1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NET3 Protocol independent device support routines.
4 *
5 * Derived from the non IP parts of dev.c 1.0.19
6 * Authors: Ross Biro
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
9 *
10 * Additional Authors:
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
17 *
18 * Changes:
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
30 * drivers
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
40 * call a packet.
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
46 * changes.
47 * Rudi Cilibrasi : Pass the right thing to
48 * set_mac_address()
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
54 * 1 device.
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
62 * the backlog queue.
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
69 */
70
71 #include <linux/uaccess.h>
72 #include <linux/bitmap.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/isolation.h>
81 #include <linux/sched/mm.h>
82 #include <linux/smpboot.h>
83 #include <linux/mutex.h>
84 #include <linux/rwsem.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/skbuff.h>
96 #include <linux/kthread.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
104 #include <net/dsa.h>
105 #include <net/dst.h>
106 #include <net/dst_metadata.h>
107 #include <net/gro.h>
108 #include <net/pkt_sched.h>
109 #include <net/pkt_cls.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <net/tcx.h>
113 #include <linux/highmem.h>
114 #include <linux/init.h>
115 #include <linux/module.h>
116 #include <linux/netpoll.h>
117 #include <linux/rcupdate.h>
118 #include <linux/delay.h>
119 #include <net/iw_handler.h>
120 #include <asm/current.h>
121 #include <linux/audit.h>
122 #include <linux/dmaengine.h>
123 #include <linux/err.h>
124 #include <linux/ctype.h>
125 #include <linux/if_arp.h>
126 #include <linux/if_vlan.h>
127 #include <linux/ip.h>
128 #include <net/ip.h>
129 #include <net/mpls.h>
130 #include <linux/ipv6.h>
131 #include <linux/in.h>
132 #include <linux/jhash.h>
133 #include <linux/random.h>
134 #include <trace/events/napi.h>
135 #include <trace/events/net.h>
136 #include <trace/events/skb.h>
137 #include <trace/events/qdisc.h>
138 #include <trace/events/xdp.h>
139 #include <linux/inetdevice.h>
140 #include <linux/cpu_rmap.h>
141 #include <linux/static_key.h>
142 #include <linux/hashtable.h>
143 #include <linux/vmalloc.h>
144 #include <linux/if_macvlan.h>
145 #include <linux/errqueue.h>
146 #include <linux/hrtimer.h>
147 #include <linux/netfilter_netdev.h>
148 #include <linux/crash_dump.h>
149 #include <linux/sctp.h>
150 #include <net/udp_tunnel.h>
151 #include <linux/net_namespace.h>
152 #include <linux/indirect_call_wrapper.h>
153 #include <net/devlink.h>
154 #include <linux/pm_runtime.h>
155 #include <linux/prandom.h>
156 #include <linux/once_lite.h>
157 #include <net/netdev_rx_queue.h>
158 #include <net/page_pool/types.h>
159 #include <net/page_pool/helpers.h>
160 #include <net/rps.h>
161
162 #include "dev.h"
163 #include "net-sysfs.h"
164
165 static DEFINE_SPINLOCK(ptype_lock);
166 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
167
168 static int netif_rx_internal(struct sk_buff *skb);
169 static int call_netdevice_notifiers_extack(unsigned long val,
170 struct net_device *dev,
171 struct netlink_ext_ack *extack);
172
173 static DEFINE_MUTEX(ifalias_mutex);
174
175 /* protects napi_hash addition/deletion and napi_gen_id */
176 static DEFINE_SPINLOCK(napi_hash_lock);
177
178 static unsigned int napi_gen_id = NR_CPUS;
179 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
180
181 static DECLARE_RWSEM(devnet_rename_sem);
182
dev_base_seq_inc(struct net * net)183 static inline void dev_base_seq_inc(struct net *net)
184 {
185 unsigned int val = net->dev_base_seq + 1;
186
187 WRITE_ONCE(net->dev_base_seq, val ?: 1);
188 }
189
dev_name_hash(struct net * net,const char * name)190 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
191 {
192 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
193
194 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
195 }
196
dev_index_hash(struct net * net,int ifindex)197 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
198 {
199 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
200 }
201
202 #ifndef CONFIG_PREEMPT_RT
203
204 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key);
205
setup_backlog_napi_threads(char * arg)206 static int __init setup_backlog_napi_threads(char *arg)
207 {
208 static_branch_enable(&use_backlog_threads_key);
209 return 0;
210 }
211 early_param("thread_backlog_napi", setup_backlog_napi_threads);
212
use_backlog_threads(void)213 static bool use_backlog_threads(void)
214 {
215 return static_branch_unlikely(&use_backlog_threads_key);
216 }
217
218 #else
219
use_backlog_threads(void)220 static bool use_backlog_threads(void)
221 {
222 return true;
223 }
224
225 #endif
226
backlog_lock_irq_save(struct softnet_data * sd,unsigned long * flags)227 static inline void backlog_lock_irq_save(struct softnet_data *sd,
228 unsigned long *flags)
229 {
230 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
231 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
232 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
233 local_irq_save(*flags);
234 }
235
backlog_lock_irq_disable(struct softnet_data * sd)236 static inline void backlog_lock_irq_disable(struct softnet_data *sd)
237 {
238 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
239 spin_lock_irq(&sd->input_pkt_queue.lock);
240 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
241 local_irq_disable();
242 }
243
backlog_unlock_irq_restore(struct softnet_data * sd,unsigned long * flags)244 static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
245 unsigned long *flags)
246 {
247 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
248 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
249 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
250 local_irq_restore(*flags);
251 }
252
backlog_unlock_irq_enable(struct softnet_data * sd)253 static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
254 {
255 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
256 spin_unlock_irq(&sd->input_pkt_queue.lock);
257 else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
258 local_irq_enable();
259 }
260
netdev_name_node_alloc(struct net_device * dev,const char * name)261 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
262 const char *name)
263 {
264 struct netdev_name_node *name_node;
265
266 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
267 if (!name_node)
268 return NULL;
269 INIT_HLIST_NODE(&name_node->hlist);
270 name_node->dev = dev;
271 name_node->name = name;
272 return name_node;
273 }
274
275 static struct netdev_name_node *
netdev_name_node_head_alloc(struct net_device * dev)276 netdev_name_node_head_alloc(struct net_device *dev)
277 {
278 struct netdev_name_node *name_node;
279
280 name_node = netdev_name_node_alloc(dev, dev->name);
281 if (!name_node)
282 return NULL;
283 INIT_LIST_HEAD(&name_node->list);
284 return name_node;
285 }
286
netdev_name_node_free(struct netdev_name_node * name_node)287 static void netdev_name_node_free(struct netdev_name_node *name_node)
288 {
289 kfree(name_node);
290 }
291
netdev_name_node_add(struct net * net,struct netdev_name_node * name_node)292 static void netdev_name_node_add(struct net *net,
293 struct netdev_name_node *name_node)
294 {
295 hlist_add_head_rcu(&name_node->hlist,
296 dev_name_hash(net, name_node->name));
297 }
298
netdev_name_node_del(struct netdev_name_node * name_node)299 static void netdev_name_node_del(struct netdev_name_node *name_node)
300 {
301 hlist_del_rcu(&name_node->hlist);
302 }
303
netdev_name_node_lookup(struct net * net,const char * name)304 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
305 const char *name)
306 {
307 struct hlist_head *head = dev_name_hash(net, name);
308 struct netdev_name_node *name_node;
309
310 hlist_for_each_entry(name_node, head, hlist)
311 if (!strcmp(name_node->name, name))
312 return name_node;
313 return NULL;
314 }
315
netdev_name_node_lookup_rcu(struct net * net,const char * name)316 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
317 const char *name)
318 {
319 struct hlist_head *head = dev_name_hash(net, name);
320 struct netdev_name_node *name_node;
321
322 hlist_for_each_entry_rcu(name_node, head, hlist)
323 if (!strcmp(name_node->name, name))
324 return name_node;
325 return NULL;
326 }
327
netdev_name_in_use(struct net * net,const char * name)328 bool netdev_name_in_use(struct net *net, const char *name)
329 {
330 return netdev_name_node_lookup(net, name);
331 }
332 EXPORT_SYMBOL(netdev_name_in_use);
333
netdev_name_node_alt_create(struct net_device * dev,const char * name)334 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
335 {
336 struct netdev_name_node *name_node;
337 struct net *net = dev_net(dev);
338
339 name_node = netdev_name_node_lookup(net, name);
340 if (name_node)
341 return -EEXIST;
342 name_node = netdev_name_node_alloc(dev, name);
343 if (!name_node)
344 return -ENOMEM;
345 netdev_name_node_add(net, name_node);
346 /* The node that holds dev->name acts as a head of per-device list. */
347 list_add_tail_rcu(&name_node->list, &dev->name_node->list);
348
349 return 0;
350 }
351
netdev_name_node_alt_free(struct rcu_head * head)352 static void netdev_name_node_alt_free(struct rcu_head *head)
353 {
354 struct netdev_name_node *name_node =
355 container_of(head, struct netdev_name_node, rcu);
356
357 kfree(name_node->name);
358 netdev_name_node_free(name_node);
359 }
360
__netdev_name_node_alt_destroy(struct netdev_name_node * name_node)361 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
362 {
363 netdev_name_node_del(name_node);
364 list_del(&name_node->list);
365 call_rcu(&name_node->rcu, netdev_name_node_alt_free);
366 }
367
netdev_name_node_alt_destroy(struct net_device * dev,const char * name)368 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
369 {
370 struct netdev_name_node *name_node;
371 struct net *net = dev_net(dev);
372
373 name_node = netdev_name_node_lookup(net, name);
374 if (!name_node)
375 return -ENOENT;
376 /* lookup might have found our primary name or a name belonging
377 * to another device.
378 */
379 if (name_node == dev->name_node || name_node->dev != dev)
380 return -EINVAL;
381
382 __netdev_name_node_alt_destroy(name_node);
383 return 0;
384 }
385
netdev_name_node_alt_flush(struct net_device * dev)386 static void netdev_name_node_alt_flush(struct net_device *dev)
387 {
388 struct netdev_name_node *name_node, *tmp;
389
390 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) {
391 list_del(&name_node->list);
392 netdev_name_node_alt_free(&name_node->rcu);
393 }
394 }
395
396 /* Device list insertion */
list_netdevice(struct net_device * dev)397 static void list_netdevice(struct net_device *dev)
398 {
399 struct netdev_name_node *name_node;
400 struct net *net = dev_net(dev);
401
402 ASSERT_RTNL();
403
404 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
405 netdev_name_node_add(net, dev->name_node);
406 hlist_add_head_rcu(&dev->index_hlist,
407 dev_index_hash(net, dev->ifindex));
408
409 netdev_for_each_altname(dev, name_node)
410 netdev_name_node_add(net, name_node);
411
412 /* We reserved the ifindex, this can't fail */
413 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL));
414
415 dev_base_seq_inc(net);
416 }
417
418 /* Device list removal
419 * caller must respect a RCU grace period before freeing/reusing dev
420 */
unlist_netdevice(struct net_device * dev)421 static void unlist_netdevice(struct net_device *dev)
422 {
423 struct netdev_name_node *name_node;
424 struct net *net = dev_net(dev);
425
426 ASSERT_RTNL();
427
428 xa_erase(&net->dev_by_index, dev->ifindex);
429
430 netdev_for_each_altname(dev, name_node)
431 netdev_name_node_del(name_node);
432
433 /* Unlink dev from the device chain */
434 list_del_rcu(&dev->dev_list);
435 netdev_name_node_del(dev->name_node);
436 hlist_del_rcu(&dev->index_hlist);
437
438 dev_base_seq_inc(dev_net(dev));
439 }
440
441 /*
442 * Our notifier list
443 */
444
445 static RAW_NOTIFIER_HEAD(netdev_chain);
446
447 /*
448 * Device drivers call our routines to queue packets here. We empty the
449 * queue in the local softnet handler.
450 */
451
452 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
453 EXPORT_PER_CPU_SYMBOL(softnet_data);
454
455 /* Page_pool has a lockless array/stack to alloc/recycle pages.
456 * PP consumers must pay attention to run APIs in the appropriate context
457 * (e.g. NAPI context).
458 */
459 static DEFINE_PER_CPU(struct page_pool *, system_page_pool);
460
461 #ifdef CONFIG_LOCKDEP
462 /*
463 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
464 * according to dev->type
465 */
466 static const unsigned short netdev_lock_type[] = {
467 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
468 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
469 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
470 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
471 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
472 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
473 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
474 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
475 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
476 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
477 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
478 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
479 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
480 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
481 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
482
483 static const char *const netdev_lock_name[] = {
484 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
485 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
486 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
487 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
488 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
489 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
490 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
491 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
492 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
493 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
494 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
495 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
496 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
497 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
498 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
499
500 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
501 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
502
netdev_lock_pos(unsigned short dev_type)503 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
504 {
505 int i;
506
507 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
508 if (netdev_lock_type[i] == dev_type)
509 return i;
510 /* the last key is used by default */
511 return ARRAY_SIZE(netdev_lock_type) - 1;
512 }
513
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)514 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
515 unsigned short dev_type)
516 {
517 int i;
518
519 i = netdev_lock_pos(dev_type);
520 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
521 netdev_lock_name[i]);
522 }
523
netdev_set_addr_lockdep_class(struct net_device * dev)524 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
525 {
526 int i;
527
528 i = netdev_lock_pos(dev->type);
529 lockdep_set_class_and_name(&dev->addr_list_lock,
530 &netdev_addr_lock_key[i],
531 netdev_lock_name[i]);
532 }
533 #else
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)534 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
535 unsigned short dev_type)
536 {
537 }
538
netdev_set_addr_lockdep_class(struct net_device * dev)539 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
540 {
541 }
542 #endif
543
544 /*******************************************************************************
545 *
546 * Protocol management and registration routines
547 *
548 *******************************************************************************/
549
550
551 /*
552 * Add a protocol ID to the list. Now that the input handler is
553 * smarter we can dispense with all the messy stuff that used to be
554 * here.
555 *
556 * BEWARE!!! Protocol handlers, mangling input packets,
557 * MUST BE last in hash buckets and checking protocol handlers
558 * MUST start from promiscuous ptype_all chain in net_bh.
559 * It is true now, do not change it.
560 * Explanation follows: if protocol handler, mangling packet, will
561 * be the first on list, it is not able to sense, that packet
562 * is cloned and should be copied-on-write, so that it will
563 * change it and subsequent readers will get broken packet.
564 * --ANK (980803)
565 */
566
ptype_head(const struct packet_type * pt)567 static inline struct list_head *ptype_head(const struct packet_type *pt)
568 {
569 if (pt->type == htons(ETH_P_ALL))
570 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all;
571 else
572 return pt->dev ? &pt->dev->ptype_specific :
573 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
574 }
575
576 /**
577 * dev_add_pack - add packet handler
578 * @pt: packet type declaration
579 *
580 * Add a protocol handler to the networking stack. The passed &packet_type
581 * is linked into kernel lists and may not be freed until it has been
582 * removed from the kernel lists.
583 *
584 * This call does not sleep therefore it can not
585 * guarantee all CPU's that are in middle of receiving packets
586 * will see the new packet type (until the next received packet).
587 */
588
dev_add_pack(struct packet_type * pt)589 void dev_add_pack(struct packet_type *pt)
590 {
591 struct list_head *head = ptype_head(pt);
592
593 spin_lock(&ptype_lock);
594 list_add_rcu(&pt->list, head);
595 spin_unlock(&ptype_lock);
596 }
597 EXPORT_SYMBOL(dev_add_pack);
598
599 /**
600 * __dev_remove_pack - remove packet handler
601 * @pt: packet type declaration
602 *
603 * Remove a protocol handler that was previously added to the kernel
604 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
605 * from the kernel lists and can be freed or reused once this function
606 * returns.
607 *
608 * The packet type might still be in use by receivers
609 * and must not be freed until after all the CPU's have gone
610 * through a quiescent state.
611 */
__dev_remove_pack(struct packet_type * pt)612 void __dev_remove_pack(struct packet_type *pt)
613 {
614 struct list_head *head = ptype_head(pt);
615 struct packet_type *pt1;
616
617 spin_lock(&ptype_lock);
618
619 list_for_each_entry(pt1, head, list) {
620 if (pt == pt1) {
621 list_del_rcu(&pt->list);
622 goto out;
623 }
624 }
625
626 pr_warn("dev_remove_pack: %p not found\n", pt);
627 out:
628 spin_unlock(&ptype_lock);
629 }
630 EXPORT_SYMBOL(__dev_remove_pack);
631
632 /**
633 * dev_remove_pack - remove packet handler
634 * @pt: packet type declaration
635 *
636 * Remove a protocol handler that was previously added to the kernel
637 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
638 * from the kernel lists and can be freed or reused once this function
639 * returns.
640 *
641 * This call sleeps to guarantee that no CPU is looking at the packet
642 * type after return.
643 */
dev_remove_pack(struct packet_type * pt)644 void dev_remove_pack(struct packet_type *pt)
645 {
646 __dev_remove_pack(pt);
647
648 synchronize_net();
649 }
650 EXPORT_SYMBOL(dev_remove_pack);
651
652
653 /*******************************************************************************
654 *
655 * Device Interface Subroutines
656 *
657 *******************************************************************************/
658
659 /**
660 * dev_get_iflink - get 'iflink' value of a interface
661 * @dev: targeted interface
662 *
663 * Indicates the ifindex the interface is linked to.
664 * Physical interfaces have the same 'ifindex' and 'iflink' values.
665 */
666
dev_get_iflink(const struct net_device * dev)667 int dev_get_iflink(const struct net_device *dev)
668 {
669 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
670 return dev->netdev_ops->ndo_get_iflink(dev);
671
672 return READ_ONCE(dev->ifindex);
673 }
674 EXPORT_SYMBOL(dev_get_iflink);
675
676 /**
677 * dev_fill_metadata_dst - Retrieve tunnel egress information.
678 * @dev: targeted interface
679 * @skb: The packet.
680 *
681 * For better visibility of tunnel traffic OVS needs to retrieve
682 * egress tunnel information for a packet. Following API allows
683 * user to get this info.
684 */
dev_fill_metadata_dst(struct net_device * dev,struct sk_buff * skb)685 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
686 {
687 struct ip_tunnel_info *info;
688
689 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
690 return -EINVAL;
691
692 info = skb_tunnel_info_unclone(skb);
693 if (!info)
694 return -ENOMEM;
695 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
696 return -EINVAL;
697
698 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
699 }
700 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
701
dev_fwd_path(struct net_device_path_stack * stack)702 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
703 {
704 int k = stack->num_paths++;
705
706 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
707 return NULL;
708
709 return &stack->path[k];
710 }
711
dev_fill_forward_path(const struct net_device * dev,const u8 * daddr,struct net_device_path_stack * stack)712 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
713 struct net_device_path_stack *stack)
714 {
715 const struct net_device *last_dev;
716 struct net_device_path_ctx ctx = {
717 .dev = dev,
718 };
719 struct net_device_path *path;
720 int ret = 0;
721
722 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
723 stack->num_paths = 0;
724 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
725 last_dev = ctx.dev;
726 path = dev_fwd_path(stack);
727 if (!path)
728 return -1;
729
730 memset(path, 0, sizeof(struct net_device_path));
731 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
732 if (ret < 0)
733 return -1;
734
735 if (WARN_ON_ONCE(last_dev == ctx.dev))
736 return -1;
737 }
738
739 if (!ctx.dev)
740 return ret;
741
742 path = dev_fwd_path(stack);
743 if (!path)
744 return -1;
745 path->type = DEV_PATH_ETHERNET;
746 path->dev = ctx.dev;
747
748 return ret;
749 }
750 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
751
752 /**
753 * __dev_get_by_name - find a device by its name
754 * @net: the applicable net namespace
755 * @name: name to find
756 *
757 * Find an interface by name. Must be called under RTNL semaphore.
758 * If the name is found a pointer to the device is returned.
759 * If the name is not found then %NULL is returned. The
760 * reference counters are not incremented so the caller must be
761 * careful with locks.
762 */
763
__dev_get_by_name(struct net * net,const char * name)764 struct net_device *__dev_get_by_name(struct net *net, const char *name)
765 {
766 struct netdev_name_node *node_name;
767
768 node_name = netdev_name_node_lookup(net, name);
769 return node_name ? node_name->dev : NULL;
770 }
771 EXPORT_SYMBOL(__dev_get_by_name);
772
773 /**
774 * dev_get_by_name_rcu - find a device by its name
775 * @net: the applicable net namespace
776 * @name: name to find
777 *
778 * Find an interface by name.
779 * If the name is found a pointer to the device is returned.
780 * If the name is not found then %NULL is returned.
781 * The reference counters are not incremented so the caller must be
782 * careful with locks. The caller must hold RCU lock.
783 */
784
dev_get_by_name_rcu(struct net * net,const char * name)785 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
786 {
787 struct netdev_name_node *node_name;
788
789 node_name = netdev_name_node_lookup_rcu(net, name);
790 return node_name ? node_name->dev : NULL;
791 }
792 EXPORT_SYMBOL(dev_get_by_name_rcu);
793
794 /* Deprecated for new users, call netdev_get_by_name() instead */
dev_get_by_name(struct net * net,const char * name)795 struct net_device *dev_get_by_name(struct net *net, const char *name)
796 {
797 struct net_device *dev;
798
799 rcu_read_lock();
800 dev = dev_get_by_name_rcu(net, name);
801 dev_hold(dev);
802 rcu_read_unlock();
803 return dev;
804 }
805 EXPORT_SYMBOL(dev_get_by_name);
806
807 /**
808 * netdev_get_by_name() - find a device by its name
809 * @net: the applicable net namespace
810 * @name: name to find
811 * @tracker: tracking object for the acquired reference
812 * @gfp: allocation flags for the tracker
813 *
814 * Find an interface by name. This can be called from any
815 * context and does its own locking. The returned handle has
816 * the usage count incremented and the caller must use netdev_put() to
817 * release it when it is no longer needed. %NULL is returned if no
818 * matching device is found.
819 */
netdev_get_by_name(struct net * net,const char * name,netdevice_tracker * tracker,gfp_t gfp)820 struct net_device *netdev_get_by_name(struct net *net, const char *name,
821 netdevice_tracker *tracker, gfp_t gfp)
822 {
823 struct net_device *dev;
824
825 dev = dev_get_by_name(net, name);
826 if (dev)
827 netdev_tracker_alloc(dev, tracker, gfp);
828 return dev;
829 }
830 EXPORT_SYMBOL(netdev_get_by_name);
831
832 /**
833 * __dev_get_by_index - find a device by its ifindex
834 * @net: the applicable net namespace
835 * @ifindex: index of device
836 *
837 * Search for an interface by index. Returns %NULL if the device
838 * is not found or a pointer to the device. The device has not
839 * had its reference counter increased so the caller must be careful
840 * about locking. The caller must hold the RTNL semaphore.
841 */
842
__dev_get_by_index(struct net * net,int ifindex)843 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
844 {
845 struct net_device *dev;
846 struct hlist_head *head = dev_index_hash(net, ifindex);
847
848 hlist_for_each_entry(dev, head, index_hlist)
849 if (dev->ifindex == ifindex)
850 return dev;
851
852 return NULL;
853 }
854 EXPORT_SYMBOL(__dev_get_by_index);
855
856 /**
857 * dev_get_by_index_rcu - find a device by its ifindex
858 * @net: the applicable net namespace
859 * @ifindex: index of device
860 *
861 * Search for an interface by index. Returns %NULL if the device
862 * is not found or a pointer to the device. The device has not
863 * had its reference counter increased so the caller must be careful
864 * about locking. The caller must hold RCU lock.
865 */
866
dev_get_by_index_rcu(struct net * net,int ifindex)867 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
868 {
869 struct net_device *dev;
870 struct hlist_head *head = dev_index_hash(net, ifindex);
871
872 hlist_for_each_entry_rcu(dev, head, index_hlist)
873 if (dev->ifindex == ifindex)
874 return dev;
875
876 return NULL;
877 }
878 EXPORT_SYMBOL(dev_get_by_index_rcu);
879
880 /* Deprecated for new users, call netdev_get_by_index() instead */
dev_get_by_index(struct net * net,int ifindex)881 struct net_device *dev_get_by_index(struct net *net, int ifindex)
882 {
883 struct net_device *dev;
884
885 rcu_read_lock();
886 dev = dev_get_by_index_rcu(net, ifindex);
887 dev_hold(dev);
888 rcu_read_unlock();
889 return dev;
890 }
891 EXPORT_SYMBOL(dev_get_by_index);
892
893 /**
894 * netdev_get_by_index() - find a device by its ifindex
895 * @net: the applicable net namespace
896 * @ifindex: index of device
897 * @tracker: tracking object for the acquired reference
898 * @gfp: allocation flags for the tracker
899 *
900 * Search for an interface by index. Returns NULL if the device
901 * is not found or a pointer to the device. The device returned has
902 * had a reference added and the pointer is safe until the user calls
903 * netdev_put() to indicate they have finished with it.
904 */
netdev_get_by_index(struct net * net,int ifindex,netdevice_tracker * tracker,gfp_t gfp)905 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
906 netdevice_tracker *tracker, gfp_t gfp)
907 {
908 struct net_device *dev;
909
910 dev = dev_get_by_index(net, ifindex);
911 if (dev)
912 netdev_tracker_alloc(dev, tracker, gfp);
913 return dev;
914 }
915 EXPORT_SYMBOL(netdev_get_by_index);
916
917 /**
918 * dev_get_by_napi_id - find a device by napi_id
919 * @napi_id: ID of the NAPI struct
920 *
921 * Search for an interface by NAPI ID. Returns %NULL if the device
922 * is not found or a pointer to the device. The device has not had
923 * its reference counter increased so the caller must be careful
924 * about locking. The caller must hold RCU lock.
925 */
926
dev_get_by_napi_id(unsigned int napi_id)927 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
928 {
929 struct napi_struct *napi;
930
931 WARN_ON_ONCE(!rcu_read_lock_held());
932
933 if (napi_id < MIN_NAPI_ID)
934 return NULL;
935
936 napi = napi_by_id(napi_id);
937
938 return napi ? napi->dev : NULL;
939 }
940 EXPORT_SYMBOL(dev_get_by_napi_id);
941
942 static DEFINE_SEQLOCK(netdev_rename_lock);
943
netdev_copy_name(struct net_device * dev,char * name)944 void netdev_copy_name(struct net_device *dev, char *name)
945 {
946 unsigned int seq;
947
948 do {
949 seq = read_seqbegin(&netdev_rename_lock);
950 strscpy(name, dev->name, IFNAMSIZ);
951 } while (read_seqretry(&netdev_rename_lock, seq));
952 }
953
954 /**
955 * netdev_get_name - get a netdevice name, knowing its ifindex.
956 * @net: network namespace
957 * @name: a pointer to the buffer where the name will be stored.
958 * @ifindex: the ifindex of the interface to get the name from.
959 */
netdev_get_name(struct net * net,char * name,int ifindex)960 int netdev_get_name(struct net *net, char *name, int ifindex)
961 {
962 struct net_device *dev;
963 int ret;
964
965 rcu_read_lock();
966
967 dev = dev_get_by_index_rcu(net, ifindex);
968 if (!dev) {
969 ret = -ENODEV;
970 goto out;
971 }
972
973 netdev_copy_name(dev, name);
974
975 ret = 0;
976 out:
977 rcu_read_unlock();
978 return ret;
979 }
980
981 /**
982 * dev_getbyhwaddr_rcu - find a device by its hardware address
983 * @net: the applicable net namespace
984 * @type: media type of device
985 * @ha: hardware address
986 *
987 * Search for an interface by MAC address. Returns NULL if the device
988 * is not found or a pointer to the device.
989 * The caller must hold RCU or RTNL.
990 * The returned device has not had its ref count increased
991 * and the caller must therefore be careful about locking
992 *
993 */
994
dev_getbyhwaddr_rcu(struct net * net,unsigned short type,const char * ha)995 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
996 const char *ha)
997 {
998 struct net_device *dev;
999
1000 for_each_netdev_rcu(net, dev)
1001 if (dev->type == type &&
1002 !memcmp(dev->dev_addr, ha, dev->addr_len))
1003 return dev;
1004
1005 return NULL;
1006 }
1007 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1008
dev_getfirstbyhwtype(struct net * net,unsigned short type)1009 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1010 {
1011 struct net_device *dev, *ret = NULL;
1012
1013 rcu_read_lock();
1014 for_each_netdev_rcu(net, dev)
1015 if (dev->type == type) {
1016 dev_hold(dev);
1017 ret = dev;
1018 break;
1019 }
1020 rcu_read_unlock();
1021 return ret;
1022 }
1023 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1024
1025 /**
1026 * __dev_get_by_flags - find any device with given flags
1027 * @net: the applicable net namespace
1028 * @if_flags: IFF_* values
1029 * @mask: bitmask of bits in if_flags to check
1030 *
1031 * Search for any interface with the given flags. Returns NULL if a device
1032 * is not found or a pointer to the device. Must be called inside
1033 * rtnl_lock(), and result refcount is unchanged.
1034 */
1035
__dev_get_by_flags(struct net * net,unsigned short if_flags,unsigned short mask)1036 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1037 unsigned short mask)
1038 {
1039 struct net_device *dev, *ret;
1040
1041 ASSERT_RTNL();
1042
1043 ret = NULL;
1044 for_each_netdev(net, dev) {
1045 if (((dev->flags ^ if_flags) & mask) == 0) {
1046 ret = dev;
1047 break;
1048 }
1049 }
1050 return ret;
1051 }
1052 EXPORT_SYMBOL(__dev_get_by_flags);
1053
1054 /**
1055 * dev_valid_name - check if name is okay for network device
1056 * @name: name string
1057 *
1058 * Network device names need to be valid file names to
1059 * allow sysfs to work. We also disallow any kind of
1060 * whitespace.
1061 */
dev_valid_name(const char * name)1062 bool dev_valid_name(const char *name)
1063 {
1064 if (*name == '\0')
1065 return false;
1066 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1067 return false;
1068 if (!strcmp(name, ".") || !strcmp(name, ".."))
1069 return false;
1070
1071 while (*name) {
1072 if (*name == '/' || *name == ':' || isspace(*name))
1073 return false;
1074 name++;
1075 }
1076 return true;
1077 }
1078 EXPORT_SYMBOL(dev_valid_name);
1079
1080 /**
1081 * __dev_alloc_name - allocate a name for a device
1082 * @net: network namespace to allocate the device name in
1083 * @name: name format string
1084 * @res: result name string
1085 *
1086 * Passed a format string - eg "lt%d" it will try and find a suitable
1087 * id. It scans list of devices to build up a free map, then chooses
1088 * the first empty slot. The caller must hold the dev_base or rtnl lock
1089 * while allocating the name and adding the device in order to avoid
1090 * duplicates.
1091 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1092 * Returns the number of the unit assigned or a negative errno code.
1093 */
1094
__dev_alloc_name(struct net * net,const char * name,char * res)1095 static int __dev_alloc_name(struct net *net, const char *name, char *res)
1096 {
1097 int i = 0;
1098 const char *p;
1099 const int max_netdevices = 8*PAGE_SIZE;
1100 unsigned long *inuse;
1101 struct net_device *d;
1102 char buf[IFNAMSIZ];
1103
1104 /* Verify the string as this thing may have come from the user.
1105 * There must be one "%d" and no other "%" characters.
1106 */
1107 p = strchr(name, '%');
1108 if (!p || p[1] != 'd' || strchr(p + 2, '%'))
1109 return -EINVAL;
1110
1111 /* Use one page as a bit array of possible slots */
1112 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC);
1113 if (!inuse)
1114 return -ENOMEM;
1115
1116 for_each_netdev(net, d) {
1117 struct netdev_name_node *name_node;
1118
1119 netdev_for_each_altname(d, name_node) {
1120 if (!sscanf(name_node->name, name, &i))
1121 continue;
1122 if (i < 0 || i >= max_netdevices)
1123 continue;
1124
1125 /* avoid cases where sscanf is not exact inverse of printf */
1126 snprintf(buf, IFNAMSIZ, name, i);
1127 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1128 __set_bit(i, inuse);
1129 }
1130 if (!sscanf(d->name, name, &i))
1131 continue;
1132 if (i < 0 || i >= max_netdevices)
1133 continue;
1134
1135 /* avoid cases where sscanf is not exact inverse of printf */
1136 snprintf(buf, IFNAMSIZ, name, i);
1137 if (!strncmp(buf, d->name, IFNAMSIZ))
1138 __set_bit(i, inuse);
1139 }
1140
1141 i = find_first_zero_bit(inuse, max_netdevices);
1142 bitmap_free(inuse);
1143 if (i == max_netdevices)
1144 return -ENFILE;
1145
1146 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */
1147 strscpy(buf, name, IFNAMSIZ);
1148 snprintf(res, IFNAMSIZ, buf, i);
1149 return i;
1150 }
1151
1152 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */
dev_prep_valid_name(struct net * net,struct net_device * dev,const char * want_name,char * out_name,int dup_errno)1153 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1154 const char *want_name, char *out_name,
1155 int dup_errno)
1156 {
1157 if (!dev_valid_name(want_name))
1158 return -EINVAL;
1159
1160 if (strchr(want_name, '%'))
1161 return __dev_alloc_name(net, want_name, out_name);
1162
1163 if (netdev_name_in_use(net, want_name))
1164 return -dup_errno;
1165 if (out_name != want_name)
1166 strscpy(out_name, want_name, IFNAMSIZ);
1167 return 0;
1168 }
1169
1170 /**
1171 * dev_alloc_name - allocate a name for a device
1172 * @dev: device
1173 * @name: name format string
1174 *
1175 * Passed a format string - eg "lt%d" it will try and find a suitable
1176 * id. It scans list of devices to build up a free map, then chooses
1177 * the first empty slot. The caller must hold the dev_base or rtnl lock
1178 * while allocating the name and adding the device in order to avoid
1179 * duplicates.
1180 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1181 * Returns the number of the unit assigned or a negative errno code.
1182 */
1183
dev_alloc_name(struct net_device * dev,const char * name)1184 int dev_alloc_name(struct net_device *dev, const char *name)
1185 {
1186 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE);
1187 }
1188 EXPORT_SYMBOL(dev_alloc_name);
1189
dev_get_valid_name(struct net * net,struct net_device * dev,const char * name)1190 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1191 const char *name)
1192 {
1193 int ret;
1194
1195 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST);
1196 return ret < 0 ? ret : 0;
1197 }
1198
1199 /**
1200 * dev_change_name - change name of a device
1201 * @dev: device
1202 * @newname: name (or format string) must be at least IFNAMSIZ
1203 *
1204 * Change name of a device, can pass format strings "eth%d".
1205 * for wildcarding.
1206 */
dev_change_name(struct net_device * dev,const char * newname)1207 int dev_change_name(struct net_device *dev, const char *newname)
1208 {
1209 unsigned char old_assign_type;
1210 char oldname[IFNAMSIZ];
1211 int err = 0;
1212 int ret;
1213 struct net *net;
1214
1215 ASSERT_RTNL();
1216 BUG_ON(!dev_net(dev));
1217
1218 net = dev_net(dev);
1219
1220 down_write(&devnet_rename_sem);
1221
1222 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1223 up_write(&devnet_rename_sem);
1224 return 0;
1225 }
1226
1227 memcpy(oldname, dev->name, IFNAMSIZ);
1228
1229 write_seqlock(&netdev_rename_lock);
1230 err = dev_get_valid_name(net, dev, newname);
1231 write_sequnlock(&netdev_rename_lock);
1232
1233 if (err < 0) {
1234 up_write(&devnet_rename_sem);
1235 return err;
1236 }
1237
1238 if (oldname[0] && !strchr(oldname, '%'))
1239 netdev_info(dev, "renamed from %s%s\n", oldname,
1240 dev->flags & IFF_UP ? " (while UP)" : "");
1241
1242 old_assign_type = dev->name_assign_type;
1243 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED);
1244
1245 rollback:
1246 ret = device_rename(&dev->dev, dev->name);
1247 if (ret) {
1248 memcpy(dev->name, oldname, IFNAMSIZ);
1249 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1250 up_write(&devnet_rename_sem);
1251 return ret;
1252 }
1253
1254 up_write(&devnet_rename_sem);
1255
1256 netdev_adjacent_rename_links(dev, oldname);
1257
1258 netdev_name_node_del(dev->name_node);
1259
1260 synchronize_net();
1261
1262 netdev_name_node_add(net, dev->name_node);
1263
1264 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1265 ret = notifier_to_errno(ret);
1266
1267 if (ret) {
1268 /* err >= 0 after dev_alloc_name() or stores the first errno */
1269 if (err >= 0) {
1270 err = ret;
1271 down_write(&devnet_rename_sem);
1272 write_seqlock(&netdev_rename_lock);
1273 memcpy(dev->name, oldname, IFNAMSIZ);
1274 write_sequnlock(&netdev_rename_lock);
1275 memcpy(oldname, newname, IFNAMSIZ);
1276 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1277 old_assign_type = NET_NAME_RENAMED;
1278 goto rollback;
1279 } else {
1280 netdev_err(dev, "name change rollback failed: %d\n",
1281 ret);
1282 }
1283 }
1284
1285 return err;
1286 }
1287
1288 /**
1289 * dev_set_alias - change ifalias of a device
1290 * @dev: device
1291 * @alias: name up to IFALIASZ
1292 * @len: limit of bytes to copy from info
1293 *
1294 * Set ifalias for a device,
1295 */
dev_set_alias(struct net_device * dev,const char * alias,size_t len)1296 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1297 {
1298 struct dev_ifalias *new_alias = NULL;
1299
1300 if (len >= IFALIASZ)
1301 return -EINVAL;
1302
1303 if (len) {
1304 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1305 if (!new_alias)
1306 return -ENOMEM;
1307
1308 memcpy(new_alias->ifalias, alias, len);
1309 new_alias->ifalias[len] = 0;
1310 }
1311
1312 mutex_lock(&ifalias_mutex);
1313 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1314 mutex_is_locked(&ifalias_mutex));
1315 mutex_unlock(&ifalias_mutex);
1316
1317 if (new_alias)
1318 kfree_rcu(new_alias, rcuhead);
1319
1320 return len;
1321 }
1322 EXPORT_SYMBOL(dev_set_alias);
1323
1324 /**
1325 * dev_get_alias - get ifalias of a device
1326 * @dev: device
1327 * @name: buffer to store name of ifalias
1328 * @len: size of buffer
1329 *
1330 * get ifalias for a device. Caller must make sure dev cannot go
1331 * away, e.g. rcu read lock or own a reference count to device.
1332 */
dev_get_alias(const struct net_device * dev,char * name,size_t len)1333 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1334 {
1335 const struct dev_ifalias *alias;
1336 int ret = 0;
1337
1338 rcu_read_lock();
1339 alias = rcu_dereference(dev->ifalias);
1340 if (alias)
1341 ret = snprintf(name, len, "%s", alias->ifalias);
1342 rcu_read_unlock();
1343
1344 return ret;
1345 }
1346
1347 /**
1348 * netdev_features_change - device changes features
1349 * @dev: device to cause notification
1350 *
1351 * Called to indicate a device has changed features.
1352 */
netdev_features_change(struct net_device * dev)1353 void netdev_features_change(struct net_device *dev)
1354 {
1355 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1356 }
1357 EXPORT_SYMBOL(netdev_features_change);
1358
1359 /**
1360 * netdev_state_change - device changes state
1361 * @dev: device to cause notification
1362 *
1363 * Called to indicate a device has changed state. This function calls
1364 * the notifier chains for netdev_chain and sends a NEWLINK message
1365 * to the routing socket.
1366 */
netdev_state_change(struct net_device * dev)1367 void netdev_state_change(struct net_device *dev)
1368 {
1369 if (dev->flags & IFF_UP) {
1370 struct netdev_notifier_change_info change_info = {
1371 .info.dev = dev,
1372 };
1373
1374 call_netdevice_notifiers_info(NETDEV_CHANGE,
1375 &change_info.info);
1376 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1377 }
1378 }
1379 EXPORT_SYMBOL(netdev_state_change);
1380
1381 /**
1382 * __netdev_notify_peers - notify network peers about existence of @dev,
1383 * to be called when rtnl lock is already held.
1384 * @dev: network device
1385 *
1386 * Generate traffic such that interested network peers are aware of
1387 * @dev, such as by generating a gratuitous ARP. This may be used when
1388 * a device wants to inform the rest of the network about some sort of
1389 * reconfiguration such as a failover event or virtual machine
1390 * migration.
1391 */
__netdev_notify_peers(struct net_device * dev)1392 void __netdev_notify_peers(struct net_device *dev)
1393 {
1394 ASSERT_RTNL();
1395 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1396 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1397 }
1398 EXPORT_SYMBOL(__netdev_notify_peers);
1399
1400 /**
1401 * netdev_notify_peers - notify network peers about existence of @dev
1402 * @dev: network device
1403 *
1404 * Generate traffic such that interested network peers are aware of
1405 * @dev, such as by generating a gratuitous ARP. This may be used when
1406 * a device wants to inform the rest of the network about some sort of
1407 * reconfiguration such as a failover event or virtual machine
1408 * migration.
1409 */
netdev_notify_peers(struct net_device * dev)1410 void netdev_notify_peers(struct net_device *dev)
1411 {
1412 rtnl_lock();
1413 __netdev_notify_peers(dev);
1414 rtnl_unlock();
1415 }
1416 EXPORT_SYMBOL(netdev_notify_peers);
1417
1418 static int napi_threaded_poll(void *data);
1419
napi_kthread_create(struct napi_struct * n)1420 static int napi_kthread_create(struct napi_struct *n)
1421 {
1422 int err = 0;
1423
1424 /* Create and wake up the kthread once to put it in
1425 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1426 * warning and work with loadavg.
1427 */
1428 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1429 n->dev->name, n->napi_id);
1430 if (IS_ERR(n->thread)) {
1431 err = PTR_ERR(n->thread);
1432 pr_err("kthread_run failed with err %d\n", err);
1433 n->thread = NULL;
1434 }
1435
1436 return err;
1437 }
1438
__dev_open(struct net_device * dev,struct netlink_ext_ack * extack)1439 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1440 {
1441 const struct net_device_ops *ops = dev->netdev_ops;
1442 int ret;
1443
1444 ASSERT_RTNL();
1445 dev_addr_check(dev);
1446
1447 if (!netif_device_present(dev)) {
1448 /* may be detached because parent is runtime-suspended */
1449 if (dev->dev.parent)
1450 pm_runtime_resume(dev->dev.parent);
1451 if (!netif_device_present(dev))
1452 return -ENODEV;
1453 }
1454
1455 /* Block netpoll from trying to do any rx path servicing.
1456 * If we don't do this there is a chance ndo_poll_controller
1457 * or ndo_poll may be running while we open the device
1458 */
1459 netpoll_poll_disable(dev);
1460
1461 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1462 ret = notifier_to_errno(ret);
1463 if (ret)
1464 return ret;
1465
1466 set_bit(__LINK_STATE_START, &dev->state);
1467
1468 if (ops->ndo_validate_addr)
1469 ret = ops->ndo_validate_addr(dev);
1470
1471 if (!ret && ops->ndo_open)
1472 ret = ops->ndo_open(dev);
1473
1474 netpoll_poll_enable(dev);
1475
1476 if (ret)
1477 clear_bit(__LINK_STATE_START, &dev->state);
1478 else {
1479 dev->flags |= IFF_UP;
1480 dev_set_rx_mode(dev);
1481 dev_activate(dev);
1482 add_device_randomness(dev->dev_addr, dev->addr_len);
1483 }
1484
1485 return ret;
1486 }
1487
1488 /**
1489 * dev_open - prepare an interface for use.
1490 * @dev: device to open
1491 * @extack: netlink extended ack
1492 *
1493 * Takes a device from down to up state. The device's private open
1494 * function is invoked and then the multicast lists are loaded. Finally
1495 * the device is moved into the up state and a %NETDEV_UP message is
1496 * sent to the netdev notifier chain.
1497 *
1498 * Calling this function on an active interface is a nop. On a failure
1499 * a negative errno code is returned.
1500 */
dev_open(struct net_device * dev,struct netlink_ext_ack * extack)1501 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1502 {
1503 int ret;
1504
1505 if (dev->flags & IFF_UP)
1506 return 0;
1507
1508 ret = __dev_open(dev, extack);
1509 if (ret < 0)
1510 return ret;
1511
1512 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1513 call_netdevice_notifiers(NETDEV_UP, dev);
1514
1515 return ret;
1516 }
1517 EXPORT_SYMBOL(dev_open);
1518
__dev_close_many(struct list_head * head)1519 static void __dev_close_many(struct list_head *head)
1520 {
1521 struct net_device *dev;
1522
1523 ASSERT_RTNL();
1524 might_sleep();
1525
1526 list_for_each_entry(dev, head, close_list) {
1527 /* Temporarily disable netpoll until the interface is down */
1528 netpoll_poll_disable(dev);
1529
1530 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1531
1532 clear_bit(__LINK_STATE_START, &dev->state);
1533
1534 /* Synchronize to scheduled poll. We cannot touch poll list, it
1535 * can be even on different cpu. So just clear netif_running().
1536 *
1537 * dev->stop() will invoke napi_disable() on all of it's
1538 * napi_struct instances on this device.
1539 */
1540 smp_mb__after_atomic(); /* Commit netif_running(). */
1541 }
1542
1543 dev_deactivate_many(head);
1544
1545 list_for_each_entry(dev, head, close_list) {
1546 const struct net_device_ops *ops = dev->netdev_ops;
1547
1548 /*
1549 * Call the device specific close. This cannot fail.
1550 * Only if device is UP
1551 *
1552 * We allow it to be called even after a DETACH hot-plug
1553 * event.
1554 */
1555 if (ops->ndo_stop)
1556 ops->ndo_stop(dev);
1557
1558 dev->flags &= ~IFF_UP;
1559 netpoll_poll_enable(dev);
1560 }
1561 }
1562
__dev_close(struct net_device * dev)1563 static void __dev_close(struct net_device *dev)
1564 {
1565 LIST_HEAD(single);
1566
1567 list_add(&dev->close_list, &single);
1568 __dev_close_many(&single);
1569 list_del(&single);
1570 }
1571
dev_close_many(struct list_head * head,bool unlink)1572 void dev_close_many(struct list_head *head, bool unlink)
1573 {
1574 struct net_device *dev, *tmp;
1575
1576 /* Remove the devices that don't need to be closed */
1577 list_for_each_entry_safe(dev, tmp, head, close_list)
1578 if (!(dev->flags & IFF_UP))
1579 list_del_init(&dev->close_list);
1580
1581 __dev_close_many(head);
1582
1583 list_for_each_entry_safe(dev, tmp, head, close_list) {
1584 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1585 call_netdevice_notifiers(NETDEV_DOWN, dev);
1586 if (unlink)
1587 list_del_init(&dev->close_list);
1588 }
1589 }
1590 EXPORT_SYMBOL(dev_close_many);
1591
1592 /**
1593 * dev_close - shutdown an interface.
1594 * @dev: device to shutdown
1595 *
1596 * This function moves an active device into down state. A
1597 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1598 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1599 * chain.
1600 */
dev_close(struct net_device * dev)1601 void dev_close(struct net_device *dev)
1602 {
1603 if (dev->flags & IFF_UP) {
1604 LIST_HEAD(single);
1605
1606 list_add(&dev->close_list, &single);
1607 dev_close_many(&single, true);
1608 list_del(&single);
1609 }
1610 }
1611 EXPORT_SYMBOL(dev_close);
1612
1613
1614 /**
1615 * dev_disable_lro - disable Large Receive Offload on a device
1616 * @dev: device
1617 *
1618 * Disable Large Receive Offload (LRO) on a net device. Must be
1619 * called under RTNL. This is needed if received packets may be
1620 * forwarded to another interface.
1621 */
dev_disable_lro(struct net_device * dev)1622 void dev_disable_lro(struct net_device *dev)
1623 {
1624 struct net_device *lower_dev;
1625 struct list_head *iter;
1626
1627 dev->wanted_features &= ~NETIF_F_LRO;
1628 netdev_update_features(dev);
1629
1630 if (unlikely(dev->features & NETIF_F_LRO))
1631 netdev_WARN(dev, "failed to disable LRO!\n");
1632
1633 netdev_for_each_lower_dev(dev, lower_dev, iter)
1634 dev_disable_lro(lower_dev);
1635 }
1636 EXPORT_SYMBOL(dev_disable_lro);
1637
1638 /**
1639 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1640 * @dev: device
1641 *
1642 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1643 * called under RTNL. This is needed if Generic XDP is installed on
1644 * the device.
1645 */
dev_disable_gro_hw(struct net_device * dev)1646 static void dev_disable_gro_hw(struct net_device *dev)
1647 {
1648 dev->wanted_features &= ~NETIF_F_GRO_HW;
1649 netdev_update_features(dev);
1650
1651 if (unlikely(dev->features & NETIF_F_GRO_HW))
1652 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1653 }
1654
netdev_cmd_to_name(enum netdev_cmd cmd)1655 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1656 {
1657 #define N(val) \
1658 case NETDEV_##val: \
1659 return "NETDEV_" __stringify(val);
1660 switch (cmd) {
1661 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1662 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1663 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1664 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1665 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1666 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1667 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1668 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1669 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1670 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1671 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1672 N(XDP_FEAT_CHANGE)
1673 }
1674 #undef N
1675 return "UNKNOWN_NETDEV_EVENT";
1676 }
1677 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1678
call_netdevice_notifier(struct notifier_block * nb,unsigned long val,struct net_device * dev)1679 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1680 struct net_device *dev)
1681 {
1682 struct netdev_notifier_info info = {
1683 .dev = dev,
1684 };
1685
1686 return nb->notifier_call(nb, val, &info);
1687 }
1688
call_netdevice_register_notifiers(struct notifier_block * nb,struct net_device * dev)1689 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1690 struct net_device *dev)
1691 {
1692 int err;
1693
1694 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1695 err = notifier_to_errno(err);
1696 if (err)
1697 return err;
1698
1699 if (!(dev->flags & IFF_UP))
1700 return 0;
1701
1702 call_netdevice_notifier(nb, NETDEV_UP, dev);
1703 return 0;
1704 }
1705
call_netdevice_unregister_notifiers(struct notifier_block * nb,struct net_device * dev)1706 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1707 struct net_device *dev)
1708 {
1709 if (dev->flags & IFF_UP) {
1710 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1711 dev);
1712 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1713 }
1714 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1715 }
1716
call_netdevice_register_net_notifiers(struct notifier_block * nb,struct net * net)1717 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1718 struct net *net)
1719 {
1720 struct net_device *dev;
1721 int err;
1722
1723 for_each_netdev(net, dev) {
1724 err = call_netdevice_register_notifiers(nb, dev);
1725 if (err)
1726 goto rollback;
1727 }
1728 return 0;
1729
1730 rollback:
1731 for_each_netdev_continue_reverse(net, dev)
1732 call_netdevice_unregister_notifiers(nb, dev);
1733 return err;
1734 }
1735
call_netdevice_unregister_net_notifiers(struct notifier_block * nb,struct net * net)1736 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1737 struct net *net)
1738 {
1739 struct net_device *dev;
1740
1741 for_each_netdev(net, dev)
1742 call_netdevice_unregister_notifiers(nb, dev);
1743 }
1744
1745 static int dev_boot_phase = 1;
1746
1747 /**
1748 * register_netdevice_notifier - register a network notifier block
1749 * @nb: notifier
1750 *
1751 * Register a notifier to be called when network device events occur.
1752 * The notifier passed is linked into the kernel structures and must
1753 * not be reused until it has been unregistered. A negative errno code
1754 * is returned on a failure.
1755 *
1756 * When registered all registration and up events are replayed
1757 * to the new notifier to allow device to have a race free
1758 * view of the network device list.
1759 */
1760
register_netdevice_notifier(struct notifier_block * nb)1761 int register_netdevice_notifier(struct notifier_block *nb)
1762 {
1763 struct net *net;
1764 int err;
1765
1766 /* Close race with setup_net() and cleanup_net() */
1767 down_write(&pernet_ops_rwsem);
1768 rtnl_lock();
1769 err = raw_notifier_chain_register(&netdev_chain, nb);
1770 if (err)
1771 goto unlock;
1772 if (dev_boot_phase)
1773 goto unlock;
1774 for_each_net(net) {
1775 err = call_netdevice_register_net_notifiers(nb, net);
1776 if (err)
1777 goto rollback;
1778 }
1779
1780 unlock:
1781 rtnl_unlock();
1782 up_write(&pernet_ops_rwsem);
1783 return err;
1784
1785 rollback:
1786 for_each_net_continue_reverse(net)
1787 call_netdevice_unregister_net_notifiers(nb, net);
1788
1789 raw_notifier_chain_unregister(&netdev_chain, nb);
1790 goto unlock;
1791 }
1792 EXPORT_SYMBOL(register_netdevice_notifier);
1793
1794 /**
1795 * unregister_netdevice_notifier - unregister a network notifier block
1796 * @nb: notifier
1797 *
1798 * Unregister a notifier previously registered by
1799 * register_netdevice_notifier(). The notifier is unlinked into the
1800 * kernel structures and may then be reused. A negative errno code
1801 * is returned on a failure.
1802 *
1803 * After unregistering unregister and down device events are synthesized
1804 * for all devices on the device list to the removed notifier to remove
1805 * the need for special case cleanup code.
1806 */
1807
unregister_netdevice_notifier(struct notifier_block * nb)1808 int unregister_netdevice_notifier(struct notifier_block *nb)
1809 {
1810 struct net *net;
1811 int err;
1812
1813 /* Close race with setup_net() and cleanup_net() */
1814 down_write(&pernet_ops_rwsem);
1815 rtnl_lock();
1816 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1817 if (err)
1818 goto unlock;
1819
1820 for_each_net(net)
1821 call_netdevice_unregister_net_notifiers(nb, net);
1822
1823 unlock:
1824 rtnl_unlock();
1825 up_write(&pernet_ops_rwsem);
1826 return err;
1827 }
1828 EXPORT_SYMBOL(unregister_netdevice_notifier);
1829
__register_netdevice_notifier_net(struct net * net,struct notifier_block * nb,bool ignore_call_fail)1830 static int __register_netdevice_notifier_net(struct net *net,
1831 struct notifier_block *nb,
1832 bool ignore_call_fail)
1833 {
1834 int err;
1835
1836 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1837 if (err)
1838 return err;
1839 if (dev_boot_phase)
1840 return 0;
1841
1842 err = call_netdevice_register_net_notifiers(nb, net);
1843 if (err && !ignore_call_fail)
1844 goto chain_unregister;
1845
1846 return 0;
1847
1848 chain_unregister:
1849 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1850 return err;
1851 }
1852
__unregister_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1853 static int __unregister_netdevice_notifier_net(struct net *net,
1854 struct notifier_block *nb)
1855 {
1856 int err;
1857
1858 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1859 if (err)
1860 return err;
1861
1862 call_netdevice_unregister_net_notifiers(nb, net);
1863 return 0;
1864 }
1865
1866 /**
1867 * register_netdevice_notifier_net - register a per-netns network notifier block
1868 * @net: network namespace
1869 * @nb: notifier
1870 *
1871 * Register a notifier to be called when network device events occur.
1872 * The notifier passed is linked into the kernel structures and must
1873 * not be reused until it has been unregistered. A negative errno code
1874 * is returned on a failure.
1875 *
1876 * When registered all registration and up events are replayed
1877 * to the new notifier to allow device to have a race free
1878 * view of the network device list.
1879 */
1880
register_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1881 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1882 {
1883 int err;
1884
1885 rtnl_lock();
1886 err = __register_netdevice_notifier_net(net, nb, false);
1887 rtnl_unlock();
1888 return err;
1889 }
1890 EXPORT_SYMBOL(register_netdevice_notifier_net);
1891
1892 /**
1893 * unregister_netdevice_notifier_net - unregister a per-netns
1894 * network notifier block
1895 * @net: network namespace
1896 * @nb: notifier
1897 *
1898 * Unregister a notifier previously registered by
1899 * register_netdevice_notifier_net(). The notifier is unlinked from the
1900 * kernel structures and may then be reused. A negative errno code
1901 * is returned on a failure.
1902 *
1903 * After unregistering unregister and down device events are synthesized
1904 * for all devices on the device list to the removed notifier to remove
1905 * the need for special case cleanup code.
1906 */
1907
unregister_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1908 int unregister_netdevice_notifier_net(struct net *net,
1909 struct notifier_block *nb)
1910 {
1911 int err;
1912
1913 rtnl_lock();
1914 err = __unregister_netdevice_notifier_net(net, nb);
1915 rtnl_unlock();
1916 return err;
1917 }
1918 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1919
__move_netdevice_notifier_net(struct net * src_net,struct net * dst_net,struct notifier_block * nb)1920 static void __move_netdevice_notifier_net(struct net *src_net,
1921 struct net *dst_net,
1922 struct notifier_block *nb)
1923 {
1924 __unregister_netdevice_notifier_net(src_net, nb);
1925 __register_netdevice_notifier_net(dst_net, nb, true);
1926 }
1927
register_netdevice_notifier_dev_net(struct net_device * dev,struct notifier_block * nb,struct netdev_net_notifier * nn)1928 int register_netdevice_notifier_dev_net(struct net_device *dev,
1929 struct notifier_block *nb,
1930 struct netdev_net_notifier *nn)
1931 {
1932 int err;
1933
1934 rtnl_lock();
1935 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1936 if (!err) {
1937 nn->nb = nb;
1938 list_add(&nn->list, &dev->net_notifier_list);
1939 }
1940 rtnl_unlock();
1941 return err;
1942 }
1943 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1944
unregister_netdevice_notifier_dev_net(struct net_device * dev,struct notifier_block * nb,struct netdev_net_notifier * nn)1945 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1946 struct notifier_block *nb,
1947 struct netdev_net_notifier *nn)
1948 {
1949 int err;
1950
1951 rtnl_lock();
1952 list_del(&nn->list);
1953 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1954 rtnl_unlock();
1955 return err;
1956 }
1957 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1958
move_netdevice_notifiers_dev_net(struct net_device * dev,struct net * net)1959 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1960 struct net *net)
1961 {
1962 struct netdev_net_notifier *nn;
1963
1964 list_for_each_entry(nn, &dev->net_notifier_list, list)
1965 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1966 }
1967
1968 /**
1969 * call_netdevice_notifiers_info - call all network notifier blocks
1970 * @val: value passed unmodified to notifier function
1971 * @info: notifier information data
1972 *
1973 * Call all network notifier blocks. Parameters and return value
1974 * are as for raw_notifier_call_chain().
1975 */
1976
call_netdevice_notifiers_info(unsigned long val,struct netdev_notifier_info * info)1977 int call_netdevice_notifiers_info(unsigned long val,
1978 struct netdev_notifier_info *info)
1979 {
1980 struct net *net = dev_net(info->dev);
1981 int ret;
1982
1983 ASSERT_RTNL();
1984
1985 /* Run per-netns notifier block chain first, then run the global one.
1986 * Hopefully, one day, the global one is going to be removed after
1987 * all notifier block registrators get converted to be per-netns.
1988 */
1989 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1990 if (ret & NOTIFY_STOP_MASK)
1991 return ret;
1992 return raw_notifier_call_chain(&netdev_chain, val, info);
1993 }
1994
1995 /**
1996 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1997 * for and rollback on error
1998 * @val_up: value passed unmodified to notifier function
1999 * @val_down: value passed unmodified to the notifier function when
2000 * recovering from an error on @val_up
2001 * @info: notifier information data
2002 *
2003 * Call all per-netns network notifier blocks, but not notifier blocks on
2004 * the global notifier chain. Parameters and return value are as for
2005 * raw_notifier_call_chain_robust().
2006 */
2007
2008 static int
call_netdevice_notifiers_info_robust(unsigned long val_up,unsigned long val_down,struct netdev_notifier_info * info)2009 call_netdevice_notifiers_info_robust(unsigned long val_up,
2010 unsigned long val_down,
2011 struct netdev_notifier_info *info)
2012 {
2013 struct net *net = dev_net(info->dev);
2014
2015 ASSERT_RTNL();
2016
2017 return raw_notifier_call_chain_robust(&net->netdev_chain,
2018 val_up, val_down, info);
2019 }
2020
call_netdevice_notifiers_extack(unsigned long val,struct net_device * dev,struct netlink_ext_ack * extack)2021 static int call_netdevice_notifiers_extack(unsigned long val,
2022 struct net_device *dev,
2023 struct netlink_ext_ack *extack)
2024 {
2025 struct netdev_notifier_info info = {
2026 .dev = dev,
2027 .extack = extack,
2028 };
2029
2030 return call_netdevice_notifiers_info(val, &info);
2031 }
2032
2033 /**
2034 * call_netdevice_notifiers - call all network notifier blocks
2035 * @val: value passed unmodified to notifier function
2036 * @dev: net_device pointer passed unmodified to notifier function
2037 *
2038 * Call all network notifier blocks. Parameters and return value
2039 * are as for raw_notifier_call_chain().
2040 */
2041
call_netdevice_notifiers(unsigned long val,struct net_device * dev)2042 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2043 {
2044 return call_netdevice_notifiers_extack(val, dev, NULL);
2045 }
2046 EXPORT_SYMBOL(call_netdevice_notifiers);
2047
2048 /**
2049 * call_netdevice_notifiers_mtu - call all network notifier blocks
2050 * @val: value passed unmodified to notifier function
2051 * @dev: net_device pointer passed unmodified to notifier function
2052 * @arg: additional u32 argument passed to the notifier function
2053 *
2054 * Call all network notifier blocks. Parameters and return value
2055 * are as for raw_notifier_call_chain().
2056 */
call_netdevice_notifiers_mtu(unsigned long val,struct net_device * dev,u32 arg)2057 static int call_netdevice_notifiers_mtu(unsigned long val,
2058 struct net_device *dev, u32 arg)
2059 {
2060 struct netdev_notifier_info_ext info = {
2061 .info.dev = dev,
2062 .ext.mtu = arg,
2063 };
2064
2065 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2066
2067 return call_netdevice_notifiers_info(val, &info.info);
2068 }
2069
2070 #ifdef CONFIG_NET_INGRESS
2071 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2072
net_inc_ingress_queue(void)2073 void net_inc_ingress_queue(void)
2074 {
2075 static_branch_inc(&ingress_needed_key);
2076 }
2077 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2078
net_dec_ingress_queue(void)2079 void net_dec_ingress_queue(void)
2080 {
2081 static_branch_dec(&ingress_needed_key);
2082 }
2083 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2084 #endif
2085
2086 #ifdef CONFIG_NET_EGRESS
2087 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2088
net_inc_egress_queue(void)2089 void net_inc_egress_queue(void)
2090 {
2091 static_branch_inc(&egress_needed_key);
2092 }
2093 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2094
net_dec_egress_queue(void)2095 void net_dec_egress_queue(void)
2096 {
2097 static_branch_dec(&egress_needed_key);
2098 }
2099 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2100 #endif
2101
2102 #ifdef CONFIG_NET_CLS_ACT
2103 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key);
2104 EXPORT_SYMBOL(tcf_bypass_check_needed_key);
2105 #endif
2106
2107 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2108 EXPORT_SYMBOL(netstamp_needed_key);
2109 #ifdef CONFIG_JUMP_LABEL
2110 static atomic_t netstamp_needed_deferred;
2111 static atomic_t netstamp_wanted;
netstamp_clear(struct work_struct * work)2112 static void netstamp_clear(struct work_struct *work)
2113 {
2114 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2115 int wanted;
2116
2117 wanted = atomic_add_return(deferred, &netstamp_wanted);
2118 if (wanted > 0)
2119 static_branch_enable(&netstamp_needed_key);
2120 else
2121 static_branch_disable(&netstamp_needed_key);
2122 }
2123 static DECLARE_WORK(netstamp_work, netstamp_clear);
2124 #endif
2125
net_enable_timestamp(void)2126 void net_enable_timestamp(void)
2127 {
2128 #ifdef CONFIG_JUMP_LABEL
2129 int wanted = atomic_read(&netstamp_wanted);
2130
2131 while (wanted > 0) {
2132 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2133 return;
2134 }
2135 atomic_inc(&netstamp_needed_deferred);
2136 schedule_work(&netstamp_work);
2137 #else
2138 static_branch_inc(&netstamp_needed_key);
2139 #endif
2140 }
2141 EXPORT_SYMBOL(net_enable_timestamp);
2142
net_disable_timestamp(void)2143 void net_disable_timestamp(void)
2144 {
2145 #ifdef CONFIG_JUMP_LABEL
2146 int wanted = atomic_read(&netstamp_wanted);
2147
2148 while (wanted > 1) {
2149 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2150 return;
2151 }
2152 atomic_dec(&netstamp_needed_deferred);
2153 schedule_work(&netstamp_work);
2154 #else
2155 static_branch_dec(&netstamp_needed_key);
2156 #endif
2157 }
2158 EXPORT_SYMBOL(net_disable_timestamp);
2159
net_timestamp_set(struct sk_buff * skb)2160 static inline void net_timestamp_set(struct sk_buff *skb)
2161 {
2162 skb->tstamp = 0;
2163 skb->mono_delivery_time = 0;
2164 if (static_branch_unlikely(&netstamp_needed_key))
2165 skb->tstamp = ktime_get_real();
2166 }
2167
2168 #define net_timestamp_check(COND, SKB) \
2169 if (static_branch_unlikely(&netstamp_needed_key)) { \
2170 if ((COND) && !(SKB)->tstamp) \
2171 (SKB)->tstamp = ktime_get_real(); \
2172 } \
2173
is_skb_forwardable(const struct net_device * dev,const struct sk_buff * skb)2174 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2175 {
2176 return __is_skb_forwardable(dev, skb, true);
2177 }
2178 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2179
__dev_forward_skb2(struct net_device * dev,struct sk_buff * skb,bool check_mtu)2180 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2181 bool check_mtu)
2182 {
2183 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2184
2185 if (likely(!ret)) {
2186 skb->protocol = eth_type_trans(skb, dev);
2187 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2188 }
2189
2190 return ret;
2191 }
2192
__dev_forward_skb(struct net_device * dev,struct sk_buff * skb)2193 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2194 {
2195 return __dev_forward_skb2(dev, skb, true);
2196 }
2197 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2198
2199 /**
2200 * dev_forward_skb - loopback an skb to another netif
2201 *
2202 * @dev: destination network device
2203 * @skb: buffer to forward
2204 *
2205 * return values:
2206 * NET_RX_SUCCESS (no congestion)
2207 * NET_RX_DROP (packet was dropped, but freed)
2208 *
2209 * dev_forward_skb can be used for injecting an skb from the
2210 * start_xmit function of one device into the receive queue
2211 * of another device.
2212 *
2213 * The receiving device may be in another namespace, so
2214 * we have to clear all information in the skb that could
2215 * impact namespace isolation.
2216 */
dev_forward_skb(struct net_device * dev,struct sk_buff * skb)2217 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2218 {
2219 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2220 }
2221 EXPORT_SYMBOL_GPL(dev_forward_skb);
2222
dev_forward_skb_nomtu(struct net_device * dev,struct sk_buff * skb)2223 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2224 {
2225 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2226 }
2227
deliver_skb(struct sk_buff * skb,struct packet_type * pt_prev,struct net_device * orig_dev)2228 static inline int deliver_skb(struct sk_buff *skb,
2229 struct packet_type *pt_prev,
2230 struct net_device *orig_dev)
2231 {
2232 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2233 return -ENOMEM;
2234 refcount_inc(&skb->users);
2235 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2236 }
2237
deliver_ptype_list_skb(struct sk_buff * skb,struct packet_type ** pt,struct net_device * orig_dev,__be16 type,struct list_head * ptype_list)2238 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2239 struct packet_type **pt,
2240 struct net_device *orig_dev,
2241 __be16 type,
2242 struct list_head *ptype_list)
2243 {
2244 struct packet_type *ptype, *pt_prev = *pt;
2245
2246 list_for_each_entry_rcu(ptype, ptype_list, list) {
2247 if (ptype->type != type)
2248 continue;
2249 if (pt_prev)
2250 deliver_skb(skb, pt_prev, orig_dev);
2251 pt_prev = ptype;
2252 }
2253 *pt = pt_prev;
2254 }
2255
skb_loop_sk(struct packet_type * ptype,struct sk_buff * skb)2256 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2257 {
2258 if (!ptype->af_packet_priv || !skb->sk)
2259 return false;
2260
2261 if (ptype->id_match)
2262 return ptype->id_match(ptype, skb->sk);
2263 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2264 return true;
2265
2266 return false;
2267 }
2268
2269 /**
2270 * dev_nit_active - return true if any network interface taps are in use
2271 *
2272 * @dev: network device to check for the presence of taps
2273 */
dev_nit_active(struct net_device * dev)2274 bool dev_nit_active(struct net_device *dev)
2275 {
2276 return !list_empty(&net_hotdata.ptype_all) ||
2277 !list_empty(&dev->ptype_all);
2278 }
2279 EXPORT_SYMBOL_GPL(dev_nit_active);
2280
2281 /*
2282 * Support routine. Sends outgoing frames to any network
2283 * taps currently in use.
2284 */
2285
dev_queue_xmit_nit(struct sk_buff * skb,struct net_device * dev)2286 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2287 {
2288 struct list_head *ptype_list = &net_hotdata.ptype_all;
2289 struct packet_type *ptype, *pt_prev = NULL;
2290 struct sk_buff *skb2 = NULL;
2291
2292 rcu_read_lock();
2293 again:
2294 list_for_each_entry_rcu(ptype, ptype_list, list) {
2295 if (READ_ONCE(ptype->ignore_outgoing))
2296 continue;
2297
2298 /* Never send packets back to the socket
2299 * they originated from - MvS (miquels@drinkel.ow.org)
2300 */
2301 if (skb_loop_sk(ptype, skb))
2302 continue;
2303
2304 if (pt_prev) {
2305 deliver_skb(skb2, pt_prev, skb->dev);
2306 pt_prev = ptype;
2307 continue;
2308 }
2309
2310 /* need to clone skb, done only once */
2311 skb2 = skb_clone(skb, GFP_ATOMIC);
2312 if (!skb2)
2313 goto out_unlock;
2314
2315 net_timestamp_set(skb2);
2316
2317 /* skb->nh should be correctly
2318 * set by sender, so that the second statement is
2319 * just protection against buggy protocols.
2320 */
2321 skb_reset_mac_header(skb2);
2322
2323 if (skb_network_header(skb2) < skb2->data ||
2324 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2325 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2326 ntohs(skb2->protocol),
2327 dev->name);
2328 skb_reset_network_header(skb2);
2329 }
2330
2331 skb2->transport_header = skb2->network_header;
2332 skb2->pkt_type = PACKET_OUTGOING;
2333 pt_prev = ptype;
2334 }
2335
2336 if (ptype_list == &net_hotdata.ptype_all) {
2337 ptype_list = &dev->ptype_all;
2338 goto again;
2339 }
2340 out_unlock:
2341 if (pt_prev) {
2342 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2343 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2344 else
2345 kfree_skb(skb2);
2346 }
2347 rcu_read_unlock();
2348 }
2349 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2350
2351 /**
2352 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2353 * @dev: Network device
2354 * @txq: number of queues available
2355 *
2356 * If real_num_tx_queues is changed the tc mappings may no longer be
2357 * valid. To resolve this verify the tc mapping remains valid and if
2358 * not NULL the mapping. With no priorities mapping to this
2359 * offset/count pair it will no longer be used. In the worst case TC0
2360 * is invalid nothing can be done so disable priority mappings. If is
2361 * expected that drivers will fix this mapping if they can before
2362 * calling netif_set_real_num_tx_queues.
2363 */
netif_setup_tc(struct net_device * dev,unsigned int txq)2364 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2365 {
2366 int i;
2367 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2368
2369 /* If TC0 is invalidated disable TC mapping */
2370 if (tc->offset + tc->count > txq) {
2371 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2372 dev->num_tc = 0;
2373 return;
2374 }
2375
2376 /* Invalidated prio to tc mappings set to TC0 */
2377 for (i = 1; i < TC_BITMASK + 1; i++) {
2378 int q = netdev_get_prio_tc_map(dev, i);
2379
2380 tc = &dev->tc_to_txq[q];
2381 if (tc->offset + tc->count > txq) {
2382 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",
2383 i, q);
2384 netdev_set_prio_tc_map(dev, i, 0);
2385 }
2386 }
2387 }
2388
netdev_txq_to_tc(struct net_device * dev,unsigned int txq)2389 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2390 {
2391 if (dev->num_tc) {
2392 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2393 int i;
2394
2395 /* walk through the TCs and see if it falls into any of them */
2396 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2397 if ((txq - tc->offset) < tc->count)
2398 return i;
2399 }
2400
2401 /* didn't find it, just return -1 to indicate no match */
2402 return -1;
2403 }
2404
2405 return 0;
2406 }
2407 EXPORT_SYMBOL(netdev_txq_to_tc);
2408
2409 #ifdef CONFIG_XPS
2410 static struct static_key xps_needed __read_mostly;
2411 static struct static_key xps_rxqs_needed __read_mostly;
2412 static DEFINE_MUTEX(xps_map_mutex);
2413 #define xmap_dereference(P) \
2414 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2415
remove_xps_queue(struct xps_dev_maps * dev_maps,struct xps_dev_maps * old_maps,int tci,u16 index)2416 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2417 struct xps_dev_maps *old_maps, int tci, u16 index)
2418 {
2419 struct xps_map *map = NULL;
2420 int pos;
2421
2422 map = xmap_dereference(dev_maps->attr_map[tci]);
2423 if (!map)
2424 return false;
2425
2426 for (pos = map->len; pos--;) {
2427 if (map->queues[pos] != index)
2428 continue;
2429
2430 if (map->len > 1) {
2431 map->queues[pos] = map->queues[--map->len];
2432 break;
2433 }
2434
2435 if (old_maps)
2436 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2437 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2438 kfree_rcu(map, rcu);
2439 return false;
2440 }
2441
2442 return true;
2443 }
2444
remove_xps_queue_cpu(struct net_device * dev,struct xps_dev_maps * dev_maps,int cpu,u16 offset,u16 count)2445 static bool remove_xps_queue_cpu(struct net_device *dev,
2446 struct xps_dev_maps *dev_maps,
2447 int cpu, u16 offset, u16 count)
2448 {
2449 int num_tc = dev_maps->num_tc;
2450 bool active = false;
2451 int tci;
2452
2453 for (tci = cpu * num_tc; num_tc--; tci++) {
2454 int i, j;
2455
2456 for (i = count, j = offset; i--; j++) {
2457 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2458 break;
2459 }
2460
2461 active |= i < 0;
2462 }
2463
2464 return active;
2465 }
2466
reset_xps_maps(struct net_device * dev,struct xps_dev_maps * dev_maps,enum xps_map_type type)2467 static void reset_xps_maps(struct net_device *dev,
2468 struct xps_dev_maps *dev_maps,
2469 enum xps_map_type type)
2470 {
2471 static_key_slow_dec_cpuslocked(&xps_needed);
2472 if (type == XPS_RXQS)
2473 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2474
2475 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2476
2477 kfree_rcu(dev_maps, rcu);
2478 }
2479
clean_xps_maps(struct net_device * dev,enum xps_map_type type,u16 offset,u16 count)2480 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2481 u16 offset, u16 count)
2482 {
2483 struct xps_dev_maps *dev_maps;
2484 bool active = false;
2485 int i, j;
2486
2487 dev_maps = xmap_dereference(dev->xps_maps[type]);
2488 if (!dev_maps)
2489 return;
2490
2491 for (j = 0; j < dev_maps->nr_ids; j++)
2492 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2493 if (!active)
2494 reset_xps_maps(dev, dev_maps, type);
2495
2496 if (type == XPS_CPUS) {
2497 for (i = offset + (count - 1); count--; i--)
2498 netdev_queue_numa_node_write(
2499 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2500 }
2501 }
2502
netif_reset_xps_queues(struct net_device * dev,u16 offset,u16 count)2503 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2504 u16 count)
2505 {
2506 if (!static_key_false(&xps_needed))
2507 return;
2508
2509 cpus_read_lock();
2510 mutex_lock(&xps_map_mutex);
2511
2512 if (static_key_false(&xps_rxqs_needed))
2513 clean_xps_maps(dev, XPS_RXQS, offset, count);
2514
2515 clean_xps_maps(dev, XPS_CPUS, offset, count);
2516
2517 mutex_unlock(&xps_map_mutex);
2518 cpus_read_unlock();
2519 }
2520
netif_reset_xps_queues_gt(struct net_device * dev,u16 index)2521 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2522 {
2523 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2524 }
2525
expand_xps_map(struct xps_map * map,int attr_index,u16 index,bool is_rxqs_map)2526 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2527 u16 index, bool is_rxqs_map)
2528 {
2529 struct xps_map *new_map;
2530 int alloc_len = XPS_MIN_MAP_ALLOC;
2531 int i, pos;
2532
2533 for (pos = 0; map && pos < map->len; pos++) {
2534 if (map->queues[pos] != index)
2535 continue;
2536 return map;
2537 }
2538
2539 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2540 if (map) {
2541 if (pos < map->alloc_len)
2542 return map;
2543
2544 alloc_len = map->alloc_len * 2;
2545 }
2546
2547 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2548 * map
2549 */
2550 if (is_rxqs_map)
2551 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2552 else
2553 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2554 cpu_to_node(attr_index));
2555 if (!new_map)
2556 return NULL;
2557
2558 for (i = 0; i < pos; i++)
2559 new_map->queues[i] = map->queues[i];
2560 new_map->alloc_len = alloc_len;
2561 new_map->len = pos;
2562
2563 return new_map;
2564 }
2565
2566 /* Copy xps maps at a given index */
xps_copy_dev_maps(struct xps_dev_maps * dev_maps,struct xps_dev_maps * new_dev_maps,int index,int tc,bool skip_tc)2567 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2568 struct xps_dev_maps *new_dev_maps, int index,
2569 int tc, bool skip_tc)
2570 {
2571 int i, tci = index * dev_maps->num_tc;
2572 struct xps_map *map;
2573
2574 /* copy maps belonging to foreign traffic classes */
2575 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2576 if (i == tc && skip_tc)
2577 continue;
2578
2579 /* fill in the new device map from the old device map */
2580 map = xmap_dereference(dev_maps->attr_map[tci]);
2581 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2582 }
2583 }
2584
2585 /* Must be called under cpus_read_lock */
__netif_set_xps_queue(struct net_device * dev,const unsigned long * mask,u16 index,enum xps_map_type type)2586 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2587 u16 index, enum xps_map_type type)
2588 {
2589 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2590 const unsigned long *online_mask = NULL;
2591 bool active = false, copy = false;
2592 int i, j, tci, numa_node_id = -2;
2593 int maps_sz, num_tc = 1, tc = 0;
2594 struct xps_map *map, *new_map;
2595 unsigned int nr_ids;
2596
2597 WARN_ON_ONCE(index >= dev->num_tx_queues);
2598
2599 if (dev->num_tc) {
2600 /* Do not allow XPS on subordinate device directly */
2601 num_tc = dev->num_tc;
2602 if (num_tc < 0)
2603 return -EINVAL;
2604
2605 /* If queue belongs to subordinate dev use its map */
2606 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2607
2608 tc = netdev_txq_to_tc(dev, index);
2609 if (tc < 0)
2610 return -EINVAL;
2611 }
2612
2613 mutex_lock(&xps_map_mutex);
2614
2615 dev_maps = xmap_dereference(dev->xps_maps[type]);
2616 if (type == XPS_RXQS) {
2617 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2618 nr_ids = dev->num_rx_queues;
2619 } else {
2620 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2621 if (num_possible_cpus() > 1)
2622 online_mask = cpumask_bits(cpu_online_mask);
2623 nr_ids = nr_cpu_ids;
2624 }
2625
2626 if (maps_sz < L1_CACHE_BYTES)
2627 maps_sz = L1_CACHE_BYTES;
2628
2629 /* The old dev_maps could be larger or smaller than the one we're
2630 * setting up now, as dev->num_tc or nr_ids could have been updated in
2631 * between. We could try to be smart, but let's be safe instead and only
2632 * copy foreign traffic classes if the two map sizes match.
2633 */
2634 if (dev_maps &&
2635 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2636 copy = true;
2637
2638 /* allocate memory for queue storage */
2639 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2640 j < nr_ids;) {
2641 if (!new_dev_maps) {
2642 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2643 if (!new_dev_maps) {
2644 mutex_unlock(&xps_map_mutex);
2645 return -ENOMEM;
2646 }
2647
2648 new_dev_maps->nr_ids = nr_ids;
2649 new_dev_maps->num_tc = num_tc;
2650 }
2651
2652 tci = j * num_tc + tc;
2653 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2654
2655 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2656 if (!map)
2657 goto error;
2658
2659 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2660 }
2661
2662 if (!new_dev_maps)
2663 goto out_no_new_maps;
2664
2665 if (!dev_maps) {
2666 /* Increment static keys at most once per type */
2667 static_key_slow_inc_cpuslocked(&xps_needed);
2668 if (type == XPS_RXQS)
2669 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2670 }
2671
2672 for (j = 0; j < nr_ids; j++) {
2673 bool skip_tc = false;
2674
2675 tci = j * num_tc + tc;
2676 if (netif_attr_test_mask(j, mask, nr_ids) &&
2677 netif_attr_test_online(j, online_mask, nr_ids)) {
2678 /* add tx-queue to CPU/rx-queue maps */
2679 int pos = 0;
2680
2681 skip_tc = true;
2682
2683 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2684 while ((pos < map->len) && (map->queues[pos] != index))
2685 pos++;
2686
2687 if (pos == map->len)
2688 map->queues[map->len++] = index;
2689 #ifdef CONFIG_NUMA
2690 if (type == XPS_CPUS) {
2691 if (numa_node_id == -2)
2692 numa_node_id = cpu_to_node(j);
2693 else if (numa_node_id != cpu_to_node(j))
2694 numa_node_id = -1;
2695 }
2696 #endif
2697 }
2698
2699 if (copy)
2700 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2701 skip_tc);
2702 }
2703
2704 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2705
2706 /* Cleanup old maps */
2707 if (!dev_maps)
2708 goto out_no_old_maps;
2709
2710 for (j = 0; j < dev_maps->nr_ids; j++) {
2711 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2712 map = xmap_dereference(dev_maps->attr_map[tci]);
2713 if (!map)
2714 continue;
2715
2716 if (copy) {
2717 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2718 if (map == new_map)
2719 continue;
2720 }
2721
2722 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2723 kfree_rcu(map, rcu);
2724 }
2725 }
2726
2727 old_dev_maps = dev_maps;
2728
2729 out_no_old_maps:
2730 dev_maps = new_dev_maps;
2731 active = true;
2732
2733 out_no_new_maps:
2734 if (type == XPS_CPUS)
2735 /* update Tx queue numa node */
2736 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2737 (numa_node_id >= 0) ?
2738 numa_node_id : NUMA_NO_NODE);
2739
2740 if (!dev_maps)
2741 goto out_no_maps;
2742
2743 /* removes tx-queue from unused CPUs/rx-queues */
2744 for (j = 0; j < dev_maps->nr_ids; j++) {
2745 tci = j * dev_maps->num_tc;
2746
2747 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2748 if (i == tc &&
2749 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2750 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2751 continue;
2752
2753 active |= remove_xps_queue(dev_maps,
2754 copy ? old_dev_maps : NULL,
2755 tci, index);
2756 }
2757 }
2758
2759 if (old_dev_maps)
2760 kfree_rcu(old_dev_maps, rcu);
2761
2762 /* free map if not active */
2763 if (!active)
2764 reset_xps_maps(dev, dev_maps, type);
2765
2766 out_no_maps:
2767 mutex_unlock(&xps_map_mutex);
2768
2769 return 0;
2770 error:
2771 /* remove any maps that we added */
2772 for (j = 0; j < nr_ids; j++) {
2773 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2774 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2775 map = copy ?
2776 xmap_dereference(dev_maps->attr_map[tci]) :
2777 NULL;
2778 if (new_map && new_map != map)
2779 kfree(new_map);
2780 }
2781 }
2782
2783 mutex_unlock(&xps_map_mutex);
2784
2785 kfree(new_dev_maps);
2786 return -ENOMEM;
2787 }
2788 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2789
netif_set_xps_queue(struct net_device * dev,const struct cpumask * mask,u16 index)2790 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2791 u16 index)
2792 {
2793 int ret;
2794
2795 cpus_read_lock();
2796 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2797 cpus_read_unlock();
2798
2799 return ret;
2800 }
2801 EXPORT_SYMBOL(netif_set_xps_queue);
2802
2803 #endif
netdev_unbind_all_sb_channels(struct net_device * dev)2804 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2805 {
2806 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2807
2808 /* Unbind any subordinate channels */
2809 while (txq-- != &dev->_tx[0]) {
2810 if (txq->sb_dev)
2811 netdev_unbind_sb_channel(dev, txq->sb_dev);
2812 }
2813 }
2814
netdev_reset_tc(struct net_device * dev)2815 void netdev_reset_tc(struct net_device *dev)
2816 {
2817 #ifdef CONFIG_XPS
2818 netif_reset_xps_queues_gt(dev, 0);
2819 #endif
2820 netdev_unbind_all_sb_channels(dev);
2821
2822 /* Reset TC configuration of device */
2823 dev->num_tc = 0;
2824 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2825 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2826 }
2827 EXPORT_SYMBOL(netdev_reset_tc);
2828
netdev_set_tc_queue(struct net_device * dev,u8 tc,u16 count,u16 offset)2829 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2830 {
2831 if (tc >= dev->num_tc)
2832 return -EINVAL;
2833
2834 #ifdef CONFIG_XPS
2835 netif_reset_xps_queues(dev, offset, count);
2836 #endif
2837 dev->tc_to_txq[tc].count = count;
2838 dev->tc_to_txq[tc].offset = offset;
2839 return 0;
2840 }
2841 EXPORT_SYMBOL(netdev_set_tc_queue);
2842
netdev_set_num_tc(struct net_device * dev,u8 num_tc)2843 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2844 {
2845 if (num_tc > TC_MAX_QUEUE)
2846 return -EINVAL;
2847
2848 #ifdef CONFIG_XPS
2849 netif_reset_xps_queues_gt(dev, 0);
2850 #endif
2851 netdev_unbind_all_sb_channels(dev);
2852
2853 dev->num_tc = num_tc;
2854 return 0;
2855 }
2856 EXPORT_SYMBOL(netdev_set_num_tc);
2857
netdev_unbind_sb_channel(struct net_device * dev,struct net_device * sb_dev)2858 void netdev_unbind_sb_channel(struct net_device *dev,
2859 struct net_device *sb_dev)
2860 {
2861 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2862
2863 #ifdef CONFIG_XPS
2864 netif_reset_xps_queues_gt(sb_dev, 0);
2865 #endif
2866 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2867 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2868
2869 while (txq-- != &dev->_tx[0]) {
2870 if (txq->sb_dev == sb_dev)
2871 txq->sb_dev = NULL;
2872 }
2873 }
2874 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2875
netdev_bind_sb_channel_queue(struct net_device * dev,struct net_device * sb_dev,u8 tc,u16 count,u16 offset)2876 int netdev_bind_sb_channel_queue(struct net_device *dev,
2877 struct net_device *sb_dev,
2878 u8 tc, u16 count, u16 offset)
2879 {
2880 /* Make certain the sb_dev and dev are already configured */
2881 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2882 return -EINVAL;
2883
2884 /* We cannot hand out queues we don't have */
2885 if ((offset + count) > dev->real_num_tx_queues)
2886 return -EINVAL;
2887
2888 /* Record the mapping */
2889 sb_dev->tc_to_txq[tc].count = count;
2890 sb_dev->tc_to_txq[tc].offset = offset;
2891
2892 /* Provide a way for Tx queue to find the tc_to_txq map or
2893 * XPS map for itself.
2894 */
2895 while (count--)
2896 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2897
2898 return 0;
2899 }
2900 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2901
netdev_set_sb_channel(struct net_device * dev,u16 channel)2902 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2903 {
2904 /* Do not use a multiqueue device to represent a subordinate channel */
2905 if (netif_is_multiqueue(dev))
2906 return -ENODEV;
2907
2908 /* We allow channels 1 - 32767 to be used for subordinate channels.
2909 * Channel 0 is meant to be "native" mode and used only to represent
2910 * the main root device. We allow writing 0 to reset the device back
2911 * to normal mode after being used as a subordinate channel.
2912 */
2913 if (channel > S16_MAX)
2914 return -EINVAL;
2915
2916 dev->num_tc = -channel;
2917
2918 return 0;
2919 }
2920 EXPORT_SYMBOL(netdev_set_sb_channel);
2921
2922 /*
2923 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2924 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2925 */
netif_set_real_num_tx_queues(struct net_device * dev,unsigned int txq)2926 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2927 {
2928 bool disabling;
2929 int rc;
2930
2931 disabling = txq < dev->real_num_tx_queues;
2932
2933 if (txq < 1 || txq > dev->num_tx_queues)
2934 return -EINVAL;
2935
2936 if (dev->reg_state == NETREG_REGISTERED ||
2937 dev->reg_state == NETREG_UNREGISTERING) {
2938 ASSERT_RTNL();
2939
2940 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2941 txq);
2942 if (rc)
2943 return rc;
2944
2945 if (dev->num_tc)
2946 netif_setup_tc(dev, txq);
2947
2948 dev_qdisc_change_real_num_tx(dev, txq);
2949
2950 dev->real_num_tx_queues = txq;
2951
2952 if (disabling) {
2953 synchronize_net();
2954 qdisc_reset_all_tx_gt(dev, txq);
2955 #ifdef CONFIG_XPS
2956 netif_reset_xps_queues_gt(dev, txq);
2957 #endif
2958 }
2959 } else {
2960 dev->real_num_tx_queues = txq;
2961 }
2962
2963 return 0;
2964 }
2965 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2966
2967 #ifdef CONFIG_SYSFS
2968 /**
2969 * netif_set_real_num_rx_queues - set actual number of RX queues used
2970 * @dev: Network device
2971 * @rxq: Actual number of RX queues
2972 *
2973 * This must be called either with the rtnl_lock held or before
2974 * registration of the net device. Returns 0 on success, or a
2975 * negative error code. If called before registration, it always
2976 * succeeds.
2977 */
netif_set_real_num_rx_queues(struct net_device * dev,unsigned int rxq)2978 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2979 {
2980 int rc;
2981
2982 if (rxq < 1 || rxq > dev->num_rx_queues)
2983 return -EINVAL;
2984
2985 if (dev->reg_state == NETREG_REGISTERED) {
2986 ASSERT_RTNL();
2987
2988 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2989 rxq);
2990 if (rc)
2991 return rc;
2992 }
2993
2994 dev->real_num_rx_queues = rxq;
2995 return 0;
2996 }
2997 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2998 #endif
2999
3000 /**
3001 * netif_set_real_num_queues - set actual number of RX and TX queues used
3002 * @dev: Network device
3003 * @txq: Actual number of TX queues
3004 * @rxq: Actual number of RX queues
3005 *
3006 * Set the real number of both TX and RX queues.
3007 * Does nothing if the number of queues is already correct.
3008 */
netif_set_real_num_queues(struct net_device * dev,unsigned int txq,unsigned int rxq)3009 int netif_set_real_num_queues(struct net_device *dev,
3010 unsigned int txq, unsigned int rxq)
3011 {
3012 unsigned int old_rxq = dev->real_num_rx_queues;
3013 int err;
3014
3015 if (txq < 1 || txq > dev->num_tx_queues ||
3016 rxq < 1 || rxq > dev->num_rx_queues)
3017 return -EINVAL;
3018
3019 /* Start from increases, so the error path only does decreases -
3020 * decreases can't fail.
3021 */
3022 if (rxq > dev->real_num_rx_queues) {
3023 err = netif_set_real_num_rx_queues(dev, rxq);
3024 if (err)
3025 return err;
3026 }
3027 if (txq > dev->real_num_tx_queues) {
3028 err = netif_set_real_num_tx_queues(dev, txq);
3029 if (err)
3030 goto undo_rx;
3031 }
3032 if (rxq < dev->real_num_rx_queues)
3033 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3034 if (txq < dev->real_num_tx_queues)
3035 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3036
3037 return 0;
3038 undo_rx:
3039 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3040 return err;
3041 }
3042 EXPORT_SYMBOL(netif_set_real_num_queues);
3043
3044 /**
3045 * netif_set_tso_max_size() - set the max size of TSO frames supported
3046 * @dev: netdev to update
3047 * @size: max skb->len of a TSO frame
3048 *
3049 * Set the limit on the size of TSO super-frames the device can handle.
3050 * Unless explicitly set the stack will assume the value of
3051 * %GSO_LEGACY_MAX_SIZE.
3052 */
netif_set_tso_max_size(struct net_device * dev,unsigned int size)3053 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3054 {
3055 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3056 if (size < READ_ONCE(dev->gso_max_size))
3057 netif_set_gso_max_size(dev, size);
3058 if (size < READ_ONCE(dev->gso_ipv4_max_size))
3059 netif_set_gso_ipv4_max_size(dev, size);
3060 }
3061 EXPORT_SYMBOL(netif_set_tso_max_size);
3062
3063 /**
3064 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3065 * @dev: netdev to update
3066 * @segs: max number of TCP segments
3067 *
3068 * Set the limit on the number of TCP segments the device can generate from
3069 * a single TSO super-frame.
3070 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3071 */
netif_set_tso_max_segs(struct net_device * dev,unsigned int segs)3072 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3073 {
3074 dev->tso_max_segs = segs;
3075 if (segs < READ_ONCE(dev->gso_max_segs))
3076 netif_set_gso_max_segs(dev, segs);
3077 }
3078 EXPORT_SYMBOL(netif_set_tso_max_segs);
3079
3080 /**
3081 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3082 * @to: netdev to update
3083 * @from: netdev from which to copy the limits
3084 */
netif_inherit_tso_max(struct net_device * to,const struct net_device * from)3085 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3086 {
3087 netif_set_tso_max_size(to, from->tso_max_size);
3088 netif_set_tso_max_segs(to, from->tso_max_segs);
3089 }
3090 EXPORT_SYMBOL(netif_inherit_tso_max);
3091
3092 /**
3093 * netif_get_num_default_rss_queues - default number of RSS queues
3094 *
3095 * Default value is the number of physical cores if there are only 1 or 2, or
3096 * divided by 2 if there are more.
3097 */
netif_get_num_default_rss_queues(void)3098 int netif_get_num_default_rss_queues(void)
3099 {
3100 cpumask_var_t cpus;
3101 int cpu, count = 0;
3102
3103 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3104 return 1;
3105
3106 cpumask_copy(cpus, cpu_online_mask);
3107 for_each_cpu(cpu, cpus) {
3108 ++count;
3109 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3110 }
3111 free_cpumask_var(cpus);
3112
3113 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3114 }
3115 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3116
__netif_reschedule(struct Qdisc * q)3117 static void __netif_reschedule(struct Qdisc *q)
3118 {
3119 struct softnet_data *sd;
3120 unsigned long flags;
3121
3122 local_irq_save(flags);
3123 sd = this_cpu_ptr(&softnet_data);
3124 q->next_sched = NULL;
3125 *sd->output_queue_tailp = q;
3126 sd->output_queue_tailp = &q->next_sched;
3127 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3128 local_irq_restore(flags);
3129 }
3130
__netif_schedule(struct Qdisc * q)3131 void __netif_schedule(struct Qdisc *q)
3132 {
3133 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3134 __netif_reschedule(q);
3135 }
3136 EXPORT_SYMBOL(__netif_schedule);
3137
3138 struct dev_kfree_skb_cb {
3139 enum skb_drop_reason reason;
3140 };
3141
get_kfree_skb_cb(const struct sk_buff * skb)3142 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3143 {
3144 return (struct dev_kfree_skb_cb *)skb->cb;
3145 }
3146
netif_schedule_queue(struct netdev_queue * txq)3147 void netif_schedule_queue(struct netdev_queue *txq)
3148 {
3149 rcu_read_lock();
3150 if (!netif_xmit_stopped(txq)) {
3151 struct Qdisc *q = rcu_dereference(txq->qdisc);
3152
3153 __netif_schedule(q);
3154 }
3155 rcu_read_unlock();
3156 }
3157 EXPORT_SYMBOL(netif_schedule_queue);
3158
netif_tx_wake_queue(struct netdev_queue * dev_queue)3159 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3160 {
3161 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3162 struct Qdisc *q;
3163
3164 rcu_read_lock();
3165 q = rcu_dereference(dev_queue->qdisc);
3166 __netif_schedule(q);
3167 rcu_read_unlock();
3168 }
3169 }
3170 EXPORT_SYMBOL(netif_tx_wake_queue);
3171
dev_kfree_skb_irq_reason(struct sk_buff * skb,enum skb_drop_reason reason)3172 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3173 {
3174 unsigned long flags;
3175
3176 if (unlikely(!skb))
3177 return;
3178
3179 if (likely(refcount_read(&skb->users) == 1)) {
3180 smp_rmb();
3181 refcount_set(&skb->users, 0);
3182 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3183 return;
3184 }
3185 get_kfree_skb_cb(skb)->reason = reason;
3186 local_irq_save(flags);
3187 skb->next = __this_cpu_read(softnet_data.completion_queue);
3188 __this_cpu_write(softnet_data.completion_queue, skb);
3189 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3190 local_irq_restore(flags);
3191 }
3192 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3193
dev_kfree_skb_any_reason(struct sk_buff * skb,enum skb_drop_reason reason)3194 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3195 {
3196 if (in_hardirq() || irqs_disabled())
3197 dev_kfree_skb_irq_reason(skb, reason);
3198 else
3199 kfree_skb_reason(skb, reason);
3200 }
3201 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3202
3203
3204 /**
3205 * netif_device_detach - mark device as removed
3206 * @dev: network device
3207 *
3208 * Mark device as removed from system and therefore no longer available.
3209 */
netif_device_detach(struct net_device * dev)3210 void netif_device_detach(struct net_device *dev)
3211 {
3212 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3213 netif_running(dev)) {
3214 netif_tx_stop_all_queues(dev);
3215 }
3216 }
3217 EXPORT_SYMBOL(netif_device_detach);
3218
3219 /**
3220 * netif_device_attach - mark device as attached
3221 * @dev: network device
3222 *
3223 * Mark device as attached from system and restart if needed.
3224 */
netif_device_attach(struct net_device * dev)3225 void netif_device_attach(struct net_device *dev)
3226 {
3227 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3228 netif_running(dev)) {
3229 netif_tx_wake_all_queues(dev);
3230 __netdev_watchdog_up(dev);
3231 }
3232 }
3233 EXPORT_SYMBOL(netif_device_attach);
3234
3235 /*
3236 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3237 * to be used as a distribution range.
3238 */
skb_tx_hash(const struct net_device * dev,const struct net_device * sb_dev,struct sk_buff * skb)3239 static u16 skb_tx_hash(const struct net_device *dev,
3240 const struct net_device *sb_dev,
3241 struct sk_buff *skb)
3242 {
3243 u32 hash;
3244 u16 qoffset = 0;
3245 u16 qcount = dev->real_num_tx_queues;
3246
3247 if (dev->num_tc) {
3248 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3249
3250 qoffset = sb_dev->tc_to_txq[tc].offset;
3251 qcount = sb_dev->tc_to_txq[tc].count;
3252 if (unlikely(!qcount)) {
3253 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3254 sb_dev->name, qoffset, tc);
3255 qoffset = 0;
3256 qcount = dev->real_num_tx_queues;
3257 }
3258 }
3259
3260 if (skb_rx_queue_recorded(skb)) {
3261 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3262 hash = skb_get_rx_queue(skb);
3263 if (hash >= qoffset)
3264 hash -= qoffset;
3265 while (unlikely(hash >= qcount))
3266 hash -= qcount;
3267 return hash + qoffset;
3268 }
3269
3270 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3271 }
3272
skb_warn_bad_offload(const struct sk_buff * skb)3273 void skb_warn_bad_offload(const struct sk_buff *skb)
3274 {
3275 static const netdev_features_t null_features;
3276 struct net_device *dev = skb->dev;
3277 const char *name = "";
3278
3279 if (!net_ratelimit())
3280 return;
3281
3282 if (dev) {
3283 if (dev->dev.parent)
3284 name = dev_driver_string(dev->dev.parent);
3285 else
3286 name = netdev_name(dev);
3287 }
3288 skb_dump(KERN_WARNING, skb, false);
3289 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3290 name, dev ? &dev->features : &null_features,
3291 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3292 }
3293
3294 /*
3295 * Invalidate hardware checksum when packet is to be mangled, and
3296 * complete checksum manually on outgoing path.
3297 */
skb_checksum_help(struct sk_buff * skb)3298 int skb_checksum_help(struct sk_buff *skb)
3299 {
3300 __wsum csum;
3301 int ret = 0, offset;
3302
3303 if (skb->ip_summed == CHECKSUM_COMPLETE)
3304 goto out_set_summed;
3305
3306 if (unlikely(skb_is_gso(skb))) {
3307 skb_warn_bad_offload(skb);
3308 return -EINVAL;
3309 }
3310
3311 /* Before computing a checksum, we should make sure no frag could
3312 * be modified by an external entity : checksum could be wrong.
3313 */
3314 if (skb_has_shared_frag(skb)) {
3315 ret = __skb_linearize(skb);
3316 if (ret)
3317 goto out;
3318 }
3319
3320 offset = skb_checksum_start_offset(skb);
3321 ret = -EINVAL;
3322 if (unlikely(offset >= skb_headlen(skb))) {
3323 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3324 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3325 offset, skb_headlen(skb));
3326 goto out;
3327 }
3328 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3329
3330 offset += skb->csum_offset;
3331 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3332 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3333 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3334 offset + sizeof(__sum16), skb_headlen(skb));
3335 goto out;
3336 }
3337 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3338 if (ret)
3339 goto out;
3340
3341 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3342 out_set_summed:
3343 skb->ip_summed = CHECKSUM_NONE;
3344 out:
3345 return ret;
3346 }
3347 EXPORT_SYMBOL(skb_checksum_help);
3348
skb_crc32c_csum_help(struct sk_buff * skb)3349 int skb_crc32c_csum_help(struct sk_buff *skb)
3350 {
3351 __le32 crc32c_csum;
3352 int ret = 0, offset, start;
3353
3354 if (skb->ip_summed != CHECKSUM_PARTIAL)
3355 goto out;
3356
3357 if (unlikely(skb_is_gso(skb)))
3358 goto out;
3359
3360 /* Before computing a checksum, we should make sure no frag could
3361 * be modified by an external entity : checksum could be wrong.
3362 */
3363 if (unlikely(skb_has_shared_frag(skb))) {
3364 ret = __skb_linearize(skb);
3365 if (ret)
3366 goto out;
3367 }
3368 start = skb_checksum_start_offset(skb);
3369 offset = start + offsetof(struct sctphdr, checksum);
3370 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3371 ret = -EINVAL;
3372 goto out;
3373 }
3374
3375 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3376 if (ret)
3377 goto out;
3378
3379 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3380 skb->len - start, ~(__u32)0,
3381 crc32c_csum_stub));
3382 *(__le32 *)(skb->data + offset) = crc32c_csum;
3383 skb_reset_csum_not_inet(skb);
3384 out:
3385 return ret;
3386 }
3387
skb_network_protocol(struct sk_buff * skb,int * depth)3388 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3389 {
3390 __be16 type = skb->protocol;
3391
3392 /* Tunnel gso handlers can set protocol to ethernet. */
3393 if (type == htons(ETH_P_TEB)) {
3394 struct ethhdr *eth;
3395
3396 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3397 return 0;
3398
3399 eth = (struct ethhdr *)skb->data;
3400 type = eth->h_proto;
3401 }
3402
3403 return vlan_get_protocol_and_depth(skb, type, depth);
3404 }
3405
3406
3407 /* Take action when hardware reception checksum errors are detected. */
3408 #ifdef CONFIG_BUG
do_netdev_rx_csum_fault(struct net_device * dev,struct sk_buff * skb)3409 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3410 {
3411 netdev_err(dev, "hw csum failure\n");
3412 skb_dump(KERN_ERR, skb, true);
3413 dump_stack();
3414 }
3415
netdev_rx_csum_fault(struct net_device * dev,struct sk_buff * skb)3416 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3417 {
3418 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3419 }
3420 EXPORT_SYMBOL(netdev_rx_csum_fault);
3421 #endif
3422
3423 /* XXX: check that highmem exists at all on the given machine. */
illegal_highdma(struct net_device * dev,struct sk_buff * skb)3424 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3425 {
3426 #ifdef CONFIG_HIGHMEM
3427 int i;
3428
3429 if (!(dev->features & NETIF_F_HIGHDMA)) {
3430 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3431 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3432
3433 if (PageHighMem(skb_frag_page(frag)))
3434 return 1;
3435 }
3436 }
3437 #endif
3438 return 0;
3439 }
3440
3441 /* If MPLS offload request, verify we are testing hardware MPLS features
3442 * instead of standard features for the netdev.
3443 */
3444 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
net_mpls_features(struct sk_buff * skb,netdev_features_t features,__be16 type)3445 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3446 netdev_features_t features,
3447 __be16 type)
3448 {
3449 if (eth_p_mpls(type))
3450 features &= skb->dev->mpls_features;
3451
3452 return features;
3453 }
3454 #else
net_mpls_features(struct sk_buff * skb,netdev_features_t features,__be16 type)3455 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3456 netdev_features_t features,
3457 __be16 type)
3458 {
3459 return features;
3460 }
3461 #endif
3462
harmonize_features(struct sk_buff * skb,netdev_features_t features)3463 static netdev_features_t harmonize_features(struct sk_buff *skb,
3464 netdev_features_t features)
3465 {
3466 __be16 type;
3467
3468 type = skb_network_protocol(skb, NULL);
3469 features = net_mpls_features(skb, features, type);
3470
3471 if (skb->ip_summed != CHECKSUM_NONE &&
3472 !can_checksum_protocol(features, type)) {
3473 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3474 }
3475 if (illegal_highdma(skb->dev, skb))
3476 features &= ~NETIF_F_SG;
3477
3478 return features;
3479 }
3480
passthru_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3481 netdev_features_t passthru_features_check(struct sk_buff *skb,
3482 struct net_device *dev,
3483 netdev_features_t features)
3484 {
3485 return features;
3486 }
3487 EXPORT_SYMBOL(passthru_features_check);
3488
dflt_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3489 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3490 struct net_device *dev,
3491 netdev_features_t features)
3492 {
3493 return vlan_features_check(skb, features);
3494 }
3495
gso_features_check(const struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3496 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3497 struct net_device *dev,
3498 netdev_features_t features)
3499 {
3500 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3501
3502 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3503 return features & ~NETIF_F_GSO_MASK;
3504
3505 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size)))
3506 return features & ~NETIF_F_GSO_MASK;
3507
3508 if (!skb_shinfo(skb)->gso_type) {
3509 skb_warn_bad_offload(skb);
3510 return features & ~NETIF_F_GSO_MASK;
3511 }
3512
3513 /* Support for GSO partial features requires software
3514 * intervention before we can actually process the packets
3515 * so we need to strip support for any partial features now
3516 * and we can pull them back in after we have partially
3517 * segmented the frame.
3518 */
3519 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3520 features &= ~dev->gso_partial_features;
3521
3522 /* Make sure to clear the IPv4 ID mangling feature if the
3523 * IPv4 header has the potential to be fragmented.
3524 */
3525 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3526 struct iphdr *iph = skb->encapsulation ?
3527 inner_ip_hdr(skb) : ip_hdr(skb);
3528
3529 if (!(iph->frag_off & htons(IP_DF)))
3530 features &= ~NETIF_F_TSO_MANGLEID;
3531 }
3532
3533 return features;
3534 }
3535
netif_skb_features(struct sk_buff * skb)3536 netdev_features_t netif_skb_features(struct sk_buff *skb)
3537 {
3538 struct net_device *dev = skb->dev;
3539 netdev_features_t features = dev->features;
3540
3541 if (skb_is_gso(skb))
3542 features = gso_features_check(skb, dev, features);
3543
3544 /* If encapsulation offload request, verify we are testing
3545 * hardware encapsulation features instead of standard
3546 * features for the netdev
3547 */
3548 if (skb->encapsulation)
3549 features &= dev->hw_enc_features;
3550
3551 if (skb_vlan_tagged(skb))
3552 features = netdev_intersect_features(features,
3553 dev->vlan_features |
3554 NETIF_F_HW_VLAN_CTAG_TX |
3555 NETIF_F_HW_VLAN_STAG_TX);
3556
3557 if (dev->netdev_ops->ndo_features_check)
3558 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3559 features);
3560 else
3561 features &= dflt_features_check(skb, dev, features);
3562
3563 return harmonize_features(skb, features);
3564 }
3565 EXPORT_SYMBOL(netif_skb_features);
3566
xmit_one(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq,bool more)3567 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3568 struct netdev_queue *txq, bool more)
3569 {
3570 unsigned int len;
3571 int rc;
3572
3573 if (dev_nit_active(dev))
3574 dev_queue_xmit_nit(skb, dev);
3575
3576 len = skb->len;
3577 trace_net_dev_start_xmit(skb, dev);
3578 rc = netdev_start_xmit(skb, dev, txq, more);
3579 trace_net_dev_xmit(skb, rc, dev, len);
3580
3581 return rc;
3582 }
3583
dev_hard_start_xmit(struct sk_buff * first,struct net_device * dev,struct netdev_queue * txq,int * ret)3584 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3585 struct netdev_queue *txq, int *ret)
3586 {
3587 struct sk_buff *skb = first;
3588 int rc = NETDEV_TX_OK;
3589
3590 while (skb) {
3591 struct sk_buff *next = skb->next;
3592
3593 skb_mark_not_on_list(skb);
3594 rc = xmit_one(skb, dev, txq, next != NULL);
3595 if (unlikely(!dev_xmit_complete(rc))) {
3596 skb->next = next;
3597 goto out;
3598 }
3599
3600 skb = next;
3601 if (netif_tx_queue_stopped(txq) && skb) {
3602 rc = NETDEV_TX_BUSY;
3603 break;
3604 }
3605 }
3606
3607 out:
3608 *ret = rc;
3609 return skb;
3610 }
3611
validate_xmit_vlan(struct sk_buff * skb,netdev_features_t features)3612 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3613 netdev_features_t features)
3614 {
3615 if (skb_vlan_tag_present(skb) &&
3616 !vlan_hw_offload_capable(features, skb->vlan_proto))
3617 skb = __vlan_hwaccel_push_inside(skb);
3618 return skb;
3619 }
3620
skb_csum_hwoffload_help(struct sk_buff * skb,const netdev_features_t features)3621 int skb_csum_hwoffload_help(struct sk_buff *skb,
3622 const netdev_features_t features)
3623 {
3624 if (unlikely(skb_csum_is_sctp(skb)))
3625 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3626 skb_crc32c_csum_help(skb);
3627
3628 if (features & NETIF_F_HW_CSUM)
3629 return 0;
3630
3631 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3632 switch (skb->csum_offset) {
3633 case offsetof(struct tcphdr, check):
3634 case offsetof(struct udphdr, check):
3635 return 0;
3636 }
3637 }
3638
3639 return skb_checksum_help(skb);
3640 }
3641 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3642
validate_xmit_skb(struct sk_buff * skb,struct net_device * dev,bool * again)3643 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3644 {
3645 netdev_features_t features;
3646
3647 features = netif_skb_features(skb);
3648 skb = validate_xmit_vlan(skb, features);
3649 if (unlikely(!skb))
3650 goto out_null;
3651
3652 skb = sk_validate_xmit_skb(skb, dev);
3653 if (unlikely(!skb))
3654 goto out_null;
3655
3656 if (netif_needs_gso(skb, features)) {
3657 struct sk_buff *segs;
3658
3659 segs = skb_gso_segment(skb, features);
3660 if (IS_ERR(segs)) {
3661 goto out_kfree_skb;
3662 } else if (segs) {
3663 consume_skb(skb);
3664 skb = segs;
3665 }
3666 } else {
3667 if (skb_needs_linearize(skb, features) &&
3668 __skb_linearize(skb))
3669 goto out_kfree_skb;
3670
3671 /* If packet is not checksummed and device does not
3672 * support checksumming for this protocol, complete
3673 * checksumming here.
3674 */
3675 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3676 if (skb->encapsulation)
3677 skb_set_inner_transport_header(skb,
3678 skb_checksum_start_offset(skb));
3679 else
3680 skb_set_transport_header(skb,
3681 skb_checksum_start_offset(skb));
3682 if (skb_csum_hwoffload_help(skb, features))
3683 goto out_kfree_skb;
3684 }
3685 }
3686
3687 skb = validate_xmit_xfrm(skb, features, again);
3688
3689 return skb;
3690
3691 out_kfree_skb:
3692 kfree_skb(skb);
3693 out_null:
3694 dev_core_stats_tx_dropped_inc(dev);
3695 return NULL;
3696 }
3697
validate_xmit_skb_list(struct sk_buff * skb,struct net_device * dev,bool * again)3698 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3699 {
3700 struct sk_buff *next, *head = NULL, *tail;
3701
3702 for (; skb != NULL; skb = next) {
3703 next = skb->next;
3704 skb_mark_not_on_list(skb);
3705
3706 /* in case skb wont be segmented, point to itself */
3707 skb->prev = skb;
3708
3709 skb = validate_xmit_skb(skb, dev, again);
3710 if (!skb)
3711 continue;
3712
3713 if (!head)
3714 head = skb;
3715 else
3716 tail->next = skb;
3717 /* If skb was segmented, skb->prev points to
3718 * the last segment. If not, it still contains skb.
3719 */
3720 tail = skb->prev;
3721 }
3722 return head;
3723 }
3724 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3725
qdisc_pkt_len_init(struct sk_buff * skb)3726 static void qdisc_pkt_len_init(struct sk_buff *skb)
3727 {
3728 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3729
3730 qdisc_skb_cb(skb)->pkt_len = skb->len;
3731
3732 /* To get more precise estimation of bytes sent on wire,
3733 * we add to pkt_len the headers size of all segments
3734 */
3735 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3736 u16 gso_segs = shinfo->gso_segs;
3737 unsigned int hdr_len;
3738
3739 /* mac layer + network layer */
3740 hdr_len = skb_transport_offset(skb);
3741
3742 /* + transport layer */
3743 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3744 const struct tcphdr *th;
3745 struct tcphdr _tcphdr;
3746
3747 th = skb_header_pointer(skb, hdr_len,
3748 sizeof(_tcphdr), &_tcphdr);
3749 if (likely(th))
3750 hdr_len += __tcp_hdrlen(th);
3751 } else {
3752 struct udphdr _udphdr;
3753
3754 if (skb_header_pointer(skb, hdr_len,
3755 sizeof(_udphdr), &_udphdr))
3756 hdr_len += sizeof(struct udphdr);
3757 }
3758
3759 if (shinfo->gso_type & SKB_GSO_DODGY)
3760 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3761 shinfo->gso_size);
3762
3763 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3764 }
3765 }
3766
dev_qdisc_enqueue(struct sk_buff * skb,struct Qdisc * q,struct sk_buff ** to_free,struct netdev_queue * txq)3767 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3768 struct sk_buff **to_free,
3769 struct netdev_queue *txq)
3770 {
3771 int rc;
3772
3773 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3774 if (rc == NET_XMIT_SUCCESS)
3775 trace_qdisc_enqueue(q, txq, skb);
3776 return rc;
3777 }
3778
__dev_xmit_skb(struct sk_buff * skb,struct Qdisc * q,struct net_device * dev,struct netdev_queue * txq)3779 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3780 struct net_device *dev,
3781 struct netdev_queue *txq)
3782 {
3783 spinlock_t *root_lock = qdisc_lock(q);
3784 struct sk_buff *to_free = NULL;
3785 bool contended;
3786 int rc;
3787
3788 qdisc_calculate_pkt_len(skb, q);
3789
3790 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP);
3791
3792 if (q->flags & TCQ_F_NOLOCK) {
3793 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3794 qdisc_run_begin(q)) {
3795 /* Retest nolock_qdisc_is_empty() within the protection
3796 * of q->seqlock to protect from racing with requeuing.
3797 */
3798 if (unlikely(!nolock_qdisc_is_empty(q))) {
3799 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3800 __qdisc_run(q);
3801 qdisc_run_end(q);
3802
3803 goto no_lock_out;
3804 }
3805
3806 qdisc_bstats_cpu_update(q, skb);
3807 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3808 !nolock_qdisc_is_empty(q))
3809 __qdisc_run(q);
3810
3811 qdisc_run_end(q);
3812 return NET_XMIT_SUCCESS;
3813 }
3814
3815 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3816 qdisc_run(q);
3817
3818 no_lock_out:
3819 if (unlikely(to_free))
3820 kfree_skb_list_reason(to_free,
3821 tcf_get_drop_reason(to_free));
3822 return rc;
3823 }
3824
3825 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) {
3826 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
3827 return NET_XMIT_DROP;
3828 }
3829 /*
3830 * Heuristic to force contended enqueues to serialize on a
3831 * separate lock before trying to get qdisc main lock.
3832 * This permits qdisc->running owner to get the lock more
3833 * often and dequeue packets faster.
3834 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3835 * and then other tasks will only enqueue packets. The packets will be
3836 * sent after the qdisc owner is scheduled again. To prevent this
3837 * scenario the task always serialize on the lock.
3838 */
3839 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3840 if (unlikely(contended))
3841 spin_lock(&q->busylock);
3842
3843 spin_lock(root_lock);
3844 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3845 __qdisc_drop(skb, &to_free);
3846 rc = NET_XMIT_DROP;
3847 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3848 qdisc_run_begin(q)) {
3849 /*
3850 * This is a work-conserving queue; there are no old skbs
3851 * waiting to be sent out; and the qdisc is not running -
3852 * xmit the skb directly.
3853 */
3854
3855 qdisc_bstats_update(q, skb);
3856
3857 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3858 if (unlikely(contended)) {
3859 spin_unlock(&q->busylock);
3860 contended = false;
3861 }
3862 __qdisc_run(q);
3863 }
3864
3865 qdisc_run_end(q);
3866 rc = NET_XMIT_SUCCESS;
3867 } else {
3868 WRITE_ONCE(q->owner, smp_processor_id());
3869 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3870 WRITE_ONCE(q->owner, -1);
3871 if (qdisc_run_begin(q)) {
3872 if (unlikely(contended)) {
3873 spin_unlock(&q->busylock);
3874 contended = false;
3875 }
3876 __qdisc_run(q);
3877 qdisc_run_end(q);
3878 }
3879 }
3880 spin_unlock(root_lock);
3881 if (unlikely(to_free))
3882 kfree_skb_list_reason(to_free,
3883 tcf_get_drop_reason(to_free));
3884 if (unlikely(contended))
3885 spin_unlock(&q->busylock);
3886 return rc;
3887 }
3888
3889 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
skb_update_prio(struct sk_buff * skb)3890 static void skb_update_prio(struct sk_buff *skb)
3891 {
3892 const struct netprio_map *map;
3893 const struct sock *sk;
3894 unsigned int prioidx;
3895
3896 if (skb->priority)
3897 return;
3898 map = rcu_dereference_bh(skb->dev->priomap);
3899 if (!map)
3900 return;
3901 sk = skb_to_full_sk(skb);
3902 if (!sk)
3903 return;
3904
3905 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3906
3907 if (prioidx < map->priomap_len)
3908 skb->priority = map->priomap[prioidx];
3909 }
3910 #else
3911 #define skb_update_prio(skb)
3912 #endif
3913
3914 /**
3915 * dev_loopback_xmit - loop back @skb
3916 * @net: network namespace this loopback is happening in
3917 * @sk: sk needed to be a netfilter okfn
3918 * @skb: buffer to transmit
3919 */
dev_loopback_xmit(struct net * net,struct sock * sk,struct sk_buff * skb)3920 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3921 {
3922 skb_reset_mac_header(skb);
3923 __skb_pull(skb, skb_network_offset(skb));
3924 skb->pkt_type = PACKET_LOOPBACK;
3925 if (skb->ip_summed == CHECKSUM_NONE)
3926 skb->ip_summed = CHECKSUM_UNNECESSARY;
3927 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3928 skb_dst_force(skb);
3929 netif_rx(skb);
3930 return 0;
3931 }
3932 EXPORT_SYMBOL(dev_loopback_xmit);
3933
3934 #ifdef CONFIG_NET_EGRESS
3935 static struct netdev_queue *
netdev_tx_queue_mapping(struct net_device * dev,struct sk_buff * skb)3936 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3937 {
3938 int qm = skb_get_queue_mapping(skb);
3939
3940 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3941 }
3942
netdev_xmit_txqueue_skipped(void)3943 static bool netdev_xmit_txqueue_skipped(void)
3944 {
3945 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3946 }
3947
netdev_xmit_skip_txqueue(bool skip)3948 void netdev_xmit_skip_txqueue(bool skip)
3949 {
3950 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3951 }
3952 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3953 #endif /* CONFIG_NET_EGRESS */
3954
3955 #ifdef CONFIG_NET_XGRESS
tc_run(struct tcx_entry * entry,struct sk_buff * skb,enum skb_drop_reason * drop_reason)3956 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb,
3957 enum skb_drop_reason *drop_reason)
3958 {
3959 int ret = TC_ACT_UNSPEC;
3960 #ifdef CONFIG_NET_CLS_ACT
3961 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
3962 struct tcf_result res;
3963
3964 if (!miniq)
3965 return ret;
3966
3967 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) {
3968 if (tcf_block_bypass_sw(miniq->block))
3969 return ret;
3970 }
3971
3972 tc_skb_cb(skb)->mru = 0;
3973 tc_skb_cb(skb)->post_ct = false;
3974 tcf_set_drop_reason(skb, *drop_reason);
3975
3976 mini_qdisc_bstats_cpu_update(miniq, skb);
3977 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
3978 /* Only tcf related quirks below. */
3979 switch (ret) {
3980 case TC_ACT_SHOT:
3981 *drop_reason = tcf_get_drop_reason(skb);
3982 mini_qdisc_qstats_cpu_drop(miniq);
3983 break;
3984 case TC_ACT_OK:
3985 case TC_ACT_RECLASSIFY:
3986 skb->tc_index = TC_H_MIN(res.classid);
3987 break;
3988 }
3989 #endif /* CONFIG_NET_CLS_ACT */
3990 return ret;
3991 }
3992
3993 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
3994
tcx_inc(void)3995 void tcx_inc(void)
3996 {
3997 static_branch_inc(&tcx_needed_key);
3998 }
3999
tcx_dec(void)4000 void tcx_dec(void)
4001 {
4002 static_branch_dec(&tcx_needed_key);
4003 }
4004
4005 static __always_inline enum tcx_action_base
tcx_run(const struct bpf_mprog_entry * entry,struct sk_buff * skb,const bool needs_mac)4006 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
4007 const bool needs_mac)
4008 {
4009 const struct bpf_mprog_fp *fp;
4010 const struct bpf_prog *prog;
4011 int ret = TCX_NEXT;
4012
4013 if (needs_mac)
4014 __skb_push(skb, skb->mac_len);
4015 bpf_mprog_foreach_prog(entry, fp, prog) {
4016 bpf_compute_data_pointers(skb);
4017 ret = bpf_prog_run(prog, skb);
4018 if (ret != TCX_NEXT)
4019 break;
4020 }
4021 if (needs_mac)
4022 __skb_pull(skb, skb->mac_len);
4023 return tcx_action_code(skb, ret);
4024 }
4025
4026 static __always_inline struct sk_buff *
sch_handle_ingress(struct sk_buff * skb,struct packet_type ** pt_prev,int * ret,struct net_device * orig_dev,bool * another)4027 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4028 struct net_device *orig_dev, bool *another)
4029 {
4030 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
4031 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
4032 int sch_ret;
4033
4034 if (!entry)
4035 return skb;
4036 if (*pt_prev) {
4037 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4038 *pt_prev = NULL;
4039 }
4040
4041 qdisc_skb_cb(skb)->pkt_len = skb->len;
4042 tcx_set_ingress(skb, true);
4043
4044 if (static_branch_unlikely(&tcx_needed_key)) {
4045 sch_ret = tcx_run(entry, skb, true);
4046 if (sch_ret != TC_ACT_UNSPEC)
4047 goto ingress_verdict;
4048 }
4049 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4050 ingress_verdict:
4051 switch (sch_ret) {
4052 case TC_ACT_REDIRECT:
4053 /* skb_mac_header check was done by BPF, so we can safely
4054 * push the L2 header back before redirecting to another
4055 * netdev.
4056 */
4057 __skb_push(skb, skb->mac_len);
4058 if (skb_do_redirect(skb) == -EAGAIN) {
4059 __skb_pull(skb, skb->mac_len);
4060 *another = true;
4061 break;
4062 }
4063 *ret = NET_RX_SUCCESS;
4064 return NULL;
4065 case TC_ACT_SHOT:
4066 kfree_skb_reason(skb, drop_reason);
4067 *ret = NET_RX_DROP;
4068 return NULL;
4069 /* used by tc_run */
4070 case TC_ACT_STOLEN:
4071 case TC_ACT_QUEUED:
4072 case TC_ACT_TRAP:
4073 consume_skb(skb);
4074 fallthrough;
4075 case TC_ACT_CONSUMED:
4076 *ret = NET_RX_SUCCESS;
4077 return NULL;
4078 }
4079
4080 return skb;
4081 }
4082
4083 static __always_inline struct sk_buff *
sch_handle_egress(struct sk_buff * skb,int * ret,struct net_device * dev)4084 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4085 {
4086 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4087 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
4088 int sch_ret;
4089
4090 if (!entry)
4091 return skb;
4092
4093 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4094 * already set by the caller.
4095 */
4096 if (static_branch_unlikely(&tcx_needed_key)) {
4097 sch_ret = tcx_run(entry, skb, false);
4098 if (sch_ret != TC_ACT_UNSPEC)
4099 goto egress_verdict;
4100 }
4101 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4102 egress_verdict:
4103 switch (sch_ret) {
4104 case TC_ACT_REDIRECT:
4105 /* No need to push/pop skb's mac_header here on egress! */
4106 skb_do_redirect(skb);
4107 *ret = NET_XMIT_SUCCESS;
4108 return NULL;
4109 case TC_ACT_SHOT:
4110 kfree_skb_reason(skb, drop_reason);
4111 *ret = NET_XMIT_DROP;
4112 return NULL;
4113 /* used by tc_run */
4114 case TC_ACT_STOLEN:
4115 case TC_ACT_QUEUED:
4116 case TC_ACT_TRAP:
4117 consume_skb(skb);
4118 fallthrough;
4119 case TC_ACT_CONSUMED:
4120 *ret = NET_XMIT_SUCCESS;
4121 return NULL;
4122 }
4123
4124 return skb;
4125 }
4126 #else
4127 static __always_inline struct sk_buff *
sch_handle_ingress(struct sk_buff * skb,struct packet_type ** pt_prev,int * ret,struct net_device * orig_dev,bool * another)4128 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4129 struct net_device *orig_dev, bool *another)
4130 {
4131 return skb;
4132 }
4133
4134 static __always_inline struct sk_buff *
sch_handle_egress(struct sk_buff * skb,int * ret,struct net_device * dev)4135 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4136 {
4137 return skb;
4138 }
4139 #endif /* CONFIG_NET_XGRESS */
4140
4141 #ifdef CONFIG_XPS
__get_xps_queue_idx(struct net_device * dev,struct sk_buff * skb,struct xps_dev_maps * dev_maps,unsigned int tci)4142 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4143 struct xps_dev_maps *dev_maps, unsigned int tci)
4144 {
4145 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4146 struct xps_map *map;
4147 int queue_index = -1;
4148
4149 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4150 return queue_index;
4151
4152 tci *= dev_maps->num_tc;
4153 tci += tc;
4154
4155 map = rcu_dereference(dev_maps->attr_map[tci]);
4156 if (map) {
4157 if (map->len == 1)
4158 queue_index = map->queues[0];
4159 else
4160 queue_index = map->queues[reciprocal_scale(
4161 skb_get_hash(skb), map->len)];
4162 if (unlikely(queue_index >= dev->real_num_tx_queues))
4163 queue_index = -1;
4164 }
4165 return queue_index;
4166 }
4167 #endif
4168
get_xps_queue(struct net_device * dev,struct net_device * sb_dev,struct sk_buff * skb)4169 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4170 struct sk_buff *skb)
4171 {
4172 #ifdef CONFIG_XPS
4173 struct xps_dev_maps *dev_maps;
4174 struct sock *sk = skb->sk;
4175 int queue_index = -1;
4176
4177 if (!static_key_false(&xps_needed))
4178 return -1;
4179
4180 rcu_read_lock();
4181 if (!static_key_false(&xps_rxqs_needed))
4182 goto get_cpus_map;
4183
4184 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4185 if (dev_maps) {
4186 int tci = sk_rx_queue_get(sk);
4187
4188 if (tci >= 0)
4189 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4190 tci);
4191 }
4192
4193 get_cpus_map:
4194 if (queue_index < 0) {
4195 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4196 if (dev_maps) {
4197 unsigned int tci = skb->sender_cpu - 1;
4198
4199 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4200 tci);
4201 }
4202 }
4203 rcu_read_unlock();
4204
4205 return queue_index;
4206 #else
4207 return -1;
4208 #endif
4209 }
4210
dev_pick_tx_zero(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4211 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4212 struct net_device *sb_dev)
4213 {
4214 return 0;
4215 }
4216 EXPORT_SYMBOL(dev_pick_tx_zero);
4217
dev_pick_tx_cpu_id(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4218 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4219 struct net_device *sb_dev)
4220 {
4221 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4222 }
4223 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4224
netdev_pick_tx(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4225 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4226 struct net_device *sb_dev)
4227 {
4228 struct sock *sk = skb->sk;
4229 int queue_index = sk_tx_queue_get(sk);
4230
4231 sb_dev = sb_dev ? : dev;
4232
4233 if (queue_index < 0 || skb->ooo_okay ||
4234 queue_index >= dev->real_num_tx_queues) {
4235 int new_index = get_xps_queue(dev, sb_dev, skb);
4236
4237 if (new_index < 0)
4238 new_index = skb_tx_hash(dev, sb_dev, skb);
4239
4240 if (queue_index != new_index && sk &&
4241 sk_fullsock(sk) &&
4242 rcu_access_pointer(sk->sk_dst_cache))
4243 sk_tx_queue_set(sk, new_index);
4244
4245 queue_index = new_index;
4246 }
4247
4248 return queue_index;
4249 }
4250 EXPORT_SYMBOL(netdev_pick_tx);
4251
netdev_core_pick_tx(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4252 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4253 struct sk_buff *skb,
4254 struct net_device *sb_dev)
4255 {
4256 int queue_index = 0;
4257
4258 #ifdef CONFIG_XPS
4259 u32 sender_cpu = skb->sender_cpu - 1;
4260
4261 if (sender_cpu >= (u32)NR_CPUS)
4262 skb->sender_cpu = raw_smp_processor_id() + 1;
4263 #endif
4264
4265 if (dev->real_num_tx_queues != 1) {
4266 const struct net_device_ops *ops = dev->netdev_ops;
4267
4268 if (ops->ndo_select_queue)
4269 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4270 else
4271 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4272
4273 queue_index = netdev_cap_txqueue(dev, queue_index);
4274 }
4275
4276 skb_set_queue_mapping(skb, queue_index);
4277 return netdev_get_tx_queue(dev, queue_index);
4278 }
4279
4280 /**
4281 * __dev_queue_xmit() - transmit a buffer
4282 * @skb: buffer to transmit
4283 * @sb_dev: suboordinate device used for L2 forwarding offload
4284 *
4285 * Queue a buffer for transmission to a network device. The caller must
4286 * have set the device and priority and built the buffer before calling
4287 * this function. The function can be called from an interrupt.
4288 *
4289 * When calling this method, interrupts MUST be enabled. This is because
4290 * the BH enable code must have IRQs enabled so that it will not deadlock.
4291 *
4292 * Regardless of the return value, the skb is consumed, so it is currently
4293 * difficult to retry a send to this method. (You can bump the ref count
4294 * before sending to hold a reference for retry if you are careful.)
4295 *
4296 * Return:
4297 * * 0 - buffer successfully transmitted
4298 * * positive qdisc return code - NET_XMIT_DROP etc.
4299 * * negative errno - other errors
4300 */
__dev_queue_xmit(struct sk_buff * skb,struct net_device * sb_dev)4301 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4302 {
4303 struct net_device *dev = skb->dev;
4304 struct netdev_queue *txq = NULL;
4305 struct Qdisc *q;
4306 int rc = -ENOMEM;
4307 bool again = false;
4308
4309 skb_reset_mac_header(skb);
4310 skb_assert_len(skb);
4311
4312 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4313 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4314
4315 /* Disable soft irqs for various locks below. Also
4316 * stops preemption for RCU.
4317 */
4318 rcu_read_lock_bh();
4319
4320 skb_update_prio(skb);
4321
4322 qdisc_pkt_len_init(skb);
4323 tcx_set_ingress(skb, false);
4324 #ifdef CONFIG_NET_EGRESS
4325 if (static_branch_unlikely(&egress_needed_key)) {
4326 if (nf_hook_egress_active()) {
4327 skb = nf_hook_egress(skb, &rc, dev);
4328 if (!skb)
4329 goto out;
4330 }
4331
4332 netdev_xmit_skip_txqueue(false);
4333
4334 nf_skip_egress(skb, true);
4335 skb = sch_handle_egress(skb, &rc, dev);
4336 if (!skb)
4337 goto out;
4338 nf_skip_egress(skb, false);
4339
4340 if (netdev_xmit_txqueue_skipped())
4341 txq = netdev_tx_queue_mapping(dev, skb);
4342 }
4343 #endif
4344 /* If device/qdisc don't need skb->dst, release it right now while
4345 * its hot in this cpu cache.
4346 */
4347 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4348 skb_dst_drop(skb);
4349 else
4350 skb_dst_force(skb);
4351
4352 if (!txq)
4353 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4354
4355 q = rcu_dereference_bh(txq->qdisc);
4356
4357 trace_net_dev_queue(skb);
4358 if (q->enqueue) {
4359 rc = __dev_xmit_skb(skb, q, dev, txq);
4360 goto out;
4361 }
4362
4363 /* The device has no queue. Common case for software devices:
4364 * loopback, all the sorts of tunnels...
4365
4366 * Really, it is unlikely that netif_tx_lock protection is necessary
4367 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4368 * counters.)
4369 * However, it is possible, that they rely on protection
4370 * made by us here.
4371
4372 * Check this and shot the lock. It is not prone from deadlocks.
4373 *Either shot noqueue qdisc, it is even simpler 8)
4374 */
4375 if (dev->flags & IFF_UP) {
4376 int cpu = smp_processor_id(); /* ok because BHs are off */
4377
4378 /* Other cpus might concurrently change txq->xmit_lock_owner
4379 * to -1 or to their cpu id, but not to our id.
4380 */
4381 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4382 if (dev_xmit_recursion())
4383 goto recursion_alert;
4384
4385 skb = validate_xmit_skb(skb, dev, &again);
4386 if (!skb)
4387 goto out;
4388
4389 HARD_TX_LOCK(dev, txq, cpu);
4390
4391 if (!netif_xmit_stopped(txq)) {
4392 dev_xmit_recursion_inc();
4393 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4394 dev_xmit_recursion_dec();
4395 if (dev_xmit_complete(rc)) {
4396 HARD_TX_UNLOCK(dev, txq);
4397 goto out;
4398 }
4399 }
4400 HARD_TX_UNLOCK(dev, txq);
4401 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4402 dev->name);
4403 } else {
4404 /* Recursion is detected! It is possible,
4405 * unfortunately
4406 */
4407 recursion_alert:
4408 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4409 dev->name);
4410 }
4411 }
4412
4413 rc = -ENETDOWN;
4414 rcu_read_unlock_bh();
4415
4416 dev_core_stats_tx_dropped_inc(dev);
4417 kfree_skb_list(skb);
4418 return rc;
4419 out:
4420 rcu_read_unlock_bh();
4421 return rc;
4422 }
4423 EXPORT_SYMBOL(__dev_queue_xmit);
4424
__dev_direct_xmit(struct sk_buff * skb,u16 queue_id)4425 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4426 {
4427 struct net_device *dev = skb->dev;
4428 struct sk_buff *orig_skb = skb;
4429 struct netdev_queue *txq;
4430 int ret = NETDEV_TX_BUSY;
4431 bool again = false;
4432
4433 if (unlikely(!netif_running(dev) ||
4434 !netif_carrier_ok(dev)))
4435 goto drop;
4436
4437 skb = validate_xmit_skb_list(skb, dev, &again);
4438 if (skb != orig_skb)
4439 goto drop;
4440
4441 skb_set_queue_mapping(skb, queue_id);
4442 txq = skb_get_tx_queue(dev, skb);
4443
4444 local_bh_disable();
4445
4446 dev_xmit_recursion_inc();
4447 HARD_TX_LOCK(dev, txq, smp_processor_id());
4448 if (!netif_xmit_frozen_or_drv_stopped(txq))
4449 ret = netdev_start_xmit(skb, dev, txq, false);
4450 HARD_TX_UNLOCK(dev, txq);
4451 dev_xmit_recursion_dec();
4452
4453 local_bh_enable();
4454 return ret;
4455 drop:
4456 dev_core_stats_tx_dropped_inc(dev);
4457 kfree_skb_list(skb);
4458 return NET_XMIT_DROP;
4459 }
4460 EXPORT_SYMBOL(__dev_direct_xmit);
4461
4462 /*************************************************************************
4463 * Receiver routines
4464 *************************************************************************/
4465 static DEFINE_PER_CPU(struct task_struct *, backlog_napi);
4466
4467 int weight_p __read_mostly = 64; /* old backlog weight */
4468 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4469 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4470
4471 /* Called with irq disabled */
____napi_schedule(struct softnet_data * sd,struct napi_struct * napi)4472 static inline void ____napi_schedule(struct softnet_data *sd,
4473 struct napi_struct *napi)
4474 {
4475 struct task_struct *thread;
4476
4477 lockdep_assert_irqs_disabled();
4478
4479 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4480 /* Paired with smp_mb__before_atomic() in
4481 * napi_enable()/dev_set_threaded().
4482 * Use READ_ONCE() to guarantee a complete
4483 * read on napi->thread. Only call
4484 * wake_up_process() when it's not NULL.
4485 */
4486 thread = READ_ONCE(napi->thread);
4487 if (thread) {
4488 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi))
4489 goto use_local_napi;
4490
4491 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4492 wake_up_process(thread);
4493 return;
4494 }
4495 }
4496
4497 use_local_napi:
4498 list_add_tail(&napi->poll_list, &sd->poll_list);
4499 WRITE_ONCE(napi->list_owner, smp_processor_id());
4500 /* If not called from net_rx_action()
4501 * we have to raise NET_RX_SOFTIRQ.
4502 */
4503 if (!sd->in_net_rx_action)
4504 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4505 }
4506
4507 #ifdef CONFIG_RPS
4508
4509 struct static_key_false rps_needed __read_mostly;
4510 EXPORT_SYMBOL(rps_needed);
4511 struct static_key_false rfs_needed __read_mostly;
4512 EXPORT_SYMBOL(rfs_needed);
4513
4514 static struct rps_dev_flow *
set_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow * rflow,u16 next_cpu)4515 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4516 struct rps_dev_flow *rflow, u16 next_cpu)
4517 {
4518 if (next_cpu < nr_cpu_ids) {
4519 #ifdef CONFIG_RFS_ACCEL
4520 struct netdev_rx_queue *rxqueue;
4521 struct rps_dev_flow_table *flow_table;
4522 struct rps_dev_flow *old_rflow;
4523 u32 flow_id, head;
4524 u16 rxq_index;
4525 int rc;
4526
4527 /* Should we steer this flow to a different hardware queue? */
4528 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4529 !(dev->features & NETIF_F_NTUPLE))
4530 goto out;
4531 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4532 if (rxq_index == skb_get_rx_queue(skb))
4533 goto out;
4534
4535 rxqueue = dev->_rx + rxq_index;
4536 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4537 if (!flow_table)
4538 goto out;
4539 flow_id = skb_get_hash(skb) & flow_table->mask;
4540 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4541 rxq_index, flow_id);
4542 if (rc < 0)
4543 goto out;
4544 old_rflow = rflow;
4545 rflow = &flow_table->flows[flow_id];
4546 WRITE_ONCE(rflow->filter, rc);
4547 if (old_rflow->filter == rc)
4548 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
4549 out:
4550 #endif
4551 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
4552 rps_input_queue_tail_save(&rflow->last_qtail, head);
4553 }
4554
4555 WRITE_ONCE(rflow->cpu, next_cpu);
4556 return rflow;
4557 }
4558
4559 /*
4560 * get_rps_cpu is called from netif_receive_skb and returns the target
4561 * CPU from the RPS map of the receiving queue for a given skb.
4562 * rcu_read_lock must be held on entry.
4563 */
get_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow ** rflowp)4564 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4565 struct rps_dev_flow **rflowp)
4566 {
4567 const struct rps_sock_flow_table *sock_flow_table;
4568 struct netdev_rx_queue *rxqueue = dev->_rx;
4569 struct rps_dev_flow_table *flow_table;
4570 struct rps_map *map;
4571 int cpu = -1;
4572 u32 tcpu;
4573 u32 hash;
4574
4575 if (skb_rx_queue_recorded(skb)) {
4576 u16 index = skb_get_rx_queue(skb);
4577
4578 if (unlikely(index >= dev->real_num_rx_queues)) {
4579 WARN_ONCE(dev->real_num_rx_queues > 1,
4580 "%s received packet on queue %u, but number "
4581 "of RX queues is %u\n",
4582 dev->name, index, dev->real_num_rx_queues);
4583 goto done;
4584 }
4585 rxqueue += index;
4586 }
4587
4588 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4589
4590 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4591 map = rcu_dereference(rxqueue->rps_map);
4592 if (!flow_table && !map)
4593 goto done;
4594
4595 skb_reset_network_header(skb);
4596 hash = skb_get_hash(skb);
4597 if (!hash)
4598 goto done;
4599
4600 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
4601 if (flow_table && sock_flow_table) {
4602 struct rps_dev_flow *rflow;
4603 u32 next_cpu;
4604 u32 ident;
4605
4606 /* First check into global flow table if there is a match.
4607 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4608 */
4609 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4610 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
4611 goto try_rps;
4612
4613 next_cpu = ident & net_hotdata.rps_cpu_mask;
4614
4615 /* OK, now we know there is a match,
4616 * we can look at the local (per receive queue) flow table
4617 */
4618 rflow = &flow_table->flows[hash & flow_table->mask];
4619 tcpu = rflow->cpu;
4620
4621 /*
4622 * If the desired CPU (where last recvmsg was done) is
4623 * different from current CPU (one in the rx-queue flow
4624 * table entry), switch if one of the following holds:
4625 * - Current CPU is unset (>= nr_cpu_ids).
4626 * - Current CPU is offline.
4627 * - The current CPU's queue tail has advanced beyond the
4628 * last packet that was enqueued using this table entry.
4629 * This guarantees that all previous packets for the flow
4630 * have been dequeued, thus preserving in order delivery.
4631 */
4632 if (unlikely(tcpu != next_cpu) &&
4633 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4634 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
4635 rflow->last_qtail)) >= 0)) {
4636 tcpu = next_cpu;
4637 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4638 }
4639
4640 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4641 *rflowp = rflow;
4642 cpu = tcpu;
4643 goto done;
4644 }
4645 }
4646
4647 try_rps:
4648
4649 if (map) {
4650 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4651 if (cpu_online(tcpu)) {
4652 cpu = tcpu;
4653 goto done;
4654 }
4655 }
4656
4657 done:
4658 return cpu;
4659 }
4660
4661 #ifdef CONFIG_RFS_ACCEL
4662
4663 /**
4664 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4665 * @dev: Device on which the filter was set
4666 * @rxq_index: RX queue index
4667 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4668 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4669 *
4670 * Drivers that implement ndo_rx_flow_steer() should periodically call
4671 * this function for each installed filter and remove the filters for
4672 * which it returns %true.
4673 */
rps_may_expire_flow(struct net_device * dev,u16 rxq_index,u32 flow_id,u16 filter_id)4674 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4675 u32 flow_id, u16 filter_id)
4676 {
4677 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4678 struct rps_dev_flow_table *flow_table;
4679 struct rps_dev_flow *rflow;
4680 bool expire = true;
4681 unsigned int cpu;
4682
4683 rcu_read_lock();
4684 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4685 if (flow_table && flow_id <= flow_table->mask) {
4686 rflow = &flow_table->flows[flow_id];
4687 cpu = READ_ONCE(rflow->cpu);
4688 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids &&
4689 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) -
4690 READ_ONCE(rflow->last_qtail)) <
4691 (int)(10 * flow_table->mask)))
4692 expire = false;
4693 }
4694 rcu_read_unlock();
4695 return expire;
4696 }
4697 EXPORT_SYMBOL(rps_may_expire_flow);
4698
4699 #endif /* CONFIG_RFS_ACCEL */
4700
4701 /* Called from hardirq (IPI) context */
rps_trigger_softirq(void * data)4702 static void rps_trigger_softirq(void *data)
4703 {
4704 struct softnet_data *sd = data;
4705
4706 ____napi_schedule(sd, &sd->backlog);
4707 sd->received_rps++;
4708 }
4709
4710 #endif /* CONFIG_RPS */
4711
4712 /* Called from hardirq (IPI) context */
trigger_rx_softirq(void * data)4713 static void trigger_rx_softirq(void *data)
4714 {
4715 struct softnet_data *sd = data;
4716
4717 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4718 smp_store_release(&sd->defer_ipi_scheduled, 0);
4719 }
4720
4721 /*
4722 * After we queued a packet into sd->input_pkt_queue,
4723 * we need to make sure this queue is serviced soon.
4724 *
4725 * - If this is another cpu queue, link it to our rps_ipi_list,
4726 * and make sure we will process rps_ipi_list from net_rx_action().
4727 *
4728 * - If this is our own queue, NAPI schedule our backlog.
4729 * Note that this also raises NET_RX_SOFTIRQ.
4730 */
napi_schedule_rps(struct softnet_data * sd)4731 static void napi_schedule_rps(struct softnet_data *sd)
4732 {
4733 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4734
4735 #ifdef CONFIG_RPS
4736 if (sd != mysd) {
4737 if (use_backlog_threads()) {
4738 __napi_schedule_irqoff(&sd->backlog);
4739 return;
4740 }
4741
4742 sd->rps_ipi_next = mysd->rps_ipi_list;
4743 mysd->rps_ipi_list = sd;
4744
4745 /* If not called from net_rx_action() or napi_threaded_poll()
4746 * we have to raise NET_RX_SOFTIRQ.
4747 */
4748 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4749 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4750 return;
4751 }
4752 #endif /* CONFIG_RPS */
4753 __napi_schedule_irqoff(&mysd->backlog);
4754 }
4755
kick_defer_list_purge(struct softnet_data * sd,unsigned int cpu)4756 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu)
4757 {
4758 unsigned long flags;
4759
4760 if (use_backlog_threads()) {
4761 backlog_lock_irq_save(sd, &flags);
4762
4763 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4764 __napi_schedule_irqoff(&sd->backlog);
4765
4766 backlog_unlock_irq_restore(sd, &flags);
4767
4768 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) {
4769 smp_call_function_single_async(cpu, &sd->defer_csd);
4770 }
4771 }
4772
4773 #ifdef CONFIG_NET_FLOW_LIMIT
4774 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4775 #endif
4776
skb_flow_limit(struct sk_buff * skb,unsigned int qlen)4777 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4778 {
4779 #ifdef CONFIG_NET_FLOW_LIMIT
4780 struct sd_flow_limit *fl;
4781 struct softnet_data *sd;
4782 unsigned int old_flow, new_flow;
4783
4784 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
4785 return false;
4786
4787 sd = this_cpu_ptr(&softnet_data);
4788
4789 rcu_read_lock();
4790 fl = rcu_dereference(sd->flow_limit);
4791 if (fl) {
4792 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4793 old_flow = fl->history[fl->history_head];
4794 fl->history[fl->history_head] = new_flow;
4795
4796 fl->history_head++;
4797 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4798
4799 if (likely(fl->buckets[old_flow]))
4800 fl->buckets[old_flow]--;
4801
4802 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4803 fl->count++;
4804 rcu_read_unlock();
4805 return true;
4806 }
4807 }
4808 rcu_read_unlock();
4809 #endif
4810 return false;
4811 }
4812
4813 /*
4814 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4815 * queue (may be a remote CPU queue).
4816 */
enqueue_to_backlog(struct sk_buff * skb,int cpu,unsigned int * qtail)4817 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4818 unsigned int *qtail)
4819 {
4820 enum skb_drop_reason reason;
4821 struct softnet_data *sd;
4822 unsigned long flags;
4823 unsigned int qlen;
4824 int max_backlog;
4825 u32 tail;
4826
4827 reason = SKB_DROP_REASON_DEV_READY;
4828 if (!netif_running(skb->dev))
4829 goto bad_dev;
4830
4831 reason = SKB_DROP_REASON_CPU_BACKLOG;
4832 sd = &per_cpu(softnet_data, cpu);
4833
4834 qlen = skb_queue_len_lockless(&sd->input_pkt_queue);
4835 max_backlog = READ_ONCE(net_hotdata.max_backlog);
4836 if (unlikely(qlen > max_backlog))
4837 goto cpu_backlog_drop;
4838 backlog_lock_irq_save(sd, &flags);
4839 qlen = skb_queue_len(&sd->input_pkt_queue);
4840 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) {
4841 if (!qlen) {
4842 /* Schedule NAPI for backlog device. We can use
4843 * non atomic operation as we own the queue lock.
4844 */
4845 if (!__test_and_set_bit(NAPI_STATE_SCHED,
4846 &sd->backlog.state))
4847 napi_schedule_rps(sd);
4848 }
4849 __skb_queue_tail(&sd->input_pkt_queue, skb);
4850 tail = rps_input_queue_tail_incr(sd);
4851 backlog_unlock_irq_restore(sd, &flags);
4852
4853 /* save the tail outside of the critical section */
4854 rps_input_queue_tail_save(qtail, tail);
4855 return NET_RX_SUCCESS;
4856 }
4857
4858 backlog_unlock_irq_restore(sd, &flags);
4859
4860 cpu_backlog_drop:
4861 atomic_inc(&sd->dropped);
4862 bad_dev:
4863 dev_core_stats_rx_dropped_inc(skb->dev);
4864 kfree_skb_reason(skb, reason);
4865 return NET_RX_DROP;
4866 }
4867
netif_get_rxqueue(struct sk_buff * skb)4868 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4869 {
4870 struct net_device *dev = skb->dev;
4871 struct netdev_rx_queue *rxqueue;
4872
4873 rxqueue = dev->_rx;
4874
4875 if (skb_rx_queue_recorded(skb)) {
4876 u16 index = skb_get_rx_queue(skb);
4877
4878 if (unlikely(index >= dev->real_num_rx_queues)) {
4879 WARN_ONCE(dev->real_num_rx_queues > 1,
4880 "%s received packet on queue %u, but number "
4881 "of RX queues is %u\n",
4882 dev->name, index, dev->real_num_rx_queues);
4883
4884 return rxqueue; /* Return first rxqueue */
4885 }
4886 rxqueue += index;
4887 }
4888 return rxqueue;
4889 }
4890
bpf_prog_run_generic_xdp(struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4891 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4892 struct bpf_prog *xdp_prog)
4893 {
4894 void *orig_data, *orig_data_end, *hard_start;
4895 struct netdev_rx_queue *rxqueue;
4896 bool orig_bcast, orig_host;
4897 u32 mac_len, frame_sz;
4898 __be16 orig_eth_type;
4899 struct ethhdr *eth;
4900 u32 metalen, act;
4901 int off;
4902
4903 /* The XDP program wants to see the packet starting at the MAC
4904 * header.
4905 */
4906 mac_len = skb->data - skb_mac_header(skb);
4907 hard_start = skb->data - skb_headroom(skb);
4908
4909 /* SKB "head" area always have tailroom for skb_shared_info */
4910 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4911 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4912
4913 rxqueue = netif_get_rxqueue(skb);
4914 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4915 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4916 skb_headlen(skb) + mac_len, true);
4917 if (skb_is_nonlinear(skb)) {
4918 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
4919 xdp_buff_set_frags_flag(xdp);
4920 } else {
4921 xdp_buff_clear_frags_flag(xdp);
4922 }
4923
4924 orig_data_end = xdp->data_end;
4925 orig_data = xdp->data;
4926 eth = (struct ethhdr *)xdp->data;
4927 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4928 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4929 orig_eth_type = eth->h_proto;
4930
4931 act = bpf_prog_run_xdp(xdp_prog, xdp);
4932
4933 /* check if bpf_xdp_adjust_head was used */
4934 off = xdp->data - orig_data;
4935 if (off) {
4936 if (off > 0)
4937 __skb_pull(skb, off);
4938 else if (off < 0)
4939 __skb_push(skb, -off);
4940
4941 skb->mac_header += off;
4942 skb_reset_network_header(skb);
4943 }
4944
4945 /* check if bpf_xdp_adjust_tail was used */
4946 off = xdp->data_end - orig_data_end;
4947 if (off != 0) {
4948 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4949 skb->len += off; /* positive on grow, negative on shrink */
4950 }
4951
4952 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
4953 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
4954 */
4955 if (xdp_buff_has_frags(xdp))
4956 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
4957 else
4958 skb->data_len = 0;
4959
4960 /* check if XDP changed eth hdr such SKB needs update */
4961 eth = (struct ethhdr *)xdp->data;
4962 if ((orig_eth_type != eth->h_proto) ||
4963 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4964 skb->dev->dev_addr)) ||
4965 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4966 __skb_push(skb, ETH_HLEN);
4967 skb->pkt_type = PACKET_HOST;
4968 skb->protocol = eth_type_trans(skb, skb->dev);
4969 }
4970
4971 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4972 * before calling us again on redirect path. We do not call do_redirect
4973 * as we leave that up to the caller.
4974 *
4975 * Caller is responsible for managing lifetime of skb (i.e. calling
4976 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4977 */
4978 switch (act) {
4979 case XDP_REDIRECT:
4980 case XDP_TX:
4981 __skb_push(skb, mac_len);
4982 break;
4983 case XDP_PASS:
4984 metalen = xdp->data - xdp->data_meta;
4985 if (metalen)
4986 skb_metadata_set(skb, metalen);
4987 break;
4988 }
4989
4990 return act;
4991 }
4992
4993 static int
netif_skb_check_for_xdp(struct sk_buff ** pskb,struct bpf_prog * prog)4994 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog)
4995 {
4996 struct sk_buff *skb = *pskb;
4997 int err, hroom, troom;
4998
4999 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog))
5000 return 0;
5001
5002 /* In case we have to go down the path and also linearize,
5003 * then lets do the pskb_expand_head() work just once here.
5004 */
5005 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
5006 troom = skb->tail + skb->data_len - skb->end;
5007 err = pskb_expand_head(skb,
5008 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
5009 troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
5010 if (err)
5011 return err;
5012
5013 return skb_linearize(skb);
5014 }
5015
netif_receive_generic_xdp(struct sk_buff ** pskb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)5016 static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
5017 struct xdp_buff *xdp,
5018 struct bpf_prog *xdp_prog)
5019 {
5020 struct sk_buff *skb = *pskb;
5021 u32 mac_len, act = XDP_DROP;
5022
5023 /* Reinjected packets coming from act_mirred or similar should
5024 * not get XDP generic processing.
5025 */
5026 if (skb_is_redirected(skb))
5027 return XDP_PASS;
5028
5029 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
5030 * bytes. This is the guarantee that also native XDP provides,
5031 * thus we need to do it here as well.
5032 */
5033 mac_len = skb->data - skb_mac_header(skb);
5034 __skb_push(skb, mac_len);
5035
5036 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
5037 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
5038 if (netif_skb_check_for_xdp(pskb, xdp_prog))
5039 goto do_drop;
5040 }
5041
5042 __skb_pull(*pskb, mac_len);
5043
5044 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
5045 switch (act) {
5046 case XDP_REDIRECT:
5047 case XDP_TX:
5048 case XDP_PASS:
5049 break;
5050 default:
5051 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
5052 fallthrough;
5053 case XDP_ABORTED:
5054 trace_xdp_exception((*pskb)->dev, xdp_prog, act);
5055 fallthrough;
5056 case XDP_DROP:
5057 do_drop:
5058 kfree_skb(*pskb);
5059 break;
5060 }
5061
5062 return act;
5063 }
5064
5065 /* When doing generic XDP we have to bypass the qdisc layer and the
5066 * network taps in order to match in-driver-XDP behavior. This also means
5067 * that XDP packets are able to starve other packets going through a qdisc,
5068 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
5069 * queues, so they do not have this starvation issue.
5070 */
generic_xdp_tx(struct sk_buff * skb,struct bpf_prog * xdp_prog)5071 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
5072 {
5073 struct net_device *dev = skb->dev;
5074 struct netdev_queue *txq;
5075 bool free_skb = true;
5076 int cpu, rc;
5077
5078 txq = netdev_core_pick_tx(dev, skb, NULL);
5079 cpu = smp_processor_id();
5080 HARD_TX_LOCK(dev, txq, cpu);
5081 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
5082 rc = netdev_start_xmit(skb, dev, txq, 0);
5083 if (dev_xmit_complete(rc))
5084 free_skb = false;
5085 }
5086 HARD_TX_UNLOCK(dev, txq);
5087 if (free_skb) {
5088 trace_xdp_exception(dev, xdp_prog, XDP_TX);
5089 dev_core_stats_tx_dropped_inc(dev);
5090 kfree_skb(skb);
5091 }
5092 }
5093
5094 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
5095
do_xdp_generic(struct bpf_prog * xdp_prog,struct sk_buff ** pskb)5096 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb)
5097 {
5098 if (xdp_prog) {
5099 struct xdp_buff xdp;
5100 u32 act;
5101 int err;
5102
5103 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
5104 if (act != XDP_PASS) {
5105 switch (act) {
5106 case XDP_REDIRECT:
5107 err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
5108 &xdp, xdp_prog);
5109 if (err)
5110 goto out_redir;
5111 break;
5112 case XDP_TX:
5113 generic_xdp_tx(*pskb, xdp_prog);
5114 break;
5115 }
5116 return XDP_DROP;
5117 }
5118 }
5119 return XDP_PASS;
5120 out_redir:
5121 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
5122 return XDP_DROP;
5123 }
5124 EXPORT_SYMBOL_GPL(do_xdp_generic);
5125
netif_rx_internal(struct sk_buff * skb)5126 static int netif_rx_internal(struct sk_buff *skb)
5127 {
5128 int ret;
5129
5130 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5131
5132 trace_netif_rx(skb);
5133
5134 #ifdef CONFIG_RPS
5135 if (static_branch_unlikely(&rps_needed)) {
5136 struct rps_dev_flow voidflow, *rflow = &voidflow;
5137 int cpu;
5138
5139 rcu_read_lock();
5140
5141 cpu = get_rps_cpu(skb->dev, skb, &rflow);
5142 if (cpu < 0)
5143 cpu = smp_processor_id();
5144
5145 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5146
5147 rcu_read_unlock();
5148 } else
5149 #endif
5150 {
5151 unsigned int qtail;
5152
5153 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5154 }
5155 return ret;
5156 }
5157
5158 /**
5159 * __netif_rx - Slightly optimized version of netif_rx
5160 * @skb: buffer to post
5161 *
5162 * This behaves as netif_rx except that it does not disable bottom halves.
5163 * As a result this function may only be invoked from the interrupt context
5164 * (either hard or soft interrupt).
5165 */
__netif_rx(struct sk_buff * skb)5166 int __netif_rx(struct sk_buff *skb)
5167 {
5168 int ret;
5169
5170 lockdep_assert_once(hardirq_count() | softirq_count());
5171
5172 trace_netif_rx_entry(skb);
5173 ret = netif_rx_internal(skb);
5174 trace_netif_rx_exit(ret);
5175 return ret;
5176 }
5177 EXPORT_SYMBOL(__netif_rx);
5178
5179 /**
5180 * netif_rx - post buffer to the network code
5181 * @skb: buffer to post
5182 *
5183 * This function receives a packet from a device driver and queues it for
5184 * the upper (protocol) levels to process via the backlog NAPI device. It
5185 * always succeeds. The buffer may be dropped during processing for
5186 * congestion control or by the protocol layers.
5187 * The network buffer is passed via the backlog NAPI device. Modern NIC
5188 * driver should use NAPI and GRO.
5189 * This function can used from interrupt and from process context. The
5190 * caller from process context must not disable interrupts before invoking
5191 * this function.
5192 *
5193 * return values:
5194 * NET_RX_SUCCESS (no congestion)
5195 * NET_RX_DROP (packet was dropped)
5196 *
5197 */
netif_rx(struct sk_buff * skb)5198 int netif_rx(struct sk_buff *skb)
5199 {
5200 bool need_bh_off = !(hardirq_count() | softirq_count());
5201 int ret;
5202
5203 if (need_bh_off)
5204 local_bh_disable();
5205 trace_netif_rx_entry(skb);
5206 ret = netif_rx_internal(skb);
5207 trace_netif_rx_exit(ret);
5208 if (need_bh_off)
5209 local_bh_enable();
5210 return ret;
5211 }
5212 EXPORT_SYMBOL(netif_rx);
5213
net_tx_action(struct softirq_action * h)5214 static __latent_entropy void net_tx_action(struct softirq_action *h)
5215 {
5216 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5217
5218 if (sd->completion_queue) {
5219 struct sk_buff *clist;
5220
5221 local_irq_disable();
5222 clist = sd->completion_queue;
5223 sd->completion_queue = NULL;
5224 local_irq_enable();
5225
5226 while (clist) {
5227 struct sk_buff *skb = clist;
5228
5229 clist = clist->next;
5230
5231 WARN_ON(refcount_read(&skb->users));
5232 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5233 trace_consume_skb(skb, net_tx_action);
5234 else
5235 trace_kfree_skb(skb, net_tx_action,
5236 get_kfree_skb_cb(skb)->reason);
5237
5238 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5239 __kfree_skb(skb);
5240 else
5241 __napi_kfree_skb(skb,
5242 get_kfree_skb_cb(skb)->reason);
5243 }
5244 }
5245
5246 if (sd->output_queue) {
5247 struct Qdisc *head;
5248
5249 local_irq_disable();
5250 head = sd->output_queue;
5251 sd->output_queue = NULL;
5252 sd->output_queue_tailp = &sd->output_queue;
5253 local_irq_enable();
5254
5255 rcu_read_lock();
5256
5257 while (head) {
5258 struct Qdisc *q = head;
5259 spinlock_t *root_lock = NULL;
5260
5261 head = head->next_sched;
5262
5263 /* We need to make sure head->next_sched is read
5264 * before clearing __QDISC_STATE_SCHED
5265 */
5266 smp_mb__before_atomic();
5267
5268 if (!(q->flags & TCQ_F_NOLOCK)) {
5269 root_lock = qdisc_lock(q);
5270 spin_lock(root_lock);
5271 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5272 &q->state))) {
5273 /* There is a synchronize_net() between
5274 * STATE_DEACTIVATED flag being set and
5275 * qdisc_reset()/some_qdisc_is_busy() in
5276 * dev_deactivate(), so we can safely bail out
5277 * early here to avoid data race between
5278 * qdisc_deactivate() and some_qdisc_is_busy()
5279 * for lockless qdisc.
5280 */
5281 clear_bit(__QDISC_STATE_SCHED, &q->state);
5282 continue;
5283 }
5284
5285 clear_bit(__QDISC_STATE_SCHED, &q->state);
5286 qdisc_run(q);
5287 if (root_lock)
5288 spin_unlock(root_lock);
5289 }
5290
5291 rcu_read_unlock();
5292 }
5293
5294 xfrm_dev_backlog(sd);
5295 }
5296
5297 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5298 /* This hook is defined here for ATM LANE */
5299 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5300 unsigned char *addr) __read_mostly;
5301 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5302 #endif
5303
5304 /**
5305 * netdev_is_rx_handler_busy - check if receive handler is registered
5306 * @dev: device to check
5307 *
5308 * Check if a receive handler is already registered for a given device.
5309 * Return true if there one.
5310 *
5311 * The caller must hold the rtnl_mutex.
5312 */
netdev_is_rx_handler_busy(struct net_device * dev)5313 bool netdev_is_rx_handler_busy(struct net_device *dev)
5314 {
5315 ASSERT_RTNL();
5316 return dev && rtnl_dereference(dev->rx_handler);
5317 }
5318 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5319
5320 /**
5321 * netdev_rx_handler_register - register receive handler
5322 * @dev: device to register a handler for
5323 * @rx_handler: receive handler to register
5324 * @rx_handler_data: data pointer that is used by rx handler
5325 *
5326 * Register a receive handler for a device. This handler will then be
5327 * called from __netif_receive_skb. A negative errno code is returned
5328 * on a failure.
5329 *
5330 * The caller must hold the rtnl_mutex.
5331 *
5332 * For a general description of rx_handler, see enum rx_handler_result.
5333 */
netdev_rx_handler_register(struct net_device * dev,rx_handler_func_t * rx_handler,void * rx_handler_data)5334 int netdev_rx_handler_register(struct net_device *dev,
5335 rx_handler_func_t *rx_handler,
5336 void *rx_handler_data)
5337 {
5338 if (netdev_is_rx_handler_busy(dev))
5339 return -EBUSY;
5340
5341 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5342 return -EINVAL;
5343
5344 /* Note: rx_handler_data must be set before rx_handler */
5345 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5346 rcu_assign_pointer(dev->rx_handler, rx_handler);
5347
5348 return 0;
5349 }
5350 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5351
5352 /**
5353 * netdev_rx_handler_unregister - unregister receive handler
5354 * @dev: device to unregister a handler from
5355 *
5356 * Unregister a receive handler from a device.
5357 *
5358 * The caller must hold the rtnl_mutex.
5359 */
netdev_rx_handler_unregister(struct net_device * dev)5360 void netdev_rx_handler_unregister(struct net_device *dev)
5361 {
5362
5363 ASSERT_RTNL();
5364 RCU_INIT_POINTER(dev->rx_handler, NULL);
5365 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5366 * section has a guarantee to see a non NULL rx_handler_data
5367 * as well.
5368 */
5369 synchronize_net();
5370 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5371 }
5372 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5373
5374 /*
5375 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5376 * the special handling of PFMEMALLOC skbs.
5377 */
skb_pfmemalloc_protocol(struct sk_buff * skb)5378 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5379 {
5380 switch (skb->protocol) {
5381 case htons(ETH_P_ARP):
5382 case htons(ETH_P_IP):
5383 case htons(ETH_P_IPV6):
5384 case htons(ETH_P_8021Q):
5385 case htons(ETH_P_8021AD):
5386 return true;
5387 default:
5388 return false;
5389 }
5390 }
5391
nf_ingress(struct sk_buff * skb,struct packet_type ** pt_prev,int * ret,struct net_device * orig_dev)5392 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5393 int *ret, struct net_device *orig_dev)
5394 {
5395 if (nf_hook_ingress_active(skb)) {
5396 int ingress_retval;
5397
5398 if (*pt_prev) {
5399 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5400 *pt_prev = NULL;
5401 }
5402
5403 rcu_read_lock();
5404 ingress_retval = nf_hook_ingress(skb);
5405 rcu_read_unlock();
5406 return ingress_retval;
5407 }
5408 return 0;
5409 }
5410
__netif_receive_skb_core(struct sk_buff ** pskb,bool pfmemalloc,struct packet_type ** ppt_prev)5411 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5412 struct packet_type **ppt_prev)
5413 {
5414 struct packet_type *ptype, *pt_prev;
5415 rx_handler_func_t *rx_handler;
5416 struct sk_buff *skb = *pskb;
5417 struct net_device *orig_dev;
5418 bool deliver_exact = false;
5419 int ret = NET_RX_DROP;
5420 __be16 type;
5421
5422 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5423
5424 trace_netif_receive_skb(skb);
5425
5426 orig_dev = skb->dev;
5427
5428 skb_reset_network_header(skb);
5429 if (!skb_transport_header_was_set(skb))
5430 skb_reset_transport_header(skb);
5431 skb_reset_mac_len(skb);
5432
5433 pt_prev = NULL;
5434
5435 another_round:
5436 skb->skb_iif = skb->dev->ifindex;
5437
5438 __this_cpu_inc(softnet_data.processed);
5439
5440 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5441 int ret2;
5442
5443 migrate_disable();
5444 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
5445 &skb);
5446 migrate_enable();
5447
5448 if (ret2 != XDP_PASS) {
5449 ret = NET_RX_DROP;
5450 goto out;
5451 }
5452 }
5453
5454 if (eth_type_vlan(skb->protocol)) {
5455 skb = skb_vlan_untag(skb);
5456 if (unlikely(!skb))
5457 goto out;
5458 }
5459
5460 if (skb_skip_tc_classify(skb))
5461 goto skip_classify;
5462
5463 if (pfmemalloc)
5464 goto skip_taps;
5465
5466 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) {
5467 if (pt_prev)
5468 ret = deliver_skb(skb, pt_prev, orig_dev);
5469 pt_prev = ptype;
5470 }
5471
5472 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5473 if (pt_prev)
5474 ret = deliver_skb(skb, pt_prev, orig_dev);
5475 pt_prev = ptype;
5476 }
5477
5478 skip_taps:
5479 #ifdef CONFIG_NET_INGRESS
5480 if (static_branch_unlikely(&ingress_needed_key)) {
5481 bool another = false;
5482
5483 nf_skip_egress(skb, true);
5484 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5485 &another);
5486 if (another)
5487 goto another_round;
5488 if (!skb)
5489 goto out;
5490
5491 nf_skip_egress(skb, false);
5492 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5493 goto out;
5494 }
5495 #endif
5496 skb_reset_redirect(skb);
5497 skip_classify:
5498 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5499 goto drop;
5500
5501 if (skb_vlan_tag_present(skb)) {
5502 if (pt_prev) {
5503 ret = deliver_skb(skb, pt_prev, orig_dev);
5504 pt_prev = NULL;
5505 }
5506 if (vlan_do_receive(&skb))
5507 goto another_round;
5508 else if (unlikely(!skb))
5509 goto out;
5510 }
5511
5512 rx_handler = rcu_dereference(skb->dev->rx_handler);
5513 if (rx_handler) {
5514 if (pt_prev) {
5515 ret = deliver_skb(skb, pt_prev, orig_dev);
5516 pt_prev = NULL;
5517 }
5518 switch (rx_handler(&skb)) {
5519 case RX_HANDLER_CONSUMED:
5520 ret = NET_RX_SUCCESS;
5521 goto out;
5522 case RX_HANDLER_ANOTHER:
5523 goto another_round;
5524 case RX_HANDLER_EXACT:
5525 deliver_exact = true;
5526 break;
5527 case RX_HANDLER_PASS:
5528 break;
5529 default:
5530 BUG();
5531 }
5532 }
5533
5534 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5535 check_vlan_id:
5536 if (skb_vlan_tag_get_id(skb)) {
5537 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5538 * find vlan device.
5539 */
5540 skb->pkt_type = PACKET_OTHERHOST;
5541 } else if (eth_type_vlan(skb->protocol)) {
5542 /* Outer header is 802.1P with vlan 0, inner header is
5543 * 802.1Q or 802.1AD and vlan_do_receive() above could
5544 * not find vlan dev for vlan id 0.
5545 */
5546 __vlan_hwaccel_clear_tag(skb);
5547 skb = skb_vlan_untag(skb);
5548 if (unlikely(!skb))
5549 goto out;
5550 if (vlan_do_receive(&skb))
5551 /* After stripping off 802.1P header with vlan 0
5552 * vlan dev is found for inner header.
5553 */
5554 goto another_round;
5555 else if (unlikely(!skb))
5556 goto out;
5557 else
5558 /* We have stripped outer 802.1P vlan 0 header.
5559 * But could not find vlan dev.
5560 * check again for vlan id to set OTHERHOST.
5561 */
5562 goto check_vlan_id;
5563 }
5564 /* Note: we might in the future use prio bits
5565 * and set skb->priority like in vlan_do_receive()
5566 * For the time being, just ignore Priority Code Point
5567 */
5568 __vlan_hwaccel_clear_tag(skb);
5569 }
5570
5571 type = skb->protocol;
5572
5573 /* deliver only exact match when indicated */
5574 if (likely(!deliver_exact)) {
5575 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5576 &ptype_base[ntohs(type) &
5577 PTYPE_HASH_MASK]);
5578 }
5579
5580 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5581 &orig_dev->ptype_specific);
5582
5583 if (unlikely(skb->dev != orig_dev)) {
5584 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5585 &skb->dev->ptype_specific);
5586 }
5587
5588 if (pt_prev) {
5589 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5590 goto drop;
5591 *ppt_prev = pt_prev;
5592 } else {
5593 drop:
5594 if (!deliver_exact)
5595 dev_core_stats_rx_dropped_inc(skb->dev);
5596 else
5597 dev_core_stats_rx_nohandler_inc(skb->dev);
5598 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5599 /* Jamal, now you will not able to escape explaining
5600 * me how you were going to use this. :-)
5601 */
5602 ret = NET_RX_DROP;
5603 }
5604
5605 out:
5606 /* The invariant here is that if *ppt_prev is not NULL
5607 * then skb should also be non-NULL.
5608 *
5609 * Apparently *ppt_prev assignment above holds this invariant due to
5610 * skb dereferencing near it.
5611 */
5612 *pskb = skb;
5613 return ret;
5614 }
5615
__netif_receive_skb_one_core(struct sk_buff * skb,bool pfmemalloc)5616 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5617 {
5618 struct net_device *orig_dev = skb->dev;
5619 struct packet_type *pt_prev = NULL;
5620 int ret;
5621
5622 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5623 if (pt_prev)
5624 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5625 skb->dev, pt_prev, orig_dev);
5626 return ret;
5627 }
5628
5629 /**
5630 * netif_receive_skb_core - special purpose version of netif_receive_skb
5631 * @skb: buffer to process
5632 *
5633 * More direct receive version of netif_receive_skb(). It should
5634 * only be used by callers that have a need to skip RPS and Generic XDP.
5635 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5636 *
5637 * This function may only be called from softirq context and interrupts
5638 * should be enabled.
5639 *
5640 * Return values (usually ignored):
5641 * NET_RX_SUCCESS: no congestion
5642 * NET_RX_DROP: packet was dropped
5643 */
netif_receive_skb_core(struct sk_buff * skb)5644 int netif_receive_skb_core(struct sk_buff *skb)
5645 {
5646 int ret;
5647
5648 rcu_read_lock();
5649 ret = __netif_receive_skb_one_core(skb, false);
5650 rcu_read_unlock();
5651
5652 return ret;
5653 }
5654 EXPORT_SYMBOL(netif_receive_skb_core);
5655
__netif_receive_skb_list_ptype(struct list_head * head,struct packet_type * pt_prev,struct net_device * orig_dev)5656 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5657 struct packet_type *pt_prev,
5658 struct net_device *orig_dev)
5659 {
5660 struct sk_buff *skb, *next;
5661
5662 if (!pt_prev)
5663 return;
5664 if (list_empty(head))
5665 return;
5666 if (pt_prev->list_func != NULL)
5667 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5668 ip_list_rcv, head, pt_prev, orig_dev);
5669 else
5670 list_for_each_entry_safe(skb, next, head, list) {
5671 skb_list_del_init(skb);
5672 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5673 }
5674 }
5675
__netif_receive_skb_list_core(struct list_head * head,bool pfmemalloc)5676 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5677 {
5678 /* Fast-path assumptions:
5679 * - There is no RX handler.
5680 * - Only one packet_type matches.
5681 * If either of these fails, we will end up doing some per-packet
5682 * processing in-line, then handling the 'last ptype' for the whole
5683 * sublist. This can't cause out-of-order delivery to any single ptype,
5684 * because the 'last ptype' must be constant across the sublist, and all
5685 * other ptypes are handled per-packet.
5686 */
5687 /* Current (common) ptype of sublist */
5688 struct packet_type *pt_curr = NULL;
5689 /* Current (common) orig_dev of sublist */
5690 struct net_device *od_curr = NULL;
5691 struct list_head sublist;
5692 struct sk_buff *skb, *next;
5693
5694 INIT_LIST_HEAD(&sublist);
5695 list_for_each_entry_safe(skb, next, head, list) {
5696 struct net_device *orig_dev = skb->dev;
5697 struct packet_type *pt_prev = NULL;
5698
5699 skb_list_del_init(skb);
5700 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5701 if (!pt_prev)
5702 continue;
5703 if (pt_curr != pt_prev || od_curr != orig_dev) {
5704 /* dispatch old sublist */
5705 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5706 /* start new sublist */
5707 INIT_LIST_HEAD(&sublist);
5708 pt_curr = pt_prev;
5709 od_curr = orig_dev;
5710 }
5711 list_add_tail(&skb->list, &sublist);
5712 }
5713
5714 /* dispatch final sublist */
5715 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5716 }
5717
__netif_receive_skb(struct sk_buff * skb)5718 static int __netif_receive_skb(struct sk_buff *skb)
5719 {
5720 int ret;
5721
5722 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5723 unsigned int noreclaim_flag;
5724
5725 /*
5726 * PFMEMALLOC skbs are special, they should
5727 * - be delivered to SOCK_MEMALLOC sockets only
5728 * - stay away from userspace
5729 * - have bounded memory usage
5730 *
5731 * Use PF_MEMALLOC as this saves us from propagating the allocation
5732 * context down to all allocation sites.
5733 */
5734 noreclaim_flag = memalloc_noreclaim_save();
5735 ret = __netif_receive_skb_one_core(skb, true);
5736 memalloc_noreclaim_restore(noreclaim_flag);
5737 } else
5738 ret = __netif_receive_skb_one_core(skb, false);
5739
5740 return ret;
5741 }
5742
__netif_receive_skb_list(struct list_head * head)5743 static void __netif_receive_skb_list(struct list_head *head)
5744 {
5745 unsigned long noreclaim_flag = 0;
5746 struct sk_buff *skb, *next;
5747 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5748
5749 list_for_each_entry_safe(skb, next, head, list) {
5750 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5751 struct list_head sublist;
5752
5753 /* Handle the previous sublist */
5754 list_cut_before(&sublist, head, &skb->list);
5755 if (!list_empty(&sublist))
5756 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5757 pfmemalloc = !pfmemalloc;
5758 /* See comments in __netif_receive_skb */
5759 if (pfmemalloc)
5760 noreclaim_flag = memalloc_noreclaim_save();
5761 else
5762 memalloc_noreclaim_restore(noreclaim_flag);
5763 }
5764 }
5765 /* Handle the remaining sublist */
5766 if (!list_empty(head))
5767 __netif_receive_skb_list_core(head, pfmemalloc);
5768 /* Restore pflags */
5769 if (pfmemalloc)
5770 memalloc_noreclaim_restore(noreclaim_flag);
5771 }
5772
generic_xdp_install(struct net_device * dev,struct netdev_bpf * xdp)5773 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5774 {
5775 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5776 struct bpf_prog *new = xdp->prog;
5777 int ret = 0;
5778
5779 switch (xdp->command) {
5780 case XDP_SETUP_PROG:
5781 rcu_assign_pointer(dev->xdp_prog, new);
5782 if (old)
5783 bpf_prog_put(old);
5784
5785 if (old && !new) {
5786 static_branch_dec(&generic_xdp_needed_key);
5787 } else if (new && !old) {
5788 static_branch_inc(&generic_xdp_needed_key);
5789 dev_disable_lro(dev);
5790 dev_disable_gro_hw(dev);
5791 }
5792 break;
5793
5794 default:
5795 ret = -EINVAL;
5796 break;
5797 }
5798
5799 return ret;
5800 }
5801
netif_receive_skb_internal(struct sk_buff * skb)5802 static int netif_receive_skb_internal(struct sk_buff *skb)
5803 {
5804 int ret;
5805
5806 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5807
5808 if (skb_defer_rx_timestamp(skb))
5809 return NET_RX_SUCCESS;
5810
5811 rcu_read_lock();
5812 #ifdef CONFIG_RPS
5813 if (static_branch_unlikely(&rps_needed)) {
5814 struct rps_dev_flow voidflow, *rflow = &voidflow;
5815 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5816
5817 if (cpu >= 0) {
5818 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5819 rcu_read_unlock();
5820 return ret;
5821 }
5822 }
5823 #endif
5824 ret = __netif_receive_skb(skb);
5825 rcu_read_unlock();
5826 return ret;
5827 }
5828
netif_receive_skb_list_internal(struct list_head * head)5829 void netif_receive_skb_list_internal(struct list_head *head)
5830 {
5831 struct sk_buff *skb, *next;
5832 struct list_head sublist;
5833
5834 INIT_LIST_HEAD(&sublist);
5835 list_for_each_entry_safe(skb, next, head, list) {
5836 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
5837 skb);
5838 skb_list_del_init(skb);
5839 if (!skb_defer_rx_timestamp(skb))
5840 list_add_tail(&skb->list, &sublist);
5841 }
5842 list_splice_init(&sublist, head);
5843
5844 rcu_read_lock();
5845 #ifdef CONFIG_RPS
5846 if (static_branch_unlikely(&rps_needed)) {
5847 list_for_each_entry_safe(skb, next, head, list) {
5848 struct rps_dev_flow voidflow, *rflow = &voidflow;
5849 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5850
5851 if (cpu >= 0) {
5852 /* Will be handled, remove from list */
5853 skb_list_del_init(skb);
5854 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5855 }
5856 }
5857 }
5858 #endif
5859 __netif_receive_skb_list(head);
5860 rcu_read_unlock();
5861 }
5862
5863 /**
5864 * netif_receive_skb - process receive buffer from network
5865 * @skb: buffer to process
5866 *
5867 * netif_receive_skb() is the main receive data processing function.
5868 * It always succeeds. The buffer may be dropped during processing
5869 * for congestion control or by the protocol layers.
5870 *
5871 * This function may only be called from softirq context and interrupts
5872 * should be enabled.
5873 *
5874 * Return values (usually ignored):
5875 * NET_RX_SUCCESS: no congestion
5876 * NET_RX_DROP: packet was dropped
5877 */
netif_receive_skb(struct sk_buff * skb)5878 int netif_receive_skb(struct sk_buff *skb)
5879 {
5880 int ret;
5881
5882 trace_netif_receive_skb_entry(skb);
5883
5884 ret = netif_receive_skb_internal(skb);
5885 trace_netif_receive_skb_exit(ret);
5886
5887 return ret;
5888 }
5889 EXPORT_SYMBOL(netif_receive_skb);
5890
5891 /**
5892 * netif_receive_skb_list - process many receive buffers from network
5893 * @head: list of skbs to process.
5894 *
5895 * Since return value of netif_receive_skb() is normally ignored, and
5896 * wouldn't be meaningful for a list, this function returns void.
5897 *
5898 * This function may only be called from softirq context and interrupts
5899 * should be enabled.
5900 */
netif_receive_skb_list(struct list_head * head)5901 void netif_receive_skb_list(struct list_head *head)
5902 {
5903 struct sk_buff *skb;
5904
5905 if (list_empty(head))
5906 return;
5907 if (trace_netif_receive_skb_list_entry_enabled()) {
5908 list_for_each_entry(skb, head, list)
5909 trace_netif_receive_skb_list_entry(skb);
5910 }
5911 netif_receive_skb_list_internal(head);
5912 trace_netif_receive_skb_list_exit(0);
5913 }
5914 EXPORT_SYMBOL(netif_receive_skb_list);
5915
5916 static DEFINE_PER_CPU(struct work_struct, flush_works);
5917
5918 /* Network device is going away, flush any packets still pending */
flush_backlog(struct work_struct * work)5919 static void flush_backlog(struct work_struct *work)
5920 {
5921 struct sk_buff *skb, *tmp;
5922 struct softnet_data *sd;
5923
5924 local_bh_disable();
5925 sd = this_cpu_ptr(&softnet_data);
5926
5927 backlog_lock_irq_disable(sd);
5928 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5929 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5930 __skb_unlink(skb, &sd->input_pkt_queue);
5931 dev_kfree_skb_irq(skb);
5932 rps_input_queue_head_incr(sd);
5933 }
5934 }
5935 backlog_unlock_irq_enable(sd);
5936
5937 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5938 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5939 __skb_unlink(skb, &sd->process_queue);
5940 kfree_skb(skb);
5941 rps_input_queue_head_incr(sd);
5942 }
5943 }
5944 local_bh_enable();
5945 }
5946
flush_required(int cpu)5947 static bool flush_required(int cpu)
5948 {
5949 #if IS_ENABLED(CONFIG_RPS)
5950 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5951 bool do_flush;
5952
5953 backlog_lock_irq_disable(sd);
5954
5955 /* as insertion into process_queue happens with the rps lock held,
5956 * process_queue access may race only with dequeue
5957 */
5958 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5959 !skb_queue_empty_lockless(&sd->process_queue);
5960 backlog_unlock_irq_enable(sd);
5961
5962 return do_flush;
5963 #endif
5964 /* without RPS we can't safely check input_pkt_queue: during a
5965 * concurrent remote skb_queue_splice() we can detect as empty both
5966 * input_pkt_queue and process_queue even if the latter could end-up
5967 * containing a lot of packets.
5968 */
5969 return true;
5970 }
5971
flush_all_backlogs(void)5972 static void flush_all_backlogs(void)
5973 {
5974 static cpumask_t flush_cpus;
5975 unsigned int cpu;
5976
5977 /* since we are under rtnl lock protection we can use static data
5978 * for the cpumask and avoid allocating on stack the possibly
5979 * large mask
5980 */
5981 ASSERT_RTNL();
5982
5983 cpus_read_lock();
5984
5985 cpumask_clear(&flush_cpus);
5986 for_each_online_cpu(cpu) {
5987 if (flush_required(cpu)) {
5988 queue_work_on(cpu, system_highpri_wq,
5989 per_cpu_ptr(&flush_works, cpu));
5990 cpumask_set_cpu(cpu, &flush_cpus);
5991 }
5992 }
5993
5994 /* we can have in flight packet[s] on the cpus we are not flushing,
5995 * synchronize_net() in unregister_netdevice_many() will take care of
5996 * them
5997 */
5998 for_each_cpu(cpu, &flush_cpus)
5999 flush_work(per_cpu_ptr(&flush_works, cpu));
6000
6001 cpus_read_unlock();
6002 }
6003
net_rps_send_ipi(struct softnet_data * remsd)6004 static void net_rps_send_ipi(struct softnet_data *remsd)
6005 {
6006 #ifdef CONFIG_RPS
6007 while (remsd) {
6008 struct softnet_data *next = remsd->rps_ipi_next;
6009
6010 if (cpu_online(remsd->cpu))
6011 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6012 remsd = next;
6013 }
6014 #endif
6015 }
6016
6017 /*
6018 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6019 * Note: called with local irq disabled, but exits with local irq enabled.
6020 */
net_rps_action_and_irq_enable(struct softnet_data * sd)6021 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6022 {
6023 #ifdef CONFIG_RPS
6024 struct softnet_data *remsd = sd->rps_ipi_list;
6025
6026 if (!use_backlog_threads() && remsd) {
6027 sd->rps_ipi_list = NULL;
6028
6029 local_irq_enable();
6030
6031 /* Send pending IPI's to kick RPS processing on remote cpus. */
6032 net_rps_send_ipi(remsd);
6033 } else
6034 #endif
6035 local_irq_enable();
6036 }
6037
sd_has_rps_ipi_waiting(struct softnet_data * sd)6038 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6039 {
6040 #ifdef CONFIG_RPS
6041 return !use_backlog_threads() && sd->rps_ipi_list;
6042 #else
6043 return false;
6044 #endif
6045 }
6046
process_backlog(struct napi_struct * napi,int quota)6047 static int process_backlog(struct napi_struct *napi, int quota)
6048 {
6049 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6050 bool again = true;
6051 int work = 0;
6052
6053 /* Check if we have pending ipi, its better to send them now,
6054 * not waiting net_rx_action() end.
6055 */
6056 if (sd_has_rps_ipi_waiting(sd)) {
6057 local_irq_disable();
6058 net_rps_action_and_irq_enable(sd);
6059 }
6060
6061 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
6062 while (again) {
6063 struct sk_buff *skb;
6064
6065 while ((skb = __skb_dequeue(&sd->process_queue))) {
6066 rcu_read_lock();
6067 __netif_receive_skb(skb);
6068 rcu_read_unlock();
6069 if (++work >= quota) {
6070 rps_input_queue_head_add(sd, work);
6071 return work;
6072 }
6073
6074 }
6075
6076 backlog_lock_irq_disable(sd);
6077 if (skb_queue_empty(&sd->input_pkt_queue)) {
6078 /*
6079 * Inline a custom version of __napi_complete().
6080 * only current cpu owns and manipulates this napi,
6081 * and NAPI_STATE_SCHED is the only possible flag set
6082 * on backlog.
6083 * We can use a plain write instead of clear_bit(),
6084 * and we dont need an smp_mb() memory barrier.
6085 */
6086 napi->state &= NAPIF_STATE_THREADED;
6087 again = false;
6088 } else {
6089 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6090 &sd->process_queue);
6091 }
6092 backlog_unlock_irq_enable(sd);
6093 }
6094
6095 if (work)
6096 rps_input_queue_head_add(sd, work);
6097 return work;
6098 }
6099
6100 /**
6101 * __napi_schedule - schedule for receive
6102 * @n: entry to schedule
6103 *
6104 * The entry's receive function will be scheduled to run.
6105 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6106 */
__napi_schedule(struct napi_struct * n)6107 void __napi_schedule(struct napi_struct *n)
6108 {
6109 unsigned long flags;
6110
6111 local_irq_save(flags);
6112 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6113 local_irq_restore(flags);
6114 }
6115 EXPORT_SYMBOL(__napi_schedule);
6116
6117 /**
6118 * napi_schedule_prep - check if napi can be scheduled
6119 * @n: napi context
6120 *
6121 * Test if NAPI routine is already running, and if not mark
6122 * it as running. This is used as a condition variable to
6123 * insure only one NAPI poll instance runs. We also make
6124 * sure there is no pending NAPI disable.
6125 */
napi_schedule_prep(struct napi_struct * n)6126 bool napi_schedule_prep(struct napi_struct *n)
6127 {
6128 unsigned long new, val = READ_ONCE(n->state);
6129
6130 do {
6131 if (unlikely(val & NAPIF_STATE_DISABLE))
6132 return false;
6133 new = val | NAPIF_STATE_SCHED;
6134
6135 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6136 * This was suggested by Alexander Duyck, as compiler
6137 * emits better code than :
6138 * if (val & NAPIF_STATE_SCHED)
6139 * new |= NAPIF_STATE_MISSED;
6140 */
6141 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6142 NAPIF_STATE_MISSED;
6143 } while (!try_cmpxchg(&n->state, &val, new));
6144
6145 return !(val & NAPIF_STATE_SCHED);
6146 }
6147 EXPORT_SYMBOL(napi_schedule_prep);
6148
6149 /**
6150 * __napi_schedule_irqoff - schedule for receive
6151 * @n: entry to schedule
6152 *
6153 * Variant of __napi_schedule() assuming hard irqs are masked.
6154 *
6155 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6156 * because the interrupt disabled assumption might not be true
6157 * due to force-threaded interrupts and spinlock substitution.
6158 */
__napi_schedule_irqoff(struct napi_struct * n)6159 void __napi_schedule_irqoff(struct napi_struct *n)
6160 {
6161 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6162 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6163 else
6164 __napi_schedule(n);
6165 }
6166 EXPORT_SYMBOL(__napi_schedule_irqoff);
6167
napi_complete_done(struct napi_struct * n,int work_done)6168 bool napi_complete_done(struct napi_struct *n, int work_done)
6169 {
6170 unsigned long flags, val, new, timeout = 0;
6171 bool ret = true;
6172
6173 /*
6174 * 1) Don't let napi dequeue from the cpu poll list
6175 * just in case its running on a different cpu.
6176 * 2) If we are busy polling, do nothing here, we have
6177 * the guarantee we will be called later.
6178 */
6179 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6180 NAPIF_STATE_IN_BUSY_POLL)))
6181 return false;
6182
6183 if (work_done) {
6184 if (n->gro_bitmask)
6185 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6186 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6187 }
6188 if (n->defer_hard_irqs_count > 0) {
6189 n->defer_hard_irqs_count--;
6190 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6191 if (timeout)
6192 ret = false;
6193 }
6194 if (n->gro_bitmask) {
6195 /* When the NAPI instance uses a timeout and keeps postponing
6196 * it, we need to bound somehow the time packets are kept in
6197 * the GRO layer
6198 */
6199 napi_gro_flush(n, !!timeout);
6200 }
6201
6202 gro_normal_list(n);
6203
6204 if (unlikely(!list_empty(&n->poll_list))) {
6205 /* If n->poll_list is not empty, we need to mask irqs */
6206 local_irq_save(flags);
6207 list_del_init(&n->poll_list);
6208 local_irq_restore(flags);
6209 }
6210 WRITE_ONCE(n->list_owner, -1);
6211
6212 val = READ_ONCE(n->state);
6213 do {
6214 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6215
6216 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6217 NAPIF_STATE_SCHED_THREADED |
6218 NAPIF_STATE_PREFER_BUSY_POLL);
6219
6220 /* If STATE_MISSED was set, leave STATE_SCHED set,
6221 * because we will call napi->poll() one more time.
6222 * This C code was suggested by Alexander Duyck to help gcc.
6223 */
6224 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6225 NAPIF_STATE_SCHED;
6226 } while (!try_cmpxchg(&n->state, &val, new));
6227
6228 if (unlikely(val & NAPIF_STATE_MISSED)) {
6229 __napi_schedule(n);
6230 return false;
6231 }
6232
6233 if (timeout)
6234 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6235 HRTIMER_MODE_REL_PINNED);
6236 return ret;
6237 }
6238 EXPORT_SYMBOL(napi_complete_done);
6239
6240 /* must be called under rcu_read_lock(), as we dont take a reference */
napi_by_id(unsigned int napi_id)6241 struct napi_struct *napi_by_id(unsigned int napi_id)
6242 {
6243 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6244 struct napi_struct *napi;
6245
6246 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6247 if (napi->napi_id == napi_id)
6248 return napi;
6249
6250 return NULL;
6251 }
6252
skb_defer_free_flush(struct softnet_data * sd)6253 static void skb_defer_free_flush(struct softnet_data *sd)
6254 {
6255 struct sk_buff *skb, *next;
6256
6257 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6258 if (!READ_ONCE(sd->defer_list))
6259 return;
6260
6261 spin_lock(&sd->defer_lock);
6262 skb = sd->defer_list;
6263 sd->defer_list = NULL;
6264 sd->defer_count = 0;
6265 spin_unlock(&sd->defer_lock);
6266
6267 while (skb != NULL) {
6268 next = skb->next;
6269 napi_consume_skb(skb, 1);
6270 skb = next;
6271 }
6272 }
6273
6274 #if defined(CONFIG_NET_RX_BUSY_POLL)
6275
__busy_poll_stop(struct napi_struct * napi,bool skip_schedule)6276 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6277 {
6278 if (!skip_schedule) {
6279 gro_normal_list(napi);
6280 __napi_schedule(napi);
6281 return;
6282 }
6283
6284 if (napi->gro_bitmask) {
6285 /* flush too old packets
6286 * If HZ < 1000, flush all packets.
6287 */
6288 napi_gro_flush(napi, HZ >= 1000);
6289 }
6290
6291 gro_normal_list(napi);
6292 clear_bit(NAPI_STATE_SCHED, &napi->state);
6293 }
6294
6295 enum {
6296 NAPI_F_PREFER_BUSY_POLL = 1,
6297 NAPI_F_END_ON_RESCHED = 2,
6298 };
6299
busy_poll_stop(struct napi_struct * napi,void * have_poll_lock,unsigned flags,u16 budget)6300 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock,
6301 unsigned flags, u16 budget)
6302 {
6303 bool skip_schedule = false;
6304 unsigned long timeout;
6305 int rc;
6306
6307 /* Busy polling means there is a high chance device driver hard irq
6308 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6309 * set in napi_schedule_prep().
6310 * Since we are about to call napi->poll() once more, we can safely
6311 * clear NAPI_STATE_MISSED.
6312 *
6313 * Note: x86 could use a single "lock and ..." instruction
6314 * to perform these two clear_bit()
6315 */
6316 clear_bit(NAPI_STATE_MISSED, &napi->state);
6317 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6318
6319 local_bh_disable();
6320
6321 if (flags & NAPI_F_PREFER_BUSY_POLL) {
6322 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6323 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6324 if (napi->defer_hard_irqs_count && timeout) {
6325 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6326 skip_schedule = true;
6327 }
6328 }
6329
6330 /* All we really want here is to re-enable device interrupts.
6331 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6332 */
6333 rc = napi->poll(napi, budget);
6334 /* We can't gro_normal_list() here, because napi->poll() might have
6335 * rearmed the napi (napi_complete_done()) in which case it could
6336 * already be running on another CPU.
6337 */
6338 trace_napi_poll(napi, rc, budget);
6339 netpoll_poll_unlock(have_poll_lock);
6340 if (rc == budget)
6341 __busy_poll_stop(napi, skip_schedule);
6342 local_bh_enable();
6343 }
6344
__napi_busy_loop(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,unsigned flags,u16 budget)6345 static void __napi_busy_loop(unsigned int napi_id,
6346 bool (*loop_end)(void *, unsigned long),
6347 void *loop_end_arg, unsigned flags, u16 budget)
6348 {
6349 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6350 int (*napi_poll)(struct napi_struct *napi, int budget);
6351 void *have_poll_lock = NULL;
6352 struct napi_struct *napi;
6353
6354 WARN_ON_ONCE(!rcu_read_lock_held());
6355
6356 restart:
6357 napi_poll = NULL;
6358
6359 napi = napi_by_id(napi_id);
6360 if (!napi)
6361 return;
6362
6363 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6364 preempt_disable();
6365 for (;;) {
6366 int work = 0;
6367
6368 local_bh_disable();
6369 if (!napi_poll) {
6370 unsigned long val = READ_ONCE(napi->state);
6371
6372 /* If multiple threads are competing for this napi,
6373 * we avoid dirtying napi->state as much as we can.
6374 */
6375 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6376 NAPIF_STATE_IN_BUSY_POLL)) {
6377 if (flags & NAPI_F_PREFER_BUSY_POLL)
6378 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6379 goto count;
6380 }
6381 if (cmpxchg(&napi->state, val,
6382 val | NAPIF_STATE_IN_BUSY_POLL |
6383 NAPIF_STATE_SCHED) != val) {
6384 if (flags & NAPI_F_PREFER_BUSY_POLL)
6385 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6386 goto count;
6387 }
6388 have_poll_lock = netpoll_poll_lock(napi);
6389 napi_poll = napi->poll;
6390 }
6391 work = napi_poll(napi, budget);
6392 trace_napi_poll(napi, work, budget);
6393 gro_normal_list(napi);
6394 count:
6395 if (work > 0)
6396 __NET_ADD_STATS(dev_net(napi->dev),
6397 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6398 skb_defer_free_flush(this_cpu_ptr(&softnet_data));
6399 local_bh_enable();
6400
6401 if (!loop_end || loop_end(loop_end_arg, start_time))
6402 break;
6403
6404 if (unlikely(need_resched())) {
6405 if (flags & NAPI_F_END_ON_RESCHED)
6406 break;
6407 if (napi_poll)
6408 busy_poll_stop(napi, have_poll_lock, flags, budget);
6409 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6410 preempt_enable();
6411 rcu_read_unlock();
6412 cond_resched();
6413 rcu_read_lock();
6414 if (loop_end(loop_end_arg, start_time))
6415 return;
6416 goto restart;
6417 }
6418 cpu_relax();
6419 }
6420 if (napi_poll)
6421 busy_poll_stop(napi, have_poll_lock, flags, budget);
6422 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6423 preempt_enable();
6424 }
6425
napi_busy_loop_rcu(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,bool prefer_busy_poll,u16 budget)6426 void napi_busy_loop_rcu(unsigned int napi_id,
6427 bool (*loop_end)(void *, unsigned long),
6428 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6429 {
6430 unsigned flags = NAPI_F_END_ON_RESCHED;
6431
6432 if (prefer_busy_poll)
6433 flags |= NAPI_F_PREFER_BUSY_POLL;
6434
6435 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6436 }
6437
napi_busy_loop(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,bool prefer_busy_poll,u16 budget)6438 void napi_busy_loop(unsigned int napi_id,
6439 bool (*loop_end)(void *, unsigned long),
6440 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6441 {
6442 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0;
6443
6444 rcu_read_lock();
6445 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6446 rcu_read_unlock();
6447 }
6448 EXPORT_SYMBOL(napi_busy_loop);
6449
6450 #endif /* CONFIG_NET_RX_BUSY_POLL */
6451
napi_hash_add(struct napi_struct * napi)6452 static void napi_hash_add(struct napi_struct *napi)
6453 {
6454 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6455 return;
6456
6457 spin_lock(&napi_hash_lock);
6458
6459 /* 0..NR_CPUS range is reserved for sender_cpu use */
6460 do {
6461 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6462 napi_gen_id = MIN_NAPI_ID;
6463 } while (napi_by_id(napi_gen_id));
6464 napi->napi_id = napi_gen_id;
6465
6466 hlist_add_head_rcu(&napi->napi_hash_node,
6467 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6468
6469 spin_unlock(&napi_hash_lock);
6470 }
6471
6472 /* Warning : caller is responsible to make sure rcu grace period
6473 * is respected before freeing memory containing @napi
6474 */
napi_hash_del(struct napi_struct * napi)6475 static void napi_hash_del(struct napi_struct *napi)
6476 {
6477 spin_lock(&napi_hash_lock);
6478
6479 hlist_del_init_rcu(&napi->napi_hash_node);
6480
6481 spin_unlock(&napi_hash_lock);
6482 }
6483
napi_watchdog(struct hrtimer * timer)6484 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6485 {
6486 struct napi_struct *napi;
6487
6488 napi = container_of(timer, struct napi_struct, timer);
6489
6490 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6491 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6492 */
6493 if (!napi_disable_pending(napi) &&
6494 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6495 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6496 __napi_schedule_irqoff(napi);
6497 }
6498
6499 return HRTIMER_NORESTART;
6500 }
6501
init_gro_hash(struct napi_struct * napi)6502 static void init_gro_hash(struct napi_struct *napi)
6503 {
6504 int i;
6505
6506 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6507 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6508 napi->gro_hash[i].count = 0;
6509 }
6510 napi->gro_bitmask = 0;
6511 }
6512
dev_set_threaded(struct net_device * dev,bool threaded)6513 int dev_set_threaded(struct net_device *dev, bool threaded)
6514 {
6515 struct napi_struct *napi;
6516 int err = 0;
6517
6518 if (dev->threaded == threaded)
6519 return 0;
6520
6521 if (threaded) {
6522 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6523 if (!napi->thread) {
6524 err = napi_kthread_create(napi);
6525 if (err) {
6526 threaded = false;
6527 break;
6528 }
6529 }
6530 }
6531 }
6532
6533 WRITE_ONCE(dev->threaded, threaded);
6534
6535 /* Make sure kthread is created before THREADED bit
6536 * is set.
6537 */
6538 smp_mb__before_atomic();
6539
6540 /* Setting/unsetting threaded mode on a napi might not immediately
6541 * take effect, if the current napi instance is actively being
6542 * polled. In this case, the switch between threaded mode and
6543 * softirq mode will happen in the next round of napi_schedule().
6544 * This should not cause hiccups/stalls to the live traffic.
6545 */
6546 list_for_each_entry(napi, &dev->napi_list, dev_list)
6547 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6548
6549 return err;
6550 }
6551 EXPORT_SYMBOL(dev_set_threaded);
6552
6553 /**
6554 * netif_queue_set_napi - Associate queue with the napi
6555 * @dev: device to which NAPI and queue belong
6556 * @queue_index: Index of queue
6557 * @type: queue type as RX or TX
6558 * @napi: NAPI context, pass NULL to clear previously set NAPI
6559 *
6560 * Set queue with its corresponding napi context. This should be done after
6561 * registering the NAPI handler for the queue-vector and the queues have been
6562 * mapped to the corresponding interrupt vector.
6563 */
netif_queue_set_napi(struct net_device * dev,unsigned int queue_index,enum netdev_queue_type type,struct napi_struct * napi)6564 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index,
6565 enum netdev_queue_type type, struct napi_struct *napi)
6566 {
6567 struct netdev_rx_queue *rxq;
6568 struct netdev_queue *txq;
6569
6570 if (WARN_ON_ONCE(napi && !napi->dev))
6571 return;
6572 if (dev->reg_state >= NETREG_REGISTERED)
6573 ASSERT_RTNL();
6574
6575 switch (type) {
6576 case NETDEV_QUEUE_TYPE_RX:
6577 rxq = __netif_get_rx_queue(dev, queue_index);
6578 rxq->napi = napi;
6579 return;
6580 case NETDEV_QUEUE_TYPE_TX:
6581 txq = netdev_get_tx_queue(dev, queue_index);
6582 txq->napi = napi;
6583 return;
6584 default:
6585 return;
6586 }
6587 }
6588 EXPORT_SYMBOL(netif_queue_set_napi);
6589
netif_napi_add_weight(struct net_device * dev,struct napi_struct * napi,int (* poll)(struct napi_struct *,int),int weight)6590 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6591 int (*poll)(struct napi_struct *, int), int weight)
6592 {
6593 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6594 return;
6595
6596 INIT_LIST_HEAD(&napi->poll_list);
6597 INIT_HLIST_NODE(&napi->napi_hash_node);
6598 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6599 napi->timer.function = napi_watchdog;
6600 init_gro_hash(napi);
6601 napi->skb = NULL;
6602 INIT_LIST_HEAD(&napi->rx_list);
6603 napi->rx_count = 0;
6604 napi->poll = poll;
6605 if (weight > NAPI_POLL_WEIGHT)
6606 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6607 weight);
6608 napi->weight = weight;
6609 napi->dev = dev;
6610 #ifdef CONFIG_NETPOLL
6611 napi->poll_owner = -1;
6612 #endif
6613 napi->list_owner = -1;
6614 set_bit(NAPI_STATE_SCHED, &napi->state);
6615 set_bit(NAPI_STATE_NPSVC, &napi->state);
6616 list_add_rcu(&napi->dev_list, &dev->napi_list);
6617 napi_hash_add(napi);
6618 napi_get_frags_check(napi);
6619 /* Create kthread for this napi if dev->threaded is set.
6620 * Clear dev->threaded if kthread creation failed so that
6621 * threaded mode will not be enabled in napi_enable().
6622 */
6623 if (dev->threaded && napi_kthread_create(napi))
6624 dev->threaded = false;
6625 netif_napi_set_irq(napi, -1);
6626 }
6627 EXPORT_SYMBOL(netif_napi_add_weight);
6628
napi_disable(struct napi_struct * n)6629 void napi_disable(struct napi_struct *n)
6630 {
6631 unsigned long val, new;
6632
6633 might_sleep();
6634 set_bit(NAPI_STATE_DISABLE, &n->state);
6635
6636 val = READ_ONCE(n->state);
6637 do {
6638 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6639 usleep_range(20, 200);
6640 val = READ_ONCE(n->state);
6641 }
6642
6643 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6644 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6645 } while (!try_cmpxchg(&n->state, &val, new));
6646
6647 hrtimer_cancel(&n->timer);
6648
6649 clear_bit(NAPI_STATE_DISABLE, &n->state);
6650 }
6651 EXPORT_SYMBOL(napi_disable);
6652
6653 /**
6654 * napi_enable - enable NAPI scheduling
6655 * @n: NAPI context
6656 *
6657 * Resume NAPI from being scheduled on this context.
6658 * Must be paired with napi_disable.
6659 */
napi_enable(struct napi_struct * n)6660 void napi_enable(struct napi_struct *n)
6661 {
6662 unsigned long new, val = READ_ONCE(n->state);
6663
6664 do {
6665 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6666
6667 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6668 if (n->dev->threaded && n->thread)
6669 new |= NAPIF_STATE_THREADED;
6670 } while (!try_cmpxchg(&n->state, &val, new));
6671 }
6672 EXPORT_SYMBOL(napi_enable);
6673
flush_gro_hash(struct napi_struct * napi)6674 static void flush_gro_hash(struct napi_struct *napi)
6675 {
6676 int i;
6677
6678 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6679 struct sk_buff *skb, *n;
6680
6681 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6682 kfree_skb(skb);
6683 napi->gro_hash[i].count = 0;
6684 }
6685 }
6686
6687 /* Must be called in process context */
__netif_napi_del(struct napi_struct * napi)6688 void __netif_napi_del(struct napi_struct *napi)
6689 {
6690 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6691 return;
6692
6693 napi_hash_del(napi);
6694 list_del_rcu(&napi->dev_list);
6695 napi_free_frags(napi);
6696
6697 flush_gro_hash(napi);
6698 napi->gro_bitmask = 0;
6699
6700 if (napi->thread) {
6701 kthread_stop(napi->thread);
6702 napi->thread = NULL;
6703 }
6704 }
6705 EXPORT_SYMBOL(__netif_napi_del);
6706
__napi_poll(struct napi_struct * n,bool * repoll)6707 static int __napi_poll(struct napi_struct *n, bool *repoll)
6708 {
6709 int work, weight;
6710
6711 weight = n->weight;
6712
6713 /* This NAPI_STATE_SCHED test is for avoiding a race
6714 * with netpoll's poll_napi(). Only the entity which
6715 * obtains the lock and sees NAPI_STATE_SCHED set will
6716 * actually make the ->poll() call. Therefore we avoid
6717 * accidentally calling ->poll() when NAPI is not scheduled.
6718 */
6719 work = 0;
6720 if (napi_is_scheduled(n)) {
6721 work = n->poll(n, weight);
6722 trace_napi_poll(n, work, weight);
6723
6724 xdp_do_check_flushed(n);
6725 }
6726
6727 if (unlikely(work > weight))
6728 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6729 n->poll, work, weight);
6730
6731 if (likely(work < weight))
6732 return work;
6733
6734 /* Drivers must not modify the NAPI state if they
6735 * consume the entire weight. In such cases this code
6736 * still "owns" the NAPI instance and therefore can
6737 * move the instance around on the list at-will.
6738 */
6739 if (unlikely(napi_disable_pending(n))) {
6740 napi_complete(n);
6741 return work;
6742 }
6743
6744 /* The NAPI context has more processing work, but busy-polling
6745 * is preferred. Exit early.
6746 */
6747 if (napi_prefer_busy_poll(n)) {
6748 if (napi_complete_done(n, work)) {
6749 /* If timeout is not set, we need to make sure
6750 * that the NAPI is re-scheduled.
6751 */
6752 napi_schedule(n);
6753 }
6754 return work;
6755 }
6756
6757 if (n->gro_bitmask) {
6758 /* flush too old packets
6759 * If HZ < 1000, flush all packets.
6760 */
6761 napi_gro_flush(n, HZ >= 1000);
6762 }
6763
6764 gro_normal_list(n);
6765
6766 /* Some drivers may have called napi_schedule
6767 * prior to exhausting their budget.
6768 */
6769 if (unlikely(!list_empty(&n->poll_list))) {
6770 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6771 n->dev ? n->dev->name : "backlog");
6772 return work;
6773 }
6774
6775 *repoll = true;
6776
6777 return work;
6778 }
6779
napi_poll(struct napi_struct * n,struct list_head * repoll)6780 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6781 {
6782 bool do_repoll = false;
6783 void *have;
6784 int work;
6785
6786 list_del_init(&n->poll_list);
6787
6788 have = netpoll_poll_lock(n);
6789
6790 work = __napi_poll(n, &do_repoll);
6791
6792 if (do_repoll)
6793 list_add_tail(&n->poll_list, repoll);
6794
6795 netpoll_poll_unlock(have);
6796
6797 return work;
6798 }
6799
napi_thread_wait(struct napi_struct * napi)6800 static int napi_thread_wait(struct napi_struct *napi)
6801 {
6802 set_current_state(TASK_INTERRUPTIBLE);
6803
6804 while (!kthread_should_stop()) {
6805 /* Testing SCHED_THREADED bit here to make sure the current
6806 * kthread owns this napi and could poll on this napi.
6807 * Testing SCHED bit is not enough because SCHED bit might be
6808 * set by some other busy poll thread or by napi_disable().
6809 */
6810 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) {
6811 WARN_ON(!list_empty(&napi->poll_list));
6812 __set_current_state(TASK_RUNNING);
6813 return 0;
6814 }
6815
6816 schedule();
6817 set_current_state(TASK_INTERRUPTIBLE);
6818 }
6819 __set_current_state(TASK_RUNNING);
6820
6821 return -1;
6822 }
6823
napi_threaded_poll_loop(struct napi_struct * napi)6824 static void napi_threaded_poll_loop(struct napi_struct *napi)
6825 {
6826 struct softnet_data *sd;
6827 unsigned long last_qs = jiffies;
6828
6829 for (;;) {
6830 bool repoll = false;
6831 void *have;
6832
6833 local_bh_disable();
6834 sd = this_cpu_ptr(&softnet_data);
6835 sd->in_napi_threaded_poll = true;
6836
6837 have = netpoll_poll_lock(napi);
6838 __napi_poll(napi, &repoll);
6839 netpoll_poll_unlock(have);
6840
6841 sd->in_napi_threaded_poll = false;
6842 barrier();
6843
6844 if (sd_has_rps_ipi_waiting(sd)) {
6845 local_irq_disable();
6846 net_rps_action_and_irq_enable(sd);
6847 }
6848 skb_defer_free_flush(sd);
6849 local_bh_enable();
6850
6851 if (!repoll)
6852 break;
6853
6854 rcu_softirq_qs_periodic(last_qs);
6855 cond_resched();
6856 }
6857 }
6858
napi_threaded_poll(void * data)6859 static int napi_threaded_poll(void *data)
6860 {
6861 struct napi_struct *napi = data;
6862
6863 while (!napi_thread_wait(napi))
6864 napi_threaded_poll_loop(napi);
6865
6866 return 0;
6867 }
6868
net_rx_action(struct softirq_action * h)6869 static __latent_entropy void net_rx_action(struct softirq_action *h)
6870 {
6871 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6872 unsigned long time_limit = jiffies +
6873 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
6874 int budget = READ_ONCE(net_hotdata.netdev_budget);
6875 LIST_HEAD(list);
6876 LIST_HEAD(repoll);
6877
6878 start:
6879 sd->in_net_rx_action = true;
6880 local_irq_disable();
6881 list_splice_init(&sd->poll_list, &list);
6882 local_irq_enable();
6883
6884 for (;;) {
6885 struct napi_struct *n;
6886
6887 skb_defer_free_flush(sd);
6888
6889 if (list_empty(&list)) {
6890 if (list_empty(&repoll)) {
6891 sd->in_net_rx_action = false;
6892 barrier();
6893 /* We need to check if ____napi_schedule()
6894 * had refilled poll_list while
6895 * sd->in_net_rx_action was true.
6896 */
6897 if (!list_empty(&sd->poll_list))
6898 goto start;
6899 if (!sd_has_rps_ipi_waiting(sd))
6900 goto end;
6901 }
6902 break;
6903 }
6904
6905 n = list_first_entry(&list, struct napi_struct, poll_list);
6906 budget -= napi_poll(n, &repoll);
6907
6908 /* If softirq window is exhausted then punt.
6909 * Allow this to run for 2 jiffies since which will allow
6910 * an average latency of 1.5/HZ.
6911 */
6912 if (unlikely(budget <= 0 ||
6913 time_after_eq(jiffies, time_limit))) {
6914 sd->time_squeeze++;
6915 break;
6916 }
6917 }
6918
6919 local_irq_disable();
6920
6921 list_splice_tail_init(&sd->poll_list, &list);
6922 list_splice_tail(&repoll, &list);
6923 list_splice(&list, &sd->poll_list);
6924 if (!list_empty(&sd->poll_list))
6925 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6926 else
6927 sd->in_net_rx_action = false;
6928
6929 net_rps_action_and_irq_enable(sd);
6930 end:;
6931 }
6932
6933 struct netdev_adjacent {
6934 struct net_device *dev;
6935 netdevice_tracker dev_tracker;
6936
6937 /* upper master flag, there can only be one master device per list */
6938 bool master;
6939
6940 /* lookup ignore flag */
6941 bool ignore;
6942
6943 /* counter for the number of times this device was added to us */
6944 u16 ref_nr;
6945
6946 /* private field for the users */
6947 void *private;
6948
6949 struct list_head list;
6950 struct rcu_head rcu;
6951 };
6952
__netdev_find_adj(struct net_device * adj_dev,struct list_head * adj_list)6953 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6954 struct list_head *adj_list)
6955 {
6956 struct netdev_adjacent *adj;
6957
6958 list_for_each_entry(adj, adj_list, list) {
6959 if (adj->dev == adj_dev)
6960 return adj;
6961 }
6962 return NULL;
6963 }
6964
____netdev_has_upper_dev(struct net_device * upper_dev,struct netdev_nested_priv * priv)6965 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6966 struct netdev_nested_priv *priv)
6967 {
6968 struct net_device *dev = (struct net_device *)priv->data;
6969
6970 return upper_dev == dev;
6971 }
6972
6973 /**
6974 * netdev_has_upper_dev - Check if device is linked to an upper device
6975 * @dev: device
6976 * @upper_dev: upper device to check
6977 *
6978 * Find out if a device is linked to specified upper device and return true
6979 * in case it is. Note that this checks only immediate upper device,
6980 * not through a complete stack of devices. The caller must hold the RTNL lock.
6981 */
netdev_has_upper_dev(struct net_device * dev,struct net_device * upper_dev)6982 bool netdev_has_upper_dev(struct net_device *dev,
6983 struct net_device *upper_dev)
6984 {
6985 struct netdev_nested_priv priv = {
6986 .data = (void *)upper_dev,
6987 };
6988
6989 ASSERT_RTNL();
6990
6991 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6992 &priv);
6993 }
6994 EXPORT_SYMBOL(netdev_has_upper_dev);
6995
6996 /**
6997 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6998 * @dev: device
6999 * @upper_dev: upper device to check
7000 *
7001 * Find out if a device is linked to specified upper device and return true
7002 * in case it is. Note that this checks the entire upper device chain.
7003 * The caller must hold rcu lock.
7004 */
7005
netdev_has_upper_dev_all_rcu(struct net_device * dev,struct net_device * upper_dev)7006 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7007 struct net_device *upper_dev)
7008 {
7009 struct netdev_nested_priv priv = {
7010 .data = (void *)upper_dev,
7011 };
7012
7013 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7014 &priv);
7015 }
7016 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7017
7018 /**
7019 * netdev_has_any_upper_dev - Check if device is linked to some device
7020 * @dev: device
7021 *
7022 * Find out if a device is linked to an upper device and return true in case
7023 * it is. The caller must hold the RTNL lock.
7024 */
netdev_has_any_upper_dev(struct net_device * dev)7025 bool netdev_has_any_upper_dev(struct net_device *dev)
7026 {
7027 ASSERT_RTNL();
7028
7029 return !list_empty(&dev->adj_list.upper);
7030 }
7031 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7032
7033 /**
7034 * netdev_master_upper_dev_get - Get master upper device
7035 * @dev: device
7036 *
7037 * Find a master upper device and return pointer to it or NULL in case
7038 * it's not there. The caller must hold the RTNL lock.
7039 */
netdev_master_upper_dev_get(struct net_device * dev)7040 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7041 {
7042 struct netdev_adjacent *upper;
7043
7044 ASSERT_RTNL();
7045
7046 if (list_empty(&dev->adj_list.upper))
7047 return NULL;
7048
7049 upper = list_first_entry(&dev->adj_list.upper,
7050 struct netdev_adjacent, list);
7051 if (likely(upper->master))
7052 return upper->dev;
7053 return NULL;
7054 }
7055 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7056
__netdev_master_upper_dev_get(struct net_device * dev)7057 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7058 {
7059 struct netdev_adjacent *upper;
7060
7061 ASSERT_RTNL();
7062
7063 if (list_empty(&dev->adj_list.upper))
7064 return NULL;
7065
7066 upper = list_first_entry(&dev->adj_list.upper,
7067 struct netdev_adjacent, list);
7068 if (likely(upper->master) && !upper->ignore)
7069 return upper->dev;
7070 return NULL;
7071 }
7072
7073 /**
7074 * netdev_has_any_lower_dev - Check if device is linked to some device
7075 * @dev: device
7076 *
7077 * Find out if a device is linked to a lower device and return true in case
7078 * it is. The caller must hold the RTNL lock.
7079 */
netdev_has_any_lower_dev(struct net_device * dev)7080 static bool netdev_has_any_lower_dev(struct net_device *dev)
7081 {
7082 ASSERT_RTNL();
7083
7084 return !list_empty(&dev->adj_list.lower);
7085 }
7086
netdev_adjacent_get_private(struct list_head * adj_list)7087 void *netdev_adjacent_get_private(struct list_head *adj_list)
7088 {
7089 struct netdev_adjacent *adj;
7090
7091 adj = list_entry(adj_list, struct netdev_adjacent, list);
7092
7093 return adj->private;
7094 }
7095 EXPORT_SYMBOL(netdev_adjacent_get_private);
7096
7097 /**
7098 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7099 * @dev: device
7100 * @iter: list_head ** of the current position
7101 *
7102 * Gets the next device from the dev's upper list, starting from iter
7103 * position. The caller must hold RCU read lock.
7104 */
netdev_upper_get_next_dev_rcu(struct net_device * dev,struct list_head ** iter)7105 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7106 struct list_head **iter)
7107 {
7108 struct netdev_adjacent *upper;
7109
7110 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7111
7112 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7113
7114 if (&upper->list == &dev->adj_list.upper)
7115 return NULL;
7116
7117 *iter = &upper->list;
7118
7119 return upper->dev;
7120 }
7121 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7122
__netdev_next_upper_dev(struct net_device * dev,struct list_head ** iter,bool * ignore)7123 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7124 struct list_head **iter,
7125 bool *ignore)
7126 {
7127 struct netdev_adjacent *upper;
7128
7129 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7130
7131 if (&upper->list == &dev->adj_list.upper)
7132 return NULL;
7133
7134 *iter = &upper->list;
7135 *ignore = upper->ignore;
7136
7137 return upper->dev;
7138 }
7139
netdev_next_upper_dev_rcu(struct net_device * dev,struct list_head ** iter)7140 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7141 struct list_head **iter)
7142 {
7143 struct netdev_adjacent *upper;
7144
7145 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7146
7147 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7148
7149 if (&upper->list == &dev->adj_list.upper)
7150 return NULL;
7151
7152 *iter = &upper->list;
7153
7154 return upper->dev;
7155 }
7156
__netdev_walk_all_upper_dev(struct net_device * dev,int (* fn)(struct net_device * dev,struct netdev_nested_priv * priv),struct netdev_nested_priv * priv)7157 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7158 int (*fn)(struct net_device *dev,
7159 struct netdev_nested_priv *priv),
7160 struct netdev_nested_priv *priv)
7161 {
7162 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7163 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7164 int ret, cur = 0;
7165 bool ignore;
7166
7167 now = dev;
7168 iter = &dev->adj_list.upper;
7169
7170 while (1) {
7171 if (now != dev) {
7172 ret = fn(now, priv);
7173 if (ret)
7174 return ret;
7175 }
7176
7177 next = NULL;
7178 while (1) {
7179 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7180 if (!udev)
7181 break;
7182 if (ignore)
7183 continue;
7184
7185 next = udev;
7186 niter = &udev->adj_list.upper;
7187 dev_stack[cur] = now;
7188 iter_stack[cur++] = iter;
7189 break;
7190 }
7191
7192 if (!next) {
7193 if (!cur)
7194 return 0;
7195 next = dev_stack[--cur];
7196 niter = iter_stack[cur];
7197 }
7198
7199 now = next;
7200 iter = niter;
7201 }
7202
7203 return 0;
7204 }
7205
netdev_walk_all_upper_dev_rcu(struct net_device * dev,int (* fn)(struct net_device * dev,struct netdev_nested_priv * priv),struct netdev_nested_priv * priv)7206 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7207 int (*fn)(struct net_device *dev,
7208 struct netdev_nested_priv *priv),
7209 struct netdev_nested_priv *priv)
7210 {
7211 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7212 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7213 int ret, cur = 0;
7214
7215 now = dev;
7216 iter = &dev->adj_list.upper;
7217
7218 while (1) {
7219 if (now != dev) {
7220 ret = fn(now, priv);
7221 if (ret)
7222 return ret;
7223 }
7224
7225 next = NULL;
7226 while (1) {
7227 udev = netdev_next_upper_dev_rcu(now, &iter);
7228 if (!udev)
7229 break;
7230
7231 next = udev;
7232 niter = &udev->adj_list.upper;
7233 dev_stack[cur] = now;
7234 iter_stack[cur++] = iter;
7235 break;
7236 }
7237
7238 if (!next) {
7239 if (!cur)
7240 return 0;
7241 next = dev_stack[--cur];
7242 niter = iter_stack[cur];
7243 }
7244
7245 now = next;
7246 iter = niter;
7247 }
7248
7249 return 0;
7250 }
7251 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7252
__netdev_has_upper_dev(struct net_device * dev,struct net_device * upper_dev)7253 static bool __netdev_has_upper_dev(struct net_device *dev,
7254 struct net_device *upper_dev)
7255 {
7256 struct netdev_nested_priv priv = {
7257 .flags = 0,
7258 .data = (void *)upper_dev,
7259 };
7260
7261 ASSERT_RTNL();
7262
7263 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7264 &priv);
7265 }
7266
7267 /**
7268 * netdev_lower_get_next_private - Get the next ->private from the
7269 * lower neighbour list
7270 * @dev: device
7271 * @iter: list_head ** of the current position
7272 *
7273 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7274 * list, starting from iter position. The caller must hold either hold the
7275 * RTNL lock or its own locking that guarantees that the neighbour lower
7276 * list will remain unchanged.
7277 */
netdev_lower_get_next_private(struct net_device * dev,struct list_head ** iter)7278 void *netdev_lower_get_next_private(struct net_device *dev,
7279 struct list_head **iter)
7280 {
7281 struct netdev_adjacent *lower;
7282
7283 lower = list_entry(*iter, struct netdev_adjacent, list);
7284
7285 if (&lower->list == &dev->adj_list.lower)
7286 return NULL;
7287
7288 *iter = lower->list.next;
7289
7290 return lower->private;
7291 }
7292 EXPORT_SYMBOL(netdev_lower_get_next_private);
7293
7294 /**
7295 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7296 * lower neighbour list, RCU
7297 * variant
7298 * @dev: device
7299 * @iter: list_head ** of the current position
7300 *
7301 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7302 * list, starting from iter position. The caller must hold RCU read lock.
7303 */
netdev_lower_get_next_private_rcu(struct net_device * dev,struct list_head ** iter)7304 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7305 struct list_head **iter)
7306 {
7307 struct netdev_adjacent *lower;
7308
7309 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7310
7311 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7312
7313 if (&lower->list == &dev->adj_list.lower)
7314 return NULL;
7315
7316 *iter = &lower->list;
7317
7318 return lower->private;
7319 }
7320 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7321
7322 /**
7323 * netdev_lower_get_next - Get the next device from the lower neighbour
7324 * list
7325 * @dev: device
7326 * @iter: list_head ** of the current position
7327 *
7328 * Gets the next netdev_adjacent from the dev's lower neighbour
7329 * list, starting from iter position. The caller must hold RTNL lock or
7330 * its own locking that guarantees that the neighbour lower
7331 * list will remain unchanged.
7332 */
netdev_lower_get_next(struct net_device * dev,struct list_head ** iter)7333 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7334 {
7335 struct netdev_adjacent *lower;
7336
7337 lower = list_entry(*iter, struct netdev_adjacent, list);
7338
7339 if (&lower->list == &dev->adj_list.lower)
7340 return NULL;
7341
7342 *iter = lower->list.next;
7343
7344 return lower->dev;
7345 }
7346 EXPORT_SYMBOL(netdev_lower_get_next);
7347
netdev_next_lower_dev(struct net_device * dev,struct list_head ** iter)7348 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7349 struct list_head **iter)
7350 {
7351 struct netdev_adjacent *lower;
7352
7353 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7354
7355 if (&lower->list == &dev->adj_list.lower)
7356 return NULL;
7357
7358 *iter = &lower->list;
7359
7360 return lower->dev;
7361 }
7362
__netdev_next_lower_dev(struct net_device * dev,struct list_head ** iter,bool * ignore)7363 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7364 struct list_head **iter,
7365 bool *ignore)
7366 {
7367 struct netdev_adjacent *lower;
7368
7369 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7370
7371 if (&lower->list == &dev->adj_list.lower)
7372 return NULL;
7373
7374 *iter = &lower->list;
7375 *ignore = lower->ignore;
7376
7377 return lower->dev;
7378 }
7379
netdev_walk_all_lower_dev(struct net_device * dev,int (* fn)(struct net_device * dev,struct netdev_nested_priv * priv),struct netdev_nested_priv * priv)7380 int netdev_walk_all_lower_dev(struct net_device *dev,
7381 int (*fn)(struct net_device *dev,
7382 struct netdev_nested_priv *priv),
7383 struct netdev_nested_priv *priv)
7384 {
7385 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7386 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7387 int ret, cur = 0;
7388
7389 now = dev;
7390 iter = &dev->adj_list.lower;
7391
7392 while (1) {
7393 if (now != dev) {
7394 ret = fn(now, priv);
7395 if (ret)
7396 return ret;
7397 }
7398
7399 next = NULL;
7400 while (1) {
7401 ldev = netdev_next_lower_dev(now, &iter);
7402 if (!ldev)
7403 break;
7404
7405 next = ldev;
7406 niter = &ldev->adj_list.lower;
7407 dev_stack[cur] = now;
7408 iter_stack[cur++] = iter;
7409 break;
7410 }
7411
7412 if (!next) {
7413 if (!cur)
7414 return 0;
7415 next = dev_stack[--cur];
7416 niter = iter_stack[cur];
7417 }
7418
7419 now = next;
7420 iter = niter;
7421 }
7422
7423 return 0;
7424 }
7425 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7426
__netdev_walk_all_lower_dev(struct net_device * dev,int (* fn)(struct net_device * dev,struct netdev_nested_priv * priv),struct netdev_nested_priv * priv)7427 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7428 int (*fn)(struct net_device *dev,
7429 struct netdev_nested_priv *priv),
7430 struct netdev_nested_priv *priv)
7431 {
7432 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7433 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7434 int ret, cur = 0;
7435 bool ignore;
7436
7437 now = dev;
7438 iter = &dev->adj_list.lower;
7439
7440 while (1) {
7441 if (now != dev) {
7442 ret = fn(now, priv);
7443 if (ret)
7444 return ret;
7445 }
7446
7447 next = NULL;
7448 while (1) {
7449 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7450 if (!ldev)
7451 break;
7452 if (ignore)
7453 continue;
7454
7455 next = ldev;
7456 niter = &ldev->adj_list.lower;
7457 dev_stack[cur] = now;
7458 iter_stack[cur++] = iter;
7459 break;
7460 }
7461
7462 if (!next) {
7463 if (!cur)
7464 return 0;
7465 next = dev_stack[--cur];
7466 niter = iter_stack[cur];
7467 }
7468
7469 now = next;
7470 iter = niter;
7471 }
7472
7473 return 0;
7474 }
7475
netdev_next_lower_dev_rcu(struct net_device * dev,struct list_head ** iter)7476 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7477 struct list_head **iter)
7478 {
7479 struct netdev_adjacent *lower;
7480
7481 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7482 if (&lower->list == &dev->adj_list.lower)
7483 return NULL;
7484
7485 *iter = &lower->list;
7486
7487 return lower->dev;
7488 }
7489 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7490
__netdev_upper_depth(struct net_device * dev)7491 static u8 __netdev_upper_depth(struct net_device *dev)
7492 {
7493 struct net_device *udev;
7494 struct list_head *iter;
7495 u8 max_depth = 0;
7496 bool ignore;
7497
7498 for (iter = &dev->adj_list.upper,
7499 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7500 udev;
7501 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7502 if (ignore)
7503 continue;
7504 if (max_depth < udev->upper_level)
7505 max_depth = udev->upper_level;
7506 }
7507
7508 return max_depth;
7509 }
7510
__netdev_lower_depth(struct net_device * dev)7511 static u8 __netdev_lower_depth(struct net_device *dev)
7512 {
7513 struct net_device *ldev;
7514 struct list_head *iter;
7515 u8 max_depth = 0;
7516 bool ignore;
7517
7518 for (iter = &dev->adj_list.lower,
7519 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7520 ldev;
7521 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7522 if (ignore)
7523 continue;
7524 if (max_depth < ldev->lower_level)
7525 max_depth = ldev->lower_level;
7526 }
7527
7528 return max_depth;
7529 }
7530
__netdev_update_upper_level(struct net_device * dev,struct netdev_nested_priv * __unused)7531 static int __netdev_update_upper_level(struct net_device *dev,
7532 struct netdev_nested_priv *__unused)
7533 {
7534 dev->upper_level = __netdev_upper_depth(dev) + 1;
7535 return 0;
7536 }
7537
7538 #ifdef CONFIG_LOCKDEP
7539 static LIST_HEAD(net_unlink_list);
7540
net_unlink_todo(struct net_device * dev)7541 static void net_unlink_todo(struct net_device *dev)
7542 {
7543 if (list_empty(&dev->unlink_list))
7544 list_add_tail(&dev->unlink_list, &net_unlink_list);
7545 }
7546 #endif
7547
__netdev_update_lower_level(struct net_device * dev,struct netdev_nested_priv * priv)7548 static int __netdev_update_lower_level(struct net_device *dev,
7549 struct netdev_nested_priv *priv)
7550 {
7551 dev->lower_level = __netdev_lower_depth(dev) + 1;
7552
7553 #ifdef CONFIG_LOCKDEP
7554 if (!priv)
7555 return 0;
7556
7557 if (priv->flags & NESTED_SYNC_IMM)
7558 dev->nested_level = dev->lower_level - 1;
7559 if (priv->flags & NESTED_SYNC_TODO)
7560 net_unlink_todo(dev);
7561 #endif
7562 return 0;
7563 }
7564
netdev_walk_all_lower_dev_rcu(struct net_device * dev,int (* fn)(struct net_device * dev,struct netdev_nested_priv * priv),struct netdev_nested_priv * priv)7565 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7566 int (*fn)(struct net_device *dev,
7567 struct netdev_nested_priv *priv),
7568 struct netdev_nested_priv *priv)
7569 {
7570 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7571 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7572 int ret, cur = 0;
7573
7574 now = dev;
7575 iter = &dev->adj_list.lower;
7576
7577 while (1) {
7578 if (now != dev) {
7579 ret = fn(now, priv);
7580 if (ret)
7581 return ret;
7582 }
7583
7584 next = NULL;
7585 while (1) {
7586 ldev = netdev_next_lower_dev_rcu(now, &iter);
7587 if (!ldev)
7588 break;
7589
7590 next = ldev;
7591 niter = &ldev->adj_list.lower;
7592 dev_stack[cur] = now;
7593 iter_stack[cur++] = iter;
7594 break;
7595 }
7596
7597 if (!next) {
7598 if (!cur)
7599 return 0;
7600 next = dev_stack[--cur];
7601 niter = iter_stack[cur];
7602 }
7603
7604 now = next;
7605 iter = niter;
7606 }
7607
7608 return 0;
7609 }
7610 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7611
7612 /**
7613 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7614 * lower neighbour list, RCU
7615 * variant
7616 * @dev: device
7617 *
7618 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7619 * list. The caller must hold RCU read lock.
7620 */
netdev_lower_get_first_private_rcu(struct net_device * dev)7621 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7622 {
7623 struct netdev_adjacent *lower;
7624
7625 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7626 struct netdev_adjacent, list);
7627 if (lower)
7628 return lower->private;
7629 return NULL;
7630 }
7631 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7632
7633 /**
7634 * netdev_master_upper_dev_get_rcu - Get master upper device
7635 * @dev: device
7636 *
7637 * Find a master upper device and return pointer to it or NULL in case
7638 * it's not there. The caller must hold the RCU read lock.
7639 */
netdev_master_upper_dev_get_rcu(struct net_device * dev)7640 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7641 {
7642 struct netdev_adjacent *upper;
7643
7644 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7645 struct netdev_adjacent, list);
7646 if (upper && likely(upper->master))
7647 return upper->dev;
7648 return NULL;
7649 }
7650 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7651
netdev_adjacent_sysfs_add(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)7652 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7653 struct net_device *adj_dev,
7654 struct list_head *dev_list)
7655 {
7656 char linkname[IFNAMSIZ+7];
7657
7658 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7659 "upper_%s" : "lower_%s", adj_dev->name);
7660 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7661 linkname);
7662 }
netdev_adjacent_sysfs_del(struct net_device * dev,char * name,struct list_head * dev_list)7663 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7664 char *name,
7665 struct list_head *dev_list)
7666 {
7667 char linkname[IFNAMSIZ+7];
7668
7669 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7670 "upper_%s" : "lower_%s", name);
7671 sysfs_remove_link(&(dev->dev.kobj), linkname);
7672 }
7673
netdev_adjacent_is_neigh_list(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)7674 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7675 struct net_device *adj_dev,
7676 struct list_head *dev_list)
7677 {
7678 return (dev_list == &dev->adj_list.upper ||
7679 dev_list == &dev->adj_list.lower) &&
7680 net_eq(dev_net(dev), dev_net(adj_dev));
7681 }
7682
__netdev_adjacent_dev_insert(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list,void * private,bool master)7683 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7684 struct net_device *adj_dev,
7685 struct list_head *dev_list,
7686 void *private, bool master)
7687 {
7688 struct netdev_adjacent *adj;
7689 int ret;
7690
7691 adj = __netdev_find_adj(adj_dev, dev_list);
7692
7693 if (adj) {
7694 adj->ref_nr += 1;
7695 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7696 dev->name, adj_dev->name, adj->ref_nr);
7697
7698 return 0;
7699 }
7700
7701 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7702 if (!adj)
7703 return -ENOMEM;
7704
7705 adj->dev = adj_dev;
7706 adj->master = master;
7707 adj->ref_nr = 1;
7708 adj->private = private;
7709 adj->ignore = false;
7710 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7711
7712 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7713 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7714
7715 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7716 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7717 if (ret)
7718 goto free_adj;
7719 }
7720
7721 /* Ensure that master link is always the first item in list. */
7722 if (master) {
7723 ret = sysfs_create_link(&(dev->dev.kobj),
7724 &(adj_dev->dev.kobj), "master");
7725 if (ret)
7726 goto remove_symlinks;
7727
7728 list_add_rcu(&adj->list, dev_list);
7729 } else {
7730 list_add_tail_rcu(&adj->list, dev_list);
7731 }
7732
7733 return 0;
7734
7735 remove_symlinks:
7736 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7737 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7738 free_adj:
7739 netdev_put(adj_dev, &adj->dev_tracker);
7740 kfree(adj);
7741
7742 return ret;
7743 }
7744
__netdev_adjacent_dev_remove(struct net_device * dev,struct net_device * adj_dev,u16 ref_nr,struct list_head * dev_list)7745 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7746 struct net_device *adj_dev,
7747 u16 ref_nr,
7748 struct list_head *dev_list)
7749 {
7750 struct netdev_adjacent *adj;
7751
7752 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7753 dev->name, adj_dev->name, ref_nr);
7754
7755 adj = __netdev_find_adj(adj_dev, dev_list);
7756
7757 if (!adj) {
7758 pr_err("Adjacency does not exist for device %s from %s\n",
7759 dev->name, adj_dev->name);
7760 WARN_ON(1);
7761 return;
7762 }
7763
7764 if (adj->ref_nr > ref_nr) {
7765 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7766 dev->name, adj_dev->name, ref_nr,
7767 adj->ref_nr - ref_nr);
7768 adj->ref_nr -= ref_nr;
7769 return;
7770 }
7771
7772 if (adj->master)
7773 sysfs_remove_link(&(dev->dev.kobj), "master");
7774
7775 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7776 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7777
7778 list_del_rcu(&adj->list);
7779 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7780 adj_dev->name, dev->name, adj_dev->name);
7781 netdev_put(adj_dev, &adj->dev_tracker);
7782 kfree_rcu(adj, rcu);
7783 }
7784
__netdev_adjacent_dev_link_lists(struct net_device * dev,struct net_device * upper_dev,struct list_head * up_list,struct list_head * down_list,void * private,bool master)7785 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7786 struct net_device *upper_dev,
7787 struct list_head *up_list,
7788 struct list_head *down_list,
7789 void *private, bool master)
7790 {
7791 int ret;
7792
7793 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7794 private, master);
7795 if (ret)
7796 return ret;
7797
7798 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7799 private, false);
7800 if (ret) {
7801 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7802 return ret;
7803 }
7804
7805 return 0;
7806 }
7807
__netdev_adjacent_dev_unlink_lists(struct net_device * dev,struct net_device * upper_dev,u16 ref_nr,struct list_head * up_list,struct list_head * down_list)7808 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7809 struct net_device *upper_dev,
7810 u16 ref_nr,
7811 struct list_head *up_list,
7812 struct list_head *down_list)
7813 {
7814 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7815 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7816 }
7817
__netdev_adjacent_dev_link_neighbour(struct net_device * dev,struct net_device * upper_dev,void * private,bool master)7818 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7819 struct net_device *upper_dev,
7820 void *private, bool master)
7821 {
7822 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7823 &dev->adj_list.upper,
7824 &upper_dev->adj_list.lower,
7825 private, master);
7826 }
7827
__netdev_adjacent_dev_unlink_neighbour(struct net_device * dev,struct net_device * upper_dev)7828 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7829 struct net_device *upper_dev)
7830 {
7831 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7832 &dev->adj_list.upper,
7833 &upper_dev->adj_list.lower);
7834 }
7835
__netdev_upper_dev_link(struct net_device * dev,struct net_device * upper_dev,bool master,void * upper_priv,void * upper_info,struct netdev_nested_priv * priv,struct netlink_ext_ack * extack)7836 static int __netdev_upper_dev_link(struct net_device *dev,
7837 struct net_device *upper_dev, bool master,
7838 void *upper_priv, void *upper_info,
7839 struct netdev_nested_priv *priv,
7840 struct netlink_ext_ack *extack)
7841 {
7842 struct netdev_notifier_changeupper_info changeupper_info = {
7843 .info = {
7844 .dev = dev,
7845 .extack = extack,
7846 },
7847 .upper_dev = upper_dev,
7848 .master = master,
7849 .linking = true,
7850 .upper_info = upper_info,
7851 };
7852 struct net_device *master_dev;
7853 int ret = 0;
7854
7855 ASSERT_RTNL();
7856
7857 if (dev == upper_dev)
7858 return -EBUSY;
7859
7860 /* To prevent loops, check if dev is not upper device to upper_dev. */
7861 if (__netdev_has_upper_dev(upper_dev, dev))
7862 return -EBUSY;
7863
7864 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7865 return -EMLINK;
7866
7867 if (!master) {
7868 if (__netdev_has_upper_dev(dev, upper_dev))
7869 return -EEXIST;
7870 } else {
7871 master_dev = __netdev_master_upper_dev_get(dev);
7872 if (master_dev)
7873 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7874 }
7875
7876 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7877 &changeupper_info.info);
7878 ret = notifier_to_errno(ret);
7879 if (ret)
7880 return ret;
7881
7882 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7883 master);
7884 if (ret)
7885 return ret;
7886
7887 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7888 &changeupper_info.info);
7889 ret = notifier_to_errno(ret);
7890 if (ret)
7891 goto rollback;
7892
7893 __netdev_update_upper_level(dev, NULL);
7894 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7895
7896 __netdev_update_lower_level(upper_dev, priv);
7897 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7898 priv);
7899
7900 return 0;
7901
7902 rollback:
7903 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7904
7905 return ret;
7906 }
7907
7908 /**
7909 * netdev_upper_dev_link - Add a link to the upper device
7910 * @dev: device
7911 * @upper_dev: new upper device
7912 * @extack: netlink extended ack
7913 *
7914 * Adds a link to device which is upper to this one. The caller must hold
7915 * the RTNL lock. On a failure a negative errno code is returned.
7916 * On success the reference counts are adjusted and the function
7917 * returns zero.
7918 */
netdev_upper_dev_link(struct net_device * dev,struct net_device * upper_dev,struct netlink_ext_ack * extack)7919 int netdev_upper_dev_link(struct net_device *dev,
7920 struct net_device *upper_dev,
7921 struct netlink_ext_ack *extack)
7922 {
7923 struct netdev_nested_priv priv = {
7924 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7925 .data = NULL,
7926 };
7927
7928 return __netdev_upper_dev_link(dev, upper_dev, false,
7929 NULL, NULL, &priv, extack);
7930 }
7931 EXPORT_SYMBOL(netdev_upper_dev_link);
7932
7933 /**
7934 * netdev_master_upper_dev_link - Add a master link to the upper device
7935 * @dev: device
7936 * @upper_dev: new upper device
7937 * @upper_priv: upper device private
7938 * @upper_info: upper info to be passed down via notifier
7939 * @extack: netlink extended ack
7940 *
7941 * Adds a link to device which is upper to this one. In this case, only
7942 * one master upper device can be linked, although other non-master devices
7943 * might be linked as well. The caller must hold the RTNL lock.
7944 * On a failure a negative errno code is returned. On success the reference
7945 * counts are adjusted and the function returns zero.
7946 */
netdev_master_upper_dev_link(struct net_device * dev,struct net_device * upper_dev,void * upper_priv,void * upper_info,struct netlink_ext_ack * extack)7947 int netdev_master_upper_dev_link(struct net_device *dev,
7948 struct net_device *upper_dev,
7949 void *upper_priv, void *upper_info,
7950 struct netlink_ext_ack *extack)
7951 {
7952 struct netdev_nested_priv priv = {
7953 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7954 .data = NULL,
7955 };
7956
7957 return __netdev_upper_dev_link(dev, upper_dev, true,
7958 upper_priv, upper_info, &priv, extack);
7959 }
7960 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7961
__netdev_upper_dev_unlink(struct net_device * dev,struct net_device * upper_dev,struct netdev_nested_priv * priv)7962 static void __netdev_upper_dev_unlink(struct net_device *dev,
7963 struct net_device *upper_dev,
7964 struct netdev_nested_priv *priv)
7965 {
7966 struct netdev_notifier_changeupper_info changeupper_info = {
7967 .info = {
7968 .dev = dev,
7969 },
7970 .upper_dev = upper_dev,
7971 .linking = false,
7972 };
7973
7974 ASSERT_RTNL();
7975
7976 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7977
7978 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7979 &changeupper_info.info);
7980
7981 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7982
7983 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7984 &changeupper_info.info);
7985
7986 __netdev_update_upper_level(dev, NULL);
7987 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7988
7989 __netdev_update_lower_level(upper_dev, priv);
7990 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7991 priv);
7992 }
7993
7994 /**
7995 * netdev_upper_dev_unlink - Removes a link to upper device
7996 * @dev: device
7997 * @upper_dev: new upper device
7998 *
7999 * Removes a link to device which is upper to this one. The caller must hold
8000 * the RTNL lock.
8001 */
netdev_upper_dev_unlink(struct net_device * dev,struct net_device * upper_dev)8002 void netdev_upper_dev_unlink(struct net_device *dev,
8003 struct net_device *upper_dev)
8004 {
8005 struct netdev_nested_priv priv = {
8006 .flags = NESTED_SYNC_TODO,
8007 .data = NULL,
8008 };
8009
8010 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8011 }
8012 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8013
__netdev_adjacent_dev_set(struct net_device * upper_dev,struct net_device * lower_dev,bool val)8014 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8015 struct net_device *lower_dev,
8016 bool val)
8017 {
8018 struct netdev_adjacent *adj;
8019
8020 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8021 if (adj)
8022 adj->ignore = val;
8023
8024 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8025 if (adj)
8026 adj->ignore = val;
8027 }
8028
netdev_adjacent_dev_disable(struct net_device * upper_dev,struct net_device * lower_dev)8029 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8030 struct net_device *lower_dev)
8031 {
8032 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8033 }
8034
netdev_adjacent_dev_enable(struct net_device * upper_dev,struct net_device * lower_dev)8035 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8036 struct net_device *lower_dev)
8037 {
8038 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8039 }
8040
netdev_adjacent_change_prepare(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev,struct netlink_ext_ack * extack)8041 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8042 struct net_device *new_dev,
8043 struct net_device *dev,
8044 struct netlink_ext_ack *extack)
8045 {
8046 struct netdev_nested_priv priv = {
8047 .flags = 0,
8048 .data = NULL,
8049 };
8050 int err;
8051
8052 if (!new_dev)
8053 return 0;
8054
8055 if (old_dev && new_dev != old_dev)
8056 netdev_adjacent_dev_disable(dev, old_dev);
8057 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8058 extack);
8059 if (err) {
8060 if (old_dev && new_dev != old_dev)
8061 netdev_adjacent_dev_enable(dev, old_dev);
8062 return err;
8063 }
8064
8065 return 0;
8066 }
8067 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8068
netdev_adjacent_change_commit(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev)8069 void netdev_adjacent_change_commit(struct net_device *old_dev,
8070 struct net_device *new_dev,
8071 struct net_device *dev)
8072 {
8073 struct netdev_nested_priv priv = {
8074 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8075 .data = NULL,
8076 };
8077
8078 if (!new_dev || !old_dev)
8079 return;
8080
8081 if (new_dev == old_dev)
8082 return;
8083
8084 netdev_adjacent_dev_enable(dev, old_dev);
8085 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8086 }
8087 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8088
netdev_adjacent_change_abort(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev)8089 void netdev_adjacent_change_abort(struct net_device *old_dev,
8090 struct net_device *new_dev,
8091 struct net_device *dev)
8092 {
8093 struct netdev_nested_priv priv = {
8094 .flags = 0,
8095 .data = NULL,
8096 };
8097
8098 if (!new_dev)
8099 return;
8100
8101 if (old_dev && new_dev != old_dev)
8102 netdev_adjacent_dev_enable(dev, old_dev);
8103
8104 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8105 }
8106 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8107
8108 /**
8109 * netdev_bonding_info_change - Dispatch event about slave change
8110 * @dev: device
8111 * @bonding_info: info to dispatch
8112 *
8113 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8114 * The caller must hold the RTNL lock.
8115 */
netdev_bonding_info_change(struct net_device * dev,struct netdev_bonding_info * bonding_info)8116 void netdev_bonding_info_change(struct net_device *dev,
8117 struct netdev_bonding_info *bonding_info)
8118 {
8119 struct netdev_notifier_bonding_info info = {
8120 .info.dev = dev,
8121 };
8122
8123 memcpy(&info.bonding_info, bonding_info,
8124 sizeof(struct netdev_bonding_info));
8125 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8126 &info.info);
8127 }
8128 EXPORT_SYMBOL(netdev_bonding_info_change);
8129
netdev_offload_xstats_enable_l3(struct net_device * dev,struct netlink_ext_ack * extack)8130 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
8131 struct netlink_ext_ack *extack)
8132 {
8133 struct netdev_notifier_offload_xstats_info info = {
8134 .info.dev = dev,
8135 .info.extack = extack,
8136 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8137 };
8138 int err;
8139 int rc;
8140
8141 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
8142 GFP_KERNEL);
8143 if (!dev->offload_xstats_l3)
8144 return -ENOMEM;
8145
8146 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
8147 NETDEV_OFFLOAD_XSTATS_DISABLE,
8148 &info.info);
8149 err = notifier_to_errno(rc);
8150 if (err)
8151 goto free_stats;
8152
8153 return 0;
8154
8155 free_stats:
8156 kfree(dev->offload_xstats_l3);
8157 dev->offload_xstats_l3 = NULL;
8158 return err;
8159 }
8160
netdev_offload_xstats_enable(struct net_device * dev,enum netdev_offload_xstats_type type,struct netlink_ext_ack * extack)8161 int netdev_offload_xstats_enable(struct net_device *dev,
8162 enum netdev_offload_xstats_type type,
8163 struct netlink_ext_ack *extack)
8164 {
8165 ASSERT_RTNL();
8166
8167 if (netdev_offload_xstats_enabled(dev, type))
8168 return -EALREADY;
8169
8170 switch (type) {
8171 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8172 return netdev_offload_xstats_enable_l3(dev, extack);
8173 }
8174
8175 WARN_ON(1);
8176 return -EINVAL;
8177 }
8178 EXPORT_SYMBOL(netdev_offload_xstats_enable);
8179
netdev_offload_xstats_disable_l3(struct net_device * dev)8180 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
8181 {
8182 struct netdev_notifier_offload_xstats_info info = {
8183 .info.dev = dev,
8184 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8185 };
8186
8187 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
8188 &info.info);
8189 kfree(dev->offload_xstats_l3);
8190 dev->offload_xstats_l3 = NULL;
8191 }
8192
netdev_offload_xstats_disable(struct net_device * dev,enum netdev_offload_xstats_type type)8193 int netdev_offload_xstats_disable(struct net_device *dev,
8194 enum netdev_offload_xstats_type type)
8195 {
8196 ASSERT_RTNL();
8197
8198 if (!netdev_offload_xstats_enabled(dev, type))
8199 return -EALREADY;
8200
8201 switch (type) {
8202 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8203 netdev_offload_xstats_disable_l3(dev);
8204 return 0;
8205 }
8206
8207 WARN_ON(1);
8208 return -EINVAL;
8209 }
8210 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8211
netdev_offload_xstats_disable_all(struct net_device * dev)8212 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8213 {
8214 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8215 }
8216
8217 static struct rtnl_hw_stats64 *
netdev_offload_xstats_get_ptr(const struct net_device * dev,enum netdev_offload_xstats_type type)8218 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8219 enum netdev_offload_xstats_type type)
8220 {
8221 switch (type) {
8222 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8223 return dev->offload_xstats_l3;
8224 }
8225
8226 WARN_ON(1);
8227 return NULL;
8228 }
8229
netdev_offload_xstats_enabled(const struct net_device * dev,enum netdev_offload_xstats_type type)8230 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8231 enum netdev_offload_xstats_type type)
8232 {
8233 ASSERT_RTNL();
8234
8235 return netdev_offload_xstats_get_ptr(dev, type);
8236 }
8237 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8238
8239 struct netdev_notifier_offload_xstats_ru {
8240 bool used;
8241 };
8242
8243 struct netdev_notifier_offload_xstats_rd {
8244 struct rtnl_hw_stats64 stats;
8245 bool used;
8246 };
8247
netdev_hw_stats64_add(struct rtnl_hw_stats64 * dest,const struct rtnl_hw_stats64 * src)8248 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8249 const struct rtnl_hw_stats64 *src)
8250 {
8251 dest->rx_packets += src->rx_packets;
8252 dest->tx_packets += src->tx_packets;
8253 dest->rx_bytes += src->rx_bytes;
8254 dest->tx_bytes += src->tx_bytes;
8255 dest->rx_errors += src->rx_errors;
8256 dest->tx_errors += src->tx_errors;
8257 dest->rx_dropped += src->rx_dropped;
8258 dest->tx_dropped += src->tx_dropped;
8259 dest->multicast += src->multicast;
8260 }
8261
netdev_offload_xstats_get_used(struct net_device * dev,enum netdev_offload_xstats_type type,bool * p_used,struct netlink_ext_ack * extack)8262 static int netdev_offload_xstats_get_used(struct net_device *dev,
8263 enum netdev_offload_xstats_type type,
8264 bool *p_used,
8265 struct netlink_ext_ack *extack)
8266 {
8267 struct netdev_notifier_offload_xstats_ru report_used = {};
8268 struct netdev_notifier_offload_xstats_info info = {
8269 .info.dev = dev,
8270 .info.extack = extack,
8271 .type = type,
8272 .report_used = &report_used,
8273 };
8274 int rc;
8275
8276 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8277 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8278 &info.info);
8279 *p_used = report_used.used;
8280 return notifier_to_errno(rc);
8281 }
8282
netdev_offload_xstats_get_stats(struct net_device * dev,enum netdev_offload_xstats_type type,struct rtnl_hw_stats64 * p_stats,bool * p_used,struct netlink_ext_ack * extack)8283 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8284 enum netdev_offload_xstats_type type,
8285 struct rtnl_hw_stats64 *p_stats,
8286 bool *p_used,
8287 struct netlink_ext_ack *extack)
8288 {
8289 struct netdev_notifier_offload_xstats_rd report_delta = {};
8290 struct netdev_notifier_offload_xstats_info info = {
8291 .info.dev = dev,
8292 .info.extack = extack,
8293 .type = type,
8294 .report_delta = &report_delta,
8295 };
8296 struct rtnl_hw_stats64 *stats;
8297 int rc;
8298
8299 stats = netdev_offload_xstats_get_ptr(dev, type);
8300 if (WARN_ON(!stats))
8301 return -EINVAL;
8302
8303 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8304 &info.info);
8305
8306 /* Cache whatever we got, even if there was an error, otherwise the
8307 * successful stats retrievals would get lost.
8308 */
8309 netdev_hw_stats64_add(stats, &report_delta.stats);
8310
8311 if (p_stats)
8312 *p_stats = *stats;
8313 *p_used = report_delta.used;
8314
8315 return notifier_to_errno(rc);
8316 }
8317
netdev_offload_xstats_get(struct net_device * dev,enum netdev_offload_xstats_type type,struct rtnl_hw_stats64 * p_stats,bool * p_used,struct netlink_ext_ack * extack)8318 int netdev_offload_xstats_get(struct net_device *dev,
8319 enum netdev_offload_xstats_type type,
8320 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8321 struct netlink_ext_ack *extack)
8322 {
8323 ASSERT_RTNL();
8324
8325 if (p_stats)
8326 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8327 p_used, extack);
8328 else
8329 return netdev_offload_xstats_get_used(dev, type, p_used,
8330 extack);
8331 }
8332 EXPORT_SYMBOL(netdev_offload_xstats_get);
8333
8334 void
netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd * report_delta,const struct rtnl_hw_stats64 * stats)8335 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8336 const struct rtnl_hw_stats64 *stats)
8337 {
8338 report_delta->used = true;
8339 netdev_hw_stats64_add(&report_delta->stats, stats);
8340 }
8341 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8342
8343 void
netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru * report_used)8344 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8345 {
8346 report_used->used = true;
8347 }
8348 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8349
netdev_offload_xstats_push_delta(struct net_device * dev,enum netdev_offload_xstats_type type,const struct rtnl_hw_stats64 * p_stats)8350 void netdev_offload_xstats_push_delta(struct net_device *dev,
8351 enum netdev_offload_xstats_type type,
8352 const struct rtnl_hw_stats64 *p_stats)
8353 {
8354 struct rtnl_hw_stats64 *stats;
8355
8356 ASSERT_RTNL();
8357
8358 stats = netdev_offload_xstats_get_ptr(dev, type);
8359 if (WARN_ON(!stats))
8360 return;
8361
8362 netdev_hw_stats64_add(stats, p_stats);
8363 }
8364 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8365
8366 /**
8367 * netdev_get_xmit_slave - Get the xmit slave of master device
8368 * @dev: device
8369 * @skb: The packet
8370 * @all_slaves: assume all the slaves are active
8371 *
8372 * The reference counters are not incremented so the caller must be
8373 * careful with locks. The caller must hold RCU lock.
8374 * %NULL is returned if no slave is found.
8375 */
8376
netdev_get_xmit_slave(struct net_device * dev,struct sk_buff * skb,bool all_slaves)8377 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8378 struct sk_buff *skb,
8379 bool all_slaves)
8380 {
8381 const struct net_device_ops *ops = dev->netdev_ops;
8382
8383 if (!ops->ndo_get_xmit_slave)
8384 return NULL;
8385 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8386 }
8387 EXPORT_SYMBOL(netdev_get_xmit_slave);
8388
netdev_sk_get_lower_dev(struct net_device * dev,struct sock * sk)8389 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8390 struct sock *sk)
8391 {
8392 const struct net_device_ops *ops = dev->netdev_ops;
8393
8394 if (!ops->ndo_sk_get_lower_dev)
8395 return NULL;
8396 return ops->ndo_sk_get_lower_dev(dev, sk);
8397 }
8398
8399 /**
8400 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8401 * @dev: device
8402 * @sk: the socket
8403 *
8404 * %NULL is returned if no lower device is found.
8405 */
8406
netdev_sk_get_lowest_dev(struct net_device * dev,struct sock * sk)8407 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8408 struct sock *sk)
8409 {
8410 struct net_device *lower;
8411
8412 lower = netdev_sk_get_lower_dev(dev, sk);
8413 while (lower) {
8414 dev = lower;
8415 lower = netdev_sk_get_lower_dev(dev, sk);
8416 }
8417
8418 return dev;
8419 }
8420 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8421
netdev_adjacent_add_links(struct net_device * dev)8422 static void netdev_adjacent_add_links(struct net_device *dev)
8423 {
8424 struct netdev_adjacent *iter;
8425
8426 struct net *net = dev_net(dev);
8427
8428 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8429 if (!net_eq(net, dev_net(iter->dev)))
8430 continue;
8431 netdev_adjacent_sysfs_add(iter->dev, dev,
8432 &iter->dev->adj_list.lower);
8433 netdev_adjacent_sysfs_add(dev, iter->dev,
8434 &dev->adj_list.upper);
8435 }
8436
8437 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8438 if (!net_eq(net, dev_net(iter->dev)))
8439 continue;
8440 netdev_adjacent_sysfs_add(iter->dev, dev,
8441 &iter->dev->adj_list.upper);
8442 netdev_adjacent_sysfs_add(dev, iter->dev,
8443 &dev->adj_list.lower);
8444 }
8445 }
8446
netdev_adjacent_del_links(struct net_device * dev)8447 static void netdev_adjacent_del_links(struct net_device *dev)
8448 {
8449 struct netdev_adjacent *iter;
8450
8451 struct net *net = dev_net(dev);
8452
8453 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8454 if (!net_eq(net, dev_net(iter->dev)))
8455 continue;
8456 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8457 &iter->dev->adj_list.lower);
8458 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8459 &dev->adj_list.upper);
8460 }
8461
8462 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8463 if (!net_eq(net, dev_net(iter->dev)))
8464 continue;
8465 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8466 &iter->dev->adj_list.upper);
8467 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8468 &dev->adj_list.lower);
8469 }
8470 }
8471
netdev_adjacent_rename_links(struct net_device * dev,char * oldname)8472 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8473 {
8474 struct netdev_adjacent *iter;
8475
8476 struct net *net = dev_net(dev);
8477
8478 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8479 if (!net_eq(net, dev_net(iter->dev)))
8480 continue;
8481 netdev_adjacent_sysfs_del(iter->dev, oldname,
8482 &iter->dev->adj_list.lower);
8483 netdev_adjacent_sysfs_add(iter->dev, dev,
8484 &iter->dev->adj_list.lower);
8485 }
8486
8487 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8488 if (!net_eq(net, dev_net(iter->dev)))
8489 continue;
8490 netdev_adjacent_sysfs_del(iter->dev, oldname,
8491 &iter->dev->adj_list.upper);
8492 netdev_adjacent_sysfs_add(iter->dev, dev,
8493 &iter->dev->adj_list.upper);
8494 }
8495 }
8496
netdev_lower_dev_get_private(struct net_device * dev,struct net_device * lower_dev)8497 void *netdev_lower_dev_get_private(struct net_device *dev,
8498 struct net_device *lower_dev)
8499 {
8500 struct netdev_adjacent *lower;
8501
8502 if (!lower_dev)
8503 return NULL;
8504 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8505 if (!lower)
8506 return NULL;
8507
8508 return lower->private;
8509 }
8510 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8511
8512
8513 /**
8514 * netdev_lower_state_changed - Dispatch event about lower device state change
8515 * @lower_dev: device
8516 * @lower_state_info: state to dispatch
8517 *
8518 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8519 * The caller must hold the RTNL lock.
8520 */
netdev_lower_state_changed(struct net_device * lower_dev,void * lower_state_info)8521 void netdev_lower_state_changed(struct net_device *lower_dev,
8522 void *lower_state_info)
8523 {
8524 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8525 .info.dev = lower_dev,
8526 };
8527
8528 ASSERT_RTNL();
8529 changelowerstate_info.lower_state_info = lower_state_info;
8530 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8531 &changelowerstate_info.info);
8532 }
8533 EXPORT_SYMBOL(netdev_lower_state_changed);
8534
dev_change_rx_flags(struct net_device * dev,int flags)8535 static void dev_change_rx_flags(struct net_device *dev, int flags)
8536 {
8537 const struct net_device_ops *ops = dev->netdev_ops;
8538
8539 if (ops->ndo_change_rx_flags)
8540 ops->ndo_change_rx_flags(dev, flags);
8541 }
8542
__dev_set_promiscuity(struct net_device * dev,int inc,bool notify)8543 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8544 {
8545 unsigned int old_flags = dev->flags;
8546 unsigned int promiscuity, flags;
8547 kuid_t uid;
8548 kgid_t gid;
8549
8550 ASSERT_RTNL();
8551
8552 promiscuity = dev->promiscuity + inc;
8553 if (promiscuity == 0) {
8554 /*
8555 * Avoid overflow.
8556 * If inc causes overflow, untouch promisc and return error.
8557 */
8558 if (unlikely(inc > 0)) {
8559 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8560 return -EOVERFLOW;
8561 }
8562 flags = old_flags & ~IFF_PROMISC;
8563 } else {
8564 flags = old_flags | IFF_PROMISC;
8565 }
8566 WRITE_ONCE(dev->promiscuity, promiscuity);
8567 if (flags != old_flags) {
8568 WRITE_ONCE(dev->flags, flags);
8569 netdev_info(dev, "%s promiscuous mode\n",
8570 dev->flags & IFF_PROMISC ? "entered" : "left");
8571 if (audit_enabled) {
8572 current_uid_gid(&uid, &gid);
8573 audit_log(audit_context(), GFP_ATOMIC,
8574 AUDIT_ANOM_PROMISCUOUS,
8575 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8576 dev->name, (dev->flags & IFF_PROMISC),
8577 (old_flags & IFF_PROMISC),
8578 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8579 from_kuid(&init_user_ns, uid),
8580 from_kgid(&init_user_ns, gid),
8581 audit_get_sessionid(current));
8582 }
8583
8584 dev_change_rx_flags(dev, IFF_PROMISC);
8585 }
8586 if (notify)
8587 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8588 return 0;
8589 }
8590
8591 /**
8592 * dev_set_promiscuity - update promiscuity count on a device
8593 * @dev: device
8594 * @inc: modifier
8595 *
8596 * Add or remove promiscuity from a device. While the count in the device
8597 * remains above zero the interface remains promiscuous. Once it hits zero
8598 * the device reverts back to normal filtering operation. A negative inc
8599 * value is used to drop promiscuity on the device.
8600 * Return 0 if successful or a negative errno code on error.
8601 */
dev_set_promiscuity(struct net_device * dev,int inc)8602 int dev_set_promiscuity(struct net_device *dev, int inc)
8603 {
8604 unsigned int old_flags = dev->flags;
8605 int err;
8606
8607 err = __dev_set_promiscuity(dev, inc, true);
8608 if (err < 0)
8609 return err;
8610 if (dev->flags != old_flags)
8611 dev_set_rx_mode(dev);
8612 return err;
8613 }
8614 EXPORT_SYMBOL(dev_set_promiscuity);
8615
__dev_set_allmulti(struct net_device * dev,int inc,bool notify)8616 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8617 {
8618 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8619 unsigned int allmulti, flags;
8620
8621 ASSERT_RTNL();
8622
8623 allmulti = dev->allmulti + inc;
8624 if (allmulti == 0) {
8625 /*
8626 * Avoid overflow.
8627 * If inc causes overflow, untouch allmulti and return error.
8628 */
8629 if (unlikely(inc > 0)) {
8630 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8631 return -EOVERFLOW;
8632 }
8633 flags = old_flags & ~IFF_ALLMULTI;
8634 } else {
8635 flags = old_flags | IFF_ALLMULTI;
8636 }
8637 WRITE_ONCE(dev->allmulti, allmulti);
8638 if (flags != old_flags) {
8639 WRITE_ONCE(dev->flags, flags);
8640 netdev_info(dev, "%s allmulticast mode\n",
8641 dev->flags & IFF_ALLMULTI ? "entered" : "left");
8642 dev_change_rx_flags(dev, IFF_ALLMULTI);
8643 dev_set_rx_mode(dev);
8644 if (notify)
8645 __dev_notify_flags(dev, old_flags,
8646 dev->gflags ^ old_gflags, 0, NULL);
8647 }
8648 return 0;
8649 }
8650
8651 /**
8652 * dev_set_allmulti - update allmulti count on a device
8653 * @dev: device
8654 * @inc: modifier
8655 *
8656 * Add or remove reception of all multicast frames to a device. While the
8657 * count in the device remains above zero the interface remains listening
8658 * to all interfaces. Once it hits zero the device reverts back to normal
8659 * filtering operation. A negative @inc value is used to drop the counter
8660 * when releasing a resource needing all multicasts.
8661 * Return 0 if successful or a negative errno code on error.
8662 */
8663
dev_set_allmulti(struct net_device * dev,int inc)8664 int dev_set_allmulti(struct net_device *dev, int inc)
8665 {
8666 return __dev_set_allmulti(dev, inc, true);
8667 }
8668 EXPORT_SYMBOL(dev_set_allmulti);
8669
8670 /*
8671 * Upload unicast and multicast address lists to device and
8672 * configure RX filtering. When the device doesn't support unicast
8673 * filtering it is put in promiscuous mode while unicast addresses
8674 * are present.
8675 */
__dev_set_rx_mode(struct net_device * dev)8676 void __dev_set_rx_mode(struct net_device *dev)
8677 {
8678 const struct net_device_ops *ops = dev->netdev_ops;
8679
8680 /* dev_open will call this function so the list will stay sane. */
8681 if (!(dev->flags&IFF_UP))
8682 return;
8683
8684 if (!netif_device_present(dev))
8685 return;
8686
8687 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8688 /* Unicast addresses changes may only happen under the rtnl,
8689 * therefore calling __dev_set_promiscuity here is safe.
8690 */
8691 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8692 __dev_set_promiscuity(dev, 1, false);
8693 dev->uc_promisc = true;
8694 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8695 __dev_set_promiscuity(dev, -1, false);
8696 dev->uc_promisc = false;
8697 }
8698 }
8699
8700 if (ops->ndo_set_rx_mode)
8701 ops->ndo_set_rx_mode(dev);
8702 }
8703
dev_set_rx_mode(struct net_device * dev)8704 void dev_set_rx_mode(struct net_device *dev)
8705 {
8706 netif_addr_lock_bh(dev);
8707 __dev_set_rx_mode(dev);
8708 netif_addr_unlock_bh(dev);
8709 }
8710
8711 /**
8712 * dev_get_flags - get flags reported to userspace
8713 * @dev: device
8714 *
8715 * Get the combination of flag bits exported through APIs to userspace.
8716 */
dev_get_flags(const struct net_device * dev)8717 unsigned int dev_get_flags(const struct net_device *dev)
8718 {
8719 unsigned int flags;
8720
8721 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC |
8722 IFF_ALLMULTI |
8723 IFF_RUNNING |
8724 IFF_LOWER_UP |
8725 IFF_DORMANT)) |
8726 (READ_ONCE(dev->gflags) & (IFF_PROMISC |
8727 IFF_ALLMULTI));
8728
8729 if (netif_running(dev)) {
8730 if (netif_oper_up(dev))
8731 flags |= IFF_RUNNING;
8732 if (netif_carrier_ok(dev))
8733 flags |= IFF_LOWER_UP;
8734 if (netif_dormant(dev))
8735 flags |= IFF_DORMANT;
8736 }
8737
8738 return flags;
8739 }
8740 EXPORT_SYMBOL(dev_get_flags);
8741
__dev_change_flags(struct net_device * dev,unsigned int flags,struct netlink_ext_ack * extack)8742 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8743 struct netlink_ext_ack *extack)
8744 {
8745 unsigned int old_flags = dev->flags;
8746 int ret;
8747
8748 ASSERT_RTNL();
8749
8750 /*
8751 * Set the flags on our device.
8752 */
8753
8754 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8755 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8756 IFF_AUTOMEDIA)) |
8757 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8758 IFF_ALLMULTI));
8759
8760 /*
8761 * Load in the correct multicast list now the flags have changed.
8762 */
8763
8764 if ((old_flags ^ flags) & IFF_MULTICAST)
8765 dev_change_rx_flags(dev, IFF_MULTICAST);
8766
8767 dev_set_rx_mode(dev);
8768
8769 /*
8770 * Have we downed the interface. We handle IFF_UP ourselves
8771 * according to user attempts to set it, rather than blindly
8772 * setting it.
8773 */
8774
8775 ret = 0;
8776 if ((old_flags ^ flags) & IFF_UP) {
8777 if (old_flags & IFF_UP)
8778 __dev_close(dev);
8779 else
8780 ret = __dev_open(dev, extack);
8781 }
8782
8783 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8784 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8785 unsigned int old_flags = dev->flags;
8786
8787 dev->gflags ^= IFF_PROMISC;
8788
8789 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8790 if (dev->flags != old_flags)
8791 dev_set_rx_mode(dev);
8792 }
8793
8794 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8795 * is important. Some (broken) drivers set IFF_PROMISC, when
8796 * IFF_ALLMULTI is requested not asking us and not reporting.
8797 */
8798 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8799 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8800
8801 dev->gflags ^= IFF_ALLMULTI;
8802 __dev_set_allmulti(dev, inc, false);
8803 }
8804
8805 return ret;
8806 }
8807
__dev_notify_flags(struct net_device * dev,unsigned int old_flags,unsigned int gchanges,u32 portid,const struct nlmsghdr * nlh)8808 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8809 unsigned int gchanges, u32 portid,
8810 const struct nlmsghdr *nlh)
8811 {
8812 unsigned int changes = dev->flags ^ old_flags;
8813
8814 if (gchanges)
8815 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8816
8817 if (changes & IFF_UP) {
8818 if (dev->flags & IFF_UP)
8819 call_netdevice_notifiers(NETDEV_UP, dev);
8820 else
8821 call_netdevice_notifiers(NETDEV_DOWN, dev);
8822 }
8823
8824 if (dev->flags & IFF_UP &&
8825 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8826 struct netdev_notifier_change_info change_info = {
8827 .info = {
8828 .dev = dev,
8829 },
8830 .flags_changed = changes,
8831 };
8832
8833 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8834 }
8835 }
8836
8837 /**
8838 * dev_change_flags - change device settings
8839 * @dev: device
8840 * @flags: device state flags
8841 * @extack: netlink extended ack
8842 *
8843 * Change settings on device based state flags. The flags are
8844 * in the userspace exported format.
8845 */
dev_change_flags(struct net_device * dev,unsigned int flags,struct netlink_ext_ack * extack)8846 int dev_change_flags(struct net_device *dev, unsigned int flags,
8847 struct netlink_ext_ack *extack)
8848 {
8849 int ret;
8850 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8851
8852 ret = __dev_change_flags(dev, flags, extack);
8853 if (ret < 0)
8854 return ret;
8855
8856 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8857 __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8858 return ret;
8859 }
8860 EXPORT_SYMBOL(dev_change_flags);
8861
__dev_set_mtu(struct net_device * dev,int new_mtu)8862 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8863 {
8864 const struct net_device_ops *ops = dev->netdev_ops;
8865
8866 if (ops->ndo_change_mtu)
8867 return ops->ndo_change_mtu(dev, new_mtu);
8868
8869 /* Pairs with all the lockless reads of dev->mtu in the stack */
8870 WRITE_ONCE(dev->mtu, new_mtu);
8871 return 0;
8872 }
8873 EXPORT_SYMBOL(__dev_set_mtu);
8874
dev_validate_mtu(struct net_device * dev,int new_mtu,struct netlink_ext_ack * extack)8875 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8876 struct netlink_ext_ack *extack)
8877 {
8878 /* MTU must be positive, and in range */
8879 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8880 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8881 return -EINVAL;
8882 }
8883
8884 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8885 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8886 return -EINVAL;
8887 }
8888 return 0;
8889 }
8890
8891 /**
8892 * dev_set_mtu_ext - Change maximum transfer unit
8893 * @dev: device
8894 * @new_mtu: new transfer unit
8895 * @extack: netlink extended ack
8896 *
8897 * Change the maximum transfer size of the network device.
8898 */
dev_set_mtu_ext(struct net_device * dev,int new_mtu,struct netlink_ext_ack * extack)8899 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8900 struct netlink_ext_ack *extack)
8901 {
8902 int err, orig_mtu;
8903
8904 if (new_mtu == dev->mtu)
8905 return 0;
8906
8907 err = dev_validate_mtu(dev, new_mtu, extack);
8908 if (err)
8909 return err;
8910
8911 if (!netif_device_present(dev))
8912 return -ENODEV;
8913
8914 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8915 err = notifier_to_errno(err);
8916 if (err)
8917 return err;
8918
8919 orig_mtu = dev->mtu;
8920 err = __dev_set_mtu(dev, new_mtu);
8921
8922 if (!err) {
8923 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8924 orig_mtu);
8925 err = notifier_to_errno(err);
8926 if (err) {
8927 /* setting mtu back and notifying everyone again,
8928 * so that they have a chance to revert changes.
8929 */
8930 __dev_set_mtu(dev, orig_mtu);
8931 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8932 new_mtu);
8933 }
8934 }
8935 return err;
8936 }
8937
dev_set_mtu(struct net_device * dev,int new_mtu)8938 int dev_set_mtu(struct net_device *dev, int new_mtu)
8939 {
8940 struct netlink_ext_ack extack;
8941 int err;
8942
8943 memset(&extack, 0, sizeof(extack));
8944 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8945 if (err && extack._msg)
8946 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8947 return err;
8948 }
8949 EXPORT_SYMBOL(dev_set_mtu);
8950
8951 /**
8952 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8953 * @dev: device
8954 * @new_len: new tx queue length
8955 */
dev_change_tx_queue_len(struct net_device * dev,unsigned long new_len)8956 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8957 {
8958 unsigned int orig_len = dev->tx_queue_len;
8959 int res;
8960
8961 if (new_len != (unsigned int)new_len)
8962 return -ERANGE;
8963
8964 if (new_len != orig_len) {
8965 WRITE_ONCE(dev->tx_queue_len, new_len);
8966 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8967 res = notifier_to_errno(res);
8968 if (res)
8969 goto err_rollback;
8970 res = dev_qdisc_change_tx_queue_len(dev);
8971 if (res)
8972 goto err_rollback;
8973 }
8974
8975 return 0;
8976
8977 err_rollback:
8978 netdev_err(dev, "refused to change device tx_queue_len\n");
8979 WRITE_ONCE(dev->tx_queue_len, orig_len);
8980 return res;
8981 }
8982
8983 /**
8984 * dev_set_group - Change group this device belongs to
8985 * @dev: device
8986 * @new_group: group this device should belong to
8987 */
dev_set_group(struct net_device * dev,int new_group)8988 void dev_set_group(struct net_device *dev, int new_group)
8989 {
8990 dev->group = new_group;
8991 }
8992
8993 /**
8994 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8995 * @dev: device
8996 * @addr: new address
8997 * @extack: netlink extended ack
8998 */
dev_pre_changeaddr_notify(struct net_device * dev,const char * addr,struct netlink_ext_ack * extack)8999 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9000 struct netlink_ext_ack *extack)
9001 {
9002 struct netdev_notifier_pre_changeaddr_info info = {
9003 .info.dev = dev,
9004 .info.extack = extack,
9005 .dev_addr = addr,
9006 };
9007 int rc;
9008
9009 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9010 return notifier_to_errno(rc);
9011 }
9012 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9013
9014 /**
9015 * dev_set_mac_address - Change Media Access Control Address
9016 * @dev: device
9017 * @sa: new address
9018 * @extack: netlink extended ack
9019 *
9020 * Change the hardware (MAC) address of the device
9021 */
dev_set_mac_address(struct net_device * dev,struct sockaddr * sa,struct netlink_ext_ack * extack)9022 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9023 struct netlink_ext_ack *extack)
9024 {
9025 const struct net_device_ops *ops = dev->netdev_ops;
9026 int err;
9027
9028 if (!ops->ndo_set_mac_address)
9029 return -EOPNOTSUPP;
9030 if (sa->sa_family != dev->type)
9031 return -EINVAL;
9032 if (!netif_device_present(dev))
9033 return -ENODEV;
9034 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9035 if (err)
9036 return err;
9037 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) {
9038 err = ops->ndo_set_mac_address(dev, sa);
9039 if (err)
9040 return err;
9041 }
9042 dev->addr_assign_type = NET_ADDR_SET;
9043 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9044 add_device_randomness(dev->dev_addr, dev->addr_len);
9045 return 0;
9046 }
9047 EXPORT_SYMBOL(dev_set_mac_address);
9048
9049 DECLARE_RWSEM(dev_addr_sem);
9050
dev_set_mac_address_user(struct net_device * dev,struct sockaddr * sa,struct netlink_ext_ack * extack)9051 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9052 struct netlink_ext_ack *extack)
9053 {
9054 int ret;
9055
9056 down_write(&dev_addr_sem);
9057 ret = dev_set_mac_address(dev, sa, extack);
9058 up_write(&dev_addr_sem);
9059 return ret;
9060 }
9061 EXPORT_SYMBOL(dev_set_mac_address_user);
9062
dev_get_mac_address(struct sockaddr * sa,struct net * net,char * dev_name)9063 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9064 {
9065 size_t size = sizeof(sa->sa_data_min);
9066 struct net_device *dev;
9067 int ret = 0;
9068
9069 down_read(&dev_addr_sem);
9070 rcu_read_lock();
9071
9072 dev = dev_get_by_name_rcu(net, dev_name);
9073 if (!dev) {
9074 ret = -ENODEV;
9075 goto unlock;
9076 }
9077 if (!dev->addr_len)
9078 memset(sa->sa_data, 0, size);
9079 else
9080 memcpy(sa->sa_data, dev->dev_addr,
9081 min_t(size_t, size, dev->addr_len));
9082 sa->sa_family = dev->type;
9083
9084 unlock:
9085 rcu_read_unlock();
9086 up_read(&dev_addr_sem);
9087 return ret;
9088 }
9089 EXPORT_SYMBOL(dev_get_mac_address);
9090
9091 /**
9092 * dev_change_carrier - Change device carrier
9093 * @dev: device
9094 * @new_carrier: new value
9095 *
9096 * Change device carrier
9097 */
dev_change_carrier(struct net_device * dev,bool new_carrier)9098 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9099 {
9100 const struct net_device_ops *ops = dev->netdev_ops;
9101
9102 if (!ops->ndo_change_carrier)
9103 return -EOPNOTSUPP;
9104 if (!netif_device_present(dev))
9105 return -ENODEV;
9106 return ops->ndo_change_carrier(dev, new_carrier);
9107 }
9108
9109 /**
9110 * dev_get_phys_port_id - Get device physical port ID
9111 * @dev: device
9112 * @ppid: port ID
9113 *
9114 * Get device physical port ID
9115 */
dev_get_phys_port_id(struct net_device * dev,struct netdev_phys_item_id * ppid)9116 int dev_get_phys_port_id(struct net_device *dev,
9117 struct netdev_phys_item_id *ppid)
9118 {
9119 const struct net_device_ops *ops = dev->netdev_ops;
9120
9121 if (!ops->ndo_get_phys_port_id)
9122 return -EOPNOTSUPP;
9123 return ops->ndo_get_phys_port_id(dev, ppid);
9124 }
9125
9126 /**
9127 * dev_get_phys_port_name - Get device physical port name
9128 * @dev: device
9129 * @name: port name
9130 * @len: limit of bytes to copy to name
9131 *
9132 * Get device physical port name
9133 */
dev_get_phys_port_name(struct net_device * dev,char * name,size_t len)9134 int dev_get_phys_port_name(struct net_device *dev,
9135 char *name, size_t len)
9136 {
9137 const struct net_device_ops *ops = dev->netdev_ops;
9138 int err;
9139
9140 if (ops->ndo_get_phys_port_name) {
9141 err = ops->ndo_get_phys_port_name(dev, name, len);
9142 if (err != -EOPNOTSUPP)
9143 return err;
9144 }
9145 return devlink_compat_phys_port_name_get(dev, name, len);
9146 }
9147
9148 /**
9149 * dev_get_port_parent_id - Get the device's port parent identifier
9150 * @dev: network device
9151 * @ppid: pointer to a storage for the port's parent identifier
9152 * @recurse: allow/disallow recursion to lower devices
9153 *
9154 * Get the devices's port parent identifier
9155 */
dev_get_port_parent_id(struct net_device * dev,struct netdev_phys_item_id * ppid,bool recurse)9156 int dev_get_port_parent_id(struct net_device *dev,
9157 struct netdev_phys_item_id *ppid,
9158 bool recurse)
9159 {
9160 const struct net_device_ops *ops = dev->netdev_ops;
9161 struct netdev_phys_item_id first = { };
9162 struct net_device *lower_dev;
9163 struct list_head *iter;
9164 int err;
9165
9166 if (ops->ndo_get_port_parent_id) {
9167 err = ops->ndo_get_port_parent_id(dev, ppid);
9168 if (err != -EOPNOTSUPP)
9169 return err;
9170 }
9171
9172 err = devlink_compat_switch_id_get(dev, ppid);
9173 if (!recurse || err != -EOPNOTSUPP)
9174 return err;
9175
9176 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9177 err = dev_get_port_parent_id(lower_dev, ppid, true);
9178 if (err)
9179 break;
9180 if (!first.id_len)
9181 first = *ppid;
9182 else if (memcmp(&first, ppid, sizeof(*ppid)))
9183 return -EOPNOTSUPP;
9184 }
9185
9186 return err;
9187 }
9188 EXPORT_SYMBOL(dev_get_port_parent_id);
9189
9190 /**
9191 * netdev_port_same_parent_id - Indicate if two network devices have
9192 * the same port parent identifier
9193 * @a: first network device
9194 * @b: second network device
9195 */
netdev_port_same_parent_id(struct net_device * a,struct net_device * b)9196 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9197 {
9198 struct netdev_phys_item_id a_id = { };
9199 struct netdev_phys_item_id b_id = { };
9200
9201 if (dev_get_port_parent_id(a, &a_id, true) ||
9202 dev_get_port_parent_id(b, &b_id, true))
9203 return false;
9204
9205 return netdev_phys_item_id_same(&a_id, &b_id);
9206 }
9207 EXPORT_SYMBOL(netdev_port_same_parent_id);
9208
9209 /**
9210 * dev_change_proto_down - set carrier according to proto_down.
9211 *
9212 * @dev: device
9213 * @proto_down: new value
9214 */
dev_change_proto_down(struct net_device * dev,bool proto_down)9215 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9216 {
9217 if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
9218 return -EOPNOTSUPP;
9219 if (!netif_device_present(dev))
9220 return -ENODEV;
9221 if (proto_down)
9222 netif_carrier_off(dev);
9223 else
9224 netif_carrier_on(dev);
9225 WRITE_ONCE(dev->proto_down, proto_down);
9226 return 0;
9227 }
9228
9229 /**
9230 * dev_change_proto_down_reason - proto down reason
9231 *
9232 * @dev: device
9233 * @mask: proto down mask
9234 * @value: proto down value
9235 */
dev_change_proto_down_reason(struct net_device * dev,unsigned long mask,u32 value)9236 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9237 u32 value)
9238 {
9239 u32 proto_down_reason;
9240 int b;
9241
9242 if (!mask) {
9243 proto_down_reason = value;
9244 } else {
9245 proto_down_reason = dev->proto_down_reason;
9246 for_each_set_bit(b, &mask, 32) {
9247 if (value & (1 << b))
9248 proto_down_reason |= BIT(b);
9249 else
9250 proto_down_reason &= ~BIT(b);
9251 }
9252 }
9253 WRITE_ONCE(dev->proto_down_reason, proto_down_reason);
9254 }
9255
9256 struct bpf_xdp_link {
9257 struct bpf_link link;
9258 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9259 int flags;
9260 };
9261
dev_xdp_mode(struct net_device * dev,u32 flags)9262 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9263 {
9264 if (flags & XDP_FLAGS_HW_MODE)
9265 return XDP_MODE_HW;
9266 if (flags & XDP_FLAGS_DRV_MODE)
9267 return XDP_MODE_DRV;
9268 if (flags & XDP_FLAGS_SKB_MODE)
9269 return XDP_MODE_SKB;
9270 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9271 }
9272
dev_xdp_bpf_op(struct net_device * dev,enum bpf_xdp_mode mode)9273 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9274 {
9275 switch (mode) {
9276 case XDP_MODE_SKB:
9277 return generic_xdp_install;
9278 case XDP_MODE_DRV:
9279 case XDP_MODE_HW:
9280 return dev->netdev_ops->ndo_bpf;
9281 default:
9282 return NULL;
9283 }
9284 }
9285
dev_xdp_link(struct net_device * dev,enum bpf_xdp_mode mode)9286 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9287 enum bpf_xdp_mode mode)
9288 {
9289 return dev->xdp_state[mode].link;
9290 }
9291
dev_xdp_prog(struct net_device * dev,enum bpf_xdp_mode mode)9292 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9293 enum bpf_xdp_mode mode)
9294 {
9295 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9296
9297 if (link)
9298 return link->link.prog;
9299 return dev->xdp_state[mode].prog;
9300 }
9301
dev_xdp_prog_count(struct net_device * dev)9302 u8 dev_xdp_prog_count(struct net_device *dev)
9303 {
9304 u8 count = 0;
9305 int i;
9306
9307 for (i = 0; i < __MAX_XDP_MODE; i++)
9308 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9309 count++;
9310 return count;
9311 }
9312 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9313
dev_xdp_prog_id(struct net_device * dev,enum bpf_xdp_mode mode)9314 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9315 {
9316 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9317
9318 return prog ? prog->aux->id : 0;
9319 }
9320
dev_xdp_set_link(struct net_device * dev,enum bpf_xdp_mode mode,struct bpf_xdp_link * link)9321 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9322 struct bpf_xdp_link *link)
9323 {
9324 dev->xdp_state[mode].link = link;
9325 dev->xdp_state[mode].prog = NULL;
9326 }
9327
dev_xdp_set_prog(struct net_device * dev,enum bpf_xdp_mode mode,struct bpf_prog * prog)9328 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9329 struct bpf_prog *prog)
9330 {
9331 dev->xdp_state[mode].link = NULL;
9332 dev->xdp_state[mode].prog = prog;
9333 }
9334
dev_xdp_install(struct net_device * dev,enum bpf_xdp_mode mode,bpf_op_t bpf_op,struct netlink_ext_ack * extack,u32 flags,struct bpf_prog * prog)9335 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9336 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9337 u32 flags, struct bpf_prog *prog)
9338 {
9339 struct netdev_bpf xdp;
9340 int err;
9341
9342 memset(&xdp, 0, sizeof(xdp));
9343 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9344 xdp.extack = extack;
9345 xdp.flags = flags;
9346 xdp.prog = prog;
9347
9348 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9349 * "moved" into driver), so they don't increment it on their own, but
9350 * they do decrement refcnt when program is detached or replaced.
9351 * Given net_device also owns link/prog, we need to bump refcnt here
9352 * to prevent drivers from underflowing it.
9353 */
9354 if (prog)
9355 bpf_prog_inc(prog);
9356 err = bpf_op(dev, &xdp);
9357 if (err) {
9358 if (prog)
9359 bpf_prog_put(prog);
9360 return err;
9361 }
9362
9363 if (mode != XDP_MODE_HW)
9364 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9365
9366 return 0;
9367 }
9368
dev_xdp_uninstall(struct net_device * dev)9369 static void dev_xdp_uninstall(struct net_device *dev)
9370 {
9371 struct bpf_xdp_link *link;
9372 struct bpf_prog *prog;
9373 enum bpf_xdp_mode mode;
9374 bpf_op_t bpf_op;
9375
9376 ASSERT_RTNL();
9377
9378 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9379 prog = dev_xdp_prog(dev, mode);
9380 if (!prog)
9381 continue;
9382
9383 bpf_op = dev_xdp_bpf_op(dev, mode);
9384 if (!bpf_op)
9385 continue;
9386
9387 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9388
9389 /* auto-detach link from net device */
9390 link = dev_xdp_link(dev, mode);
9391 if (link)
9392 link->dev = NULL;
9393 else
9394 bpf_prog_put(prog);
9395
9396 dev_xdp_set_link(dev, mode, NULL);
9397 }
9398 }
9399
dev_xdp_attach(struct net_device * dev,struct netlink_ext_ack * extack,struct bpf_xdp_link * link,struct bpf_prog * new_prog,struct bpf_prog * old_prog,u32 flags)9400 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9401 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9402 struct bpf_prog *old_prog, u32 flags)
9403 {
9404 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9405 struct bpf_prog *cur_prog;
9406 struct net_device *upper;
9407 struct list_head *iter;
9408 enum bpf_xdp_mode mode;
9409 bpf_op_t bpf_op;
9410 int err;
9411
9412 ASSERT_RTNL();
9413
9414 /* either link or prog attachment, never both */
9415 if (link && (new_prog || old_prog))
9416 return -EINVAL;
9417 /* link supports only XDP mode flags */
9418 if (link && (flags & ~XDP_FLAGS_MODES)) {
9419 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9420 return -EINVAL;
9421 }
9422 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9423 if (num_modes > 1) {
9424 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9425 return -EINVAL;
9426 }
9427 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9428 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9429 NL_SET_ERR_MSG(extack,
9430 "More than one program loaded, unset mode is ambiguous");
9431 return -EINVAL;
9432 }
9433 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9434 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9435 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9436 return -EINVAL;
9437 }
9438
9439 mode = dev_xdp_mode(dev, flags);
9440 /* can't replace attached link */
9441 if (dev_xdp_link(dev, mode)) {
9442 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9443 return -EBUSY;
9444 }
9445
9446 /* don't allow if an upper device already has a program */
9447 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9448 if (dev_xdp_prog_count(upper) > 0) {
9449 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9450 return -EEXIST;
9451 }
9452 }
9453
9454 cur_prog = dev_xdp_prog(dev, mode);
9455 /* can't replace attached prog with link */
9456 if (link && cur_prog) {
9457 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9458 return -EBUSY;
9459 }
9460 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9461 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9462 return -EEXIST;
9463 }
9464
9465 /* put effective new program into new_prog */
9466 if (link)
9467 new_prog = link->link.prog;
9468
9469 if (new_prog) {
9470 bool offload = mode == XDP_MODE_HW;
9471 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9472 ? XDP_MODE_DRV : XDP_MODE_SKB;
9473
9474 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9475 NL_SET_ERR_MSG(extack, "XDP program already attached");
9476 return -EBUSY;
9477 }
9478 if (!offload && dev_xdp_prog(dev, other_mode)) {
9479 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9480 return -EEXIST;
9481 }
9482 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9483 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9484 return -EINVAL;
9485 }
9486 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9487 NL_SET_ERR_MSG(extack, "Program bound to different device");
9488 return -EINVAL;
9489 }
9490 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9491 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9492 return -EINVAL;
9493 }
9494 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9495 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9496 return -EINVAL;
9497 }
9498 }
9499
9500 /* don't call drivers if the effective program didn't change */
9501 if (new_prog != cur_prog) {
9502 bpf_op = dev_xdp_bpf_op(dev, mode);
9503 if (!bpf_op) {
9504 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9505 return -EOPNOTSUPP;
9506 }
9507
9508 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9509 if (err)
9510 return err;
9511 }
9512
9513 if (link)
9514 dev_xdp_set_link(dev, mode, link);
9515 else
9516 dev_xdp_set_prog(dev, mode, new_prog);
9517 if (cur_prog)
9518 bpf_prog_put(cur_prog);
9519
9520 return 0;
9521 }
9522
dev_xdp_attach_link(struct net_device * dev,struct netlink_ext_ack * extack,struct bpf_xdp_link * link)9523 static int dev_xdp_attach_link(struct net_device *dev,
9524 struct netlink_ext_ack *extack,
9525 struct bpf_xdp_link *link)
9526 {
9527 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9528 }
9529
dev_xdp_detach_link(struct net_device * dev,struct netlink_ext_ack * extack,struct bpf_xdp_link * link)9530 static int dev_xdp_detach_link(struct net_device *dev,
9531 struct netlink_ext_ack *extack,
9532 struct bpf_xdp_link *link)
9533 {
9534 enum bpf_xdp_mode mode;
9535 bpf_op_t bpf_op;
9536
9537 ASSERT_RTNL();
9538
9539 mode = dev_xdp_mode(dev, link->flags);
9540 if (dev_xdp_link(dev, mode) != link)
9541 return -EINVAL;
9542
9543 bpf_op = dev_xdp_bpf_op(dev, mode);
9544 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9545 dev_xdp_set_link(dev, mode, NULL);
9546 return 0;
9547 }
9548
bpf_xdp_link_release(struct bpf_link * link)9549 static void bpf_xdp_link_release(struct bpf_link *link)
9550 {
9551 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9552
9553 rtnl_lock();
9554
9555 /* if racing with net_device's tear down, xdp_link->dev might be
9556 * already NULL, in which case link was already auto-detached
9557 */
9558 if (xdp_link->dev) {
9559 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9560 xdp_link->dev = NULL;
9561 }
9562
9563 rtnl_unlock();
9564 }
9565
bpf_xdp_link_detach(struct bpf_link * link)9566 static int bpf_xdp_link_detach(struct bpf_link *link)
9567 {
9568 bpf_xdp_link_release(link);
9569 return 0;
9570 }
9571
bpf_xdp_link_dealloc(struct bpf_link * link)9572 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9573 {
9574 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9575
9576 kfree(xdp_link);
9577 }
9578
bpf_xdp_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)9579 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9580 struct seq_file *seq)
9581 {
9582 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9583 u32 ifindex = 0;
9584
9585 rtnl_lock();
9586 if (xdp_link->dev)
9587 ifindex = xdp_link->dev->ifindex;
9588 rtnl_unlock();
9589
9590 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9591 }
9592
bpf_xdp_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)9593 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9594 struct bpf_link_info *info)
9595 {
9596 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9597 u32 ifindex = 0;
9598
9599 rtnl_lock();
9600 if (xdp_link->dev)
9601 ifindex = xdp_link->dev->ifindex;
9602 rtnl_unlock();
9603
9604 info->xdp.ifindex = ifindex;
9605 return 0;
9606 }
9607
bpf_xdp_link_update(struct bpf_link * link,struct bpf_prog * new_prog,struct bpf_prog * old_prog)9608 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9609 struct bpf_prog *old_prog)
9610 {
9611 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9612 enum bpf_xdp_mode mode;
9613 bpf_op_t bpf_op;
9614 int err = 0;
9615
9616 rtnl_lock();
9617
9618 /* link might have been auto-released already, so fail */
9619 if (!xdp_link->dev) {
9620 err = -ENOLINK;
9621 goto out_unlock;
9622 }
9623
9624 if (old_prog && link->prog != old_prog) {
9625 err = -EPERM;
9626 goto out_unlock;
9627 }
9628 old_prog = link->prog;
9629 if (old_prog->type != new_prog->type ||
9630 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9631 err = -EINVAL;
9632 goto out_unlock;
9633 }
9634
9635 if (old_prog == new_prog) {
9636 /* no-op, don't disturb drivers */
9637 bpf_prog_put(new_prog);
9638 goto out_unlock;
9639 }
9640
9641 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9642 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9643 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9644 xdp_link->flags, new_prog);
9645 if (err)
9646 goto out_unlock;
9647
9648 old_prog = xchg(&link->prog, new_prog);
9649 bpf_prog_put(old_prog);
9650
9651 out_unlock:
9652 rtnl_unlock();
9653 return err;
9654 }
9655
9656 static const struct bpf_link_ops bpf_xdp_link_lops = {
9657 .release = bpf_xdp_link_release,
9658 .dealloc = bpf_xdp_link_dealloc,
9659 .detach = bpf_xdp_link_detach,
9660 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9661 .fill_link_info = bpf_xdp_link_fill_link_info,
9662 .update_prog = bpf_xdp_link_update,
9663 };
9664
bpf_xdp_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)9665 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9666 {
9667 struct net *net = current->nsproxy->net_ns;
9668 struct bpf_link_primer link_primer;
9669 struct netlink_ext_ack extack = {};
9670 struct bpf_xdp_link *link;
9671 struct net_device *dev;
9672 int err, fd;
9673
9674 rtnl_lock();
9675 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9676 if (!dev) {
9677 rtnl_unlock();
9678 return -EINVAL;
9679 }
9680
9681 link = kzalloc(sizeof(*link), GFP_USER);
9682 if (!link) {
9683 err = -ENOMEM;
9684 goto unlock;
9685 }
9686
9687 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9688 link->dev = dev;
9689 link->flags = attr->link_create.flags;
9690
9691 err = bpf_link_prime(&link->link, &link_primer);
9692 if (err) {
9693 kfree(link);
9694 goto unlock;
9695 }
9696
9697 err = dev_xdp_attach_link(dev, &extack, link);
9698 rtnl_unlock();
9699
9700 if (err) {
9701 link->dev = NULL;
9702 bpf_link_cleanup(&link_primer);
9703 trace_bpf_xdp_link_attach_failed(extack._msg);
9704 goto out_put_dev;
9705 }
9706
9707 fd = bpf_link_settle(&link_primer);
9708 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9709 dev_put(dev);
9710 return fd;
9711
9712 unlock:
9713 rtnl_unlock();
9714
9715 out_put_dev:
9716 dev_put(dev);
9717 return err;
9718 }
9719
9720 /**
9721 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9722 * @dev: device
9723 * @extack: netlink extended ack
9724 * @fd: new program fd or negative value to clear
9725 * @expected_fd: old program fd that userspace expects to replace or clear
9726 * @flags: xdp-related flags
9727 *
9728 * Set or clear a bpf program for a device
9729 */
dev_change_xdp_fd(struct net_device * dev,struct netlink_ext_ack * extack,int fd,int expected_fd,u32 flags)9730 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9731 int fd, int expected_fd, u32 flags)
9732 {
9733 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9734 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9735 int err;
9736
9737 ASSERT_RTNL();
9738
9739 if (fd >= 0) {
9740 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9741 mode != XDP_MODE_SKB);
9742 if (IS_ERR(new_prog))
9743 return PTR_ERR(new_prog);
9744 }
9745
9746 if (expected_fd >= 0) {
9747 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9748 mode != XDP_MODE_SKB);
9749 if (IS_ERR(old_prog)) {
9750 err = PTR_ERR(old_prog);
9751 old_prog = NULL;
9752 goto err_out;
9753 }
9754 }
9755
9756 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9757
9758 err_out:
9759 if (err && new_prog)
9760 bpf_prog_put(new_prog);
9761 if (old_prog)
9762 bpf_prog_put(old_prog);
9763 return err;
9764 }
9765
9766 /**
9767 * dev_index_reserve() - allocate an ifindex in a namespace
9768 * @net: the applicable net namespace
9769 * @ifindex: requested ifindex, pass %0 to get one allocated
9770 *
9771 * Allocate a ifindex for a new device. Caller must either use the ifindex
9772 * to store the device (via list_netdevice()) or call dev_index_release()
9773 * to give the index up.
9774 *
9775 * Return: a suitable unique value for a new device interface number or -errno.
9776 */
dev_index_reserve(struct net * net,u32 ifindex)9777 static int dev_index_reserve(struct net *net, u32 ifindex)
9778 {
9779 int err;
9780
9781 if (ifindex > INT_MAX) {
9782 DEBUG_NET_WARN_ON_ONCE(1);
9783 return -EINVAL;
9784 }
9785
9786 if (!ifindex)
9787 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL,
9788 xa_limit_31b, &net->ifindex, GFP_KERNEL);
9789 else
9790 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL);
9791 if (err < 0)
9792 return err;
9793
9794 return ifindex;
9795 }
9796
dev_index_release(struct net * net,int ifindex)9797 static void dev_index_release(struct net *net, int ifindex)
9798 {
9799 /* Expect only unused indexes, unlist_netdevice() removes the used */
9800 WARN_ON(xa_erase(&net->dev_by_index, ifindex));
9801 }
9802
9803 /* Delayed registration/unregisteration */
9804 LIST_HEAD(net_todo_list);
9805 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9806 atomic_t dev_unreg_count = ATOMIC_INIT(0);
9807
net_set_todo(struct net_device * dev)9808 static void net_set_todo(struct net_device *dev)
9809 {
9810 list_add_tail(&dev->todo_list, &net_todo_list);
9811 }
9812
netdev_sync_upper_features(struct net_device * lower,struct net_device * upper,netdev_features_t features)9813 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9814 struct net_device *upper, netdev_features_t features)
9815 {
9816 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9817 netdev_features_t feature;
9818 int feature_bit;
9819
9820 for_each_netdev_feature(upper_disables, feature_bit) {
9821 feature = __NETIF_F_BIT(feature_bit);
9822 if (!(upper->wanted_features & feature)
9823 && (features & feature)) {
9824 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9825 &feature, upper->name);
9826 features &= ~feature;
9827 }
9828 }
9829
9830 return features;
9831 }
9832
netdev_sync_lower_features(struct net_device * upper,struct net_device * lower,netdev_features_t features)9833 static void netdev_sync_lower_features(struct net_device *upper,
9834 struct net_device *lower, netdev_features_t features)
9835 {
9836 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9837 netdev_features_t feature;
9838 int feature_bit;
9839
9840 for_each_netdev_feature(upper_disables, feature_bit) {
9841 feature = __NETIF_F_BIT(feature_bit);
9842 if (!(features & feature) && (lower->features & feature)) {
9843 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9844 &feature, lower->name);
9845 lower->wanted_features &= ~feature;
9846 __netdev_update_features(lower);
9847
9848 if (unlikely(lower->features & feature))
9849 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9850 &feature, lower->name);
9851 else
9852 netdev_features_change(lower);
9853 }
9854 }
9855 }
9856
netdev_fix_features(struct net_device * dev,netdev_features_t features)9857 static netdev_features_t netdev_fix_features(struct net_device *dev,
9858 netdev_features_t features)
9859 {
9860 /* Fix illegal checksum combinations */
9861 if ((features & NETIF_F_HW_CSUM) &&
9862 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9863 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9864 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9865 }
9866
9867 /* TSO requires that SG is present as well. */
9868 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9869 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9870 features &= ~NETIF_F_ALL_TSO;
9871 }
9872
9873 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9874 !(features & NETIF_F_IP_CSUM)) {
9875 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9876 features &= ~NETIF_F_TSO;
9877 features &= ~NETIF_F_TSO_ECN;
9878 }
9879
9880 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9881 !(features & NETIF_F_IPV6_CSUM)) {
9882 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9883 features &= ~NETIF_F_TSO6;
9884 }
9885
9886 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9887 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9888 features &= ~NETIF_F_TSO_MANGLEID;
9889
9890 /* TSO ECN requires that TSO is present as well. */
9891 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9892 features &= ~NETIF_F_TSO_ECN;
9893
9894 /* Software GSO depends on SG. */
9895 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9896 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9897 features &= ~NETIF_F_GSO;
9898 }
9899
9900 /* GSO partial features require GSO partial be set */
9901 if ((features & dev->gso_partial_features) &&
9902 !(features & NETIF_F_GSO_PARTIAL)) {
9903 netdev_dbg(dev,
9904 "Dropping partially supported GSO features since no GSO partial.\n");
9905 features &= ~dev->gso_partial_features;
9906 }
9907
9908 if (!(features & NETIF_F_RXCSUM)) {
9909 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9910 * successfully merged by hardware must also have the
9911 * checksum verified by hardware. If the user does not
9912 * want to enable RXCSUM, logically, we should disable GRO_HW.
9913 */
9914 if (features & NETIF_F_GRO_HW) {
9915 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9916 features &= ~NETIF_F_GRO_HW;
9917 }
9918 }
9919
9920 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9921 if (features & NETIF_F_RXFCS) {
9922 if (features & NETIF_F_LRO) {
9923 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9924 features &= ~NETIF_F_LRO;
9925 }
9926
9927 if (features & NETIF_F_GRO_HW) {
9928 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9929 features &= ~NETIF_F_GRO_HW;
9930 }
9931 }
9932
9933 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9934 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9935 features &= ~NETIF_F_LRO;
9936 }
9937
9938 if (features & NETIF_F_HW_TLS_TX) {
9939 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9940 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9941 bool hw_csum = features & NETIF_F_HW_CSUM;
9942
9943 if (!ip_csum && !hw_csum) {
9944 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9945 features &= ~NETIF_F_HW_TLS_TX;
9946 }
9947 }
9948
9949 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9950 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9951 features &= ~NETIF_F_HW_TLS_RX;
9952 }
9953
9954 return features;
9955 }
9956
__netdev_update_features(struct net_device * dev)9957 int __netdev_update_features(struct net_device *dev)
9958 {
9959 struct net_device *upper, *lower;
9960 netdev_features_t features;
9961 struct list_head *iter;
9962 int err = -1;
9963
9964 ASSERT_RTNL();
9965
9966 features = netdev_get_wanted_features(dev);
9967
9968 if (dev->netdev_ops->ndo_fix_features)
9969 features = dev->netdev_ops->ndo_fix_features(dev, features);
9970
9971 /* driver might be less strict about feature dependencies */
9972 features = netdev_fix_features(dev, features);
9973
9974 /* some features can't be enabled if they're off on an upper device */
9975 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9976 features = netdev_sync_upper_features(dev, upper, features);
9977
9978 if (dev->features == features)
9979 goto sync_lower;
9980
9981 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9982 &dev->features, &features);
9983
9984 if (dev->netdev_ops->ndo_set_features)
9985 err = dev->netdev_ops->ndo_set_features(dev, features);
9986 else
9987 err = 0;
9988
9989 if (unlikely(err < 0)) {
9990 netdev_err(dev,
9991 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9992 err, &features, &dev->features);
9993 /* return non-0 since some features might have changed and
9994 * it's better to fire a spurious notification than miss it
9995 */
9996 return -1;
9997 }
9998
9999 sync_lower:
10000 /* some features must be disabled on lower devices when disabled
10001 * on an upper device (think: bonding master or bridge)
10002 */
10003 netdev_for_each_lower_dev(dev, lower, iter)
10004 netdev_sync_lower_features(dev, lower, features);
10005
10006 if (!err) {
10007 netdev_features_t diff = features ^ dev->features;
10008
10009 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10010 /* udp_tunnel_{get,drop}_rx_info both need
10011 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10012 * device, or they won't do anything.
10013 * Thus we need to update dev->features
10014 * *before* calling udp_tunnel_get_rx_info,
10015 * but *after* calling udp_tunnel_drop_rx_info.
10016 */
10017 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10018 dev->features = features;
10019 udp_tunnel_get_rx_info(dev);
10020 } else {
10021 udp_tunnel_drop_rx_info(dev);
10022 }
10023 }
10024
10025 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10026 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10027 dev->features = features;
10028 err |= vlan_get_rx_ctag_filter_info(dev);
10029 } else {
10030 vlan_drop_rx_ctag_filter_info(dev);
10031 }
10032 }
10033
10034 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10035 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10036 dev->features = features;
10037 err |= vlan_get_rx_stag_filter_info(dev);
10038 } else {
10039 vlan_drop_rx_stag_filter_info(dev);
10040 }
10041 }
10042
10043 dev->features = features;
10044 }
10045
10046 return err < 0 ? 0 : 1;
10047 }
10048
10049 /**
10050 * netdev_update_features - recalculate device features
10051 * @dev: the device to check
10052 *
10053 * Recalculate dev->features set and send notifications if it
10054 * has changed. Should be called after driver or hardware dependent
10055 * conditions might have changed that influence the features.
10056 */
netdev_update_features(struct net_device * dev)10057 void netdev_update_features(struct net_device *dev)
10058 {
10059 if (__netdev_update_features(dev))
10060 netdev_features_change(dev);
10061 }
10062 EXPORT_SYMBOL(netdev_update_features);
10063
10064 /**
10065 * netdev_change_features - recalculate device features
10066 * @dev: the device to check
10067 *
10068 * Recalculate dev->features set and send notifications even
10069 * if they have not changed. Should be called instead of
10070 * netdev_update_features() if also dev->vlan_features might
10071 * have changed to allow the changes to be propagated to stacked
10072 * VLAN devices.
10073 */
netdev_change_features(struct net_device * dev)10074 void netdev_change_features(struct net_device *dev)
10075 {
10076 __netdev_update_features(dev);
10077 netdev_features_change(dev);
10078 }
10079 EXPORT_SYMBOL(netdev_change_features);
10080
10081 /**
10082 * netif_stacked_transfer_operstate - transfer operstate
10083 * @rootdev: the root or lower level device to transfer state from
10084 * @dev: the device to transfer operstate to
10085 *
10086 * Transfer operational state from root to device. This is normally
10087 * called when a stacking relationship exists between the root
10088 * device and the device(a leaf device).
10089 */
netif_stacked_transfer_operstate(const struct net_device * rootdev,struct net_device * dev)10090 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10091 struct net_device *dev)
10092 {
10093 if (rootdev->operstate == IF_OPER_DORMANT)
10094 netif_dormant_on(dev);
10095 else
10096 netif_dormant_off(dev);
10097
10098 if (rootdev->operstate == IF_OPER_TESTING)
10099 netif_testing_on(dev);
10100 else
10101 netif_testing_off(dev);
10102
10103 if (netif_carrier_ok(rootdev))
10104 netif_carrier_on(dev);
10105 else
10106 netif_carrier_off(dev);
10107 }
10108 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10109
netif_alloc_rx_queues(struct net_device * dev)10110 static int netif_alloc_rx_queues(struct net_device *dev)
10111 {
10112 unsigned int i, count = dev->num_rx_queues;
10113 struct netdev_rx_queue *rx;
10114 size_t sz = count * sizeof(*rx);
10115 int err = 0;
10116
10117 BUG_ON(count < 1);
10118
10119 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10120 if (!rx)
10121 return -ENOMEM;
10122
10123 dev->_rx = rx;
10124
10125 for (i = 0; i < count; i++) {
10126 rx[i].dev = dev;
10127
10128 /* XDP RX-queue setup */
10129 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10130 if (err < 0)
10131 goto err_rxq_info;
10132 }
10133 return 0;
10134
10135 err_rxq_info:
10136 /* Rollback successful reg's and free other resources */
10137 while (i--)
10138 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10139 kvfree(dev->_rx);
10140 dev->_rx = NULL;
10141 return err;
10142 }
10143
netif_free_rx_queues(struct net_device * dev)10144 static void netif_free_rx_queues(struct net_device *dev)
10145 {
10146 unsigned int i, count = dev->num_rx_queues;
10147
10148 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10149 if (!dev->_rx)
10150 return;
10151
10152 for (i = 0; i < count; i++)
10153 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10154
10155 kvfree(dev->_rx);
10156 }
10157
netdev_init_one_queue(struct net_device * dev,struct netdev_queue * queue,void * _unused)10158 static void netdev_init_one_queue(struct net_device *dev,
10159 struct netdev_queue *queue, void *_unused)
10160 {
10161 /* Initialize queue lock */
10162 spin_lock_init(&queue->_xmit_lock);
10163 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10164 queue->xmit_lock_owner = -1;
10165 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10166 queue->dev = dev;
10167 #ifdef CONFIG_BQL
10168 dql_init(&queue->dql, HZ);
10169 #endif
10170 }
10171
netif_free_tx_queues(struct net_device * dev)10172 static void netif_free_tx_queues(struct net_device *dev)
10173 {
10174 kvfree(dev->_tx);
10175 }
10176
netif_alloc_netdev_queues(struct net_device * dev)10177 static int netif_alloc_netdev_queues(struct net_device *dev)
10178 {
10179 unsigned int count = dev->num_tx_queues;
10180 struct netdev_queue *tx;
10181 size_t sz = count * sizeof(*tx);
10182
10183 if (count < 1 || count > 0xffff)
10184 return -EINVAL;
10185
10186 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10187 if (!tx)
10188 return -ENOMEM;
10189
10190 dev->_tx = tx;
10191
10192 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10193 spin_lock_init(&dev->tx_global_lock);
10194
10195 return 0;
10196 }
10197
netif_tx_stop_all_queues(struct net_device * dev)10198 void netif_tx_stop_all_queues(struct net_device *dev)
10199 {
10200 unsigned int i;
10201
10202 for (i = 0; i < dev->num_tx_queues; i++) {
10203 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10204
10205 netif_tx_stop_queue(txq);
10206 }
10207 }
10208 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10209
netdev_do_alloc_pcpu_stats(struct net_device * dev)10210 static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
10211 {
10212 void __percpu *v;
10213
10214 /* Drivers implementing ndo_get_peer_dev must support tstat
10215 * accounting, so that skb_do_redirect() can bump the dev's
10216 * RX stats upon network namespace switch.
10217 */
10218 if (dev->netdev_ops->ndo_get_peer_dev &&
10219 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
10220 return -EOPNOTSUPP;
10221
10222 switch (dev->pcpu_stat_type) {
10223 case NETDEV_PCPU_STAT_NONE:
10224 return 0;
10225 case NETDEV_PCPU_STAT_LSTATS:
10226 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
10227 break;
10228 case NETDEV_PCPU_STAT_TSTATS:
10229 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
10230 break;
10231 case NETDEV_PCPU_STAT_DSTATS:
10232 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
10233 break;
10234 default:
10235 return -EINVAL;
10236 }
10237
10238 return v ? 0 : -ENOMEM;
10239 }
10240
netdev_do_free_pcpu_stats(struct net_device * dev)10241 static void netdev_do_free_pcpu_stats(struct net_device *dev)
10242 {
10243 switch (dev->pcpu_stat_type) {
10244 case NETDEV_PCPU_STAT_NONE:
10245 return;
10246 case NETDEV_PCPU_STAT_LSTATS:
10247 free_percpu(dev->lstats);
10248 break;
10249 case NETDEV_PCPU_STAT_TSTATS:
10250 free_percpu(dev->tstats);
10251 break;
10252 case NETDEV_PCPU_STAT_DSTATS:
10253 free_percpu(dev->dstats);
10254 break;
10255 }
10256 }
10257
10258 /**
10259 * register_netdevice() - register a network device
10260 * @dev: device to register
10261 *
10262 * Take a prepared network device structure and make it externally accessible.
10263 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
10264 * Callers must hold the rtnl lock - you may want register_netdev()
10265 * instead of this.
10266 */
register_netdevice(struct net_device * dev)10267 int register_netdevice(struct net_device *dev)
10268 {
10269 int ret;
10270 struct net *net = dev_net(dev);
10271
10272 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10273 NETDEV_FEATURE_COUNT);
10274 BUG_ON(dev_boot_phase);
10275 ASSERT_RTNL();
10276
10277 might_sleep();
10278
10279 /* When net_device's are persistent, this will be fatal. */
10280 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10281 BUG_ON(!net);
10282
10283 ret = ethtool_check_ops(dev->ethtool_ops);
10284 if (ret)
10285 return ret;
10286
10287 spin_lock_init(&dev->addr_list_lock);
10288 netdev_set_addr_lockdep_class(dev);
10289
10290 ret = dev_get_valid_name(net, dev, dev->name);
10291 if (ret < 0)
10292 goto out;
10293
10294 ret = -ENOMEM;
10295 dev->name_node = netdev_name_node_head_alloc(dev);
10296 if (!dev->name_node)
10297 goto out;
10298
10299 /* Init, if this function is available */
10300 if (dev->netdev_ops->ndo_init) {
10301 ret = dev->netdev_ops->ndo_init(dev);
10302 if (ret) {
10303 if (ret > 0)
10304 ret = -EIO;
10305 goto err_free_name;
10306 }
10307 }
10308
10309 if (((dev->hw_features | dev->features) &
10310 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10311 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10312 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10313 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10314 ret = -EINVAL;
10315 goto err_uninit;
10316 }
10317
10318 ret = netdev_do_alloc_pcpu_stats(dev);
10319 if (ret)
10320 goto err_uninit;
10321
10322 ret = dev_index_reserve(net, dev->ifindex);
10323 if (ret < 0)
10324 goto err_free_pcpu;
10325 dev->ifindex = ret;
10326
10327 /* Transfer changeable features to wanted_features and enable
10328 * software offloads (GSO and GRO).
10329 */
10330 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10331 dev->features |= NETIF_F_SOFT_FEATURES;
10332
10333 if (dev->udp_tunnel_nic_info) {
10334 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10335 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10336 }
10337
10338 dev->wanted_features = dev->features & dev->hw_features;
10339
10340 if (!(dev->flags & IFF_LOOPBACK))
10341 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10342
10343 /* If IPv4 TCP segmentation offload is supported we should also
10344 * allow the device to enable segmenting the frame with the option
10345 * of ignoring a static IP ID value. This doesn't enable the
10346 * feature itself but allows the user to enable it later.
10347 */
10348 if (dev->hw_features & NETIF_F_TSO)
10349 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10350 if (dev->vlan_features & NETIF_F_TSO)
10351 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10352 if (dev->mpls_features & NETIF_F_TSO)
10353 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10354 if (dev->hw_enc_features & NETIF_F_TSO)
10355 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10356
10357 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10358 */
10359 dev->vlan_features |= NETIF_F_HIGHDMA;
10360
10361 /* Make NETIF_F_SG inheritable to tunnel devices.
10362 */
10363 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10364
10365 /* Make NETIF_F_SG inheritable to MPLS.
10366 */
10367 dev->mpls_features |= NETIF_F_SG;
10368
10369 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10370 ret = notifier_to_errno(ret);
10371 if (ret)
10372 goto err_ifindex_release;
10373
10374 ret = netdev_register_kobject(dev);
10375
10376 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED);
10377
10378 if (ret)
10379 goto err_uninit_notify;
10380
10381 __netdev_update_features(dev);
10382
10383 /*
10384 * Default initial state at registry is that the
10385 * device is present.
10386 */
10387
10388 set_bit(__LINK_STATE_PRESENT, &dev->state);
10389
10390 linkwatch_init_dev(dev);
10391
10392 dev_init_scheduler(dev);
10393
10394 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10395 list_netdevice(dev);
10396
10397 add_device_randomness(dev->dev_addr, dev->addr_len);
10398
10399 /* If the device has permanent device address, driver should
10400 * set dev_addr and also addr_assign_type should be set to
10401 * NET_ADDR_PERM (default value).
10402 */
10403 if (dev->addr_assign_type == NET_ADDR_PERM)
10404 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10405
10406 /* Notify protocols, that a new device appeared. */
10407 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10408 ret = notifier_to_errno(ret);
10409 if (ret) {
10410 /* Expect explicit free_netdev() on failure */
10411 dev->needs_free_netdev = false;
10412 unregister_netdevice_queue(dev, NULL);
10413 goto out;
10414 }
10415 /*
10416 * Prevent userspace races by waiting until the network
10417 * device is fully setup before sending notifications.
10418 */
10419 if (!dev->rtnl_link_ops ||
10420 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10421 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10422
10423 out:
10424 return ret;
10425
10426 err_uninit_notify:
10427 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10428 err_ifindex_release:
10429 dev_index_release(net, dev->ifindex);
10430 err_free_pcpu:
10431 netdev_do_free_pcpu_stats(dev);
10432 err_uninit:
10433 if (dev->netdev_ops->ndo_uninit)
10434 dev->netdev_ops->ndo_uninit(dev);
10435 if (dev->priv_destructor)
10436 dev->priv_destructor(dev);
10437 err_free_name:
10438 netdev_name_node_free(dev->name_node);
10439 goto out;
10440 }
10441 EXPORT_SYMBOL(register_netdevice);
10442
10443 /* Initialize the core of a dummy net device.
10444 * This is useful if you are calling this function after alloc_netdev(),
10445 * since it does not memset the net_device fields.
10446 */
init_dummy_netdev_core(struct net_device * dev)10447 static void init_dummy_netdev_core(struct net_device *dev)
10448 {
10449 /* make sure we BUG if trying to hit standard
10450 * register/unregister code path
10451 */
10452 dev->reg_state = NETREG_DUMMY;
10453
10454 /* NAPI wants this */
10455 INIT_LIST_HEAD(&dev->napi_list);
10456
10457 /* a dummy interface is started by default */
10458 set_bit(__LINK_STATE_PRESENT, &dev->state);
10459 set_bit(__LINK_STATE_START, &dev->state);
10460
10461 /* napi_busy_loop stats accounting wants this */
10462 dev_net_set(dev, &init_net);
10463
10464 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10465 * because users of this 'device' dont need to change
10466 * its refcount.
10467 */
10468 }
10469
10470 /**
10471 * init_dummy_netdev - init a dummy network device for NAPI
10472 * @dev: device to init
10473 *
10474 * This takes a network device structure and initializes the minimum
10475 * amount of fields so it can be used to schedule NAPI polls without
10476 * registering a full blown interface. This is to be used by drivers
10477 * that need to tie several hardware interfaces to a single NAPI
10478 * poll scheduler due to HW limitations.
10479 */
init_dummy_netdev(struct net_device * dev)10480 void init_dummy_netdev(struct net_device *dev)
10481 {
10482 /* Clear everything. Note we don't initialize spinlocks
10483 * as they aren't supposed to be taken by any of the
10484 * NAPI code and this dummy netdev is supposed to be
10485 * only ever used for NAPI polls
10486 */
10487 memset(dev, 0, sizeof(struct net_device));
10488 init_dummy_netdev_core(dev);
10489 }
10490 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10491
10492 /**
10493 * register_netdev - register a network device
10494 * @dev: device to register
10495 *
10496 * Take a completed network device structure and add it to the kernel
10497 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10498 * chain. 0 is returned on success. A negative errno code is returned
10499 * on a failure to set up the device, or if the name is a duplicate.
10500 *
10501 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10502 * and expands the device name if you passed a format string to
10503 * alloc_netdev.
10504 */
register_netdev(struct net_device * dev)10505 int register_netdev(struct net_device *dev)
10506 {
10507 int err;
10508
10509 if (rtnl_lock_killable())
10510 return -EINTR;
10511 err = register_netdevice(dev);
10512 rtnl_unlock();
10513 return err;
10514 }
10515 EXPORT_SYMBOL(register_netdev);
10516
netdev_refcnt_read(const struct net_device * dev)10517 int netdev_refcnt_read(const struct net_device *dev)
10518 {
10519 #ifdef CONFIG_PCPU_DEV_REFCNT
10520 int i, refcnt = 0;
10521
10522 for_each_possible_cpu(i)
10523 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10524 return refcnt;
10525 #else
10526 return refcount_read(&dev->dev_refcnt);
10527 #endif
10528 }
10529 EXPORT_SYMBOL(netdev_refcnt_read);
10530
10531 int netdev_unregister_timeout_secs __read_mostly = 10;
10532
10533 #define WAIT_REFS_MIN_MSECS 1
10534 #define WAIT_REFS_MAX_MSECS 250
10535 /**
10536 * netdev_wait_allrefs_any - wait until all references are gone.
10537 * @list: list of net_devices to wait on
10538 *
10539 * This is called when unregistering network devices.
10540 *
10541 * Any protocol or device that holds a reference should register
10542 * for netdevice notification, and cleanup and put back the
10543 * reference if they receive an UNREGISTER event.
10544 * We can get stuck here if buggy protocols don't correctly
10545 * call dev_put.
10546 */
netdev_wait_allrefs_any(struct list_head * list)10547 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10548 {
10549 unsigned long rebroadcast_time, warning_time;
10550 struct net_device *dev;
10551 int wait = 0;
10552
10553 rebroadcast_time = warning_time = jiffies;
10554
10555 list_for_each_entry(dev, list, todo_list)
10556 if (netdev_refcnt_read(dev) == 1)
10557 return dev;
10558
10559 while (true) {
10560 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10561 rtnl_lock();
10562
10563 /* Rebroadcast unregister notification */
10564 list_for_each_entry(dev, list, todo_list)
10565 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10566
10567 __rtnl_unlock();
10568 rcu_barrier();
10569 rtnl_lock();
10570
10571 list_for_each_entry(dev, list, todo_list)
10572 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10573 &dev->state)) {
10574 /* We must not have linkwatch events
10575 * pending on unregister. If this
10576 * happens, we simply run the queue
10577 * unscheduled, resulting in a noop
10578 * for this device.
10579 */
10580 linkwatch_run_queue();
10581 break;
10582 }
10583
10584 __rtnl_unlock();
10585
10586 rebroadcast_time = jiffies;
10587 }
10588
10589 rcu_barrier();
10590
10591 if (!wait) {
10592 wait = WAIT_REFS_MIN_MSECS;
10593 } else {
10594 msleep(wait);
10595 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10596 }
10597
10598 list_for_each_entry(dev, list, todo_list)
10599 if (netdev_refcnt_read(dev) == 1)
10600 return dev;
10601
10602 if (time_after(jiffies, warning_time +
10603 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10604 list_for_each_entry(dev, list, todo_list) {
10605 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10606 dev->name, netdev_refcnt_read(dev));
10607 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10608 }
10609
10610 warning_time = jiffies;
10611 }
10612 }
10613 }
10614
10615 /* The sequence is:
10616 *
10617 * rtnl_lock();
10618 * ...
10619 * register_netdevice(x1);
10620 * register_netdevice(x2);
10621 * ...
10622 * unregister_netdevice(y1);
10623 * unregister_netdevice(y2);
10624 * ...
10625 * rtnl_unlock();
10626 * free_netdev(y1);
10627 * free_netdev(y2);
10628 *
10629 * We are invoked by rtnl_unlock().
10630 * This allows us to deal with problems:
10631 * 1) We can delete sysfs objects which invoke hotplug
10632 * without deadlocking with linkwatch via keventd.
10633 * 2) Since we run with the RTNL semaphore not held, we can sleep
10634 * safely in order to wait for the netdev refcnt to drop to zero.
10635 *
10636 * We must not return until all unregister events added during
10637 * the interval the lock was held have been completed.
10638 */
netdev_run_todo(void)10639 void netdev_run_todo(void)
10640 {
10641 struct net_device *dev, *tmp;
10642 struct list_head list;
10643 int cnt;
10644 #ifdef CONFIG_LOCKDEP
10645 struct list_head unlink_list;
10646
10647 list_replace_init(&net_unlink_list, &unlink_list);
10648
10649 while (!list_empty(&unlink_list)) {
10650 struct net_device *dev = list_first_entry(&unlink_list,
10651 struct net_device,
10652 unlink_list);
10653 list_del_init(&dev->unlink_list);
10654 dev->nested_level = dev->lower_level - 1;
10655 }
10656 #endif
10657
10658 /* Snapshot list, allow later requests */
10659 list_replace_init(&net_todo_list, &list);
10660
10661 __rtnl_unlock();
10662
10663 /* Wait for rcu callbacks to finish before next phase */
10664 if (!list_empty(&list))
10665 rcu_barrier();
10666
10667 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10668 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10669 netdev_WARN(dev, "run_todo but not unregistering\n");
10670 list_del(&dev->todo_list);
10671 continue;
10672 }
10673
10674 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED);
10675 linkwatch_sync_dev(dev);
10676 }
10677
10678 cnt = 0;
10679 while (!list_empty(&list)) {
10680 dev = netdev_wait_allrefs_any(&list);
10681 list_del(&dev->todo_list);
10682
10683 /* paranoia */
10684 BUG_ON(netdev_refcnt_read(dev) != 1);
10685 BUG_ON(!list_empty(&dev->ptype_all));
10686 BUG_ON(!list_empty(&dev->ptype_specific));
10687 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10688 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10689
10690 netdev_do_free_pcpu_stats(dev);
10691 if (dev->priv_destructor)
10692 dev->priv_destructor(dev);
10693 if (dev->needs_free_netdev)
10694 free_netdev(dev);
10695
10696 cnt++;
10697
10698 /* Free network device */
10699 kobject_put(&dev->dev.kobj);
10700 }
10701 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count))
10702 wake_up(&netdev_unregistering_wq);
10703 }
10704
10705 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10706 * all the same fields in the same order as net_device_stats, with only
10707 * the type differing, but rtnl_link_stats64 may have additional fields
10708 * at the end for newer counters.
10709 */
netdev_stats_to_stats64(struct rtnl_link_stats64 * stats64,const struct net_device_stats * netdev_stats)10710 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10711 const struct net_device_stats *netdev_stats)
10712 {
10713 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10714 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10715 u64 *dst = (u64 *)stats64;
10716
10717 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10718 for (i = 0; i < n; i++)
10719 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10720 /* zero out counters that only exist in rtnl_link_stats64 */
10721 memset((char *)stats64 + n * sizeof(u64), 0,
10722 sizeof(*stats64) - n * sizeof(u64));
10723 }
10724 EXPORT_SYMBOL(netdev_stats_to_stats64);
10725
netdev_core_stats_alloc(struct net_device * dev)10726 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc(
10727 struct net_device *dev)
10728 {
10729 struct net_device_core_stats __percpu *p;
10730
10731 p = alloc_percpu_gfp(struct net_device_core_stats,
10732 GFP_ATOMIC | __GFP_NOWARN);
10733
10734 if (p && cmpxchg(&dev->core_stats, NULL, p))
10735 free_percpu(p);
10736
10737 /* This READ_ONCE() pairs with the cmpxchg() above */
10738 return READ_ONCE(dev->core_stats);
10739 }
10740
netdev_core_stats_inc(struct net_device * dev,u32 offset)10741 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset)
10742 {
10743 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10744 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
10745 unsigned long __percpu *field;
10746
10747 if (unlikely(!p)) {
10748 p = netdev_core_stats_alloc(dev);
10749 if (!p)
10750 return;
10751 }
10752
10753 field = (__force unsigned long __percpu *)((__force void *)p + offset);
10754 this_cpu_inc(*field);
10755 }
10756 EXPORT_SYMBOL_GPL(netdev_core_stats_inc);
10757
10758 /**
10759 * dev_get_stats - get network device statistics
10760 * @dev: device to get statistics from
10761 * @storage: place to store stats
10762 *
10763 * Get network statistics from device. Return @storage.
10764 * The device driver may provide its own method by setting
10765 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10766 * otherwise the internal statistics structure is used.
10767 */
dev_get_stats(struct net_device * dev,struct rtnl_link_stats64 * storage)10768 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10769 struct rtnl_link_stats64 *storage)
10770 {
10771 const struct net_device_ops *ops = dev->netdev_ops;
10772 const struct net_device_core_stats __percpu *p;
10773
10774 if (ops->ndo_get_stats64) {
10775 memset(storage, 0, sizeof(*storage));
10776 ops->ndo_get_stats64(dev, storage);
10777 } else if (ops->ndo_get_stats) {
10778 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10779 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) {
10780 dev_get_tstats64(dev, storage);
10781 } else {
10782 netdev_stats_to_stats64(storage, &dev->stats);
10783 }
10784
10785 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10786 p = READ_ONCE(dev->core_stats);
10787 if (p) {
10788 const struct net_device_core_stats *core_stats;
10789 int i;
10790
10791 for_each_possible_cpu(i) {
10792 core_stats = per_cpu_ptr(p, i);
10793 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10794 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10795 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10796 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10797 }
10798 }
10799 return storage;
10800 }
10801 EXPORT_SYMBOL(dev_get_stats);
10802
10803 /**
10804 * dev_fetch_sw_netstats - get per-cpu network device statistics
10805 * @s: place to store stats
10806 * @netstats: per-cpu network stats to read from
10807 *
10808 * Read per-cpu network statistics and populate the related fields in @s.
10809 */
dev_fetch_sw_netstats(struct rtnl_link_stats64 * s,const struct pcpu_sw_netstats __percpu * netstats)10810 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10811 const struct pcpu_sw_netstats __percpu *netstats)
10812 {
10813 int cpu;
10814
10815 for_each_possible_cpu(cpu) {
10816 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10817 const struct pcpu_sw_netstats *stats;
10818 unsigned int start;
10819
10820 stats = per_cpu_ptr(netstats, cpu);
10821 do {
10822 start = u64_stats_fetch_begin(&stats->syncp);
10823 rx_packets = u64_stats_read(&stats->rx_packets);
10824 rx_bytes = u64_stats_read(&stats->rx_bytes);
10825 tx_packets = u64_stats_read(&stats->tx_packets);
10826 tx_bytes = u64_stats_read(&stats->tx_bytes);
10827 } while (u64_stats_fetch_retry(&stats->syncp, start));
10828
10829 s->rx_packets += rx_packets;
10830 s->rx_bytes += rx_bytes;
10831 s->tx_packets += tx_packets;
10832 s->tx_bytes += tx_bytes;
10833 }
10834 }
10835 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10836
10837 /**
10838 * dev_get_tstats64 - ndo_get_stats64 implementation
10839 * @dev: device to get statistics from
10840 * @s: place to store stats
10841 *
10842 * Populate @s from dev->stats and dev->tstats. Can be used as
10843 * ndo_get_stats64() callback.
10844 */
dev_get_tstats64(struct net_device * dev,struct rtnl_link_stats64 * s)10845 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10846 {
10847 netdev_stats_to_stats64(s, &dev->stats);
10848 dev_fetch_sw_netstats(s, dev->tstats);
10849 }
10850 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10851
dev_ingress_queue_create(struct net_device * dev)10852 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10853 {
10854 struct netdev_queue *queue = dev_ingress_queue(dev);
10855
10856 #ifdef CONFIG_NET_CLS_ACT
10857 if (queue)
10858 return queue;
10859 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10860 if (!queue)
10861 return NULL;
10862 netdev_init_one_queue(dev, queue, NULL);
10863 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10864 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10865 rcu_assign_pointer(dev->ingress_queue, queue);
10866 #endif
10867 return queue;
10868 }
10869
10870 static const struct ethtool_ops default_ethtool_ops;
10871
netdev_set_default_ethtool_ops(struct net_device * dev,const struct ethtool_ops * ops)10872 void netdev_set_default_ethtool_ops(struct net_device *dev,
10873 const struct ethtool_ops *ops)
10874 {
10875 if (dev->ethtool_ops == &default_ethtool_ops)
10876 dev->ethtool_ops = ops;
10877 }
10878 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10879
10880 /**
10881 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10882 * @dev: netdev to enable the IRQ coalescing on
10883 *
10884 * Sets a conservative default for SW IRQ coalescing. Users can use
10885 * sysfs attributes to override the default values.
10886 */
netdev_sw_irq_coalesce_default_on(struct net_device * dev)10887 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10888 {
10889 WARN_ON(dev->reg_state == NETREG_REGISTERED);
10890
10891 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
10892 dev->gro_flush_timeout = 20000;
10893 dev->napi_defer_hard_irqs = 1;
10894 }
10895 }
10896 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10897
netdev_freemem(struct net_device * dev)10898 void netdev_freemem(struct net_device *dev)
10899 {
10900 char *addr = (char *)dev - dev->padded;
10901
10902 kvfree(addr);
10903 }
10904
10905 /**
10906 * alloc_netdev_mqs - allocate network device
10907 * @sizeof_priv: size of private data to allocate space for
10908 * @name: device name format string
10909 * @name_assign_type: origin of device name
10910 * @setup: callback to initialize device
10911 * @txqs: the number of TX subqueues to allocate
10912 * @rxqs: the number of RX subqueues to allocate
10913 *
10914 * Allocates a struct net_device with private data area for driver use
10915 * and performs basic initialization. Also allocates subqueue structs
10916 * for each queue on the device.
10917 */
alloc_netdev_mqs(int sizeof_priv,const char * name,unsigned char name_assign_type,void (* setup)(struct net_device *),unsigned int txqs,unsigned int rxqs)10918 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10919 unsigned char name_assign_type,
10920 void (*setup)(struct net_device *),
10921 unsigned int txqs, unsigned int rxqs)
10922 {
10923 struct net_device *dev;
10924 unsigned int alloc_size;
10925 struct net_device *p;
10926
10927 BUG_ON(strlen(name) >= sizeof(dev->name));
10928
10929 if (txqs < 1) {
10930 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10931 return NULL;
10932 }
10933
10934 if (rxqs < 1) {
10935 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10936 return NULL;
10937 }
10938
10939 alloc_size = sizeof(struct net_device);
10940 if (sizeof_priv) {
10941 /* ensure 32-byte alignment of private area */
10942 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10943 alloc_size += sizeof_priv;
10944 }
10945 /* ensure 32-byte alignment of whole construct */
10946 alloc_size += NETDEV_ALIGN - 1;
10947
10948 p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10949 if (!p)
10950 return NULL;
10951
10952 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10953 dev->padded = (char *)dev - (char *)p;
10954
10955 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
10956 #ifdef CONFIG_PCPU_DEV_REFCNT
10957 dev->pcpu_refcnt = alloc_percpu(int);
10958 if (!dev->pcpu_refcnt)
10959 goto free_dev;
10960 __dev_hold(dev);
10961 #else
10962 refcount_set(&dev->dev_refcnt, 1);
10963 #endif
10964
10965 if (dev_addr_init(dev))
10966 goto free_pcpu;
10967
10968 dev_mc_init(dev);
10969 dev_uc_init(dev);
10970
10971 dev_net_set(dev, &init_net);
10972
10973 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10974 dev->xdp_zc_max_segs = 1;
10975 dev->gso_max_segs = GSO_MAX_SEGS;
10976 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10977 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
10978 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
10979 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10980 dev->tso_max_segs = TSO_MAX_SEGS;
10981 dev->upper_level = 1;
10982 dev->lower_level = 1;
10983 #ifdef CONFIG_LOCKDEP
10984 dev->nested_level = 0;
10985 INIT_LIST_HEAD(&dev->unlink_list);
10986 #endif
10987
10988 INIT_LIST_HEAD(&dev->napi_list);
10989 INIT_LIST_HEAD(&dev->unreg_list);
10990 INIT_LIST_HEAD(&dev->close_list);
10991 INIT_LIST_HEAD(&dev->link_watch_list);
10992 INIT_LIST_HEAD(&dev->adj_list.upper);
10993 INIT_LIST_HEAD(&dev->adj_list.lower);
10994 INIT_LIST_HEAD(&dev->ptype_all);
10995 INIT_LIST_HEAD(&dev->ptype_specific);
10996 INIT_LIST_HEAD(&dev->net_notifier_list);
10997 #ifdef CONFIG_NET_SCHED
10998 hash_init(dev->qdisc_hash);
10999 #endif
11000 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
11001 setup(dev);
11002
11003 if (!dev->tx_queue_len) {
11004 dev->priv_flags |= IFF_NO_QUEUE;
11005 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
11006 }
11007
11008 dev->num_tx_queues = txqs;
11009 dev->real_num_tx_queues = txqs;
11010 if (netif_alloc_netdev_queues(dev))
11011 goto free_all;
11012
11013 dev->num_rx_queues = rxqs;
11014 dev->real_num_rx_queues = rxqs;
11015 if (netif_alloc_rx_queues(dev))
11016 goto free_all;
11017
11018 strcpy(dev->name, name);
11019 dev->name_assign_type = name_assign_type;
11020 dev->group = INIT_NETDEV_GROUP;
11021 if (!dev->ethtool_ops)
11022 dev->ethtool_ops = &default_ethtool_ops;
11023
11024 nf_hook_netdev_init(dev);
11025
11026 return dev;
11027
11028 free_all:
11029 free_netdev(dev);
11030 return NULL;
11031
11032 free_pcpu:
11033 #ifdef CONFIG_PCPU_DEV_REFCNT
11034 free_percpu(dev->pcpu_refcnt);
11035 free_dev:
11036 #endif
11037 netdev_freemem(dev);
11038 return NULL;
11039 }
11040 EXPORT_SYMBOL(alloc_netdev_mqs);
11041
11042 /**
11043 * free_netdev - free network device
11044 * @dev: device
11045 *
11046 * This function does the last stage of destroying an allocated device
11047 * interface. The reference to the device object is released. If this
11048 * is the last reference then it will be freed.Must be called in process
11049 * context.
11050 */
free_netdev(struct net_device * dev)11051 void free_netdev(struct net_device *dev)
11052 {
11053 struct napi_struct *p, *n;
11054
11055 might_sleep();
11056
11057 /* When called immediately after register_netdevice() failed the unwind
11058 * handling may still be dismantling the device. Handle that case by
11059 * deferring the free.
11060 */
11061 if (dev->reg_state == NETREG_UNREGISTERING) {
11062 ASSERT_RTNL();
11063 dev->needs_free_netdev = true;
11064 return;
11065 }
11066
11067 netif_free_tx_queues(dev);
11068 netif_free_rx_queues(dev);
11069
11070 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
11071
11072 /* Flush device addresses */
11073 dev_addr_flush(dev);
11074
11075 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
11076 netif_napi_del(p);
11077
11078 ref_tracker_dir_exit(&dev->refcnt_tracker);
11079 #ifdef CONFIG_PCPU_DEV_REFCNT
11080 free_percpu(dev->pcpu_refcnt);
11081 dev->pcpu_refcnt = NULL;
11082 #endif
11083 free_percpu(dev->core_stats);
11084 dev->core_stats = NULL;
11085 free_percpu(dev->xdp_bulkq);
11086 dev->xdp_bulkq = NULL;
11087
11088 /* Compatibility with error handling in drivers */
11089 if (dev->reg_state == NETREG_UNINITIALIZED ||
11090 dev->reg_state == NETREG_DUMMY) {
11091 netdev_freemem(dev);
11092 return;
11093 }
11094
11095 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
11096 WRITE_ONCE(dev->reg_state, NETREG_RELEASED);
11097
11098 /* will free via device release */
11099 put_device(&dev->dev);
11100 }
11101 EXPORT_SYMBOL(free_netdev);
11102
11103 /**
11104 * alloc_netdev_dummy - Allocate and initialize a dummy net device.
11105 * @sizeof_priv: size of private data to allocate space for
11106 *
11107 * Return: the allocated net_device on success, NULL otherwise
11108 */
alloc_netdev_dummy(int sizeof_priv)11109 struct net_device *alloc_netdev_dummy(int sizeof_priv)
11110 {
11111 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN,
11112 init_dummy_netdev_core);
11113 }
11114 EXPORT_SYMBOL_GPL(alloc_netdev_dummy);
11115
11116 /**
11117 * synchronize_net - Synchronize with packet receive processing
11118 *
11119 * Wait for packets currently being received to be done.
11120 * Does not block later packets from starting.
11121 */
synchronize_net(void)11122 void synchronize_net(void)
11123 {
11124 might_sleep();
11125 if (rtnl_is_locked())
11126 synchronize_rcu_expedited();
11127 else
11128 synchronize_rcu();
11129 }
11130 EXPORT_SYMBOL(synchronize_net);
11131
11132 /**
11133 * unregister_netdevice_queue - remove device from the kernel
11134 * @dev: device
11135 * @head: list
11136 *
11137 * This function shuts down a device interface and removes it
11138 * from the kernel tables.
11139 * If head not NULL, device is queued to be unregistered later.
11140 *
11141 * Callers must hold the rtnl semaphore. You may want
11142 * unregister_netdev() instead of this.
11143 */
11144
unregister_netdevice_queue(struct net_device * dev,struct list_head * head)11145 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11146 {
11147 ASSERT_RTNL();
11148
11149 if (head) {
11150 list_move_tail(&dev->unreg_list, head);
11151 } else {
11152 LIST_HEAD(single);
11153
11154 list_add(&dev->unreg_list, &single);
11155 unregister_netdevice_many(&single);
11156 }
11157 }
11158 EXPORT_SYMBOL(unregister_netdevice_queue);
11159
unregister_netdevice_many_notify(struct list_head * head,u32 portid,const struct nlmsghdr * nlh)11160 void unregister_netdevice_many_notify(struct list_head *head,
11161 u32 portid, const struct nlmsghdr *nlh)
11162 {
11163 struct net_device *dev, *tmp;
11164 LIST_HEAD(close_head);
11165 int cnt = 0;
11166
11167 BUG_ON(dev_boot_phase);
11168 ASSERT_RTNL();
11169
11170 if (list_empty(head))
11171 return;
11172
11173 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11174 /* Some devices call without registering
11175 * for initialization unwind. Remove those
11176 * devices and proceed with the remaining.
11177 */
11178 if (dev->reg_state == NETREG_UNINITIALIZED) {
11179 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11180 dev->name, dev);
11181
11182 WARN_ON(1);
11183 list_del(&dev->unreg_list);
11184 continue;
11185 }
11186 dev->dismantle = true;
11187 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11188 }
11189
11190 /* If device is running, close it first. */
11191 list_for_each_entry(dev, head, unreg_list)
11192 list_add_tail(&dev->close_list, &close_head);
11193 dev_close_many(&close_head, true);
11194
11195 list_for_each_entry(dev, head, unreg_list) {
11196 /* And unlink it from device chain. */
11197 unlist_netdevice(dev);
11198 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING);
11199 }
11200 flush_all_backlogs();
11201
11202 synchronize_net();
11203
11204 list_for_each_entry(dev, head, unreg_list) {
11205 struct sk_buff *skb = NULL;
11206
11207 /* Shutdown queueing discipline. */
11208 dev_shutdown(dev);
11209 dev_tcx_uninstall(dev);
11210 dev_xdp_uninstall(dev);
11211 bpf_dev_bound_netdev_unregister(dev);
11212
11213 netdev_offload_xstats_disable_all(dev);
11214
11215 /* Notify protocols, that we are about to destroy
11216 * this device. They should clean all the things.
11217 */
11218 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11219
11220 if (!dev->rtnl_link_ops ||
11221 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11222 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11223 GFP_KERNEL, NULL, 0,
11224 portid, nlh);
11225
11226 /*
11227 * Flush the unicast and multicast chains
11228 */
11229 dev_uc_flush(dev);
11230 dev_mc_flush(dev);
11231
11232 netdev_name_node_alt_flush(dev);
11233 netdev_name_node_free(dev->name_node);
11234
11235 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
11236
11237 if (dev->netdev_ops->ndo_uninit)
11238 dev->netdev_ops->ndo_uninit(dev);
11239
11240 if (skb)
11241 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
11242
11243 /* Notifier chain MUST detach us all upper devices. */
11244 WARN_ON(netdev_has_any_upper_dev(dev));
11245 WARN_ON(netdev_has_any_lower_dev(dev));
11246
11247 /* Remove entries from kobject tree */
11248 netdev_unregister_kobject(dev);
11249 #ifdef CONFIG_XPS
11250 /* Remove XPS queueing entries */
11251 netif_reset_xps_queues_gt(dev, 0);
11252 #endif
11253 }
11254
11255 synchronize_net();
11256
11257 list_for_each_entry(dev, head, unreg_list) {
11258 netdev_put(dev, &dev->dev_registered_tracker);
11259 net_set_todo(dev);
11260 cnt++;
11261 }
11262 atomic_add(cnt, &dev_unreg_count);
11263
11264 list_del(head);
11265 }
11266
11267 /**
11268 * unregister_netdevice_many - unregister many devices
11269 * @head: list of devices
11270 *
11271 * Note: As most callers use a stack allocated list_head,
11272 * we force a list_del() to make sure stack wont be corrupted later.
11273 */
unregister_netdevice_many(struct list_head * head)11274 void unregister_netdevice_many(struct list_head *head)
11275 {
11276 unregister_netdevice_many_notify(head, 0, NULL);
11277 }
11278 EXPORT_SYMBOL(unregister_netdevice_many);
11279
11280 /**
11281 * unregister_netdev - remove device from the kernel
11282 * @dev: device
11283 *
11284 * This function shuts down a device interface and removes it
11285 * from the kernel tables.
11286 *
11287 * This is just a wrapper for unregister_netdevice that takes
11288 * the rtnl semaphore. In general you want to use this and not
11289 * unregister_netdevice.
11290 */
unregister_netdev(struct net_device * dev)11291 void unregister_netdev(struct net_device *dev)
11292 {
11293 rtnl_lock();
11294 unregister_netdevice(dev);
11295 rtnl_unlock();
11296 }
11297 EXPORT_SYMBOL(unregister_netdev);
11298
11299 /**
11300 * __dev_change_net_namespace - move device to different nethost namespace
11301 * @dev: device
11302 * @net: network namespace
11303 * @pat: If not NULL name pattern to try if the current device name
11304 * is already taken in the destination network namespace.
11305 * @new_ifindex: If not zero, specifies device index in the target
11306 * namespace.
11307 *
11308 * This function shuts down a device interface and moves it
11309 * to a new network namespace. On success 0 is returned, on
11310 * a failure a netagive errno code is returned.
11311 *
11312 * Callers must hold the rtnl semaphore.
11313 */
11314
__dev_change_net_namespace(struct net_device * dev,struct net * net,const char * pat,int new_ifindex)11315 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11316 const char *pat, int new_ifindex)
11317 {
11318 struct netdev_name_node *name_node;
11319 struct net *net_old = dev_net(dev);
11320 char new_name[IFNAMSIZ] = {};
11321 int err, new_nsid;
11322
11323 ASSERT_RTNL();
11324
11325 /* Don't allow namespace local devices to be moved. */
11326 err = -EINVAL;
11327 if (dev->features & NETIF_F_NETNS_LOCAL)
11328 goto out;
11329
11330 /* Ensure the device has been registrered */
11331 if (dev->reg_state != NETREG_REGISTERED)
11332 goto out;
11333
11334 /* Get out if there is nothing todo */
11335 err = 0;
11336 if (net_eq(net_old, net))
11337 goto out;
11338
11339 /* Pick the destination device name, and ensure
11340 * we can use it in the destination network namespace.
11341 */
11342 err = -EEXIST;
11343 if (netdev_name_in_use(net, dev->name)) {
11344 /* We get here if we can't use the current device name */
11345 if (!pat)
11346 goto out;
11347 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST);
11348 if (err < 0)
11349 goto out;
11350 }
11351 /* Check that none of the altnames conflicts. */
11352 err = -EEXIST;
11353 netdev_for_each_altname(dev, name_node)
11354 if (netdev_name_in_use(net, name_node->name))
11355 goto out;
11356
11357 /* Check that new_ifindex isn't used yet. */
11358 if (new_ifindex) {
11359 err = dev_index_reserve(net, new_ifindex);
11360 if (err < 0)
11361 goto out;
11362 } else {
11363 /* If there is an ifindex conflict assign a new one */
11364 err = dev_index_reserve(net, dev->ifindex);
11365 if (err == -EBUSY)
11366 err = dev_index_reserve(net, 0);
11367 if (err < 0)
11368 goto out;
11369 new_ifindex = err;
11370 }
11371
11372 /*
11373 * And now a mini version of register_netdevice unregister_netdevice.
11374 */
11375
11376 /* If device is running close it first. */
11377 dev_close(dev);
11378
11379 /* And unlink it from device chain */
11380 unlist_netdevice(dev);
11381
11382 synchronize_net();
11383
11384 /* Shutdown queueing discipline. */
11385 dev_shutdown(dev);
11386
11387 /* Notify protocols, that we are about to destroy
11388 * this device. They should clean all the things.
11389 *
11390 * Note that dev->reg_state stays at NETREG_REGISTERED.
11391 * This is wanted because this way 8021q and macvlan know
11392 * the device is just moving and can keep their slaves up.
11393 */
11394 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11395 rcu_barrier();
11396
11397 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11398
11399 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11400 new_ifindex);
11401
11402 /*
11403 * Flush the unicast and multicast chains
11404 */
11405 dev_uc_flush(dev);
11406 dev_mc_flush(dev);
11407
11408 /* Send a netdev-removed uevent to the old namespace */
11409 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11410 netdev_adjacent_del_links(dev);
11411
11412 /* Move per-net netdevice notifiers that are following the netdevice */
11413 move_netdevice_notifiers_dev_net(dev, net);
11414
11415 /* Actually switch the network namespace */
11416 dev_net_set(dev, net);
11417 dev->ifindex = new_ifindex;
11418
11419 if (new_name[0]) {
11420 /* Rename the netdev to prepared name */
11421 write_seqlock(&netdev_rename_lock);
11422 strscpy(dev->name, new_name, IFNAMSIZ);
11423 write_sequnlock(&netdev_rename_lock);
11424 }
11425
11426 /* Fixup kobjects */
11427 dev_set_uevent_suppress(&dev->dev, 1);
11428 err = device_rename(&dev->dev, dev->name);
11429 dev_set_uevent_suppress(&dev->dev, 0);
11430 WARN_ON(err);
11431
11432 /* Send a netdev-add uevent to the new namespace */
11433 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11434 netdev_adjacent_add_links(dev);
11435
11436 /* Adapt owner in case owning user namespace of target network
11437 * namespace is different from the original one.
11438 */
11439 err = netdev_change_owner(dev, net_old, net);
11440 WARN_ON(err);
11441
11442 /* Add the device back in the hashes */
11443 list_netdevice(dev);
11444
11445 /* Notify protocols, that a new device appeared. */
11446 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11447
11448 /*
11449 * Prevent userspace races by waiting until the network
11450 * device is fully setup before sending notifications.
11451 */
11452 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11453
11454 synchronize_net();
11455 err = 0;
11456 out:
11457 return err;
11458 }
11459 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11460
dev_cpu_dead(unsigned int oldcpu)11461 static int dev_cpu_dead(unsigned int oldcpu)
11462 {
11463 struct sk_buff **list_skb;
11464 struct sk_buff *skb;
11465 unsigned int cpu;
11466 struct softnet_data *sd, *oldsd, *remsd = NULL;
11467
11468 local_irq_disable();
11469 cpu = smp_processor_id();
11470 sd = &per_cpu(softnet_data, cpu);
11471 oldsd = &per_cpu(softnet_data, oldcpu);
11472
11473 /* Find end of our completion_queue. */
11474 list_skb = &sd->completion_queue;
11475 while (*list_skb)
11476 list_skb = &(*list_skb)->next;
11477 /* Append completion queue from offline CPU. */
11478 *list_skb = oldsd->completion_queue;
11479 oldsd->completion_queue = NULL;
11480
11481 /* Append output queue from offline CPU. */
11482 if (oldsd->output_queue) {
11483 *sd->output_queue_tailp = oldsd->output_queue;
11484 sd->output_queue_tailp = oldsd->output_queue_tailp;
11485 oldsd->output_queue = NULL;
11486 oldsd->output_queue_tailp = &oldsd->output_queue;
11487 }
11488 /* Append NAPI poll list from offline CPU, with one exception :
11489 * process_backlog() must be called by cpu owning percpu backlog.
11490 * We properly handle process_queue & input_pkt_queue later.
11491 */
11492 while (!list_empty(&oldsd->poll_list)) {
11493 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11494 struct napi_struct,
11495 poll_list);
11496
11497 list_del_init(&napi->poll_list);
11498 if (napi->poll == process_backlog)
11499 napi->state &= NAPIF_STATE_THREADED;
11500 else
11501 ____napi_schedule(sd, napi);
11502 }
11503
11504 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11505 local_irq_enable();
11506
11507 if (!use_backlog_threads()) {
11508 #ifdef CONFIG_RPS
11509 remsd = oldsd->rps_ipi_list;
11510 oldsd->rps_ipi_list = NULL;
11511 #endif
11512 /* send out pending IPI's on offline CPU */
11513 net_rps_send_ipi(remsd);
11514 }
11515
11516 /* Process offline CPU's input_pkt_queue */
11517 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11518 netif_rx(skb);
11519 rps_input_queue_head_incr(oldsd);
11520 }
11521 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11522 netif_rx(skb);
11523 rps_input_queue_head_incr(oldsd);
11524 }
11525
11526 return 0;
11527 }
11528
11529 /**
11530 * netdev_increment_features - increment feature set by one
11531 * @all: current feature set
11532 * @one: new feature set
11533 * @mask: mask feature set
11534 *
11535 * Computes a new feature set after adding a device with feature set
11536 * @one to the master device with current feature set @all. Will not
11537 * enable anything that is off in @mask. Returns the new feature set.
11538 */
netdev_increment_features(netdev_features_t all,netdev_features_t one,netdev_features_t mask)11539 netdev_features_t netdev_increment_features(netdev_features_t all,
11540 netdev_features_t one, netdev_features_t mask)
11541 {
11542 if (mask & NETIF_F_HW_CSUM)
11543 mask |= NETIF_F_CSUM_MASK;
11544 mask |= NETIF_F_VLAN_CHALLENGED;
11545
11546 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11547 all &= one | ~NETIF_F_ALL_FOR_ALL;
11548
11549 /* If one device supports hw checksumming, set for all. */
11550 if (all & NETIF_F_HW_CSUM)
11551 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11552
11553 return all;
11554 }
11555 EXPORT_SYMBOL(netdev_increment_features);
11556
netdev_create_hash(void)11557 static struct hlist_head * __net_init netdev_create_hash(void)
11558 {
11559 int i;
11560 struct hlist_head *hash;
11561
11562 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11563 if (hash != NULL)
11564 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11565 INIT_HLIST_HEAD(&hash[i]);
11566
11567 return hash;
11568 }
11569
11570 /* Initialize per network namespace state */
netdev_init(struct net * net)11571 static int __net_init netdev_init(struct net *net)
11572 {
11573 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11574 8 * sizeof_field(struct napi_struct, gro_bitmask));
11575
11576 INIT_LIST_HEAD(&net->dev_base_head);
11577
11578 net->dev_name_head = netdev_create_hash();
11579 if (net->dev_name_head == NULL)
11580 goto err_name;
11581
11582 net->dev_index_head = netdev_create_hash();
11583 if (net->dev_index_head == NULL)
11584 goto err_idx;
11585
11586 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1);
11587
11588 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11589
11590 return 0;
11591
11592 err_idx:
11593 kfree(net->dev_name_head);
11594 err_name:
11595 return -ENOMEM;
11596 }
11597
11598 /**
11599 * netdev_drivername - network driver for the device
11600 * @dev: network device
11601 *
11602 * Determine network driver for device.
11603 */
netdev_drivername(const struct net_device * dev)11604 const char *netdev_drivername(const struct net_device *dev)
11605 {
11606 const struct device_driver *driver;
11607 const struct device *parent;
11608 const char *empty = "";
11609
11610 parent = dev->dev.parent;
11611 if (!parent)
11612 return empty;
11613
11614 driver = parent->driver;
11615 if (driver && driver->name)
11616 return driver->name;
11617 return empty;
11618 }
11619
__netdev_printk(const char * level,const struct net_device * dev,struct va_format * vaf)11620 static void __netdev_printk(const char *level, const struct net_device *dev,
11621 struct va_format *vaf)
11622 {
11623 if (dev && dev->dev.parent) {
11624 dev_printk_emit(level[1] - '0',
11625 dev->dev.parent,
11626 "%s %s %s%s: %pV",
11627 dev_driver_string(dev->dev.parent),
11628 dev_name(dev->dev.parent),
11629 netdev_name(dev), netdev_reg_state(dev),
11630 vaf);
11631 } else if (dev) {
11632 printk("%s%s%s: %pV",
11633 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11634 } else {
11635 printk("%s(NULL net_device): %pV", level, vaf);
11636 }
11637 }
11638
netdev_printk(const char * level,const struct net_device * dev,const char * format,...)11639 void netdev_printk(const char *level, const struct net_device *dev,
11640 const char *format, ...)
11641 {
11642 struct va_format vaf;
11643 va_list args;
11644
11645 va_start(args, format);
11646
11647 vaf.fmt = format;
11648 vaf.va = &args;
11649
11650 __netdev_printk(level, dev, &vaf);
11651
11652 va_end(args);
11653 }
11654 EXPORT_SYMBOL(netdev_printk);
11655
11656 #define define_netdev_printk_level(func, level) \
11657 void func(const struct net_device *dev, const char *fmt, ...) \
11658 { \
11659 struct va_format vaf; \
11660 va_list args; \
11661 \
11662 va_start(args, fmt); \
11663 \
11664 vaf.fmt = fmt; \
11665 vaf.va = &args; \
11666 \
11667 __netdev_printk(level, dev, &vaf); \
11668 \
11669 va_end(args); \
11670 } \
11671 EXPORT_SYMBOL(func);
11672
11673 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11674 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11675 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11676 define_netdev_printk_level(netdev_err, KERN_ERR);
11677 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11678 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11679 define_netdev_printk_level(netdev_info, KERN_INFO);
11680
netdev_exit(struct net * net)11681 static void __net_exit netdev_exit(struct net *net)
11682 {
11683 kfree(net->dev_name_head);
11684 kfree(net->dev_index_head);
11685 xa_destroy(&net->dev_by_index);
11686 if (net != &init_net)
11687 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11688 }
11689
11690 static struct pernet_operations __net_initdata netdev_net_ops = {
11691 .init = netdev_init,
11692 .exit = netdev_exit,
11693 };
11694
default_device_exit_net(struct net * net)11695 static void __net_exit default_device_exit_net(struct net *net)
11696 {
11697 struct netdev_name_node *name_node, *tmp;
11698 struct net_device *dev, *aux;
11699 /*
11700 * Push all migratable network devices back to the
11701 * initial network namespace
11702 */
11703 ASSERT_RTNL();
11704 for_each_netdev_safe(net, dev, aux) {
11705 int err;
11706 char fb_name[IFNAMSIZ];
11707
11708 /* Ignore unmoveable devices (i.e. loopback) */
11709 if (dev->features & NETIF_F_NETNS_LOCAL)
11710 continue;
11711
11712 /* Leave virtual devices for the generic cleanup */
11713 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11714 continue;
11715
11716 /* Push remaining network devices to init_net */
11717 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11718 if (netdev_name_in_use(&init_net, fb_name))
11719 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11720
11721 netdev_for_each_altname_safe(dev, name_node, tmp)
11722 if (netdev_name_in_use(&init_net, name_node->name))
11723 __netdev_name_node_alt_destroy(name_node);
11724
11725 err = dev_change_net_namespace(dev, &init_net, fb_name);
11726 if (err) {
11727 pr_emerg("%s: failed to move %s to init_net: %d\n",
11728 __func__, dev->name, err);
11729 BUG();
11730 }
11731 }
11732 }
11733
default_device_exit_batch(struct list_head * net_list)11734 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11735 {
11736 /* At exit all network devices most be removed from a network
11737 * namespace. Do this in the reverse order of registration.
11738 * Do this across as many network namespaces as possible to
11739 * improve batching efficiency.
11740 */
11741 struct net_device *dev;
11742 struct net *net;
11743 LIST_HEAD(dev_kill_list);
11744
11745 rtnl_lock();
11746 list_for_each_entry(net, net_list, exit_list) {
11747 default_device_exit_net(net);
11748 cond_resched();
11749 }
11750
11751 list_for_each_entry(net, net_list, exit_list) {
11752 for_each_netdev_reverse(net, dev) {
11753 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11754 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11755 else
11756 unregister_netdevice_queue(dev, &dev_kill_list);
11757 }
11758 }
11759 unregister_netdevice_many(&dev_kill_list);
11760 rtnl_unlock();
11761 }
11762
11763 static struct pernet_operations __net_initdata default_device_ops = {
11764 .exit_batch = default_device_exit_batch,
11765 };
11766
net_dev_struct_check(void)11767 static void __init net_dev_struct_check(void)
11768 {
11769 /* TX read-mostly hotpath */
11770 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags);
11771 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops);
11772 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops);
11773 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx);
11774 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues);
11775 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size);
11776 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size);
11777 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs);
11778 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features);
11779 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc);
11780 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu);
11781 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom);
11782 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq);
11783 #ifdef CONFIG_XPS
11784 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps);
11785 #endif
11786 #ifdef CONFIG_NETFILTER_EGRESS
11787 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress);
11788 #endif
11789 #ifdef CONFIG_NET_XGRESS
11790 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress);
11791 #endif
11792 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160);
11793
11794 /* TXRX read-mostly hotpath */
11795 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats);
11796 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state);
11797 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags);
11798 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len);
11799 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features);
11800 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr);
11801 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46);
11802
11803 /* RX read-mostly hotpath */
11804 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific);
11805 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex);
11806 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues);
11807 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx);
11808 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout);
11809 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs);
11810 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size);
11811 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size);
11812 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler);
11813 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data);
11814 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net);
11815 #ifdef CONFIG_NETPOLL
11816 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo);
11817 #endif
11818 #ifdef CONFIG_NET_XGRESS
11819 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress);
11820 #endif
11821 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104);
11822 }
11823
11824 /*
11825 * Initialize the DEV module. At boot time this walks the device list and
11826 * unhooks any devices that fail to initialise (normally hardware not
11827 * present) and leaves us with a valid list of present and active devices.
11828 *
11829 */
11830
11831 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
11832 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE)
11833
net_page_pool_create(int cpuid)11834 static int net_page_pool_create(int cpuid)
11835 {
11836 #if IS_ENABLED(CONFIG_PAGE_POOL)
11837 struct page_pool_params page_pool_params = {
11838 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
11839 .flags = PP_FLAG_SYSTEM_POOL,
11840 .nid = cpu_to_mem(cpuid),
11841 };
11842 struct page_pool *pp_ptr;
11843
11844 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
11845 if (IS_ERR(pp_ptr))
11846 return -ENOMEM;
11847
11848 per_cpu(system_page_pool, cpuid) = pp_ptr;
11849 #endif
11850 return 0;
11851 }
11852
backlog_napi_should_run(unsigned int cpu)11853 static int backlog_napi_should_run(unsigned int cpu)
11854 {
11855 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11856 struct napi_struct *napi = &sd->backlog;
11857
11858 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
11859 }
11860
run_backlog_napi(unsigned int cpu)11861 static void run_backlog_napi(unsigned int cpu)
11862 {
11863 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11864
11865 napi_threaded_poll_loop(&sd->backlog);
11866 }
11867
backlog_napi_setup(unsigned int cpu)11868 static void backlog_napi_setup(unsigned int cpu)
11869 {
11870 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11871 struct napi_struct *napi = &sd->backlog;
11872
11873 napi->thread = this_cpu_read(backlog_napi);
11874 set_bit(NAPI_STATE_THREADED, &napi->state);
11875 }
11876
11877 static struct smp_hotplug_thread backlog_threads = {
11878 .store = &backlog_napi,
11879 .thread_should_run = backlog_napi_should_run,
11880 .thread_fn = run_backlog_napi,
11881 .thread_comm = "backlog_napi/%u",
11882 .setup = backlog_napi_setup,
11883 };
11884
11885 /*
11886 * This is called single threaded during boot, so no need
11887 * to take the rtnl semaphore.
11888 */
net_dev_init(void)11889 static int __init net_dev_init(void)
11890 {
11891 int i, rc = -ENOMEM;
11892
11893 BUG_ON(!dev_boot_phase);
11894
11895 net_dev_struct_check();
11896
11897 if (dev_proc_init())
11898 goto out;
11899
11900 if (netdev_kobject_init())
11901 goto out;
11902
11903 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11904 INIT_LIST_HEAD(&ptype_base[i]);
11905
11906 if (register_pernet_subsys(&netdev_net_ops))
11907 goto out;
11908
11909 /*
11910 * Initialise the packet receive queues.
11911 */
11912
11913 for_each_possible_cpu(i) {
11914 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11915 struct softnet_data *sd = &per_cpu(softnet_data, i);
11916
11917 INIT_WORK(flush, flush_backlog);
11918
11919 skb_queue_head_init(&sd->input_pkt_queue);
11920 skb_queue_head_init(&sd->process_queue);
11921 #ifdef CONFIG_XFRM_OFFLOAD
11922 skb_queue_head_init(&sd->xfrm_backlog);
11923 #endif
11924 INIT_LIST_HEAD(&sd->poll_list);
11925 sd->output_queue_tailp = &sd->output_queue;
11926 #ifdef CONFIG_RPS
11927 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11928 sd->cpu = i;
11929 #endif
11930 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11931 spin_lock_init(&sd->defer_lock);
11932
11933 init_gro_hash(&sd->backlog);
11934 sd->backlog.poll = process_backlog;
11935 sd->backlog.weight = weight_p;
11936 INIT_LIST_HEAD(&sd->backlog.poll_list);
11937
11938 if (net_page_pool_create(i))
11939 goto out;
11940 }
11941 if (use_backlog_threads())
11942 smpboot_register_percpu_thread(&backlog_threads);
11943
11944 dev_boot_phase = 0;
11945
11946 /* The loopback device is special if any other network devices
11947 * is present in a network namespace the loopback device must
11948 * be present. Since we now dynamically allocate and free the
11949 * loopback device ensure this invariant is maintained by
11950 * keeping the loopback device as the first device on the
11951 * list of network devices. Ensuring the loopback devices
11952 * is the first device that appears and the last network device
11953 * that disappears.
11954 */
11955 if (register_pernet_device(&loopback_net_ops))
11956 goto out;
11957
11958 if (register_pernet_device(&default_device_ops))
11959 goto out;
11960
11961 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11962 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11963
11964 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11965 NULL, dev_cpu_dead);
11966 WARN_ON(rc < 0);
11967 rc = 0;
11968
11969 /* avoid static key IPIs to isolated CPUs */
11970 if (housekeeping_enabled(HK_TYPE_MISC))
11971 net_enable_timestamp();
11972 out:
11973 if (rc < 0) {
11974 for_each_possible_cpu(i) {
11975 struct page_pool *pp_ptr;
11976
11977 pp_ptr = per_cpu(system_page_pool, i);
11978 if (!pp_ptr)
11979 continue;
11980
11981 page_pool_destroy(pp_ptr);
11982 per_cpu(system_page_pool, i) = NULL;
11983 }
11984 }
11985
11986 return rc;
11987 }
11988
11989 subsys_initcall(net_dev_init);
11990