rcu, netdev_name_node_alt_free);
370 }
371
372 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
373 {
374 struct netdev_name_node *name_node;
375 struct net *net = dev_net(dev);
376
377 name_node = netdev_name_node_lookup(net, name);
378 if (!name_node)
379 return -ENOENT;
380 /* lookup might have found our primary name or a name belonging
381 * to another device.
382 */
383 if (name_node == dev->name_node || name_node->dev != dev)
384 return -EINVAL;
385
386 __netdev_name_node_alt_destroy(name_node);
387 return 0;
388 }
389
390 static void netdev_name_node_alt_flush(struct net_device *dev)
391 {
392 struct netdev_name_node *name_node, *tmp;
393
394 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) {
395 list_del(&name_node->list);
396 netdev_name_node_alt_free(&name_node->rcu);
397 }
398 }
399
400 /* Device list insertion */
401 static void list_netdevice(struct net_device *dev)
402 {
403 struct netdev_name_node *name_node;
404 struct net *net = dev_net(dev);
405
406 ASSERT_RTNL();
407
408 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
409 netdev_name_node_add(net, dev->name_node);
410 hlist_add_head_rcu(&dev->index_hlist,
411 dev_index_hash(net, dev->ifindex));
412
413 netdev_for_each_altname(dev, name_node)
414 netdev_name_node_add(net, name_node);
415
416 /* We reserved the ifindex, this can't fail */
417 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL));
418
419 dev_base_seq_inc(net);
420 }
421
422 /* Device list removal
423 * caller must respect a RCU grace period before freeing/reusing dev
424 */
425 static void unlist_netdevice(struct net_device *dev)
426 {
427 struct netdev_name_node *name_node;
428 struct net *net = dev_net(dev);
429
430 ASSERT_RTNL();
431
432 xa_erase(&net->dev_by_index, dev->ifindex);
433
434 netdev_for_each_altname(dev, name_node)
435 netdev_name_node_del(name_node);
436
437 /* Unlink dev from the device chain */
438 list_del_rcu(&dev->dev_list);
439 netdev_name_node_del(dev->name_node);
440 hlist_del_rcu(&dev->index_hlist);
441
442 dev_base_seq_inc(dev_net(dev));
443 }
444
445 /*
446 * Our notifier list
447 */
448
449 static RAW_NOTIFIER_HEAD(netdev_chain);
450
451 /*
452 * Device drivers call our routines to queue packets here. We empty the
453 * queue in the local softnet handler.
454 */
455
456 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = {
457 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock),
458 };
459 EXPORT_PER_CPU_SYMBOL(softnet_data);
460
461 /* Page_pool has a lockless array/stack to alloc/recycle pages.
462 * PP consumers must pay attention to run APIs in the appropriate context
463 * (e.g. NAPI context).
464 */
465 DEFINE_PER_CPU(struct page_pool_bh, system_page_pool) = {
466 .bh_lock = INIT_LOCAL_LOCK(bh_lock),
467 };
468
469 #ifdef CONFIG_LOCKDEP
470 /*
471 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
472 * according to dev->type
473 */
474 static const unsigned short netdev_lock_type[] = {
475 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
476 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
477 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
478 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
479 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
480 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
481 ARPHRD_CAN, ARPHRD_MCTP,
482 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
483 ARPHRD_RAWHDLC, ARPHRD_RAWIP,
484 ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
485 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
486 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
487 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
488 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
489 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
490 ARPHRD_IEEE80211_RADIOTAP,
491 ARPHRD_IEEE802154, ARPHRD_IEEE802154_MONITOR,
492 ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
493 ARPHRD_CAIF, ARPHRD_IP6GRE, ARPHRD_NETLINK, ARPHRD_6LOWPAN,
494 ARPHRD_VSOCKMON,
495 ARPHRD_VOID, ARPHRD_NONE};
496
497 static const char *const netdev_lock_name[] = {
498 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
499 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
500 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
501 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
502 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
503 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
504 "_xmit_CAN", "_xmit_MCTP",
505 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
506 "_xmit_RAWHDLC", "_xmit_RAWIP",
507 "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
508 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
509 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
510 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
511 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
512 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
513 "_xmit_IEEE80211_RADIOTAP",
514 "_xmit_IEEE802154", "_xmit_IEEE802154_MONITOR",
515 "_xmit_PHONET", "_xmit_PHONET_PIPE",
516 "_xmit_CAIF", "_xmit_IP6GRE", "_xmit_NETLINK", "_xmit_6LOWPAN",
517 "_xmit_VSOCKMON",
518 "_xmit_VOID", "_xmit_NONE"};
519
520 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
521 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
522
523 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
524 {
525 int i;
526
527 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
528 if (netdev_lock_type[i] == dev_type)
529 return i;
530 /* the last key is used by default */
531 WARN_ONCE(1, "netdev_lock_pos() could not find dev_type=%u\n", dev_type);
532 return ARRAY_SIZE(netdev_lock_type) - 1;
533 }
534
535 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
536 unsigned short dev_type)
537 {
538 int i;
539
540 i = netdev_lock_pos(dev_type);
541 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
542 netdev_lock_name[i]);
543 }
544
545 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
546 {
547 int i;
548
549 i = netdev_lock_pos(dev->type);
550 lockdep_set_class_and_name(&dev->addr_list_lock,
551 &netdev_addr_lock_key[i],
552 netdev_lock_name[i]);
553 }
554 #else
555 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
556 unsigned short dev_type)
557 {
558 }
559
560 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
561 {
562 }
563 #endif
564
565 /*******************************************************************************
566 *
567 * Protocol management and registration routines
568 *
569 *******************************************************************************/
570
571
572 /*
573 * Add a protocol ID to the list. Now that the input handler is
574 * smarter we can dispense with all the messy stuff that used to be
575 * here.
576 *
577 * BEWARE!!! Protocol handlers, mangling input packets,
578 * MUST BE last in hash buckets and checking protocol handlers
579 * MUST start from promiscuous ptype_all chain in net_bh.
580 * It is true now, do not change it.
581 * Explanation follows: if protocol handler, mangling packet, will
582 * be the first on list, it is not able to sense, that packet
583 * is cloned and should be copied-on-write, so that it will
584 * change it and subsequent readers will get broken packet.
585 * --ANK (980803)
586 */
587
588 static inline struct list_head *ptype_head(const struct packet_type *pt)
589 {
590 if (pt->type == htons(ETH_P_ALL)) {
591 if (!pt->af_packet_net && !pt->dev)
592 return NULL;
593
594 return pt->dev ? &pt->dev->ptype_all :
595 &pt->af_packet_net->ptype_all;
596 }
597
598 if (pt->dev)
599 return &pt->dev->ptype_specific;
600
601 return pt->af_packet_net ? &pt->af_packet_net->ptype_specific :
602 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
603 }
604
605 /**
606 * dev_add_pack - add packet handler
607 * @pt: packet type declaration
608 *
609 * Add a protocol handler to the networking stack. The passed &packet_type
610 * is linked into kernel lists and may not be freed until it has been
611 * removed from the kernel lists.
612 *
613 * This call does not sleep therefore it can not
614 * guarantee all CPU's that are in middle of receiving packets
615 * will see the new packet type (until the next received packet).
616 */
617
618 void dev_add_pack(struct packet_type *pt)
619 {
620 struct list_head *head = ptype_head(pt);
621
622 if (WARN_ON_ONCE(!head))
623 return;
624
625 spin_lock(&ptype_lock);
626 list_add_rcu(&pt->list, head);
627 spin_unlock(&ptype_lock);
628 }
629 EXPORT_SYMBOL(dev_add_pack);
630
631 /**
632 * __dev_remove_pack - remove packet handler
633 * @pt: packet type declaration
634 *
635 * Remove a protocol handler that was previously added to the kernel
636 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
637 * from the kernel lists and can be freed or reused once this function
638 * returns.
639 *
640 * The packet type might still be in use by receivers
641 * and must not be freed until after all the CPU's have gone
642 * through a quiescent state.
643 */
644 void __dev_remove_pack(struct packet_type *pt)
645 {
646 struct list_head *head = ptype_head(pt);
647 struct packet_type *pt1;
648
649 if (!head)
650 return;
651
652 spin_lock(&ptype_lock);
653
654 list_for_each_entry(pt1, head, list) {
655 if (pt == pt1) {
656 list_del_rcu(&pt->list);
657 goto out;
658 }
659 }
660
661 pr_warn("dev_remove_pack: %p not found\n", pt);
662 out:
663 spin_unlock(&ptype_lock);
664 }
665 EXPORT_SYMBOL(__dev_remove_pack);
666
667 /**
668 * dev_remove_pack - remove packet handler
669 * @pt: packet type declaration
670 *
671 * Remove a protocol handler that was previously added to the kernel
672 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
673 * from the kernel lists and can be freed or reused once this function
674 * returns.
675 *
676 * This call sleeps to guarantee that no CPU is looking at the packet
677 * type after return.
678 */
679 void dev_remove_pack(struct packet_type *pt)
680 {
681 __dev_remove_pack(pt);
682
683 synchronize_net();
684 }
685 EXPORT_SYMBOL(dev_remove_pack);
686
687
688 /*******************************************************************************
689 *
690 * Device Interface Subroutines
691 *
692 *******************************************************************************/
693
694 /**
695 * dev_get_iflink - get 'iflink' value of a interface
696 * @dev: targeted interface
697 *
698 * Indicates the ifindex the interface is linked to.
699 * Physical interfaces have the same 'ifindex' and 'iflink' values.
700 */
701
702 int dev_get_iflink(const struct net_device *dev)
703 {
704 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
705 return dev->netdev_ops->ndo_get_iflink(dev);
706
707 return READ_ONCE(dev->ifindex);
708 }
709 EXPORT_SYMBOL(dev_get_iflink);
710
711 /**
712 * dev_fill_metadata_dst - Retrieve tunnel egress information.
713 * @dev: targeted interface
714 * @skb: The packet.
715 *
716 * For better visibility of tunnel traffic OVS needs to retrieve
717 * egress tunnel information for a packet. Following API allows
718 * user to get this info.
719 */
720 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
721 {
722 struct ip_tunnel_info *info;
723
724 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
725 return -EINVAL;
726
727 info = skb_tunnel_info_unclone(skb);
728 if (!info)
729 return -ENOMEM;
730 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
731 return -EINVAL;
732
733 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
734 }
735 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
736
737 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
738 {
739 int k = stack->num_paths++;
740
741 if (k >= NET_DEVICE_PATH_STACK_MAX)
742 return NULL;
743
744 return &stack->path[k];
745 }
746
747 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
748 struct net_device_path_stack *stack)
749 {
750 const struct net_device *last_dev;
751 struct net_device_path_ctx ctx = {
752 .dev = dev,
753 };
754 struct net_device_path *path;
755 int ret = 0;
756
757 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
758 stack->num_paths = 0;
759 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
760 last_dev = ctx.dev;
761 path = dev_fwd_path(stack);
762 if (!path)
763 return -1;
764
765 memset(path, 0, sizeof(struct net_device_path));
766 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
767 if (ret < 0)
768 return -1;
769
770 if (WARN_ON_ONCE(last_dev == ctx.dev))
771 return -1;
772 }
773
774 if (!ctx.dev)
775 return ret;
776
777 path = dev_fwd_path(stack);
778 if (!path)
779 return -1;
780 path->type = DEV_PATH_ETHERNET;
781 path->dev = ctx.dev;
782
783 return ret;
784 }
785 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
786
787 /* must be called under rcu_read_lock(), as we dont take a reference */
788 static struct napi_struct *napi_by_id(unsigned int napi_id)
789 {
790 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
791 struct napi_struct *napi;
792
793 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
794 if (napi->napi_id == napi_id)
795 return napi;
796
797 return NULL;
798 }
799
800 /* must be called under rcu_read_lock(), as we dont take a reference */
801 static struct napi_struct *
802 netdev_napi_by_id(struct net *net, unsigned int napi_id)
803 {
804 struct napi_struct *napi;
805
806 napi = napi_by_id(napi_id);
807 if (!napi)
808 return NULL;
809
810 if (WARN_ON_ONCE(!napi->dev))
811 return NULL;
812 if (!net_eq(net, dev_net(napi->dev)))
813 return NULL;
814
815 return napi;
816 }
817
818 /**
819 * netdev_napi_by_id_lock() - find a device by NAPI ID and lock it
820 * @net: the applicable net namespace
821 * @napi_id: ID of a NAPI of a target device
822 *
823 * Find a NAPI instance with @napi_id. Lock its device.
824 * The device must be in %NETREG_REGISTERED state for lookup to succeed.
825 * netdev_unlock() must be called to release it.
826 *
827 * Return: pointer to NAPI, its device with lock held, NULL if not found.
828 */
829 struct napi_struct *
830 netdev_napi_by_id_lock(struct net *net, unsigned int napi_id)
831 {
832 struct napi_struct *napi;
833 struct net_device *dev;
834
835 rcu_read_lock();
836 napi = netdev_napi_by_id(net, napi_id);
837 if (!napi || READ_ONCE(napi->dev->reg_state) != NETREG_REGISTERED) {
838 rcu_read_unlock();
839 return NULL;
840 }
841
842 dev = napi->dev;
843 dev_hold(dev);
844 rcu_read_unlock();
845
846 dev = __netdev_put_lock(dev, net);
847 if (!dev)
848 return NULL;
849
850 rcu_read_lock();
851 napi = netdev_napi_by_id(net, napi_id);
852 if (napi && napi->dev != dev)
853 napi = NULL;
854 rcu_read_unlock();
855
856 if (!napi)
857 netdev_unlock(dev);
858 return napi;
859 }
860
861 /**
862 * __dev_get_by_name - find a device by its name
863 * @net: the applicable net namespace
864 * @name: name to find
865 *
866 * Find an interface by name. Must be called under RTNL semaphore.
867 * If the name is found a pointer to the device is returned.
868 * If the name is not found then %NULL is returned. The
869 * reference counters are not incremented so the caller must be
870 * careful with locks.
871 */
872
873 struct net_device *__dev_get_by_name(struct net *net, const char *name)
874 {
875 struct netdev_name_node *node_name;
876
877 node_name = netdev_name_node_lookup(net, name);
878 return node_name ? node_name->dev : NULL;
879 }
880 EXPORT_SYMBOL(__dev_get_by_name);
881
882 /**
883 * dev_get_by_name_rcu - find a device by its name
884 * @net: the applicable net namespace
885 * @name: name to find
886 *
887 * Find an interface by name.
888 * If the name is found a pointer to the device is returned.
889 * If the name is not found then %NULL is returned.
890 * The reference counters are not incremented so the caller must be
891 * careful with locks. The caller must hold RCU lock.
892 */
893
894 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
895 {
896 struct netdev_name_node *node_name;
897
898 node_name = netdev_name_node_lookup_rcu(net, name);
899 return node_name ? node_name->dev : NULL;
900 }
901 EXPORT_SYMBOL(dev_get_by_name_rcu);
902
903 /* Deprecated for new users, call netdev_get_by_name() instead */
904 struct net_device *dev_get_by_name(struct net *net, const char *name)
905 {
906 struct net_device *dev;
907
908 rcu_read_lock();
909 dev = dev_get_by_name_rcu(net, name);
910 dev_hold(dev);
911 rcu_read_unlock();
912 return dev;
913 }
914 EXPORT_SYMBOL(dev_get_by_name);
915
916 /**
917 * netdev_get_by_name() - find a device by its name
918 * @net: the applicable net namespace
919 * @name: name to find
920 * @tracker: tracking object for the acquired reference
921 * @gfp: allocation flags for the tracker
922 *
923 * Find an interface by name. This can be called from any
924 * context and does its own locking. The returned handle has
925 * the usage count incremented and the caller must use netdev_put() to
926 * release it when it is no longer needed. %NULL is returned if no
927 * matching device is found.
928 */
929 struct net_device *netdev_get_by_name(struct net *net, const char *name,
930 netdevice_tracker *tracker, gfp_t gfp)
931 {
932 struct net_device *dev;
933
934 dev = dev_get_by_name(net, name);
935 if (dev)
936 netdev_tracker_alloc(dev, tracker, gfp);
937 return dev;
938 }
939 EXPORT_SYMBOL(netdev_get_by_name);
940
941 /**
942 * __dev_get_by_index - find a device by its ifindex
943 * @net: the applicable net namespace
944 * @ifindex: index of device
945 *
946 * Search for an interface by index. Returns %NULL if the device
947 * is not found or a pointer to the device. The device has not
948 * had its reference counter increased so the caller must be careful
949 * about locking. The caller must hold the RTNL semaphore.
950 */
951
952 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
953 {
954 struct net_device *dev;
955 struct hlist_head *head = dev_index_hash(net, ifindex);
956
957 hlist_for_each_entry(dev, head, index_hlist)
958 if (dev->ifindex == ifindex)
959 return dev;
960
961 return NULL;
962 }
963 EXPORT_SYMBOL(__dev_get_by_index);
964
965 /**
966 * dev_get_by_index_rcu - find a device by its ifindex
967 * @net: the applicable net namespace
968 * @ifindex: index of device
969 *
970 * Search for an interface by index. Returns %NULL if the device
971 * is not found or a pointer to the device. The device has not
972 * had its reference counter increased so the caller must be careful
973 * about locking. The caller must hold RCU lock.
974 */
975
976 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
977 {
978 struct net_device *dev;
979 struct hlist_head *head = dev_index_hash(net, ifindex);
980
981 hlist_for_each_entry_rcu(dev, head, index_hlist)
982 if (dev->ifindex == ifindex)
983 return dev;
984
985 return NULL;
986 }
987 EXPORT_SYMBOL(dev_get_by_index_rcu);
988
989 /* Deprecated for new users, call netdev_get_by_index() instead */
990 struct net_device *dev_get_by_index(struct net *net, int ifindex)
991 {
992 struct net_device *dev;
993
994 rcu_read_lock();
995 dev = dev_get_by_index_rcu(net, ifindex);
996 dev_hold(dev);
997 rcu_read_unlock();
998 return dev;
999 }
1000 EXPORT_SYMBOL(dev_get_by_index);
1001
1002 /**
1003 * netdev_get_by_index() - find a device by its ifindex
1004 * @net: the applicable net namespace
1005 * @ifindex: index of device
1006 * @tracker: tracking object for the acquired reference
1007 * @gfp: allocation flags for the tracker
1008 *
1009 * Search for an interface by index. Returns NULL if the device
1010 * is not found or a pointer to the device. The device returned has
1011 * had a reference added and the pointer is safe until the user calls
1012 * netdev_put() to indicate they have finished with it.
1013 */
1014 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
1015 netdevice_tracker *tracker, gfp_t gfp)
1016 {
1017 struct net_device *dev;
1018
1019 dev = dev_get_by_index(net, ifindex);
1020 if (dev)
1021 netdev_tracker_alloc(dev, tracker, gfp);
1022 return dev;
1023 }
1024 EXPORT_SYMBOL(netdev_get_by_index);
1025
1026 /**
1027 * dev_get_by_napi_id - find a device by napi_id
1028 * @napi_id: ID of the NAPI struct
1029 *
1030 * Search for an interface by NAPI ID. Returns %NULL if the device
1031 * is not found or a pointer to the device. The device has not had
1032 * its reference counter increased so the caller must be careful
1033 * about locking. The caller must hold RCU lock.
1034 */
1035 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
1036 {
1037 struct napi_struct *napi;
1038
1039 WARN_ON_ONCE(!rcu_read_lock_held());
1040
1041 if (!napi_id_valid(napi_id))
1042 return NULL;
1043
1044 napi = napi_by_id(napi_id);
1045
1046 return napi ? napi->dev : NULL;
1047 }
1048
1049 /* Release the held reference on the net_device, and if the net_device
1050 * is still registered try to lock the instance lock. If device is being
1051 * unregistered NULL will be returned (but the reference has been released,
1052 * either way!)
1053 *
1054 * This helper is intended for locking net_device after it has been looked up
1055 * using a lockless lookup helper. Lock prevents the instance from going away.
1056 */
1057 struct net_device *__netdev_put_lock(struct net_device *dev, struct net *net)
1058 {
1059 netdev_lock(dev);
1060 if (dev->reg_state > NETREG_REGISTERED ||
1061 dev->moving_ns || !net_eq(dev_net(dev), net)) {
1062 netdev_unlock(dev);
1063 dev_put(dev);
1064 return NULL;
1065 }
1066 dev_put(dev);
1067 return dev;
1068 }
1069
1070 static struct net_device *
1071 __netdev_put_lock_ops_compat(struct net_device *dev, struct net *net)
1072 {
1073 netdev_lock_ops_compat(dev);
1074 if (dev->reg_state > NETREG_REGISTERED ||
1075 dev->moving_ns || !net_eq(dev_net(dev), net)) {
1076 netdev_unlock_ops_compat(dev);
1077 dev_put(dev);
1078 return NULL;
1079 }
1080 dev_put(dev);
1081 return dev;
1082 }
1083
1084 /**
1085 * netdev_get_by_index_lock() - find a device by its ifindex
1086 * @net: the applicable net namespace
1087 * @ifindex: index of device
1088 *
1089 * Search for an interface by index. If a valid device
1090 * with @ifindex is found it will be returned with netdev->lock held.
1091 * netdev_unlock() must be called to release it.
1092 *
1093 * Return: pointer to a device with lock held, NULL if not found.
1094 */
1095 struct net_device *netdev_get_by_index_lock(struct net *net, int ifindex)
1096 {
1097 struct net_device *dev;
1098
1099 dev = dev_get_by_index(net, ifindex);
1100 if (!dev)
1101 return NULL;
1102
1103 return __netdev_put_lock(dev, net);
1104 }
1105
1106 struct net_device *
1107 netdev_get_by_index_lock_ops_compat(struct net *net, int ifindex)
1108 {
1109 struct net_device *dev;
1110
1111 dev = dev_get_by_index(net, ifindex);
1112 if (!dev)
1113 return NULL;
1114
1115 return __netdev_put_lock_ops_compat(dev, net);
1116 }
1117
1118 struct net_device *
1119 netdev_xa_find_lock(struct net *net, struct net_device *dev,
1120 unsigned long *index)
1121 {
1122 if (dev)
1123 netdev_unlock(dev);
1124
1125 do {
1126 rcu_read_lock();
1127 dev = xa_find(&net->dev_by_index, index, ULONG_MAX, XA_PRESENT);
1128 if (!dev) {
1129 rcu_read_unlock();
1130 return NULL;
1131 }
1132 dev_hold(dev);
1133 rcu_read_unlock();
1134
1135 dev = __netdev_put_lock(dev, net);
1136 if (dev)
1137 return dev;
1138
1139 (*index)++;
1140 } while (true);
1141 }
1142
1143 struct net_device *
1144 netdev_xa_find_lock_ops_compat(struct net *net, struct net_device *dev,
1145 unsigned long *index)
1146 {
1147 if (dev)
1148 netdev_unlock_ops_compat(dev);
1149
1150 do {
1151 rcu_read_lock();
1152 dev = xa_find(&net->dev_by_index, index, ULONG_MAX, XA_PRESENT);
1153 if (!dev) {
1154 rcu_read_unlock();
1155 return NULL;
1156 }
1157 dev_hold(dev);
1158 rcu_read_unlock();
1159
1160 dev = __netdev_put_lock_ops_compat(dev, net);
1161 if (dev)
1162 return dev;
1163
1164 (*index)++;
1165 } while (true);
1166 }
1167
1168 static DEFINE_SEQLOCK(netdev_rename_lock);
1169
1170 void netdev_copy_name(struct net_device *dev, char *name)
1171 {
1172 unsigned int seq;
1173
1174 do {
1175 seq = read_seqbegin(&netdev_rename_lock);
1176 strscpy(name, dev->name, IFNAMSIZ);
1177 } while (read_seqretry(&netdev_rename_lock, seq));
1178 }
1179
1180 /**
1181 * netdev_get_name - get a netdevice name, knowing its ifindex.
1182 * @net: network namespace
1183 * @name: a pointer to the buffer where the name will be stored.
1184 * @ifindex: the ifindex of the interface to get the name from.
1185 */
1186 int netdev_get_name(struct net *net, char *name, int ifindex)
1187 {
1188 struct net_device *dev;
1189 int ret;
1190
1191 rcu_read_lock();
1192
1193 dev = dev_get_by_index_rcu(net, ifindex);
1194 if (!dev) {
1195 ret = -ENODEV;
1196 goto out;
1197 }
1198
1199 netdev_copy_name(dev, name);
1200
1201 ret = 0;
1202 out:
1203 rcu_read_unlock();
1204 return ret;
1205 }
1206
1207 static bool dev_addr_cmp(struct net_device *dev, unsigned short type,
1208 const char *ha)
1209 {
1210 return dev->type == type && !memcmp(dev->dev_addr, ha, dev->addr_len);
1211 }
1212
1213 /**
1214 * dev_getbyhwaddr_rcu - find a device by its hardware address
1215 * @net: the applicable net namespace
1216 * @type: media type of device
1217 * @ha: hardware address
1218 *
1219 * Search for an interface by MAC address. Returns NULL if the device
1220 * is not found or a pointer to the device.
1221 * The caller must hold RCU.
1222 * The returned device has not had its ref count increased
1223 * and the caller must therefore be careful about locking
1224 *
1225 */
1226
1227 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1228 const char *ha)
1229 {
1230 struct net_device *dev;
1231
1232 for_each_netdev_rcu(net, dev)
1233 if (dev_addr_cmp(dev, type, ha))
1234 return dev;
1235
1236 return NULL;
1237 }
1238 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1239
1240 /**
1241 * dev_getbyhwaddr() - find a device by its hardware address
1242 * @net: the applicable net namespace
1243 * @type: media type of device
1244 * @ha: hardware address
1245 *
1246 * Similar to dev_getbyhwaddr_rcu(), but the owner needs to hold
1247 * rtnl_lock.
1248 *
1249 * Context: rtnl_lock() must be held.
1250 * Return: pointer to the net_device, or NULL if not found
1251 */
1252 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type,
1253 const char *ha)
1254 {
1255 struct net_device *dev;
1256
1257 ASSERT_RTNL();
1258 for_each_netdev(net, dev)
1259 if (dev_addr_cmp(dev, type, ha))
1260 return dev;
1261
1262 return NULL;
1263 }
1264 EXPORT_SYMBOL(dev_getbyhwaddr);
1265
1266 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1267 {
1268 struct net_device *dev, *ret = NULL;
1269
1270 rcu_read_lock();
1271 for_each_netdev_rcu(net, dev)
1272 if (dev->type == type) {
1273 dev_hold(dev);
1274 ret = dev;
1275 break;
1276 }
1277 rcu_read_unlock();
1278 return ret;
1279 }
1280 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1281
1282 /**
1283 * netdev_get_by_flags_rcu - find any device with given flags
1284 * @net: the applicable net namespace
1285 * @tracker: tracking object for the acquired reference
1286 * @if_flags: IFF_* values
1287 * @mask: bitmask of bits in if_flags to check
1288 *
1289 * Search for any interface with the given flags.
1290 *
1291 * Context: rcu_read_lock() must be held.
1292 * Returns: NULL if a device is not found or a pointer to the device.
1293 */
1294 struct net_device *netdev_get_by_flags_rcu(struct net *net, netdevice_tracker *tracker,
1295 unsigned short if_flags, unsigned short mask)
1296 {
1297 struct net_device *dev;
1298
1299 for_each_netdev_rcu(net, dev) {
1300 if (((READ_ONCE(dev->flags) ^ if_flags) & mask) == 0) {
1301 netdev_hold(dev, tracker, GFP_ATOMIC);
1302 return dev;
1303 }
1304 }
1305
1306 return NULL;
1307 }
1308 EXPORT_IPV6_MOD(netdev_get_by_flags_rcu);
1309
1310 /**
1311 * dev_valid_name - check if name is okay for network device
1312 * @name: name string
1313 *
1314 * Network device names need to be valid file names to
1315 * allow sysfs to work. We also disallow any kind of
1316 * whitespace.
1317 */
1318 bool dev_valid_name(const char *name)
1319 {
1320 if (*name == '\0')
1321 return false;
1322 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1323 return false;
1324 if (!strcmp(name, ".") || !strcmp(name, ".."))
1325 return false;
1326
1327 while (*name) {
1328 if (*name == '/' || *name == ':' || isspace(*name))
1329 return false;
1330 name++;
1331 }
1332 return true;
1333 }
1334 EXPORT_SYMBOL(dev_valid_name);
1335
1336 /**
1337 * __dev_alloc_name - allocate a name for a device
1338 * @net: network namespace to allocate the device name in
1339 * @name: name format string
1340 * @res: result name string
1341 *
1342 * Passed a format string - eg "lt%d" it will try and find a suitable
1343 * id. It scans list of devices to build up a free map, then chooses
1344 * the first empty slot. The caller must hold the dev_base or rtnl lock
1345 * while allocating the name and adding the device in order to avoid
1346 * duplicates.
1347 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1348 * Returns the number of the unit assigned or a negative errno code.
1349 */
1350
1351 static int __dev_alloc_name(struct net *net, const char *name, char *res)
1352 {
1353 int i = 0;
1354 const char *p;
1355 const int max_netdevices = 8*PAGE_SIZE;
1356 unsigned long *inuse;
1357 struct net_device *d;
1358 char buf[IFNAMSIZ];
1359
1360 /* Verify the string as this thing may have come from the user.
1361 * There must be one "%d" and no other "%" characters.
1362 */
1363 p = strchr(name, '%');
1364 if (!p || p[1] != 'd' || strchr(p + 2, '%'))
1365 return -EINVAL;
1366
1367 /* Use one page as a bit array of possible slots */
1368 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC);
1369 if (!inuse)
1370 return -ENOMEM;
1371
1372 for_each_netdev(net, d) {
1373 struct netdev_name_node *name_node;
1374
1375 netdev_for_each_altname(d, name_node) {
1376 if (!sscanf(name_node->name, name, &i))
1377 continue;
1378 if (i < 0 || i >= max_netdevices)
1379 continue;
1380
1381 /* avoid cases where sscanf is not exact inverse of printf */
1382 snprintf(buf, IFNAMSIZ, name, i);
1383 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1384 __set_bit(i, inuse);
1385 }
1386 if (!sscanf(d->name, name, &i))
1387 continue;
1388 if (i < 0 || i >= max_netdevices)
1389 continue;
1390
1391 /* avoid cases where sscanf is not exact inverse of printf */
1392 snprintf(buf, IFNAMSIZ, name, i);
1393 if (!strncmp(buf, d->name, IFNAMSIZ))
1394 __set_bit(i, inuse);
1395 }
1396
1397 i = find_first_zero_bit(inuse, max_netdevices);
1398 bitmap_free(inuse);
1399 if (i == max_netdevices)
1400 return -ENFILE;
1401
1402 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */
1403 strscpy(buf, name, IFNAMSIZ);
1404 snprintf(res, IFNAMSIZ, buf, i);
1405 return i;
1406 }
1407
1408 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */
1409 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1410 const char *want_name, char *out_name,
1411 int dup_errno)
1412 {
1413 if (!dev_valid_name(want_name))
1414 return -EINVAL;
1415
1416 if (strchr(want_name, '%'))
1417 return __dev_alloc_name(net, want_name, out_name);
1418
1419 if (netdev_name_in_use(net, want_name))
1420 return -dup_errno;
1421 if (out_name != want_name)
1422 strscpy(out_name, want_name, IFNAMSIZ);
1423 return 0;
1424 }
1425
1426 /**
1427 * dev_alloc_name - allocate a name for a device
1428 * @dev: device
1429 * @name: name format string
1430 *
1431 * Passed a format string - eg "lt%d" it will try and find a suitable
1432 * id. It scans list of devices to build up a free map, then chooses
1433 * the first empty slot. The caller must hold the dev_base or rtnl lock
1434 * while allocating the name and adding the device in order to avoid
1435 * duplicates.
1436 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1437 * Returns the number of the unit assigned or a negative errno code.
1438 */
1439
1440 int dev_alloc_name(struct net_device *dev, const char *name)
1441 {
1442 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE);
1443 }
1444 EXPORT_SYMBOL(dev_alloc_name);
1445
1446 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1447 const char *name)
1448 {
1449 int ret;
1450
1451 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST);
1452 return ret < 0 ? ret : 0;
1453 }
1454
1455 int netif_change_name(struct net_device *dev, const char *newname)
1456 {
1457 struct net *net = dev_net(dev);
1458 unsigned char old_assign_type;
1459 char oldname[IFNAMSIZ];
1460 int err = 0;
1461 int ret;
1462
1463 ASSERT_RTNL_NET(net);
1464
1465 if (!strncmp(newname, dev->name, IFNAMSIZ))
1466 return 0;
1467
1468 memcpy(oldname, dev->name, IFNAMSIZ);
1469
1470 write_seqlock_bh(&netdev_rename_lock);
1471 err = dev_get_valid_name(net, dev, newname);
1472 write_sequnlock_bh(&netdev_rename_lock);
1473
1474 if (err < 0)
1475 return err;
1476
1477 if (oldname[0] && !strchr(oldname, '%'))
1478 netdev_info(dev, "renamed from %s%s\n", oldname,
1479 dev->flags & IFF_UP ? " (while UP)" : "");
1480
1481 old_assign_type = dev->name_assign_type;
1482 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED);
1483
1484 rollback:
1485 ret = device_rename(&dev->dev, dev->name);
1486 if (ret) {
1487 write_seqlock_bh(&netdev_rename_lock);
1488 memcpy(dev->name, oldname, IFNAMSIZ);
1489 write_sequnlock_bh(&netdev_rename_lock);
1490 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1491 return ret;
1492 }
1493
1494 netdev_adjacent_rename_links(dev, oldname);
1495
1496 netdev_name_node_del(dev->name_node);
1497
1498 synchronize_net();
1499
1500 netdev_name_node_add(net, dev->name_node);
1501
1502 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1503 ret = notifier_to_errno(ret);
1504
1505 if (ret) {
1506 /* err >= 0 after dev_alloc_name() or stores the first errno */
1507 if (err >= 0) {
1508 err = ret;
1509 write_seqlock_bh(&netdev_rename_lock);
1510 memcpy(dev->name, oldname, IFNAMSIZ);
1511 write_sequnlock_bh(&netdev_rename_lock);
1512 memcpy(oldname, newname, IFNAMSIZ);
1513 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1514 old_assign_type = NET_NAME_RENAMED;
1515 goto rollback;
1516 } else {
1517 netdev_err(dev, "name change rollback failed: %d\n",
1518 ret);
1519 }
1520 }
1521
1522 return err;
1523 }
1524
1525 int netif_set_alias(struct net_device *dev, const char *alias, size_t len)
1526 {
1527 struct dev_ifalias *new_alias = NULL;
1528
1529 if (len >= IFALIASZ)
1530 return -EINVAL;
1531
1532 if (len) {
1533 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1534 if (!new_alias)
1535 return -ENOMEM;
1536
1537 memcpy(new_alias->ifalias, alias, len);
1538 new_alias->ifalias[len] = 0;
1539 }
1540
1541 mutex_lock(&ifalias_mutex);
1542 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1543 mutex_is_locked(&ifalias_mutex));
1544 mutex_unlock(&ifalias_mutex);
1545
1546 if (new_alias)
1547 kfree_rcu(new_alias, rcuhead);
1548
1549 return len;
1550 }
1551
1552 /**
1553 * dev_get_alias - get ifalias of a device
1554 * @dev: device
1555 * @name: buffer to store name of ifalias
1556 * @len: size of buffer
1557 *
1558 * get ifalias for a device. Caller must make sure dev cannot go
1559 * away, e.g. rcu read lock or own a reference count to device.
1560 */
1561 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1562 {
1563 const struct dev_ifalias *alias;
1564 int ret = 0;
1565
1566 rcu_read_lock();
1567 alias = rcu_dereference(dev->ifalias);
1568 if (alias)
1569 ret = snprintf(name, len, "%s", alias->ifalias);
1570 rcu_read_unlock();
1571
1572 return ret;
1573 }
1574
1575 /**
1576 * netdev_features_change - device changes features
1577 * @dev: device to cause notification
1578 *
1579 * Called to indicate a device has changed features.
1580 */
1581 void netdev_features_change(struct net_device *dev)
1582 {
1583 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1584 }
1585 EXPORT_SYMBOL(netdev_features_change);
1586
1587 void netif_state_change(struct net_device *dev)
1588 {
1589 netdev_ops_assert_locked_or_invisible(dev);
1590
1591 if (dev->flags & IFF_UP) {
1592 struct netdev_notifier_change_info change_info = {
1593 .info.dev = dev,
1594 };
1595
1596 call_netdevice_notifiers_info(NETDEV_CHANGE,
1597 &change_info.info);
1598 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1599 }
1600 }
1601
1602 /**
1603 * __netdev_notify_peers - notify network peers about existence of @dev,
1604 * to be called when rtnl lock is already held.
1605 * @dev: network device
1606 *
1607 * Generate traffic such that interested network peers are aware of
1608 * @dev, such as by generating a gratuitous ARP. This may be used when
1609 * a device wants to inform the rest of the network about some sort of
1610 * reconfiguration such as a failover event or virtual machine
1611 * migration.
1612 */
1613 void __netdev_notify_peers(struct net_device *dev)
1614 {
1615 ASSERT_RTNL();
1616 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1617 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1618 }
1619 EXPORT_SYMBOL(__netdev_notify_peers);
1620
1621 /**
1622 * netdev_notify_peers - notify network peers about existence of @dev
1623 * @dev: network device
1624 *
1625 * Generate traffic such that interested network peers are aware of
1626 * @dev, such as by generating a gratuitous ARP. This may be used when
1627 * a device wants to inform the rest of the network about some sort of
1628 * reconfiguration such as a failover event or virtual machine
1629 * migration.
1630 */
1631 void netdev_notify_peers(struct net_device *dev)
1632 {
1633 rtnl_lock();
1634 __netdev_notify_peers(dev);
1635 rtnl_unlock();
1636 }
1637 EXPORT_SYMBOL(netdev_notify_peers);
1638
1639 static int napi_threaded_poll(void *data);
1640
1641 static int napi_kthread_create(struct napi_struct *n)
1642 {
1643 int err = 0;
1644
1645 /* Create and wake up the kthread once to put it in
1646 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1647 * warning and work with loadavg.
1648 */
1649 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1650 n->dev->name, n->napi_id);
1651 if (IS_ERR(n->thread)) {
1652 err = PTR_ERR(n->thread);
1653 pr_err("kthread_run failed with err %d\n", err);
1654 n->thread = NULL;
1655 }
1656
1657 return err;
1658 }
1659
1660 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1661 {
1662 const struct net_device_ops *ops = dev->netdev_ops;
1663 int ret;
1664
1665 ASSERT_RTNL();
1666 dev_addr_check(dev);
1667
1668 if (!netif_device_present(dev)) {
1669 /* may be detached because parent is runtime-suspended */
1670 if (dev->dev.parent)
1671 pm_runtime_resume(dev->dev.parent);
1672 if (!netif_device_present(dev))
1673 return -ENODEV;
1674 }
1675
1676 /* Block netpoll from trying to do any rx path servicing.
1677 * If we don't do this there is a chance ndo_poll_controller
1678 * or ndo_poll may be running while we open the device
1679 */
1680 netpoll_poll_disable(dev);
1681
1682 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1683 ret = notifier_to_errno(ret);
1684 if (ret)
1685 return ret;
1686
1687 set_bit(__LINK_STATE_START, &dev->state);
1688
1689 netdev_ops_assert_locked(dev);
1690
1691 if (ops->ndo_validate_addr)
1692 ret = ops->ndo_validate_addr(dev);
1693
1694 if (!ret && ops->ndo_open)
1695 ret = ops->ndo_open(dev);
1696
1697 netpoll_poll_enable(dev);
1698
1699 if (ret)
1700 clear_bit(__LINK_STATE_START, &dev->state);
1701 else {
1702 netif_set_up(dev, true);
1703 dev_set_rx_mode(dev);
1704 dev_activate(dev);
1705 add_device_randomness(dev->dev_addr, dev->addr_len);
1706 }
1707
1708 return ret;
1709 }
1710
1711 int netif_open(struct net_device *dev, struct netlink_ext_ack *extack)
1712 {
1713 int ret;
1714
1715 if (dev->flags & IFF_UP)
1716 return 0;
1717
1718 ret = __dev_open(dev, extack);
1719 if (ret < 0)
1720 return ret;
1721
1722 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1723 call_netdevice_notifiers(NETDEV_UP, dev);
1724
1725 return ret;
1726 }
1727 EXPORT_SYMBOL(netif_open);
1728
1729 static void __dev_close_many(struct list_head *head)
1730 {
1731 struct net_device *dev;
1732
1733 ASSERT_RTNL();
1734 might_sleep();
1735
1736 list_for_each_entry(dev, head, close_list) {
1737 /* Temporarily disable netpoll until the interface is down */
1738 netpoll_poll_disable(dev);
1739
1740 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1741
1742 clear_bit(__LINK_STATE_START, &dev->state);
1743
1744 /* Synchronize to scheduled poll. We cannot touch poll list, it
1745 * can be even on different cpu. So just clear netif_running().
1746 *
1747 * dev->stop() will invoke napi_disable() on all of it's
1748 * napi_struct instances on this device.
1749 */
1750 smp_mb__after_atomic(); /* Commit netif_running(). */
1751 }
1752
1753 dev_deactivate_many(head);
1754
1755 list_for_each_entry(dev, head, close_list) {
1756 const struct net_device_ops *ops = dev->netdev_ops;
1757
1758 /*
1759 * Call the device specific close. This cannot fail.
1760 * Only if device is UP
1761 *
1762 * We allow it to be called even after a DETACH hot-plug
1763 * event.
1764 */
1765
1766 netdev_ops_assert_locked(dev);
1767
1768 if (ops->ndo_stop)
1769 ops->ndo_stop(dev);
1770
1771 netif_set_up(dev, false);
1772 netpoll_poll_enable(dev);
1773 }
1774 }
1775
1776 static void __dev_close(struct net_device *dev)
1777 {
1778 LIST_HEAD(single);
1779
1780 list_add(&dev->close_list, &single);
1781 __dev_close_many(&single);
1782 list_del(&single);
1783 }
1784
1785 void netif_close_many(struct list_head *head, bool unlink)
1786 {
1787 struct net_device *dev, *tmp;
1788
1789 /* Remove the devices that don't need to be closed */
1790 list_for_each_entry_safe(dev, tmp, head, close_list)
1791 if (!(dev->flags & IFF_UP))
1792 list_del_init(&dev->close_list);
1793
1794 __dev_close_many(head);
1795
1796 list_for_each_entry_safe(dev, tmp, head, close_list) {
1797 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1798 call_netdevice_notifiers(NETDEV_DOWN, dev);
1799 if (unlink)
1800 list_del_init(&dev->close_list);
1801 }
1802 }
1803 EXPORT_SYMBOL_NS_GPL(netif_close_many, "NETDEV_INTERNAL");
1804
1805 void netif_close(struct net_device *dev)
1806 {
1807 if (dev->flags & IFF_UP) {
1808 LIST_HEAD(single);
1809
1810 list_add(&dev->close_list, &single);
1811 netif_close_many(&single, true);
1812 list_del(&single);
1813 }
1814 }
1815 EXPORT_SYMBOL(netif_close);
1816
1817 void netif_disable_lro(struct net_device *dev)
1818 {
1819 struct net_device *lower_dev;
1820 struct list_head *iter;
1821
1822 dev->wanted_features &= ~NETIF_F_LRO;
1823 netdev_update_features(dev);
1824
1825 if (unlikely(dev->features & NETIF_F_LRO))
1826 netdev_WARN(dev, "failed to disable LRO!\n");
1827
1828 netdev_for_each_lower_dev(dev, lower_dev, iter) {
1829 netdev_lock_ops(lower_dev);
1830 netif_disable_lro(lower_dev);
1831 netdev_unlock_ops(lower_dev);
1832 }
1833 }
1834 EXPORT_IPV6_MOD(netif_disable_lro);
1835
1836 /**
1837 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1838 * @dev: device
1839 *
1840 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1841 * called under RTNL. This is needed if Generic XDP is installed on
1842 * the device.
1843 */
1844 static void dev_disable_gro_hw(struct net_device *dev)
1845 {
1846 dev->wanted_features &= ~NETIF_F_GRO_HW;
1847 netdev_update_features(dev);
1848
1849 if (unlikely(dev->features & NETIF_F_GRO_HW))
1850 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1851 }
1852
1853 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1854 {
1855 #define N(val) \
1856 case NETDEV_##val: \
1857 return "NETDEV_" __stringify(val);
1858 switch (cmd) {
1859 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1860 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1861 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1862 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1863 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1864 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1865 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1866 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1867 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1868 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1869 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1870 N(XDP_FEAT_CHANGE)
1871 }
1872 #undef N
1873 return "UNKNOWN_NETDEV_EVENT";
1874 }
1875 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1876
1877 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1878 struct net_device *dev)
1879 {
1880 struct netdev_notifier_info info = {
1881 .dev = dev,
1882 };
1883
1884 return nb->notifier_call(nb, val, &info);
1885 }
1886
1887 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1888 struct net_device *dev)
1889 {
1890 int err;
1891
1892 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1893 err = notifier_to_errno(err);
1894 if (err)
1895 return err;
1896
1897 if (!(dev->flags & IFF_UP))
1898 return 0;
1899
1900 call_netdevice_notifier(nb, NETDEV_UP, dev);
1901 return 0;
1902 }
1903
1904 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1905 struct net_device *dev)
1906 {
1907 if (dev->flags & IFF_UP) {
1908 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1909 dev);
1910 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1911 }
1912 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1913 }
1914
1915 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1916 struct net *net)
1917 {
1918 struct net_device *dev;
1919 int err;
1920
1921 for_each_netdev(net, dev) {
1922 netdev_lock_ops(dev);
1923 err = call_netdevice_register_notifiers(nb, dev);
1924 netdev_unlock_ops(dev);
1925 if (err)
1926 goto rollback;
1927 }
1928 return 0;
1929
1930 rollback:
1931 for_each_netdev_continue_reverse(net, dev)
1932 call_netdevice_unregister_notifiers(nb, dev);
1933 return err;
1934 }
1935
1936 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1937 struct net *net)
1938 {
1939 struct net_device *dev;
1940
1941 for_each_netdev(net, dev)
1942 call_netdevice_unregister_notifiers(nb, dev);
1943 }
1944
1945 static int dev_boot_phase = 1;
1946
1947 /**
1948 * register_netdevice_notifier - register a network notifier block
1949 * @nb: notifier
1950 *
1951 * Register a notifier to be called when network device events occur.
1952 * The notifier passed is linked into the kernel structures and must
1953 * not be reused until it has been unregistered. A negative errno code
1954 * is returned on a failure.
1955 *
1956 * When registered all registration and up events are replayed
1957 * to the new notifier to allow device to have a race free
1958 * view of the network device list.
1959 */
1960
1961 int register_netdevice_notifier(struct notifier_block *nb)
1962 {
1963 struct net *net;
1964 int err;
1965
1966 /* Close race with setup_net() and cleanup_net() */
1967 down_write(&pernet_ops_rwsem);
1968
1969 /* When RTNL is removed, we need protection for netdev_chain. */
1970 rtnl_lock();
1971
1972 err = raw_notifier_chain_register(&netdev_chain, nb);
1973 if (err)
1974 goto unlock;
1975 if (dev_boot_phase)
1976 goto unlock;
1977 for_each_net(net) {
1978 __rtnl_net_lock(net);
1979 err = call_netdevice_register_net_notifiers(nb, net);
1980 __rtnl_net_unlock(net);
1981 if (err)
1982 goto rollback;
1983 }
1984
1985 unlock:
1986 rtnl_unlock();
1987 up_write(&pernet_ops_rwsem);
1988 return err;
1989
1990 rollback:
1991 for_each_net_continue_reverse(net) {
1992 __rtnl_net_lock(net);
1993 call_netdevice_unregister_net_notifiers(nb, net);
1994 __rtnl_net_unlock(net);
1995 }
1996
1997 raw_notifier_chain_unregister(&netdev_chain, nb);
1998 goto unlock;
1999 }
2000 EXPORT_SYMBOL(register_netdevice_notifier);
2001
2002 /**
2003 * unregister_netdevice_notifier - unregister a network notifier block
2004 * @nb: notifier
2005 *
2006 * Unregister a notifier previously registered by
2007 * register_netdevice_notifier(). The notifier is unlinked into the
2008 * kernel structures and may then be reused. A negative errno code
2009 * is returned on a failure.
2010 *
2011 * After unregistering unregister and down device events are synthesized
2012 * for all devices on the device list to the removed notifier to remove
2013 * the need for special case cleanup code.
2014 */
2015
2016 int unregister_netdevice_notifier(struct notifier_block *nb)
2017 {
2018 struct net *net;
2019 int err;
2020
2021 /* Close race with setup_net() and cleanup_net() */
2022 down_write(&pernet_ops_rwsem);
2023 rtnl_lock();
2024 err = raw_notifier_chain_unregister(&netdev_chain, nb);
2025 if (err)
2026 goto unlock;
2027
2028 for_each_net(net) {
2029 __rtnl_net_lock(net);
2030 call_netdevice_unregister_net_notifiers(nb, net);
2031 __rtnl_net_unlock(net);
2032 }
2033
2034 unlock:
2035 rtnl_unlock();
2036 up_write(&pernet_ops_rwsem);
2037 return err;
2038 }
2039 EXPORT_SYMBOL(unregister_netdevice_notifier);
2040
2041 static int __register_netdevice_notifier_net(struct net *net,
2042 struct notifier_block *nb,
2043 bool ignore_call_fail)
2044 {
2045 int err;
2046
2047 err = raw_notifier_chain_register(&net->netdev_chain, nb);
2048 if (err)
2049 return err;
2050 if (dev_boot_phase)
2051 return 0;
2052
2053 err = call_netdevice_register_net_notifiers(nb, net);
2054 if (err && !ignore_call_fail)
2055 goto chain_unregister;
2056
2057 return 0;
2058
2059 chain_unregister:
2060 raw_notifier_chain_unregister(&net->netdev_chain, nb);
2061 return err;
2062 }
2063
2064 static int __unregister_netdevice_notifier_net(struct net *net,
2065 struct notifier_block *nb)
2066 {
2067 int err;
2068
2069 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
2070 if (err)
2071 return err;
2072
2073 call_netdevice_unregister_net_notifiers(nb, net);
2074 return 0;
2075 }
2076
2077 /**
2078 * register_netdevice_notifier_net - register a per-netns network notifier block
2079 * @net: network namespace
2080 * @nb: notifier
2081 *
2082 * Register a notifier to be called when network device events occur.
2083 * The notifier passed is linked into the kernel structures and must
2084 * not be reused until it has been unregistered. A negative errno code
2085 * is returned on a failure.
2086 *
2087 * When registered all registration and up events are replayed
2088 * to the new notifier to allow device to have a race free
2089 * view of the network device list.
2090 */
2091
2092 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
2093 {
2094 int err;
2095
2096 rtnl_net_lock(net);
2097 err = __register_netdevice_notifier_net(net, nb, false);
2098 rtnl_net_unlock(net);
2099
2100 return err;
2101 }
2102 EXPORT_SYMBOL(register_netdevice_notifier_net);
2103
2104 /**
2105 * unregister_netdevice_notifier_net - unregister a per-netns
2106 * network notifier block
2107 * @net: network namespace
2108 * @nb: notifier
2109 *
2110 * Unregister a notifier previously registered by
2111 * register_netdevice_notifier_net(). The notifier is unlinked from the
2112 * kernel structures and may then be reused. A negative errno code
2113 * is returned on a failure.
2114 *
2115 * After unregistering unregister and down device events are synthesized
2116 * for all devices on the device list to the removed notifier to remove
2117 * the need for special case cleanup code.
2118 */
2119
2120 int unregister_netdevice_notifier_net(struct net *net,
2121 struct notifier_block *nb)
2122 {
2123 int err;
2124
2125 rtnl_net_lock(net);
2126 err = __unregister_netdevice_notifier_net(net, nb);
2127 rtnl_net_unlock(net);
2128
2129 return err;
2130 }
2131 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
2132
2133 static void __move_netdevice_notifier_net(struct net *src_net,
2134 struct net *dst_net,
2135 struct notifier_block *nb)
2136 {
2137 __unregister_netdevice_notifier_net(src_net, nb);
2138 __register_netdevice_notifier_net(dst_net, nb, true);
2139 }
2140
2141 static void rtnl_net_dev_lock(struct net_device *dev)
2142 {
2143 bool again;
2144
2145 do {
2146 struct net *net;
2147
2148 again = false;
2149
2150 /* netns might be being dismantled. */
2151 rcu_read_lock();
2152 net = dev_net_rcu(dev);
2153 net_passive_inc(net);
2154 rcu_read_unlock();
2155
2156 rtnl_net_lock(net);
2157
2158 #ifdef CONFIG_NET_NS
2159 /* dev might have been moved to another netns. */
2160 if (!net_eq(net, rcu_access_pointer(dev->nd_net.net))) {
2161 rtnl_net_unlock(net);
2162 net_passive_dec(net);
2163 again = true;
2164 }
2165 #endif
2166 } while (again);
2167 }
2168
2169 static void rtnl_net_dev_unlock(struct net_device *dev)
2170 {
2171 struct net *net = dev_net(dev);
2172
2173 rtnl_net_unlock(net);
2174 net_passive_dec(net);
2175 }
2176
2177 int register_netdevice_notifier_dev_net(struct net_device *dev,
2178 struct notifier_block *nb,
2179 struct netdev_net_notifier *nn)
2180 {
2181 int err;
2182
2183 rtnl_net_dev_lock(dev);
2184 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
2185 if (!err) {
2186 nn->nb = nb;
2187 list_add(&nn->list, &dev->net_notifier_list);
2188 }
2189 rtnl_net_dev_unlock(dev);
2190
2191 return err;
2192 }
2193 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
2194
2195 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
2196 struct notifier_block *nb,
2197 struct netdev_net_notifier *nn)
2198 {
2199 int err;
2200
2201 rtnl_net_dev_lock(dev);
2202 list_del(&nn->list);
2203 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
2204 rtnl_net_dev_unlock(dev);
2205
2206 return err;
2207 }
2208 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
2209
2210 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
2211 struct net *net)
2212 {
2213 struct netdev_net_notifier *nn;
2214
2215 list_for_each_entry(nn, &dev->net_notifier_list, list)
2216 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
2217 }
2218
2219 /**
2220 * call_netdevice_notifiers_info - call all network notifier blocks
2221 * @val: value passed unmodified to notifier function
2222 * @info: notifier information data
2223 *
2224 * Call all network notifier blocks. Parameters and return value
2225 * are as for raw_notifier_call_chain().
2226 */
2227
2228 int call_netdevice_notifiers_info(unsigned long val,
2229 struct netdev_notifier_info *info)
2230 {
2231 struct net *net = dev_net(info->dev);
2232 int ret;
2233
2234 ASSERT_RTNL();
2235
2236 /* Run per-netns notifier block chain first, then run the global one.
2237 * Hopefully, one day, the global one is going to be removed after
2238 * all notifier block registrators get converted to be per-netns.
2239 */
2240 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2241 if (ret & NOTIFY_STOP_MASK)
2242 return ret;
2243 return raw_notifier_call_chain(&netdev_chain, val, info);
2244 }
2245
2246 /**
2247 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
2248 * for and rollback on error
2249 * @val_up: value passed unmodified to notifier function
2250 * @val_down: value passed unmodified to the notifier function when
2251 * recovering from an error on @val_up
2252 * @info: notifier information data
2253 *
2254 * Call all per-netns network notifier blocks, but not notifier blocks on
2255 * the global notifier chain. Parameters and return value are as for
2256 * raw_notifier_call_chain_robust().
2257 */
2258
2259 static int
2260 call_netdevice_notifiers_info_robust(unsigned long val_up,
2261 unsigned long val_down,
2262 struct netdev_notifier_info *info)
2263 {
2264 struct net *net = dev_net(info->dev);
2265
2266 ASSERT_RTNL();
2267
2268 return raw_notifier_call_chain_robust(&net->netdev_chain,
2269 val_up, val_down, info);
2270 }
2271
2272 static int call_netdevice_notifiers_extack(unsigned long val,
2273 struct net_device *dev,
2274 struct netlink_ext_ack *extack)
2275 {
2276 struct netdev_notifier_info info = {
2277 .dev = dev,
2278 .extack = extack,
2279 };
2280
2281 return call_netdevice_notifiers_info(val, &info);
2282 }
2283
2284 /**
2285 * call_netdevice_notifiers - call all network notifier blocks
2286 * @val: value passed unmodified to notifier function
2287 * @dev: net_device pointer passed unmodified to notifier function
2288 *
2289 * Call all network notifier blocks. Parameters and return value
2290 * are as for raw_notifier_call_chain().
2291 */
2292
2293 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2294 {
2295 return call_netdevice_notifiers_extack(val, dev, NULL);
2296 }
2297 EXPORT_SYMBOL(call_netdevice_notifiers);
2298
2299 /**
2300 * call_netdevice_notifiers_mtu - call all network notifier blocks
2301 * @val: value passed unmodified to notifier function
2302 * @dev: net_device pointer passed unmodified to notifier function
2303 * @arg: additional u32 argument passed to the notifier function
2304 *
2305 * Call all network notifier blocks. Parameters and return value
2306 * are as for raw_notifier_call_chain().
2307 */
2308 static int call_netdevice_notifiers_mtu(unsigned long val,
2309 struct net_device *dev, u32 arg)
2310 {
2311 struct netdev_notifier_info_ext info = {
2312 .info.dev = dev,
2313 .ext.mtu = arg,
2314 };
2315
2316 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2317
2318 return call_netdevice_notifiers_info(val, &info.info);
2319 }
2320
2321 #ifdef CONFIG_NET_INGRESS
2322 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2323
2324 void net_inc_ingress_queue(void)
2325 {
2326 static_branch_inc(&ingress_needed_key);
2327 }
2328 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2329
2330 void net_dec_ingress_queue(void)
2331 {
2332 static_branch_dec(&ingress_needed_key);
2333 }
2334 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2335 #endif
2336
2337 #ifdef CONFIG_NET_EGRESS
2338 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2339
2340 void net_inc_egress_queue(void)
2341 {
2342 static_branch_inc(&egress_needed_key);
2343 }
2344 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2345
2346 void net_dec_egress_queue(void)
2347 {
2348 static_branch_dec(&egress_needed_key);
2349 }
2350 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2351 #endif
2352
2353 #ifdef CONFIG_NET_CLS_ACT
2354 DEFINE_STATIC_KEY_FALSE(tcf_sw_enabled_key);
2355 EXPORT_SYMBOL(tcf_sw_enabled_key);
2356 #endif
2357
2358 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2359 EXPORT_SYMBOL(netstamp_needed_key);
2360 #ifdef CONFIG_JUMP_LABEL
2361 static atomic_t netstamp_needed_deferred;
2362 static atomic_t netstamp_wanted;
2363 static void netstamp_clear(struct work_struct *work)
2364 {
2365 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2366 int wanted;
2367
2368 wanted = atomic_add_return(deferred, &netstamp_wanted);
2369 if (wanted > 0)
2370 static_branch_enable(&netstamp_needed_key);
2371 else
2372 static_branch_disable(&netstamp_needed_key);
2373 }
2374 static DECLARE_WORK(netstamp_work, netstamp_clear);
2375 #endif
2376
2377 void net_enable_timestamp(void)
2378 {
2379 #ifdef CONFIG_JUMP_LABEL
2380 int wanted = atomic_read(&netstamp_wanted);
2381
2382 while (wanted > 0) {
2383 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2384 return;
2385 }
2386 atomic_inc(&netstamp_needed_deferred);
2387 schedule_work(&netstamp_work);
2388 #else
2389 static_branch_inc(&netstamp_needed_key);
2390 #endif
2391 }
2392 EXPORT_SYMBOL(net_enable_timestamp);
2393
2394 void net_disable_timestamp(void)
2395 {
2396 #ifdef CONFIG_JUMP_LABEL
2397 int wanted = atomic_read(&netstamp_wanted);
2398
2399 while (wanted > 1) {
2400 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2401 return;
2402 }
2403 atomic_dec(&netstamp_needed_deferred);
2404 schedule_work(&netstamp_work);
2405 #else
2406 static_branch_dec(&netstamp_needed_key);
2407 #endif
2408 }
2409 EXPORT_SYMBOL(net_disable_timestamp);
2410
2411 static inline void net_timestamp_set(struct sk_buff *skb)
2412 {
2413 skb->tstamp = 0;
2414 skb->tstamp_type = SKB_CLOCK_REALTIME;
2415 if (static_branch_unlikely(&netstamp_needed_key))
2416 skb->tstamp = ktime_get_real();
2417 }
2418
2419 #define net_timestamp_check(COND, SKB) \
2420 if (static_branch_unlikely(&netstamp_needed_key)) { \
2421 if ((COND) && !(SKB)->tstamp) \
2422 (SKB)->tstamp = ktime_get_real(); \
2423 } \
2424
2425 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2426 {
2427 return __is_skb_forwardable(dev, skb, true);
2428 }
2429 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2430
2431 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2432 bool check_mtu)
2433 {
2434 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2435
2436 if (likely(!ret)) {
2437 skb->protocol = eth_type_trans(skb, dev);
2438 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2439 }
2440
2441 return ret;
2442 }
2443
2444 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2445 {
2446 return __dev_forward_skb2(dev, skb, true);
2447 }
2448 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2449
2450 /**
2451 * dev_forward_skb - loopback an skb to another netif
2452 *
2453 * @dev: destination network device
2454 * @skb: buffer to forward
2455 *
2456 * return values:
2457 * NET_RX_SUCCESS (no congestion)
2458 * NET_RX_DROP (packet was dropped, but freed)
2459 *
2460 * dev_forward_skb can be used for injecting an skb from the
2461 * start_xmit function of one device into the receive queue
2462 * of another device.
2463 *
2464 * The receiving device may be in another namespace, so
2465 * we have to clear all information in the skb that could
2466 * impact namespace isolation.
2467 */
2468 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2469 {
2470 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2471 }
2472 EXPORT_SYMBOL_GPL(dev_forward_skb);
2473
2474 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2475 {
2476 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2477 }
2478
2479 static inline int deliver_skb(struct sk_buff *skb,
2480 struct packet_type *pt_prev,
2481 struct net_device *orig_dev)
2482 {
2483 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2484 return -ENOMEM;
2485 refcount_inc(&skb->users);
2486 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2487 }
2488
2489 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2490 struct packet_type **pt,
2491 struct net_device *orig_dev,
2492 __be16 type,
2493 struct list_head *ptype_list)
2494 {
2495 struct packet_type *ptype, *pt_prev = *pt;
2496
2497 list_for_each_entry_rcu(ptype, ptype_list, list) {
2498 if (ptype->type != type)
2499 continue;
2500 if (pt_prev)
2501 deliver_skb(skb, pt_prev, orig_dev);
2502 pt_prev = ptype;
2503 }
2504 *pt = pt_prev;
2505 }
2506
2507 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2508 {
2509 if (!ptype->af_packet_priv || !skb->sk)
2510 return false;
2511
2512 if (ptype->id_match)
2513 return ptype->id_match(ptype, skb->sk);
2514 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2515 return true;
2516
2517 return false;
2518 }
2519
2520 /**
2521 * dev_nit_active_rcu - return true if any network interface taps are in use
2522 *
2523 * The caller must hold the RCU lock
2524 *
2525 * @dev: network device to check for the presence of taps
2526 */
2527 bool dev_nit_active_rcu(const struct net_device *dev)
2528 {
2529 /* Callers may hold either RCU or RCU BH lock */
2530 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
2531
2532 return !list_empty(&dev_net(dev)->ptype_all) ||
2533 !list_empty(&dev->ptype_all);
2534 }
2535 EXPORT_SYMBOL_GPL(dev_nit_active_rcu);
2536
2537 /*
2538 * Support routine. Sends outgoing frames to any network
2539 * taps currently in use.
2540 */
2541
2542 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2543 {
2544 struct packet_type *ptype, *pt_prev = NULL;
2545 struct list_head *ptype_list;
2546 struct sk_buff *skb2 = NULL;
2547
2548 rcu_read_lock();
2549 ptype_list = &dev_net_rcu(dev)->ptype_all;
2550 again:
2551 list_for_each_entry_rcu(ptype, ptype_list, list) {
2552 if (READ_ONCE(ptype->ignore_outgoing))
2553 continue;
2554
2555 /* Never send packets back to the socket
2556 * they originated from - MvS ([email protected])
2557 */
2558 if (skb_loop_sk(ptype, skb))
2559 continue;
2560
2561 if (pt_prev) {
2562 deliver_skb(skb2, pt_prev, skb->dev);
2563 pt_prev = ptype;
2564 continue;
2565 }
2566
2567 /* need to clone skb, done only once */
2568 skb2 = skb_clone(skb, GFP_ATOMIC);
2569 if (!skb2)
2570 goto out_unlock;
2571
2572 net_timestamp_set(skb2);
2573
2574 /* skb->nh should be correctly
2575 * set by sender, so that the second statement is
2576 * just protection against buggy protocols.
2577 */
2578 skb_reset_mac_header(skb2);
2579
2580 if (skb_network_header(skb2) < skb2->data ||
2581 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2582 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2583 ntohs(skb2->protocol),
2584 dev->name);
2585 skb_reset_network_header(skb2);
2586 }
2587
2588 skb2->transport_header = skb2->network_header;
2589 skb2->pkt_type = PACKET_OUTGOING;
2590 pt_prev = ptype;
2591 }
2592
2593 if (ptype_list != &dev->ptype_all) {
2594 ptype_list = &dev->ptype_all;
2595 goto again;
2596 }
2597 out_unlock:
2598 if (pt_prev) {
2599 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2600 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2601 else
2602 kfree_skb(skb2);
2603 }
2604 rcu_read_unlock();
2605 }
2606 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2607
2608 /**
2609 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2610 * @dev: Network device
2611 * @txq: number of queues available
2612 *
2613 * If real_num_tx_queues is changed the tc mappings may no longer be
2614 * valid. To resolve this verify the tc mapping remains valid and if
2615 * not NULL the mapping. With no priorities mapping to this
2616 * offset/count pair it will no longer be used. In the worst case TC0
2617 * is invalid nothing can be done so disable priority mappings. If is
2618 * expected that drivers will fix this mapping if they can before
2619 * calling netif_set_real_num_tx_queues.
2620 */
2621 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2622 {
2623 int i;
2624 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2625
2626 /* If TC0 is invalidated disable TC mapping */
2627 if (tc->offset + tc->count > txq) {
2628 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2629 dev->num_tc = 0;
2630 return;
2631 }
2632
2633 /* Invalidated prio to tc mappings set to TC0 */
2634 for (i = 1; i < TC_BITMASK + 1; i++) {
2635 int q = netdev_get_prio_tc_map(dev, i);
2636
2637 tc = &dev->tc_to_txq[q];
2638 if (tc->offset + tc->count > txq) {
2639 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",
2640 i, q);
2641 netdev_set_prio_tc_map(dev, i, 0);
2642 }
2643 }
2644 }
2645
2646 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2647 {
2648 if (dev->num_tc) {
2649 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2650 int i;
2651
2652 /* walk through the TCs and see if it falls into any of them */
2653 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2654 if ((txq - tc->offset) < tc->count)
2655 return i;
2656 }
2657
2658 /* didn't find it, just return -1 to indicate no match */
2659 return -1;
2660 }
2661
2662 return 0;
2663 }
2664 EXPORT_SYMBOL(netdev_txq_to_tc);
2665
2666 #ifdef CONFIG_XPS
2667 static struct static_key xps_needed __read_mostly;
2668 static struct static_key xps_rxqs_needed __read_mostly;
2669 static DEFINE_MUTEX(xps_map_mutex);
2670 #define xmap_dereference(P) \
2671 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2672
2673 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2674 struct xps_dev_maps *old_maps, int tci, u16 index)
2675 {
2676 struct xps_map *map = NULL;
2677 int pos;
2678
2679 map = xmap_dereference(dev_maps->attr_map[tci]);
2680 if (!map)
2681 return false;
2682
2683 for (pos = map->len; pos--;) {
2684 if (map->queues[pos] != index)
2685 continue;
2686
2687 if (map->len > 1) {
2688 map->queues[pos] = map->queues[--map->len];
2689 break;
2690 }
2691
2692 if (old_maps)
2693 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2694 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2695 kfree_rcu(map, rcu);
2696 return false;
2697 }
2698
2699 return true;
2700 }
2701
2702 static bool remove_xps_queue_cpu(struct net_device *dev,
2703 struct xps_dev_maps *dev_maps,
2704 int cpu, u16 offset, u16 count)
2705 {
2706 int num_tc = dev_maps->num_tc;
2707 bool active = false;
2708 int tci;
2709
2710 for (tci = cpu * num_tc; num_tc--; tci++) {
2711 int i, j;
2712
2713 for (i = count, j = offset; i--; j++) {
2714 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2715 break;
2716 }
2717
2718 active |= i < 0;
2719 }
2720
2721 return active;
2722 }
2723
2724 static void reset_xps_maps(struct net_device *dev,
2725 struct xps_dev_maps *dev_maps,
2726 enum xps_map_type type)
2727 {
2728 static_key_slow_dec_cpuslocked(&xps_needed);
2729 if (type == XPS_RXQS)
2730 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2731
2732 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2733
2734 kfree_rcu(dev_maps, rcu);
2735 }
2736
2737 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2738 u16 offset, u16 count)
2739 {
2740 struct xps_dev_maps *dev_maps;
2741 bool active = false;
2742 int i, j;
2743
2744 dev_maps = xmap_dereference(dev->xps_maps[type]);
2745 if (!dev_maps)
2746 return;
2747
2748 for (j = 0; j < dev_maps->nr_ids; j++)
2749 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2750 if (!active)
2751 reset_xps_maps(dev, dev_maps, type);
2752
2753 if (type == XPS_CPUS) {
2754 for (i = offset + (count - 1); count--; i--)
2755 netdev_queue_numa_node_write(
2756 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2757 }
2758 }
2759
2760 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2761 u16 count)
2762 {
2763 if (!static_key_false(&xps_needed))
2764 return;
2765
2766 cpus_read_lock();
2767 mutex_lock(&xps_map_mutex);
2768
2769 if (static_key_false(&xps_rxqs_needed))
2770 clean_xps_maps(dev, XPS_RXQS, offset, count);
2771
2772 clean_xps_maps(dev, XPS_CPUS, offset, count);
2773
2774 mutex_unlock(&xps_map_mutex);
2775 cpus_read_unlock();
2776 }
2777
2778 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2779 {
2780 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2781 }
2782
2783 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2784 u16 index, bool is_rxqs_map)
2785 {
2786 struct xps_map *new_map;
2787 int alloc_len = XPS_MIN_MAP_ALLOC;
2788 int i, pos;
2789
2790 for (pos = 0; map && pos < map->len; pos++) {
2791 if (map->queues[pos] != index)
2792 continue;
2793 return map;
2794 }
2795
2796 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2797 if (map) {
2798 if (pos < map->alloc_len)
2799 return map;
2800
2801 alloc_len = map->alloc_len * 2;
2802 }
2803
2804 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2805 * map
2806 */
2807 if (is_rxqs_map)
2808 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2809 else
2810 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2811 cpu_to_node(attr_index));
2812 if (!new_map)
2813 return NULL;
2814
2815 for (i = 0; i < pos; i++)
2816 new_map->queues[i] = map->queues[i];
2817 new_map->alloc_len = alloc_len;
2818 new_map->len = pos;
2819
2820 return new_map;
2821 }
2822
2823 /* Copy xps maps at a given index */
2824 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2825 struct xps_dev_maps *new_dev_maps, int index,
2826 int tc, bool skip_tc)
2827 {
2828 int i, tci = index * dev_maps->num_tc;
2829 struct xps_map *map;
2830
2831 /* copy maps belonging to foreign traffic classes */
2832 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2833 if (i == tc && skip_tc)
2834 continue;
2835
2836 /* fill in the new device map from the old device map */
2837 map = xmap_dereference(dev_maps->attr_map[tci]);
2838 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2839 }
2840 }
2841
2842 /* Must be called under cpus_read_lock */
2843 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2844 u16 index, enum xps_map_type type)
2845 {
2846 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2847 const unsigned long *online_mask = NULL;
2848 bool active = false, copy = false;
2849 int i, j, tci, numa_node_id = -2;
2850 int maps_sz, num_tc = 1, tc = 0;
2851 struct xps_map *map, *new_map;
2852 unsigned int nr_ids;
2853
2854 WARN_ON_ONCE(index >= dev->num_tx_queues);
2855
2856 if (dev->num_tc) {
2857 /* Do not allow XPS on subordinate device directly */
2858 num_tc = dev->num_tc;
2859 if (num_tc < 0)
2860 return -EINVAL;
2861
2862 /* If queue belongs to subordinate dev use its map */
2863 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2864
2865 tc = netdev_txq_to_tc(dev, index);
2866 if (tc < 0)
2867 return -EINVAL;
2868 }
2869
2870 mutex_lock(&xps_map_mutex);
2871
2872 dev_maps = xmap_dereference(dev->xps_maps[type]);
2873 if (type == XPS_RXQS) {
2874 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2875 nr_ids = dev->num_rx_queues;
2876 } else {
2877 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2878 if (num_possible_cpus() > 1)
2879 online_mask = cpumask_bits(cpu_online_mask);
2880 nr_ids = nr_cpu_ids;
2881 }
2882
2883 if (maps_sz < L1_CACHE_BYTES)
2884 maps_sz = L1_CACHE_BYTES;
2885
2886 /* The old dev_maps could be larger or smaller than the one we're
2887 * setting up now, as dev->num_tc or nr_ids could have been updated in
2888 * between. We could try to be smart, but let's be safe instead and only
2889 * copy foreign traffic classes if the two map sizes match.
2890 */
2891 if (dev_maps &&
2892 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2893 copy = true;
2894
2895 /* allocate memory for queue storage */
2896 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2897 j < nr_ids;) {
2898 if (!new_dev_maps) {
2899 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2900 if (!new_dev_maps) {
2901 mutex_unlock(&xps_map_mutex);
2902 return -ENOMEM;
2903 }
2904
2905 new_dev_maps->nr_ids = nr_ids;
2906 new_dev_maps->num_tc = num_tc;
2907 }
2908
2909 tci = j * num_tc + tc;
2910 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2911
2912 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2913 if (!map)
2914 goto error;
2915
2916 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2917 }
2918
2919 if (!new_dev_maps)
2920 goto out_no_new_maps;
2921
2922 if (!dev_maps) {
2923 /* Increment static keys at most once per type */
2924 static_key_slow_inc_cpuslocked(&xps_needed);
2925 if (type == XPS_RXQS)
2926 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2927 }
2928
2929 for (j = 0; j < nr_ids; j++) {
2930 bool skip_tc = false;
2931
2932 tci = j * num_tc + tc;
2933 if (netif_attr_test_mask(j, mask, nr_ids) &&
2934 netif_attr_test_online(j, online_mask, nr_ids)) {
2935 /* add tx-queue to CPU/rx-queue maps */
2936 int pos = 0;
2937
2938 skip_tc = true;
2939
2940 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2941 while ((pos < map->len) && (map->queues[pos] != index))
2942 pos++;
2943
2944 if (pos == map->len)
2945 map->queues[map->len++] = index;
2946 #ifdef CONFIG_NUMA
2947 if (type == XPS_CPUS) {
2948 if (numa_node_id == -2)
2949 numa_node_id = cpu_to_node(j);
2950 else if (numa_node_id != cpu_to_node(j))
2951 numa_node_id = -1;
2952 }
2953 #endif
2954 }
2955
2956 if (copy)
2957 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2958 skip_tc);
2959 }
2960
2961 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2962
2963 /* Cleanup old maps */
2964 if (!dev_maps)
2965 goto out_no_old_maps;
2966
2967 for (j = 0; j < dev_maps->nr_ids; j++) {
2968 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2969 map = xmap_dereference(dev_maps->attr_map[tci]);
2970 if (!map)
2971 continue;
2972
2973 if (copy) {
2974 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2975 if (map == new_map)
2976 continue;
2977 }
2978
2979 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2980 kfree_rcu(map, rcu);
2981 }
2982 }
2983
2984 old_dev_maps = dev_maps;
2985
2986 out_no_old_maps:
2987 dev_maps = new_dev_maps;
2988 active = true;
2989
2990 out_no_new_maps:
2991 if (type == XPS_CPUS)
2992 /* update Tx queue numa node */
2993 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2994 (numa_node_id >= 0) ?
2995 numa_node_id : NUMA_NO_NODE);
2996
2997 if (!dev_maps)
2998 goto out_no_maps;
2999
3000 /* removes tx-queue from unused CPUs/rx-queues */
3001 for (j = 0; j < dev_maps->nr_ids; j++) {
3002 tci = j * dev_maps->num_tc;
3003
3004 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
3005 if (i == tc &&
3006 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
3007 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
3008 continue;
3009
3010 active |= remove_xps_queue(dev_maps,
3011 copy ? old_dev_maps : NULL,
3012 tci, index);
3013 }
3014 }
3015
3016 if (old_dev_maps)
3017 kfree_rcu(old_dev_maps, rcu);
3018
3019 /* free map if not active */
3020 if (!active)
3021 reset_xps_maps(dev, dev_maps, type);
3022
3023 out_no_maps:
3024 mutex_unlock(&xps_map_mutex);
3025
3026 return 0;
3027 error:
3028 /* remove any maps that we added */
3029 for (j = 0; j < nr_ids; j++) {
3030 for (i = num_tc, tci = j * num_tc; i--; tci++) {
3031 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
3032 map = copy ?
3033 xmap_dereference(dev_maps->attr_map[tci]) :
3034 NULL;
3035 if (new_map && new_map != map)
3036 kfree(new_map);
3037 }
3038 }
3039
3040 mutex_unlock(&xps_map_mutex);
3041
3042 kfree(new_dev_maps);
3043 return -ENOMEM;
3044 }
3045 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
3046
3047 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
3048 u16 index)
3049 {
3050 int ret;
3051
3052 cpus_read_lock();
3053 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
3054 cpus_read_unlock();
3055
3056 return ret;
3057 }
3058 EXPORT_SYMBOL(netif_set_xps_queue);
3059
3060 #endif
3061 static void netdev_unbind_all_sb_channels(struct net_device *dev)
3062 {
3063 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
3064
3065 /* Unbind any subordinate channels */
3066 while (txq-- != &dev->_tx[0]) {
3067 if (txq->sb_dev)
3068 netdev_unbind_sb_channel(dev, txq->sb_dev);
3069 }
3070 }
3071
3072 void netdev_reset_tc(struct net_device *dev)
3073 {
3074 #ifdef CONFIG_XPS
3075 netif_reset_xps_queues_gt(dev, 0);
3076 #endif
3077 netdev_unbind_all_sb_channels(dev);
3078
3079 /* Reset TC configuration of device */
3080 dev->num_tc = 0;
3081 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
3082 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
3083 }
3084 EXPORT_SYMBOL(netdev_reset_tc);
3085
3086 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
3087 {
3088 if (tc >= dev->num_tc)
3089 return -EINVAL;
3090
3091 #ifdef CONFIG_XPS
3092 netif_reset_xps_queues(dev, offset, count);
3093 #endif
3094 dev->tc_to_txq[tc].count = count;
3095 dev->tc_to_txq[tc].offset = offset;
3096 return 0;
3097 }
3098 EXPORT_SYMBOL(netdev_set_tc_queue);
3099
3100 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
3101 {
3102 if (num_tc > TC_MAX_QUEUE)
3103 return -EINVAL;
3104
3105 #ifdef CONFIG_XPS
3106 netif_reset_xps_queues_gt(dev, 0);
3107 #endif
3108 netdev_unbind_all_sb_channels(dev);
3109
3110 dev->num_tc = num_tc;
3111 return 0;
3112 }
3113 EXPORT_SYMBOL(netdev_set_num_tc);
3114
3115 void netdev_unbind_sb_channel(struct net_device *dev,
3116 struct net_device *sb_dev)
3117 {
3118 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
3119
3120 #ifdef CONFIG_XPS
3121 netif_reset_xps_queues_gt(sb_dev, 0);
3122 #endif
3123 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
3124 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
3125
3126 while (txq-- != &dev->_tx[0]) {
3127 if (txq->sb_dev == sb_dev)
3128 txq->sb_dev = NULL;
3129 }
3130 }
3131 EXPORT_SYMBOL(netdev_unbind_sb_channel);
3132
3133 int netdev_bind_sb_channel_queue(struct net_device *dev,
3134 struct net_device *sb_dev,
3135 u8 tc, u16 count, u16 offset)
3136 {
3137 /* Make certain the sb_dev and dev are already configured */
3138 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
3139 return -EINVAL;
3140
3141 /* We cannot hand out queues we don't have */
3142 if ((offset + count) > dev->real_num_tx_queues)
3143 return -EINVAL;
3144
3145 /* Record the mapping */
3146 sb_dev->tc_to_txq[tc].count = count;
3147 sb_dev->tc_to_txq[tc].offset = offset;
3148
3149 /* Provide a way for Tx queue to find the tc_to_txq map or
3150 * XPS map for itself.
3151 */
3152 while (count--)
3153 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
3154
3155 return 0;
3156 }
3157 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
3158
3159 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
3160 {
3161 /* Do not use a multiqueue device to represent a subordinate channel */
3162 if (netif_is_multiqueue(dev))
3163 return -ENODEV;
3164
3165 /* We allow channels 1 - 32767 to be used for subordinate channels.
3166 * Channel 0 is meant to be "native" mode and used only to represent
3167 * the main root device. We allow writing 0 to reset the device back
3168 * to normal mode after being used as a subordinate channel.
3169 */
3170 if (channel > S16_MAX)
3171 return -EINVAL;
3172
3173 dev->num_tc = -channel;
3174
3175 return 0;
3176 }
3177 EXPORT_SYMBOL(netdev_set_sb_channel);
3178
3179 /*
3180 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
3181 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
3182 */
3183 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
3184 {
3185 bool disabling;
3186 int rc;
3187
3188 disabling = txq < dev->real_num_tx_queues;
3189
3190 if (txq < 1 || txq > dev->num_tx_queues)
3191 return -EINVAL;
3192
3193 if (dev->reg_state == NETREG_REGISTERED ||
3194 dev->reg_state == NETREG_UNREGISTERING) {
3195 netdev_ops_assert_locked(dev);
3196
3197 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
3198 txq);
3199 if (rc)
3200 return rc;
3201
3202 if (dev->num_tc)
3203 netif_setup_tc(dev, txq);
3204
3205 net_shaper_set_real_num_tx_queues(dev, txq);
3206
3207 dev_qdisc_change_real_num_tx(dev, txq);
3208
3209 dev->real_num_tx_queues = txq;
3210
3211 if (disabling) {
3212 synchronize_net();
3213 qdisc_reset_all_tx_gt(dev, txq);
3214 #ifdef CONFIG_XPS
3215 netif_reset_xps_queues_gt(dev, txq);
3216 #endif
3217 }
3218 } else {
3219 dev->real_num_tx_queues = txq;
3220 }
3221
3222 return 0;
3223 }
3224 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
3225
3226 /**
3227 * netif_set_real_num_rx_queues - set actual number of RX queues used
3228 * @dev: Network device
3229 * @rxq: Actual number of RX queues
3230 *
3231 * This must be called either with the rtnl_lock held or before
3232 * registration of the net device. Returns 0 on success, or a
3233 * negative error code. If called before registration, it always
3234 * succeeds.
3235 */
3236 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
3237 {
3238 int rc;
3239
3240 if (rxq < 1 || rxq > dev->num_rx_queues)
3241 return -EINVAL;
3242
3243 if (dev->reg_state == NETREG_REGISTERED) {
3244 netdev_ops_assert_locked(dev);
3245
3246 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3247 rxq);
3248 if (rc)
3249 return rc;
3250 }
3251
3252 dev->real_num_rx_queues = rxq;
3253 return 0;
3254 }
3255 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3256
3257 /**
3258 * netif_set_real_num_queues - set actual number of RX and TX queues used
3259 * @dev: Network device
3260 * @txq: Actual number of TX queues
3261 * @rxq: Actual number of RX queues
3262 *
3263 * Set the real number of both TX and RX queues.
3264 * Does nothing if the number of queues is already correct.
3265 */
3266 int netif_set_real_num_queues(struct net_device *dev,
3267 unsigned int txq, unsigned int rxq)
3268 {
3269 unsigned int old_rxq = dev->real_num_rx_queues;
3270 int err;
3271
3272 if (txq < 1 || txq > dev->num_tx_queues ||
3273 rxq < 1 || rxq > dev->num_rx_queues)
3274 return -EINVAL;
3275
3276 /* Start from increases, so the error path only does decreases -
3277 * decreases can't fail.
3278 */
3279 if (rxq > dev->real_num_rx_queues) {
3280 err = netif_set_real_num_rx_queues(dev, rxq);
3281 if (err)
3282 return err;
3283 }
3284 if (txq > dev->real_num_tx_queues) {
3285 err = netif_set_real_num_tx_queues(dev, txq);
3286 if (err)
3287 goto undo_rx;
3288 }
3289 if (rxq < dev->real_num_rx_queues)
3290 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3291 if (txq < dev->real_num_tx_queues)
3292 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3293
3294 return 0;
3295 undo_rx:
3296 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3297 return err;
3298 }
3299 EXPORT_SYMBOL(netif_set_real_num_queues);
3300
3301 /**
3302 * netif_set_tso_max_size() - set the max size of TSO frames supported
3303 * @dev: netdev to update
3304 * @size: max skb->len of a TSO frame
3305 *
3306 * Set the limit on the size of TSO super-frames the device can handle.
3307 * Unless explicitly set the stack will assume the value of
3308 * %GSO_LEGACY_MAX_SIZE.
3309 */
3310 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3311 {
3312 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3313 if (size < READ_ONCE(dev->gso_max_size))
3314 netif_set_gso_max_size(dev, size);
3315 if (size < READ_ONCE(dev->gso_ipv4_max_size))
3316 netif_set_gso_ipv4_max_size(dev, size);
3317 }
3318 EXPORT_SYMBOL(netif_set_tso_max_size);
3319
3320 /**
3321 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3322 * @dev: netdev to update
3323 * @segs: max number of TCP segments
3324 *
3325 * Set the limit on the number of TCP segments the device can generate from
3326 * a single TSO super-frame.
3327 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3328 */
3329 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3330 {
3331 dev->tso_max_segs = segs;
3332 if (segs < READ_ONCE(dev->gso_max_segs))
3333 netif_set_gso_max_segs(dev, segs);
3334 }
3335 EXPORT_SYMBOL(netif_set_tso_max_segs);
3336
3337 /**
3338 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3339 * @to: netdev to update
3340 * @from: netdev from which to copy the limits
3341 */
3342 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3343 {
3344 netif_set_tso_max_size(to, from->tso_max_size);
3345 netif_set_tso_max_segs(to, from->tso_max_segs);
3346 }
3347 EXPORT_SYMBOL(netif_inherit_tso_max);
3348
3349 /**
3350 * netif_get_num_default_rss_queues - default number of RSS queues
3351 *
3352 * Default value is the number of physical cores if there are only 1 or 2, or
3353 * divided by 2 if there are more.
3354 */
3355 int netif_get_num_default_rss_queues(void)
3356 {
3357 cpumask_var_t cpus;
3358 int cpu, count = 0;
3359
3360 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3361 return 1;
3362
3363 cpumask_copy(cpus, cpu_online_mask);
3364 for_each_cpu(cpu, cpus) {
3365 ++count;
3366 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3367 }
3368 free_cpumask_var(cpus);
3369
3370 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3371 }
3372 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3373
3374 static void __netif_reschedule(struct Qdisc *q)
3375 {
3376 struct softnet_data *sd;
3377 unsigned long flags;
3378
3379 local_irq_save(flags);
3380 sd = this_cpu_ptr(&softnet_data);
3381 q->next_sched = NULL;
3382 *sd->output_queue_tailp = q;
3383 sd->output_queue_tailp = &q->next_sched;
3384 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3385 local_irq_restore(flags);
3386 }
3387
3388 void __netif_schedule(struct Qdisc *q)
3389 {
3390 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3391 __netif_reschedule(q);
3392 }
3393 EXPORT_SYMBOL(__netif_schedule);
3394
3395 struct dev_kfree_skb_cb {
3396 enum skb_drop_reason reason;
3397 };
3398
3399 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3400 {
3401 return (struct dev_kfree_skb_cb *)skb->cb;
3402 }
3403
3404 void netif_schedule_queue(struct netdev_queue *txq)
3405 {
3406 rcu_read_lock();
3407 if (!netif_xmit_stopped(txq)) {
3408 struct Qdisc *q = rcu_dereference(txq->qdisc);
3409
3410 __netif_schedule(q);
3411 }
3412 rcu_read_unlock();
3413 }
3414 EXPORT_SYMBOL(netif_schedule_queue);
3415
3416 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3417 {
3418 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3419 struct Qdisc *q;
3420
3421 rcu_read_lock();
3422 q = rcu_dereference(dev_queue->qdisc);
3423 __netif_schedule(q);
3424 rcu_read_unlock();
3425 }
3426 }
3427 EXPORT_SYMBOL(netif_tx_wake_queue);
3428
3429 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3430 {
3431 unsigned long flags;
3432
3433 if (unlikely(!skb))
3434 return;
3435
3436 if (likely(refcount_read(&skb->users) == 1)) {
3437 smp_rmb();
3438 refcount_set(&skb->users, 0);
3439 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3440 return;
3441 }
3442 get_kfree_skb_cb(skb)->reason = reason;
3443 local_irq_save(flags);
3444 skb->next = __this_cpu_read(softnet_data.completion_queue);
3445 __this_cpu_write(softnet_data.completion_queue, skb);
3446 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3447 local_irq_restore(flags);
3448 }
3449 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3450
3451 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3452 {
3453 if (in_hardirq() || irqs_disabled())
3454 dev_kfree_skb_irq_reason(skb, reason);
3455 else
3456 kfree_skb_reason(skb, reason);
3457 }
3458 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3459
3460
3461 /**
3462 * netif_device_detach - mark device as removed
3463 * @dev: network device
3464 *
3465 * Mark device as removed from system and therefore no longer available.
3466 */
3467 void netif_device_detach(struct net_device *dev)
3468 {
3469 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3470 netif_running(dev)) {
3471 netif_tx_stop_all_queues(dev);
3472 }
3473 }
3474 EXPORT_SYMBOL(netif_device_detach);
3475
3476 /**
3477 * netif_device_attach - mark device as attached
3478 * @dev: network device
3479 *
3480 * Mark device as attached from system and restart if needed.
3481 */
3482 void netif_device_attach(struct net_device *dev)
3483 {
3484 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3485 netif_running(dev)) {
3486 netif_tx_wake_all_queues(dev);
3487 netdev_watchdog_up(dev);
3488 }
3489 }
3490 EXPORT_SYMBOL(netif_device_attach);
3491
3492 /*
3493 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3494 * to be used as a distribution range.
3495 */
3496 static u16 skb_tx_hash(const struct net_device *dev,
3497 const struct net_device *sb_dev,
3498 struct sk_buff *skb)
3499 {
3500 u32 hash;
3501 u16 qoffset = 0;
3502 u16 qcount = dev->real_num_tx_queues;
3503
3504 if (dev->num_tc) {
3505 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3506
3507 qoffset = sb_dev->tc_to_txq[tc].offset;
3508 qcount = sb_dev->tc_to_txq[tc].count;
3509 if (unlikely(!qcount)) {
3510 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3511 sb_dev->name, qoffset, tc);
3512 qoffset = 0;
3513 qcount = dev->real_num_tx_queues;
3514 }
3515 }
3516
3517 if (skb_rx_queue_recorded(skb)) {
3518 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3519 hash = skb_get_rx_queue(skb);
3520 if (hash >= qoffset)
3521 hash -= qoffset;
3522 while (unlikely(hash >= qcount))
3523 hash -= qcount;
3524 return hash + qoffset;
3525 }
3526
3527 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3528 }
3529
3530 void skb_warn_bad_offload(const struct sk_buff *skb)
3531 {
3532 static const netdev_features_t null_features;
3533 struct net_device *dev = skb->dev;
3534 const char *name = "";
3535
3536 if (!net_ratelimit())
3537 return;
3538
3539 if (dev) {
3540 if (dev->dev.parent)
3541 name = dev_driver_string(dev->dev.parent);
3542 else
3543 name = netdev_name(dev);
3544 }
3545 skb_dump(KERN_WARNING, skb, false);
3546 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3547 name, dev ? &dev->features : &null_features,
3548 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3549 }
3550
3551 /*
3552 * Invalidate hardware checksum when packet is to be mangled, and
3553 * complete checksum manually on outgoing path.
3554 */
3555 int skb_checksum_help(struct sk_buff *skb)
3556 {
3557 __wsum csum;
3558 int ret = 0, offset;
3559
3560 if (skb->ip_summed == CHECKSUM_COMPLETE)
3561 goto out_set_summed;
3562
3563 if (unlikely(skb_is_gso(skb))) {
3564 skb_warn_bad_offload(skb);
3565 return -EINVAL;
3566 }
3567
3568 if (!skb_frags_readable(skb)) {
3569 return -EFAULT;
3570 }
3571
3572 /* Before computing a checksum, we should make sure no frag could
3573 * be modified by an external entity : checksum could be wrong.
3574 */
3575 if (skb_has_shared_frag(skb)) {
3576 ret = __skb_linearize(skb);
3577 if (ret)
3578 goto out;
3579 }
3580
3581 offset = skb_checksum_start_offset(skb);
3582 ret = -EINVAL;
3583 if (unlikely(offset >= skb_headlen(skb))) {
3584 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3585 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3586 offset, skb_headlen(skb));
3587 goto out;
3588 }
3589 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3590
3591 offset += skb->csum_offset;
3592 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3593 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3594 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3595 offset + sizeof(__sum16), skb_headlen(skb));
3596 goto out;
3597 }
3598 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3599 if (ret)
3600 goto out;
3601
3602 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3603 out_set_summed:
3604 skb->ip_summed = CHECKSUM_NONE;
3605 out:
3606 return ret;
3607 }
3608 EXPORT_SYMBOL(skb_checksum_help);
3609
3610 #ifdef CONFIG_NET_CRC32C
3611 int skb_crc32c_csum_help(struct sk_buff *skb)
3612 {
3613 u32 crc;
3614 int ret = 0, offset, start;
3615
3616 if (skb->ip_summed != CHECKSUM_PARTIAL)
3617 goto out;
3618
3619 if (unlikely(skb_is_gso(skb)))
3620 goto out;
3621
3622 /* Before computing a checksum, we should make sure no frag could
3623 * be modified by an external entity : checksum could be wrong.
3624 */
3625 if (unlikely(skb_has_shared_frag(skb))) {
3626 ret = __skb_linearize(skb);
3627 if (ret)
3628 goto out;
3629 }
3630 start = skb_checksum_start_offset(skb);
3631 offset = start + offsetof(struct sctphdr, checksum);
3632 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3633 ret = -EINVAL;
3634 goto out;
3635 }
3636
3637 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3638 if (ret)
3639 goto out;
3640
3641 crc = ~skb_crc32c(skb, start, skb->len - start, ~0);
3642 *(__le32 *)(skb->data + offset) = cpu_to_le32(crc);
3643 skb_reset_csum_not_inet(skb);
3644 out:
3645 return ret;
3646 }
3647 EXPORT_SYMBOL(skb_crc32c_csum_help);
3648 #endif /* CONFIG_NET_CRC32C */
3649
3650 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3651 {
3652 __be16 type = skb->protocol;
3653
3654 /* Tunnel gso handlers can set protocol to ethernet. */
3655 if (type == htons(ETH_P_TEB)) {
3656 struct ethhdr *eth;
3657
3658 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3659 return 0;
3660
3661 eth = (struct ethhdr *)skb->data;
3662 type = eth->h_proto;
3663 }
3664
3665 return vlan_get_protocol_and_depth(skb, type, depth);
3666 }
3667
3668
3669 /* Take action when hardware reception checksum errors are detected. */
3670 #ifdef CONFIG_BUG
3671 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3672 {
3673 netdev_err(dev, "hw csum failure\n");
3674 skb_dump(KERN_ERR, skb, true);
3675 dump_stack();
3676 }
3677
3678 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3679 {
3680 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3681 }
3682 EXPORT_SYMBOL(netdev_rx_csum_fault);
3683 #endif
3684
3685 /* XXX: check that highmem exists at all on the given machine. */
3686 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3687 {
3688 #ifdef CONFIG_HIGHMEM
3689 int i;
3690
3691 if (!(dev->features & NETIF_F_HIGHDMA)) {
3692 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3693 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3694 struct page *page = skb_frag_page(frag);
3695
3696 if (page && PageHighMem(page))
3697 return 1;
3698 }
3699 }
3700 #endif
3701 return 0;
3702 }
3703
3704 /* If MPLS offload request, verify we are testing hardware MPLS features
3705 * instead of standard features for the netdev.
3706 */
3707 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3708 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3709 netdev_features_t features,
3710 __be16 type)
3711 {
3712 if (eth_p_mpls(type))
3713 features &= skb->dev->mpls_features;
3714
3715 return features;
3716 }
3717 #else
3718 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3719 netdev_features_t features,
3720 __be16 type)
3721 {
3722 return features;
3723 }
3724 #endif
3725
3726 static netdev_features_t harmonize_features(struct sk_buff *skb,
3727 netdev_features_t features)
3728 {
3729 __be16 type;
3730
3731 type = skb_network_protocol(skb, NULL);
3732 features = net_mpls_features(skb, features, type);
3733
3734 if (skb->ip_summed != CHECKSUM_NONE &&
3735 !can_checksum_protocol(features, type)) {
3736 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3737 }
3738 if (illegal_highdma(skb->dev, skb))
3739 features &= ~NETIF_F_SG;
3740
3741 return features;
3742 }
3743
3744 netdev_features_t passthru_features_check(struct sk_buff *skb,
3745 struct net_device *dev,
3746 netdev_features_t features)
3747 {
3748 return features;
3749 }
3750 EXPORT_SYMBOL(passthru_features_check);
3751
3752 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3753 struct net_device *dev,
3754 netdev_features_t features)
3755 {
3756 return vlan_features_check(skb, features);
3757 }
3758
3759 static bool skb_gso_has_extension_hdr(const struct sk_buff *skb)
3760 {
3761 if (!skb->encapsulation)
3762 return ((skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6 ||
3763 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 &&
3764 vlan_get_protocol(skb) == htons(ETH_P_IPV6))) &&
3765 skb_transport_header_was_set(skb) &&
3766 skb_network_header_len(skb) != sizeof(struct ipv6hdr));
3767 else
3768 return (!skb_inner_network_header_was_set(skb) ||
3769 ((skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6 ||
3770 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 &&
3771 inner_ip_hdr(skb)->version == 6)) &&
3772 skb_inner_network_header_len(skb) != sizeof(struct ipv6hdr)));
3773 }
3774
3775 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3776 struct net_device *dev,
3777 netdev_features_t features)
3778 {
3779 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3780
3781 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3782 return features & ~NETIF_F_GSO_MASK;
3783
3784 if (unlikely(skb->len >= netif_get_gso_max_size(dev, skb)))
3785 return features & ~NETIF_F_GSO_MASK;
3786
3787 if (!skb_shinfo(skb)->gso_type) {
3788 skb_warn_bad_offload(skb);
3789 return features & ~NETIF_F_GSO_MASK;
3790 }
3791
3792 /* Support for GSO partial features requires software
3793 * intervention before we can actually process the packets
3794 * so we need to strip support for any partial features now
3795 * and we can pull them back in after we have partially
3796 * segmented the frame.
3797 */
3798 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3799 features &= ~dev->gso_partial_features;
3800
3801 /* Make sure to clear the IPv4 ID mangling feature if the IPv4 header
3802 * has the potential to be fragmented so that TSO does not generate
3803 * segments with the same ID. For encapsulated packets, the ID mangling
3804 * feature is guaranteed not to use the same ID for the outer IPv4
3805 * headers of the generated segments if the headers have the potential
3806 * to be fragmented, so there is no need to clear the IPv4 ID mangling
3807 * feature (see the section about NETIF_F_TSO_MANGLEID in
3808 * segmentation-offloads.rst).
3809 */
3810 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3811 const struct iphdr *iph;
3812 struct iphdr _iph;
3813 int nhoff = skb->encapsulation ?
3814 skb_inner_network_offset(skb) :
3815 skb_network_offset(skb);
3816
3817 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
3818
3819 if (!iph || !(iph->frag_off & htons(IP_DF)))
3820 features &= ~dev->mangleid_features;
3821 }
3822
3823 /* NETIF_F_IPV6_CSUM does not support IPv6 extension headers,
3824 * so neither does TSO that depends on it.
3825 */
3826 if (features & NETIF_F_IPV6_CSUM &&
3827 skb_gso_has_extension_hdr(skb) &&
3828 !ipv6_has_hopopt_jumbo(skb))
3829 features &= ~(NETIF_F_IPV6_CSUM | NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4);
3830
3831 return features;
3832 }
3833
3834 netdev_features_t netif_skb_features(struct sk_buff *skb)
3835 {
3836 struct net_device *dev = skb->dev;
3837 netdev_features_t features = dev->features;
3838
3839 if (skb_is_gso(skb))
3840 features = gso_features_check(skb, dev, features);
3841
3842 /* If encapsulation offload request, verify we are testing
3843 * hardware encapsulation features instead of standard
3844 * features for the netdev
3845 */
3846 if (skb->encapsulation)
3847 features &= dev->hw_enc_features;
3848
3849 if (skb_vlan_tagged(skb))
3850 features = netdev_intersect_features(features,
3851 dev->vlan_features |
3852 NETIF_F_HW_VLAN_CTAG_TX |
3853 NETIF_F_HW_VLAN_STAG_TX);
3854
3855 if (dev->netdev_ops->ndo_features_check)
3856 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3857 features);
3858 else
3859 features &= dflt_features_check(skb, dev, features);
3860
3861 return harmonize_features(skb, features);
3862 }
3863 EXPORT_SYMBOL(netif_skb_features);
3864
3865 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3866 struct netdev_queue *txq, bool more)
3867 {
3868 unsigned int len;
3869 int rc;
3870
3871 if (dev_nit_active_rcu(dev))
3872 dev_queue_xmit_nit(skb, dev);
3873
3874 len = skb->len;
3875 trace_net_dev_start_xmit(skb, dev);
3876 rc = netdev_start_xmit(skb, dev, txq, more);
3877 trace_net_dev_xmit(skb, rc, dev, len);
3878
3879 return rc;
3880 }
3881
3882 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3883 struct netdev_queue *txq, int *ret)
3884 {
3885 struct sk_buff *skb = first;
3886 int rc = NETDEV_TX_OK;
3887
3888 while (skb) {
3889 struct sk_buff *next = skb->next;
3890
3891 skb_mark_not_on_list(skb);
3892 rc = xmit_one(skb, dev, txq, next != NULL);
3893 if (unlikely(!dev_xmit_complete(rc))) {
3894 skb->next = next;
3895 goto out;
3896 }
3897
3898 skb = next;
3899 if (netif_tx_queue_stopped(txq) && skb) {
3900 rc = NETDEV_TX_BUSY;
3901 break;
3902 }
3903 }
3904
3905 out:
3906 *ret = rc;
3907 return skb;
3908 }
3909
3910 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3911 netdev_features_t features)
3912 {
3913 if (skb_vlan_tag_present(skb) &&
3914 !vlan_hw_offload_capable(features, skb->vlan_proto))
3915 skb = __vlan_hwaccel_push_inside(skb);
3916 return skb;
3917 }
3918
3919 int skb_csum_hwoffload_help(struct sk_buff *skb,
3920 const netdev_features_t features)
3921 {
3922 if (unlikely(skb_csum_is_sctp(skb)))
3923 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3924 skb_crc32c_csum_help(skb);
3925
3926 if (features & NETIF_F_HW_CSUM)
3927 return 0;
3928
3929 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3930 if (vlan_get_protocol(skb) == htons(ETH_P_IPV6) &&
3931 skb_network_header_len(skb) != sizeof(struct ipv6hdr) &&
3932 !ipv6_has_hopopt_jumbo(skb))
3933 goto sw_checksum;
3934
3935 switch (skb->csum_offset) {
3936 case offsetof(struct tcphdr, check):
3937 case offsetof(struct udphdr, check):
3938 return 0;
3939 }
3940 }
3941
3942 sw_checksum:
3943 return skb_checksum_help(skb);
3944 }
3945 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3946
3947 /* Checks if this SKB belongs to an HW offloaded socket
3948 * and whether any SW fallbacks are required based on dev.
3949 * Check decrypted mark in case skb_orphan() cleared socket.
3950 */
3951 static struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
3952 struct net_device *dev)
3953 {
3954 #ifdef CONFIG_SOCK_VALIDATE_XMIT
3955 struct sk_buff *(*sk_validate)(struct sock *sk, struct net_device *dev,
3956 struct sk_buff *skb);
3957 struct sock *sk = skb->sk;
3958
3959 sk_validate = NULL;
3960 if (sk) {
3961 if (sk_fullsock(sk))
3962 sk_validate = sk->sk_validate_xmit_skb;
3963 else if (sk_is_inet(sk) && sk->sk_state == TCP_TIME_WAIT)
3964 sk_validate = inet_twsk(sk)->tw_validate_xmit_skb;
3965 }
3966
3967 if (sk_validate) {
3968 skb = sk_validate(sk, dev, skb);
3969 } else if (unlikely(skb_is_decrypted(skb))) {
3970 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
3971 kfree_skb(skb);
3972 skb = NULL;
3973 }
3974 #endif
3975
3976 return skb;
3977 }
3978
3979 static struct sk_buff *validate_xmit_unreadable_skb(struct sk_buff *skb,
3980 struct net_device *dev)
3981 {
3982 struct skb_shared_info *shinfo;
3983 struct net_iov *niov;
3984
3985 if (likely(skb_frags_readable(skb)))
3986 goto out;
3987
3988 if (!dev->netmem_tx)
3989 goto out_free;
3990
3991 shinfo = skb_shinfo(skb);
3992
3993 if (shinfo->nr_frags > 0) {
3994 niov = netmem_to_net_iov(skb_frag_netmem(&shinfo->frags[0]));
3995 if (net_is_devmem_iov(niov) &&
3996 READ_ONCE(net_devmem_iov_binding(niov)->dev) != dev)
3997 goto out_free;
3998 }
3999
4000 out:
4001 return skb;
4002
4003 out_free:
4004 kfree_skb(skb);
4005 return NULL;
4006 }
4007
4008 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
4009 {
4010 netdev_features_t features;
4011
4012 skb = validate_xmit_unreadable_skb(skb, dev);
4013 if (unlikely(!skb))
4014 goto out_null;
4015
4016 features = netif_skb_features(skb);
4017 skb = validate_xmit_vlan(skb, features);
4018 if (unlikely(!skb))
4019 goto out_null;
4020
4021 skb = sk_validate_xmit_skb(skb, dev);
4022 if (unlikely(!skb))
4023 goto out_null;
4024
4025 if (netif_needs_gso(skb, features)) {
4026 struct sk_buff *segs;
4027
4028 segs = skb_gso_segment(skb, features);
4029 if (IS_ERR(segs)) {
4030 goto out_kfree_skb;
4031 } else if (segs) {
4032 consume_skb(skb);
4033 skb = segs;
4034 }
4035 } else {
4036 if (skb_needs_linearize(skb, features) &&
4037 __skb_linearize(skb))
4038 goto out_kfree_skb;
4039
4040 /* If packet is not checksummed and device does not
4041 * support checksumming for this protocol, complete
4042 * checksumming here.
4043 */
4044 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4045 if (skb->encapsulation)
4046 skb_set_inner_transport_header(skb,
4047 skb_checksum_start_offset(skb));
4048 else
4049 skb_set_transport_header(skb,
4050 skb_checksum_start_offset(skb));
4051 if (skb_csum_hwoffload_help(skb, features))
4052 goto out_kfree_skb;
4053 }
4054 }
4055
4056 skb = validate_xmit_xfrm(skb, features, again);
4057
4058 return skb;
4059
4060 out_kfree_skb:
4061 kfree_skb(skb);
4062 out_null:
4063 dev_core_stats_tx_dropped_inc(dev);
4064 return NULL;
4065 }
4066
4067 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
4068 {
4069 struct sk_buff *next, *head = NULL, *tail;
4070
4071 for (; skb != NULL; skb = next) {
4072 next = skb->next;
4073 skb_mark_not_on_list(skb);
4074
4075 /* in case skb won't be segmented, point to itself */
4076 skb->prev = skb;
4077
4078 skb = validate_xmit_skb(skb, dev, again);
4079 if (!skb)
4080 continue;
4081
4082 if (!head)
4083 head = skb;
4084 else
4085 tail->next = skb;
4086 /* If skb was segmented, skb->prev points to
4087 * the last segment. If not, it still contains skb.
4088 */
4089 tail = skb->prev;
4090 }
4091 return head;
4092 }
4093 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
4094
4095 static void qdisc_pkt_len_init(struct sk_buff *skb)
4096 {
4097 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4098
4099 qdisc_skb_cb(skb)->pkt_len = skb->len;
4100
4101 /* To get more precise estimation of bytes sent on wire,
4102 * we add to pkt_len the headers size of all segments
4103 */
4104 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
4105 u16 gso_segs = shinfo->gso_segs;
4106 unsigned int hdr_len;
4107
4108 /* mac layer + network layer */
4109 if (!skb->encapsulation)
4110 hdr_len = skb_transport_offset(skb);
4111 else
4112 hdr_len = skb_inner_transport_offset(skb);
4113
4114 /* + transport layer */
4115 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4116 const struct tcphdr *th;
4117 struct tcphdr _tcphdr;
4118
4119 th = skb_header_pointer(skb, hdr_len,
4120 sizeof(_tcphdr), &_tcphdr);
4121 if (likely(th))
4122 hdr_len += __tcp_hdrlen(th);
4123 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
4124 struct udphdr _udphdr;
4125
4126 if (skb_header_pointer(skb, hdr_len,
4127 sizeof(_udphdr), &_udphdr))
4128 hdr_len += sizeof(struct udphdr);
4129 }
4130
4131 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) {
4132 int payload = skb->len - hdr_len;
4133
4134 /* Malicious packet. */
4135 if (payload <= 0)
4136 return;
4137 gso_segs = DIV_ROUND_UP(payload, shinfo->gso_size);
4138 }
4139 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
4140 }
4141 }
4142
4143 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
4144 struct sk_buff **to_free,
4145 struct netdev_queue *txq)
4146 {
4147 int rc;
4148
4149 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
4150 if (rc == NET_XMIT_SUCCESS)
4151 trace_qdisc_enqueue(q, txq, skb);
4152 return rc;
4153 }
4154
4155 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
4156 struct net_device *dev,
4157 struct netdev_queue *txq)
4158 {
4159 spinlock_t *root_lock = qdisc_lock(q);
4160 struct sk_buff *to_free = NULL;
4161 bool contended;
4162 int rc;
4163
4164 qdisc_calculate_pkt_len(skb, q);
4165
4166 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP);
4167
4168 if (q->flags & TCQ_F_NOLOCK) {
4169 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
4170 qdisc_run_begin(q)) {
4171 /* Retest nolock_qdisc_is_empty() within the protection
4172 * of q->seqlock to protect from racing with requeuing.
4173 */
4174 if (unlikely(!nolock_qdisc_is_empty(q))) {
4175 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
4176 __qdisc_run(q);
4177 qdisc_run_end(q);
4178
4179 goto no_lock_out;
4180 }
4181
4182 qdisc_bstats_cpu_update(q, skb);
4183 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
4184 !nolock_qdisc_is_empty(q))
4185 __qdisc_run(q);
4186
4187 qdisc_run_end(q);
4188 return NET_XMIT_SUCCESS;
4189 }
4190
4191 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
4192 qdisc_run(q);
4193
4194 no_lock_out:
4195 if (unlikely(to_free))
4196 kfree_skb_list_reason(to_free,
4197 tcf_get_drop_reason(to_free));
4198 return rc;
4199 }
4200
4201 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) {
4202 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
4203 return NET_XMIT_DROP;
4204 }
4205 /*
4206 * Heuristic to force contended enqueues to serialize on a
4207 * separate lock before trying to get qdisc main lock.
4208 * This permits qdisc->running owner to get the lock more
4209 * often and dequeue packets faster.
4210 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
4211 * and then other tasks will only enqueue packets. The packets will be
4212 * sent after the qdisc owner is scheduled again. To prevent this
4213 * scenario the task always serialize on the lock.
4214 */
4215 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
4216 if (unlikely(contended))
4217 spin_lock(&q->busylock);
4218
4219 spin_lock(root_lock);
4220 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
4221 __qdisc_drop(skb, &to_free);
4222 rc = NET_XMIT_DROP;
4223 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
4224 qdisc_run_begin(q)) {
4225 /*
4226 * This is a work-conserving queue; there are no old skbs
4227 * waiting to be sent out; and the qdisc is not running -
4228 * xmit the skb directly.
4229 */
4230
4231 qdisc_bstats_update(q, skb);
4232
4233 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
4234 if (unlikely(contended)) {
4235 spin_unlock(&q->busylock);
4236 contended = false;
4237 }
4238 __qdisc_run(q);
4239 }
4240
4241 qdisc_run_end(q);
4242 rc = NET_XMIT_SUCCESS;
4243 } else {
4244 WRITE_ONCE(q->owner, smp_processor_id());
4245 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
4246 WRITE_ONCE(q->owner, -1);
4247 if (qdisc_run_begin(q)) {
4248 if (unlikely(contended)) {
4249 spin_unlock(&q->busylock);
4250 contended = false;
4251 }
4252 __qdisc_run(q);
4253 qdisc_run_end(q);
4254 }
4255 }
4256 spin_unlock(root_lock);
4257 if (unlikely(to_free))
4258 kfree_skb_list_reason(to_free,
4259 tcf_get_drop_reason(to_free));
4260 if (unlikely(contended))
4261 spin_unlock(&q->busylock);
4262 return rc;
4263 }
4264
4265 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
4266 static void skb_update_prio(struct sk_buff *skb)
4267 {
4268 const struct netprio_map *map;
4269 const struct sock *sk;
4270 unsigned int prioidx;
4271
4272 if (skb->priority)
4273 return;
4274 map = rcu_dereference_bh(skb->dev->priomap);
4275 if (!map)
4276 return;
4277 sk = skb_to_full_sk(skb);
4278 if (!sk)
4279 return;
4280
4281 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
4282
4283 if (prioidx < map->priomap_len)
4284 skb->priority = map->priomap[prioidx];
4285 }
4286 #else
4287 #define skb_update_prio(skb)
4288 #endif
4289
4290 /**
4291 * dev_loopback_xmit - loop back @skb
4292 * @net: network namespace this loopback is happening in
4293 * @sk: sk needed to be a netfilter okfn
4294 * @skb: buffer to transmit
4295 */
4296 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
4297 {
4298 skb_reset_mac_header(skb);
4299 __skb_pull(skb, skb_network_offset(skb));
4300 skb->pkt_type = PACKET_LOOPBACK;
4301 if (skb->ip_summed == CHECKSUM_NONE)
4302 skb->ip_summed = CHECKSUM_UNNECESSARY;
4303 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
4304 skb_dst_force(skb);
4305 netif_rx(skb);
4306 return 0;
4307 }
4308 EXPORT_SYMBOL(dev_loopback_xmit);
4309
4310 #ifdef CONFIG_NET_EGRESS
4311 static struct netdev_queue *
4312 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
4313 {
4314 int qm = skb_get_queue_mapping(skb);
4315
4316 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
4317 }
4318
4319 #ifndef CONFIG_PREEMPT_RT
4320 static bool netdev_xmit_txqueue_skipped(void)
4321 {
4322 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
4323 }
4324
4325 void netdev_xmit_skip_txqueue(bool skip)
4326 {
4327 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
4328 }
4329 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
4330
4331 #else
4332 static bool netdev_xmit_txqueue_skipped(void)
4333 {
4334 return current->net_xmit.skip_txqueue;
4335 }
4336
4337 void netdev_xmit_skip_txqueue(bool skip)
4338 {
4339 current->net_xmit.skip_txqueue = skip;
4340 }
4341 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
4342 #endif
4343 #endif /* CONFIG_NET_EGRESS */
4344
4345 #ifdef CONFIG_NET_XGRESS
4346 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb,
4347 enum skb_drop_reason *drop_reason)
4348 {
4349 int ret = TC_ACT_UNSPEC;
4350 #ifdef CONFIG_NET_CLS_ACT
4351 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
4352 struct tcf_result res;
4353
4354 if (!miniq)
4355 return ret;
4356
4357 /* Global bypass */
4358 if (!static_branch_likely(&tcf_sw_enabled_key))
4359 return ret;
4360
4361 /* Block-wise bypass */
4362 if (tcf_block_bypass_sw(miniq->block))
4363 return ret;
4364
4365 tc_skb_cb(skb)->mru = 0;
4366 tc_skb_cb(skb)->post_ct = false;
4367 tcf_set_drop_reason(skb, *drop_reason);
4368
4369 mini_qdisc_bstats_cpu_update(miniq, skb);
4370 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
4371 /* Only tcf related quirks below. */
4372 switch (ret) {
4373 case TC_ACT_SHOT:
4374 *drop_reason = tcf_get_drop_reason(skb);
4375 mini_qdisc_qstats_cpu_drop(miniq);
4376 break;
4377 case TC_ACT_OK:
4378 case TC_ACT_RECLASSIFY:
4379 skb->tc_index = TC_H_MIN(res.classid);
4380 break;
4381 }
4382 #endif /* CONFIG_NET_CLS_ACT */
4383 return ret;
4384 }
4385
4386 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
4387
4388 void tcx_inc(void)
4389 {
4390 static_branch_inc(&tcx_needed_key);
4391 }
4392
4393 void tcx_dec(void)
4394 {
4395 static_branch_dec(&tcx_needed_key);
4396 }
4397
4398 static __always_inline enum tcx_action_base
4399 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
4400 const bool needs_mac)
4401 {
4402 const struct bpf_mprog_fp *fp;
4403 const struct bpf_prog *prog;
4404 int ret = TCX_NEXT;
4405
4406 if (needs_mac)
4407 __skb_push(skb, skb->mac_len);
4408 bpf_mprog_foreach_prog(entry, fp, prog) {
4409 bpf_compute_data_pointers(skb);
4410 ret = bpf_prog_run(prog, skb);
4411 if (ret != TCX_NEXT)
4412 break;
4413 }
4414 if (needs_mac)
4415 __skb_pull(skb, skb->mac_len);
4416 return tcx_action_code(skb, ret);
4417 }
4418
4419 static __always_inline struct sk_buff *
4420 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4421 struct net_device *orig_dev, bool *another)
4422 {
4423 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
4424 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
4425 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4426 int sch_ret;
4427
4428 if (!entry)
4429 return skb;
4430
4431 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4432 if (*pt_prev) {
4433 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4434 *pt_prev = NULL;
4435 }
4436
4437 qdisc_skb_cb(skb)->pkt_len = skb->len;
4438 tcx_set_ingress(skb, true);
4439
4440 if (static_branch_unlikely(&tcx_needed_key)) {
4441 sch_ret = tcx_run(entry, skb, true);
4442 if (sch_ret != TC_ACT_UNSPEC)
4443 goto ingress_verdict;
4444 }
4445 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4446 ingress_verdict:
4447 switch (sch_ret) {
4448 case TC_ACT_REDIRECT:
4449 /* skb_mac_header check was done by BPF, so we can safely
4450 * push the L2 header back before redirecting to another
4451 * netdev.
4452 */
4453 __skb_push(skb, skb->mac_len);
4454 if (skb_do_redirect(skb) == -EAGAIN) {
4455 __skb_pull(skb, skb->mac_len);
4456 *another = true;
4457 break;
4458 }
4459 *ret = NET_RX_SUCCESS;
4460 bpf_net_ctx_clear(bpf_net_ctx);
4461 return NULL;
4462 case TC_ACT_SHOT:
4463 kfree_skb_reason(skb, drop_reason);
4464 *ret = NET_RX_DROP;
4465 bpf_net_ctx_clear(bpf_net_ctx);
4466 return NULL;
4467 /* used by tc_run */
4468 case TC_ACT_STOLEN:
4469 case TC_ACT_QUEUED:
4470 case TC_ACT_TRAP:
4471 consume_skb(skb);
4472 fallthrough;
4473 case TC_ACT_CONSUMED:
4474 *ret = NET_RX_SUCCESS;
4475 bpf_net_ctx_clear(bpf_net_ctx);
4476 return NULL;
4477 }
4478 bpf_net_ctx_clear(bpf_net_ctx);
4479
4480 return skb;
4481 }
4482
4483 static __always_inline struct sk_buff *
4484 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4485 {
4486 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4487 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
4488 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4489 int sch_ret;
4490
4491 if (!entry)
4492 return skb;
4493
4494 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4495
4496 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4497 * already set by the caller.
4498 */
4499 if (static_branch_unlikely(&tcx_needed_key)) {
4500 sch_ret = tcx_run(entry, skb, false);
4501 if (sch_ret != TC_ACT_UNSPEC)
4502 goto egress_verdict;
4503 }
4504 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4505 egress_verdict:
4506 switch (sch_ret) {
4507 case TC_ACT_REDIRECT:
4508 /* No need to push/pop skb's mac_header here on egress! */
4509 skb_do_redirect(skb);
4510 *ret = NET_XMIT_SUCCESS;
4511 bpf_net_ctx_clear(bpf_net_ctx);
4512 return NULL;
4513 case TC_ACT_SHOT:
4514 kfree_skb_reason(skb, drop_reason);
4515 *ret = NET_XMIT_DROP;
4516 bpf_net_ctx_clear(bpf_net_ctx);
4517 return NULL;
4518 /* used by tc_run */
4519 case TC_ACT_STOLEN:
4520 case TC_ACT_QUEUED:
4521 case TC_ACT_TRAP:
4522 consume_skb(skb);
4523 fallthrough;
4524 case TC_ACT_CONSUMED:
4525 *ret = NET_XMIT_SUCCESS;
4526 bpf_net_ctx_clear(bpf_net_ctx);
4527 return NULL;
4528 }
4529 bpf_net_ctx_clear(bpf_net_ctx);
4530
4531 return skb;
4532 }
4533 #else
4534 static __always_inline struct sk_buff *
4535 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4536 struct net_device *orig_dev, bool *another)
4537 {
4538 return skb;
4539 }
4540
4541 static __always_inline struct sk_buff *
4542 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4543 {
4544 return skb;
4545 }
4546 #endif /* CONFIG_NET_XGRESS */
4547
4548 #ifdef CONFIG_XPS
4549 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4550 struct xps_dev_maps *dev_maps, unsigned int tci)
4551 {
4552 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4553 struct xps_map *map;
4554 int queue_index = -1;
4555
4556 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4557 return queue_index;
4558
4559 tci *= dev_maps->num_tc;
4560 tci += tc;
4561
4562 map = rcu_dereference(dev_maps->attr_map[tci]);
4563 if (map) {
4564 if (map->len == 1)
4565 queue_index = map->queues[0];
4566 else
4567 queue_index = map->queues[reciprocal_scale(
4568 skb_get_hash(skb), map->len)];
4569 if (unlikely(queue_index >= dev->real_num_tx_queues))
4570 queue_index = -1;
4571 }
4572 return queue_index;
4573 }
4574 #endif
4575
4576 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4577 struct sk_buff *skb)
4578 {
4579 #ifdef CONFIG_XPS
4580 struct xps_dev_maps *dev_maps;
4581 struct sock *sk = skb->sk;
4582 int queue_index = -1;
4583
4584 if (!static_key_false(&xps_needed))
4585 return -1;
4586
4587 rcu_read_lock();
4588 if (!static_key_false(&xps_rxqs_needed))
4589 goto get_cpus_map;
4590
4591 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4592 if (dev_maps) {
4593 int tci = sk_rx_queue_get(sk);
4594
4595 if (tci >= 0)
4596 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4597 tci);
4598 }
4599
4600 get_cpus_map:
4601 if (queue_index < 0) {
4602 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4603 if (dev_maps) {
4604 unsigned int tci = skb->sender_cpu - 1;
4605
4606 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4607 tci);
4608 }
4609 }
4610 rcu_read_unlock();
4611
4612 return queue_index;
4613 #else
4614 return -1;
4615 #endif
4616 }
4617
4618 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4619 struct net_device *sb_dev)
4620 {
4621 return 0;
4622 }
4623 EXPORT_SYMBOL(dev_pick_tx_zero);
4624
4625 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4626 struct net_device *sb_dev)
4627 {
4628 struct sock *sk = skb->sk;
4629 int queue_index = sk_tx_queue_get(sk);
4630
4631 sb_dev = sb_dev ? : dev;
4632
4633 if (queue_index < 0 || skb->ooo_okay ||
4634 queue_index >= dev->real_num_tx_queues) {
4635 int new_index = get_xps_queue(dev, sb_dev, skb);
4636
4637 if (new_index < 0)
4638 new_index = skb_tx_hash(dev, sb_dev, skb);
4639
4640 if (queue_index != new_index && sk &&
4641 sk_fullsock(sk) &&
4642 rcu_access_pointer(sk->sk_dst_cache))
4643 sk_tx_queue_set(sk, new_index);
4644
4645 queue_index = new_index;
4646 }
4647
4648 return queue_index;
4649 }
4650 EXPORT_SYMBOL(netdev_pick_tx);
4651
4652 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4653 struct sk_buff *skb,
4654 struct net_device *sb_dev)
4655 {
4656 int queue_index = 0;
4657
4658 #ifdef CONFIG_XPS
4659 u32 sender_cpu = skb->sender_cpu - 1;
4660
4661 if (sender_cpu >= (u32)NR_CPUS)
4662 skb->sender_cpu = raw_smp_processor_id() + 1;
4663 #endif
4664
4665 if (dev->real_num_tx_queues != 1) {
4666 const struct net_device_ops *ops = dev->netdev_ops;
4667
4668 if (ops->ndo_select_queue)
4669 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4670 else
4671 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4672
4673 queue_index = netdev_cap_txqueue(dev, queue_index);
4674 }
4675
4676 skb_set_queue_mapping(skb, queue_index);
4677 return netdev_get_tx_queue(dev, queue_index);
4678 }
4679
4680 /**
4681 * __dev_queue_xmit() - transmit a buffer
4682 * @skb: buffer to transmit
4683 * @sb_dev: suboordinate device used for L2 forwarding offload
4684 *
4685 * Queue a buffer for transmission to a network device. The caller must
4686 * have set the device and priority and built the buffer before calling
4687 * this function. The function can be called from an interrupt.
4688 *
4689 * When calling this method, interrupts MUST be enabled. This is because
4690 * the BH enable code must have IRQs enabled so that it will not deadlock.
4691 *
4692 * Regardless of the return value, the skb is consumed, so it is currently
4693 * difficult to retry a send to this method. (You can bump the ref count
4694 * before sending to hold a reference for retry if you are careful.)
4695 *
4696 * Return:
4697 * * 0 - buffer successfully transmitted
4698 * * positive qdisc return code - NET_XMIT_DROP etc.
4699 * * negative errno - other errors
4700 */
4701 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4702 {
4703 struct net_device *dev = skb->dev;
4704 struct netdev_queue *txq = NULL;
4705 struct Qdisc *q;
4706 int rc = -ENOMEM;
4707 bool again = false;
4708
4709 skb_reset_mac_header(skb);
4710 skb_assert_len(skb);
4711
4712 if (unlikely(skb_shinfo(skb)->tx_flags &
4713 (SKBTX_SCHED_TSTAMP | SKBTX_BPF)))
4714 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4715
4716 /* Disable soft irqs for various locks below. Also
4717 * stops preemption for RCU.
4718 */
4719 rcu_read_lock_bh();
4720
4721 skb_update_prio(skb);
4722
4723 qdisc_pkt_len_init(skb);
4724 tcx_set_ingress(skb, false);
4725 #ifdef CONFIG_NET_EGRESS
4726 if (static_branch_unlikely(&egress_needed_key)) {
4727 if (nf_hook_egress_active()) {
4728 skb = nf_hook_egress(skb, &rc, dev);
4729 if (!skb)
4730 goto out;
4731 }
4732
4733 netdev_xmit_skip_txqueue(false);
4734
4735 nf_skip_egress(skb, true);
4736 skb = sch_handle_egress(skb, &rc, dev);
4737 if (!skb)
4738 goto out;
4739 nf_skip_egress(skb, false);
4740
4741 if (netdev_xmit_txqueue_skipped())
4742 txq = netdev_tx_queue_mapping(dev, skb);
4743 }
4744 #endif
4745 /* If device/qdisc don't need skb->dst, release it right now while
4746 * its hot in this cpu cache.
4747 */
4748 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4749 skb_dst_drop(skb);
4750 else
4751 skb_dst_force(skb);
4752
4753 if (!txq)
4754 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4755
4756 q = rcu_dereference_bh(txq->qdisc);
4757
4758 trace_net_dev_queue(skb);
4759 if (q->enqueue) {
4760 rc = __dev_xmit_skb(skb, q, dev, txq);
4761 goto out;
4762 }
4763
4764 /* The device has no queue. Common case for software devices:
4765 * loopback, all the sorts of tunnels...
4766
4767 * Really, it is unlikely that netif_tx_lock protection is necessary
4768 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4769 * counters.)
4770 * However, it is possible, that they rely on protection
4771 * made by us here.
4772
4773 * Check this and shot the lock. It is not prone from deadlocks.
4774 *Either shot noqueue qdisc, it is even simpler 8)
4775 */
4776 if (dev->flags & IFF_UP) {
4777 int cpu = smp_processor_id(); /* ok because BHs are off */
4778
4779 if (!netif_tx_owned(txq, cpu)) {
4780 bool is_list = false;
4781
4782 if (dev_xmit_recursion())
4783 goto recursion_alert;
4784
4785 skb = validate_xmit_skb(skb, dev, &again);
4786 if (!skb)
4787 goto out;
4788
4789 HARD_TX_LOCK(dev, txq, cpu);
4790
4791 if (!netif_xmit_stopped(txq)) {
4792 is_list = !!skb->next;
4793
4794 dev_xmit_recursion_inc();
4795 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4796 dev_xmit_recursion_dec();
4797
4798 /* GSO segments a single SKB into
4799 * a list of frames. TCP expects error
4800 * to mean none of the data was sent.
4801 */
4802 if (is_list)
4803 rc = NETDEV_TX_OK;
4804 }
4805 HARD_TX_UNLOCK(dev, txq);
4806 if (!skb) /* xmit completed */
4807 goto out;
4808
4809 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4810 dev->name);
4811 /* NETDEV_TX_BUSY or queue was stopped */
4812 if (!is_list)
4813 rc = -ENETDOWN;
4814 } else {
4815 /* Recursion is detected! It is possible,
4816 * unfortunately
4817 */
4818 recursion_alert:
4819 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4820 dev->name);
4821 rc = -ENETDOWN;
4822 }
4823 }
4824
4825 rcu_read_unlock_bh();
4826
4827 dev_core_stats_tx_dropped_inc(dev);
4828 kfree_skb_list(skb);
4829 return rc;
4830 out:
4831 rcu_read_unlock_bh();
4832 return rc;
4833 }
4834 EXPORT_SYMBOL(__dev_queue_xmit);
4835
4836 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4837 {
4838 struct net_device *dev = skb->dev;
4839 struct sk_buff *orig_skb = skb;
4840 struct netdev_queue *txq;
4841 int ret = NETDEV_TX_BUSY;
4842 bool again = false;
4843
4844 if (unlikely(!netif_running(dev) ||
4845 !netif_carrier_ok(dev)))
4846 goto drop;
4847
4848 skb = validate_xmit_skb_list(skb, dev, &again);
4849 if (skb != orig_skb)
4850 goto drop;
4851
4852 skb_set_queue_mapping(skb, queue_id);
4853 txq = skb_get_tx_queue(dev, skb);
4854
4855 local_bh_disable();
4856
4857 dev_xmit_recursion_inc();
4858 HARD_TX_LOCK(dev, txq, smp_processor_id());
4859 if (!netif_xmit_frozen_or_drv_stopped(txq))
4860 ret = netdev_start_xmit(skb, dev, txq, false);
4861 HARD_TX_UNLOCK(dev, txq);
4862 dev_xmit_recursion_dec();
4863
4864 local_bh_enable();
4865 return ret;
4866 drop:
4867 dev_core_stats_tx_dropped_inc(dev);
4868 kfree_skb_list(skb);
4869 return NET_XMIT_DROP;
4870 }
4871 EXPORT_SYMBOL(__dev_direct_xmit);
4872
4873 /*************************************************************************
4874 * Receiver routines
4875 *************************************************************************/
4876 static DEFINE_PER_CPU(struct task_struct *, backlog_napi);
4877
4878 int weight_p __read_mostly = 64; /* old backlog weight */
4879 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4880 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4881
4882 /* Called with irq disabled */
4883 static inline void ____napi_schedule(struct softnet_data *sd,
4884 struct napi_struct *napi)
4885 {
4886 struct task_struct *thread;
4887
4888 lockdep_assert_irqs_disabled();
4889
4890 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4891 /* Paired with smp_mb__before_atomic() in
4892 * napi_enable()/netif_set_threaded().
4893 * Use READ_ONCE() to guarantee a complete
4894 * read on napi->thread. Only call
4895 * wake_up_process() when it's not NULL.
4896 */
4897 thread = READ_ONCE(napi->thread);
4898 if (thread) {
4899 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi))
4900 goto use_local_napi;
4901
4902 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4903 wake_up_process(thread);
4904 return;
4905 }
4906 }
4907
4908 use_local_napi:
4909 DEBUG_NET_WARN_ON_ONCE(!list_empty(&napi->poll_list));
4910 list_add_tail(&napi->poll_list, &sd->poll_list);
4911 WRITE_ONCE(napi->list_owner, smp_processor_id());
4912 /* If not called from net_rx_action()
4913 * we have to raise NET_RX_SOFTIRQ.
4914 */
4915 if (!sd->in_net_rx_action)
4916 raise_softirq_irqoff(NET_RX_SOFTIRQ);
4917 }
4918
4919 #ifdef CONFIG_RPS
4920
4921 struct static_key_false rps_needed __read_mostly;
4922 EXPORT_SYMBOL(rps_needed);
4923 struct static_key_false rfs_needed __read_mostly;
4924 EXPORT_SYMBOL(rfs_needed);
4925
4926 static u32 rfs_slot(u32 hash, const struct rps_dev_flow_table *flow_table)
4927 {
4928 return hash_32(hash, flow_table->log);
4929 }
4930
4931 #ifdef CONFIG_RFS_ACCEL
4932 /**
4933 * rps_flow_is_active - check whether the flow is recently active.
4934 * @rflow: Specific flow to check activity.
4935 * @flow_table: per-queue flowtable that @rflow belongs to.
4936 * @cpu: CPU saved in @rflow.
4937 *
4938 * If the CPU has processed many packets since the flow's last activity
4939 * (beyond 10 times the table size), the flow is considered stale.
4940 *
4941 * Return: true if flow was recently active.
4942 */
4943 static bool rps_flow_is_active(struct rps_dev_flow *rflow,
4944 struct rps_dev_flow_table *flow_table,
4945 unsigned int cpu)
4946 {
4947 unsigned int flow_last_active;
4948 unsigned int sd_input_head;
4949
4950 if (cpu >= nr_cpu_ids)
4951 return false;
4952
4953 sd_input_head = READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head);
4954 flow_last_active = READ_ONCE(rflow->last_qtail);
4955
4956 return (int)(sd_input_head - flow_last_active) <
4957 (int)(10 << flow_table->log);
4958 }
4959 #endif
4960
4961 static struct rps_dev_flow *
4962 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4963 struct rps_dev_flow *rflow, u16 next_cpu, u32 hash)
4964 {
4965 if (next_cpu < nr_cpu_ids) {
4966 u32 head;
4967 #ifdef CONFIG_RFS_ACCEL
4968 struct netdev_rx_queue *rxqueue;
4969 struct rps_dev_flow_table *flow_table;
4970 struct rps_dev_flow *old_rflow;
4971 struct rps_dev_flow *tmp_rflow;
4972 unsigned int tmp_cpu;
4973 u16 rxq_index;
4974 u32 flow_id;
4975 int rc;
4976
4977 /* Should we steer this flow to a different hardware queue? */
4978 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4979 !(dev->features & NETIF_F_NTUPLE))
4980 goto out;
4981 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4982 if (rxq_index == skb_get_rx_queue(skb))
4983 goto out;
4984
4985 rxqueue = dev->_rx + rxq_index;
4986 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4987 if (!flow_table)
4988 goto out;
4989
4990 flow_id = rfs_slot(hash, flow_table);
4991 tmp_rflow = &flow_table->flows[flow_id];
4992 tmp_cpu = READ_ONCE(tmp_rflow->cpu);
4993
4994 if (READ_ONCE(tmp_rflow->filter) != RPS_NO_FILTER) {
4995 if (rps_flow_is_active(tmp_rflow, flow_table,
4996 tmp_cpu)) {
4997 if (hash != READ_ONCE(tmp_rflow->hash) ||
4998 next_cpu == tmp_cpu)
4999 goto out;
5000 }
5001 }
5002
5003 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
5004 rxq_index, flow_id);
5005 if (rc < 0)
5006 goto out;
5007
5008 old_rflow = rflow;
5009 rflow = tmp_rflow;
5010 WRITE_ONCE(rflow->filter, rc);
5011 WRITE_ONCE(rflow->hash, hash);
5012
5013 if (old_rflow->filter == rc)
5014 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
5015 out:
5016 #endif
5017 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
5018 rps_input_queue_tail_save(&rflow->last_qtail, head);
5019 }
5020
5021 WRITE_ONCE(rflow->cpu, next_cpu);
5022 return rflow;
5023 }
5024
5025 /*
5026 * get_rps_cpu is called from netif_receive_skb and returns the target
5027 * CPU from the RPS map of the receiving queue for a given skb.
5028 * rcu_read_lock must be held on entry.
5029 */
5030 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
5031 struct rps_dev_flow **rflowp)
5032 {
5033 const struct rps_sock_flow_table *sock_flow_table;
5034 struct netdev_rx_queue *rxqueue = dev->_rx;
5035 struct rps_dev_flow_table *flow_table;
5036 struct rps_map *map;
5037 int cpu = -1;
5038 u32 tcpu;
5039 u32 hash;
5040
5041 if (skb_rx_queue_recorded(skb)) {
5042 u16 index = skb_get_rx_queue(skb);
5043
5044 if (unlikely(index >= dev->real_num_rx_queues)) {
5045 WARN_ONCE(dev->real_num_rx_queues > 1,
5046 "%s received packet on queue %u, but number "
5047 "of RX queues is %u\n",
5048 dev->name, index, dev->real_num_rx_queues);
5049 goto done;
5050 }
5051 rxqueue += index;
5052 }
5053
5054 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
5055
5056 flow_table = rcu_dereference(rxqueue->rps_flow_table);
5057 map = rcu_dereference(rxqueue->rps_map);
5058 if (!flow_table && !map)
5059 goto done;
5060
5061 skb_reset_network_header(skb);
5062 hash = skb_get_hash(skb);
5063 if (!hash)
5064 goto done;
5065
5066 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
5067 if (flow_table && sock_flow_table) {
5068 struct rps_dev_flow *rflow;
5069 u32 next_cpu;
5070 u32 ident;
5071
5072 /* First check into global flow table if there is a match.
5073 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
5074 */
5075 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
5076 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
5077 goto try_rps;
5078
5079 next_cpu = ident & net_hotdata.rps_cpu_mask;
5080
5081 /* OK, now we know there is a match,
5082 * we can look at the local (per receive queue) flow table
5083 */
5084 rflow = &flow_table->flows[rfs_slot(hash, flow_table)];
5085 tcpu = rflow->cpu;
5086
5087 /*
5088 * If the desired CPU (where last recvmsg was done) is
5089 * different from current CPU (one in the rx-queue flow
5090 * table entry), switch if one of the following holds:
5091 * - Current CPU is unset (>= nr_cpu_ids).
5092 * - Current CPU is offline.
5093 * - The current CPU's queue tail has advanced beyond the
5094 * last packet that was enqueued using this table entry.
5095 * This guarantees that all previous packets for the flow
5096 * have been dequeued, thus preserving in order delivery.
5097 */
5098 if (unlikely(tcpu != next_cpu) &&
5099 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
5100 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
5101 rflow->last_qtail)) >= 0)) {
5102 tcpu = next_cpu;
5103 rflow = set_rps_cpu(dev, skb, rflow, next_cpu, hash);
5104 }
5105
5106 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
5107 *rflowp = rflow;
5108 cpu = tcpu;
5109 goto done;
5110 }
5111 }
5112
5113 try_rps:
5114
5115 if (map) {
5116 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
5117 if (cpu_online(tcpu)) {
5118 cpu = tcpu;
5119 goto done;
5120 }
5121 }
5122
5123 done:
5124 return cpu;
5125 }
5126
5127 #ifdef CONFIG_RFS_ACCEL
5128
5129 /**
5130 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
5131 * @dev: Device on which the filter was set
5132 * @rxq_index: RX queue index
5133 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
5134 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
5135 *
5136 * Drivers that implement ndo_rx_flow_steer() should periodically call
5137 * this function for each installed filter and remove the filters for
5138 * which it returns %true.
5139 */
5140 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
5141 u32 flow_id, u16 filter_id)
5142 {
5143 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
5144 struct rps_dev_flow_table *flow_table;
5145 struct rps_dev_flow *rflow;
5146 bool expire = true;
5147
5148 rcu_read_lock();
5149 flow_table = rcu_dereference(rxqueue->rps_flow_table);
5150 if (flow_table && flow_id < (1UL << flow_table->log)) {
5151 unsigned int cpu;
5152
5153 rflow = &flow_table->flows[flow_id];
5154 cpu = READ_ONCE(rflow->cpu);
5155 if (READ_ONCE(rflow->filter) == filter_id &&
5156 rps_flow_is_active(rflow, flow_table, cpu))
5157 expire = false;
5158 }
5159 rcu_read_unlock();
5160 return expire;
5161 }
5162 EXPORT_SYMBOL(rps_may_expire_flow);
5163
5164 #endif /* CONFIG_RFS_ACCEL */
5165
5166 /* Called from hardirq (IPI) context */
5167 static void rps_trigger_softirq(void *data)
5168 {
5169 struct softnet_data *sd = data;
5170
5171 ____napi_schedule(sd, &sd->backlog);
5172 /* Pairs with READ_ONCE() in softnet_seq_show() */
5173 WRITE_ONCE(sd->received_rps, sd->received_rps + 1);
5174 }
5175
5176 #endif /* CONFIG_RPS */
5177
5178 /* Called from hardirq (IPI) context */
5179 static void trigger_rx_softirq(void *data)
5180 {
5181 struct softnet_data *sd = data;
5182
5183 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5184 smp_store_release(&sd->defer_ipi_scheduled, 0);
5185 }
5186
5187 /*
5188 * After we queued a packet into sd->input_pkt_queue,
5189 * we need to make sure this queue is serviced soon.
5190 *
5191 * - If this is another cpu queue, link it to our rps_ipi_list,
5192 * and make sure we will process rps_ipi_list from net_rx_action().
5193 *
5194 * - If this is our own queue, NAPI schedule our backlog.
5195 * Note that this also raises NET_RX_SOFTIRQ.
5196 */
5197 static void napi_schedule_rps(struct softnet_data *sd)
5198 {
5199 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
5200
5201 #ifdef CONFIG_RPS
5202 if (sd != mysd) {
5203 if (use_backlog_threads()) {
5204 __napi_schedule_irqoff(&sd->backlog);
5205 return;
5206 }
5207
5208 sd->rps_ipi_next = mysd->rps_ipi_list;
5209 mysd->rps_ipi_list = sd;
5210
5211 /* If not called from net_rx_action() or napi_threaded_poll()
5212 * we have to raise NET_RX_SOFTIRQ.
5213 */
5214 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
5215 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5216 return;
5217 }
5218 #endif /* CONFIG_RPS */
5219 __napi_schedule_irqoff(&mysd->backlog);
5220 }
5221
5222 void kick_defer_list_purge(unsigned int cpu)
5223 {
5224 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5225 unsigned long flags;
5226
5227 if (use_backlog_threads()) {
5228 backlog_lock_irq_save(sd, &flags);
5229
5230 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
5231 __napi_schedule_irqoff(&sd->backlog);
5232
5233 backlog_unlock_irq_restore(sd, &flags);
5234
5235 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) {
5236 smp_call_function_single_async(cpu, &sd->defer_csd);
5237 }
5238 }
5239
5240 #ifdef CONFIG_NET_FLOW_LIMIT
5241 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
5242 #endif
5243
5244 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
5245 {
5246 #ifdef CONFIG_NET_FLOW_LIMIT
5247 struct sd_flow_limit *fl;
5248 struct softnet_data *sd;
5249 unsigned int old_flow, new_flow;
5250
5251 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
5252 return false;
5253
5254 sd = this_cpu_ptr(&softnet_data);
5255
5256 rcu_read_lock();
5257 fl = rcu_dereference(sd->flow_limit);
5258 if (fl) {
5259 new_flow = hash_32(skb_get_hash(skb), fl->log_buckets);
5260 old_flow = fl->history[fl->history_head];
5261 fl->history[fl->history_head] = new_flow;
5262
5263 fl->history_head++;
5264 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
5265
5266 if (likely(fl->buckets[old_flow]))
5267 fl->buckets[old_flow]--;
5268
5269 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
5270 /* Pairs with READ_ONCE() in softnet_seq_show() */
5271 WRITE_ONCE(fl->count, fl->count + 1);
5272 rcu_read_unlock();
5273 return true;
5274 }
5275 }
5276 rcu_read_unlock();
5277 #endif
5278 return false;
5279 }
5280
5281 /*
5282 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
5283 * queue (may be a remote CPU queue).
5284 */
5285 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
5286 unsigned int *qtail)
5287 {
5288 enum skb_drop_reason reason;
5289 struct softnet_data *sd;
5290 unsigned long flags;
5291 unsigned int qlen;
5292 int max_backlog;
5293 u32 tail;
5294
5295 reason = SKB_DROP_REASON_DEV_READY;
5296 if (!netif_running(skb->dev))
5297 goto bad_dev;
5298
5299 reason = SKB_DROP_REASON_CPU_BACKLOG;
5300 sd = &per_cpu(softnet_data, cpu);
5301
5302 qlen = skb_queue_len_lockless(&sd->input_pkt_queue);
5303 max_backlog = READ_ONCE(net_hotdata.max_backlog);
5304 if (unlikely(qlen > max_backlog))
5305 goto cpu_backlog_drop;
5306 backlog_lock_irq_save(sd, &flags);
5307 qlen = skb_queue_len(&sd->input_pkt_queue);
5308 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) {
5309 if (!qlen) {
5310 /* Schedule NAPI for backlog device. We can use
5311 * non atomic operation as we own the queue lock.
5312 */
5313 if (!__test_and_set_bit(NAPI_STATE_SCHED,
5314 &sd->backlog.state))
5315 napi_schedule_rps(sd);
5316 }
5317 __skb_queue_tail(&sd->input_pkt_queue, skb);
5318 tail = rps_input_queue_tail_incr(sd);
5319 backlog_unlock_irq_restore(sd, &flags);
5320
5321 /* save the tail outside of the critical section */
5322 rps_input_queue_tail_save(qtail, tail);
5323 return NET_RX_SUCCESS;
5324 }
5325
5326 backlog_unlock_irq_restore(sd, &flags);
5327
5328 cpu_backlog_drop:
5329 numa_drop_add(&sd->drop_counters, 1);
5330 bad_dev:
5331 dev_core_stats_rx_dropped_inc(skb->dev);
5332 kfree_skb_reason(skb, reason);
5333 return NET_RX_DROP;
5334 }
5335
5336 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
5337 {
5338 struct net_device *dev = skb->dev;
5339 struct netdev_rx_queue *rxqueue;
5340
5341 rxqueue = dev->_rx;
5342
5343 if (skb_rx_queue_recorded(skb)) {
5344 u16 index = skb_get_rx_queue(skb);
5345
5346 if (unlikely(index >= dev->real_num_rx_queues)) {
5347 WARN_ONCE(dev->real_num_rx_queues > 1,
5348 "%s received packet on queue %u, but number "
5349 "of RX queues is %u\n",
5350 dev->name, index, dev->real_num_rx_queues);
5351
5352 return rxqueue; /* Return first rxqueue */
5353 }
5354 rxqueue += index;
5355 }
5356 return rxqueue;
5357 }
5358
5359 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
5360 const struct bpf_prog *xdp_prog)
5361 {
5362 void *orig_data, *orig_data_end, *hard_start;
5363 struct netdev_rx_queue *rxqueue;
5364 bool orig_bcast, orig_host;
5365 u32 mac_len, frame_sz;
5366 __be16 orig_eth_type;
5367 struct ethhdr *eth;
5368 u32 metalen, act;
5369 int off;
5370
5371 /* The XDP program wants to see the packet starting at the MAC
5372 * header.
5373 */
5374 mac_len = skb->data - skb_mac_header(skb);
5375 hard_start = skb->data - skb_headroom(skb);
5376
5377 /* SKB "head" area always have tailroom for skb_shared_info */
5378 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
5379 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
5380
5381 rxqueue = netif_get_rxqueue(skb);
5382 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
5383 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
5384 skb_headlen(skb) + mac_len, true);
5385 if (skb_is_nonlinear(skb)) {
5386 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
5387 xdp_buff_set_frags_flag(xdp);
5388 } else {
5389 xdp_buff_clear_frags_flag(xdp);
5390 }
5391
5392 orig_data_end = xdp->data_end;
5393 orig_data = xdp->data;
5394 eth = (struct ethhdr *)xdp->data;
5395 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
5396 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
5397 orig_eth_type = eth->h_proto;
5398
5399 act = bpf_prog_run_xdp(xdp_prog, xdp);
5400
5401 /* check if bpf_xdp_adjust_head was used */
5402 off = xdp->data - orig_data;
5403 if (off) {
5404 if (off > 0)
5405 __skb_pull(skb, off);
5406 else if (off < 0)
5407 __skb_push(skb, -off);
5408
5409 skb->mac_header += off;
5410 skb_reset_network_header(skb);
5411 }
5412
5413 /* check if bpf_xdp_adjust_tail was used */
5414 off = xdp->data_end - orig_data_end;
5415 if (off != 0) {
5416 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
5417 skb->len += off; /* positive on grow, negative on shrink */
5418 }
5419
5420 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
5421 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
5422 */
5423 if (xdp_buff_has_frags(xdp))
5424 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
5425 else
5426 skb->data_len = 0;
5427
5428 /* check if XDP changed eth hdr such SKB needs update */
5429 eth = (struct ethhdr *)xdp->data;
5430 if ((orig_eth_type != eth->h_proto) ||
5431 (orig_host != ether_addr_equal_64bits(eth->h_dest,
5432 skb->dev->dev_addr)) ||
5433 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
5434 __skb_push(skb, ETH_HLEN);
5435 skb->pkt_type = PACKET_HOST;
5436 skb->protocol = eth_type_trans(skb, skb->dev);
5437 }
5438
5439 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
5440 * before calling us again on redirect path. We do not call do_redirect
5441 * as we leave that up to the caller.
5442 *
5443 * Caller is responsible for managing lifetime of skb (i.e. calling
5444 * kfree_skb in response to actions it cannot handle/XDP_DROP).
5445 */
5446 switch (act) {
5447 case XDP_REDIRECT:
5448 case XDP_TX:
5449 __skb_push(skb, mac_len);
5450 break;
5451 case XDP_PASS:
5452 metalen = xdp->data - xdp->data_meta;
5453 if (metalen)
5454 skb_metadata_set(skb, metalen);
5455 break;
5456 }
5457
5458 return act;
5459 }
5460
5461 static int
5462 netif_skb_check_for_xdp(struct sk_buff **pskb, const struct bpf_prog *prog)
5463 {
5464 struct sk_buff *skb = *pskb;
5465 int err, hroom, troom;
5466
5467 local_lock_nested_bh(&system_page_pool.bh_lock);
5468 err = skb_cow_data_for_xdp(this_cpu_read(system_page_pool.pool), pskb, prog);
5469 local_unlock_nested_bh(&system_page_pool.bh_lock);
5470 if (!err)
5471 return 0;
5472
5473 /* In case we have to go down the path and also linearize,
5474 * then lets do the pskb_expand_head() work just once here.
5475 */
5476 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
5477 troom = skb->tail + skb->data_len - skb->end;
5478 err = pskb_expand_head(skb,
5479 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
5480 troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
5481 if (err)
5482 return err;
5483
5484 return skb_linearize(skb);
5485 }
5486
5487 static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
5488 struct xdp_buff *xdp,
5489 const struct bpf_prog *xdp_prog)
5490 {
5491 struct sk_buff *skb = *pskb;
5492 u32 mac_len, act = XDP_DROP;
5493
5494 /* Reinjected packets coming from act_mirred or similar should
5495 * not get XDP generic processing.
5496 */
5497 if (skb_is_redirected(skb))
5498 return XDP_PASS;
5499
5500 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
5501 * bytes. This is the guarantee that also native XDP provides,
5502 * thus we need to do it here as well.
5503 */
5504 mac_len = skb->data - skb_mac_header(skb);
5505 __skb_push(skb, mac_len);
5506
5507 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
5508 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
5509 if (netif_skb_check_for_xdp(pskb, xdp_prog))
5510 goto do_drop;
5511 }
5512
5513 __skb_pull(*pskb, mac_len);
5514
5515 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
5516 switch (act) {
5517 case XDP_REDIRECT:
5518 case XDP_TX:
5519 case XDP_PASS:
5520 break;
5521 default:
5522 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
5523 fallthrough;
5524 case XDP_ABORTED:
5525 trace_xdp_exception((*pskb)->dev, xdp_prog, act);
5526 fallthrough;
5527 case XDP_DROP:
5528 do_drop:
5529 kfree_skb(*pskb);
5530 break;
5531 }
5532
5533 return act;
5534 }
5535
5536 /* When doing generic XDP we have to bypass the qdisc layer and the
5537 * network taps in order to match in-driver-XDP behavior. This also means
5538 * that XDP packets are able to starve other packets going through a qdisc,
5539 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
5540 * queues, so they do not have this starvation issue.
5541 */
5542 void generic_xdp_tx(struct sk_buff *skb, const struct bpf_prog *xdp_prog)
5543 {
5544 struct net_device *dev = skb->dev;
5545 struct netdev_queue *txq;
5546 bool free_skb = true;
5547 int cpu, rc;
5548
5549 txq = netdev_core_pick_tx(dev, skb, NULL);
5550 cpu = smp_processor_id();
5551 HARD_TX_LOCK(dev, txq, cpu);
5552 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
5553 rc = netdev_start_xmit(skb, dev, txq, 0);
5554 if (dev_xmit_complete(rc))
5555 free_skb = false;
5556 }
5557 HARD_TX_UNLOCK(dev, txq);
5558 if (free_skb) {
5559 trace_xdp_exception(dev, xdp_prog, XDP_TX);
5560 dev_core_stats_tx_dropped_inc(dev);
5561 kfree_skb(skb);
5562 }
5563 }
5564
5565 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
5566
5567 int do_xdp_generic(const struct bpf_prog *xdp_prog, struct sk_buff **pskb)
5568 {
5569 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
5570
5571 if (xdp_prog) {
5572 struct xdp_buff xdp;
5573 u32 act;
5574 int err;
5575
5576 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
5577 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
5578 if (act != XDP_PASS) {
5579 switch (act) {
5580 case XDP_REDIRECT:
5581 err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
5582 &xdp, xdp_prog);
5583 if (err)
5584 goto out_redir;
5585 break;
5586 case XDP_TX:
5587 generic_xdp_tx(*pskb, xdp_prog);
5588 break;
5589 }
5590 bpf_net_ctx_clear(bpf_net_ctx);
5591 return XDP_DROP;
5592 }
5593 bpf_net_ctx_clear(bpf_net_ctx);
5594 }
5595 return XDP_PASS;
5596 out_redir:
5597 bpf_net_ctx_clear(bpf_net_ctx);
5598 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
5599 return XDP_DROP;
5600 }
5601 EXPORT_SYMBOL_GPL(do_xdp_generic);
5602
5603 static int netif_rx_internal(struct sk_buff *skb)
5604 {
5605 int ret;
5606
5607 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5608
5609 trace_netif_rx(skb);
5610
5611 #ifdef CONFIG_RPS
5612 if (static_branch_unlikely(&rps_needed)) {
5613 struct rps_dev_flow voidflow, *rflow = &voidflow;
5614 int cpu;
5615
5616 rcu_read_lock();
5617
5618 cpu = get_rps_cpu(skb->dev, skb, &rflow);
5619 if (cpu < 0)
5620 cpu = smp_processor_id();
5621
5622 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5623
5624 rcu_read_unlock();
5625 } else
5626 #endif
5627 {
5628 unsigned int qtail;
5629
5630 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5631 }
5632 return ret;
5633 }
5634
5635 /**
5636 * __netif_rx - Slightly optimized version of netif_rx
5637 * @skb: buffer to post
5638 *
5639 * This behaves as netif_rx except that it does not disable bottom halves.
5640 * As a result this function may only be invoked from the interrupt context
5641 * (either hard or soft interrupt).
5642 */
5643 int __netif_rx(struct sk_buff *skb)
5644 {
5645 int ret;
5646
5647 lockdep_assert_once(hardirq_count() | softirq_count());
5648
5649 trace_netif_rx_entry(skb);
5650 ret = netif_rx_internal(skb);
5651 trace_netif_rx_exit(ret);
5652 return ret;
5653 }
5654 EXPORT_SYMBOL(__netif_rx);
5655
5656 /**
5657 * netif_rx - post buffer to the network code
5658 * @skb: buffer to post
5659 *
5660 * This function receives a packet from a device driver and queues it for
5661 * the upper (protocol) levels to process via the backlog NAPI device. It
5662 * always succeeds. The buffer may be dropped during processing for
5663 * congestion control or by the protocol layers.
5664 * The network buffer is passed via the backlog NAPI device. Modern NIC
5665 * driver should use NAPI and GRO.
5666 * This function can used from interrupt and from process context. The
5667 * caller from process context must not disable interrupts before invoking
5668 * this function.
5669 *
5670 * return values:
5671 * NET_RX_SUCCESS (no congestion)
5672 * NET_RX_DROP (packet was dropped)
5673 *
5674 */
5675 int netif_rx(struct sk_buff *skb)
5676 {
5677 bool need_bh_off = !(hardirq_count() | softirq_count());
5678 int ret;
5679
5680 if (need_bh_off)
5681 local_bh_disable();
5682 trace_netif_rx_entry(skb);
5683 ret = netif_rx_internal(skb);
5684 trace_netif_rx_exit(ret);
5685 if (need_bh_off)
5686 local_bh_enable();
5687 return ret;
5688 }
5689 EXPORT_SYMBOL(netif_rx);
5690
5691 static __latent_entropy void net_tx_action(void)
5692 {
5693 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5694
5695 if (sd->completion_queue) {
5696 struct sk_buff *clist;
5697
5698 local_irq_disable();
5699 clist = sd->completion_queue;
5700 sd->completion_queue = NULL;
5701 local_irq_enable();
5702
5703 while (clist) {
5704 struct sk_buff *skb = clist;
5705
5706 clist = clist->next;
5707
5708 WARN_ON(refcount_read(&skb->users));
5709 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5710 trace_consume_skb(skb, net_tx_action);
5711 else
5712 trace_kfree_skb(skb, net_tx_action,
5713 get_kfree_skb_cb(skb)->reason, NULL);
5714
5715 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5716 __kfree_skb(skb);
5717 else
5718 __napi_kfree_skb(skb,
5719 get_kfree_skb_cb(skb)->reason);
5720 }
5721 }
5722
5723 if (sd->output_queue) {
5724 struct Qdisc *head;
5725
5726 local_irq_disable();
5727 head = sd->output_queue;
5728 sd->output_queue = NULL;
5729 sd->output_queue_tailp = &sd->output_queue;
5730 local_irq_enable();
5731
5732 rcu_read_lock();
5733
5734 while (head) {
5735 struct Qdisc *q = head;
5736 spinlock_t *root_lock = NULL;
5737
5738 head = head->next_sched;
5739
5740 /* We need to make sure head->next_sched is read
5741 * before clearing __QDISC_STATE_SCHED
5742 */
5743 smp_mb__before_atomic();
5744
5745 if (!(q->flags & TCQ_F_NOLOCK)) {
5746 root_lock = qdisc_lock(q);
5747 spin_lock(root_lock);
5748 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5749 &q->state))) {
5750 /* There is a synchronize_net() between
5751 * STATE_DEACTIVATED flag being set and
5752 * qdisc_reset()/some_qdisc_is_busy() in
5753 * dev_deactivate(), so we can safely bail out
5754 * early here to avoid data race between
5755 * qdisc_deactivate() and some_qdisc_is_busy()
5756 * for lockless qdisc.
5757 */
5758 clear_bit(__QDISC_STATE_SCHED, &q->state);
5759 continue;
5760 }
5761
5762 clear_bit(__QDISC_STATE_SCHED, &q->state);
5763 qdisc_run(q);
5764 if (root_lock)
5765 spin_unlock(root_lock);
5766 }
5767
5768 rcu_read_unlock();
5769 }
5770
5771 xfrm_dev_backlog(sd);
5772 }
5773
5774 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5775 /* This hook is defined here for ATM LANE */
5776 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5777 unsigned char *addr) __read_mostly;
5778 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5779 #endif
5780
5781 /**
5782 * netdev_is_rx_handler_busy - check if receive handler is registered
5783 * @dev: device to check
5784 *
5785 * Check if a receive handler is already registered for a given device.
5786 * Return true if there one.
5787 *
5788 * The caller must hold the rtnl_mutex.
5789 */
5790 bool netdev_is_rx_handler_busy(struct net_device *dev)
5791 {
5792 ASSERT_RTNL();
5793 return dev && rtnl_dereference(dev->rx_handler);
5794 }
5795 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5796
5797 /**
5798 * netdev_rx_handler_register - register receive handler
5799 * @dev: device to register a handler for
5800 * @rx_handler: receive handler to register
5801 * @rx_handler_data: data pointer that is used by rx handler
5802 *
5803 * Register a receive handler for a device. This handler will then be
5804 * called from __netif_receive_skb. A negative errno code is returned
5805 * on a failure.
5806 *
5807 * The caller must hold the rtnl_mutex.
5808 *
5809 * For a general description of rx_handler, see enum rx_handler_result.
5810 */
5811 int netdev_rx_handler_register(struct net_device *dev,
5812 rx_handler_func_t *rx_handler,
5813 void *rx_handler_data)
5814 {
5815 if (netdev_is_rx_handler_busy(dev))
5816 return -EBUSY;
5817
5818 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5819 return -EINVAL;
5820
5821 /* Note: rx_handler_data must be set before rx_handler */
5822 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5823 rcu_assign_pointer(dev->rx_handler, rx_handler);
5824
5825 return 0;
5826 }
5827 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5828
5829 /**
5830 * netdev_rx_handler_unregister - unregister receive handler
5831 * @dev: device to unregister a handler from
5832 *
5833 * Unregister a receive handler from a device.
5834 *
5835 * The caller must hold the rtnl_mutex.
5836 */
5837 void netdev_rx_handler_unregister(struct net_device *dev)
5838 {
5839
5840 ASSERT_RTNL();
5841 RCU_INIT_POINTER(dev->rx_handler, NULL);
5842 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5843 * section has a guarantee to see a non NULL rx_handler_data
5844 * as well.
5845 */
5846 synchronize_net();
5847 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5848 }
5849 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5850
5851 /*
5852 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5853 * the special handling of PFMEMALLOC skbs.
5854 */
5855 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5856 {
5857 switch (skb->protocol) {
5858 case htons(ETH_P_ARP):
5859 case htons(ETH_P_IP):
5860 case htons(ETH_P_IPV6):
5861 case htons(ETH_P_8021Q):
5862 case htons(ETH_P_8021AD):
5863 return true;
5864 default:
5865 return false;
5866 }
5867 }
5868
5869 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5870 int *ret, struct net_device *orig_dev)
5871 {
5872 if (nf_hook_ingress_active(skb)) {
5873 int ingress_retval;
5874
5875 if (*pt_prev) {
5876 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5877 *pt_prev = NULL;
5878 }
5879
5880 rcu_read_lock();
5881 ingress_retval = nf_hook_ingress(skb);
5882 rcu_read_unlock();
5883 return ingress_retval;
5884 }
5885 return 0;
5886 }
5887
5888 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5889 struct packet_type **ppt_prev)
5890 {
5891 enum skb_drop_reason drop_reason = SKB_DROP_REASON_UNHANDLED_PROTO;
5892 struct packet_type *ptype, *pt_prev;
5893 rx_handler_func_t *rx_handler;
5894 struct sk_buff *skb = *pskb;
5895 struct net_device *orig_dev;
5896 bool deliver_exact = false;
5897 int ret = NET_RX_DROP;
5898 __be16 type;
5899
5900 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5901
5902 trace_netif_receive_skb(skb);
5903
5904 orig_dev = skb->dev;
5905
5906 skb_reset_network_header(skb);
5907 #if !defined(CONFIG_DEBUG_NET)
5908 /* We plan to no longer reset the transport header here.
5909 * Give some time to fuzzers and dev build to catch bugs
5910 * in network stacks.
5911 */
5912 if (!skb_transport_header_was_set(skb))
5913 skb_reset_transport_header(skb);
5914 #endif
5915 skb_reset_mac_len(skb);
5916
5917 pt_prev = NULL;
5918
5919 another_round:
5920 skb->skb_iif = skb->dev->ifindex;
5921
5922 __this_cpu_inc(softnet_data.processed);
5923
5924 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5925 int ret2;
5926
5927 migrate_disable();
5928 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
5929 &skb);
5930 migrate_enable();
5931
5932 if (ret2 != XDP_PASS) {
5933 ret = NET_RX_DROP;
5934 goto out;
5935 }
5936 }
5937
5938 if (eth_type_vlan(skb->protocol)) {
5939 skb = skb_vlan_untag(skb);
5940 if (unlikely(!skb))
5941 goto out;
5942 }
5943
5944 if (skb_skip_tc_classify(skb))
5945 goto skip_classify;
5946
5947 if (pfmemalloc)
5948 goto skip_taps;
5949
5950 list_for_each_entry_rcu(ptype, &dev_net_rcu(skb->dev)->ptype_all,
5951 list) {
5952 if (pt_prev)
5953 ret = deliver_skb(skb, pt_prev, orig_dev);
5954 pt_prev = ptype;
5955 }
5956
5957 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5958 if (pt_prev)
5959 ret = deliver_skb(skb, pt_prev, orig_dev);
5960 pt_prev = ptype;
5961 }
5962
5963 skip_taps:
5964 #ifdef CONFIG_NET_INGRESS
5965 if (static_branch_unlikely(&ingress_needed_key)) {
5966 bool another = false;
5967
5968 nf_skip_egress(skb, true);
5969 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5970 &another);
5971 if (another)
5972 goto another_round;
5973 if (!skb)
5974 goto out;
5975
5976 nf_skip_egress(skb, false);
5977 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5978 goto out;
5979 }
5980 #endif
5981 skb_reset_redirect(skb);
5982 skip_classify:
5983 if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) {
5984 drop_reason = SKB_DROP_REASON_PFMEMALLOC;
5985 goto drop;
5986 }
5987
5988 if (skb_vlan_tag_present(skb)) {
5989 if (pt_prev) {
5990 ret = deliver_skb(skb, pt_prev, orig_dev);
5991 pt_prev = NULL;
5992 }
5993 if (vlan_do_receive(&skb))
5994 goto another_round;
5995 else if (unlikely(!skb))
5996 goto out;
5997 }
5998
5999 rx_handler = rcu_dereference(skb->dev->rx_handler);
6000 if (rx_handler) {
6001 if (pt_prev) {
6002 ret = deliver_skb(skb, pt_prev, orig_dev);
6003 pt_prev = NULL;
6004 }
6005 switch (rx_handler(&skb)) {
6006 case RX_HANDLER_CONSUMED:
6007 ret = NET_RX_SUCCESS;
6008 goto out;
6009 case RX_HANDLER_ANOTHER:
6010 goto another_round;
6011 case RX_HANDLER_EXACT:
6012 deliver_exact = true;
6013 break;
6014 case RX_HANDLER_PASS:
6015 break;
6016 default:
6017 BUG();
6018 }
6019 }
6020
6021 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
6022 check_vlan_id:
6023 if (skb_vlan_tag_get_id(skb)) {
6024 /* Vlan id is non 0 and vlan_do_receive() above couldn't
6025 * find vlan device.
6026 */
6027 skb->pkt_type = PACKET_OTHERHOST;
6028 } else if (eth_type_vlan(skb->protocol)) {
6029 /* Outer header is 802.1P with vlan 0, inner header is
6030 * 802.1Q or 802.1AD and vlan_do_receive() above could
6031 * not find vlan dev for vlan id 0.
6032 */
6033 __vlan_hwaccel_clear_tag(skb);
6034 skb = skb_vlan_untag(skb);
6035 if (unlikely(!skb))
6036 goto out;
6037 if (vlan_do_receive(&skb))
6038 /* After stripping off 802.1P header with vlan 0
6039 * vlan dev is found for inner header.
6040 */
6041 goto another_round;
6042 else if (unlikely(!skb))
6043 goto out;
6044 else
6045 /* We have stripped outer 802.1P vlan 0 header.
6046 * But could not find vlan dev.
6047 * check again for vlan id to set OTHERHOST.
6048 */
6049 goto check_vlan_id;
6050 }
6051 /* Note: we might in the future use prio bits
6052 * and set skb->priority like in vlan_do_receive()
6053 * For the time being, just ignore Priority Code Point
6054 */
6055 __vlan_hwaccel_clear_tag(skb);
6056 }
6057
6058 type = skb->protocol;
6059
6060 /* deliver only exact match when indicated */
6061 if (likely(!deliver_exact)) {
6062 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
6063 &ptype_base[ntohs(type) &
6064 PTYPE_HASH_MASK]);
6065
6066 /* orig_dev and skb->dev could belong to different netns;
6067 * Even in such case we need to traverse only the list
6068 * coming from skb->dev, as the ptype owner (packet socket)
6069 * will use dev_net(skb->dev) to do namespace filtering.
6070 */
6071 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
6072 &dev_net_rcu(skb->dev)->ptype_specific);
6073 }
6074
6075 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
6076 &orig_dev->ptype_specific);
6077
6078 if (unlikely(skb->dev != orig_dev)) {
6079 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
6080 &skb->dev->ptype_specific);
6081 }
6082
6083 if (pt_prev) {
6084 *ppt_prev = pt_prev;
6085 } else {
6086 drop:
6087 if (!deliver_exact)
6088 dev_core_stats_rx_dropped_inc(skb->dev);
6089 else
6090 dev_core_stats_rx_nohandler_inc(skb->dev);
6091
6092 kfree_skb_reason(skb, drop_reason);
6093 /* Jamal, now you will not able to escape explaining
6094 * me how you were going to use this. :-)
6095 */
6096 ret = NET_RX_DROP;
6097 }
6098
6099 out:
6100 /* The invariant here is that if *ppt_prev is not NULL
6101 * then skb should also be non-NULL.
6102 *
6103 * Apparently *ppt_prev assignment above holds this invariant due to
6104 * skb dereferencing near it.
6105 */
6106 *pskb = skb;
6107 return ret;
6108 }
6109
6110 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
6111 {
6112 struct net_device *orig_dev = skb->dev;
6113 struct packet_type *pt_prev = NULL;
6114 int ret;
6115
6116 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
6117 if (pt_prev)
6118 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
6119 skb->dev, pt_prev, orig_dev);
6120 return ret;
6121 }
6122
6123 /**
6124 * netif_receive_skb_core - special purpose version of netif_receive_skb
6125 * @skb: buffer to process
6126 *
6127 * More direct receive version of netif_receive_skb(). It should
6128 * only be used by callers that have a need to skip RPS and Generic XDP.
6129 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
6130 *
6131 * This function may only be called from softirq context and interrupts
6132 * should be enabled.
6133 *
6134 * Return values (usually ignored):
6135 * NET_RX_SUCCESS: no congestion
6136 * NET_RX_DROP: packet was dropped
6137 */
6138 int netif_receive_skb_core(struct sk_buff *skb)
6139 {
6140 int ret;
6141
6142 rcu_read_lock();
6143 ret = __netif_receive_skb_one_core(skb, false);
6144 rcu_read_unlock();
6145
6146 return ret;
6147 }
6148 EXPORT_SYMBOL(netif_receive_skb_core);
6149
6150 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
6151 struct packet_type *pt_prev,
6152 struct net_device *orig_dev)
6153 {
6154 struct sk_buff *skb, *next;
6155
6156 if (!pt_prev)
6157 return;
6158 if (list_empty(head))
6159 return;
6160 if (pt_prev->list_func != NULL)
6161 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
6162 ip_list_rcv, head, pt_prev, orig_dev);
6163 else
6164 list_for_each_entry_safe(skb, next, head, list) {
6165 skb_list_del_init(skb);
6166 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
6167 }
6168 }
6169
6170 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
6171 {
6172 /* Fast-path assumptions:
6173 * - There is no RX handler.
6174 * - Only one packet_type matches.
6175 * If either of these fails, we will end up doing some per-packet
6176 * processing in-line, then handling the 'last ptype' for the whole
6177 * sublist. This can't cause out-of-order delivery to any single ptype,
6178 * because the 'last ptype' must be constant across the sublist, and all
6179 * other ptypes are handled per-packet.
6180 */
6181 /* Current (common) ptype of sublist */
6182 struct packet_type *pt_curr = NULL;
6183 /* Current (common) orig_dev of sublist */
6184 struct net_device *od_curr = NULL;
6185 struct sk_buff *skb, *next;
6186 LIST_HEAD(sublist);
6187
6188 list_for_each_entry_safe(skb, next, head, list) {
6189 struct net_device *orig_dev = skb->dev;
6190 struct packet_type *pt_prev = NULL;
6191
6192 skb_list_del_init(skb);
6193 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
6194 if (!pt_prev)
6195 continue;
6196 if (pt_curr != pt_prev || od_curr != orig_dev) {
6197 /* dispatch old sublist */
6198 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
6199 /* start new sublist */
6200 INIT_LIST_HEAD(&sublist);
6201 pt_curr = pt_prev;
6202 od_curr = orig_dev;
6203 }
6204 list_add_tail(&skb->list, &sublist);
6205 }
6206
6207 /* dispatch final sublist */
6208 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
6209 }
6210
6211 static int __netif_receive_skb(struct sk_buff *skb)
6212 {
6213 int ret;
6214
6215 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
6216 unsigned int noreclaim_flag;
6217
6218 /*
6219 * PFMEMALLOC skbs are special, they should
6220 * - be delivered to SOCK_MEMALLOC sockets only
6221 * - stay away from userspace
6222 * - have bounded memory usage
6223 *
6224 * Use PF_MEMALLOC as this saves us from propagating the allocation
6225 * context down to all allocation sites.
6226 */
6227 noreclaim_flag = memalloc_noreclaim_save();
6228 ret = __netif_receive_skb_one_core(skb, true);
6229 memalloc_noreclaim_restore(noreclaim_flag);
6230 } else
6231 ret = __netif_receive_skb_one_core(skb, false);
6232
6233 return ret;
6234 }
6235
6236 static void __netif_receive_skb_list(struct list_head *head)
6237 {
6238 unsigned long noreclaim_flag = 0;
6239 struct sk_buff *skb, *next;
6240 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
6241
6242 list_for_each_entry_safe(skb, next, head, list) {
6243 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
6244 struct list_head sublist;
6245
6246 /* Handle the previous sublist */
6247 list_cut_before(&sublist, head, &skb->list);
6248 if (!list_empty(&sublist))
6249 __netif_receive_skb_list_core(&sublist, pfmemalloc);
6250 pfmemalloc = !pfmemalloc;
6251 /* See comments in __netif_receive_skb */
6252 if (pfmemalloc)
6253 noreclaim_flag = memalloc_noreclaim_save();
6254 else
6255 memalloc_noreclaim_restore(noreclaim_flag);
6256 }
6257 }
6258 /* Handle the remaining sublist */
6259 if (!list_empty(head))
6260 __netif_receive_skb_list_core(head, pfmemalloc);
6261 /* Restore pflags */
6262 if (pfmemalloc)
6263 memalloc_noreclaim_restore(noreclaim_flag);
6264 }
6265
6266 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
6267 {
6268 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
6269 struct bpf_prog *new = xdp->prog;
6270 int ret = 0;
6271
6272 switch (xdp->command) {
6273 case XDP_SETUP_PROG:
6274 rcu_assign_pointer(dev->xdp_prog, new);
6275 if (old)
6276 bpf_prog_put(old);
6277
6278 if (old && !new) {
6279 static_branch_dec(&generic_xdp_needed_key);
6280 } else if (new && !old) {
6281 static_branch_inc(&generic_xdp_needed_key);
6282 netif_disable_lro(dev);
6283 dev_disable_gro_hw(dev);
6284 }
6285 break;
6286
6287 default:
6288 ret = -EINVAL;
6289 break;
6290 }
6291
6292 return ret;
6293 }
6294
6295 static int netif_receive_skb_internal(struct sk_buff *skb)
6296 {
6297 int ret;
6298
6299 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
6300
6301 if (skb_defer_rx_timestamp(skb))
6302 return NET_RX_SUCCESS;
6303
6304 rcu_read_lock();
6305 #ifdef CONFIG_RPS
6306 if (static_branch_unlikely(&rps_needed)) {
6307 struct rps_dev_flow voidflow, *rflow = &voidflow;
6308 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
6309
6310 if (cpu >= 0) {
6311 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
6312 rcu_read_unlock();
6313 return ret;
6314 }
6315 }
6316 #endif
6317 ret = __netif_receive_skb(skb);
6318 rcu_read_unlock();
6319 return ret;
6320 }
6321
6322 void netif_receive_skb_list_internal(struct list_head *head)
6323 {
6324 struct sk_buff *skb, *next;
6325 LIST_HEAD(sublist);
6326
6327 list_for_each_entry_safe(skb, next, head, list) {
6328 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
6329 skb);
6330 skb_list_del_init(skb);
6331 if (!skb_defer_rx_timestamp(skb))
6332 list_add_tail(&skb->list, &sublist);
6333 }
6334 list_splice_init(&sublist, head);
6335
6336 rcu_read_lock();
6337 #ifdef CONFIG_RPS
6338 if (static_branch_unlikely(&rps_needed)) {
6339 list_for_each_entry_safe(skb, next, head, list) {
6340 struct rps_dev_flow voidflow, *rflow = &voidflow;
6341 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
6342
6343 if (cpu >= 0) {
6344 /* Will be handled, remove from list */
6345 skb_list_del_init(skb);
6346 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
6347 }
6348 }
6349 }
6350 #endif
6351 __netif_receive_skb_list(head);
6352 rcu_read_unlock();
6353 }
6354
6355 /**
6356 * netif_receive_skb - process receive buffer from network
6357 * @skb: buffer to process
6358 *
6359 * netif_receive_skb() is the main receive data processing function.
6360 * It always succeeds. The buffer may be dropped during processing
6361 * for congestion control or by the protocol layers.
6362 *
6363 * This function may only be called from softirq context and interrupts
6364 * should be enabled.
6365 *
6366 * Return values (usually ignored):
6367 * NET_RX_SUCCESS: no congestion
6368 * NET_RX_DROP: packet was dropped
6369 */
6370 int netif_receive_skb(struct sk_buff *skb)
6371 {
6372 int ret;
6373
6374 trace_netif_receive_skb_entry(skb);
6375
6376 ret = netif_receive_skb_internal(skb);
6377 trace_netif_receive_skb_exit(ret);
6378
6379 return ret;
6380 }
6381 EXPORT_SYMBOL(netif_receive_skb);
6382
6383 /**
6384 * netif_receive_skb_list - process many receive buffers from network
6385 * @head: list of skbs to process.
6386 *
6387 * Since return value of netif_receive_skb() is normally ignored, and
6388 * wouldn't be meaningful for a list, this function returns void.
6389 *
6390 * This function may only be called from softirq context and interrupts
6391 * should be enabled.
6392 */
6393 void netif_receive_skb_list(struct list_head *head)
6394 {
6395 struct sk_buff *skb;
6396
6397 if (list_empty(head))
6398 return;
6399 if (trace_netif_receive_skb_list_entry_enabled()) {
6400 list_for_each_entry(skb, head, list)
6401 trace_netif_receive_skb_list_entry(skb);
6402 }
6403 netif_receive_skb_list_internal(head);
6404 trace_netif_receive_skb_list_exit(0);
6405 }
6406 EXPORT_SYMBOL(netif_receive_skb_list);
6407
6408 /* Network device is going away, flush any packets still pending */
6409 static void flush_backlog(struct work_struct *work)
6410 {
6411 struct sk_buff *skb, *tmp;
6412 struct sk_buff_head list;
6413 struct softnet_data *sd;
6414
6415 __skb_queue_head_init(&list);
6416 local_bh_disable();
6417 sd = this_cpu_ptr(&softnet_data);
6418
6419 backlog_lock_irq_disable(sd);
6420 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
6421 if (READ_ONCE(skb->dev->reg_state) == NETREG_UNREGISTERING) {
6422 __skb_unlink(skb, &sd->input_pkt_queue);
6423 __skb_queue_tail(&list, skb);
6424 rps_input_queue_head_incr(sd);
6425 }
6426 }
6427 backlog_unlock_irq_enable(sd);
6428
6429 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6430 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
6431 if (READ_ONCE(skb->dev->reg_state) == NETREG_UNREGISTERING) {
6432 __skb_unlink(skb, &sd->process_queue);
6433 __skb_queue_tail(&list, skb);
6434 rps_input_queue_head_incr(sd);
6435 }
6436 }
6437 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6438 local_bh_enable();
6439
6440 __skb_queue_purge_reason(&list, SKB_DROP_REASON_DEV_READY);
6441 }
6442
6443 static bool flush_required(int cpu)
6444 {
6445 #if IS_ENABLED(CONFIG_RPS)
6446 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
6447 bool do_flush;
6448
6449 backlog_lock_irq_disable(sd);
6450
6451 /* as insertion into process_queue happens with the rps lock held,
6452 * process_queue access may race only with dequeue
6453 */
6454 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
6455 !skb_queue_empty_lockless(&sd->process_queue);
6456 backlog_unlock_irq_enable(sd);
6457
6458 return do_flush;
6459 #endif
6460 /* without RPS we can't safely check input_pkt_queue: during a
6461 * concurrent remote skb_queue_splice() we can detect as empty both
6462 * input_pkt_queue and process_queue even if the latter could end-up
6463 * containing a lot of packets.
6464 */
6465 return true;
6466 }
6467
6468 struct flush_backlogs {
6469 cpumask_t flush_cpus;
6470 struct work_struct w[];
6471 };
6472
6473 static struct flush_backlogs *flush_backlogs_alloc(void)
6474 {
6475 return kmalloc(struct_size_t(struct flush_backlogs, w, nr_cpu_ids),
6476 GFP_KERNEL);
6477 }
6478
6479 static struct flush_backlogs *flush_backlogs_fallback;
6480 static DEFINE_MUTEX(flush_backlogs_mutex);
6481
6482 static void flush_all_backlogs(void)
6483 {
6484 struct flush_backlogs *ptr = flush_backlogs_alloc();
6485 unsigned int cpu;
6486
6487 if (!ptr) {
6488 mutex_lock(&flush_backlogs_mutex);
6489 ptr = flush_backlogs_fallback;
6490 }
6491 cpumask_clear(&ptr->flush_cpus);
6492
6493 cpus_read_lock();
6494
6495 for_each_online_cpu(cpu) {
6496 if (flush_required(cpu)) {
6497 INIT_WORK(&ptr->w[cpu], flush_backlog);
6498 queue_work_on(cpu, system_highpri_wq, &ptr->w[cpu]);
6499 __cpumask_set_cpu(cpu, &ptr->flush_cpus);
6500 }
6501 }
6502
6503 /* we can have in flight packet[s] on the cpus we are not flushing,
6504 * synchronize_net() in unregister_netdevice_many() will take care of
6505 * them.
6506 */
6507 for_each_cpu(cpu, &ptr->flush_cpus)
6508 flush_work(&ptr->w[cpu]);
6509
6510 cpus_read_unlock();
6511
6512 if (ptr != flush_backlogs_fallback)
6513 kfree(ptr);
6514 else
6515 mutex_unlock(&flush_backlogs_mutex);
6516 }
6517
6518 static void net_rps_send_ipi(struct softnet_data *remsd)
6519 {
6520 #ifdef CONFIG_RPS
6521 while (remsd) {
6522 struct softnet_data *next = remsd->rps_ipi_next;
6523
6524 if (cpu_online(remsd->cpu))
6525 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6526 remsd = next;
6527 }
6528 #endif
6529 }
6530
6531 /*
6532 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6533 * Note: called with local irq disabled, but exits with local irq enabled.
6534 */
6535 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6536 {
6537 #ifdef CONFIG_RPS
6538 struct softnet_data *remsd = sd->rps_ipi_list;
6539
6540 if (!use_backlog_threads() && remsd) {
6541 sd->rps_ipi_list = NULL;
6542
6543 local_irq_enable();
6544
6545 /* Send pending IPI's to kick RPS processing on remote cpus. */
6546 net_rps_send_ipi(remsd);
6547 } else
6548 #endif
6549 local_irq_enable();
6550 }
6551
6552 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6553 {
6554 #ifdef CONFIG_RPS
6555 return !use_backlog_threads() && sd->rps_ipi_list;
6556 #else
6557 return false;
6558 #endif
6559 }
6560
6561 static int process_backlog(struct napi_struct *napi, int quota)
6562 {
6563 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6564 bool again = true;
6565 int work = 0;
6566
6567 /* Check if we have pending ipi, its better to send them now,
6568 * not waiting net_rx_action() end.
6569 */
6570 if (sd_has_rps_ipi_waiting(sd)) {
6571 local_irq_disable();
6572 net_rps_action_and_irq_enable(sd);
6573 }
6574
6575 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
6576 while (again) {
6577 struct sk_buff *skb;
6578
6579 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6580 while ((skb = __skb_dequeue(&sd->process_queue))) {
6581 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6582 rcu_read_lock();
6583 __netif_receive_skb(skb);
6584 rcu_read_unlock();
6585 if (++work >= quota) {
6586 rps_input_queue_head_add(sd, work);
6587 return work;
6588 }
6589
6590 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6591 }
6592 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6593
6594 backlog_lock_irq_disable(sd);
6595 if (skb_queue_empty(&sd->input_pkt_queue)) {
6596 /*
6597 * Inline a custom version of __napi_complete().
6598 * only current cpu owns and manipulates this napi,
6599 * and NAPI_STATE_SCHED is the only possible flag set
6600 * on backlog.
6601 * We can use a plain write instead of clear_bit(),
6602 * and we dont need an smp_mb() memory barrier.
6603 */
6604 napi->state &= NAPIF_STATE_THREADED;
6605 again = false;
6606 } else {
6607 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6608 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6609 &sd->process_queue);
6610 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6611 }
6612 backlog_unlock_irq_enable(sd);
6613 }
6614
6615 if (work)
6616 rps_input_queue_head_add(sd, work);
6617 return work;
6618 }
6619
6620 /**
6621 * __napi_schedule - schedule for receive
6622 * @n: entry to schedule
6623 *
6624 * The entry's receive function will be scheduled to run.
6625 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6626 */
6627 void __napi_schedule(struct napi_struct *n)
6628 {
6629 unsigned long flags;
6630
6631 local_irq_save(flags);
6632 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6633 local_irq_restore(flags);
6634 }
6635 EXPORT_SYMBOL(__napi_schedule);
6636
6637 /**
6638 * napi_schedule_prep - check if napi can be scheduled
6639 * @n: napi context
6640 *
6641 * Test if NAPI routine is already running, and if not mark
6642 * it as running. This is used as a condition variable to
6643 * insure only one NAPI poll instance runs. We also make
6644 * sure there is no pending NAPI disable.
6645 */
6646 bool napi_schedule_prep(struct napi_struct *n)
6647 {
6648 unsigned long new, val = READ_ONCE(n->state);
6649
6650 do {
6651 if (unlikely(val & NAPIF_STATE_DISABLE))
6652 return false;
6653 new = val | NAPIF_STATE_SCHED;
6654
6655 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6656 * This was suggested by Alexander Duyck, as compiler
6657 * emits better code than :
6658 * if (val & NAPIF_STATE_SCHED)
6659 * new |= NAPIF_STATE_MISSED;
6660 */
6661 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6662 NAPIF_STATE_MISSED;
6663 } while (!try_cmpxchg(&n->state, &val, new));
6664
6665 return !(val & NAPIF_STATE_SCHED);
6666 }
6667 EXPORT_SYMBOL(napi_schedule_prep);
6668
6669 /**
6670 * __napi_schedule_irqoff - schedule for receive
6671 * @n: entry to schedule
6672 *
6673 * Variant of __napi_schedule() assuming hard irqs are masked.
6674 *
6675 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6676 * because the interrupt disabled assumption might not be true
6677 * due to force-threaded interrupts and spinlock substitution.
6678 */
6679 void __napi_schedule_irqoff(struct napi_struct *n)
6680 {
6681 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6682 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6683 else
6684 __napi_schedule(n);
6685 }
6686 EXPORT_SYMBOL(__napi_schedule_irqoff);
6687
6688 bool napi_complete_done(struct napi_struct *n, int work_done)
6689 {
6690 unsigned long flags, val, new, timeout = 0;
6691 bool ret = true;
6692
6693 /*
6694 * 1) Don't let napi dequeue from the cpu poll list
6695 * just in case its running on a different cpu.
6696 * 2) If we are busy polling, do nothing here, we have
6697 * the guarantee we will be called later.
6698 */
6699 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6700 NAPIF_STATE_IN_BUSY_POLL)))
6701 return false;
6702
6703 if (work_done) {
6704 if (n->gro.bitmask)
6705 timeout = napi_get_gro_flush_timeout(n);
6706 n->defer_hard_irqs_count = napi_get_defer_hard_irqs(n);
6707 }
6708 if (n->defer_hard_irqs_count > 0) {
6709 n->defer_hard_irqs_count--;
6710 timeout = napi_get_gro_flush_timeout(n);
6711 if (timeout)
6712 ret = false;
6713 }
6714
6715 /*
6716 * When the NAPI instance uses a timeout and keeps postponing
6717 * it, we need to bound somehow the time packets are kept in
6718 * the GRO layer.
6719 */
6720 gro_flush_normal(&n->gro, !!timeout);
6721
6722 if (unlikely(!list_empty(&n->poll_list))) {
6723 /* If n->poll_list is not empty, we need to mask irqs */
6724 local_irq_save(flags);
6725 list_del_init(&n->poll_list);
6726 local_irq_restore(flags);
6727 }
6728 WRITE_ONCE(n->list_owner, -1);
6729
6730 val = READ_ONCE(n->state);
6731 do {
6732 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6733
6734 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6735 NAPIF_STATE_SCHED_THREADED |
6736 NAPIF_STATE_PREFER_BUSY_POLL);
6737
6738 /* If STATE_MISSED was set, leave STATE_SCHED set,
6739 * because we will call napi->poll() one more time.
6740 * This C code was suggested by Alexander Duyck to help gcc.
6741 */
6742 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6743 NAPIF_STATE_SCHED;
6744 } while (!try_cmpxchg(&n->state, &val, new));
6745
6746 if (unlikely(val & NAPIF_STATE_MISSED)) {
6747 __napi_schedule(n);
6748 return false;
6749 }
6750
6751 if (timeout)
6752 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6753 HRTIMER_MODE_REL_PINNED);
6754 return ret;
6755 }
6756 EXPORT_SYMBOL(napi_complete_done);
6757
6758 static void skb_defer_free_flush(void)
6759 {
6760 struct llist_node *free_list;
6761 struct sk_buff *skb, *next;
6762 struct skb_defer_node *sdn;
6763 int node;
6764
6765 for_each_node(node) {
6766 sdn = this_cpu_ptr(net_hotdata.skb_defer_nodes) + node;
6767
6768 if (llist_empty(&sdn->defer_list))
6769 continue;
6770 atomic_long_set(&sdn->defer_count, 0);
6771 free_list = llist_del_all(&sdn->defer_list);
6772
6773 llist_for_each_entry_safe(skb, next, free_list, ll_node) {
6774 napi_consume_skb(skb, 1);
6775 }
6776 }
6777 }
6778
6779 #if defined(CONFIG_NET_RX_BUSY_POLL)
6780
6781 static void __busy_poll_stop(struct napi_struct *napi, unsigned long timeout)
6782 {
6783 if (!timeout) {
6784 gro_normal_list(&napi->gro);
6785 __napi_schedule(napi);
6786 return;
6787 }
6788
6789 /* Flush too old packets. If HZ < 1000, flush all packets */
6790 gro_flush_normal(&napi->gro, HZ >= 1000);
6791
6792 clear_bit(NAPI_STATE_SCHED, &napi->state);
6793 hrtimer_start(&napi->timer, ns_to_ktime(timeout),
6794 HRTIMER_MODE_REL_PINNED);
6795 }
6796
6797 enum {
6798 NAPI_F_PREFER_BUSY_POLL = 1,
6799 NAPI_F_END_ON_RESCHED = 2,
6800 };
6801
6802 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock,
6803 unsigned flags, u16 budget)
6804 {
6805 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6806 unsigned long timeout = 0;
6807 int rc;
6808
6809 /* Busy polling means there is a high chance device driver hard irq
6810 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6811 * set in napi_schedule_prep().
6812 * Since we are about to call napi->poll() once more, we can safely
6813 * clear NAPI_STATE_MISSED.
6814 *
6815 * Note: x86 could use a single "lock and ..." instruction
6816 * to perform these two clear_bit()
6817 */
6818 clear_bit(NAPI_STATE_MISSED, &napi->state);
6819 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6820
6821 local_bh_disable();
6822 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6823
6824 if (flags & NAPI_F_PREFER_BUSY_POLL) {
6825 napi->defer_hard_irqs_count = napi_get_defer_hard_irqs(napi);
6826 if (napi->defer_hard_irqs_count) {
6827 /* A short enough gro flush timeout and long enough
6828 * poll can result in timer firing too early.
6829 * Timer will be armed later if necessary.
6830 */
6831 timeout = napi_get_gro_flush_timeout(napi);
6832 }
6833 }
6834
6835 /* All we really want here is to re-enable device interrupts.
6836 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6837 */
6838 rc = napi->poll(napi, budget);
6839 /* We can't gro_normal_list() here, because napi->poll() might have
6840 * rearmed the napi (napi_complete_done()) in which case it could
6841 * already be running on another CPU.
6842 */
6843 trace_napi_poll(napi, rc, budget);
6844 netpoll_poll_unlock(have_poll_lock);
6845 if (rc == budget)
6846 __busy_poll_stop(napi, timeout);
6847 bpf_net_ctx_clear(bpf_net_ctx);
6848 local_bh_enable();
6849 }
6850
6851 static void __napi_busy_loop(unsigned int napi_id,
6852 bool (*loop_end)(void *, unsigned long),
6853 void *loop_end_arg, unsigned flags, u16 budget)
6854 {
6855 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6856 int (*napi_poll)(struct napi_struct *napi, int budget);
6857 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6858 void *have_poll_lock = NULL;
6859 struct napi_struct *napi;
6860
6861 WARN_ON_ONCE(!rcu_read_lock_held());
6862
6863 restart:
6864 napi_poll = NULL;
6865
6866 napi = napi_by_id(napi_id);
6867 if (!napi)
6868 return;
6869
6870 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6871 preempt_disable();
6872 for (;;) {
6873 int work = 0;
6874
6875 local_bh_disable();
6876 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6877 if (!napi_poll) {
6878 unsigned long val = READ_ONCE(napi->state);
6879
6880 /* If multiple threads are competing for this napi,
6881 * we avoid dirtying napi->state as much as we can.
6882 */
6883 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6884 NAPIF_STATE_IN_BUSY_POLL)) {
6885 if (flags & NAPI_F_PREFER_BUSY_POLL)
6886 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6887 goto count;
6888 }
6889 if (cmpxchg(&napi->state, val,
6890 val | NAPIF_STATE_IN_BUSY_POLL |
6891 NAPIF_STATE_SCHED) != val) {
6892 if (flags & NAPI_F_PREFER_BUSY_POLL)
6893 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6894 goto count;
6895 }
6896 have_poll_lock = netpoll_poll_lock(napi);
6897 napi_poll = napi->poll;
6898 }
6899 work = napi_poll(napi, budget);
6900 trace_napi_poll(napi, work, budget);
6901 gro_normal_list(&napi->gro);
6902 count:
6903 if (work > 0)
6904 __NET_ADD_STATS(dev_net(napi->dev),
6905 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6906 skb_defer_free_flush();
6907 bpf_net_ctx_clear(bpf_net_ctx);
6908 local_bh_enable();
6909
6910 if (!loop_end || loop_end(loop_end_arg, start_time))
6911 break;
6912
6913 if (unlikely(need_resched())) {
6914 if (flags & NAPI_F_END_ON_RESCHED)
6915 break;
6916 if (napi_poll)
6917 busy_poll_stop(napi, have_poll_lock, flags, budget);
6918 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6919 preempt_enable();
6920 rcu_read_unlock();
6921 cond_resched();
6922 rcu_read_lock();
6923 if (loop_end(loop_end_arg, start_time))
6924 return;
6925 goto restart;
6926 }
6927 cpu_relax();
6928 }
6929 if (napi_poll)
6930 busy_poll_stop(napi, have_poll_lock, flags, budget);
6931 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6932 preempt_enable();
6933 }
6934
6935 void napi_busy_loop_rcu(unsigned int napi_id,
6936 bool (*loop_end)(void *, unsigned long),
6937 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6938 {
6939 unsigned flags = NAPI_F_END_ON_RESCHED;
6940
6941 if (prefer_busy_poll)
6942 flags |= NAPI_F_PREFER_BUSY_POLL;
6943
6944 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6945 }
6946
6947 void napi_busy_loop(unsigned int napi_id,
6948 bool (*loop_end)(void *, unsigned long),
6949 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6950 {
6951 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0;
6952
6953 rcu_read_lock();
6954 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6955 rcu_read_unlock();
6956 }
6957 EXPORT_SYMBOL(napi_busy_loop);
6958
6959 void napi_suspend_irqs(unsigned int napi_id)
6960 {
6961 struct napi_struct *napi;
6962
6963 rcu_read_lock();
6964 napi = napi_by_id(napi_id);
6965 if (napi) {
6966 unsigned long timeout = napi_get_irq_suspend_timeout(napi);
6967
6968 if (timeout)
6969 hrtimer_start(&napi->timer, ns_to_ktime(timeout),
6970 HRTIMER_MODE_REL_PINNED);
6971 }
6972 rcu_read_unlock();
6973 }
6974
6975 void napi_resume_irqs(unsigned int napi_id)
6976 {
6977 struct napi_struct *napi;
6978
6979 rcu_read_lock();
6980 napi = napi_by_id(napi_id);
6981 if (napi) {
6982 /* If irq_suspend_timeout is set to 0 between the call to
6983 * napi_suspend_irqs and now, the original value still
6984 * determines the safety timeout as intended and napi_watchdog
6985 * will resume irq processing.
6986 */
6987 if (napi_get_irq_suspend_timeout(napi)) {
6988 local_bh_disable();
6989 napi_schedule(napi);
6990 local_bh_enable();
6991 }
6992 }
6993 rcu_read_unlock();
6994 }
6995
6996 #endif /* CONFIG_NET_RX_BUSY_POLL */
6997
6998 static void __napi_hash_add_with_id(struct napi_struct *napi,
6999 unsigned int napi_id)
7000 {
7001 napi->gro.cached_napi_id = napi_id;
7002
7003 WRITE_ONCE(napi->napi_id, napi_id);
7004 hlist_add_head_rcu(&napi->napi_hash_node,
7005 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
7006 }
7007
7008 static void napi_hash_add_with_id(struct napi_struct *napi,
7009 unsigned int napi_id)
7010 {
7011 unsigned long flags;
7012
7013 spin_lock_irqsave(&napi_hash_lock, flags);
7014 WARN_ON_ONCE(napi_by_id(napi_id));
7015 __napi_hash_add_with_id(napi, napi_id);
7016 spin_unlock_irqrestore(&napi_hash_lock, flags);
7017 }
7018
7019 static void napi_hash_add(struct napi_struct *napi)
7020 {
7021 unsigned long flags;
7022
7023 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
7024 return;
7025
7026 spin_lock_irqsave(&napi_hash_lock, flags);
7027
7028 /* 0..NR_CPUS range is reserved for sender_cpu use */
7029 do {
7030 if (unlikely(!napi_id_valid(++napi_gen_id)))
7031 napi_gen_id = MIN_NAPI_ID;
7032 } while (napi_by_id(napi_gen_id));
7033
7034 __napi_hash_add_with_id(napi, napi_gen_id);
7035
7036 spin_unlock_irqrestore(&napi_hash_lock, flags);
7037 }
7038
7039 /* Warning : caller is responsible to make sure rcu grace period
7040 * is respected before freeing memory containing @napi
7041 */
7042 static void napi_hash_del(struct napi_struct *napi)
7043 {
7044 unsigned long flags;
7045
7046 spin_lock_irqsave(&napi_hash_lock, flags);
7047
7048 hlist_del_init_rcu(&napi->napi_hash_node);
7049
7050 spin_unlock_irqrestore(&napi_hash_lock, flags);
7051 }
7052
7053 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
7054 {
7055 struct napi_struct *napi;
7056
7057 napi = container_of(timer, struct napi_struct, timer);
7058
7059 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
7060 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
7061 */
7062 if (!napi_disable_pending(napi) &&
7063 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
7064 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
7065 __napi_schedule_irqoff(napi);
7066 }
7067
7068 return HRTIMER_NORESTART;
7069 }
7070
7071 static void napi_stop_kthread(struct napi_struct *napi)
7072 {
7073 unsigned long val, new;
7074
7075 /* Wait until the napi STATE_THREADED is unset. */
7076 while (true) {
7077 val = READ_ONCE(napi->state);
7078
7079 /* If napi kthread own this napi or the napi is idle,
7080 * STATE_THREADED can be unset here.
7081 */
7082 if ((val & NAPIF_STATE_SCHED_THREADED) ||
7083 !(val & NAPIF_STATE_SCHED)) {
7084 new = val & (~NAPIF_STATE_THREADED);
7085 } else {
7086 msleep(20);
7087 continue;
7088 }
7089
7090 if (try_cmpxchg(&napi->state, &val, new))
7091 break;
7092 }
7093
7094 /* Once STATE_THREADED is unset, wait for SCHED_THREADED to be unset by
7095 * the kthread.
7096 */
7097 while (true) {
7098 if (!test_bit(NAPI_STATE_SCHED_THREADED, &napi->state))
7099 break;
7100
7101 msleep(20);
7102 }
7103
7104 kthread_stop(napi->thread);
7105 napi->thread = NULL;
7106 }
7107
7108 int napi_set_threaded(struct napi_struct *napi,
7109 enum netdev_napi_threaded threaded)
7110 {
7111 if (threaded) {
7112 if (!napi->thread) {
7113 int err = napi_kthread_create(napi);
7114
7115 if (err)
7116 return err;
7117 }
7118 }
7119
7120 if (napi->config)
7121 napi->config->threaded = threaded;
7122
7123 /* Setting/unsetting threaded mode on a napi might not immediately
7124 * take effect, if the current napi instance is actively being
7125 * polled. In this case, the switch between threaded mode and
7126 * softirq mode will happen in the next round of napi_schedule().
7127 * This should not cause hiccups/stalls to the live traffic.
7128 */
7129 if (!threaded && napi->thread) {
7130 napi_stop_kthread(napi);
7131 } else {
7132 /* Make sure kthread is created before THREADED bit is set. */
7133 smp_mb__before_atomic();
7134 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
7135 }
7136
7137 return 0;
7138 }
7139
7140 int netif_set_threaded(struct net_device *dev,
7141 enum netdev_napi_threaded threaded)
7142 {
7143 struct napi_struct *napi;
7144 int i, err = 0;
7145
7146 netdev_assert_locked_or_invisible(dev);
7147
7148 if (threaded) {
7149 list_for_each_entry(napi, &dev->napi_list, dev_list) {
7150 if (!napi->thread) {
7151 err = napi_kthread_create(napi);
7152 if (err) {
7153 threaded = NETDEV_NAPI_THREADED_DISABLED;
7154 break;
7155 }
7156 }
7157 }
7158 }
7159
7160 WRITE_ONCE(dev->threaded, threaded);
7161
7162 /* The error should not occur as the kthreads are already created. */
7163 list_for_each_entry(napi, &dev->napi_list, dev_list)
7164 WARN_ON_ONCE(napi_set_threaded(napi, threaded));
7165
7166 /* Override the config for all NAPIs even if currently not listed */
7167 for (i = 0; i < dev->num_napi_configs; i++)
7168 dev->napi_config[i].threaded = threaded;
7169
7170 return err;
7171 }
7172
7173 /**
7174 * netif_threaded_enable() - enable threaded NAPIs
7175 * @dev: net_device instance
7176 *
7177 * Enable threaded mode for the NAPI instances of the device. This may be useful
7178 * for devices where multiple NAPI instances get scheduled by a single
7179 * interrupt. Threaded NAPI allows moving the NAPI processing to cores other
7180 * than the core where IRQ is mapped.
7181 *
7182 * This function should be called before @dev is registered.
7183 */
7184 void netif_threaded_enable(struct net_device *dev)
7185 {
7186 WARN_ON_ONCE(netif_set_threaded(dev, NETDEV_NAPI_THREADED_ENABLED));
7187 }
7188 EXPORT_SYMBOL(netif_threaded_enable);
7189
7190 /**
7191 * netif_queue_set_napi - Associate queue with the napi
7192 * @dev: device to which NAPI and queue belong
7193 * @queue_index: Index of queue
7194 * @type: queue type as RX or TX
7195 * @napi: NAPI context, pass NULL to clear previously set NAPI
7196 *
7197 * Set queue with its corresponding napi context. This should be done after
7198 * registering the NAPI handler for the queue-vector and the queues have been
7199 * mapped to the corresponding interrupt vector.
7200 */
7201 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index,
7202 enum netdev_queue_type type, struct napi_struct *napi)
7203 {
7204 struct netdev_rx_queue *rxq;
7205 struct netdev_queue *txq;
7206
7207 if (WARN_ON_ONCE(napi && !napi->dev))
7208 return;
7209 netdev_ops_assert_locked_or_invisible(dev);
7210
7211 switch (type) {
7212 case NETDEV_QUEUE_TYPE_RX:
7213 rxq = __netif_get_rx_queue(dev, queue_index);
7214 rxq->napi = napi;
7215 return;
7216 case NETDEV_QUEUE_TYPE_TX:
7217 txq = netdev_get_tx_queue(dev, queue_index);
7218 txq->napi = napi;
7219 return;
7220 default:
7221 return;
7222 }
7223 }
7224 EXPORT_SYMBOL(netif_queue_set_napi);
7225
7226 static void
7227 netif_napi_irq_notify(struct irq_affinity_notify *notify,
7228 const cpumask_t *mask)
7229 {
7230 struct napi_struct *napi =
7231 container_of(notify, struct napi_struct, notify);
7232 #ifdef CONFIG_RFS_ACCEL
7233 struct cpu_rmap *rmap = napi->dev->rx_cpu_rmap;
7234 int err;
7235 #endif
7236
7237 if (napi->config && napi->dev->irq_affinity_auto)
7238 cpumask_copy(&napi->config->affinity_mask, mask);
7239
7240 #ifdef CONFIG_RFS_ACCEL
7241 if (napi->dev->rx_cpu_rmap_auto) {
7242 err = cpu_rmap_update(rmap, napi->napi_rmap_idx, mask);
7243 if (err)
7244 netdev_warn(napi->dev, "RMAP update failed (%d)\n",
7245 err);
7246 }
7247 #endif
7248 }
7249
7250 #ifdef CONFIG_RFS_ACCEL
7251 static void netif_napi_affinity_release(struct kref *ref)
7252 {
7253 struct napi_struct *napi =
7254 container_of(ref, struct napi_struct, notify.kref);
7255 struct cpu_rmap *rmap = napi->dev->rx_cpu_rmap;
7256
7257 netdev_assert_locked(napi->dev);
7258 WARN_ON(test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER,
7259 &napi->state));
7260
7261 if (!napi->dev->rx_cpu_rmap_auto)
7262 return;
7263 rmap->obj[napi->napi_rmap_idx] = NULL;
7264 napi->napi_rmap_idx = -1;
7265 cpu_rmap_put(rmap);
7266 }
7267
7268 int netif_enable_cpu_rmap(struct net_device *dev, unsigned int num_irqs)
7269 {
7270 if (dev->rx_cpu_rmap_auto)
7271 return 0;
7272
7273 dev->rx_cpu_rmap = alloc_irq_cpu_rmap(num_irqs);
7274 if (!dev->rx_cpu_rmap)
7275 return -ENOMEM;
7276
7277 dev->rx_cpu_rmap_auto = true;
7278 return 0;
7279 }
7280 EXPORT_SYMBOL(netif_enable_cpu_rmap);
7281
7282 static void netif_del_cpu_rmap(struct net_device *dev)
7283 {
7284 struct cpu_rmap *rmap = dev->rx_cpu_rmap;
7285
7286 if (!dev->rx_cpu_rmap_auto)
7287 return;
7288
7289 /* Free the rmap */
7290 cpu_rmap_put(rmap);
7291 dev->rx_cpu_rmap = NULL;
7292 dev->rx_cpu_rmap_auto = false;
7293 }
7294
7295 #else
7296 static void netif_napi_affinity_release(struct kref *ref)
7297 {
7298 }
7299
7300 int netif_enable_cpu_rmap(struct net_device *dev, unsigned int num_irqs)
7301 {
7302 return 0;
7303 }
7304 EXPORT_SYMBOL(netif_enable_cpu_rmap);
7305
7306 static void netif_del_cpu_rmap(struct net_device *dev)
7307 {
7308 }
7309 #endif
7310
7311 void netif_set_affinity_auto(struct net_device *dev)
7312 {
7313 unsigned int i, maxqs, numa;
7314
7315 maxqs = max(dev->num_tx_queues, dev->num_rx_queues);
7316 numa = dev_to_node(&dev->dev);
7317
7318 for (i = 0; i < maxqs; i++)
7319 cpumask_set_cpu(cpumask_local_spread(i, numa),
7320 &dev->napi_config[i].affinity_mask);
7321
7322 dev->irq_affinity_auto = true;
7323 }
7324 EXPORT_SYMBOL(netif_set_affinity_auto);
7325
7326 void netif_napi_set_irq_locked(struct napi_struct *napi, int irq)
7327 {
7328 int rc;
7329
7330 netdev_assert_locked_or_invisible(napi->dev);
7331
7332 if (napi->irq == irq)
7333 return;
7334
7335 /* Remove existing resources */
7336 if (test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state))
7337 irq_set_affinity_notifier(napi->irq, NULL);
7338
7339 napi->irq = irq;
7340 if (irq < 0 ||
7341 (!napi->dev->rx_cpu_rmap_auto && !napi->dev->irq_affinity_auto))
7342 return;
7343
7344 /* Abort for buggy drivers */
7345 if (napi->dev->irq_affinity_auto && WARN_ON_ONCE(!napi->config))
7346 return;
7347
7348 #ifdef CONFIG_RFS_ACCEL
7349 if (napi->dev->rx_cpu_rmap_auto) {
7350 rc = cpu_rmap_add(napi->dev->rx_cpu_rmap, napi);
7351 if (rc < 0)
7352 return;
7353
7354 cpu_rmap_get(napi->dev->rx_cpu_rmap);
7355 napi->napi_rmap_idx = rc;
7356 }
7357 #endif
7358
7359 /* Use core IRQ notifier */
7360 napi->notify.notify = netif_napi_irq_notify;
7361 napi->notify.release = netif_napi_affinity_release;
7362 rc = irq_set_affinity_notifier(irq, &napi->notify);
7363 if (rc) {
7364 netdev_warn(napi->dev, "Unable to set IRQ notifier (%d)\n",
7365 rc);
7366 goto put_rmap;
7367 }
7368
7369 set_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state);
7370 return;
7371
7372 put_rmap:
7373 #ifdef CONFIG_RFS_ACCEL
7374 if (napi->dev->rx_cpu_rmap_auto) {
7375 napi->dev->rx_cpu_rmap->obj[napi->napi_rmap_idx] = NULL;
7376 cpu_rmap_put(napi->dev->rx_cpu_rmap);
7377 napi->napi_rmap_idx = -1;
7378 }
7379 #endif
7380 napi->notify.notify = NULL;
7381 napi->notify.release = NULL;
7382 }
7383 EXPORT_SYMBOL(netif_napi_set_irq_locked);
7384
7385 static void napi_restore_config(struct napi_struct *n)
7386 {
7387 n->defer_hard_irqs = n->config->defer_hard_irqs;
7388 n->gro_flush_timeout = n->config->gro_flush_timeout;
7389 n->irq_suspend_timeout = n->config->irq_suspend_timeout;
7390
7391 if (n->dev->irq_affinity_auto &&
7392 test_bit(NAPI_STATE_HAS_NOTIFIER, &n->state))
7393 irq_set_affinity(n->irq, &n->config->affinity_mask);
7394
7395 /* a NAPI ID might be stored in the config, if so use it. if not, use
7396 * napi_hash_add to generate one for us.
7397 */
7398 if (n->config->napi_id) {
7399 napi_hash_add_with_id(n, n->config->napi_id);
7400 } else {
7401 napi_hash_add(n);
7402 n->config->napi_id = n->napi_id;
7403 }
7404
7405 WARN_ON_ONCE(napi_set_threaded(n, n->config->threaded));
7406 }
7407
7408 static void napi_save_config(struct napi_struct *n)
7409 {
7410 n->config->defer_hard_irqs = n->defer_hard_irqs;
7411 n->config->gro_flush_timeout = n->gro_flush_timeout;
7412 n->config->irq_suspend_timeout = n->irq_suspend_timeout;
7413 napi_hash_del(n);
7414 }
7415
7416 /* Netlink wants the NAPI list to be sorted by ID, if adding a NAPI which will
7417 * inherit an existing ID try to insert it at the right position.
7418 */
7419 static void
7420 netif_napi_dev_list_add(struct net_device *dev, struct napi_struct *napi)
7421 {
7422 unsigned int new_id, pos_id;
7423 struct list_head *higher;
7424 struct napi_struct *pos;
7425
7426 new_id = UINT_MAX;
7427 if (napi->config && napi->config->napi_id)
7428 new_id = napi->config->napi_id;
7429
7430 higher = &dev->napi_list;
7431 list_for_each_entry(pos, &dev->napi_list, dev_list) {
7432 if (napi_id_valid(pos->napi_id))
7433 pos_id = pos->napi_id;
7434 else if (pos->config)
7435 pos_id = pos->config->napi_id;
7436 else
7437 pos_id = UINT_MAX;
7438
7439 if (pos_id <= new_id)
7440 break;
7441 higher = &pos->dev_list;
7442 }
7443 list_add_rcu(&napi->dev_list, higher); /* adds after higher */
7444 }
7445
7446 /* Double check that napi_get_frags() allocates skbs with
7447 * skb->head being backed by slab, not a page fragment.
7448 * This is to make sure bug fixed in 3226b158e67c
7449 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
7450 * does not accidentally come back.
7451 */
7452 static void napi_get_frags_check(struct napi_struct *napi)
7453 {
7454 struct sk_buff *skb;
7455
7456 local_bh_disable();
7457 skb = napi_get_frags(napi);
7458 WARN_ON_ONCE(skb && skb->head_frag);
7459 napi_free_frags(napi);
7460 local_bh_enable();
7461 }
7462
7463 void netif_napi_add_weight_locked(struct net_device *dev,
7464 struct napi_struct *napi,
7465 int (*poll)(struct napi_struct *, int),
7466 int weight)
7467 {
7468 netdev_assert_locked(dev);
7469 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
7470 return;
7471
7472 INIT_LIST_HEAD(&napi->poll_list);
7473 INIT_HLIST_NODE(&napi->napi_hash_node);
7474 hrtimer_setup(&napi->timer, napi_watchdog, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
7475 gro_init(&napi->gro);
7476 napi->skb = NULL;
7477 napi->poll = poll;
7478 if (weight > NAPI_POLL_WEIGHT)
7479 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
7480 weight);
7481 napi->weight = weight;
7482 napi->dev = dev;
7483 #ifdef CONFIG_NETPOLL
7484 napi->poll_owner = -1;
7485 #endif
7486 napi->list_owner = -1;
7487 set_bit(NAPI_STATE_SCHED, &napi->state);
7488 set_bit(NAPI_STATE_NPSVC, &napi->state);
7489 netif_napi_dev_list_add(dev, napi);
7490
7491 /* default settings from sysfs are applied to all NAPIs. any per-NAPI
7492 * configuration will be loaded in napi_enable
7493 */
7494 napi_set_defer_hard_irqs(napi, READ_ONCE(dev->napi_defer_hard_irqs));
7495 napi_set_gro_flush_timeout(napi, READ_ONCE(dev->gro_flush_timeout));
7496
7497 napi_get_frags_check(napi);
7498 /* Create kthread for this napi if dev->threaded is set.
7499 * Clear dev->threaded if kthread creation failed so that
7500 * threaded mode will not be enabled in napi_enable().
7501 */
7502 if (napi_get_threaded_config(dev, napi))
7503 if (napi_kthread_create(napi))
7504 dev->threaded = NETDEV_NAPI_THREADED_DISABLED;
7505 netif_napi_set_irq_locked(napi, -1);
7506 }
7507 EXPORT_SYMBOL(netif_napi_add_weight_locked);
7508
7509 void napi_disable_locked(struct napi_struct *n)
7510 {
7511 unsigned long val, new;
7512
7513 might_sleep();
7514 netdev_assert_locked(n->dev);
7515
7516 set_bit(NAPI_STATE_DISABLE, &n->state);
7517
7518 val = READ_ONCE(n->state);
7519 do {
7520 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
7521 usleep_range(20, 200);
7522 val = READ_ONCE(n->state);
7523 }
7524
7525 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
7526 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
7527 } while (!try_cmpxchg(&n->state, &val, new));
7528
7529 hrtimer_cancel(&n->timer);
7530
7531 if (n->config)
7532 napi_save_config(n);
7533 else
7534 napi_hash_del(n);
7535
7536 clear_bit(NAPI_STATE_DISABLE, &n->state);
7537 }
7538 EXPORT_SYMBOL(napi_disable_locked);
7539
7540 /**
7541 * napi_disable() - prevent NAPI from scheduling
7542 * @n: NAPI context
7543 *
7544 * Stop NAPI from being scheduled on this context.
7545 * Waits till any outstanding processing completes.
7546 * Takes netdev_lock() for associated net_device.
7547 */
7548 void napi_disable(struct napi_struct *n)
7549 {
7550 netdev_lock(n->dev);
7551 napi_disable_locked(n);
7552 netdev_unlock(n->dev);
7553 }
7554 EXPORT_SYMBOL(napi_disable);
7555
7556 void napi_enable_locked(struct napi_struct *n)
7557 {
7558 unsigned long new, val = READ_ONCE(n->state);
7559
7560 if (n->config)
7561 napi_restore_config(n);
7562 else
7563 napi_hash_add(n);
7564
7565 do {
7566 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
7567
7568 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
7569 if (n->dev->threaded && n->thread)
7570 new |= NAPIF_STATE_THREADED;
7571 } while (!try_cmpxchg(&n->state, &val, new));
7572 }
7573 EXPORT_SYMBOL(napi_enable_locked);
7574
7575 /**
7576 * napi_enable() - enable NAPI scheduling
7577 * @n: NAPI context
7578 *
7579 * Enable scheduling of a NAPI instance.
7580 * Must be paired with napi_disable().
7581 * Takes netdev_lock() for associated net_device.
7582 */
7583 void napi_enable(struct napi_struct *n)
7584 {
7585 netdev_lock(n->dev);
7586 napi_enable_locked(n);
7587 netdev_unlock(n->dev);
7588 }
7589 EXPORT_SYMBOL(napi_enable);
7590
7591 /* Must be called in process context */
7592 void __netif_napi_del_locked(struct napi_struct *napi)
7593 {
7594 netdev_assert_locked(napi->dev);
7595
7596 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
7597 return;
7598
7599 /* Make sure NAPI is disabled (or was never enabled). */
7600 WARN_ON(!test_bit(NAPI_STATE_SCHED, &napi->state));
7601
7602 if (test_and_clear_bit(NAPI_STATE_HAS_NOTIFIER, &napi->state))
7603 irq_set_affinity_notifier(napi->irq, NULL);
7604
7605 if (napi->config) {
7606 napi->index = -1;
7607 napi->config = NULL;
7608 }
7609
7610 list_del_rcu(&napi->dev_list);
7611 napi_free_frags(napi);
7612
7613 gro_cleanup(&napi->gro);
7614
7615 if (napi->thread) {
7616 kthread_stop(napi->thread);
7617 napi->thread = NULL;
7618 }
7619 }
7620 EXPORT_SYMBOL(__netif_napi_del_locked);
7621
7622 static int __napi_poll(struct napi_struct *n, bool *repoll)
7623 {
7624 int work, weight;
7625
7626 weight = n->weight;
7627
7628 /* This NAPI_STATE_SCHED test is for avoiding a race
7629 * with netpoll's poll_napi(). Only the entity which
7630 * obtains the lock and sees NAPI_STATE_SCHED set will
7631 * actually make the ->poll() call. Therefore we avoid
7632 * accidentally calling ->poll() when NAPI is not scheduled.
7633 */
7634 work = 0;
7635 if (napi_is_scheduled(n)) {
7636 work = n->poll(n, weight);
7637 trace_napi_poll(n, work, weight);
7638
7639 xdp_do_check_flushed(n);
7640 }
7641
7642 if (unlikely(work > weight))
7643 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
7644 n->poll, work, weight);
7645
7646 if (likely(work < weight))
7647 return work;
7648
7649 /* Drivers must not modify the NAPI state if they
7650 * consume the entire weight. In such cases this code
7651 * still "owns" the NAPI instance and therefore can
7652 * move the instance around on the list at-will.
7653 */
7654 if (unlikely(napi_disable_pending(n))) {
7655 napi_complete(n);
7656 return work;
7657 }
7658
7659 /* The NAPI context has more processing work, but busy-polling
7660 * is preferred. Exit early.
7661 */
7662 if (napi_prefer_busy_poll(n)) {
7663 if (napi_complete_done(n, work)) {
7664 /* If timeout is not set, we need to make sure
7665 * that the NAPI is re-scheduled.
7666 */
7667 napi_schedule(n);
7668 }
7669 return work;
7670 }
7671
7672 /* Flush too old packets. If HZ < 1000, flush all packets */
7673 gro_flush_normal(&n->gro, HZ >= 1000);
7674
7675 /* Some drivers may have called napi_schedule
7676 * prior to exhausting their budget.
7677 */
7678 if (unlikely(!list_empty(&n->poll_list))) {
7679 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7680 n->dev ? n->dev->name : "backlog");
7681 return work;
7682 }
7683
7684 *repoll = true;
7685
7686 return work;
7687 }
7688
7689 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7690 {
7691 bool do_repoll = false;
7692 void *have;
7693 int work;
7694
7695 list_del_init(&n->poll_list);
7696
7697 have = netpoll_poll_lock(n);
7698
7699 work = __napi_poll(n, &do_repoll);
7700
7701 if (do_repoll) {
7702 #if defined(CONFIG_DEBUG_NET)
7703 if (unlikely(!napi_is_scheduled(n)))
7704 pr_crit("repoll requested for device %s %ps but napi is not scheduled.\n",
7705 n->dev->name, n->poll);
7706 #endif
7707 list_add_tail(&n->poll_list, repoll);
7708 }
7709 netpoll_poll_unlock(have);
7710
7711 return work;
7712 }
7713
7714 static int napi_thread_wait(struct napi_struct *napi)
7715 {
7716 set_current_state(TASK_INTERRUPTIBLE);
7717
7718 while (!kthread_should_stop()) {
7719 /* Testing SCHED_THREADED bit here to make sure the current
7720 * kthread owns this napi and could poll on this napi.
7721 * Testing SCHED bit is not enough because SCHED bit might be
7722 * set by some other busy poll thread or by napi_disable().
7723 */
7724 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) {
7725 WARN_ON(!list_empty(&napi->poll_list));
7726 __set_current_state(TASK_RUNNING);
7727 return 0;
7728 }
7729
7730 schedule();
7731 set_current_state(TASK_INTERRUPTIBLE);
7732 }
7733 __set_current_state(TASK_RUNNING);
7734
7735 return -1;
7736 }
7737
7738 static void napi_threaded_poll_loop(struct napi_struct *napi)
7739 {
7740 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
7741 struct softnet_data *sd;
7742 unsigned long last_qs = jiffies;
7743
7744 for (;;) {
7745 bool repoll = false;
7746 void *have;
7747
7748 local_bh_disable();
7749 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
7750
7751 sd = this_cpu_ptr(&softnet_data);
7752 sd->in_napi_threaded_poll = true;
7753
7754 have = netpoll_poll_lock(napi);
7755 __napi_poll(napi, &repoll);
7756 netpoll_poll_unlock(have);
7757
7758 sd->in_napi_threaded_poll = false;
7759 barrier();
7760
7761 if (sd_has_rps_ipi_waiting(sd)) {
7762 local_irq_disable();
7763 net_rps_action_and_irq_enable(sd);
7764 }
7765 skb_defer_free_flush();
7766 bpf_net_ctx_clear(bpf_net_ctx);
7767 local_bh_enable();
7768
7769 if (!repoll)
7770 break;
7771
7772 rcu_softirq_qs_periodic(last_qs);
7773 cond_resched();
7774 }
7775 }
7776
7777 static int napi_threaded_poll(void *data)
7778 {
7779 struct napi_struct *napi = data;
7780
7781 while (!napi_thread_wait(napi))
7782 napi_threaded_poll_loop(napi);
7783
7784 return 0;
7785 }
7786
7787 static __latent_entropy void net_rx_action(void)
7788 {
7789 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7790 unsigned long time_limit = jiffies +
7791 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
7792 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
7793 int budget = READ_ONCE(net_hotdata.netdev_budget);
7794 LIST_HEAD(list);
7795 LIST_HEAD(repoll);
7796
7797 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
7798 start:
7799 sd->in_net_rx_action = true;
7800 local_irq_disable();
7801 list_splice_init(&sd->poll_list, &list);
7802 local_irq_enable();
7803
7804 for (;;) {
7805 struct napi_struct *n;
7806
7807 skb_defer_free_flush();
7808
7809 if (list_empty(&list)) {
7810 if (list_empty(&repoll)) {
7811 sd->in_net_rx_action = false;
7812 barrier();
7813 /* We need to check if ____napi_schedule()
7814 * had refilled poll_list while
7815 * sd->in_net_rx_action was true.
7816 */
7817 if (!list_empty(&sd->poll_list))
7818 goto start;
7819 if (!sd_has_rps_ipi_waiting(sd))
7820 goto end;
7821 }
7822 break;
7823 }
7824
7825 n = list_first_entry(&list, struct napi_struct, poll_list);
7826 budget -= napi_poll(n, &repoll);
7827
7828 /* If softirq window is exhausted then punt.
7829 * Allow this to run for 2 jiffies since which will allow
7830 * an average latency of 1.5/HZ.
7831 */
7832 if (unlikely(budget <= 0 ||
7833 time_after_eq(jiffies, time_limit))) {
7834 /* Pairs with READ_ONCE() in softnet_seq_show() */
7835 WRITE_ONCE(sd->time_squeeze, sd->time_squeeze + 1);
7836 break;
7837 }
7838 }
7839
7840 local_irq_disable();
7841
7842 list_splice_tail_init(&sd->poll_list, &list);
7843 list_splice_tail(&repoll, &list);
7844 list_splice(&list, &sd->poll_list);
7845 if (!list_empty(&sd->poll_list))
7846 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7847 else
7848 sd->in_net_rx_action = false;
7849
7850 net_rps_action_and_irq_enable(sd);
7851 end:
7852 bpf_net_ctx_clear(bpf_net_ctx);
7853 }
7854
7855 struct netdev_adjacent {
7856 struct net_device *dev;
7857 netdevice_tracker dev_tracker;
7858
7859 /* upper master flag, there can only be one master device per list */
7860 bool master;
7861
7862 /* lookup ignore flag */
7863 bool ignore;
7864
7865 /* counter for the number of times this device was added to us */
7866 u16 ref_nr;
7867
7868 /* private field for the users */
7869 void *private;
7870
7871 struct list_head list;
7872 struct rcu_head rcu;
7873 };
7874
7875 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7876 struct list_head *adj_list)
7877 {
7878 struct netdev_adjacent *adj;
7879
7880 list_for_each_entry(adj, adj_list, list) {
7881 if (adj->dev == adj_dev)
7882 return adj;
7883 }
7884 return NULL;
7885 }
7886
7887 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7888 struct netdev_nested_priv *priv)
7889 {
7890 struct net_device *dev = (struct net_device *)priv->data;
7891
7892 return upper_dev == dev;
7893 }
7894
7895 /**
7896 * netdev_has_upper_dev - Check if device is linked to an upper device
7897 * @dev: device
7898 * @upper_dev: upper device to check
7899 *
7900 * Find out if a device is linked to specified upper device and return true
7901 * in case it is. Note that this checks only immediate upper device,
7902 * not through a complete stack of devices. The caller must hold the RTNL lock.
7903 */
7904 bool netdev_has_upper_dev(struct net_device *dev,
7905 struct net_device *upper_dev)
7906 {
7907 struct netdev_nested_priv priv = {
7908 .data = (void *)upper_dev,
7909 };
7910
7911 ASSERT_RTNL();
7912
7913 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7914 &priv);
7915 }
7916 EXPORT_SYMBOL(netdev_has_upper_dev);
7917
7918 /**
7919 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7920 * @dev: device
7921 * @upper_dev: upper device to check
7922 *
7923 * Find out if a device is linked to specified upper device and return true
7924 * in case it is. Note that this checks the entire upper device chain.
7925 * The caller must hold rcu lock.
7926 */
7927
7928 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7929 struct net_device *upper_dev)
7930 {
7931 struct netdev_nested_priv priv = {
7932 .data = (void *)upper_dev,
7933 };
7934
7935 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7936 &priv);
7937 }
7938 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7939
7940 /**
7941 * netdev_has_any_upper_dev - Check if device is linked to some device
7942 * @dev: device
7943 *
7944 * Find out if a device is linked to an upper device and return true in case
7945 * it is. The caller must hold the RTNL lock.
7946 */
7947 bool netdev_has_any_upper_dev(struct net_device *dev)
7948 {
7949 ASSERT_RTNL();
7950
7951 return !list_empty(&dev->adj_list.upper);
7952 }
7953 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7954
7955 /**
7956 * netdev_master_upper_dev_get - Get master upper device
7957 * @dev: device
7958 *
7959 * Find a master upper device and return pointer to it or NULL in case
7960 * it's not there. The caller must hold the RTNL lock.
7961 */
7962 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7963 {
7964 struct netdev_adjacent *upper;
7965
7966 ASSERT_RTNL();
7967
7968 if (list_empty(&dev->adj_list.upper))
7969 return NULL;
7970
7971 upper = list_first_entry(&dev->adj_list.upper,
7972 struct netdev_adjacent, list);
7973 if (likely(upper->master))
7974 return upper->dev;
7975 return NULL;
7976 }
7977 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7978
7979 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7980 {
7981 struct netdev_adjacent *upper;
7982
7983 ASSERT_RTNL();
7984
7985 if (list_empty(&dev->adj_list.upper))
7986 return NULL;
7987
7988 upper = list_first_entry(&dev->adj_list.upper,
7989 struct netdev_adjacent, list);
7990 if (likely(upper->master) && !upper->ignore)
7991 return upper->dev;
7992 return NULL;
7993 }
7994
7995 /**
7996 * netdev_has_any_lower_dev - Check if device is linked to some device
7997 * @dev: device
7998 *
7999 * Find out if a device is linked to a lower device and return true in case
8000 * it is. The caller must hold the RTNL lock.
8001 */
8002 static bool netdev_has_any_lower_dev(struct net_device *dev)
8003 {
8004 ASSERT_RTNL();
8005
8006 return !list_empty(&dev->adj_list.lower);
8007 }
8008
8009 void *netdev_adjacent_get_private(struct list_head *adj_list)
8010 {
8011 struct netdev_adjacent *adj;
8012
8013 adj = list_entry(adj_list, struct netdev_adjacent, list);
8014
8015 return adj->private;
8016 }
8017 EXPORT_SYMBOL(netdev_adjacent_get_private);
8018
8019 /**
8020 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
8021 * @dev: device
8022 * @iter: list_head ** of the current position
8023 *
8024 * Gets the next device from the dev's upper list, starting from iter
8025 * position. The caller must hold RCU read lock.
8026 */
8027 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
8028 struct list_head **iter)
8029 {
8030 struct netdev_adjacent *upper;
8031
8032 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
8033
8034 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
8035
8036 if (&upper->list == &dev->adj_list.upper)
8037 return NULL;
8038
8039 *iter = &upper->list;
8040
8041 return upper->dev;
8042 }
8043 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
8044
8045 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
8046 struct list_head **iter,
8047 bool *ignore)
8048 {
8049 struct netdev_adjacent *upper;
8050
8051 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
8052
8053 if (&upper->list == &dev->adj_list.upper)
8054 return NULL;
8055
8056 *iter = &upper->list;
8057 *ignore = upper->ignore;
8058
8059 return upper->dev;
8060 }
8061
8062 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
8063 struct list_head **iter)
8064 {
8065 struct netdev_adjacent *upper;
8066
8067 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
8068
8069 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
8070
8071 if (&upper->list == &dev->adj_list.upper)
8072 return NULL;
8073
8074 *iter = &upper->list;
8075
8076 return upper->dev;
8077 }
8078
8079 static int __netdev_walk_all_upper_dev(struct net_device *dev,
8080 int (*fn)(struct net_device *dev,
8081 struct netdev_nested_priv *priv),
8082 struct netdev_nested_priv *priv)
8083 {
8084 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
8085 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
8086 int ret, cur = 0;
8087 bool ignore;
8088
8089 now = dev;
8090 iter = &dev->adj_list.upper;
8091
8092 while (1) {
8093 if (now != dev) {
8094 ret = fn(now, priv);
8095 if (ret)
8096 return ret;
8097 }
8098
8099 next = NULL;
8100 while (1) {
8101 udev = __netdev_next_upper_dev(now, &iter, &ignore);
8102 if (!udev)
8103 break;
8104 if (ignore)
8105 continue;
8106
8107 next = udev;
8108 niter = &udev->adj_list.upper;
8109 dev_stack[cur] = now;
8110 iter_stack[cur++] = iter;
8111 break;
8112 }
8113
8114 if (!next) {
8115 if (!cur)
8116 return 0;
8117 next = dev_stack[--cur];
8118 niter = iter_stack[cur];
8119 }
8120
8121 now = next;
8122 iter = niter;
8123 }
8124
8125 return 0;
8126 }
8127
8128 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
8129 int (*fn)(struct net_device *dev,
8130 struct netdev_nested_priv *priv),
8131 struct netdev_nested_priv *priv)
8132 {
8133 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
8134 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
8135 int ret, cur = 0;
8136
8137 now = dev;
8138 iter = &dev->adj_list.upper;
8139
8140 while (1) {
8141 if (now != dev) {
8142 ret = fn(now, priv);
8143 if (ret)
8144 return ret;
8145 }
8146
8147 next = NULL;
8148 while (1) {
8149 udev = netdev_next_upper_dev_rcu(now, &iter);
8150 if (!udev)
8151 break;
8152
8153 next = udev;
8154 niter = &udev->adj_list.upper;
8155 dev_stack[cur] = now;
8156 iter_stack[cur++] = iter;
8157 break;
8158 }
8159
8160 if (!next) {
8161 if (!cur)
8162 return 0;
8163 next = dev_stack[--cur];
8164 niter = iter_stack[cur];
8165 }
8166
8167 now = next;
8168 iter = niter;
8169 }
8170
8171 return 0;
8172 }
8173 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
8174
8175 static bool __netdev_has_upper_dev(struct net_device *dev,
8176 struct net_device *upper_dev)
8177 {
8178 struct netdev_nested_priv priv = {
8179 .flags = 0,
8180 .data = (void *)upper_dev,
8181 };
8182
8183 ASSERT_RTNL();
8184
8185 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
8186 &priv);
8187 }
8188
8189 /**
8190 * netdev_lower_get_next_private - Get the next ->private from the
8191 * lower neighbour list
8192 * @dev: device
8193 * @iter: list_head ** of the current position
8194 *
8195 * Gets the next netdev_adjacent->private from the dev's lower neighbour
8196 * list, starting from iter position. The caller must hold either hold the
8197 * RTNL lock or its own locking that guarantees that the neighbour lower
8198 * list will remain unchanged.
8199 */
8200 void *netdev_lower_get_next_private(struct net_device *dev,
8201 struct list_head **iter)
8202 {
8203 struct netdev_adjacent *lower;
8204
8205 lower = list_entry(*iter, struct netdev_adjacent, list);
8206
8207 if (&lower->list == &dev->adj_list.lower)
8208 return NULL;
8209
8210 *iter = lower->list.next;
8211
8212 return lower->private;
8213 }
8214 EXPORT_SYMBOL(netdev_lower_get_next_private);
8215
8216 /**
8217 * netdev_lower_get_next_private_rcu - Get the next ->private from the
8218 * lower neighbour list, RCU
8219 * variant
8220 * @dev: device
8221 * @iter: list_head ** of the current position
8222 *
8223 * Gets the next netdev_adjacent->private from the dev's lower neighbour
8224 * list, starting from iter position. The caller must hold RCU read lock.
8225 */
8226 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
8227 struct list_head **iter)
8228 {
8229 struct netdev_adjacent *lower;
8230
8231 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
8232
8233 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
8234
8235 if (&lower->list == &dev->adj_list.lower)
8236 return NULL;
8237
8238 *iter = &lower->list;
8239
8240 return lower->private;
8241 }
8242 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
8243
8244 /**
8245 * netdev_lower_get_next - Get the next device from the lower neighbour
8246 * list
8247 * @dev: device
8248 * @iter: list_head ** of the current position
8249 *
8250 * Gets the next netdev_adjacent from the dev's lower neighbour
8251 * list, starting from iter position. The caller must hold RTNL lock or
8252 * its own locking that guarantees that the neighbour lower
8253 * list will remain unchanged.
8254 */
8255 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
8256 {
8257 struct netdev_adjacent *lower;
8258
8259 lower = list_entry(*iter, struct netdev_adjacent, list);
8260
8261 if (&lower->list == &dev->adj_list.lower)
8262 return NULL;
8263
8264 *iter = lower->list.next;
8265
8266 return lower->dev;
8267 }
8268 EXPORT_SYMBOL(netdev_lower_get_next);
8269
8270 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
8271 struct list_head **iter)
8272 {
8273 struct netdev_adjacent *lower;
8274
8275 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
8276
8277 if (&lower->list == &dev->adj_list.lower)
8278 return NULL;
8279
8280 *iter = &lower->list;
8281
8282 return lower->dev;
8283 }
8284
8285 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
8286 struct list_head **iter,
8287 bool *ignore)
8288 {
8289 struct netdev_adjacent *lower;
8290
8291 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
8292
8293 if (&lower->list == &dev->adj_list.lower)
8294 return NULL;
8295
8296 *iter = &lower->list;
8297 *ignore = lower->ignore;
8298
8299 return lower->dev;
8300 }
8301
8302 int netdev_walk_all_lower_dev(struct net_device *dev,
8303 int (*fn)(struct net_device *dev,
8304 struct netdev_nested_priv *priv),
8305 struct netdev_nested_priv *priv)
8306 {
8307 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
8308 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
8309 int ret, cur = 0;
8310
8311 now = dev;
8312 iter = &dev->adj_list.lower;
8313
8314 while (1) {
8315 if (now != dev) {
8316 ret = fn(now, priv);
8317 if (ret)
8318 return ret;
8319 }
8320
8321 next = NULL;
8322 while (1) {
8323 ldev = netdev_next_lower_dev(now, &iter);
8324 if (!ldev)
8325 break;
8326
8327 next = ldev;
8328 niter = &ldev->adj_list.lower;
8329 dev_stack[cur] = now;
8330 iter_stack[cur++] = iter;
8331 break;
8332 }
8333
8334 if (!next) {
8335 if (!cur)
8336 return 0;
8337 next = dev_stack[--cur];
8338 niter = iter_stack[cur];
8339 }
8340
8341 now = next;
8342 iter = niter;
8343 }
8344
8345 return 0;
8346 }
8347 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
8348
8349 static int __netdev_walk_all_lower_dev(struct net_device *dev,
8350 int (*fn)(struct net_device *dev,
8351 struct netdev_nested_priv *priv),
8352 struct netdev_nested_priv *priv)
8353 {
8354 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
8355 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
8356 int ret, cur = 0;
8357 bool ignore;
8358
8359 now = dev;
8360 iter = &dev->adj_list.lower;
8361
8362 while (1) {
8363 if (now != dev) {
8364 ret = fn(now, priv);
8365 if (ret)
8366 return ret;
8367 }
8368
8369 next = NULL;
8370 while (1) {
8371 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
8372 if (!ldev)
8373 break;
8374 if (ignore)
8375 continue;
8376
8377 next = ldev;
8378 niter = &ldev->adj_list.lower;
8379 dev_stack[cur] = now;
8380 iter_stack[cur++] = iter;
8381 break;
8382 }
8383
8384 if (!next) {
8385 if (!cur)
8386 return 0;
8387 next = dev_stack[--cur];
8388 niter = iter_stack[cur];
8389 }
8390
8391 now = next;
8392 iter = niter;
8393 }
8394
8395 return 0;
8396 }
8397
8398 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
8399 struct list_head **iter)
8400 {
8401 struct netdev_adjacent *lower;
8402
8403 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
8404 if (&lower->list == &dev->adj_list.lower)
8405 return NULL;
8406
8407 *iter = &lower->list;
8408
8409 return lower->dev;
8410 }
8411 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
8412
8413 static u8 __netdev_upper_depth(struct net_device *dev)
8414 {
8415 struct net_device *udev;
8416 struct list_head *iter;
8417 u8 max_depth = 0;
8418 bool ignore;
8419
8420 for (iter = &dev->adj_list.upper,
8421 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
8422 udev;
8423 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
8424 if (ignore)
8425 continue;
8426 if (max_depth < udev->upper_level)
8427 max_depth = udev->upper_level;
8428 }
8429
8430 return max_depth;
8431 }
8432
8433 static u8 __netdev_lower_depth(struct net_device *dev)
8434 {
8435 struct net_device *ldev;
8436 struct list_head *iter;
8437 u8 max_depth = 0;
8438 bool ignore;
8439
8440 for (iter = &dev->adj_list.lower,
8441 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
8442 ldev;
8443 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
8444 if (ignore)
8445 continue;
8446 if (max_depth < ldev->lower_level)
8447 max_depth = ldev->lower_level;
8448 }
8449
8450 return max_depth;
8451 }
8452
8453 static int __netdev_update_upper_level(struct net_device *dev,
8454 struct netdev_nested_priv *__unused)
8455 {
8456 dev->upper_level = __netdev_upper_depth(dev) + 1;
8457 return 0;
8458 }
8459
8460 #ifdef CONFIG_LOCKDEP
8461 static LIST_HEAD(net_unlink_list);
8462
8463 static void net_unlink_todo(struct net_device *dev)
8464 {
8465 if (list_empty(&dev->unlink_list))
8466 list_add_tail(&dev->unlink_list, &net_unlink_list);
8467 }
8468 #endif
8469
8470 static int __netdev_update_lower_level(struct net_device *dev,
8471 struct netdev_nested_priv *priv)
8472 {
8473 dev->lower_level = __netdev_lower_depth(dev) + 1;
8474
8475 #ifdef CONFIG_LOCKDEP
8476 if (!priv)
8477 return 0;
8478
8479 if (priv->flags & NESTED_SYNC_IMM)
8480 dev->nested_level = dev->lower_level - 1;
8481 if (priv->flags & NESTED_SYNC_TODO)
8482 net_unlink_todo(dev);
8483 #endif
8484 return 0;
8485 }
8486
8487 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
8488 int (*fn)(struct net_device *dev,
8489 struct netdev_nested_priv *priv),
8490 struct netdev_nested_priv *priv)
8491 {
8492 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
8493 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
8494 int ret, cur = 0;
8495
8496 now = dev;
8497 iter = &dev->adj_list.lower;
8498
8499 while (1) {
8500 if (now != dev) {
8501 ret = fn(now, priv);
8502 if (ret)
8503 return ret;
8504 }
8505
8506 next = NULL;
8507 while (1) {
8508 ldev = netdev_next_lower_dev_rcu(now, &iter);
8509 if (!ldev)
8510 break;
8511
8512 next = ldev;
8513 niter = &ldev->adj_list.lower;
8514 dev_stack[cur] = now;
8515 iter_stack[cur++] = iter;
8516 break;
8517 }
8518
8519 if (!next) {
8520 if (!cur)
8521 return 0;
8522 next = dev_stack[--cur];
8523 niter = iter_stack[cur];
8524 }
8525
8526 now = next;
8527 iter = niter;
8528 }
8529
8530 return 0;
8531 }
8532 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
8533
8534 /**
8535 * netdev_lower_get_first_private_rcu - Get the first ->private from the
8536 * lower neighbour list, RCU
8537 * variant
8538 * @dev: device
8539 *
8540 * Gets the first netdev_adjacent->private from the dev's lower neighbour
8541 * list. The caller must hold RCU read lock.
8542 */
8543 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
8544 {
8545 struct netdev_adjacent *lower;
8546
8547 lower = list_first_or_null_rcu(&dev->adj_list.lower,
8548 struct netdev_adjacent, list);
8549 if (lower)
8550 return lower->private;
8551 return NULL;
8552 }
8553 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
8554
8555 /**
8556 * netdev_master_upper_dev_get_rcu - Get master upper device
8557 * @dev: device
8558 *
8559 * Find a master upper device and return pointer to it or NULL in case
8560 * it's not there. The caller must hold the RCU read lock.
8561 */
8562 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
8563 {
8564 struct netdev_adjacent *upper;
8565
8566 upper = list_first_or_null_rcu(&dev->adj_list.upper,
8567 struct netdev_adjacent, list);
8568 if (upper && likely(upper->master))
8569 return upper->dev;
8570 return NULL;
8571 }
8572 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
8573
8574 static int netdev_adjacent_sysfs_add(struct net_device *dev,
8575 struct net_device *adj_dev,
8576 struct list_head *dev_list)
8577 {
8578 char linkname[IFNAMSIZ+7];
8579
8580 sprintf(linkname, dev_list == &dev->adj_list.upper ?
8581 "upper_%s" : "lower_%s", adj_dev->name);
8582 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
8583 linkname);
8584 }
8585 static void netdev_adjacent_sysfs_del(struct net_device *dev,
8586 char *name,
8587 struct list_head *dev_list)
8588 {
8589 char linkname[IFNAMSIZ+7];
8590
8591 sprintf(linkname, dev_list == &dev->adj_list.upper ?
8592 "upper_%s" : "lower_%s", name);
8593 sysfs_remove_link(&(dev->dev.kobj), linkname);
8594 }
8595
8596 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
8597 struct net_device *adj_dev,
8598 struct list_head *dev_list)
8599 {
8600 return (dev_list == &dev->adj_list.upper ||
8601 dev_list == &dev->adj_list.lower) &&
8602 net_eq(dev_net(dev), dev_net(adj_dev));
8603 }
8604
8605 static int __netdev_adjacent_dev_insert(struct net_device *dev,
8606 struct net_device *adj_dev,
8607 struct list_head *dev_list,
8608 void *private, bool master)
8609 {
8610 struct netdev_adjacent *adj;
8611 int ret;
8612
8613 adj = __netdev_find_adj(adj_dev, dev_list);
8614
8615 if (adj) {
8616 adj->ref_nr += 1;
8617 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
8618 dev->name, adj_dev->name, adj->ref_nr);
8619
8620 return 0;
8621 }
8622
8623 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
8624 if (!adj)
8625 return -ENOMEM;
8626
8627 adj->dev = adj_dev;
8628 adj->master = master;
8629 adj->ref_nr = 1;
8630 adj->private = private;
8631 adj->ignore = false;
8632 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
8633
8634 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
8635 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
8636
8637 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
8638 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
8639 if (ret)
8640 goto free_adj;
8641 }
8642
8643 /* Ensure that master link is always the first item in list. */
8644 if (master) {
8645 ret = sysfs_create_link(&(dev->dev.kobj),
8646 &(adj_dev->dev.kobj), "master");
8647 if (ret)
8648 goto remove_symlinks;
8649
8650 list_add_rcu(&adj->list, dev_list);
8651 } else {
8652 list_add_tail_rcu(&adj->list, dev_list);
8653 }
8654
8655 return 0;
8656
8657 remove_symlinks:
8658 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8659 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8660 free_adj:
8661 netdev_put(adj_dev, &adj->dev_tracker);
8662 kfree(adj);
8663
8664 return ret;
8665 }
8666
8667 static void __netdev_adjacent_dev_remove(struct net_device *dev,
8668 struct net_device *adj_dev,
8669 u16 ref_nr,
8670 struct list_head *dev_list)
8671 {
8672 struct netdev_adjacent *adj;
8673
8674 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
8675 dev->name, adj_dev->name, ref_nr);
8676
8677 adj = __netdev_find_adj(adj_dev, dev_list);
8678
8679 if (!adj) {
8680 pr_err("Adjacency does not exist for device %s from %s\n",
8681 dev->name, adj_dev->name);
8682 WARN_ON(1);
8683 return;
8684 }
8685
8686 if (adj->ref_nr > ref_nr) {
8687 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
8688 dev->name, adj_dev->name, ref_nr,
8689 adj->ref_nr - ref_nr);
8690 adj->ref_nr -= ref_nr;
8691 return;
8692 }
8693
8694 if (adj->master)
8695 sysfs_remove_link(&(dev->dev.kobj), "master");
8696
8697 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
8698 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
8699
8700 list_del_rcu(&adj->list);
8701 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
8702 adj_dev->name, dev->name, adj_dev->name);
8703 netdev_put(adj_dev, &adj->dev_tracker);
8704 kfree_rcu(adj, rcu);
8705 }
8706
8707 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
8708 struct net_device *upper_dev,
8709 struct list_head *up_list,
8710 struct list_head *down_list,
8711 void *private, bool master)
8712 {
8713 int ret;
8714
8715 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
8716 private, master);
8717 if (ret)
8718 return ret;
8719
8720 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
8721 private, false);
8722 if (ret) {
8723 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8724 return ret;
8725 }
8726
8727 return 0;
8728 }
8729
8730 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8731 struct net_device *upper_dev,
8732 u16 ref_nr,
8733 struct list_head *up_list,
8734 struct list_head *down_list)
8735 {
8736 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8737 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8738 }
8739
8740 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8741 struct net_device *upper_dev,
8742 void *private, bool master)
8743 {
8744 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8745 &dev->adj_list.upper,
8746 &upper_dev->adj_list.lower,
8747 private, master);
8748 }
8749
8750 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8751 struct net_device *upper_dev)
8752 {
8753 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8754 &dev->adj_list.upper,
8755 &upper_dev->adj_list.lower);
8756 }
8757
8758 static int __netdev_upper_dev_link(struct net_device *dev,
8759 struct net_device *upper_dev, bool master,
8760 void *upper_priv, void *upper_info,
8761 struct netdev_nested_priv *priv,
8762 struct netlink_ext_ack *extack)
8763 {
8764 struct netdev_notifier_changeupper_info changeupper_info = {
8765 .info = {
8766 .dev = dev,
8767 .extack = extack,
8768 },
8769 .upper_dev = upper_dev,
8770 .master = master,
8771 .linking = true,
8772 .upper_info = upper_info,
8773 };
8774 struct net_device *master_dev;
8775 int ret = 0;
8776
8777 ASSERT_RTNL();
8778
8779 if (dev == upper_dev)
8780 return -EBUSY;
8781
8782 /* To prevent loops, check if dev is not upper device to upper_dev. */
8783 if (__netdev_has_upper_dev(upper_dev, dev))
8784 return -EBUSY;
8785
8786 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8787 return -EMLINK;
8788
8789 if (!master) {
8790 if (__netdev_has_upper_dev(dev, upper_dev))
8791 return -EEXIST;
8792 } else {
8793 master_dev = __netdev_master_upper_dev_get(dev);
8794 if (master_dev)
8795 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8796 }
8797
8798 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8799 &changeupper_info.info);
8800 ret = notifier_to_errno(ret);
8801 if (ret)
8802 return ret;
8803
8804 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8805 master);
8806 if (ret)
8807 return ret;
8808
8809 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8810 &changeupper_info.info);
8811 ret = notifier_to_errno(ret);
8812 if (ret)
8813 goto rollback;
8814
8815 __netdev_update_upper_level(dev, NULL);
8816 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8817
8818 __netdev_update_lower_level(upper_dev, priv);
8819 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8820 priv);
8821
8822 return 0;
8823
8824 rollback:
8825 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8826
8827 return ret;
8828 }
8829
8830 /**
8831 * netdev_upper_dev_link - Add a link to the upper device
8832 * @dev: device
8833 * @upper_dev: new upper device
8834 * @extack: netlink extended ack
8835 *
8836 * Adds a link to device which is upper to this one. The caller must hold
8837 * the RTNL lock. On a failure a negative errno code is returned.
8838 * On success the reference counts are adjusted and the function
8839 * returns zero.
8840 */
8841 int netdev_upper_dev_link(struct net_device *dev,
8842 struct net_device *upper_dev,
8843 struct netlink_ext_ack *extack)
8844 {
8845 struct netdev_nested_priv priv = {
8846 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8847 .data = NULL,
8848 };
8849
8850 return __netdev_upper_dev_link(dev, upper_dev, false,
8851 NULL, NULL, &priv, extack);
8852 }
8853 EXPORT_SYMBOL(netdev_upper_dev_link);
8854
8855 /**
8856 * netdev_master_upper_dev_link - Add a master link to the upper device
8857 * @dev: device
8858 * @upper_dev: new upper device
8859 * @upper_priv: upper device private
8860 * @upper_info: upper info to be passed down via notifier
8861 * @extack: netlink extended ack
8862 *
8863 * Adds a link to device which is upper to this one. In this case, only
8864 * one master upper device can be linked, although other non-master devices
8865 * might be linked as well. The caller must hold the RTNL lock.
8866 * On a failure a negative errno code is returned. On success the reference
8867 * counts are adjusted and the function returns zero.
8868 */
8869 int netdev_master_upper_dev_link(struct net_device *dev,
8870 struct net_device *upper_dev,
8871 void *upper_priv, void *upper_info,
8872 struct netlink_ext_ack *extack)
8873 {
8874 struct netdev_nested_priv priv = {
8875 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8876 .data = NULL,
8877 };
8878
8879 return __netdev_upper_dev_link(dev, upper_dev, true,
8880 upper_priv, upper_info, &priv, extack);
8881 }
8882 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8883
8884 static void __netdev_upper_dev_unlink(struct net_device *dev,
8885 struct net_device *upper_dev,
8886 struct netdev_nested_priv *priv)
8887 {
8888 struct netdev_notifier_changeupper_info changeupper_info = {
8889 .info = {
8890 .dev = dev,
8891 },
8892 .upper_dev = upper_dev,
8893 .linking = false,
8894 };
8895
8896 ASSERT_RTNL();
8897
8898 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8899
8900 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8901 &changeupper_info.info);
8902
8903 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8904
8905 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8906 &changeupper_info.info);
8907
8908 __netdev_update_upper_level(dev, NULL);
8909 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8910
8911 __netdev_update_lower_level(upper_dev, priv);
8912 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8913 priv);
8914 }
8915
8916 /**
8917 * netdev_upper_dev_unlink - Removes a link to upper device
8918 * @dev: device
8919 * @upper_dev: new upper device
8920 *
8921 * Removes a link to device which is upper to this one. The caller must hold
8922 * the RTNL lock.
8923 */
8924 void netdev_upper_dev_unlink(struct net_device *dev,
8925 struct net_device *upper_dev)
8926 {
8927 struct netdev_nested_priv priv = {
8928 .flags = NESTED_SYNC_TODO,
8929 .data = NULL,
8930 };
8931
8932 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8933 }
8934 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8935
8936 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8937 struct net_device *lower_dev,
8938 bool val)
8939 {
8940 struct netdev_adjacent *adj;
8941
8942 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8943 if (adj)
8944 adj->ignore = val;
8945
8946 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8947 if (adj)
8948 adj->ignore = val;
8949 }
8950
8951 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8952 struct net_device *lower_dev)
8953 {
8954 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8955 }
8956
8957 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8958 struct net_device *lower_dev)
8959 {
8960 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8961 }
8962
8963 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8964 struct net_device *new_dev,
8965 struct net_device *dev,
8966 struct netlink_ext_ack *extack)
8967 {
8968 struct netdev_nested_priv priv = {
8969 .flags = 0,
8970 .data = NULL,
8971 };
8972 int err;
8973
8974 if (!new_dev)
8975 return 0;
8976
8977 if (old_dev && new_dev != old_dev)
8978 netdev_adjacent_dev_disable(dev, old_dev);
8979 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8980 extack);
8981 if (err) {
8982 if (old_dev && new_dev != old_dev)
8983 netdev_adjacent_dev_enable(dev, old_dev);
8984 return err;
8985 }
8986
8987 return 0;
8988 }
8989 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8990
8991 void netdev_adjacent_change_commit(struct net_device *old_dev,
8992 struct net_device *new_dev,
8993 struct net_device *dev)
8994 {
8995 struct netdev_nested_priv priv = {
8996 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8997 .data = NULL,
8998 };
8999
9000 if (!new_dev || !old_dev)
9001 return;
9002
9003 if (new_dev == old_dev)
9004 return;
9005
9006 netdev_adjacent_dev_enable(dev, old_dev);
9007 __netdev_upper_dev_unlink(old_dev, dev, &priv);
9008 }
9009 EXPORT_SYMBOL(netdev_adjacent_change_commit);
9010
9011 void netdev_adjacent_change_abort(struct net_device *old_dev,
9012 struct net_device *new_dev,
9013 struct net_device *dev)
9014 {
9015 struct netdev_nested_priv priv = {
9016 .flags = 0,
9017 .data = NULL,
9018 };
9019
9020 if (!new_dev)
9021 return;
9022
9023 if (old_dev && new_dev != old_dev)
9024 netdev_adjacent_dev_enable(dev, old_dev);
9025
9026 __netdev_upper_dev_unlink(new_dev, dev, &priv);
9027 }
9028 EXPORT_SYMBOL(netdev_adjacent_change_abort);
9029
9030 /**
9031 * netdev_bonding_info_change - Dispatch event about slave change
9032 * @dev: device
9033 * @bonding_info: info to dispatch
9034 *
9035 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
9036 * The caller must hold the RTNL lock.
9037 */
9038 void netdev_bonding_info_change(struct net_device *dev,
9039 struct netdev_bonding_info *bonding_info)
9040 {
9041 struct netdev_notifier_bonding_info info = {
9042 .info.dev = dev,
9043 };
9044
9045 memcpy(&info.bonding_info, bonding_info,
9046 sizeof(struct netdev_bonding_info));
9047 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
9048 &info.info);
9049 }
9050 EXPORT_SYMBOL(netdev_bonding_info_change);
9051
9052 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
9053 struct netlink_ext_ack *extack)
9054 {
9055 struct netdev_notifier_offload_xstats_info info = {
9056 .info.dev = dev,
9057 .info.extack = extack,
9058 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
9059 };
9060 int err;
9061 int rc;
9062
9063 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
9064 GFP_KERNEL);
9065 if (!dev->offload_xstats_l3)
9066 return -ENOMEM;
9067
9068 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
9069 NETDEV_OFFLOAD_XSTATS_DISABLE,
9070 &info.info);
9071 err = notifier_to_errno(rc);
9072 if (err)
9073 goto free_stats;
9074
9075 return 0;
9076
9077 free_stats:
9078 kfree(dev->offload_xstats_l3);
9079 dev->offload_xstats_l3 = NULL;
9080 return err;
9081 }
9082
9083 int netdev_offload_xstats_enable(struct net_device *dev,
9084 enum netdev_offload_xstats_type type,
9085 struct netlink_ext_ack *extack)
9086 {
9087 ASSERT_RTNL();
9088
9089 if (netdev_offload_xstats_enabled(dev, type))
9090 return -EALREADY;
9091
9092 switch (type) {
9093 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
9094 return netdev_offload_xstats_enable_l3(dev, extack);
9095 }
9096
9097 WARN_ON(1);
9098 return -EINVAL;
9099 }
9100 EXPORT_SYMBOL(netdev_offload_xstats_enable);
9101
9102 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
9103 {
9104 struct netdev_notifier_offload_xstats_info info = {
9105 .info.dev = dev,
9106 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
9107 };
9108
9109 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
9110 &info.info);
9111 kfree(dev->offload_xstats_l3);
9112 dev->offload_xstats_l3 = NULL;
9113 }
9114
9115 int netdev_offload_xstats_disable(struct net_device *dev,
9116 enum netdev_offload_xstats_type type)
9117 {
9118 ASSERT_RTNL();
9119
9120 if (!netdev_offload_xstats_enabled(dev, type))
9121 return -EALREADY;
9122
9123 switch (type) {
9124 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
9125 netdev_offload_xstats_disable_l3(dev);
9126 return 0;
9127 }
9128
9129 WARN_ON(1);
9130 return -EINVAL;
9131 }
9132 EXPORT_SYMBOL(netdev_offload_xstats_disable);
9133
9134 static void netdev_offload_xstats_disable_all(struct net_device *dev)
9135 {
9136 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
9137 }
9138
9139 static struct rtnl_hw_stats64 *
9140 netdev_offload_xstats_get_ptr(const struct net_device *dev,
9141 enum netdev_offload_xstats_type type)
9142 {
9143 switch (type) {
9144 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
9145 return dev->offload_xstats_l3;
9146 }
9147
9148 WARN_ON(1);
9149 return NULL;
9150 }
9151
9152 bool netdev_offload_xstats_enabled(const struct net_device *dev,
9153 enum netdev_offload_xstats_type type)
9154 {
9155 ASSERT_RTNL();
9156
9157 return netdev_offload_xstats_get_ptr(dev, type);
9158 }
9159 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
9160
9161 struct netdev_notifier_offload_xstats_ru {
9162 bool used;
9163 };
9164
9165 struct netdev_notifier_offload_xstats_rd {
9166 struct rtnl_hw_stats64 stats;
9167 bool used;
9168 };
9169
9170 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
9171 const struct rtnl_hw_stats64 *src)
9172 {
9173 dest->rx_packets += src->rx_packets;
9174 dest->tx_packets += src->tx_packets;
9175 dest->rx_bytes += src->rx_bytes;
9176 dest->tx_bytes += src->tx_bytes;
9177 dest->rx_errors += src->rx_errors;
9178 dest->tx_errors += src->tx_errors;
9179 dest->rx_dropped += src->rx_dropped;
9180 dest->tx_dropped += src->tx_dropped;
9181 dest->multicast += src->multicast;
9182 }
9183
9184 static int netdev_offload_xstats_get_used(struct net_device *dev,
9185 enum netdev_offload_xstats_type type,
9186 bool *p_used,
9187 struct netlink_ext_ack *extack)
9188 {
9189 struct netdev_notifier_offload_xstats_ru report_used = {};
9190 struct netdev_notifier_offload_xstats_info info = {
9191 .info.dev = dev,
9192 .info.extack = extack,
9193 .type = type,
9194 .report_used = &report_used,
9195 };
9196 int rc;
9197
9198 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
9199 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
9200 &info.info);
9201 *p_used = report_used.used;
9202 return notifier_to_errno(rc);
9203 }
9204
9205 static int netdev_offload_xstats_get_stats(struct net_device *dev,
9206 enum netdev_offload_xstats_type type,
9207 struct rtnl_hw_stats64 *p_stats,
9208 bool *p_used,
9209 struct netlink_ext_ack *extack)
9210 {
9211 struct netdev_notifier_offload_xstats_rd report_delta = {};
9212 struct netdev_notifier_offload_xstats_info info = {
9213 .info.dev = dev,
9214 .info.extack = extack,
9215 .type = type,
9216 .report_delta = &report_delta,
9217 };
9218 struct rtnl_hw_stats64 *stats;
9219 int rc;
9220
9221 stats = netdev_offload_xstats_get_ptr(dev, type);
9222 if (WARN_ON(!stats))
9223 return -EINVAL;
9224
9225 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
9226 &info.info);
9227
9228 /* Cache whatever we got, even if there was an error, otherwise the
9229 * successful stats retrievals would get lost.
9230 */
9231 netdev_hw_stats64_add(stats, &report_delta.stats);
9232
9233 if (p_stats)
9234 *p_stats = *stats;
9235 *p_used = report_delta.used;
9236
9237 return notifier_to_errno(rc);
9238 }
9239
9240 int netdev_offload_xstats_get(struct net_device *dev,
9241 enum netdev_offload_xstats_type type,
9242 struct rtnl_hw_stats64 *p_stats, bool *p_used,
9243 struct netlink_ext_ack *extack)
9244 {
9245 ASSERT_RTNL();
9246
9247 if (p_stats)
9248 return netdev_offload_xstats_get_stats(dev, type, p_stats,
9249 p_used, extack);
9250 else
9251 return netdev_offload_xstats_get_used(dev, type, p_used,
9252 extack);
9253 }
9254 EXPORT_SYMBOL(netdev_offload_xstats_get);
9255
9256 void
9257 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
9258 const struct rtnl_hw_stats64 *stats)
9259 {
9260 report_delta->used = true;
9261 netdev_hw_stats64_add(&report_delta->stats, stats);
9262 }
9263 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
9264
9265 void
9266 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
9267 {
9268 report_used->used = true;
9269 }
9270 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
9271
9272 void netdev_offload_xstats_push_delta(struct net_device *dev,
9273 enum netdev_offload_xstats_type type,
9274 const struct rtnl_hw_stats64 *p_stats)
9275 {
9276 struct rtnl_hw_stats64 *stats;
9277
9278 ASSERT_RTNL();
9279
9280 stats = netdev_offload_xstats_get_ptr(dev, type);
9281 if (WARN_ON(!stats))
9282 return;
9283
9284 netdev_hw_stats64_add(stats, p_stats);
9285 }
9286 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
9287
9288 /**
9289 * netdev_get_xmit_slave - Get the xmit slave of master device
9290 * @dev: device
9291 * @skb: The packet
9292 * @all_slaves: assume all the slaves are active
9293 *
9294 * The reference counters are not incremented so the caller must be
9295 * careful with locks. The caller must hold RCU lock.
9296 * %NULL is returned if no slave is found.
9297 */
9298
9299 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
9300 struct sk_buff *skb,
9301 bool all_slaves)
9302 {
9303 const struct net_device_ops *ops = dev->netdev_ops;
9304
9305 if (!ops->ndo_get_xmit_slave)
9306 return NULL;
9307 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
9308 }
9309 EXPORT_SYMBOL(netdev_get_xmit_slave);
9310
9311 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
9312 struct sock *sk)
9313 {
9314 const struct net_device_ops *ops = dev->netdev_ops;
9315
9316 if (!ops->ndo_sk_get_lower_dev)
9317 return NULL;
9318 return ops->ndo_sk_get_lower_dev(dev, sk);
9319 }
9320
9321 /**
9322 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
9323 * @dev: device
9324 * @sk: the socket
9325 *
9326 * %NULL is returned if no lower device is found.
9327 */
9328
9329 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
9330 struct sock *sk)
9331 {
9332 struct net_device *lower;
9333
9334 lower = netdev_sk_get_lower_dev(dev, sk);
9335 while (lower) {
9336 dev = lower;
9337 lower = netdev_sk_get_lower_dev(dev, sk);
9338 }
9339
9340 return dev;
9341 }
9342 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
9343
9344 static void netdev_adjacent_add_links(struct net_device *dev)
9345 {
9346 struct netdev_adjacent *iter;
9347
9348 struct net *net = dev_net(dev);
9349
9350 list_for_each_entry(iter, &dev->adj_list.upper, list) {
9351 if (!net_eq(net, dev_net(iter->dev)))
9352 continue;
9353 netdev_adjacent_sysfs_add(iter->dev, dev,
9354 &iter->dev->adj_list.lower);
9355 netdev_adjacent_sysfs_add(dev, iter->dev,
9356 &dev->adj_list.upper);
9357 }
9358
9359 list_for_each_entry(iter, &dev->adj_list.lower, list) {
9360 if (!net_eq(net, dev_net(iter->dev)))
9361 continue;
9362 netdev_adjacent_sysfs_add(iter->dev, dev,
9363 &iter->dev->adj_list.upper);
9364 netdev_adjacent_sysfs_add(dev, iter->dev,
9365 &dev->adj_list.lower);
9366 }
9367 }
9368
9369 static void netdev_adjacent_del_links(struct net_device *dev)
9370 {
9371 struct netdev_adjacent *iter;
9372
9373 struct net *net = dev_net(dev);
9374
9375 list_for_each_entry(iter, &dev->adj_list.upper, list) {
9376 if (!net_eq(net, dev_net(iter->dev)))
9377 continue;
9378 netdev_adjacent_sysfs_del(iter->dev, dev->name,
9379 &iter->dev->adj_list.lower);
9380 netdev_adjacent_sysfs_del(dev, iter->dev->name,
9381 &dev->adj_list.upper);
9382 }
9383
9384 list_for_each_entry(iter, &dev->adj_list.lower, list) {
9385 if (!net_eq(net, dev_net(iter->dev)))
9386 continue;
9387 netdev_adjacent_sysfs_del(iter->dev, dev->name,
9388 &iter->dev->adj_list.upper);
9389 netdev_adjacent_sysfs_del(dev, iter->dev->name,
9390 &dev->adj_list.lower);
9391 }
9392 }
9393
9394 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
9395 {
9396 struct netdev_adjacent *iter;
9397
9398 struct net *net = dev_net(dev);
9399
9400 list_for_each_entry(iter, &dev->adj_list.upper, list) {
9401 if (!net_eq(net, dev_net(iter->dev)))
9402 continue;
9403 netdev_adjacent_sysfs_del(iter->dev, oldname,
9404 &iter->dev->adj_list.lower);
9405 netdev_adjacent_sysfs_add(iter->dev, dev,
9406 &iter->dev->adj_list.lower);
9407 }
9408
9409 list_for_each_entry(iter, &dev->adj_list.lower, list) {
9410 if (!net_eq(net, dev_net(iter->dev)))
9411 continue;
9412 netdev_adjacent_sysfs_del(iter->dev, oldname,
9413 &iter->dev->adj_list.upper);
9414 netdev_adjacent_sysfs_add(iter->dev, dev,
9415 &iter->dev->adj_list.upper);
9416 }
9417 }
9418
9419 void *netdev_lower_dev_get_private(struct net_device *dev,
9420 struct net_device *lower_dev)
9421 {
9422 struct netdev_adjacent *lower;
9423
9424 if (!lower_dev)
9425 return NULL;
9426 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
9427 if (!lower)
9428 return NULL;
9429
9430 return lower->private;
9431 }
9432 EXPORT_SYMBOL(netdev_lower_dev_get_private);
9433
9434
9435 /**
9436 * netdev_lower_state_changed - Dispatch event about lower device state change
9437 * @lower_dev: device
9438 * @lower_state_info: state to dispatch
9439 *
9440 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
9441 * The caller must hold the RTNL lock.
9442 */
9443 void netdev_lower_state_changed(struct net_device *lower_dev,
9444 void *lower_state_info)
9445 {
9446 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
9447 .info.dev = lower_dev,
9448 };
9449
9450 ASSERT_RTNL();
9451 changelowerstate_info.lower_state_info = lower_state_info;
9452 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
9453 &changelowerstate_info.info);
9454 }
9455 EXPORT_SYMBOL(netdev_lower_state_changed);
9456
9457 static void dev_change_rx_flags(struct net_device *dev, int flags)
9458 {
9459 const struct net_device_ops *ops = dev->netdev_ops;
9460
9461 if (ops->ndo_change_rx_flags)
9462 ops->ndo_change_rx_flags(dev, flags);
9463 }
9464
9465 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
9466 {
9467 unsigned int old_flags = dev->flags;
9468 unsigned int promiscuity, flags;
9469 kuid_t uid;
9470 kgid_t gid;
9471
9472 ASSERT_RTNL();
9473
9474 promiscuity = dev->promiscuity + inc;
9475 if (promiscuity == 0) {
9476 /*
9477 * Avoid overflow.
9478 * If inc causes overflow, untouch promisc and return error.
9479 */
9480 if (unlikely(inc > 0)) {
9481 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
9482 return -EOVERFLOW;
9483 }
9484 flags = old_flags & ~IFF_PROMISC;
9485 } else {
9486 flags = old_flags | IFF_PROMISC;
9487 }
9488 WRITE_ONCE(dev->promiscuity, promiscuity);
9489 if (flags != old_flags) {
9490 WRITE_ONCE(dev->flags, flags);
9491 netdev_info(dev, "%s promiscuous mode\n",
9492 dev->flags & IFF_PROMISC ? "entered" : "left");
9493 if (audit_enabled) {
9494 current_uid_gid(&uid, &gid);
9495 audit_log(audit_context(), GFP_ATOMIC,
9496 AUDIT_ANOM_PROMISCUOUS,
9497 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
9498 dev->name, (dev->flags & IFF_PROMISC),
9499 (old_flags & IFF_PROMISC),
9500 from_kuid(&init_user_ns, audit_get_loginuid(current)),
9501 from_kuid(&init_user_ns, uid),
9502 from_kgid(&init_user_ns, gid),
9503 audit_get_sessionid(current));
9504 }
9505
9506 dev_change_rx_flags(dev, IFF_PROMISC);
9507 }
9508 if (notify) {
9509 /* The ops lock is only required to ensure consistent locking
9510 * for `NETDEV_CHANGE` notifiers. This function is sometimes
9511 * called without the lock, even for devices that are ops
9512 * locked, such as in `dev_uc_sync_multiple` when using
9513 * bonding or teaming.
9514 */
9515 netdev_ops_assert_locked(dev);
9516 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
9517 }
9518 return 0;
9519 }
9520
9521 int netif_set_promiscuity(struct net_device *dev, int inc)
9522 {
9523 unsigned int old_flags = dev->flags;
9524 int err;
9525
9526 err = __dev_set_promiscuity(dev, inc, true);
9527 if (err < 0)
9528 return err;
9529 if (dev->flags != old_flags)
9530 dev_set_rx_mode(dev);
9531 return err;
9532 }
9533
9534 int netif_set_allmulti(struct net_device *dev, int inc, bool notify)
9535 {
9536 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
9537 unsigned int allmulti, flags;
9538
9539 ASSERT_RTNL();
9540
9541 allmulti = dev->allmulti + inc;
9542 if (allmulti == 0) {
9543 /*
9544 * Avoid overflow.
9545 * If inc causes overflow, untouch allmulti and return error.
9546 */
9547 if (unlikely(inc > 0)) {
9548 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
9549 return -EOVERFLOW;
9550 }
9551 flags = old_flags & ~IFF_ALLMULTI;
9552 } else {
9553 flags = old_flags | IFF_ALLMULTI;
9554 }
9555 WRITE_ONCE(dev->allmulti, allmulti);
9556 if (flags != old_flags) {
9557 WRITE_ONCE(dev->flags, flags);
9558 netdev_info(dev, "%s allmulticast mode\n",
9559 dev->flags & IFF_ALLMULTI ? "entered" : "left");
9560 dev_change_rx_flags(dev, IFF_ALLMULTI);
9561 dev_set_rx_mode(dev);
9562 if (notify)
9563 __dev_notify_flags(dev, old_flags,
9564 dev->gflags ^ old_gflags, 0, NULL);
9565 }
9566 return 0;
9567 }
9568
9569 /*
9570 * Upload unicast and multicast address lists to device and
9571 * configure RX filtering. When the device doesn't support unicast
9572 * filtering it is put in promiscuous mode while unicast addresses
9573 * are present.
9574 */
9575 void __dev_set_rx_mode(struct net_device *dev)
9576 {
9577 const struct net_device_ops *ops = dev->netdev_ops;
9578
9579 /* dev_open will call this function so the list will stay sane. */
9580 if (!(dev->flags&IFF_UP))
9581 return;
9582
9583 if (!netif_device_present(dev))
9584 return;
9585
9586 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
9587 /* Unicast addresses changes may only happen under the rtnl,
9588 * therefore calling __dev_set_promiscuity here is safe.
9589 */
9590 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
9591 __dev_set_promiscuity(dev, 1, false);
9592 dev->uc_promisc = true;
9593 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
9594 __dev_set_promiscuity(dev, -1, false);
9595 dev->uc_promisc = false;
9596 }
9597 }
9598
9599 if (ops->ndo_set_rx_mode)
9600 ops->ndo_set_rx_mode(dev);
9601 }
9602
9603 void dev_set_rx_mode(struct net_device *dev)
9604 {
9605 netif_addr_lock_bh(dev);
9606 __dev_set_rx_mode(dev);
9607 netif_addr_unlock_bh(dev);
9608 }
9609
9610 /**
9611 * netif_get_flags() - get flags reported to userspace
9612 * @dev: device
9613 *
9614 * Get the combination of flag bits exported through APIs to userspace.
9615 */
9616 unsigned int netif_get_flags(const struct net_device *dev)
9617 {
9618 unsigned int flags;
9619
9620 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC |
9621 IFF_ALLMULTI |
9622 IFF_RUNNING |
9623 IFF_LOWER_UP |
9624 IFF_DORMANT)) |
9625 (READ_ONCE(dev->gflags) & (IFF_PROMISC |
9626 IFF_ALLMULTI));
9627
9628 if (netif_running(dev)) {
9629 if (netif_oper_up(dev))
9630 flags |= IFF_RUNNING;
9631 if (netif_carrier_ok(dev))
9632 flags |= IFF_LOWER_UP;
9633 if (netif_dormant(dev))
9634 flags |= IFF_DORMANT;
9635 }
9636
9637 return flags;
9638 }
9639 EXPORT_SYMBOL(netif_get_flags);
9640
9641 int __dev_change_flags(struct net_device *dev, unsigned int flags,
9642 struct netlink_ext_ack *extack)
9643 {
9644 unsigned int old_flags = dev->flags;
9645 int ret;
9646
9647 ASSERT_RTNL();
9648
9649 /*
9650 * Set the flags on our device.
9651 */
9652
9653 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
9654 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
9655 IFF_AUTOMEDIA)) |
9656 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
9657 IFF_ALLMULTI));
9658
9659 /*
9660 * Load in the correct multicast list now the flags have changed.
9661 */
9662
9663 if ((old_flags ^ flags) & IFF_MULTICAST)
9664 dev_change_rx_flags(dev, IFF_MULTICAST);
9665
9666 dev_set_rx_mode(dev);
9667
9668 /*
9669 * Have we downed the interface. We handle IFF_UP ourselves
9670 * according to user attempts to set it, rather than blindly
9671 * setting it.
9672 */
9673
9674 ret = 0;
9675 if ((old_flags ^ flags) & IFF_UP) {
9676 if (old_flags & IFF_UP)
9677 __dev_close(dev);
9678 else
9679 ret = __dev_open(dev, extack);
9680 }
9681
9682 if ((flags ^ dev->gflags) & IFF_PROMISC) {
9683 int inc = (flags & IFF_PROMISC) ? 1 : -1;
9684 old_flags = dev->flags;
9685
9686 dev->gflags ^= IFF_PROMISC;
9687
9688 if (__dev_set_promiscuity(dev, inc, false) >= 0)
9689 if (dev->flags != old_flags)
9690 dev_set_rx_mode(dev);
9691 }
9692
9693 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
9694 * is important. Some (broken) drivers set IFF_PROMISC, when
9695 * IFF_ALLMULTI is requested not asking us and not reporting.
9696 */
9697 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
9698 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
9699
9700 dev->gflags ^= IFF_ALLMULTI;
9701 netif_set_allmulti(dev, inc, false);
9702 }
9703
9704 return ret;
9705 }
9706
9707 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
9708 unsigned int gchanges, u32 portid,
9709 const struct nlmsghdr *nlh)
9710 {
9711 unsigned int changes = dev->flags ^ old_flags;
9712
9713 if (gchanges)
9714 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
9715
9716 if (changes & IFF_UP) {
9717 if (dev->flags & IFF_UP)
9718 call_netdevice_notifiers(NETDEV_UP, dev);
9719 else
9720 call_netdevice_notifiers(NETDEV_DOWN, dev);
9721 }
9722
9723 if (dev->flags & IFF_UP &&
9724 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
9725 struct netdev_notifier_change_info change_info = {
9726 .info = {
9727 .dev = dev,
9728 },
9729 .flags_changed = changes,
9730 };
9731
9732 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
9733 }
9734 }
9735
9736 int netif_change_flags(struct net_device *dev, unsigned int flags,
9737 struct netlink_ext_ack *extack)
9738 {
9739 int ret;
9740 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
9741
9742 ret = __dev_change_flags(dev, flags, extack);
9743 if (ret < 0)
9744 return ret;
9745
9746 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
9747 __dev_notify_flags(dev, old_flags, changes, 0, NULL);
9748 return ret;
9749 }
9750
9751 int __netif_set_mtu(struct net_device *dev, int new_mtu)
9752 {
9753 const struct net_device_ops *ops = dev->netdev_ops;
9754
9755 if (ops->ndo_change_mtu)
9756 return ops->ndo_change_mtu(dev, new_mtu);
9757
9758 /* Pairs with all the lockless reads of dev->mtu in the stack */
9759 WRITE_ONCE(dev->mtu, new_mtu);
9760 return 0;
9761 }
9762 EXPORT_SYMBOL_NS_GPL(__netif_set_mtu, "NETDEV_INTERNAL");
9763
9764 int dev_validate_mtu(struct net_device *dev, int new_mtu,
9765 struct netlink_ext_ack *extack)
9766 {
9767 /* MTU must be positive, and in range */
9768 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
9769 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
9770 return -EINVAL;
9771 }
9772
9773 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
9774 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
9775 return -EINVAL;
9776 }
9777 return 0;
9778 }
9779
9780 /**
9781 * netif_set_mtu_ext() - Change maximum transfer unit
9782 * @dev: device
9783 * @new_mtu: new transfer unit
9784 * @extack: netlink extended ack
9785 *
9786 * Change the maximum transfer size of the network device.
9787 *
9788 * Return: 0 on success, -errno on failure.
9789 */
9790 int netif_set_mtu_ext(struct net_device *dev, int new_mtu,
9791 struct netlink_ext_ack *extack)
9792 {
9793 int err, orig_mtu;
9794
9795 netdev_ops_assert_locked(dev);
9796
9797 if (new_mtu == dev->mtu)
9798 return 0;
9799
9800 err = dev_validate_mtu(dev, new_mtu, extack);
9801 if (err)
9802 return err;
9803
9804 if (!netif_device_present(dev))
9805 return -ENODEV;
9806
9807 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
9808 err = notifier_to_errno(err);
9809 if (err)
9810 return err;
9811
9812 orig_mtu = dev->mtu;
9813 err = __netif_set_mtu(dev, new_mtu);
9814
9815 if (!err) {
9816 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
9817 orig_mtu);
9818 err = notifier_to_errno(err);
9819 if (err) {
9820 /* setting mtu back and notifying everyone again,
9821 * so that they have a chance to revert changes.
9822 */
9823 __netif_set_mtu(dev, orig_mtu);
9824 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
9825 new_mtu);
9826 }
9827 }
9828 return err;
9829 }
9830
9831 int netif_set_mtu(struct net_device *dev, int new_mtu)
9832 {
9833 struct netlink_ext_ack extack;
9834 int err;
9835
9836 memset(&extack, 0, sizeof(extack));
9837 err = netif_set_mtu_ext(dev, new_mtu, &extack);
9838 if (err && extack._msg)
9839 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
9840 return err;
9841 }
9842 EXPORT_SYMBOL(netif_set_mtu);
9843
9844 int netif_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9845 {
9846 unsigned int orig_len = dev->tx_queue_len;
9847 int res;
9848
9849 if (new_len != (unsigned int)new_len)
9850 return -ERANGE;
9851
9852 if (new_len != orig_len) {
9853 WRITE_ONCE(dev->tx_queue_len, new_len);
9854 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9855 res = notifier_to_errno(res);
9856 if (res)
9857 goto err_rollback;
9858 res = dev_qdisc_change_tx_queue_len(dev);
9859 if (res)
9860 goto err_rollback;
9861 }
9862
9863 return 0;
9864
9865 err_rollback:
9866 netdev_err(dev, "refused to change device tx_queue_len\n");
9867 WRITE_ONCE(dev->tx_queue_len, orig_len);
9868 return res;
9869 }
9870
9871 void netif_set_group(struct net_device *dev, int new_group)
9872 {
9873 dev->group = new_group;
9874 }
9875
9876 /**
9877 * netif_pre_changeaddr_notify() - Call NETDEV_PRE_CHANGEADDR.
9878 * @dev: device
9879 * @addr: new address
9880 * @extack: netlink extended ack
9881 *
9882 * Return: 0 on success, -errno on failure.
9883 */
9884 int netif_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9885 struct netlink_ext_ack *extack)
9886 {
9887 struct netdev_notifier_pre_changeaddr_info info = {
9888 .info.dev = dev,
9889 .info.extack = extack,
9890 .dev_addr = addr,
9891 };
9892 int rc;
9893
9894 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9895 return notifier_to_errno(rc);
9896 }
9897 EXPORT_SYMBOL_NS_GPL(netif_pre_changeaddr_notify, "NETDEV_INTERNAL");
9898
9899 int netif_set_mac_address(struct net_device *dev, struct sockaddr_storage *ss,
9900 struct netlink_ext_ack *extack)
9901 {
9902 const struct net_device_ops *ops = dev->netdev_ops;
9903 int err;
9904
9905 if (!ops->ndo_set_mac_address)
9906 return -EOPNOTSUPP;
9907 if (ss->ss_family != dev->type)
9908 return -EINVAL;
9909 if (!netif_device_present(dev))
9910 return -ENODEV;
9911 err = netif_pre_changeaddr_notify(dev, ss->__data, extack);
9912 if (err)
9913 return err;
9914 if (memcmp(dev->dev_addr, ss->__data, dev->addr_len)) {
9915 err = ops->ndo_set_mac_address(dev, ss);
9916 if (err)
9917 return err;
9918 }
9919 dev->addr_assign_type = NET_ADDR_SET;
9920 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9921 add_device_randomness(dev->dev_addr, dev->addr_len);
9922 return 0;
9923 }
9924
9925 DECLARE_RWSEM(dev_addr_sem);
9926
9927 /* "sa" is a true struct sockaddr with limited "sa_data" member. */
9928 int netif_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9929 {
9930 size_t size = sizeof(sa->sa_data_min);
9931 struct net_device *dev;
9932 int ret = 0;
9933
9934 down_read(&dev_addr_sem);
9935 rcu_read_lock();
9936
9937 dev = dev_get_by_name_rcu(net, dev_name);
9938 if (!dev) {
9939 ret = -ENODEV;
9940 goto unlock;
9941 }
9942 if (!dev->addr_len)
9943 memset(sa->sa_data, 0, size);
9944 else
9945 memcpy(sa->sa_data, dev->dev_addr,
9946 min_t(size_t, size, dev->addr_len));
9947 sa->sa_family = dev->type;
9948
9949 unlock:
9950 rcu_read_unlock();
9951 up_read(&dev_addr_sem);
9952 return ret;
9953 }
9954 EXPORT_SYMBOL_NS_GPL(netif_get_mac_address, "NETDEV_INTERNAL");
9955
9956 int netif_change_carrier(struct net_device *dev, bool new_carrier)
9957 {
9958 const struct net_device_ops *ops = dev->netdev_ops;
9959
9960 if (!ops->ndo_change_carrier)
9961 return -EOPNOTSUPP;
9962 if (!netif_device_present(dev))
9963 return -ENODEV;
9964 return ops->ndo_change_carrier(dev, new_carrier);
9965 }
9966
9967 /**
9968 * dev_get_phys_port_id - Get device physical port ID
9969 * @dev: device
9970 * @ppid: port ID
9971 *
9972 * Get device physical port ID
9973 */
9974 int dev_get_phys_port_id(struct net_device *dev,
9975 struct netdev_phys_item_id *ppid)
9976 {
9977 const struct net_device_ops *ops = dev->netdev_ops;
9978
9979 if (!ops->ndo_get_phys_port_id)
9980 return -EOPNOTSUPP;
9981 return ops->ndo_get_phys_port_id(dev, ppid);
9982 }
9983
9984 /**
9985 * dev_get_phys_port_name - Get device physical port name
9986 * @dev: device
9987 * @name: port name
9988 * @len: limit of bytes to copy to name
9989 *
9990 * Get device physical port name
9991 */
9992 int dev_get_phys_port_name(struct net_device *dev,
9993 char *name, size_t len)
9994 {
9995 const struct net_device_ops *ops = dev->netdev_ops;
9996 int err;
9997
9998 if (ops->ndo_get_phys_port_name) {
9999 err = ops->ndo_get_phys_port_name(dev, name, len);
10000 if (err != -EOPNOTSUPP)
10001 return err;
10002 }
10003 return devlink_compat_phys_port_name_get(dev, name, len);
10004 }
10005
10006 /**
10007 * netif_get_port_parent_id() - Get the device's port parent identifier
10008 * @dev: network device
10009 * @ppid: pointer to a storage for the port's parent identifier
10010 * @recurse: allow/disallow recursion to lower devices
10011 *
10012 * Get the devices's port parent identifier.
10013 *
10014 * Return: 0 on success, -errno on failure.
10015 */
10016 int netif_get_port_parent_id(struct net_device *dev,
10017 struct netdev_phys_item_id *ppid, bool recurse)
10018 {
10019 const struct net_device_ops *ops = dev->netdev_ops;
10020 struct netdev_phys_item_id first = { };
10021 struct net_device *lower_dev;
10022 struct list_head *iter;
10023 int err;
10024
10025 if (ops->ndo_get_port_parent_id) {
10026 err = ops->ndo_get_port_parent_id(dev, ppid);
10027 if (err != -EOPNOTSUPP)
10028 return err;
10029 }
10030
10031 err = devlink_compat_switch_id_get(dev, ppid);
10032 if (!recurse || err != -EOPNOTSUPP)
10033 return err;
10034
10035 netdev_for_each_lower_dev(dev, lower_dev, iter) {
10036 err = netif_get_port_parent_id(lower_dev, ppid, true);
10037 if (err)
10038 break;
10039 if (!first.id_len)
10040 first = *ppid;
10041 else if (memcmp(&first, ppid, sizeof(*ppid)))
10042 return -EOPNOTSUPP;
10043 }
10044
10045 return err;
10046 }
10047 EXPORT_SYMBOL(netif_get_port_parent_id);
10048
10049 /**
10050 * netdev_port_same_parent_id - Indicate if two network devices have
10051 * the same port parent identifier
10052 * @a: first network device
10053 * @b: second network device
10054 */
10055 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
10056 {
10057 struct netdev_phys_item_id a_id = { };
10058 struct netdev_phys_item_id b_id = { };
10059
10060 if (netif_get_port_parent_id(a, &a_id, true) ||
10061 netif_get_port_parent_id(b, &b_id, true))
10062 return false;
10063
10064 return netdev_phys_item_id_same(&a_id, &b_id);
10065 }
10066 EXPORT_SYMBOL(netdev_port_same_parent_id);
10067
10068 int netif_change_proto_down(struct net_device *dev, bool proto_down)
10069 {
10070 if (!dev->change_proto_down)
10071 return -EOPNOTSUPP;
10072 if (!netif_device_present(dev))
10073 return -ENODEV;
10074 if (proto_down)
10075 netif_carrier_off(dev);
10076 else
10077 netif_carrier_on(dev);
10078 WRITE_ONCE(dev->proto_down, proto_down);
10079 return 0;
10080 }
10081
10082 /**
10083 * netdev_change_proto_down_reason_locked - proto down reason
10084 *
10085 * @dev: device
10086 * @mask: proto down mask
10087 * @value: proto down value
10088 */
10089 void netdev_change_proto_down_reason_locked(struct net_device *dev,
10090 unsigned long mask, u32 value)
10091 {
10092 u32 proto_down_reason;
10093 int b;
10094
10095 if (!mask) {
10096 proto_down_reason = value;
10097 } else {
10098 proto_down_reason = dev->proto_down_reason;
10099 for_each_set_bit(b, &mask, 32) {
10100 if (value & (1 << b))
10101 proto_down_reason |= BIT(b);
10102 else
10103 proto_down_reason &= ~BIT(b);
10104 }
10105 }
10106 WRITE_ONCE(dev->proto_down_reason, proto_down_reason);
10107 }
10108
10109 struct bpf_xdp_link {
10110 struct bpf_link link;
10111 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
10112 int flags;
10113 };
10114
10115 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
10116 {
10117 if (flags & XDP_FLAGS_HW_MODE)
10118 return XDP_MODE_HW;
10119 if (flags & XDP_FLAGS_DRV_MODE)
10120 return XDP_MODE_DRV;
10121 if (flags & XDP_FLAGS_SKB_MODE)
10122 return XDP_MODE_SKB;
10123 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
10124 }
10125
10126 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
10127 {
10128 switch (mode) {
10129 case XDP_MODE_SKB:
10130 return generic_xdp_install;
10131 case XDP_MODE_DRV:
10132 case XDP_MODE_HW:
10133 return dev->netdev_ops->ndo_bpf;
10134 default:
10135 return NULL;
10136 }
10137 }
10138
10139 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
10140 enum bpf_xdp_mode mode)
10141 {
10142 return dev->xdp_state[mode].link;
10143 }
10144
10145 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
10146 enum bpf_xdp_mode mode)
10147 {
10148 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
10149
10150 if (link)
10151 return link->link.prog;
10152 return dev->xdp_state[mode].prog;
10153 }
10154
10155 u8 dev_xdp_prog_count(struct net_device *dev)
10156 {
10157 u8 count = 0;
10158 int i;
10159
10160 for (i = 0; i < __MAX_XDP_MODE; i++)
10161 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
10162 count++;
10163 return count;
10164 }
10165 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
10166
10167 u8 dev_xdp_sb_prog_count(struct net_device *dev)
10168 {
10169 u8 count = 0;
10170 int i;
10171
10172 for (i = 0; i < __MAX_XDP_MODE; i++)
10173 if (dev->xdp_state[i].prog &&
10174 !dev->xdp_state[i].prog->aux->xdp_has_frags)
10175 count++;
10176 return count;
10177 }
10178
10179 int netif_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf)
10180 {
10181 if (!dev->netdev_ops->ndo_bpf)
10182 return -EOPNOTSUPP;
10183
10184 if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
10185 bpf->command == XDP_SETUP_PROG &&
10186 bpf->prog && !bpf->prog->aux->xdp_has_frags) {
10187 NL_SET_ERR_MSG(bpf->extack,
10188 "unable to propagate XDP to device using tcp-data-split");
10189 return -EBUSY;
10190 }
10191
10192 if (dev_get_min_mp_channel_count(dev)) {
10193 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider");
10194 return -EBUSY;
10195 }
10196
10197 return dev->netdev_ops->ndo_bpf(dev, bpf);
10198 }
10199 EXPORT_SYMBOL_GPL(netif_xdp_propagate);
10200
10201 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
10202 {
10203 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
10204
10205 return prog ? prog->aux->id : 0;
10206 }
10207
10208 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
10209 struct bpf_xdp_link *link)
10210 {
10211 dev->xdp_state[mode].link = link;
10212 dev->xdp_state[mode].prog = NULL;
10213 }
10214
10215 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
10216 struct bpf_prog *prog)
10217 {
10218 dev->xdp_state[mode].link = NULL;
10219 dev->xdp_state[mode].prog = prog;
10220 }
10221
10222 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
10223 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
10224 u32 flags, struct bpf_prog *prog)
10225 {
10226 struct netdev_bpf xdp;
10227 int err;
10228
10229 netdev_ops_assert_locked(dev);
10230
10231 if (dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
10232 prog && !prog->aux->xdp_has_frags) {
10233 NL_SET_ERR_MSG(extack, "unable to install XDP to device using tcp-data-split");
10234 return -EBUSY;
10235 }
10236
10237 if (dev_get_min_mp_channel_count(dev)) {
10238 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider");
10239 return -EBUSY;
10240 }
10241
10242 memset(&xdp, 0, sizeof(xdp));
10243 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
10244 xdp.extack = extack;
10245 xdp.flags = flags;
10246 xdp.prog = prog;
10247
10248 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
10249 * "moved" into driver), so they don't increment it on their own, but
10250 * they do decrement refcnt when program is detached or replaced.
10251 * Given net_device also owns link/prog, we need to bump refcnt here
10252 * to prevent drivers from underflowing it.
10253 */
10254 if (prog)
10255 bpf_prog_inc(prog);
10256 err = bpf_op(dev, &xdp);
10257 if (err) {
10258 if (prog)
10259 bpf_prog_put(prog);
10260 return err;
10261 }
10262
10263 if (mode != XDP_MODE_HW)
10264 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
10265
10266 return 0;
10267 }
10268
10269 static void dev_xdp_uninstall(struct net_device *dev)
10270 {
10271 struct bpf_xdp_link *link;
10272 struct bpf_prog *prog;
10273 enum bpf_xdp_mode mode;
10274 bpf_op_t bpf_op;
10275
10276 ASSERT_RTNL();
10277
10278 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
10279 prog = dev_xdp_prog(dev, mode);
10280 if (!prog)
10281 continue;
10282
10283 bpf_op = dev_xdp_bpf_op(dev, mode);
10284 if (!bpf_op)
10285 continue;
10286
10287 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
10288
10289 /* auto-detach link from net device */
10290 link = dev_xdp_link(dev, mode);
10291 if (link)
10292 link->dev = NULL;
10293 else
10294 bpf_prog_put(prog);
10295
10296 dev_xdp_set_link(dev, mode, NULL);
10297 }
10298 }
10299
10300 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
10301 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
10302 struct bpf_prog *old_prog, u32 flags)
10303 {
10304 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
10305 struct bpf_prog *cur_prog;
10306 struct net_device *upper;
10307 struct list_head *iter;
10308 enum bpf_xdp_mode mode;
10309 bpf_op_t bpf_op;
10310 int err;
10311
10312 ASSERT_RTNL();
10313
10314 /* either link or prog attachment, never both */
10315 if (link && (new_prog || old_prog))
10316 return -EINVAL;
10317 /* link supports only XDP mode flags */
10318 if (link && (flags & ~XDP_FLAGS_MODES)) {
10319 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
10320 return -EINVAL;
10321 }
10322 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
10323 if (num_modes > 1) {
10324 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
10325 return -EINVAL;
10326 }
10327 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
10328 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
10329 NL_SET_ERR_MSG(extack,
10330 "More than one program loaded, unset mode is ambiguous");
10331 return -EINVAL;
10332 }
10333 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
10334 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
10335 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
10336 return -EINVAL;
10337 }
10338
10339 mode = dev_xdp_mode(dev, flags);
10340 /* can't replace attached link */
10341 if (dev_xdp_link(dev, mode)) {
10342 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
10343 return -EBUSY;
10344 }
10345
10346 /* don't allow if an upper device already has a program */
10347 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
10348 if (dev_xdp_prog_count(upper) > 0) {
10349 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
10350 return -EEXIST;
10351 }
10352 }
10353
10354 cur_prog = dev_xdp_prog(dev, mode);
10355 /* can't replace attached prog with link */
10356 if (link && cur_prog) {
10357 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
10358 return -EBUSY;
10359 }
10360 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
10361 NL_SET_ERR_MSG(extack, "Active program does not match expected");
10362 return -EEXIST;
10363 }
10364
10365 /* put effective new program into new_prog */
10366 if (link)
10367 new_prog = link->link.prog;
10368
10369 if (new_prog) {
10370 bool offload = mode == XDP_MODE_HW;
10371 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
10372 ? XDP_MODE_DRV : XDP_MODE_SKB;
10373
10374 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
10375 NL_SET_ERR_MSG(extack, "XDP program already attached");
10376 return -EBUSY;
10377 }
10378 if (!offload && dev_xdp_prog(dev, other_mode)) {
10379 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
10380 return -EEXIST;
10381 }
10382 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
10383 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
10384 return -EINVAL;
10385 }
10386 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
10387 NL_SET_ERR_MSG(extack, "Program bound to different device");
10388 return -EINVAL;
10389 }
10390 if (bpf_prog_is_dev_bound(new_prog->aux) && mode == XDP_MODE_SKB) {
10391 NL_SET_ERR_MSG(extack, "Can't attach device-bound programs in generic mode");
10392 return -EINVAL;
10393 }
10394 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
10395 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
10396 return -EINVAL;
10397 }
10398 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
10399 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
10400 return -EINVAL;
10401 }
10402 }
10403
10404 /* don't call drivers if the effective program didn't change */
10405 if (new_prog != cur_prog) {
10406 bpf_op = dev_xdp_bpf_op(dev, mode);
10407 if (!bpf_op) {
10408 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
10409 return -EOPNOTSUPP;
10410 }
10411
10412 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
10413 if (err)
10414 return err;
10415 }
10416
10417 if (link)
10418 dev_xdp_set_link(dev, mode, link);
10419 else
10420 dev_xdp_set_prog(dev, mode, new_prog);
10421 if (cur_prog)
10422 bpf_prog_put(cur_prog);
10423
10424 return 0;
10425 }
10426
10427 static int dev_xdp_attach_link(struct net_device *dev,
10428 struct netlink_ext_ack *extack,
10429 struct bpf_xdp_link *link)
10430 {
10431 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
10432 }
10433
10434 static int dev_xdp_detach_link(struct net_device *dev,
10435 struct netlink_ext_ack *extack,
10436 struct bpf_xdp_link *link)
10437 {
10438 enum bpf_xdp_mode mode;
10439 bpf_op_t bpf_op;
10440
10441 ASSERT_RTNL();
10442
10443 mode = dev_xdp_mode(dev, link->flags);
10444 if (dev_xdp_link(dev, mode) != link)
10445 return -EINVAL;
10446
10447 bpf_op = dev_xdp_bpf_op(dev, mode);
10448 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
10449 dev_xdp_set_link(dev, mode, NULL);
10450 return 0;
10451 }
10452
10453 static void bpf_xdp_link_release(struct bpf_link *link)
10454 {
10455 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
10456
10457 rtnl_lock();
10458
10459 /* if racing with net_device's tear down, xdp_link->dev might be
10460 * already NULL, in which case link was already auto-detached
10461 */
10462 if (xdp_link->dev) {
10463 netdev_lock_ops(xdp_link->dev);
10464 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
10465 netdev_unlock_ops(xdp_link->dev);
10466 xdp_link->dev = NULL;
10467 }
10468
10469 rtnl_unlock();
10470 }
10471
10472 static int bpf_xdp_link_detach(struct bpf_link *link)
10473 {
10474 bpf_xdp_link_release(link);
10475 return 0;
10476 }
10477
10478 static void bpf_xdp_link_dealloc(struct bpf_link *link)
10479 {
10480 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
10481
10482 kfree(xdp_link);
10483 }
10484
10485 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
10486 struct seq_file *seq)
10487 {
10488 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
10489 u32 ifindex = 0;
10490
10491 rtnl_lock();
10492 if (xdp_link->dev)
10493 ifindex = xdp_link->dev->ifindex;
10494 rtnl_unlock();
10495
10496 seq_printf(seq, "ifindex:\t%u\n", ifindex);
10497 }
10498
10499 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
10500 struct bpf_link_info *info)
10501 {
10502 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
10503 u32 ifindex = 0;
10504
10505 rtnl_lock();
10506 if (xdp_link->dev)
10507 ifindex = xdp_link->dev->ifindex;
10508 rtnl_unlock();
10509
10510 info->xdp.ifindex = ifindex;
10511 return 0;
10512 }
10513
10514 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
10515 struct bpf_prog *old_prog)
10516 {
10517 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
10518 enum bpf_xdp_mode mode;
10519 bpf_op_t bpf_op;
10520 int err = 0;
10521
10522 rtnl_lock();
10523
10524 /* link might have been auto-released already, so fail */
10525 if (!xdp_link->dev) {
10526 err = -ENOLINK;
10527 goto out_unlock;
10528 }
10529
10530 if (old_prog && link->prog != old_prog) {
10531 err = -EPERM;
10532 goto out_unlock;
10533 }
10534 old_prog = link->prog;
10535 if (old_prog->type != new_prog->type ||
10536 old_prog->expected_attach_type != new_prog->expected_attach_type) {
10537 err = -EINVAL;
10538 goto out_unlock;
10539 }
10540
10541 if (old_prog == new_prog) {
10542 /* no-op, don't disturb drivers */
10543 bpf_prog_put(new_prog);
10544 goto out_unlock;
10545 }
10546
10547 netdev_lock_ops(xdp_link->dev);
10548 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
10549 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
10550 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
10551 xdp_link->flags, new_prog);
10552 netdev_unlock_ops(xdp_link->dev);
10553 if (err)
10554 goto out_unlock;
10555
10556 old_prog = xchg(&link->prog, new_prog);
10557 bpf_prog_put(old_prog);
10558
10559 out_unlock:
10560 rtnl_unlock();
10561 return err;
10562 }
10563
10564 static const struct bpf_link_ops bpf_xdp_link_lops = {
10565 .release = bpf_xdp_link_release,
10566 .dealloc = bpf_xdp_link_dealloc,
10567 .detach = bpf_xdp_link_detach,
10568 .show_fdinfo = bpf_xdp_link_show_fdinfo,
10569 .fill_link_info = bpf_xdp_link_fill_link_info,
10570 .update_prog = bpf_xdp_link_update,
10571 };
10572
10573 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
10574 {
10575 struct net *net = current->nsproxy->net_ns;
10576 struct bpf_link_primer link_primer;
10577 struct netlink_ext_ack extack = {};
10578 struct bpf_xdp_link *link;
10579 struct net_device *dev;
10580 int err, fd;
10581
10582 rtnl_lock();
10583 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
10584 if (!dev) {
10585 rtnl_unlock();
10586 return -EINVAL;
10587 }
10588
10589 link = kzalloc(sizeof(*link), GFP_USER);
10590 if (!link) {
10591 err = -ENOMEM;
10592 goto unlock;
10593 }
10594
10595 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog,
10596 attr->link_create.attach_type);
10597 link->dev = dev;
10598 link->flags = attr->link_create.flags;
10599
10600 err = bpf_link_prime(&link->link, &link_primer);
10601 if (err) {
10602 kfree(link);
10603 goto unlock;
10604 }
10605
10606 netdev_lock_ops(dev);
10607 err = dev_xdp_attach_link(dev, &extack, link);
10608 netdev_unlock_ops(dev);
10609 rtnl_unlock();
10610
10611 if (err) {
10612 link->dev = NULL;
10613 bpf_link_cleanup(&link_primer);
10614 trace_bpf_xdp_link_attach_failed(extack._msg);
10615 goto out_put_dev;
10616 }
10617
10618 fd = bpf_link_settle(&link_primer);
10619 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
10620 dev_put(dev);
10621 return fd;
10622
10623 unlock:
10624 rtnl_unlock();
10625
10626 out_put_dev:
10627 dev_put(dev);
10628 return err;
10629 }
10630
10631 /**
10632 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
10633 * @dev: device
10634 * @extack: netlink extended ack
10635 * @fd: new program fd or negative value to clear
10636 * @expected_fd: old program fd that userspace expects to replace or clear
10637 * @flags: xdp-related flags
10638 *
10639 * Set or clear a bpf program for a device
10640 */
10641 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
10642 int fd, int expected_fd, u32 flags)
10643 {
10644 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
10645 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
10646 int err;
10647
10648 ASSERT_RTNL();
10649
10650 if (fd >= 0) {
10651 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
10652 mode != XDP_MODE_SKB);
10653 if (IS_ERR(new_prog))
10654 return PTR_ERR(new_prog);
10655 }
10656
10657 if (expected_fd >= 0) {
10658 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
10659 mode != XDP_MODE_SKB);
10660 if (IS_ERR(old_prog)) {
10661 err = PTR_ERR(old_prog);
10662 old_prog = NULL;
10663 goto err_out;
10664 }
10665 }
10666
10667 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
10668
10669 err_out:
10670 if (err && new_prog)
10671 bpf_prog_put(new_prog);
10672 if (old_prog)
10673 bpf_prog_put(old_prog);
10674 return err;
10675 }
10676
10677 u32 dev_get_min_mp_channel_count(const struct net_device *dev)
10678 {
10679 int i;
10680
10681 netdev_ops_assert_locked(dev);
10682
10683 for (i = dev->real_num_rx_queues - 1; i >= 0; i--)
10684 if (dev->_rx[i].mp_params.mp_priv)
10685 /* The channel count is the idx plus 1. */
10686 return i + 1;
10687
10688 return 0;
10689 }
10690
10691 /**
10692 * dev_index_reserve() - allocate an ifindex in a namespace
10693 * @net: the applicable net namespace
10694 * @ifindex: requested ifindex, pass %0 to get one allocated
10695 *
10696 * Allocate a ifindex for a new device. Caller must either use the ifindex
10697 * to store the device (via list_netdevice()) or call dev_index_release()
10698 * to give the index up.
10699 *
10700 * Return: a suitable unique value for a new device interface number or -errno.
10701 */
10702 static int dev_index_reserve(struct net *net, u32 ifindex)
10703 {
10704 int err;
10705
10706 if (ifindex > INT_MAX) {
10707 DEBUG_NET_WARN_ON_ONCE(1);
10708 return -EINVAL;
10709 }
10710
10711 if (!ifindex)
10712 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL,
10713 xa_limit_31b, &net->ifindex, GFP_KERNEL);
10714 else
10715 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL);
10716 if (err < 0)
10717 return err;
10718
10719 return ifindex;
10720 }
10721
10722 static void dev_index_release(struct net *net, int ifindex)
10723 {
10724 /* Expect only unused indexes, unlist_netdevice() removes the used */
10725 WARN_ON(xa_erase(&net->dev_by_index, ifindex));
10726 }
10727
10728 static bool from_cleanup_net(void)
10729 {
10730 #ifdef CONFIG_NET_NS
10731 return current == READ_ONCE(cleanup_net_task);
10732 #else
10733 return false;
10734 #endif
10735 }
10736
10737 /* Delayed registration/unregisteration */
10738 LIST_HEAD(net_todo_list);
10739 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
10740 atomic_t dev_unreg_count = ATOMIC_INIT(0);
10741
10742 static void net_set_todo(struct net_device *dev)
10743 {
10744 list_add_tail(&dev->todo_list, &net_todo_list);
10745 }
10746
10747 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
10748 struct net_device *upper, netdev_features_t features)
10749 {
10750 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
10751 netdev_features_t feature;
10752 int feature_bit;
10753
10754 for_each_netdev_feature(upper_disables, feature_bit) {
10755 feature = __NETIF_F_BIT(feature_bit);
10756 if (!(upper->wanted_features & feature)
10757 && (features & feature)) {
10758 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
10759 &feature, upper->name);
10760 features &= ~feature;
10761 }
10762 }
10763
10764 return features;
10765 }
10766
10767 static void netdev_sync_lower_features(struct net_device *upper,
10768 struct net_device *lower, netdev_features_t features)
10769 {
10770 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
10771 netdev_features_t feature;
10772 int feature_bit;
10773
10774 for_each_netdev_feature(upper_disables, feature_bit) {
10775 feature = __NETIF_F_BIT(feature_bit);
10776 if (!(features & feature) && (lower->features & feature)) {
10777 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
10778 &feature, lower->name);
10779 netdev_lock_ops(lower);
10780 lower->wanted_features &= ~feature;
10781 __netdev_update_features(lower);
10782
10783 if (unlikely(lower->features & feature))
10784 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
10785 &feature, lower->name);
10786 else
10787 netdev_features_change(lower);
10788 netdev_unlock_ops(lower);
10789 }
10790 }
10791 }
10792
10793 static bool netdev_has_ip_or_hw_csum(netdev_features_t features)
10794 {
10795 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
10796 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask;
10797 bool hw_csum = features & NETIF_F_HW_CSUM;
10798
10799 return ip_csum || hw_csum;
10800 }
10801
10802 static netdev_features_t netdev_fix_features(struct net_device *dev,
10803 netdev_features_t features)
10804 {
10805 /* Fix illegal checksum combinations */
10806 if ((features & NETIF_F_HW_CSUM) &&
10807 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
10808 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
10809 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
10810 }
10811
10812 /* TSO requires that SG is present as well. */
10813 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
10814 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
10815 features &= ~NETIF_F_ALL_TSO;
10816 }
10817
10818 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
10819 !(features & NETIF_F_IP_CSUM)) {
10820 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
10821 features &= ~NETIF_F_TSO;
10822 features &= ~NETIF_F_TSO_ECN;
10823 }
10824
10825 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
10826 !(features & NETIF_F_IPV6_CSUM)) {
10827 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
10828 features &= ~NETIF_F_TSO6;
10829 }
10830
10831 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
10832 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
10833 features &= ~NETIF_F_TSO_MANGLEID;
10834
10835 /* TSO ECN requires that TSO is present as well. */
10836 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
10837 features &= ~NETIF_F_TSO_ECN;
10838
10839 /* Software GSO depends on SG. */
10840 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
10841 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
10842 features &= ~NETIF_F_GSO;
10843 }
10844
10845 /* GSO partial features require GSO partial be set */
10846 if ((features & dev->gso_partial_features) &&
10847 !(features & NETIF_F_GSO_PARTIAL)) {
10848 netdev_dbg(dev,
10849 "Dropping partially supported GSO features since no GSO partial.\n");
10850 features &= ~dev->gso_partial_features;
10851 }
10852
10853 if (!(features & NETIF_F_RXCSUM)) {
10854 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
10855 * successfully merged by hardware must also have the
10856 * checksum verified by hardware. If the user does not
10857 * want to enable RXCSUM, logically, we should disable GRO_HW.
10858 */
10859 if (features & NETIF_F_GRO_HW) {
10860 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
10861 features &= ~NETIF_F_GRO_HW;
10862 }
10863 }
10864
10865 /* LRO/HW-GRO features cannot be combined with RX-FCS */
10866 if (features & NETIF_F_RXFCS) {
10867 if (features & NETIF_F_LRO) {
10868 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
10869 features &= ~NETIF_F_LRO;
10870 }
10871
10872 if (features & NETIF_F_GRO_HW) {
10873 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
10874 features &= ~NETIF_F_GRO_HW;
10875 }
10876 }
10877
10878 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
10879 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
10880 features &= ~NETIF_F_LRO;
10881 }
10882
10883 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) {
10884 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
10885 features &= ~NETIF_F_HW_TLS_TX;
10886 }
10887
10888 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
10889 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
10890 features &= ~NETIF_F_HW_TLS_RX;
10891 }
10892
10893 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) {
10894 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n");
10895 features &= ~NETIF_F_GSO_UDP_L4;
10896 }
10897
10898 return features;
10899 }
10900
10901 int __netdev_update_features(struct net_device *dev)
10902 {
10903 struct net_device *upper, *lower;
10904 netdev_features_t features;
10905 struct list_head *iter;
10906 int err = -1;
10907
10908 ASSERT_RTNL();
10909 netdev_ops_assert_locked(dev);
10910
10911 features = netdev_get_wanted_features(dev);
10912
10913 if (dev->netdev_ops->ndo_fix_features)
10914 features = dev->netdev_ops->ndo_fix_features(dev, features);
10915
10916 /* driver might be less strict about feature dependencies */
10917 features = netdev_fix_features(dev, features);
10918
10919 /* some features can't be enabled if they're off on an upper device */
10920 netdev_for_each_upper_dev_rcu(dev, upper, iter)
10921 features = netdev_sync_upper_features(dev, upper, features);
10922
10923 if (dev->features == features)
10924 goto sync_lower;
10925
10926 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10927 &dev->features, &features);
10928
10929 if (dev->netdev_ops->ndo_set_features)
10930 err = dev->netdev_ops->ndo_set_features(dev, features);
10931 else
10932 err = 0;
10933
10934 if (unlikely(err < 0)) {
10935 netdev_err(dev,
10936 "set_features() failed (%d); wanted %pNF, left %pNF\n",
10937 err, &features, &dev->features);
10938 /* return non-0 since some features might have changed and
10939 * it's better to fire a spurious notification than miss it
10940 */
10941 return -1;
10942 }
10943
10944 sync_lower:
10945 /* some features must be disabled on lower devices when disabled
10946 * on an upper device (think: bonding master or bridge)
10947 */
10948 netdev_for_each_lower_dev(dev, lower, iter)
10949 netdev_sync_lower_features(dev, lower, features);
10950
10951 if (!err) {
10952 netdev_features_t diff = features ^ dev->features;
10953
10954 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10955 /* udp_tunnel_{get,drop}_rx_info both need
10956 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10957 * device, or they won't do anything.
10958 * Thus we need to update dev->features
10959 * *before* calling udp_tunnel_get_rx_info,
10960 * but *after* calling udp_tunnel_drop_rx_info.
10961 */
10962 udp_tunnel_nic_lock(dev);
10963 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10964 dev->features = features;
10965 udp_tunnel_get_rx_info(dev);
10966 } else {
10967 udp_tunnel_drop_rx_info(dev);
10968 }
10969 udp_tunnel_nic_unlock(dev);
10970 }
10971
10972 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10973 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10974 dev->features = features;
10975 err |= vlan_get_rx_ctag_filter_info(dev);
10976 } else {
10977 vlan_drop_rx_ctag_filter_info(dev);
10978 }
10979 }
10980
10981 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10982 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10983 dev->features = features;
10984 err |= vlan_get_rx_stag_filter_info(dev);
10985 } else {
10986 vlan_drop_rx_stag_filter_info(dev);
10987 }
10988 }
10989
10990 dev->features = features;
10991 }
10992
10993 return err < 0 ? 0 : 1;
10994 }
10995
10996 /**
10997 * netdev_update_features - recalculate device features
10998 * @dev: the device to check
10999 *
11000 * Recalculate dev->features set and send notifications if it
11001 * has changed. Should be called after driver or hardware dependent
11002 * conditions might have changed that influence the features.
11003 */
11004 void netdev_update_features(struct net_device *dev)
11005 {
11006 if (__netdev_update_features(dev))
11007 netdev_features_change(dev);
11008 }
11009 EXPORT_SYMBOL(netdev_update_features);
11010
11011 /**
11012 * netdev_change_features - recalculate device features
11013 * @dev: the device to check
11014 *
11015 * Recalculate dev->features set and send notifications even
11016 * if they have not changed. Should be called instead of
11017 * netdev_update_features() if also dev->vlan_features might
11018 * have changed to allow the changes to be propagated to stacked
11019 * VLAN devices.
11020 */
11021 void netdev_change_features(struct net_device *dev)
11022 {
11023 __netdev_update_features(dev);
11024 netdev_features_change(dev);
11025 }
11026 EXPORT_SYMBOL(netdev_change_features);
11027
11028 /**
11029 * netif_stacked_transfer_operstate - transfer operstate
11030 * @rootdev: the root or lower level device to transfer state from
11031 * @dev: the device to transfer operstate to
11032 *
11033 * Transfer operational state from root to device. This is normally
11034 * called when a stacking relationship exists between the root
11035 * device and the device(a leaf device).
11036 */
11037 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
11038 struct net_device *dev)
11039 {
11040 if (rootdev->operstate == IF_OPER_DORMANT)
11041 netif_dormant_on(dev);
11042 else
11043 netif_dormant_off(dev);
11044
11045 if (rootdev->operstate == IF_OPER_TESTING)
11046 netif_testing_on(dev);
11047 else
11048 netif_testing_off(dev);
11049
11050 if (netif_carrier_ok(rootdev))
11051 netif_carrier_on(dev);
11052 else
11053 netif_carrier_off(dev);
11054 }
11055 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
11056
11057 static int netif_alloc_rx_queues(struct net_device *dev)
11058 {
11059 unsigned int i, count = dev->num_rx_queues;
11060 struct netdev_rx_queue *rx;
11061 size_t sz = count * sizeof(*rx);
11062 int err = 0;
11063
11064 BUG_ON(count < 1);
11065
11066 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11067 if (!rx)
11068 return -ENOMEM;
11069
11070 dev->_rx = rx;
11071
11072 for (i = 0; i < count; i++) {
11073 rx[i].dev = dev;
11074
11075 /* XDP RX-queue setup */
11076 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
11077 if (err < 0)
11078 goto err_rxq_info;
11079 }
11080 return 0;
11081
11082 err_rxq_info:
11083 /* Rollback successful reg's and free other resources */
11084 while (i--)
11085 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
11086 kvfree(dev->_rx);
11087 dev->_rx = NULL;
11088 return err;
11089 }
11090
11091 static void netif_free_rx_queues(struct net_device *dev)
11092 {
11093 unsigned int i, count = dev->num_rx_queues;
11094
11095 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
11096 if (!dev->_rx)
11097 return;
11098
11099 for (i = 0; i < count; i++)
11100 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
11101
11102 kvfree(dev->_rx);
11103 }
11104
11105 static void netdev_init_one_queue(struct net_device *dev,
11106 struct netdev_queue *queue, void *_unused)
11107 {
11108 /* Initialize queue lock */
11109 spin_lock_init(&queue->_xmit_lock);
11110 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
11111 queue->xmit_lock_owner = -1;
11112 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
11113 queue->dev = dev;
11114 #ifdef CONFIG_BQL
11115 dql_init(&queue->dql, HZ);
11116 #endif
11117 }
11118
11119 static void netif_free_tx_queues(struct net_device *dev)
11120 {
11121 kvfree(dev->_tx);
11122 }
11123
11124 static int netif_alloc_netdev_queues(struct net_device *dev)
11125 {
11126 unsigned int count = dev->num_tx_queues;
11127 struct netdev_queue *tx;
11128 size_t sz = count * sizeof(*tx);
11129
11130 if (count < 1 || count > 0xffff)
11131 return -EINVAL;
11132
11133 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11134 if (!tx)
11135 return -ENOMEM;
11136
11137 dev->_tx = tx;
11138
11139 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
11140 spin_lock_init(&dev->tx_global_lock);
11141
11142 return 0;
11143 }
11144
11145 void netif_tx_stop_all_queues(struct net_device *dev)
11146 {
11147 unsigned int i;
11148
11149 for (i = 0; i < dev->num_tx_queues; i++) {
11150 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
11151
11152 netif_tx_stop_queue(txq);
11153 }
11154 }
11155 EXPORT_SYMBOL(netif_tx_stop_all_queues);
11156
11157 static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
11158 {
11159 void __percpu *v;
11160
11161 /* Drivers implementing ndo_get_peer_dev must support tstat
11162 * accounting, so that skb_do_redirect() can bump the dev's
11163 * RX stats upon network namespace switch.
11164 */
11165 if (dev->netdev_ops->ndo_get_peer_dev &&
11166 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
11167 return -EOPNOTSUPP;
11168
11169 switch (dev->pcpu_stat_type) {
11170 case NETDEV_PCPU_STAT_NONE:
11171 return 0;
11172 case NETDEV_PCPU_STAT_LSTATS:
11173 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
11174 break;
11175 case NETDEV_PCPU_STAT_TSTATS:
11176 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
11177 break;
11178 case NETDEV_PCPU_STAT_DSTATS:
11179 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
11180 break;
11181 default:
11182 return -EINVAL;
11183 }
11184
11185 return v ? 0 : -ENOMEM;
11186 }
11187
11188 static void netdev_do_free_pcpu_stats(struct net_device *dev)
11189 {
11190 switch (dev->pcpu_stat_type) {
11191 case NETDEV_PCPU_STAT_NONE:
11192 return;
11193 case NETDEV_PCPU_STAT_LSTATS:
11194 free_percpu(dev->lstats);
11195 break;
11196 case NETDEV_PCPU_STAT_TSTATS:
11197 free_percpu(dev->tstats);
11198 break;
11199 case NETDEV_PCPU_STAT_DSTATS:
11200 free_percpu(dev->dstats);
11201 break;
11202 }
11203 }
11204
11205 static void netdev_free_phy_link_topology(struct net_device *dev)
11206 {
11207 struct phy_link_topology *topo = dev->link_topo;
11208
11209 if (IS_ENABLED(CONFIG_PHYLIB) && topo) {
11210 xa_destroy(&topo->phys);
11211 kfree(topo);
11212 dev->link_topo = NULL;
11213 }
11214 }
11215
11216 /**
11217 * register_netdevice() - register a network device
11218 * @dev: device to register
11219 *
11220 * Take a prepared network device structure and make it externally accessible.
11221 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
11222 * Callers must hold the rtnl lock - you may want register_netdev()
11223 * instead of this.
11224 */
11225 int register_netdevice(struct net_device *dev)
11226 {
11227 int ret;
11228 struct net *net = dev_net(dev);
11229
11230 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
11231 NETDEV_FEATURE_COUNT);
11232 BUG_ON(dev_boot_phase);
11233 ASSERT_RTNL();
11234
11235 might_sleep();
11236
11237 /* When net_device's are persistent, this will be fatal. */
11238 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
11239 BUG_ON(!net);
11240
11241 ret = ethtool_check_ops(dev->ethtool_ops);
11242 if (ret)
11243 return ret;
11244
11245 /* rss ctx ID 0 is reserved for the default context, start from 1 */
11246 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1);
11247 mutex_init(&dev->ethtool->rss_lock);
11248
11249 spin_lock_init(&dev->addr_list_lock);
11250 netdev_set_addr_lockdep_class(dev);
11251
11252 ret = dev_get_valid_name(net, dev, dev->name);
11253 if (ret < 0)
11254 goto out;
11255
11256 ret = -ENOMEM;
11257 dev->name_node = netdev_name_node_head_alloc(dev);
11258 if (!dev->name_node)
11259 goto out;
11260
11261 /* Init, if this function is available */
11262 if (dev->netdev_ops->ndo_init) {
11263 ret = dev->netdev_ops->ndo_init(dev);
11264 if (ret) {
11265 if (ret > 0)
11266 ret = -EIO;
11267 goto err_free_name;
11268 }
11269 }
11270
11271 if (((dev->hw_features | dev->features) &
11272 NETIF_F_HW_VLAN_CTAG_FILTER) &&
11273 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
11274 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
11275 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
11276 ret = -EINVAL;
11277 goto err_uninit;
11278 }
11279
11280 ret = netdev_do_alloc_pcpu_stats(dev);
11281 if (ret)
11282 goto err_uninit;
11283
11284 ret = dev_index_reserve(net, dev->ifindex);
11285 if (ret < 0)
11286 goto err_free_pcpu;
11287 dev->ifindex = ret;
11288
11289 /* Transfer changeable features to wanted_features and enable
11290 * software offloads (GSO and GRO).
11291 */
11292 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
11293 dev->features |= NETIF_F_SOFT_FEATURES;
11294
11295 if (dev->udp_tunnel_nic_info) {
11296 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
11297 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
11298 }
11299
11300 dev->wanted_features = dev->features & dev->hw_features;
11301
11302 if (!(dev->flags & IFF_LOOPBACK))
11303 dev->hw_features |= NETIF_F_NOCACHE_COPY;
11304
11305 /* If IPv4 TCP segmentation offload is supported we should also
11306 * allow the device to enable segmenting the frame with the option
11307 * of ignoring a static IP ID value. This doesn't enable the
11308 * feature itself but allows the user to enable it later.
11309 */
11310 if (dev->hw_features & NETIF_F_TSO)
11311 dev->hw_features |= NETIF_F_TSO_MANGLEID;
11312 if (dev->vlan_features & NETIF_F_TSO)
11313 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
11314 if (dev->mpls_features & NETIF_F_TSO)
11315 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
11316 if (dev->hw_enc_features & NETIF_F_TSO)
11317 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
11318
11319 /* TSO_MANGLEID belongs in mangleid_features by definition */
11320 dev->mangleid_features |= NETIF_F_TSO_MANGLEID;
11321
11322 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
11323 */
11324 dev->vlan_features |= NETIF_F_HIGHDMA;
11325
11326 /* Make NETIF_F_SG inheritable to tunnel devices.
11327 */
11328 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
11329
11330 /* Make NETIF_F_SG inheritable to MPLS.
11331 */
11332 dev->mpls_features |= NETIF_F_SG;
11333
11334 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
11335 ret = notifier_to_errno(ret);
11336 if (ret)
11337 goto err_ifindex_release;
11338
11339 ret = netdev_register_kobject(dev);
11340
11341 netdev_lock(dev);
11342 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED);
11343 netdev_unlock(dev);
11344
11345 if (ret)
11346 goto err_uninit_notify;
11347
11348 netdev_lock_ops(dev);
11349 __netdev_update_features(dev);
11350 netdev_unlock_ops(dev);
11351
11352 /*
11353 * Default initial state at registry is that the
11354 * device is present.
11355 */
11356
11357 set_bit(__LINK_STATE_PRESENT, &dev->state);
11358
11359 linkwatch_init_dev(dev);
11360
11361 dev_init_scheduler(dev);
11362
11363 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
11364 list_netdevice(dev);
11365
11366 add_device_randomness(dev->dev_addr, dev->addr_len);
11367
11368 /* If the device has permanent device address, driver should
11369 * set dev_addr and also addr_assign_type should be set to
11370 * NET_ADDR_PERM (default value).
11371 */
11372 if (dev->addr_assign_type == NET_ADDR_PERM)
11373 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
11374
11375 /* Notify protocols, that a new device appeared. */
11376 netdev_lock_ops(dev);
11377 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
11378 netdev_unlock_ops(dev);
11379 ret = notifier_to_errno(ret);
11380 if (ret) {
11381 /* Expect explicit free_netdev() on failure */
11382 dev->needs_free_netdev = false;
11383 unregister_netdevice_queue(dev, NULL);
11384 goto out;
11385 }
11386 /*
11387 * Prevent userspace races by waiting until the network
11388 * device is fully setup before sending notifications.
11389 */
11390 if (!(dev->rtnl_link_ops && dev->rtnl_link_initializing))
11391 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11392
11393 out:
11394 return ret;
11395
11396 err_uninit_notify:
11397 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
11398 err_ifindex_release:
11399 dev_index_release(net, dev->ifindex);
11400 err_free_pcpu:
11401 netdev_do_free_pcpu_stats(dev);
11402 err_uninit:
11403 if (dev->netdev_ops->ndo_uninit)
11404 dev->netdev_ops->ndo_uninit(dev);
11405 if (dev->priv_destructor)
11406 dev->priv_destructor(dev);
11407 err_free_name:
11408 netdev_name_node_free(dev->name_node);
11409 goto out;
11410 }
11411 EXPORT_SYMBOL(register_netdevice);
11412
11413 /* Initialize the core of a dummy net device.
11414 * The setup steps dummy netdevs need which normal netdevs get by going
11415 * through register_netdevice().
11416 */
11417 static void init_dummy_netdev(struct net_device *dev)
11418 {
11419 /* make sure we BUG if trying to hit standard
11420 * register/unregister code path
11421 */
11422 dev->reg_state = NETREG_DUMMY;
11423
11424 /* a dummy interface is started by default */
11425 set_bit(__LINK_STATE_PRESENT, &dev->state);
11426 set_bit(__LINK_STATE_START, &dev->state);
11427
11428 /* Note : We dont allocate pcpu_refcnt for dummy devices,
11429 * because users of this 'device' dont need to change
11430 * its refcount.
11431 */
11432 }
11433
11434 /**
11435 * register_netdev - register a network device
11436 * @dev: device to register
11437 *
11438 * Take a completed network device structure and add it to the kernel
11439 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
11440 * chain. 0 is returned on success. A negative errno code is returned
11441 * on a failure to set up the device, or if the name is a duplicate.
11442 *
11443 * This is a wrapper around register_netdevice that takes the rtnl semaphore
11444 * and expands the device name if you passed a format string to
11445 * alloc_netdev.
11446 */
11447 int register_netdev(struct net_device *dev)
11448 {
11449 struct net *net = dev_net(dev);
11450 int err;
11451
11452 if (rtnl_net_lock_killable(net))
11453 return -EINTR;
11454
11455 err = register_netdevice(dev);
11456
11457 rtnl_net_unlock(net);
11458
11459 return err;
11460 }
11461 EXPORT_SYMBOL(register_netdev);
11462
11463 int netdev_refcnt_read(const struct net_device *dev)
11464 {
11465 #ifdef CONFIG_PCPU_DEV_REFCNT
11466 int i, refcnt = 0;
11467
11468 for_each_possible_cpu(i)
11469 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
11470 return refcnt;
11471 #else
11472 return refcount_read(&dev->dev_refcnt);
11473 #endif
11474 }
11475 EXPORT_SYMBOL(netdev_refcnt_read);
11476
11477 int netdev_unregister_timeout_secs __read_mostly = 10;
11478
11479 #define WAIT_REFS_MIN_MSECS 1
11480 #define WAIT_REFS_MAX_MSECS 250
11481 /**
11482 * netdev_wait_allrefs_any - wait until all references are gone.
11483 * @list: list of net_devices to wait on
11484 *
11485 * This is called when unregistering network devices.
11486 *
11487 * Any protocol or device that holds a reference should register
11488 * for netdevice notification, and cleanup and put back the
11489 * reference if they receive an UNREGISTER event.
11490 * We can get stuck here if buggy protocols don't correctly
11491 * call dev_put.
11492 */
11493 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
11494 {
11495 unsigned long rebroadcast_time, warning_time;
11496 struct net_device *dev;
11497 int wait = 0;
11498
11499 rebroadcast_time = warning_time = jiffies;
11500
11501 list_for_each_entry(dev, list, todo_list)
11502 if (netdev_refcnt_read(dev) == 1)
11503 return dev;
11504
11505 while (true) {
11506 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
11507 rtnl_lock();
11508
11509 /* Rebroadcast unregister notification */
11510 list_for_each_entry(dev, list, todo_list)
11511 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11512
11513 __rtnl_unlock();
11514 rcu_barrier();
11515 rtnl_lock();
11516
11517 list_for_each_entry(dev, list, todo_list)
11518 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
11519 &dev->state)) {
11520 /* We must not have linkwatch events
11521 * pending on unregister. If this
11522 * happens, we simply run the queue
11523 * unscheduled, resulting in a noop
11524 * for this device.
11525 */
11526 linkwatch_run_queue();
11527 break;
11528 }
11529
11530 __rtnl_unlock();
11531
11532 rebroadcast_time = jiffies;
11533 }
11534
11535 rcu_barrier();
11536
11537 if (!wait) {
11538 wait = WAIT_REFS_MIN_MSECS;
11539 } else {
11540 msleep(wait);
11541 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
11542 }
11543
11544 list_for_each_entry(dev, list, todo_list)
11545 if (netdev_refcnt_read(dev) == 1)
11546 return dev;
11547
11548 if (time_after(jiffies, warning_time +
11549 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
11550 list_for_each_entry(dev, list, todo_list) {
11551 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
11552 dev->name, netdev_refcnt_read(dev));
11553 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
11554 }
11555
11556 warning_time = jiffies;
11557 }
11558 }
11559 }
11560
11561 /* The sequence is:
11562 *
11563 * rtnl_lock();
11564 * ...
11565 * register_netdevice(x1);
11566 * register_netdevice(x2);
11567 * ...
11568 * unregister_netdevice(y1);
11569 * unregister_netdevice(y2);
11570 * ...
11571 * rtnl_unlock();
11572 * free_netdev(y1);
11573 * free_netdev(y2);
11574 *
11575 * We are invoked by rtnl_unlock().
11576 * This allows us to deal with problems:
11577 * 1) We can delete sysfs objects which invoke hotplug
11578 * without deadlocking with linkwatch via keventd.
11579 * 2) Since we run with the RTNL semaphore not held, we can sleep
11580 * safely in order to wait for the netdev refcnt to drop to zero.
11581 *
11582 * We must not return until all unregister events added during
11583 * the interval the lock was held have been completed.
11584 */
11585 void netdev_run_todo(void)
11586 {
11587 struct net_device *dev, *tmp;
11588 struct list_head list;
11589 int cnt;
11590 #ifdef CONFIG_LOCKDEP
11591 struct list_head unlink_list;
11592
11593 list_replace_init(&net_unlink_list, &unlink_list);
11594
11595 while (!list_empty(&unlink_list)) {
11596 dev = list_first_entry(&unlink_list, struct net_device,
11597 unlink_list);
11598 list_del_init(&dev->unlink_list);
11599 dev->nested_level = dev->lower_level - 1;
11600 }
11601 #endif
11602
11603 /* Snapshot list, allow later requests */
11604 list_replace_init(&net_todo_list, &list);
11605
11606 __rtnl_unlock();
11607
11608 /* Wait for rcu callbacks to finish before next phase */
11609 if (!list_empty(&list))
11610 rcu_barrier();
11611
11612 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
11613 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
11614 netdev_WARN(dev, "run_todo but not unregistering\n");
11615 list_del(&dev->todo_list);
11616 continue;
11617 }
11618
11619 netdev_lock(dev);
11620 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED);
11621 netdev_unlock(dev);
11622 linkwatch_sync_dev(dev);
11623 }
11624
11625 cnt = 0;
11626 while (!list_empty(&list)) {
11627 dev = netdev_wait_allrefs_any(&list);
11628 list_del(&dev->todo_list);
11629
11630 /* paranoia */
11631 BUG_ON(netdev_refcnt_read(dev) != 1);
11632 BUG_ON(!list_empty(&dev->ptype_all));
11633 BUG_ON(!list_empty(&dev->ptype_specific));
11634 WARN_ON(rcu_access_pointer(dev->ip_ptr));
11635 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
11636
11637 netdev_do_free_pcpu_stats(dev);
11638 if (dev->priv_destructor)
11639 dev->priv_destructor(dev);
11640 if (dev->needs_free_netdev)
11641 free_netdev(dev);
11642
11643 cnt++;
11644
11645 /* Free network device */
11646 kobject_put(&dev->dev.kobj);
11647 }
11648 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count))
11649 wake_up(&netdev_unregistering_wq);
11650 }
11651
11652 /* Collate per-cpu network dstats statistics
11653 *
11654 * Read per-cpu network statistics from dev->dstats and populate the related
11655 * fields in @s.
11656 */
11657 static void dev_fetch_dstats(struct rtnl_link_stats64 *s,
11658 const struct pcpu_dstats __percpu *dstats)
11659 {
11660 int cpu;
11661
11662 for_each_possible_cpu(cpu) {
11663 u64 rx_packets, rx_bytes, rx_drops;
11664 u64 tx_packets, tx_bytes, tx_drops;
11665 const struct pcpu_dstats *stats;
11666 unsigned int start;
11667
11668 stats = per_cpu_ptr(dstats, cpu);
11669 do {
11670 start = u64_stats_fetch_begin(&stats->syncp);
11671 rx_packets = u64_stats_read(&stats->rx_packets);
11672 rx_bytes = u64_stats_read(&stats->rx_bytes);
11673 rx_drops = u64_stats_read(&stats->rx_drops);
11674 tx_packets = u64_stats_read(&stats->tx_packets);
11675 tx_bytes = u64_stats_read(&stats->tx_bytes);
11676 tx_drops = u64_stats_read(&stats->tx_drops);
11677 } while (u64_stats_fetch_retry(&stats->syncp, start));
11678
11679 s->rx_packets += rx_packets;
11680 s->rx_bytes += rx_bytes;
11681 s->rx_dropped += rx_drops;
11682 s->tx_packets += tx_packets;
11683 s->tx_bytes += tx_bytes;
11684 s->tx_dropped += tx_drops;
11685 }
11686 }
11687
11688 /* ndo_get_stats64 implementation for dtstats-based accounting.
11689 *
11690 * Populate @s from dev->stats and dev->dstats. This is used internally by the
11691 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection.
11692 */
11693 static void dev_get_dstats64(const struct net_device *dev,
11694 struct rtnl_link_stats64 *s)
11695 {
11696 netdev_stats_to_stats64(s, &dev->stats);
11697 dev_fetch_dstats(s, dev->dstats);
11698 }
11699
11700 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
11701 * all the same fields in the same order as net_device_stats, with only
11702 * the type differing, but rtnl_link_stats64 may have additional fields
11703 * at the end for newer counters.
11704 */
11705 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
11706 const struct net_device_stats *netdev_stats)
11707 {
11708 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
11709 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
11710 u64 *dst = (u64 *)stats64;
11711
11712 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
11713 for (i = 0; i < n; i++)
11714 dst[i] = (unsigned long)atomic_long_read(&src[i]);
11715 /* zero out counters that only exist in rtnl_link_stats64 */
11716 memset((char *)stats64 + n * sizeof(u64), 0,
11717 sizeof(*stats64) - n * sizeof(u64));
11718 }
11719 EXPORT_SYMBOL(netdev_stats_to_stats64);
11720
11721 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc(
11722 struct net_device *dev)
11723 {
11724 struct net_device_core_stats __percpu *p;
11725
11726 p = alloc_percpu_gfp(struct net_device_core_stats,
11727 GFP_ATOMIC | __GFP_NOWARN);
11728
11729 if (p && cmpxchg(&dev->core_stats, NULL, p))
11730 free_percpu(p);
11731
11732 /* This READ_ONCE() pairs with the cmpxchg() above */
11733 return READ_ONCE(dev->core_stats);
11734 }
11735
11736 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset)
11737 {
11738 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
11739 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
11740 unsigned long __percpu *field;
11741
11742 if (unlikely(!p)) {
11743 p = netdev_core_stats_alloc(dev);
11744 if (!p)
11745 return;
11746 }
11747
11748 field = (unsigned long __percpu *)((void __percpu *)p + offset);
11749 this_cpu_inc(*field);
11750 }
11751 EXPORT_SYMBOL_GPL(netdev_core_stats_inc);
11752
11753 /**
11754 * dev_get_stats - get network device statistics
11755 * @dev: device to get statistics from
11756 * @storage: place to store stats
11757 *
11758 * Get network statistics from device. Return @storage.
11759 * The device driver may provide its own method by setting
11760 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
11761 * otherwise the internal statistics structure is used.
11762 */
11763 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
11764 struct rtnl_link_stats64 *storage)
11765 {
11766 const struct net_device_ops *ops = dev->netdev_ops;
11767 const struct net_device_core_stats __percpu *p;
11768
11769 /*
11770 * IPv{4,6} and udp tunnels share common stat helpers and use
11771 * different stat type (NETDEV_PCPU_STAT_TSTATS vs
11772 * NETDEV_PCPU_STAT_DSTATS). Ensure the accounting is consistent.
11773 */
11774 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, rx_bytes) !=
11775 offsetof(struct pcpu_dstats, rx_bytes));
11776 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, rx_packets) !=
11777 offsetof(struct pcpu_dstats, rx_packets));
11778 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, tx_bytes) !=
11779 offsetof(struct pcpu_dstats, tx_bytes));
11780 BUILD_BUG_ON(offsetof(struct pcpu_sw_netstats, tx_packets) !=
11781 offsetof(struct pcpu_dstats, tx_packets));
11782
11783 if (ops->ndo_get_stats64) {
11784 memset(storage, 0, sizeof(*storage));
11785 ops->ndo_get_stats64(dev, storage);
11786 } else if (ops->ndo_get_stats) {
11787 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
11788 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) {
11789 dev_get_tstats64(dev, storage);
11790 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) {
11791 dev_get_dstats64(dev, storage);
11792 } else {
11793 netdev_stats_to_stats64(storage, &dev->stats);
11794 }
11795
11796 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
11797 p = READ_ONCE(dev->core_stats);
11798 if (p) {
11799 const struct net_device_core_stats *core_stats;
11800 int i;
11801
11802 for_each_possible_cpu(i) {
11803 core_stats = per_cpu_ptr(p, i);
11804 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
11805 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
11806 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
11807 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
11808 }
11809 }
11810 return storage;
11811 }
11812 EXPORT_SYMBOL(dev_get_stats);
11813
11814 /**
11815 * dev_fetch_sw_netstats - get per-cpu network device statistics
11816 * @s: place to store stats
11817 * @netstats: per-cpu network stats to read from
11818 *
11819 * Read per-cpu network statistics and populate the related fields in @s.
11820 */
11821 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
11822 const struct pcpu_sw_netstats __percpu *netstats)
11823 {
11824 int cpu;
11825
11826 for_each_possible_cpu(cpu) {
11827 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
11828 const struct pcpu_sw_netstats *stats;
11829 unsigned int start;
11830
11831 stats = per_cpu_ptr(netstats, cpu);
11832 do {
11833 start = u64_stats_fetch_begin(&stats->syncp);
11834 rx_packets = u64_stats_read(&stats->rx_packets);
11835 rx_bytes = u64_stats_read(&stats->rx_bytes);
11836 tx_packets = u64_stats_read(&stats->tx_packets);
11837 tx_bytes = u64_stats_read(&stats->tx_bytes);
11838 } while (u64_stats_fetch_retry(&stats->syncp, start));
11839
11840 s->rx_packets += rx_packets;
11841 s->rx_bytes += rx_bytes;
11842 s->tx_packets += tx_packets;
11843 s->tx_bytes += tx_bytes;
11844 }
11845 }
11846 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
11847
11848 /**
11849 * dev_get_tstats64 - ndo_get_stats64 implementation
11850 * @dev: device to get statistics from
11851 * @s: place to store stats
11852 *
11853 * Populate @s from dev->stats and dev->tstats. Can be used as
11854 * ndo_get_stats64() callback.
11855 */
11856 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
11857 {
11858 netdev_stats_to_stats64(s, &dev->stats);
11859 dev_fetch_sw_netstats(s, dev->tstats);
11860 }
11861 EXPORT_SYMBOL_GPL(dev_get_tstats64);
11862
11863 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
11864 {
11865 struct netdev_queue *queue = dev_ingress_queue(dev);
11866
11867 #ifdef CONFIG_NET_CLS_ACT
11868 if (queue)
11869 return queue;
11870 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
11871 if (!queue)
11872 return NULL;
11873 netdev_init_one_queue(dev, queue, NULL);
11874 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
11875 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
11876 rcu_assign_pointer(dev->ingress_queue, queue);
11877 #endif
11878 return queue;
11879 }
11880
11881 static const struct ethtool_ops default_ethtool_ops;
11882
11883 void netdev_set_default_ethtool_ops(struct net_device *dev,
11884 const struct ethtool_ops *ops)
11885 {
11886 if (dev->ethtool_ops == &default_ethtool_ops)
11887 dev->ethtool_ops = ops;
11888 }
11889 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
11890
11891 /**
11892 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
11893 * @dev: netdev to enable the IRQ coalescing on
11894 *
11895 * Sets a conservative default for SW IRQ coalescing. Users can use
11896 * sysfs attributes to override the default values.
11897 */
11898 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
11899 {
11900 WARN_ON(dev->reg_state == NETREG_REGISTERED);
11901
11902 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
11903 netdev_set_gro_flush_timeout(dev, 20000);
11904 netdev_set_defer_hard_irqs(dev, 1);
11905 }
11906 }
11907 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
11908
11909 /**
11910 * alloc_netdev_mqs - allocate network device
11911 * @sizeof_priv: size of private data to allocate space for
11912 * @name: device name format string
11913 * @name_assign_type: origin of device name
11914 * @setup: callback to initialize device
11915 * @txqs: the number of TX subqueues to allocate
11916 * @rxqs: the number of RX subqueues to allocate
11917 *
11918 * Allocates a struct net_device with private data area for driver use
11919 * and performs basic initialization. Also allocates subqueue structs
11920 * for each queue on the device.
11921 */
11922 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
11923 unsigned char name_assign_type,
11924 void (*setup)(struct net_device *),
11925 unsigned int txqs, unsigned int rxqs)
11926 {
11927 struct net_device *dev;
11928 size_t napi_config_sz;
11929 unsigned int maxqs;
11930
11931 BUG_ON(strlen(name) >= sizeof(dev->name));
11932
11933 if (txqs < 1) {
11934 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
11935 return NULL;
11936 }
11937
11938 if (rxqs < 1) {
11939 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
11940 return NULL;
11941 }
11942
11943 maxqs = max(txqs, rxqs);
11944
11945 dev = kvzalloc(struct_size(dev, priv, sizeof_priv),
11946 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11947 if (!dev)
11948 return NULL;
11949
11950 dev->priv_len = sizeof_priv;
11951
11952 ref_tracker_dir_init(&dev->refcnt_tracker, 128, "netdev");
11953 #ifdef CONFIG_PCPU_DEV_REFCNT
11954 dev->pcpu_refcnt = alloc_percpu(int);
11955 if (!dev->pcpu_refcnt)
11956 goto free_dev;
11957 __dev_hold(dev);
11958 #else
11959 refcount_set(&dev->dev_refcnt, 1);
11960 #endif
11961
11962 if (dev_addr_init(dev))
11963 goto free_pcpu;
11964
11965 dev_mc_init(dev);
11966 dev_uc_init(dev);
11967
11968 dev_net_set(dev, &init_net);
11969
11970 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
11971 dev->xdp_zc_max_segs = 1;
11972 dev->gso_max_segs = GSO_MAX_SEGS;
11973 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
11974 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
11975 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
11976 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
11977 dev->tso_max_segs = TSO_MAX_SEGS;
11978 dev->upper_level = 1;
11979 dev->lower_level = 1;
11980 #ifdef CONFIG_LOCKDEP
11981 dev->nested_level = 0;
11982 INIT_LIST_HEAD(&dev->unlink_list);
11983 #endif
11984
11985 INIT_LIST_HEAD(&dev->napi_list);
11986 INIT_LIST_HEAD(&dev->unreg_list);
11987 INIT_LIST_HEAD(&dev->close_list);
11988 INIT_LIST_HEAD(&dev->link_watch_list);
11989 INIT_LIST_HEAD(&dev->adj_list.upper);
11990 INIT_LIST_HEAD(&dev->adj_list.lower);
11991 INIT_LIST_HEAD(&dev->ptype_all);
11992 INIT_LIST_HEAD(&dev->ptype_specific);
11993 INIT_LIST_HEAD(&dev->net_notifier_list);
11994 #ifdef CONFIG_NET_SCHED
11995 hash_init(dev->qdisc_hash);
11996 #endif
11997
11998 mutex_init(&dev->lock);
11999
12000 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
12001 setup(dev);
12002
12003 if (!dev->tx_queue_len) {
12004 dev->priv_flags |= IFF_NO_QUEUE;
12005 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
12006 }
12007
12008 dev->num_tx_queues = txqs;
12009 dev->real_num_tx_queues = txqs;
12010 if (netif_alloc_netdev_queues(dev))
12011 goto free_all;
12012
12013 dev->num_rx_queues = rxqs;
12014 dev->real_num_rx_queues = rxqs;
12015 if (netif_alloc_rx_queues(dev))
12016 goto free_all;
12017 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT);
12018 if (!dev->ethtool)
12019 goto free_all;
12020
12021 dev->cfg = kzalloc(sizeof(*dev->cfg), GFP_KERNEL_ACCOUNT);
12022 if (!dev->cfg)
12023 goto free_all;
12024 dev->cfg_pending = dev->cfg;
12025
12026 dev->num_napi_configs = maxqs;
12027 napi_config_sz = array_size(maxqs, sizeof(*dev->napi_config));
12028 dev->napi_config = kvzalloc(napi_config_sz, GFP_KERNEL_ACCOUNT);
12029 if (!dev->napi_config)
12030 goto free_all;
12031
12032 strscpy(dev->name, name);
12033 dev->name_assign_type = name_assign_type;
12034 dev->group = INIT_NETDEV_GROUP;
12035 if (!dev->ethtool_ops)
12036 dev->ethtool_ops = &default_ethtool_ops;
12037
12038 nf_hook_netdev_init(dev);
12039
12040 return dev;
12041
12042 free_all:
12043 free_netdev(dev);
12044 return NULL;
12045
12046 free_pcpu:
12047 #ifdef CONFIG_PCPU_DEV_REFCNT
12048 free_percpu(dev->pcpu_refcnt);
12049 free_dev:
12050 #endif
12051 kvfree(dev);
12052 return NULL;
12053 }
12054 EXPORT_SYMBOL(alloc_netdev_mqs);
12055
12056 static void netdev_napi_exit(struct net_device *dev)
12057 {
12058 if (!list_empty(&dev->napi_list)) {
12059 struct napi_struct *p, *n;
12060
12061 netdev_lock(dev);
12062 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
12063 __netif_napi_del_locked(p);
12064 netdev_unlock(dev);
12065
12066 synchronize_net();
12067 }
12068
12069 kvfree(dev->napi_config);
12070 }
12071
12072 /**
12073 * free_netdev - free network device
12074 * @dev: device
12075 *
12076 * This function does the last stage of destroying an allocated device
12077 * interface. The reference to the device object is released. If this
12078 * is the last reference then it will be freed.Must be called in process
12079 * context.
12080 */
12081 void free_netdev(struct net_device *dev)
12082 {
12083 might_sleep();
12084
12085 /* When called immediately after register_netdevice() failed the unwind
12086 * handling may still be dismantling the device. Handle that case by
12087 * deferring the free.
12088 */
12089 if (dev->reg_state == NETREG_UNREGISTERING) {
12090 ASSERT_RTNL();
12091 dev->needs_free_netdev = true;
12092 return;
12093 }
12094
12095 WARN_ON(dev->cfg != dev->cfg_pending);
12096 kfree(dev->cfg);
12097 kfree(dev->ethtool);
12098 netif_free_tx_queues(dev);
12099 netif_free_rx_queues(dev);
12100
12101 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
12102
12103 /* Flush device addresses */
12104 dev_addr_flush(dev);
12105
12106 netdev_napi_exit(dev);
12107
12108 netif_del_cpu_rmap(dev);
12109
12110 ref_tracker_dir_exit(&dev->refcnt_tracker);
12111 #ifdef CONFIG_PCPU_DEV_REFCNT
12112 free_percpu(dev->pcpu_refcnt);
12113 dev->pcpu_refcnt = NULL;
12114 #endif
12115 free_percpu(dev->core_stats);
12116 dev->core_stats = NULL;
12117 free_percpu(dev->xdp_bulkq);
12118 dev->xdp_bulkq = NULL;
12119
12120 netdev_free_phy_link_topology(dev);
12121
12122 mutex_destroy(&dev->lock);
12123
12124 /* Compatibility with error handling in drivers */
12125 if (dev->reg_state == NETREG_UNINITIALIZED ||
12126 dev->reg_state == NETREG_DUMMY) {
12127 kvfree(dev);
12128 return;
12129 }
12130
12131 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
12132 WRITE_ONCE(dev->reg_state, NETREG_RELEASED);
12133
12134 /* will free via device release */
12135 put_device(&dev->dev);
12136 }
12137 EXPORT_SYMBOL(free_netdev);
12138
12139 /**
12140 * alloc_netdev_dummy - Allocate and initialize a dummy net device.
12141 * @sizeof_priv: size of private data to allocate space for
12142 *
12143 * Return: the allocated net_device on success, NULL otherwise
12144 */
12145 struct net_device *alloc_netdev_dummy(int sizeof_priv)
12146 {
12147 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN,
12148 init_dummy_netdev);
12149 }
12150 EXPORT_SYMBOL_GPL(alloc_netdev_dummy);
12151
12152 /**
12153 * synchronize_net - Synchronize with packet receive processing
12154 *
12155 * Wait for packets currently being received to be done.
12156 * Does not block later packets from starting.
12157 */
12158 void synchronize_net(void)
12159 {
12160 might_sleep();
12161 if (from_cleanup_net() || rtnl_is_locked())
12162 synchronize_rcu_expedited();
12163 else
12164 synchronize_rcu();
12165 }
12166 EXPORT_SYMBOL(synchronize_net);
12167
12168 static void netdev_rss_contexts_free(struct net_device *dev)
12169 {
12170 struct ethtool_rxfh_context *ctx;
12171 unsigned long context;
12172
12173 mutex_lock(&dev->ethtool->rss_lock);
12174 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) {
12175 xa_erase(&dev->ethtool->rss_ctx, context);
12176 dev->ethtool_ops->remove_rxfh_context(dev, ctx, context, NULL);
12177 kfree(ctx);
12178 }
12179 xa_destroy(&dev->ethtool->rss_ctx);
12180 mutex_unlock(&dev->ethtool->rss_lock);
12181 }
12182
12183 /**
12184 * unregister_netdevice_queue - remove device from the kernel
12185 * @dev: device
12186 * @head: list
12187 *
12188 * This function shuts down a device interface and removes it
12189 * from the kernel tables.
12190 * If head not NULL, device is queued to be unregistered later.
12191 *
12192 * Callers must hold the rtnl semaphore. You may want
12193 * unregister_netdev() instead of this.
12194 */
12195
12196 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
12197 {
12198 ASSERT_RTNL();
12199
12200 if (head) {
12201 list_move_tail(&dev->unreg_list, head);
12202 } else {
12203 LIST_HEAD(single);
12204
12205 list_add(&dev->unreg_list, &single);
12206 unregister_netdevice_many(&single);
12207 }
12208 }
12209 EXPORT_SYMBOL(unregister_netdevice_queue);
12210
12211 static void dev_memory_provider_uninstall(struct net_device *dev)
12212 {
12213 unsigned int i;
12214
12215 for (i = 0; i < dev->real_num_rx_queues; i++) {
12216 struct netdev_rx_queue *rxq = &dev->_rx[i];
12217 struct pp_memory_provider_params *p = &rxq->mp_params;
12218
12219 if (p->mp_ops && p->mp_ops->uninstall)
12220 p->mp_ops->uninstall(rxq->mp_params.mp_priv, rxq);
12221 }
12222 }
12223
12224 /* devices must be UP and netdev_lock()'d */
12225 static void netif_close_many_and_unlock(struct list_head *close_head)
12226 {
12227 struct net_device *dev, *tmp;
12228
12229 netif_close_many(close_head, false);
12230
12231 /* ... now unlock them */
12232 list_for_each_entry_safe(dev, tmp, close_head, close_list) {
12233 netdev_unlock(dev);
12234 list_del_init(&dev->close_list);
12235 }
12236 }
12237
12238 static void netif_close_many_and_unlock_cond(struct list_head *close_head)
12239 {
12240 #ifdef CONFIG_LOCKDEP
12241 /* We can only track up to MAX_LOCK_DEPTH locks per task.
12242 *
12243 * Reserve half the available slots for additional locks possibly
12244 * taken by notifiers and (soft)irqs.
12245 */
12246 unsigned int limit = MAX_LOCK_DEPTH / 2;
12247
12248 if (lockdep_depth(current) > limit)
12249 netif_close_many_and_unlock(close_head);
12250 #endif
12251 }
12252
12253 void unregister_netdevice_many_notify(struct list_head *head,
12254 u32 portid, const struct nlmsghdr *nlh)
12255 {
12256 struct net_device *dev, *tmp;
12257 LIST_HEAD(close_head);
12258 int cnt = 0;
12259
12260 BUG_ON(dev_boot_phase);
12261 ASSERT_RTNL();
12262
12263 if (list_empty(head))
12264 return;
12265
12266 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
12267 /* Some devices call without registering
12268 * for initialization unwind. Remove those
12269 * devices and proceed with the remaining.
12270 */
12271 if (dev->reg_state == NETREG_UNINITIALIZED) {
12272 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
12273 dev->name, dev);
12274
12275 WARN_ON(1);
12276 list_del(&dev->unreg_list);
12277 continue;
12278 }
12279 dev->dismantle = true;
12280 BUG_ON(dev->reg_state != NETREG_REGISTERED);
12281 }
12282
12283 /* If device is running, close it first. Start with ops locked... */
12284 list_for_each_entry(dev, head, unreg_list) {
12285 if (!(dev->flags & IFF_UP))
12286 continue;
12287 if (netdev_need_ops_lock(dev)) {
12288 list_add_tail(&dev->close_list, &close_head);
12289 netdev_lock(dev);
12290 }
12291 netif_close_many_and_unlock_cond(&close_head);
12292 }
12293 netif_close_many_and_unlock(&close_head);
12294 /* ... now go over the rest. */
12295 list_for_each_entry(dev, head, unreg_list) {
12296 if (!netdev_need_ops_lock(dev))
12297 list_add_tail(&dev->close_list, &close_head);
12298 }
12299 netif_close_many(&close_head, true);
12300
12301 list_for_each_entry(dev, head, unreg_list) {
12302 /* And unlink it from device chain. */
12303 unlist_netdevice(dev);
12304 netdev_lock(dev);
12305 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING);
12306 netdev_unlock(dev);
12307 }
12308 flush_all_backlogs();
12309
12310 synchronize_net();
12311
12312 list_for_each_entry(dev, head, unreg_list) {
12313 struct sk_buff *skb = NULL;
12314
12315 /* Shutdown queueing discipline. */
12316 netdev_lock_ops(dev);
12317 dev_shutdown(dev);
12318 dev_tcx_uninstall(dev);
12319 dev_xdp_uninstall(dev);
12320 dev_memory_provider_uninstall(dev);
12321 netdev_unlock_ops(dev);
12322 bpf_dev_bound_netdev_unregister(dev);
12323
12324 netdev_offload_xstats_disable_all(dev);
12325
12326 /* Notify protocols, that we are about to destroy
12327 * this device. They should clean all the things.
12328 */
12329 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
12330
12331 if (!(dev->rtnl_link_ops && dev->rtnl_link_initializing))
12332 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
12333 GFP_KERNEL, NULL, 0,
12334 portid, nlh);
12335
12336 /*
12337 * Flush the unicast and multicast chains
12338 */
12339 dev_uc_flush(dev);
12340 dev_mc_flush(dev);
12341
12342 netdev_name_node_alt_flush(dev);
12343 netdev_name_node_free(dev->name_node);
12344
12345 netdev_rss_contexts_free(dev);
12346
12347 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
12348
12349 if (dev->netdev_ops->ndo_uninit)
12350 dev->netdev_ops->ndo_uninit(dev);
12351
12352 mutex_destroy(&dev->ethtool->rss_lock);
12353
12354 net_shaper_flush_netdev(dev);
12355
12356 if (skb)
12357 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
12358
12359 /* Notifier chain MUST detach us all upper devices. */
12360 WARN_ON(netdev_has_any_upper_dev(dev));
12361 WARN_ON(netdev_has_any_lower_dev(dev));
12362
12363 /* Remove entries from kobject tree */
12364 netdev_unregister_kobject(dev);
12365 #ifdef CONFIG_XPS
12366 /* Remove XPS queueing entries */
12367 netif_reset_xps_queues_gt(dev, 0);
12368 #endif
12369 }
12370
12371 synchronize_net();
12372
12373 list_for_each_entry(dev, head, unreg_list) {
12374 netdev_put(dev, &dev->dev_registered_tracker);
12375 net_set_todo(dev);
12376 cnt++;
12377 }
12378 atomic_add(cnt, &dev_unreg_count);
12379
12380 list_del(head);
12381 }
12382
12383 /**
12384 * unregister_netdevice_many - unregister many devices
12385 * @head: list of devices
12386 *
12387 * Note: As most callers use a stack allocated list_head,
12388 * we force a list_del() to make sure stack won't be corrupted later.
12389 */
12390 void unregister_netdevice_many(struct list_head *head)
12391 {
12392 unregister_netdevice_many_notify(head, 0, NULL);
12393 }
12394 EXPORT_SYMBOL(unregister_netdevice_many);
12395
12396 /**
12397 * unregister_netdev - remove device from the kernel
12398 * @dev: device
12399 *
12400 * This function shuts down a device interface and removes it
12401 * from the kernel tables.
12402 *
12403 * This is just a wrapper for unregister_netdevice that takes
12404 * the rtnl semaphore. In general you want to use this and not
12405 * unregister_netdevice.
12406 */
12407 void unregister_netdev(struct net_device *dev)
12408 {
12409 rtnl_net_dev_lock(dev);
12410 unregister_netdevice(dev);
12411 rtnl_net_dev_unlock(dev);
12412 }
12413 EXPORT_SYMBOL(unregister_netdev);
12414
12415 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
12416 const char *pat, int new_ifindex,
12417 struct netlink_ext_ack *extack)
12418 {
12419 struct netdev_name_node *name_node;
12420 struct net *net_old = dev_net(dev);
12421 char new_name[IFNAMSIZ] = {};
12422 int err, new_nsid;
12423
12424 ASSERT_RTNL();
12425
12426 /* Don't allow namespace local devices to be moved. */
12427 err = -EINVAL;
12428 if (dev->netns_immutable) {
12429 NL_SET_ERR_MSG(extack, "The interface netns is immutable");
12430 goto out;
12431 }
12432
12433 /* Ensure the device has been registered */
12434 if (dev->reg_state != NETREG_REGISTERED) {
12435 NL_SET_ERR_MSG(extack, "The interface isn't registered");
12436 goto out;
12437 }
12438
12439 /* Get out if there is nothing todo */
12440 err = 0;
12441 if (net_eq(net_old, net))
12442 goto out;
12443
12444 /* Pick the destination device name, and ensure
12445 * we can use it in the destination network namespace.
12446 */
12447 err = -EEXIST;
12448 if (netdev_name_in_use(net, dev->name)) {
12449 /* We get here if we can't use the current device name */
12450 if (!pat) {
12451 NL_SET_ERR_MSG(extack,
12452 "An interface with the same name exists in the target netns");
12453 goto out;
12454 }
12455 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST);
12456 if (err < 0) {
12457 NL_SET_ERR_MSG_FMT(extack,
12458 "Unable to use '%s' for the new interface name in the target netns",
12459 pat);
12460 goto out;
12461 }
12462 }
12463 /* Check that none of the altnames conflicts. */
12464 err = -EEXIST;
12465 netdev_for_each_altname(dev, name_node) {
12466 if (netdev_name_in_use(net, name_node->name)) {
12467 NL_SET_ERR_MSG_FMT(extack,
12468 "An interface with the altname %s exists in the target netns",
12469 name_node->name);
12470 goto out;
12471 }
12472 }
12473
12474 /* Check that new_ifindex isn't used yet. */
12475 if (new_ifindex) {
12476 err = dev_index_reserve(net, new_ifindex);
12477 if (err < 0) {
12478 NL_SET_ERR_MSG_FMT(extack,
12479 "The ifindex %d is not available in the target netns",
12480 new_ifindex);
12481 goto out;
12482 }
12483 } else {
12484 /* If there is an ifindex conflict assign a new one */
12485 err = dev_index_reserve(net, dev->ifindex);
12486 if (err == -EBUSY)
12487 err = dev_index_reserve(net, 0);
12488 if (err < 0) {
12489 NL_SET_ERR_MSG(extack,
12490 "Unable to allocate a new ifindex in the target netns");
12491 goto out;
12492 }
12493 new_ifindex = err;
12494 }
12495
12496 /*
12497 * And now a mini version of register_netdevice unregister_netdevice.
12498 */
12499
12500 netdev_lock_ops(dev);
12501 /* If device is running close it first. */
12502 netif_close(dev);
12503 /* And unlink it from device chain */
12504 unlist_netdevice(dev);
12505
12506 if (!netdev_need_ops_lock(dev))
12507 netdev_lock(dev);
12508 dev->moving_ns = true;
12509 netdev_unlock(dev);
12510
12511 synchronize_net();
12512
12513 /* Shutdown queueing discipline. */
12514 netdev_lock_ops(dev);
12515 dev_shutdown(dev);
12516 netdev_unlock_ops(dev);
12517
12518 /* Notify protocols, that we are about to destroy
12519 * this device. They should clean all the things.
12520 *
12521 * Note that dev->reg_state stays at NETREG_REGISTERED.
12522 * This is wanted because this way 8021q and macvlan know
12523 * the device is just moving and can keep their slaves up.
12524 */
12525 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
12526 rcu_barrier();
12527
12528 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
12529
12530 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
12531 new_ifindex);
12532
12533 /*
12534 * Flush the unicast and multicast chains
12535 */
12536 dev_uc_flush(dev);
12537 dev_mc_flush(dev);
12538
12539 /* Send a netdev-removed uevent to the old namespace */
12540 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
12541 netdev_adjacent_del_links(dev);
12542
12543 /* Move per-net netdevice notifiers that are following the netdevice */
12544 move_netdevice_notifiers_dev_net(dev, net);
12545
12546 /* Actually switch the network namespace */
12547 netdev_lock(dev);
12548 dev_net_set(dev, net);
12549 netdev_unlock(dev);
12550 dev->ifindex = new_ifindex;
12551
12552 if (new_name[0]) {
12553 /* Rename the netdev to prepared name */
12554 write_seqlock_bh(&netdev_rename_lock);
12555 strscpy(dev->name, new_name, IFNAMSIZ);
12556 write_sequnlock_bh(&netdev_rename_lock);
12557 }
12558
12559 /* Fixup kobjects */
12560 dev_set_uevent_suppress(&dev->dev, 1);
12561 err = device_rename(&dev->dev, dev->name);
12562 dev_set_uevent_suppress(&dev->dev, 0);
12563 WARN_ON(err);
12564
12565 /* Send a netdev-add uevent to the new namespace */
12566 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
12567 netdev_adjacent_add_links(dev);
12568
12569 /* Adapt owner in case owning user namespace of target network
12570 * namespace is different from the original one.
12571 */
12572 err = netdev_change_owner(dev, net_old, net);
12573 WARN_ON(err);
12574
12575 netdev_lock(dev);
12576 dev->moving_ns = false;
12577 if (!netdev_need_ops_lock(dev))
12578 netdev_unlock(dev);
12579
12580 /* Add the device back in the hashes */
12581 list_netdevice(dev);
12582 /* Notify protocols, that a new device appeared. */
12583 call_netdevice_notifiers(NETDEV_REGISTER, dev);
12584 netdev_unlock_ops(dev);
12585
12586 /*
12587 * Prevent userspace races by waiting until the network
12588 * device is fully setup before sending notifications.
12589 */
12590 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
12591
12592 synchronize_net();
12593 err = 0;
12594 out:
12595 return err;
12596 }
12597
12598 static int dev_cpu_dead(unsigned int oldcpu)
12599 {
12600 struct sk_buff **list_skb;
12601 struct sk_buff *skb;
12602 unsigned int cpu;
12603 struct softnet_data *sd, *oldsd, *remsd = NULL;
12604
12605 local_irq_disable();
12606 cpu = smp_processor_id();
12607 sd = &per_cpu(softnet_data, cpu);
12608 oldsd = &per_cpu(softnet_data, oldcpu);
12609
12610 /* Find end of our completion_queue. */
12611 list_skb = &sd->completion_queue;
12612 while (*list_skb)
12613 list_skb = &(*list_skb)->next;
12614 /* Append completion queue from offline CPU. */
12615 *list_skb = oldsd->completion_queue;
12616 oldsd->completion_queue = NULL;
12617
12618 /* Append output queue from offline CPU. */
12619 if (oldsd->output_queue) {
12620 *sd->output_queue_tailp = oldsd->output_queue;
12621 sd->output_queue_tailp = oldsd->output_queue_tailp;
12622 oldsd->output_queue = NULL;
12623 oldsd->output_queue_tailp = &oldsd->output_queue;
12624 }
12625 /* Append NAPI poll list from offline CPU, with one exception :
12626 * process_backlog() must be called by cpu owning percpu backlog.
12627 * We properly handle process_queue & input_pkt_queue later.
12628 */
12629 while (!list_empty(&oldsd->poll_list)) {
12630 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
12631 struct napi_struct,
12632 poll_list);
12633
12634 list_del_init(&napi->poll_list);
12635 if (napi->poll == process_backlog)
12636 napi->state &= NAPIF_STATE_THREADED;
12637 else
12638 ____napi_schedule(sd, napi);
12639 }
12640
12641 raise_softirq_irqoff(NET_TX_SOFTIRQ);
12642 local_irq_enable();
12643
12644 if (!use_backlog_threads()) {
12645 #ifdef CONFIG_RPS
12646 remsd = oldsd->rps_ipi_list;
12647 oldsd->rps_ipi_list = NULL;
12648 #endif
12649 /* send out pending IPI's on offline CPU */
12650 net_rps_send_ipi(remsd);
12651 }
12652
12653 /* Process offline CPU's input_pkt_queue */
12654 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
12655 netif_rx(skb);
12656 rps_input_queue_head_incr(oldsd);
12657 }
12658 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
12659 netif_rx(skb);
12660 rps_input_queue_head_incr(oldsd);
12661 }
12662
12663 return 0;
12664 }
12665
12666 /**
12667 * netdev_increment_features - increment feature set by one
12668 * @all: current feature set
12669 * @one: new feature set
12670 * @mask: mask feature set
12671 *
12672 * Computes a new feature set after adding a device with feature set
12673 * @one to the master device with current feature set @all. Will not
12674 * enable anything that is off in @mask. Returns the new feature set.
12675 */
12676 netdev_features_t netdev_increment_features(netdev_features_t all,
12677 netdev_features_t one, netdev_features_t mask)
12678 {
12679 if (mask & NETIF_F_HW_CSUM)
12680 mask |= NETIF_F_CSUM_MASK;
12681 mask |= NETIF_F_VLAN_CHALLENGED;
12682
12683 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
12684 all &= one | ~NETIF_F_ALL_FOR_ALL;
12685
12686 /* If one device supports hw checksumming, set for all. */
12687 if (all & NETIF_F_HW_CSUM)
12688 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
12689
12690 return all;
12691 }
12692 EXPORT_SYMBOL(netdev_increment_features);
12693
12694 /**
12695 * netdev_compute_master_upper_features - compute feature from lowers
12696 * @dev: the upper device
12697 * @update_header: whether to update upper device's header_len/headroom/tailroom
12698 *
12699 * Recompute the upper device's feature based on all lower devices.
12700 */
12701 void netdev_compute_master_upper_features(struct net_device *dev, bool update_header)
12702 {
12703 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
12704 netdev_features_t gso_partial_features = MASTER_UPPER_DEV_GSO_PARTIAL_FEATURES;
12705 netdev_features_t xfrm_features = MASTER_UPPER_DEV_XFRM_FEATURES;
12706 netdev_features_t mpls_features = MASTER_UPPER_DEV_MPLS_FEATURES;
12707 netdev_features_t vlan_features = MASTER_UPPER_DEV_VLAN_FEATURES;
12708 netdev_features_t enc_features = MASTER_UPPER_DEV_ENC_FEATURES;
12709 unsigned short max_header_len = ETH_HLEN;
12710 unsigned int tso_max_size = TSO_MAX_SIZE;
12711 unsigned short max_headroom = 0;
12712 unsigned short max_tailroom = 0;
12713 u16 tso_max_segs = TSO_MAX_SEGS;
12714 struct net_device *lower_dev;
12715 struct list_head *iter;
12716
12717 mpls_features = netdev_base_features(mpls_features);
12718 vlan_features = netdev_base_features(vlan_features);
12719 enc_features = netdev_base_features(enc_features);
12720
12721 netdev_for_each_lower_dev(dev, lower_dev, iter) {
12722 gso_partial_features = netdev_increment_features(gso_partial_features,
12723 lower_dev->gso_partial_features,
12724 MASTER_UPPER_DEV_GSO_PARTIAL_FEATURES);
12725
12726 vlan_features = netdev_increment_features(vlan_features,
12727 lower_dev->vlan_features,
12728 MASTER_UPPER_DEV_VLAN_FEATURES);
12729
12730 enc_features = netdev_increment_features(enc_features,
12731 lower_dev->hw_enc_features,
12732 MASTER_UPPER_DEV_ENC_FEATURES);
12733
12734 if (IS_ENABLED(CONFIG_XFRM_OFFLOAD))
12735 xfrm_features = netdev_increment_features(xfrm_features,
12736 lower_dev->hw_enc_features,
12737 MASTER_UPPER_DEV_XFRM_FEATURES);
12738
12739 mpls_features = netdev_increment_features(mpls_features,
12740 lower_dev->mpls_features,
12741 MASTER_UPPER_DEV_MPLS_FEATURES);
12742
12743 dst_release_flag &= lower_dev->priv_flags;
12744
12745 if (update_header) {
12746 max_header_len = max(max_header_len, lower_dev->hard_header_len);
12747 max_headroom = max(max_headroom, lower_dev->needed_headroom);
12748 max_tailroom = max(max_tailroom, lower_dev->needed_tailroom);
12749 }
12750
12751 tso_max_size = min(tso_max_size, lower_dev->tso_max_size);
12752 tso_max_segs = min(tso_max_segs, lower_dev->tso_max_segs);
12753 }
12754
12755 dev->gso_partial_features = gso_partial_features;
12756 dev->vlan_features = vlan_features;
12757 dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
12758 NETIF_F_HW_VLAN_CTAG_TX |
12759 NETIF_F_HW_VLAN_STAG_TX;
12760 if (IS_ENABLED(CONFIG_XFRM_OFFLOAD))
12761 dev->hw_enc_features |= xfrm_features;
12762 dev->mpls_features = mpls_features;
12763
12764 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
12765 if ((dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
12766 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
12767 dev->priv_flags |= IFF_XMIT_DST_RELEASE;
12768
12769 if (update_header) {
12770 dev->hard_header_len = max_header_len;
12771 dev->needed_headroom = max_headroom;
12772 dev->needed_tailroom = max_tailroom;
12773 }
12774
12775 netif_set_tso_max_segs(dev, tso_max_segs);
12776 netif_set_tso_max_size(dev, tso_max_size);
12777
12778 netdev_change_features(dev);
12779 }
12780 EXPORT_SYMBOL(netdev_compute_master_upper_features);
12781
12782 static struct hlist_head * __net_init netdev_create_hash(void)
12783 {
12784 int i;
12785 struct hlist_head *hash;
12786
12787 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
12788 if (hash != NULL)
12789 for (i = 0; i < NETDEV_HASHENTRIES; i++)
12790 INIT_HLIST_HEAD(&hash[i]);
12791
12792 return hash;
12793 }
12794
12795 /* Initialize per network namespace state */
12796 static int __net_init netdev_init(struct net *net)
12797 {
12798 BUILD_BUG_ON(GRO_HASH_BUCKETS >
12799 BITS_PER_BYTE * sizeof_field(struct gro_node, bitmask));
12800
12801 INIT_LIST_HEAD(&net->dev_base_head);
12802
12803 net->dev_name_head = netdev_create_hash();
12804 if (net->dev_name_head == NULL)
12805 goto err_name;
12806
12807 net->dev_index_head = netdev_create_hash();
12808 if (net->dev_index_head == NULL)
12809 goto err_idx;
12810
12811 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1);
12812
12813 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
12814
12815 return 0;
12816
12817 err_idx:
12818 kfree(net->dev_name_head);
12819 err_name:
12820 return -ENOMEM;
12821 }
12822
12823 /**
12824 * netdev_drivername - network driver for the device
12825 * @dev: network device
12826 *
12827 * Determine network driver for device.
12828 */
12829 const char *netdev_drivername(const struct net_device *dev)
12830 {
12831 const struct device_driver *driver;
12832 const struct device *parent;
12833 const char *empty = "";
12834
12835 parent = dev->dev.parent;
12836 if (!parent)
12837 return empty;
12838
12839 driver = parent->driver;
12840 if (driver && driver->name)
12841 return driver->name;
12842 return empty;
12843 }
12844
12845 static void __netdev_printk(const char *level, const struct net_device *dev,
12846 struct va_format *vaf)
12847 {
12848 if (dev && dev->dev.parent) {
12849 dev_printk_emit(level[1] - '0',
12850 dev->dev.parent,
12851 "%s %s %s%s: %pV",
12852 dev_driver_string(dev->dev.parent),
12853 dev_name(dev->dev.parent),
12854 netdev_name(dev), netdev_reg_state(dev),
12855 vaf);
12856 } else if (dev) {
12857 printk("%s%s%s: %pV",
12858 level, netdev_name(dev), netdev_reg_state(dev), vaf);
12859 } else {
12860 printk("%s(NULL net_device): %pV", level, vaf);
12861 }
12862 }
12863
12864 void netdev_printk(const char *level, const struct net_device *dev,
12865 const char *format, ...)
12866 {
12867 struct va_format vaf;
12868 va_list args;
12869
12870 va_start(args, format);
12871
12872 vaf.fmt = format;
12873 vaf.va = &args;
12874
12875 __netdev_printk(level, dev, &vaf);
12876
12877 va_end(args);
12878 }
12879 EXPORT_SYMBOL(netdev_printk);
12880
12881 #define define_netdev_printk_level(func, level) \
12882 void func(const struct net_device *dev, const char *fmt, ...) \
12883 { \
12884 struct va_format vaf; \
12885 va_list args; \
12886 \
12887 va_start(args, fmt); \
12888 \
12889 vaf.fmt = fmt; \
12890 vaf.va = &args; \
12891 \
12892 __netdev_printk(level, dev, &vaf); \
12893 \
12894 va_end(args); \
12895 } \
12896 EXPORT_SYMBOL(func);
12897
12898 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
12899 define_netdev_printk_level(netdev_alert, KERN_ALERT);
12900 define_netdev_printk_level(netdev_crit, KERN_CRIT);
12901 define_netdev_printk_level(netdev_err, KERN_ERR);
12902 define_netdev_printk_level(netdev_warn, KERN_WARNING);
12903 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
12904 define_netdev_printk_level(netdev_info, KERN_INFO);
12905
12906 static void __net_exit netdev_exit(struct net *net)
12907 {
12908 kfree(net->dev_name_head);
12909 kfree(net->dev_index_head);
12910 xa_destroy(&net->dev_by_index);
12911 if (net != &init_net)
12912 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
12913 }
12914
12915 static struct pernet_operations __net_initdata netdev_net_ops = {
12916 .init = netdev_init,
12917 .exit = netdev_exit,
12918 };
12919
12920 static void __net_exit default_device_exit_net(struct net *net)
12921 {
12922 struct netdev_name_node *name_node, *tmp;
12923 struct net_device *dev, *aux;
12924 /*
12925 * Push all migratable network devices back to the
12926 * initial network namespace
12927 */
12928 ASSERT_RTNL();
12929 for_each_netdev_safe(net, dev, aux) {
12930 int err;
12931 char fb_name[IFNAMSIZ];
12932
12933 /* Ignore unmoveable devices (i.e. loopback) */
12934 if (dev->netns_immutable)
12935 continue;
12936
12937 /* Leave virtual devices for the generic cleanup */
12938 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
12939 continue;
12940
12941 /* Push remaining network devices to init_net */
12942 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
12943 if (netdev_name_in_use(&init_net, fb_name))
12944 snprintf(fb_name, IFNAMSIZ, "dev%%d");
12945
12946 netdev_for_each_altname_safe(dev, name_node, tmp)
12947 if (netdev_name_in_use(&init_net, name_node->name))
12948 __netdev_name_node_alt_destroy(name_node);
12949
12950 err = dev_change_net_namespace(dev, &init_net, fb_name);
12951 if (err) {
12952 pr_emerg("%s: failed to move %s to init_net: %d\n",
12953 __func__, dev->name, err);
12954 BUG();
12955 }
12956 }
12957 }
12958
12959 static void __net_exit default_device_exit_batch(struct list_head *net_list)
12960 {
12961 /* At exit all network devices most be removed from a network
12962 * namespace. Do this in the reverse order of registration.
12963 * Do this across as many network namespaces as possible to
12964 * improve batching efficiency.
12965 */
12966 struct net_device *dev;
12967 struct net *net;
12968 LIST_HEAD(dev_kill_list);
12969
12970 rtnl_lock();
12971 list_for_each_entry(net, net_list, exit_list) {
12972 default_device_exit_net(net);
12973 cond_resched();
12974 }
12975
12976 list_for_each_entry(net, net_list, exit_list) {
12977 for_each_netdev_reverse(net, dev) {
12978 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
12979 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
12980 else
12981 unregister_netdevice_queue(dev, &dev_kill_list);
12982 }
12983 }
12984 unregister_netdevice_many(&dev_kill_list);
12985 rtnl_unlock();
12986 }
12987
12988 static struct pernet_operations __net_initdata default_device_ops = {
12989 .exit_batch = default_device_exit_batch,
12990 };
12991
12992 static void __init net_dev_struct_check(void)
12993 {
12994 /* TX read-mostly hotpath */
12995 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast);
12996 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops);
12997 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops);
12998 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx);
12999 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues);
13000 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size);
13001 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size);
13002 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs);
13003 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features);
13004 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc);
13005 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu);
13006 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom);
13007 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq);
13008 #ifdef CONFIG_XPS
13009 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps);
13010 #endif
13011 #ifdef CONFIG_NETFILTER_EGRESS
13012 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress);
13013 #endif
13014 #ifdef CONFIG_NET_XGRESS
13015 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress);
13016 #endif
13017 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160);
13018
13019 /* TXRX read-mostly hotpath */
13020 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats);
13021 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state);
13022 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags);
13023 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len);
13024 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features);
13025 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr);
13026 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46);
13027
13028 /* RX read-mostly hotpath */
13029 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific);
13030 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex);
13031 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues);
13032 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx);
13033 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size);
13034 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size);
13035 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler);
13036 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data);
13037 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net);
13038 #ifdef CONFIG_NETPOLL
13039 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo);
13040 #endif
13041 #ifdef CONFIG_NET_XGRESS
13042 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress);
13043 #endif
13044 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 92);
13045 }
13046
13047 /*
13048 * Initialize the DEV module. At boot time this walks the device list and
13049 * unhooks any devices that fail to initialise (normally hardware not
13050 * present) and leaves us with a valid list of present and active devices.
13051 *
13052 */
13053
13054 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
13055 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE)
13056
13057 static int net_page_pool_create(int cpuid)
13058 {
13059 #if IS_ENABLED(CONFIG_PAGE_POOL)
13060 struct page_pool_params page_pool_params = {
13061 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
13062 .flags = PP_FLAG_SYSTEM_POOL,
13063 .nid = cpu_to_mem(cpuid),
13064 };
13065 struct page_pool *pp_ptr;
13066 int err;
13067
13068 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
13069 if (IS_ERR(pp_ptr))
13070 return -ENOMEM;
13071
13072 err = xdp_reg_page_pool(pp_ptr);
13073 if (err) {
13074 page_pool_destroy(pp_ptr);
13075 return err;
13076 }
13077
13078 per_cpu(system_page_pool.pool, cpuid) = pp_ptr;
13079 #endif
13080 return 0;
13081 }
13082
13083 static int backlog_napi_should_run(unsigned int cpu)
13084 {
13085 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
13086 struct napi_struct *napi = &sd->backlog;
13087
13088 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
13089 }
13090
13091 static void run_backlog_napi(unsigned int cpu)
13092 {
13093 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
13094
13095 napi_threaded_poll_loop(&sd->backlog);
13096 }
13097
13098 static void backlog_napi_setup(unsigned int cpu)
13099 {
13100 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
13101 struct napi_struct *napi = &sd->backlog;
13102
13103 napi->thread = this_cpu_read(backlog_napi);
13104 set_bit(NAPI_STATE_THREADED, &napi->state);
13105 }
13106
13107 static struct smp_hotplug_thread backlog_threads = {
13108 .store = &backlog_napi,
13109 .thread_should_run = backlog_napi_should_run,
13110 .thread_fn = run_backlog_napi,
13111 .thread_comm = "backlog_napi/%u",
13112 .setup = backlog_napi_setup,
13113 };
13114
13115 /*
13116 * This is called single threaded during boot, so no need
13117 * to take the rtnl semaphore.
13118 */
13119 static int __init net_dev_init(void)
13120 {
13121 int i, rc = -ENOMEM;
13122
13123 BUG_ON(!dev_boot_phase);
13124
13125 net_dev_struct_check();
13126
13127 if (dev_proc_init())
13128 goto out;
13129
13130 if (netdev_kobject_init())
13131 goto out;
13132
13133 for (i = 0; i < PTYPE_HASH_SIZE; i++)
13134 INIT_LIST_HEAD(&ptype_base[i]);
13135
13136 if (register_pernet_subsys(&netdev_net_ops))
13137 goto out;
13138
13139 /*
13140 * Initialise the packet receive queues.
13141 */
13142
13143 flush_backlogs_fallback = flush_backlogs_alloc();
13144 if (!flush_backlogs_fallback)
13145 goto out;
13146
13147 for_each_possible_cpu(i) {
13148 struct softnet_data *sd = &per_cpu(softnet_data, i);
13149
13150 skb_queue_head_init(&sd->input_pkt_queue);
13151 skb_queue_head_init(&sd->process_queue);
13152 #ifdef CONFIG_XFRM_OFFLOAD
13153 skb_queue_head_init(&sd->xfrm_backlog);
13154 #endif
13155 INIT_LIST_HEAD(&sd->poll_list);
13156 sd->output_queue_tailp = &sd->output_queue;
13157 #ifdef CONFIG_RPS
13158 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
13159 sd->cpu = i;
13160 #endif
13161 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
13162
13163 gro_init(&sd->backlog.gro);
13164 sd->backlog.poll = process_backlog;
13165 sd->backlog.weight = weight_p;
13166 INIT_LIST_HEAD(&sd->backlog.poll_list);
13167
13168 if (net_page_pool_create(i))
13169 goto out;
13170 }
13171 net_hotdata.skb_defer_nodes =
13172 __alloc_percpu(sizeof(struct skb_defer_node) * nr_node_ids,
13173 __alignof__(struct skb_defer_node));
13174 if (!net_hotdata.skb_defer_nodes)
13175 goto out;
13176 if (use_backlog_threads())
13177 smpboot_register_percpu_thread(&backlog_threads);
13178
13179 dev_boot_phase = 0;
13180
13181 /* The loopback device is special if any other network devices
13182 * is present in a network namespace the loopback device must
13183 * be present. Since we now dynamically allocate and free the
13184 * loopback device ensure this invariant is maintained by
13185 * keeping the loopback device as the first device on the
13186 * list of network devices. Ensuring the loopback devices
13187 * is the first device that appears and the last network device
13188 * that disappears.
13189 */
13190 if (register_pernet_device(&loopback_net_ops))
13191 goto out;
13192
13193 if (register_pernet_device(&default_device_ops))
13194 goto out;
13195
13196 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
13197 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
13198
13199 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
13200 NULL, dev_cpu_dead);
13201 WARN_ON(rc < 0);
13202 rc = 0;
13203
13204 /* avoid static key IPIs to isolated CPUs */
13205 if (housekeeping_enabled(HK_TYPE_MISC))
13206 net_enable_timestamp();
13207 out:
13208 if (rc < 0) {
13209 for_each_possible_cpu(i) {
13210 struct page_pool *pp_ptr;
13211
13212 pp_ptr = per_cpu(system_page_pool.pool, i);
13213 if (!pp_ptr)
13214 continue;
13215
13216 xdp_unreg_page_pool(pp_ptr);
13217 page_pool_destroy(pp_ptr);
13218 per_cpu(system_page_pool.pool, i) = NULL;
13219 }
13220 }
13221
13222 return rc;
13223 }
13224
13225 subsys_initcall(net_dev_init);