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 #include <linux/phy_link_topology.h>
162
163 #include "dev.h"
164 #include "devmem.h"
165 #include "net-sysfs.h"
166
167 static DEFINE_SPINLOCK(ptype_lock);
168 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
169
170 static int netif_rx_internal(struct sk_buff *skb);
171 static int call_netdevice_notifiers_extack(unsigned long val,
172 struct net_device *dev,
173 struct netlink_ext_ack *extack);
174
175 static DEFINE_MUTEX(ifalias_mutex);
176
177 /* protects napi_hash addition/deletion and napi_gen_id */
178 static DEFINE_SPINLOCK(napi_hash_lock);
179
180 static unsigned int napi_gen_id = NR_CPUS;
181 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
182
183 static DECLARE_RWSEM(devnet_rename_sem);
184
dev_base_seq_inc(struct net * net)185 static inline void dev_base_seq_inc(struct net *net)
186 {
187 unsigned int val = net->dev_base_seq + 1;
188
189 WRITE_ONCE(net->dev_base_seq, val ?: 1);
190 }
191
dev_name_hash(struct net * net,const char * name)192 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
193 {
194 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
195
196 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
197 }
198
dev_index_hash(struct net * net,int ifindex)199 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
200 {
201 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
202 }
203
204 #ifndef CONFIG_PREEMPT_RT
205
206 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key);
207
setup_backlog_napi_threads(char * arg)208 static int __init setup_backlog_napi_threads(char *arg)
209 {
210 static_branch_enable(&use_backlog_threads_key);
211 return 0;
212 }
213 early_param("thread_backlog_napi", setup_backlog_napi_threads);
214
use_backlog_threads(void)215 static bool use_backlog_threads(void)
216 {
217 return static_branch_unlikely(&use_backlog_threads_key);
218 }
219
220 #else
221
use_backlog_threads(void)222 static bool use_backlog_threads(void)
223 {
224 return true;
225 }
226
227 #endif
228
backlog_lock_irq_save(struct softnet_data * sd,unsigned long * flags)229 static inline void backlog_lock_irq_save(struct softnet_data *sd,
230 unsigned long *flags)
231 {
232 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
233 spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
234 else
235 local_irq_save(*flags);
236 }
237
backlog_lock_irq_disable(struct softnet_data * sd)238 static inline void backlog_lock_irq_disable(struct softnet_data *sd)
239 {
240 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
241 spin_lock_irq(&sd->input_pkt_queue.lock);
242 else
243 local_irq_disable();
244 }
245
backlog_unlock_irq_restore(struct softnet_data * sd,unsigned long * flags)246 static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
247 unsigned long *flags)
248 {
249 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
250 spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
251 else
252 local_irq_restore(*flags);
253 }
254
backlog_unlock_irq_enable(struct softnet_data * sd)255 static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
256 {
257 if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
258 spin_unlock_irq(&sd->input_pkt_queue.lock);
259 else
260 local_irq_enable();
261 }
262
netdev_name_node_alloc(struct net_device * dev,const char * name)263 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
264 const char *name)
265 {
266 struct netdev_name_node *name_node;
267
268 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
269 if (!name_node)
270 return NULL;
271 INIT_HLIST_NODE(&name_node->hlist);
272 name_node->dev = dev;
273 name_node->name = name;
274 return name_node;
275 }
276
277 static struct netdev_name_node *
netdev_name_node_head_alloc(struct net_device * dev)278 netdev_name_node_head_alloc(struct net_device *dev)
279 {
280 struct netdev_name_node *name_node;
281
282 name_node = netdev_name_node_alloc(dev, dev->name);
283 if (!name_node)
284 return NULL;
285 INIT_LIST_HEAD(&name_node->list);
286 return name_node;
287 }
288
netdev_name_node_free(struct netdev_name_node * name_node)289 static void netdev_name_node_free(struct netdev_name_node *name_node)
290 {
291 kfree(name_node);
292 }
293
netdev_name_node_add(struct net * net,struct netdev_name_node * name_node)294 static void netdev_name_node_add(struct net *net,
295 struct netdev_name_node *name_node)
296 {
297 hlist_add_head_rcu(&name_node->hlist,
298 dev_name_hash(net, name_node->name));
299 }
300
netdev_name_node_del(struct netdev_name_node * name_node)301 static void netdev_name_node_del(struct netdev_name_node *name_node)
302 {
303 hlist_del_rcu(&name_node->hlist);
304 }
305
netdev_name_node_lookup(struct net * net,const char * name)306 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
307 const char *name)
308 {
309 struct hlist_head *head = dev_name_hash(net, name);
310 struct netdev_name_node *name_node;
311
312 hlist_for_each_entry(name_node, head, hlist)
313 if (!strcmp(name_node->name, name))
314 return name_node;
315 return NULL;
316 }
317
netdev_name_node_lookup_rcu(struct net * net,const char * name)318 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
319 const char *name)
320 {
321 struct hlist_head *head = dev_name_hash(net, name);
322 struct netdev_name_node *name_node;
323
324 hlist_for_each_entry_rcu(name_node, head, hlist)
325 if (!strcmp(name_node->name, name))
326 return name_node;
327 return NULL;
328 }
329
netdev_name_in_use(struct net * net,const char * name)330 bool netdev_name_in_use(struct net *net, const char *name)
331 {
332 return netdev_name_node_lookup(net, name);
333 }
334 EXPORT_SYMBOL(netdev_name_in_use);
335
netdev_name_node_alt_create(struct net_device * dev,const char * name)336 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
337 {
338 struct netdev_name_node *name_node;
339 struct net *net = dev_net(dev);
340
341 name_node = netdev_name_node_lookup(net, name);
342 if (name_node)
343 return -EEXIST;
344 name_node = netdev_name_node_alloc(dev, name);
345 if (!name_node)
346 return -ENOMEM;
347 netdev_name_node_add(net, name_node);
348 /* The node that holds dev->name acts as a head of per-device list. */
349 list_add_tail_rcu(&name_node->list, &dev->name_node->list);
350
351 return 0;
352 }
353
netdev_name_node_alt_free(struct rcu_head * head)354 static void netdev_name_node_alt_free(struct rcu_head *head)
355 {
356 struct netdev_name_node *name_node =
357 container_of(head, struct netdev_name_node, rcu);
358
359 kfree(name_node->name);
360 netdev_name_node_free(name_node);
361 }
362
__netdev_name_node_alt_destroy(struct netdev_name_node * name_node)363 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
364 {
365 netdev_name_node_del(name_node);
366 list_del(&name_node->list);
367 call_rcu(&name_node->rcu, netdev_name_node_alt_free);
368 }
369
netdev_name_node_alt_destroy(struct net_device * dev,const char * name)370 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
371 {
372 struct netdev_name_node *name_node;
373 struct net *net = dev_net(dev);
374
375 name_node = netdev_name_node_lookup(net, name);
376 if (!name_node)
377 return -ENOENT;
378 /* lookup might have found our primary name or a name belonging
379 * to another device.
380 */
381 if (name_node == dev->name_node || name_node->dev != dev)
382 return -EINVAL;
383
384 __netdev_name_node_alt_destroy(name_node);
385 return 0;
386 }
387
netdev_name_node_alt_flush(struct net_device * dev)388 static void netdev_name_node_alt_flush(struct net_device *dev)
389 {
390 struct netdev_name_node *name_node, *tmp;
391
392 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) {
393 list_del(&name_node->list);
394 netdev_name_node_alt_free(&name_node->rcu);
395 }
396 }
397
398 /* Device list insertion */
list_netdevice(struct net_device * dev)399 static void list_netdevice(struct net_device *dev)
400 {
401 struct netdev_name_node *name_node;
402 struct net *net = dev_net(dev);
403
404 ASSERT_RTNL();
405
406 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
407 netdev_name_node_add(net, dev->name_node);
408 hlist_add_head_rcu(&dev->index_hlist,
409 dev_index_hash(net, dev->ifindex));
410
411 netdev_for_each_altname(dev, name_node)
412 netdev_name_node_add(net, name_node);
413
414 /* We reserved the ifindex, this can't fail */
415 WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL));
416
417 dev_base_seq_inc(net);
418 }
419
420 /* Device list removal
421 * caller must respect a RCU grace period before freeing/reusing dev
422 */
unlist_netdevice(struct net_device * dev)423 static void unlist_netdevice(struct net_device *dev)
424 {
425 struct netdev_name_node *name_node;
426 struct net *net = dev_net(dev);
427
428 ASSERT_RTNL();
429
430 xa_erase(&net->dev_by_index, dev->ifindex);
431
432 netdev_for_each_altname(dev, name_node)
433 netdev_name_node_del(name_node);
434
435 /* Unlink dev from the device chain */
436 list_del_rcu(&dev->dev_list);
437 netdev_name_node_del(dev->name_node);
438 hlist_del_rcu(&dev->index_hlist);
439
440 dev_base_seq_inc(dev_net(dev));
441 }
442
443 /*
444 * Our notifier list
445 */
446
447 static RAW_NOTIFIER_HEAD(netdev_chain);
448
449 /*
450 * Device drivers call our routines to queue packets here. We empty the
451 * queue in the local softnet handler.
452 */
453
454 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = {
455 .process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock),
456 };
457 EXPORT_PER_CPU_SYMBOL(softnet_data);
458
459 /* Page_pool has a lockless array/stack to alloc/recycle pages.
460 * PP consumers must pay attention to run APIs in the appropriate context
461 * (e.g. NAPI context).
462 */
463 static DEFINE_PER_CPU(struct page_pool *, system_page_pool);
464
465 #ifdef CONFIG_LOCKDEP
466 /*
467 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
468 * according to dev->type
469 */
470 static const unsigned short netdev_lock_type[] = {
471 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
472 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
473 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
474 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
475 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
476 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
477 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
478 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
479 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
480 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
481 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
482 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
483 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
484 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
485 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
486
487 static const char *const netdev_lock_name[] = {
488 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
489 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
490 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
491 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
492 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
493 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
494 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
495 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
496 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
497 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
498 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
499 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
500 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
501 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
502 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
503
504 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
505 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
506
netdev_lock_pos(unsigned short dev_type)507 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
508 {
509 int i;
510
511 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
512 if (netdev_lock_type[i] == dev_type)
513 return i;
514 /* the last key is used by default */
515 return ARRAY_SIZE(netdev_lock_type) - 1;
516 }
517
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)518 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
519 unsigned short dev_type)
520 {
521 int i;
522
523 i = netdev_lock_pos(dev_type);
524 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
525 netdev_lock_name[i]);
526 }
527
netdev_set_addr_lockdep_class(struct net_device * dev)528 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
529 {
530 int i;
531
532 i = netdev_lock_pos(dev->type);
533 lockdep_set_class_and_name(&dev->addr_list_lock,
534 &netdev_addr_lock_key[i],
535 netdev_lock_name[i]);
536 }
537 #else
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)538 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
539 unsigned short dev_type)
540 {
541 }
542
netdev_set_addr_lockdep_class(struct net_device * dev)543 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
544 {
545 }
546 #endif
547
548 /*******************************************************************************
549 *
550 * Protocol management and registration routines
551 *
552 *******************************************************************************/
553
554
555 /*
556 * Add a protocol ID to the list. Now that the input handler is
557 * smarter we can dispense with all the messy stuff that used to be
558 * here.
559 *
560 * BEWARE!!! Protocol handlers, mangling input packets,
561 * MUST BE last in hash buckets and checking protocol handlers
562 * MUST start from promiscuous ptype_all chain in net_bh.
563 * It is true now, do not change it.
564 * Explanation follows: if protocol handler, mangling packet, will
565 * be the first on list, it is not able to sense, that packet
566 * is cloned and should be copied-on-write, so that it will
567 * change it and subsequent readers will get broken packet.
568 * --ANK (980803)
569 */
570
ptype_head(const struct packet_type * pt)571 static inline struct list_head *ptype_head(const struct packet_type *pt)
572 {
573 if (pt->type == htons(ETH_P_ALL))
574 return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all;
575 else
576 return pt->dev ? &pt->dev->ptype_specific :
577 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
578 }
579
580 /**
581 * dev_add_pack - add packet handler
582 * @pt: packet type declaration
583 *
584 * Add a protocol handler to the networking stack. The passed &packet_type
585 * is linked into kernel lists and may not be freed until it has been
586 * removed from the kernel lists.
587 *
588 * This call does not sleep therefore it can not
589 * guarantee all CPU's that are in middle of receiving packets
590 * will see the new packet type (until the next received packet).
591 */
592
dev_add_pack(struct packet_type * pt)593 void dev_add_pack(struct packet_type *pt)
594 {
595 struct list_head *head = ptype_head(pt);
596
597 spin_lock(&ptype_lock);
598 list_add_rcu(&pt->list, head);
599 spin_unlock(&ptype_lock);
600 }
601 EXPORT_SYMBOL(dev_add_pack);
602
603 /**
604 * __dev_remove_pack - remove packet handler
605 * @pt: packet type declaration
606 *
607 * Remove a protocol handler that was previously added to the kernel
608 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
609 * from the kernel lists and can be freed or reused once this function
610 * returns.
611 *
612 * The packet type might still be in use by receivers
613 * and must not be freed until after all the CPU's have gone
614 * through a quiescent state.
615 */
__dev_remove_pack(struct packet_type * pt)616 void __dev_remove_pack(struct packet_type *pt)
617 {
618 struct list_head *head = ptype_head(pt);
619 struct packet_type *pt1;
620
621 spin_lock(&ptype_lock);
622
623 list_for_each_entry(pt1, head, list) {
624 if (pt == pt1) {
625 list_del_rcu(&pt->list);
626 goto out;
627 }
628 }
629
630 pr_warn("dev_remove_pack: %p not found\n", pt);
631 out:
632 spin_unlock(&ptype_lock);
633 }
634 EXPORT_SYMBOL(__dev_remove_pack);
635
636 /**
637 * dev_remove_pack - remove packet handler
638 * @pt: packet type declaration
639 *
640 * Remove a protocol handler that was previously added to the kernel
641 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
642 * from the kernel lists and can be freed or reused once this function
643 * returns.
644 *
645 * This call sleeps to guarantee that no CPU is looking at the packet
646 * type after return.
647 */
dev_remove_pack(struct packet_type * pt)648 void dev_remove_pack(struct packet_type *pt)
649 {
650 __dev_remove_pack(pt);
651
652 synchronize_net();
653 }
654 EXPORT_SYMBOL(dev_remove_pack);
655
656
657 /*******************************************************************************
658 *
659 * Device Interface Subroutines
660 *
661 *******************************************************************************/
662
663 /**
664 * dev_get_iflink - get 'iflink' value of a interface
665 * @dev: targeted interface
666 *
667 * Indicates the ifindex the interface is linked to.
668 * Physical interfaces have the same 'ifindex' and 'iflink' values.
669 */
670
dev_get_iflink(const struct net_device * dev)671 int dev_get_iflink(const struct net_device *dev)
672 {
673 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
674 return dev->netdev_ops->ndo_get_iflink(dev);
675
676 return READ_ONCE(dev->ifindex);
677 }
678 EXPORT_SYMBOL(dev_get_iflink);
679
680 /**
681 * dev_fill_metadata_dst - Retrieve tunnel egress information.
682 * @dev: targeted interface
683 * @skb: The packet.
684 *
685 * For better visibility of tunnel traffic OVS needs to retrieve
686 * egress tunnel information for a packet. Following API allows
687 * user to get this info.
688 */
dev_fill_metadata_dst(struct net_device * dev,struct sk_buff * skb)689 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
690 {
691 struct ip_tunnel_info *info;
692
693 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
694 return -EINVAL;
695
696 info = skb_tunnel_info_unclone(skb);
697 if (!info)
698 return -ENOMEM;
699 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
700 return -EINVAL;
701
702 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
703 }
704 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
705
dev_fwd_path(struct net_device_path_stack * stack)706 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
707 {
708 int k = stack->num_paths++;
709
710 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
711 return NULL;
712
713 return &stack->path[k];
714 }
715
dev_fill_forward_path(const struct net_device * dev,const u8 * daddr,struct net_device_path_stack * stack)716 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
717 struct net_device_path_stack *stack)
718 {
719 const struct net_device *last_dev;
720 struct net_device_path_ctx ctx = {
721 .dev = dev,
722 };
723 struct net_device_path *path;
724 int ret = 0;
725
726 memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
727 stack->num_paths = 0;
728 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
729 last_dev = ctx.dev;
730 path = dev_fwd_path(stack);
731 if (!path)
732 return -1;
733
734 memset(path, 0, sizeof(struct net_device_path));
735 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
736 if (ret < 0)
737 return -1;
738
739 if (WARN_ON_ONCE(last_dev == ctx.dev))
740 return -1;
741 }
742
743 if (!ctx.dev)
744 return ret;
745
746 path = dev_fwd_path(stack);
747 if (!path)
748 return -1;
749 path->type = DEV_PATH_ETHERNET;
750 path->dev = ctx.dev;
751
752 return ret;
753 }
754 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
755
756 /**
757 * __dev_get_by_name - find a device by its name
758 * @net: the applicable net namespace
759 * @name: name to find
760 *
761 * Find an interface by name. Must be called under RTNL semaphore.
762 * If the name is found a pointer to the device is returned.
763 * If the name is not found then %NULL is returned. The
764 * reference counters are not incremented so the caller must be
765 * careful with locks.
766 */
767
__dev_get_by_name(struct net * net,const char * name)768 struct net_device *__dev_get_by_name(struct net *net, const char *name)
769 {
770 struct netdev_name_node *node_name;
771
772 node_name = netdev_name_node_lookup(net, name);
773 return node_name ? node_name->dev : NULL;
774 }
775 EXPORT_SYMBOL(__dev_get_by_name);
776
777 /**
778 * dev_get_by_name_rcu - find a device by its name
779 * @net: the applicable net namespace
780 * @name: name to find
781 *
782 * Find an interface by name.
783 * If the name is found a pointer to the device is returned.
784 * If the name is not found then %NULL is returned.
785 * The reference counters are not incremented so the caller must be
786 * careful with locks. The caller must hold RCU lock.
787 */
788
dev_get_by_name_rcu(struct net * net,const char * name)789 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
790 {
791 struct netdev_name_node *node_name;
792
793 node_name = netdev_name_node_lookup_rcu(net, name);
794 return node_name ? node_name->dev : NULL;
795 }
796 EXPORT_SYMBOL(dev_get_by_name_rcu);
797
798 /* Deprecated for new users, call netdev_get_by_name() instead */
dev_get_by_name(struct net * net,const char * name)799 struct net_device *dev_get_by_name(struct net *net, const char *name)
800 {
801 struct net_device *dev;
802
803 rcu_read_lock();
804 dev = dev_get_by_name_rcu(net, name);
805 dev_hold(dev);
806 rcu_read_unlock();
807 return dev;
808 }
809 EXPORT_SYMBOL(dev_get_by_name);
810
811 /**
812 * netdev_get_by_name() - find a device by its name
813 * @net: the applicable net namespace
814 * @name: name to find
815 * @tracker: tracking object for the acquired reference
816 * @gfp: allocation flags for the tracker
817 *
818 * Find an interface by name. This can be called from any
819 * context and does its own locking. The returned handle has
820 * the usage count incremented and the caller must use netdev_put() to
821 * release it when it is no longer needed. %NULL is returned if no
822 * matching device is found.
823 */
netdev_get_by_name(struct net * net,const char * name,netdevice_tracker * tracker,gfp_t gfp)824 struct net_device *netdev_get_by_name(struct net *net, const char *name,
825 netdevice_tracker *tracker, gfp_t gfp)
826 {
827 struct net_device *dev;
828
829 dev = dev_get_by_name(net, name);
830 if (dev)
831 netdev_tracker_alloc(dev, tracker, gfp);
832 return dev;
833 }
834 EXPORT_SYMBOL(netdev_get_by_name);
835
836 /**
837 * __dev_get_by_index - find a device by its ifindex
838 * @net: the applicable net namespace
839 * @ifindex: index of device
840 *
841 * Search for an interface by index. Returns %NULL if the device
842 * is not found or a pointer to the device. The device has not
843 * had its reference counter increased so the caller must be careful
844 * about locking. The caller must hold the RTNL semaphore.
845 */
846
__dev_get_by_index(struct net * net,int ifindex)847 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
848 {
849 struct net_device *dev;
850 struct hlist_head *head = dev_index_hash(net, ifindex);
851
852 hlist_for_each_entry(dev, head, index_hlist)
853 if (dev->ifindex == ifindex)
854 return dev;
855
856 return NULL;
857 }
858 EXPORT_SYMBOL(__dev_get_by_index);
859
860 /**
861 * dev_get_by_index_rcu - find a device by its ifindex
862 * @net: the applicable net namespace
863 * @ifindex: index of device
864 *
865 * Search for an interface by index. Returns %NULL if the device
866 * is not found or a pointer to the device. The device has not
867 * had its reference counter increased so the caller must be careful
868 * about locking. The caller must hold RCU lock.
869 */
870
dev_get_by_index_rcu(struct net * net,int ifindex)871 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
872 {
873 struct net_device *dev;
874 struct hlist_head *head = dev_index_hash(net, ifindex);
875
876 hlist_for_each_entry_rcu(dev, head, index_hlist)
877 if (dev->ifindex == ifindex)
878 return dev;
879
880 return NULL;
881 }
882 EXPORT_SYMBOL(dev_get_by_index_rcu);
883
884 /* Deprecated for new users, call netdev_get_by_index() instead */
dev_get_by_index(struct net * net,int ifindex)885 struct net_device *dev_get_by_index(struct net *net, int ifindex)
886 {
887 struct net_device *dev;
888
889 rcu_read_lock();
890 dev = dev_get_by_index_rcu(net, ifindex);
891 dev_hold(dev);
892 rcu_read_unlock();
893 return dev;
894 }
895 EXPORT_SYMBOL(dev_get_by_index);
896
897 /**
898 * netdev_get_by_index() - find a device by its ifindex
899 * @net: the applicable net namespace
900 * @ifindex: index of device
901 * @tracker: tracking object for the acquired reference
902 * @gfp: allocation flags for the tracker
903 *
904 * Search for an interface by index. Returns NULL if the device
905 * is not found or a pointer to the device. The device returned has
906 * had a reference added and the pointer is safe until the user calls
907 * netdev_put() to indicate they have finished with it.
908 */
netdev_get_by_index(struct net * net,int ifindex,netdevice_tracker * tracker,gfp_t gfp)909 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
910 netdevice_tracker *tracker, gfp_t gfp)
911 {
912 struct net_device *dev;
913
914 dev = dev_get_by_index(net, ifindex);
915 if (dev)
916 netdev_tracker_alloc(dev, tracker, gfp);
917 return dev;
918 }
919 EXPORT_SYMBOL(netdev_get_by_index);
920
921 /**
922 * dev_get_by_napi_id - find a device by napi_id
923 * @napi_id: ID of the NAPI struct
924 *
925 * Search for an interface by NAPI ID. Returns %NULL if the device
926 * is not found or a pointer to the device. The device has not had
927 * its reference counter increased so the caller must be careful
928 * about locking. The caller must hold RCU lock.
929 */
930
dev_get_by_napi_id(unsigned int napi_id)931 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
932 {
933 struct napi_struct *napi;
934
935 WARN_ON_ONCE(!rcu_read_lock_held());
936
937 if (napi_id < MIN_NAPI_ID)
938 return NULL;
939
940 napi = napi_by_id(napi_id);
941
942 return napi ? napi->dev : NULL;
943 }
944 EXPORT_SYMBOL(dev_get_by_napi_id);
945
946 static DEFINE_SEQLOCK(netdev_rename_lock);
947
netdev_copy_name(struct net_device * dev,char * name)948 void netdev_copy_name(struct net_device *dev, char *name)
949 {
950 unsigned int seq;
951
952 do {
953 seq = read_seqbegin(&netdev_rename_lock);
954 strscpy(name, dev->name, IFNAMSIZ);
955 } while (read_seqretry(&netdev_rename_lock, seq));
956 }
957
958 /**
959 * netdev_get_name - get a netdevice name, knowing its ifindex.
960 * @net: network namespace
961 * @name: a pointer to the buffer where the name will be stored.
962 * @ifindex: the ifindex of the interface to get the name from.
963 */
netdev_get_name(struct net * net,char * name,int ifindex)964 int netdev_get_name(struct net *net, char *name, int ifindex)
965 {
966 struct net_device *dev;
967 int ret;
968
969 rcu_read_lock();
970
971 dev = dev_get_by_index_rcu(net, ifindex);
972 if (!dev) {
973 ret = -ENODEV;
974 goto out;
975 }
976
977 netdev_copy_name(dev, name);
978
979 ret = 0;
980 out:
981 rcu_read_unlock();
982 return ret;
983 }
984
985 /**
986 * dev_getbyhwaddr_rcu - find a device by its hardware address
987 * @net: the applicable net namespace
988 * @type: media type of device
989 * @ha: hardware address
990 *
991 * Search for an interface by MAC address. Returns NULL if the device
992 * is not found or a pointer to the device.
993 * The caller must hold RCU or RTNL.
994 * The returned device has not had its ref count increased
995 * and the caller must therefore be careful about locking
996 *
997 */
998
dev_getbyhwaddr_rcu(struct net * net,unsigned short type,const char * ha)999 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1000 const char *ha)
1001 {
1002 struct net_device *dev;
1003
1004 for_each_netdev_rcu(net, dev)
1005 if (dev->type == type &&
1006 !memcmp(dev->dev_addr, ha, dev->addr_len))
1007 return dev;
1008
1009 return NULL;
1010 }
1011 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1012
dev_getfirstbyhwtype(struct net * net,unsigned short type)1013 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1014 {
1015 struct net_device *dev, *ret = NULL;
1016
1017 rcu_read_lock();
1018 for_each_netdev_rcu(net, dev)
1019 if (dev->type == type) {
1020 dev_hold(dev);
1021 ret = dev;
1022 break;
1023 }
1024 rcu_read_unlock();
1025 return ret;
1026 }
1027 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1028
1029 /**
1030 * __dev_get_by_flags - find any device with given flags
1031 * @net: the applicable net namespace
1032 * @if_flags: IFF_* values
1033 * @mask: bitmask of bits in if_flags to check
1034 *
1035 * Search for any interface with the given flags. Returns NULL if a device
1036 * is not found or a pointer to the device. Must be called inside
1037 * rtnl_lock(), and result refcount is unchanged.
1038 */
1039
__dev_get_by_flags(struct net * net,unsigned short if_flags,unsigned short mask)1040 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1041 unsigned short mask)
1042 {
1043 struct net_device *dev, *ret;
1044
1045 ASSERT_RTNL();
1046
1047 ret = NULL;
1048 for_each_netdev(net, dev) {
1049 if (((dev->flags ^ if_flags) & mask) == 0) {
1050 ret = dev;
1051 break;
1052 }
1053 }
1054 return ret;
1055 }
1056 EXPORT_SYMBOL(__dev_get_by_flags);
1057
1058 /**
1059 * dev_valid_name - check if name is okay for network device
1060 * @name: name string
1061 *
1062 * Network device names need to be valid file names to
1063 * allow sysfs to work. We also disallow any kind of
1064 * whitespace.
1065 */
dev_valid_name(const char * name)1066 bool dev_valid_name(const char *name)
1067 {
1068 if (*name == '\0')
1069 return false;
1070 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1071 return false;
1072 if (!strcmp(name, ".") || !strcmp(name, ".."))
1073 return false;
1074
1075 while (*name) {
1076 if (*name == '/' || *name == ':' || isspace(*name))
1077 return false;
1078 name++;
1079 }
1080 return true;
1081 }
1082 EXPORT_SYMBOL(dev_valid_name);
1083
1084 /**
1085 * __dev_alloc_name - allocate a name for a device
1086 * @net: network namespace to allocate the device name in
1087 * @name: name format string
1088 * @res: result name string
1089 *
1090 * Passed a format string - eg "lt%d" it will try and find a suitable
1091 * id. It scans list of devices to build up a free map, then chooses
1092 * the first empty slot. The caller must hold the dev_base or rtnl lock
1093 * while allocating the name and adding the device in order to avoid
1094 * duplicates.
1095 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1096 * Returns the number of the unit assigned or a negative errno code.
1097 */
1098
__dev_alloc_name(struct net * net,const char * name,char * res)1099 static int __dev_alloc_name(struct net *net, const char *name, char *res)
1100 {
1101 int i = 0;
1102 const char *p;
1103 const int max_netdevices = 8*PAGE_SIZE;
1104 unsigned long *inuse;
1105 struct net_device *d;
1106 char buf[IFNAMSIZ];
1107
1108 /* Verify the string as this thing may have come from the user.
1109 * There must be one "%d" and no other "%" characters.
1110 */
1111 p = strchr(name, '%');
1112 if (!p || p[1] != 'd' || strchr(p + 2, '%'))
1113 return -EINVAL;
1114
1115 /* Use one page as a bit array of possible slots */
1116 inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC);
1117 if (!inuse)
1118 return -ENOMEM;
1119
1120 for_each_netdev(net, d) {
1121 struct netdev_name_node *name_node;
1122
1123 netdev_for_each_altname(d, name_node) {
1124 if (!sscanf(name_node->name, name, &i))
1125 continue;
1126 if (i < 0 || i >= max_netdevices)
1127 continue;
1128
1129 /* avoid cases where sscanf is not exact inverse of printf */
1130 snprintf(buf, IFNAMSIZ, name, i);
1131 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1132 __set_bit(i, inuse);
1133 }
1134 if (!sscanf(d->name, name, &i))
1135 continue;
1136 if (i < 0 || i >= max_netdevices)
1137 continue;
1138
1139 /* avoid cases where sscanf is not exact inverse of printf */
1140 snprintf(buf, IFNAMSIZ, name, i);
1141 if (!strncmp(buf, d->name, IFNAMSIZ))
1142 __set_bit(i, inuse);
1143 }
1144
1145 i = find_first_zero_bit(inuse, max_netdevices);
1146 bitmap_free(inuse);
1147 if (i == max_netdevices)
1148 return -ENFILE;
1149
1150 /* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */
1151 strscpy(buf, name, IFNAMSIZ);
1152 snprintf(res, IFNAMSIZ, buf, i);
1153 return i;
1154 }
1155
1156 /* 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)1157 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1158 const char *want_name, char *out_name,
1159 int dup_errno)
1160 {
1161 if (!dev_valid_name(want_name))
1162 return -EINVAL;
1163
1164 if (strchr(want_name, '%'))
1165 return __dev_alloc_name(net, want_name, out_name);
1166
1167 if (netdev_name_in_use(net, want_name))
1168 return -dup_errno;
1169 if (out_name != want_name)
1170 strscpy(out_name, want_name, IFNAMSIZ);
1171 return 0;
1172 }
1173
1174 /**
1175 * dev_alloc_name - allocate a name for a device
1176 * @dev: device
1177 * @name: name format string
1178 *
1179 * Passed a format string - eg "lt%d" it will try and find a suitable
1180 * id. It scans list of devices to build up a free map, then chooses
1181 * the first empty slot. The caller must hold the dev_base or rtnl lock
1182 * while allocating the name and adding the device in order to avoid
1183 * duplicates.
1184 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1185 * Returns the number of the unit assigned or a negative errno code.
1186 */
1187
dev_alloc_name(struct net_device * dev,const char * name)1188 int dev_alloc_name(struct net_device *dev, const char *name)
1189 {
1190 return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE);
1191 }
1192 EXPORT_SYMBOL(dev_alloc_name);
1193
dev_get_valid_name(struct net * net,struct net_device * dev,const char * name)1194 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1195 const char *name)
1196 {
1197 int ret;
1198
1199 ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST);
1200 return ret < 0 ? ret : 0;
1201 }
1202
1203 /**
1204 * dev_change_name - change name of a device
1205 * @dev: device
1206 * @newname: name (or format string) must be at least IFNAMSIZ
1207 *
1208 * Change name of a device, can pass format strings "eth%d".
1209 * for wildcarding.
1210 */
dev_change_name(struct net_device * dev,const char * newname)1211 int dev_change_name(struct net_device *dev, const char *newname)
1212 {
1213 unsigned char old_assign_type;
1214 char oldname[IFNAMSIZ];
1215 int err = 0;
1216 int ret;
1217 struct net *net;
1218
1219 ASSERT_RTNL();
1220 BUG_ON(!dev_net(dev));
1221
1222 net = dev_net(dev);
1223
1224 down_write(&devnet_rename_sem);
1225
1226 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1227 up_write(&devnet_rename_sem);
1228 return 0;
1229 }
1230
1231 memcpy(oldname, dev->name, IFNAMSIZ);
1232
1233 write_seqlock_bh(&netdev_rename_lock);
1234 err = dev_get_valid_name(net, dev, newname);
1235 write_sequnlock_bh(&netdev_rename_lock);
1236
1237 if (err < 0) {
1238 up_write(&devnet_rename_sem);
1239 return err;
1240 }
1241
1242 if (oldname[0] && !strchr(oldname, '%'))
1243 netdev_info(dev, "renamed from %s%s\n", oldname,
1244 dev->flags & IFF_UP ? " (while UP)" : "");
1245
1246 old_assign_type = dev->name_assign_type;
1247 WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED);
1248
1249 rollback:
1250 ret = device_rename(&dev->dev, dev->name);
1251 if (ret) {
1252 memcpy(dev->name, oldname, IFNAMSIZ);
1253 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1254 up_write(&devnet_rename_sem);
1255 return ret;
1256 }
1257
1258 up_write(&devnet_rename_sem);
1259
1260 netdev_adjacent_rename_links(dev, oldname);
1261
1262 netdev_name_node_del(dev->name_node);
1263
1264 synchronize_net();
1265
1266 netdev_name_node_add(net, dev->name_node);
1267
1268 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1269 ret = notifier_to_errno(ret);
1270
1271 if (ret) {
1272 /* err >= 0 after dev_alloc_name() or stores the first errno */
1273 if (err >= 0) {
1274 err = ret;
1275 down_write(&devnet_rename_sem);
1276 write_seqlock_bh(&netdev_rename_lock);
1277 memcpy(dev->name, oldname, IFNAMSIZ);
1278 write_sequnlock_bh(&netdev_rename_lock);
1279 memcpy(oldname, newname, IFNAMSIZ);
1280 WRITE_ONCE(dev->name_assign_type, old_assign_type);
1281 old_assign_type = NET_NAME_RENAMED;
1282 goto rollback;
1283 } else {
1284 netdev_err(dev, "name change rollback failed: %d\n",
1285 ret);
1286 }
1287 }
1288
1289 return err;
1290 }
1291
1292 /**
1293 * dev_set_alias - change ifalias of a device
1294 * @dev: device
1295 * @alias: name up to IFALIASZ
1296 * @len: limit of bytes to copy from info
1297 *
1298 * Set ifalias for a device,
1299 */
dev_set_alias(struct net_device * dev,const char * alias,size_t len)1300 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1301 {
1302 struct dev_ifalias *new_alias = NULL;
1303
1304 if (len >= IFALIASZ)
1305 return -EINVAL;
1306
1307 if (len) {
1308 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1309 if (!new_alias)
1310 return -ENOMEM;
1311
1312 memcpy(new_alias->ifalias, alias, len);
1313 new_alias->ifalias[len] = 0;
1314 }
1315
1316 mutex_lock(&ifalias_mutex);
1317 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1318 mutex_is_locked(&ifalias_mutex));
1319 mutex_unlock(&ifalias_mutex);
1320
1321 if (new_alias)
1322 kfree_rcu(new_alias, rcuhead);
1323
1324 return len;
1325 }
1326 EXPORT_SYMBOL(dev_set_alias);
1327
1328 /**
1329 * dev_get_alias - get ifalias of a device
1330 * @dev: device
1331 * @name: buffer to store name of ifalias
1332 * @len: size of buffer
1333 *
1334 * get ifalias for a device. Caller must make sure dev cannot go
1335 * away, e.g. rcu read lock or own a reference count to device.
1336 */
dev_get_alias(const struct net_device * dev,char * name,size_t len)1337 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1338 {
1339 const struct dev_ifalias *alias;
1340 int ret = 0;
1341
1342 rcu_read_lock();
1343 alias = rcu_dereference(dev->ifalias);
1344 if (alias)
1345 ret = snprintf(name, len, "%s", alias->ifalias);
1346 rcu_read_unlock();
1347
1348 return ret;
1349 }
1350
1351 /**
1352 * netdev_features_change - device changes features
1353 * @dev: device to cause notification
1354 *
1355 * Called to indicate a device has changed features.
1356 */
netdev_features_change(struct net_device * dev)1357 void netdev_features_change(struct net_device *dev)
1358 {
1359 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1360 }
1361 EXPORT_SYMBOL(netdev_features_change);
1362
1363 /**
1364 * netdev_state_change - device changes state
1365 * @dev: device to cause notification
1366 *
1367 * Called to indicate a device has changed state. This function calls
1368 * the notifier chains for netdev_chain and sends a NEWLINK message
1369 * to the routing socket.
1370 */
netdev_state_change(struct net_device * dev)1371 void netdev_state_change(struct net_device *dev)
1372 {
1373 if (dev->flags & IFF_UP) {
1374 struct netdev_notifier_change_info change_info = {
1375 .info.dev = dev,
1376 };
1377
1378 call_netdevice_notifiers_info(NETDEV_CHANGE,
1379 &change_info.info);
1380 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1381 }
1382 }
1383 EXPORT_SYMBOL(netdev_state_change);
1384
1385 /**
1386 * __netdev_notify_peers - notify network peers about existence of @dev,
1387 * to be called when rtnl lock is already held.
1388 * @dev: network device
1389 *
1390 * Generate traffic such that interested network peers are aware of
1391 * @dev, such as by generating a gratuitous ARP. This may be used when
1392 * a device wants to inform the rest of the network about some sort of
1393 * reconfiguration such as a failover event or virtual machine
1394 * migration.
1395 */
__netdev_notify_peers(struct net_device * dev)1396 void __netdev_notify_peers(struct net_device *dev)
1397 {
1398 ASSERT_RTNL();
1399 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1400 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1401 }
1402 EXPORT_SYMBOL(__netdev_notify_peers);
1403
1404 /**
1405 * netdev_notify_peers - notify network peers about existence of @dev
1406 * @dev: network device
1407 *
1408 * Generate traffic such that interested network peers are aware of
1409 * @dev, such as by generating a gratuitous ARP. This may be used when
1410 * a device wants to inform the rest of the network about some sort of
1411 * reconfiguration such as a failover event or virtual machine
1412 * migration.
1413 */
netdev_notify_peers(struct net_device * dev)1414 void netdev_notify_peers(struct net_device *dev)
1415 {
1416 rtnl_lock();
1417 __netdev_notify_peers(dev);
1418 rtnl_unlock();
1419 }
1420 EXPORT_SYMBOL(netdev_notify_peers);
1421
1422 static int napi_threaded_poll(void *data);
1423
napi_kthread_create(struct napi_struct * n)1424 static int napi_kthread_create(struct napi_struct *n)
1425 {
1426 int err = 0;
1427
1428 /* Create and wake up the kthread once to put it in
1429 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1430 * warning and work with loadavg.
1431 */
1432 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1433 n->dev->name, n->napi_id);
1434 if (IS_ERR(n->thread)) {
1435 err = PTR_ERR(n->thread);
1436 pr_err("kthread_run failed with err %d\n", err);
1437 n->thread = NULL;
1438 }
1439
1440 return err;
1441 }
1442
__dev_open(struct net_device * dev,struct netlink_ext_ack * extack)1443 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1444 {
1445 const struct net_device_ops *ops = dev->netdev_ops;
1446 int ret;
1447
1448 ASSERT_RTNL();
1449 dev_addr_check(dev);
1450
1451 if (!netif_device_present(dev)) {
1452 /* may be detached because parent is runtime-suspended */
1453 if (dev->dev.parent)
1454 pm_runtime_resume(dev->dev.parent);
1455 if (!netif_device_present(dev))
1456 return -ENODEV;
1457 }
1458
1459 /* Block netpoll from trying to do any rx path servicing.
1460 * If we don't do this there is a chance ndo_poll_controller
1461 * or ndo_poll may be running while we open the device
1462 */
1463 netpoll_poll_disable(dev);
1464
1465 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1466 ret = notifier_to_errno(ret);
1467 if (ret)
1468 return ret;
1469
1470 set_bit(__LINK_STATE_START, &dev->state);
1471
1472 if (ops->ndo_validate_addr)
1473 ret = ops->ndo_validate_addr(dev);
1474
1475 if (!ret && ops->ndo_open)
1476 ret = ops->ndo_open(dev);
1477
1478 netpoll_poll_enable(dev);
1479
1480 if (ret)
1481 clear_bit(__LINK_STATE_START, &dev->state);
1482 else {
1483 dev->flags |= IFF_UP;
1484 dev_set_rx_mode(dev);
1485 dev_activate(dev);
1486 add_device_randomness(dev->dev_addr, dev->addr_len);
1487 }
1488
1489 return ret;
1490 }
1491
1492 /**
1493 * dev_open - prepare an interface for use.
1494 * @dev: device to open
1495 * @extack: netlink extended ack
1496 *
1497 * Takes a device from down to up state. The device's private open
1498 * function is invoked and then the multicast lists are loaded. Finally
1499 * the device is moved into the up state and a %NETDEV_UP message is
1500 * sent to the netdev notifier chain.
1501 *
1502 * Calling this function on an active interface is a nop. On a failure
1503 * a negative errno code is returned.
1504 */
dev_open(struct net_device * dev,struct netlink_ext_ack * extack)1505 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1506 {
1507 int ret;
1508
1509 if (dev->flags & IFF_UP)
1510 return 0;
1511
1512 ret = __dev_open(dev, extack);
1513 if (ret < 0)
1514 return ret;
1515
1516 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1517 call_netdevice_notifiers(NETDEV_UP, dev);
1518
1519 return ret;
1520 }
1521 EXPORT_SYMBOL(dev_open);
1522
__dev_close_many(struct list_head * head)1523 static void __dev_close_many(struct list_head *head)
1524 {
1525 struct net_device *dev;
1526
1527 ASSERT_RTNL();
1528 might_sleep();
1529
1530 list_for_each_entry(dev, head, close_list) {
1531 /* Temporarily disable netpoll until the interface is down */
1532 netpoll_poll_disable(dev);
1533
1534 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1535
1536 clear_bit(__LINK_STATE_START, &dev->state);
1537
1538 /* Synchronize to scheduled poll. We cannot touch poll list, it
1539 * can be even on different cpu. So just clear netif_running().
1540 *
1541 * dev->stop() will invoke napi_disable() on all of it's
1542 * napi_struct instances on this device.
1543 */
1544 smp_mb__after_atomic(); /* Commit netif_running(). */
1545 }
1546
1547 dev_deactivate_many(head);
1548
1549 list_for_each_entry(dev, head, close_list) {
1550 const struct net_device_ops *ops = dev->netdev_ops;
1551
1552 /*
1553 * Call the device specific close. This cannot fail.
1554 * Only if device is UP
1555 *
1556 * We allow it to be called even after a DETACH hot-plug
1557 * event.
1558 */
1559 if (ops->ndo_stop)
1560 ops->ndo_stop(dev);
1561
1562 dev->flags &= ~IFF_UP;
1563 netpoll_poll_enable(dev);
1564 }
1565 }
1566
__dev_close(struct net_device * dev)1567 static void __dev_close(struct net_device *dev)
1568 {
1569 LIST_HEAD(single);
1570
1571 list_add(&dev->close_list, &single);
1572 __dev_close_many(&single);
1573 list_del(&single);
1574 }
1575
dev_close_many(struct list_head * head,bool unlink)1576 void dev_close_many(struct list_head *head, bool unlink)
1577 {
1578 struct net_device *dev, *tmp;
1579
1580 /* Remove the devices that don't need to be closed */
1581 list_for_each_entry_safe(dev, tmp, head, close_list)
1582 if (!(dev->flags & IFF_UP))
1583 list_del_init(&dev->close_list);
1584
1585 __dev_close_many(head);
1586
1587 list_for_each_entry_safe(dev, tmp, head, close_list) {
1588 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1589 call_netdevice_notifiers(NETDEV_DOWN, dev);
1590 if (unlink)
1591 list_del_init(&dev->close_list);
1592 }
1593 }
1594 EXPORT_SYMBOL(dev_close_many);
1595
1596 /**
1597 * dev_close - shutdown an interface.
1598 * @dev: device to shutdown
1599 *
1600 * This function moves an active device into down state. A
1601 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1602 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1603 * chain.
1604 */
dev_close(struct net_device * dev)1605 void dev_close(struct net_device *dev)
1606 {
1607 if (dev->flags & IFF_UP) {
1608 LIST_HEAD(single);
1609
1610 list_add(&dev->close_list, &single);
1611 dev_close_many(&single, true);
1612 list_del(&single);
1613 }
1614 }
1615 EXPORT_SYMBOL(dev_close);
1616
1617
1618 /**
1619 * dev_disable_lro - disable Large Receive Offload on a device
1620 * @dev: device
1621 *
1622 * Disable Large Receive Offload (LRO) on a net device. Must be
1623 * called under RTNL. This is needed if received packets may be
1624 * forwarded to another interface.
1625 */
dev_disable_lro(struct net_device * dev)1626 void dev_disable_lro(struct net_device *dev)
1627 {
1628 struct net_device *lower_dev;
1629 struct list_head *iter;
1630
1631 dev->wanted_features &= ~NETIF_F_LRO;
1632 netdev_update_features(dev);
1633
1634 if (unlikely(dev->features & NETIF_F_LRO))
1635 netdev_WARN(dev, "failed to disable LRO!\n");
1636
1637 netdev_for_each_lower_dev(dev, lower_dev, iter)
1638 dev_disable_lro(lower_dev);
1639 }
1640 EXPORT_SYMBOL(dev_disable_lro);
1641
1642 /**
1643 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1644 * @dev: device
1645 *
1646 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1647 * called under RTNL. This is needed if Generic XDP is installed on
1648 * the device.
1649 */
dev_disable_gro_hw(struct net_device * dev)1650 static void dev_disable_gro_hw(struct net_device *dev)
1651 {
1652 dev->wanted_features &= ~NETIF_F_GRO_HW;
1653 netdev_update_features(dev);
1654
1655 if (unlikely(dev->features & NETIF_F_GRO_HW))
1656 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1657 }
1658
netdev_cmd_to_name(enum netdev_cmd cmd)1659 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1660 {
1661 #define N(val) \
1662 case NETDEV_##val: \
1663 return "NETDEV_" __stringify(val);
1664 switch (cmd) {
1665 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1666 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1667 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1668 N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1669 N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1670 N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1671 N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1672 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1673 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1674 N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1675 N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1676 N(XDP_FEAT_CHANGE)
1677 }
1678 #undef N
1679 return "UNKNOWN_NETDEV_EVENT";
1680 }
1681 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1682
call_netdevice_notifier(struct notifier_block * nb,unsigned long val,struct net_device * dev)1683 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1684 struct net_device *dev)
1685 {
1686 struct netdev_notifier_info info = {
1687 .dev = dev,
1688 };
1689
1690 return nb->notifier_call(nb, val, &info);
1691 }
1692
call_netdevice_register_notifiers(struct notifier_block * nb,struct net_device * dev)1693 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1694 struct net_device *dev)
1695 {
1696 int err;
1697
1698 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1699 err = notifier_to_errno(err);
1700 if (err)
1701 return err;
1702
1703 if (!(dev->flags & IFF_UP))
1704 return 0;
1705
1706 call_netdevice_notifier(nb, NETDEV_UP, dev);
1707 return 0;
1708 }
1709
call_netdevice_unregister_notifiers(struct notifier_block * nb,struct net_device * dev)1710 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1711 struct net_device *dev)
1712 {
1713 if (dev->flags & IFF_UP) {
1714 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1715 dev);
1716 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1717 }
1718 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1719 }
1720
call_netdevice_register_net_notifiers(struct notifier_block * nb,struct net * net)1721 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1722 struct net *net)
1723 {
1724 struct net_device *dev;
1725 int err;
1726
1727 for_each_netdev(net, dev) {
1728 err = call_netdevice_register_notifiers(nb, dev);
1729 if (err)
1730 goto rollback;
1731 }
1732 return 0;
1733
1734 rollback:
1735 for_each_netdev_continue_reverse(net, dev)
1736 call_netdevice_unregister_notifiers(nb, dev);
1737 return err;
1738 }
1739
call_netdevice_unregister_net_notifiers(struct notifier_block * nb,struct net * net)1740 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1741 struct net *net)
1742 {
1743 struct net_device *dev;
1744
1745 for_each_netdev(net, dev)
1746 call_netdevice_unregister_notifiers(nb, dev);
1747 }
1748
1749 static int dev_boot_phase = 1;
1750
1751 /**
1752 * register_netdevice_notifier - register a network notifier block
1753 * @nb: notifier
1754 *
1755 * Register a notifier to be called when network device events occur.
1756 * The notifier passed is linked into the kernel structures and must
1757 * not be reused until it has been unregistered. A negative errno code
1758 * is returned on a failure.
1759 *
1760 * When registered all registration and up events are replayed
1761 * to the new notifier to allow device to have a race free
1762 * view of the network device list.
1763 */
1764
register_netdevice_notifier(struct notifier_block * nb)1765 int register_netdevice_notifier(struct notifier_block *nb)
1766 {
1767 struct net *net;
1768 int err;
1769
1770 /* Close race with setup_net() and cleanup_net() */
1771 down_write(&pernet_ops_rwsem);
1772 rtnl_lock();
1773 err = raw_notifier_chain_register(&netdev_chain, nb);
1774 if (err)
1775 goto unlock;
1776 if (dev_boot_phase)
1777 goto unlock;
1778 for_each_net(net) {
1779 err = call_netdevice_register_net_notifiers(nb, net);
1780 if (err)
1781 goto rollback;
1782 }
1783
1784 unlock:
1785 rtnl_unlock();
1786 up_write(&pernet_ops_rwsem);
1787 return err;
1788
1789 rollback:
1790 for_each_net_continue_reverse(net)
1791 call_netdevice_unregister_net_notifiers(nb, net);
1792
1793 raw_notifier_chain_unregister(&netdev_chain, nb);
1794 goto unlock;
1795 }
1796 EXPORT_SYMBOL(register_netdevice_notifier);
1797
1798 /**
1799 * unregister_netdevice_notifier - unregister a network notifier block
1800 * @nb: notifier
1801 *
1802 * Unregister a notifier previously registered by
1803 * register_netdevice_notifier(). The notifier is unlinked into the
1804 * kernel structures and may then be reused. A negative errno code
1805 * is returned on a failure.
1806 *
1807 * After unregistering unregister and down device events are synthesized
1808 * for all devices on the device list to the removed notifier to remove
1809 * the need for special case cleanup code.
1810 */
1811
unregister_netdevice_notifier(struct notifier_block * nb)1812 int unregister_netdevice_notifier(struct notifier_block *nb)
1813 {
1814 struct net *net;
1815 int err;
1816
1817 /* Close race with setup_net() and cleanup_net() */
1818 down_write(&pernet_ops_rwsem);
1819 rtnl_lock();
1820 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1821 if (err)
1822 goto unlock;
1823
1824 for_each_net(net)
1825 call_netdevice_unregister_net_notifiers(nb, net);
1826
1827 unlock:
1828 rtnl_unlock();
1829 up_write(&pernet_ops_rwsem);
1830 return err;
1831 }
1832 EXPORT_SYMBOL(unregister_netdevice_notifier);
1833
__register_netdevice_notifier_net(struct net * net,struct notifier_block * nb,bool ignore_call_fail)1834 static int __register_netdevice_notifier_net(struct net *net,
1835 struct notifier_block *nb,
1836 bool ignore_call_fail)
1837 {
1838 int err;
1839
1840 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1841 if (err)
1842 return err;
1843 if (dev_boot_phase)
1844 return 0;
1845
1846 err = call_netdevice_register_net_notifiers(nb, net);
1847 if (err && !ignore_call_fail)
1848 goto chain_unregister;
1849
1850 return 0;
1851
1852 chain_unregister:
1853 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1854 return err;
1855 }
1856
__unregister_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1857 static int __unregister_netdevice_notifier_net(struct net *net,
1858 struct notifier_block *nb)
1859 {
1860 int err;
1861
1862 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1863 if (err)
1864 return err;
1865
1866 call_netdevice_unregister_net_notifiers(nb, net);
1867 return 0;
1868 }
1869
1870 /**
1871 * register_netdevice_notifier_net - register a per-netns network notifier block
1872 * @net: network namespace
1873 * @nb: notifier
1874 *
1875 * Register a notifier to be called when network device events occur.
1876 * The notifier passed is linked into the kernel structures and must
1877 * not be reused until it has been unregistered. A negative errno code
1878 * is returned on a failure.
1879 *
1880 * When registered all registration and up events are replayed
1881 * to the new notifier to allow device to have a race free
1882 * view of the network device list.
1883 */
1884
register_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1885 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1886 {
1887 int err;
1888
1889 rtnl_lock();
1890 err = __register_netdevice_notifier_net(net, nb, false);
1891 rtnl_unlock();
1892 return err;
1893 }
1894 EXPORT_SYMBOL(register_netdevice_notifier_net);
1895
1896 /**
1897 * unregister_netdevice_notifier_net - unregister a per-netns
1898 * network notifier block
1899 * @net: network namespace
1900 * @nb: notifier
1901 *
1902 * Unregister a notifier previously registered by
1903 * register_netdevice_notifier_net(). The notifier is unlinked from the
1904 * kernel structures and may then be reused. A negative errno code
1905 * is returned on a failure.
1906 *
1907 * After unregistering unregister and down device events are synthesized
1908 * for all devices on the device list to the removed notifier to remove
1909 * the need for special case cleanup code.
1910 */
1911
unregister_netdevice_notifier_net(struct net * net,struct notifier_block * nb)1912 int unregister_netdevice_notifier_net(struct net *net,
1913 struct notifier_block *nb)
1914 {
1915 int err;
1916
1917 rtnl_lock();
1918 err = __unregister_netdevice_notifier_net(net, nb);
1919 rtnl_unlock();
1920 return err;
1921 }
1922 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1923
__move_netdevice_notifier_net(struct net * src_net,struct net * dst_net,struct notifier_block * nb)1924 static void __move_netdevice_notifier_net(struct net *src_net,
1925 struct net *dst_net,
1926 struct notifier_block *nb)
1927 {
1928 __unregister_netdevice_notifier_net(src_net, nb);
1929 __register_netdevice_notifier_net(dst_net, nb, true);
1930 }
1931
register_netdevice_notifier_dev_net(struct net_device * dev,struct notifier_block * nb,struct netdev_net_notifier * nn)1932 int register_netdevice_notifier_dev_net(struct net_device *dev,
1933 struct notifier_block *nb,
1934 struct netdev_net_notifier *nn)
1935 {
1936 int err;
1937
1938 rtnl_lock();
1939 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1940 if (!err) {
1941 nn->nb = nb;
1942 list_add(&nn->list, &dev->net_notifier_list);
1943 }
1944 rtnl_unlock();
1945 return err;
1946 }
1947 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1948
unregister_netdevice_notifier_dev_net(struct net_device * dev,struct notifier_block * nb,struct netdev_net_notifier * nn)1949 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1950 struct notifier_block *nb,
1951 struct netdev_net_notifier *nn)
1952 {
1953 int err;
1954
1955 rtnl_lock();
1956 list_del(&nn->list);
1957 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1958 rtnl_unlock();
1959 return err;
1960 }
1961 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1962
move_netdevice_notifiers_dev_net(struct net_device * dev,struct net * net)1963 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1964 struct net *net)
1965 {
1966 struct netdev_net_notifier *nn;
1967
1968 list_for_each_entry(nn, &dev->net_notifier_list, list)
1969 __move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1970 }
1971
1972 /**
1973 * call_netdevice_notifiers_info - call all network notifier blocks
1974 * @val: value passed unmodified to notifier function
1975 * @info: notifier information data
1976 *
1977 * Call all network notifier blocks. Parameters and return value
1978 * are as for raw_notifier_call_chain().
1979 */
1980
call_netdevice_notifiers_info(unsigned long val,struct netdev_notifier_info * info)1981 int call_netdevice_notifiers_info(unsigned long val,
1982 struct netdev_notifier_info *info)
1983 {
1984 struct net *net = dev_net(info->dev);
1985 int ret;
1986
1987 ASSERT_RTNL();
1988
1989 /* Run per-netns notifier block chain first, then run the global one.
1990 * Hopefully, one day, the global one is going to be removed after
1991 * all notifier block registrators get converted to be per-netns.
1992 */
1993 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1994 if (ret & NOTIFY_STOP_MASK)
1995 return ret;
1996 return raw_notifier_call_chain(&netdev_chain, val, info);
1997 }
1998
1999 /**
2000 * call_netdevice_notifiers_info_robust - call per-netns notifier blocks
2001 * for and rollback on error
2002 * @val_up: value passed unmodified to notifier function
2003 * @val_down: value passed unmodified to the notifier function when
2004 * recovering from an error on @val_up
2005 * @info: notifier information data
2006 *
2007 * Call all per-netns network notifier blocks, but not notifier blocks on
2008 * the global notifier chain. Parameters and return value are as for
2009 * raw_notifier_call_chain_robust().
2010 */
2011
2012 static int
call_netdevice_notifiers_info_robust(unsigned long val_up,unsigned long val_down,struct netdev_notifier_info * info)2013 call_netdevice_notifiers_info_robust(unsigned long val_up,
2014 unsigned long val_down,
2015 struct netdev_notifier_info *info)
2016 {
2017 struct net *net = dev_net(info->dev);
2018
2019 ASSERT_RTNL();
2020
2021 return raw_notifier_call_chain_robust(&net->netdev_chain,
2022 val_up, val_down, info);
2023 }
2024
call_netdevice_notifiers_extack(unsigned long val,struct net_device * dev,struct netlink_ext_ack * extack)2025 static int call_netdevice_notifiers_extack(unsigned long val,
2026 struct net_device *dev,
2027 struct netlink_ext_ack *extack)
2028 {
2029 struct netdev_notifier_info info = {
2030 .dev = dev,
2031 .extack = extack,
2032 };
2033
2034 return call_netdevice_notifiers_info(val, &info);
2035 }
2036
2037 /**
2038 * call_netdevice_notifiers - call all network notifier blocks
2039 * @val: value passed unmodified to notifier function
2040 * @dev: net_device pointer passed unmodified to notifier function
2041 *
2042 * Call all network notifier blocks. Parameters and return value
2043 * are as for raw_notifier_call_chain().
2044 */
2045
call_netdevice_notifiers(unsigned long val,struct net_device * dev)2046 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2047 {
2048 return call_netdevice_notifiers_extack(val, dev, NULL);
2049 }
2050 EXPORT_SYMBOL(call_netdevice_notifiers);
2051
2052 /**
2053 * call_netdevice_notifiers_mtu - call all network notifier blocks
2054 * @val: value passed unmodified to notifier function
2055 * @dev: net_device pointer passed unmodified to notifier function
2056 * @arg: additional u32 argument passed to the notifier function
2057 *
2058 * Call all network notifier blocks. Parameters and return value
2059 * are as for raw_notifier_call_chain().
2060 */
call_netdevice_notifiers_mtu(unsigned long val,struct net_device * dev,u32 arg)2061 static int call_netdevice_notifiers_mtu(unsigned long val,
2062 struct net_device *dev, u32 arg)
2063 {
2064 struct netdev_notifier_info_ext info = {
2065 .info.dev = dev,
2066 .ext.mtu = arg,
2067 };
2068
2069 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2070
2071 return call_netdevice_notifiers_info(val, &info.info);
2072 }
2073
2074 #ifdef CONFIG_NET_INGRESS
2075 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2076
net_inc_ingress_queue(void)2077 void net_inc_ingress_queue(void)
2078 {
2079 static_branch_inc(&ingress_needed_key);
2080 }
2081 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2082
net_dec_ingress_queue(void)2083 void net_dec_ingress_queue(void)
2084 {
2085 static_branch_dec(&ingress_needed_key);
2086 }
2087 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2088 #endif
2089
2090 #ifdef CONFIG_NET_EGRESS
2091 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2092
net_inc_egress_queue(void)2093 void net_inc_egress_queue(void)
2094 {
2095 static_branch_inc(&egress_needed_key);
2096 }
2097 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2098
net_dec_egress_queue(void)2099 void net_dec_egress_queue(void)
2100 {
2101 static_branch_dec(&egress_needed_key);
2102 }
2103 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2104 #endif
2105
2106 #ifdef CONFIG_NET_CLS_ACT
2107 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key);
2108 EXPORT_SYMBOL(tcf_bypass_check_needed_key);
2109 #endif
2110
2111 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2112 EXPORT_SYMBOL(netstamp_needed_key);
2113 #ifdef CONFIG_JUMP_LABEL
2114 static atomic_t netstamp_needed_deferred;
2115 static atomic_t netstamp_wanted;
netstamp_clear(struct work_struct * work)2116 static void netstamp_clear(struct work_struct *work)
2117 {
2118 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2119 int wanted;
2120
2121 wanted = atomic_add_return(deferred, &netstamp_wanted);
2122 if (wanted > 0)
2123 static_branch_enable(&netstamp_needed_key);
2124 else
2125 static_branch_disable(&netstamp_needed_key);
2126 }
2127 static DECLARE_WORK(netstamp_work, netstamp_clear);
2128 #endif
2129
net_enable_timestamp(void)2130 void net_enable_timestamp(void)
2131 {
2132 #ifdef CONFIG_JUMP_LABEL
2133 int wanted = atomic_read(&netstamp_wanted);
2134
2135 while (wanted > 0) {
2136 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2137 return;
2138 }
2139 atomic_inc(&netstamp_needed_deferred);
2140 schedule_work(&netstamp_work);
2141 #else
2142 static_branch_inc(&netstamp_needed_key);
2143 #endif
2144 }
2145 EXPORT_SYMBOL(net_enable_timestamp);
2146
net_disable_timestamp(void)2147 void net_disable_timestamp(void)
2148 {
2149 #ifdef CONFIG_JUMP_LABEL
2150 int wanted = atomic_read(&netstamp_wanted);
2151
2152 while (wanted > 1) {
2153 if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2154 return;
2155 }
2156 atomic_dec(&netstamp_needed_deferred);
2157 schedule_work(&netstamp_work);
2158 #else
2159 static_branch_dec(&netstamp_needed_key);
2160 #endif
2161 }
2162 EXPORT_SYMBOL(net_disable_timestamp);
2163
net_timestamp_set(struct sk_buff * skb)2164 static inline void net_timestamp_set(struct sk_buff *skb)
2165 {
2166 skb->tstamp = 0;
2167 skb->tstamp_type = SKB_CLOCK_REALTIME;
2168 if (static_branch_unlikely(&netstamp_needed_key))
2169 skb->tstamp = ktime_get_real();
2170 }
2171
2172 #define net_timestamp_check(COND, SKB) \
2173 if (static_branch_unlikely(&netstamp_needed_key)) { \
2174 if ((COND) && !(SKB)->tstamp) \
2175 (SKB)->tstamp = ktime_get_real(); \
2176 } \
2177
is_skb_forwardable(const struct net_device * dev,const struct sk_buff * skb)2178 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2179 {
2180 return __is_skb_forwardable(dev, skb, true);
2181 }
2182 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2183
__dev_forward_skb2(struct net_device * dev,struct sk_buff * skb,bool check_mtu)2184 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2185 bool check_mtu)
2186 {
2187 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2188
2189 if (likely(!ret)) {
2190 skb->protocol = eth_type_trans(skb, dev);
2191 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2192 }
2193
2194 return ret;
2195 }
2196
__dev_forward_skb(struct net_device * dev,struct sk_buff * skb)2197 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2198 {
2199 return __dev_forward_skb2(dev, skb, true);
2200 }
2201 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2202
2203 /**
2204 * dev_forward_skb - loopback an skb to another netif
2205 *
2206 * @dev: destination network device
2207 * @skb: buffer to forward
2208 *
2209 * return values:
2210 * NET_RX_SUCCESS (no congestion)
2211 * NET_RX_DROP (packet was dropped, but freed)
2212 *
2213 * dev_forward_skb can be used for injecting an skb from the
2214 * start_xmit function of one device into the receive queue
2215 * of another device.
2216 *
2217 * The receiving device may be in another namespace, so
2218 * we have to clear all information in the skb that could
2219 * impact namespace isolation.
2220 */
dev_forward_skb(struct net_device * dev,struct sk_buff * skb)2221 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2222 {
2223 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2224 }
2225 EXPORT_SYMBOL_GPL(dev_forward_skb);
2226
dev_forward_skb_nomtu(struct net_device * dev,struct sk_buff * skb)2227 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2228 {
2229 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2230 }
2231
deliver_skb(struct sk_buff * skb,struct packet_type * pt_prev,struct net_device * orig_dev)2232 static inline int deliver_skb(struct sk_buff *skb,
2233 struct packet_type *pt_prev,
2234 struct net_device *orig_dev)
2235 {
2236 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2237 return -ENOMEM;
2238 refcount_inc(&skb->users);
2239 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2240 }
2241
deliver_ptype_list_skb(struct sk_buff * skb,struct packet_type ** pt,struct net_device * orig_dev,__be16 type,struct list_head * ptype_list)2242 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2243 struct packet_type **pt,
2244 struct net_device *orig_dev,
2245 __be16 type,
2246 struct list_head *ptype_list)
2247 {
2248 struct packet_type *ptype, *pt_prev = *pt;
2249
2250 list_for_each_entry_rcu(ptype, ptype_list, list) {
2251 if (ptype->type != type)
2252 continue;
2253 if (pt_prev)
2254 deliver_skb(skb, pt_prev, orig_dev);
2255 pt_prev = ptype;
2256 }
2257 *pt = pt_prev;
2258 }
2259
skb_loop_sk(struct packet_type * ptype,struct sk_buff * skb)2260 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2261 {
2262 if (!ptype->af_packet_priv || !skb->sk)
2263 return false;
2264
2265 if (ptype->id_match)
2266 return ptype->id_match(ptype, skb->sk);
2267 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2268 return true;
2269
2270 return false;
2271 }
2272
2273 /**
2274 * dev_nit_active - return true if any network interface taps are in use
2275 *
2276 * @dev: network device to check for the presence of taps
2277 */
dev_nit_active(struct net_device * dev)2278 bool dev_nit_active(struct net_device *dev)
2279 {
2280 return !list_empty(&net_hotdata.ptype_all) ||
2281 !list_empty(&dev->ptype_all);
2282 }
2283 EXPORT_SYMBOL_GPL(dev_nit_active);
2284
2285 /*
2286 * Support routine. Sends outgoing frames to any network
2287 * taps currently in use.
2288 */
2289
dev_queue_xmit_nit(struct sk_buff * skb,struct net_device * dev)2290 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2291 {
2292 struct list_head *ptype_list = &net_hotdata.ptype_all;
2293 struct packet_type *ptype, *pt_prev = NULL;
2294 struct sk_buff *skb2 = NULL;
2295
2296 rcu_read_lock();
2297 again:
2298 list_for_each_entry_rcu(ptype, ptype_list, list) {
2299 if (READ_ONCE(ptype->ignore_outgoing))
2300 continue;
2301
2302 /* Never send packets back to the socket
2303 * they originated from - MvS (miquels@drinkel.ow.org)
2304 */
2305 if (skb_loop_sk(ptype, skb))
2306 continue;
2307
2308 if (pt_prev) {
2309 deliver_skb(skb2, pt_prev, skb->dev);
2310 pt_prev = ptype;
2311 continue;
2312 }
2313
2314 /* need to clone skb, done only once */
2315 skb2 = skb_clone(skb, GFP_ATOMIC);
2316 if (!skb2)
2317 goto out_unlock;
2318
2319 net_timestamp_set(skb2);
2320
2321 /* skb->nh should be correctly
2322 * set by sender, so that the second statement is
2323 * just protection against buggy protocols.
2324 */
2325 skb_reset_mac_header(skb2);
2326
2327 if (skb_network_header(skb2) < skb2->data ||
2328 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2329 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2330 ntohs(skb2->protocol),
2331 dev->name);
2332 skb_reset_network_header(skb2);
2333 }
2334
2335 skb2->transport_header = skb2->network_header;
2336 skb2->pkt_type = PACKET_OUTGOING;
2337 pt_prev = ptype;
2338 }
2339
2340 if (ptype_list == &net_hotdata.ptype_all) {
2341 ptype_list = &dev->ptype_all;
2342 goto again;
2343 }
2344 out_unlock:
2345 if (pt_prev) {
2346 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2347 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2348 else
2349 kfree_skb(skb2);
2350 }
2351 rcu_read_unlock();
2352 }
2353 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2354
2355 /**
2356 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2357 * @dev: Network device
2358 * @txq: number of queues available
2359 *
2360 * If real_num_tx_queues is changed the tc mappings may no longer be
2361 * valid. To resolve this verify the tc mapping remains valid and if
2362 * not NULL the mapping. With no priorities mapping to this
2363 * offset/count pair it will no longer be used. In the worst case TC0
2364 * is invalid nothing can be done so disable priority mappings. If is
2365 * expected that drivers will fix this mapping if they can before
2366 * calling netif_set_real_num_tx_queues.
2367 */
netif_setup_tc(struct net_device * dev,unsigned int txq)2368 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2369 {
2370 int i;
2371 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2372
2373 /* If TC0 is invalidated disable TC mapping */
2374 if (tc->offset + tc->count > txq) {
2375 netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2376 dev->num_tc = 0;
2377 return;
2378 }
2379
2380 /* Invalidated prio to tc mappings set to TC0 */
2381 for (i = 1; i < TC_BITMASK + 1; i++) {
2382 int q = netdev_get_prio_tc_map(dev, i);
2383
2384 tc = &dev->tc_to_txq[q];
2385 if (tc->offset + tc->count > txq) {
2386 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",
2387 i, q);
2388 netdev_set_prio_tc_map(dev, i, 0);
2389 }
2390 }
2391 }
2392
netdev_txq_to_tc(struct net_device * dev,unsigned int txq)2393 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2394 {
2395 if (dev->num_tc) {
2396 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2397 int i;
2398
2399 /* walk through the TCs and see if it falls into any of them */
2400 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2401 if ((txq - tc->offset) < tc->count)
2402 return i;
2403 }
2404
2405 /* didn't find it, just return -1 to indicate no match */
2406 return -1;
2407 }
2408
2409 return 0;
2410 }
2411 EXPORT_SYMBOL(netdev_txq_to_tc);
2412
2413 #ifdef CONFIG_XPS
2414 static struct static_key xps_needed __read_mostly;
2415 static struct static_key xps_rxqs_needed __read_mostly;
2416 static DEFINE_MUTEX(xps_map_mutex);
2417 #define xmap_dereference(P) \
2418 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2419
remove_xps_queue(struct xps_dev_maps * dev_maps,struct xps_dev_maps * old_maps,int tci,u16 index)2420 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2421 struct xps_dev_maps *old_maps, int tci, u16 index)
2422 {
2423 struct xps_map *map = NULL;
2424 int pos;
2425
2426 map = xmap_dereference(dev_maps->attr_map[tci]);
2427 if (!map)
2428 return false;
2429
2430 for (pos = map->len; pos--;) {
2431 if (map->queues[pos] != index)
2432 continue;
2433
2434 if (map->len > 1) {
2435 map->queues[pos] = map->queues[--map->len];
2436 break;
2437 }
2438
2439 if (old_maps)
2440 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2441 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2442 kfree_rcu(map, rcu);
2443 return false;
2444 }
2445
2446 return true;
2447 }
2448
remove_xps_queue_cpu(struct net_device * dev,struct xps_dev_maps * dev_maps,int cpu,u16 offset,u16 count)2449 static bool remove_xps_queue_cpu(struct net_device *dev,
2450 struct xps_dev_maps *dev_maps,
2451 int cpu, u16 offset, u16 count)
2452 {
2453 int num_tc = dev_maps->num_tc;
2454 bool active = false;
2455 int tci;
2456
2457 for (tci = cpu * num_tc; num_tc--; tci++) {
2458 int i, j;
2459
2460 for (i = count, j = offset; i--; j++) {
2461 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2462 break;
2463 }
2464
2465 active |= i < 0;
2466 }
2467
2468 return active;
2469 }
2470
reset_xps_maps(struct net_device * dev,struct xps_dev_maps * dev_maps,enum xps_map_type type)2471 static void reset_xps_maps(struct net_device *dev,
2472 struct xps_dev_maps *dev_maps,
2473 enum xps_map_type type)
2474 {
2475 static_key_slow_dec_cpuslocked(&xps_needed);
2476 if (type == XPS_RXQS)
2477 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2478
2479 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2480
2481 kfree_rcu(dev_maps, rcu);
2482 }
2483
clean_xps_maps(struct net_device * dev,enum xps_map_type type,u16 offset,u16 count)2484 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2485 u16 offset, u16 count)
2486 {
2487 struct xps_dev_maps *dev_maps;
2488 bool active = false;
2489 int i, j;
2490
2491 dev_maps = xmap_dereference(dev->xps_maps[type]);
2492 if (!dev_maps)
2493 return;
2494
2495 for (j = 0; j < dev_maps->nr_ids; j++)
2496 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2497 if (!active)
2498 reset_xps_maps(dev, dev_maps, type);
2499
2500 if (type == XPS_CPUS) {
2501 for (i = offset + (count - 1); count--; i--)
2502 netdev_queue_numa_node_write(
2503 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2504 }
2505 }
2506
netif_reset_xps_queues(struct net_device * dev,u16 offset,u16 count)2507 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2508 u16 count)
2509 {
2510 if (!static_key_false(&xps_needed))
2511 return;
2512
2513 cpus_read_lock();
2514 mutex_lock(&xps_map_mutex);
2515
2516 if (static_key_false(&xps_rxqs_needed))
2517 clean_xps_maps(dev, XPS_RXQS, offset, count);
2518
2519 clean_xps_maps(dev, XPS_CPUS, offset, count);
2520
2521 mutex_unlock(&xps_map_mutex);
2522 cpus_read_unlock();
2523 }
2524
netif_reset_xps_queues_gt(struct net_device * dev,u16 index)2525 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2526 {
2527 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2528 }
2529
expand_xps_map(struct xps_map * map,int attr_index,u16 index,bool is_rxqs_map)2530 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2531 u16 index, bool is_rxqs_map)
2532 {
2533 struct xps_map *new_map;
2534 int alloc_len = XPS_MIN_MAP_ALLOC;
2535 int i, pos;
2536
2537 for (pos = 0; map && pos < map->len; pos++) {
2538 if (map->queues[pos] != index)
2539 continue;
2540 return map;
2541 }
2542
2543 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2544 if (map) {
2545 if (pos < map->alloc_len)
2546 return map;
2547
2548 alloc_len = map->alloc_len * 2;
2549 }
2550
2551 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2552 * map
2553 */
2554 if (is_rxqs_map)
2555 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2556 else
2557 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2558 cpu_to_node(attr_index));
2559 if (!new_map)
2560 return NULL;
2561
2562 for (i = 0; i < pos; i++)
2563 new_map->queues[i] = map->queues[i];
2564 new_map->alloc_len = alloc_len;
2565 new_map->len = pos;
2566
2567 return new_map;
2568 }
2569
2570 /* 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)2571 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2572 struct xps_dev_maps *new_dev_maps, int index,
2573 int tc, bool skip_tc)
2574 {
2575 int i, tci = index * dev_maps->num_tc;
2576 struct xps_map *map;
2577
2578 /* copy maps belonging to foreign traffic classes */
2579 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2580 if (i == tc && skip_tc)
2581 continue;
2582
2583 /* fill in the new device map from the old device map */
2584 map = xmap_dereference(dev_maps->attr_map[tci]);
2585 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2586 }
2587 }
2588
2589 /* 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)2590 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2591 u16 index, enum xps_map_type type)
2592 {
2593 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2594 const unsigned long *online_mask = NULL;
2595 bool active = false, copy = false;
2596 int i, j, tci, numa_node_id = -2;
2597 int maps_sz, num_tc = 1, tc = 0;
2598 struct xps_map *map, *new_map;
2599 unsigned int nr_ids;
2600
2601 WARN_ON_ONCE(index >= dev->num_tx_queues);
2602
2603 if (dev->num_tc) {
2604 /* Do not allow XPS on subordinate device directly */
2605 num_tc = dev->num_tc;
2606 if (num_tc < 0)
2607 return -EINVAL;
2608
2609 /* If queue belongs to subordinate dev use its map */
2610 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2611
2612 tc = netdev_txq_to_tc(dev, index);
2613 if (tc < 0)
2614 return -EINVAL;
2615 }
2616
2617 mutex_lock(&xps_map_mutex);
2618
2619 dev_maps = xmap_dereference(dev->xps_maps[type]);
2620 if (type == XPS_RXQS) {
2621 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2622 nr_ids = dev->num_rx_queues;
2623 } else {
2624 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2625 if (num_possible_cpus() > 1)
2626 online_mask = cpumask_bits(cpu_online_mask);
2627 nr_ids = nr_cpu_ids;
2628 }
2629
2630 if (maps_sz < L1_CACHE_BYTES)
2631 maps_sz = L1_CACHE_BYTES;
2632
2633 /* The old dev_maps could be larger or smaller than the one we're
2634 * setting up now, as dev->num_tc or nr_ids could have been updated in
2635 * between. We could try to be smart, but let's be safe instead and only
2636 * copy foreign traffic classes if the two map sizes match.
2637 */
2638 if (dev_maps &&
2639 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2640 copy = true;
2641
2642 /* allocate memory for queue storage */
2643 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2644 j < nr_ids;) {
2645 if (!new_dev_maps) {
2646 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2647 if (!new_dev_maps) {
2648 mutex_unlock(&xps_map_mutex);
2649 return -ENOMEM;
2650 }
2651
2652 new_dev_maps->nr_ids = nr_ids;
2653 new_dev_maps->num_tc = num_tc;
2654 }
2655
2656 tci = j * num_tc + tc;
2657 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2658
2659 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2660 if (!map)
2661 goto error;
2662
2663 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2664 }
2665
2666 if (!new_dev_maps)
2667 goto out_no_new_maps;
2668
2669 if (!dev_maps) {
2670 /* Increment static keys at most once per type */
2671 static_key_slow_inc_cpuslocked(&xps_needed);
2672 if (type == XPS_RXQS)
2673 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2674 }
2675
2676 for (j = 0; j < nr_ids; j++) {
2677 bool skip_tc = false;
2678
2679 tci = j * num_tc + tc;
2680 if (netif_attr_test_mask(j, mask, nr_ids) &&
2681 netif_attr_test_online(j, online_mask, nr_ids)) {
2682 /* add tx-queue to CPU/rx-queue maps */
2683 int pos = 0;
2684
2685 skip_tc = true;
2686
2687 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2688 while ((pos < map->len) && (map->queues[pos] != index))
2689 pos++;
2690
2691 if (pos == map->len)
2692 map->queues[map->len++] = index;
2693 #ifdef CONFIG_NUMA
2694 if (type == XPS_CPUS) {
2695 if (numa_node_id == -2)
2696 numa_node_id = cpu_to_node(j);
2697 else if (numa_node_id != cpu_to_node(j))
2698 numa_node_id = -1;
2699 }
2700 #endif
2701 }
2702
2703 if (copy)
2704 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2705 skip_tc);
2706 }
2707
2708 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2709
2710 /* Cleanup old maps */
2711 if (!dev_maps)
2712 goto out_no_old_maps;
2713
2714 for (j = 0; j < dev_maps->nr_ids; j++) {
2715 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2716 map = xmap_dereference(dev_maps->attr_map[tci]);
2717 if (!map)
2718 continue;
2719
2720 if (copy) {
2721 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2722 if (map == new_map)
2723 continue;
2724 }
2725
2726 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2727 kfree_rcu(map, rcu);
2728 }
2729 }
2730
2731 old_dev_maps = dev_maps;
2732
2733 out_no_old_maps:
2734 dev_maps = new_dev_maps;
2735 active = true;
2736
2737 out_no_new_maps:
2738 if (type == XPS_CPUS)
2739 /* update Tx queue numa node */
2740 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2741 (numa_node_id >= 0) ?
2742 numa_node_id : NUMA_NO_NODE);
2743
2744 if (!dev_maps)
2745 goto out_no_maps;
2746
2747 /* removes tx-queue from unused CPUs/rx-queues */
2748 for (j = 0; j < dev_maps->nr_ids; j++) {
2749 tci = j * dev_maps->num_tc;
2750
2751 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2752 if (i == tc &&
2753 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2754 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2755 continue;
2756
2757 active |= remove_xps_queue(dev_maps,
2758 copy ? old_dev_maps : NULL,
2759 tci, index);
2760 }
2761 }
2762
2763 if (old_dev_maps)
2764 kfree_rcu(old_dev_maps, rcu);
2765
2766 /* free map if not active */
2767 if (!active)
2768 reset_xps_maps(dev, dev_maps, type);
2769
2770 out_no_maps:
2771 mutex_unlock(&xps_map_mutex);
2772
2773 return 0;
2774 error:
2775 /* remove any maps that we added */
2776 for (j = 0; j < nr_ids; j++) {
2777 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2778 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2779 map = copy ?
2780 xmap_dereference(dev_maps->attr_map[tci]) :
2781 NULL;
2782 if (new_map && new_map != map)
2783 kfree(new_map);
2784 }
2785 }
2786
2787 mutex_unlock(&xps_map_mutex);
2788
2789 kfree(new_dev_maps);
2790 return -ENOMEM;
2791 }
2792 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2793
netif_set_xps_queue(struct net_device * dev,const struct cpumask * mask,u16 index)2794 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2795 u16 index)
2796 {
2797 int ret;
2798
2799 cpus_read_lock();
2800 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2801 cpus_read_unlock();
2802
2803 return ret;
2804 }
2805 EXPORT_SYMBOL(netif_set_xps_queue);
2806
2807 #endif
netdev_unbind_all_sb_channels(struct net_device * dev)2808 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2809 {
2810 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2811
2812 /* Unbind any subordinate channels */
2813 while (txq-- != &dev->_tx[0]) {
2814 if (txq->sb_dev)
2815 netdev_unbind_sb_channel(dev, txq->sb_dev);
2816 }
2817 }
2818
netdev_reset_tc(struct net_device * dev)2819 void netdev_reset_tc(struct net_device *dev)
2820 {
2821 #ifdef CONFIG_XPS
2822 netif_reset_xps_queues_gt(dev, 0);
2823 #endif
2824 netdev_unbind_all_sb_channels(dev);
2825
2826 /* Reset TC configuration of device */
2827 dev->num_tc = 0;
2828 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2829 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2830 }
2831 EXPORT_SYMBOL(netdev_reset_tc);
2832
netdev_set_tc_queue(struct net_device * dev,u8 tc,u16 count,u16 offset)2833 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2834 {
2835 if (tc >= dev->num_tc)
2836 return -EINVAL;
2837
2838 #ifdef CONFIG_XPS
2839 netif_reset_xps_queues(dev, offset, count);
2840 #endif
2841 dev->tc_to_txq[tc].count = count;
2842 dev->tc_to_txq[tc].offset = offset;
2843 return 0;
2844 }
2845 EXPORT_SYMBOL(netdev_set_tc_queue);
2846
netdev_set_num_tc(struct net_device * dev,u8 num_tc)2847 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2848 {
2849 if (num_tc > TC_MAX_QUEUE)
2850 return -EINVAL;
2851
2852 #ifdef CONFIG_XPS
2853 netif_reset_xps_queues_gt(dev, 0);
2854 #endif
2855 netdev_unbind_all_sb_channels(dev);
2856
2857 dev->num_tc = num_tc;
2858 return 0;
2859 }
2860 EXPORT_SYMBOL(netdev_set_num_tc);
2861
netdev_unbind_sb_channel(struct net_device * dev,struct net_device * sb_dev)2862 void netdev_unbind_sb_channel(struct net_device *dev,
2863 struct net_device *sb_dev)
2864 {
2865 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2866
2867 #ifdef CONFIG_XPS
2868 netif_reset_xps_queues_gt(sb_dev, 0);
2869 #endif
2870 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2871 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2872
2873 while (txq-- != &dev->_tx[0]) {
2874 if (txq->sb_dev == sb_dev)
2875 txq->sb_dev = NULL;
2876 }
2877 }
2878 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2879
netdev_bind_sb_channel_queue(struct net_device * dev,struct net_device * sb_dev,u8 tc,u16 count,u16 offset)2880 int netdev_bind_sb_channel_queue(struct net_device *dev,
2881 struct net_device *sb_dev,
2882 u8 tc, u16 count, u16 offset)
2883 {
2884 /* Make certain the sb_dev and dev are already configured */
2885 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2886 return -EINVAL;
2887
2888 /* We cannot hand out queues we don't have */
2889 if ((offset + count) > dev->real_num_tx_queues)
2890 return -EINVAL;
2891
2892 /* Record the mapping */
2893 sb_dev->tc_to_txq[tc].count = count;
2894 sb_dev->tc_to_txq[tc].offset = offset;
2895
2896 /* Provide a way for Tx queue to find the tc_to_txq map or
2897 * XPS map for itself.
2898 */
2899 while (count--)
2900 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2901
2902 return 0;
2903 }
2904 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2905
netdev_set_sb_channel(struct net_device * dev,u16 channel)2906 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2907 {
2908 /* Do not use a multiqueue device to represent a subordinate channel */
2909 if (netif_is_multiqueue(dev))
2910 return -ENODEV;
2911
2912 /* We allow channels 1 - 32767 to be used for subordinate channels.
2913 * Channel 0 is meant to be "native" mode and used only to represent
2914 * the main root device. We allow writing 0 to reset the device back
2915 * to normal mode after being used as a subordinate channel.
2916 */
2917 if (channel > S16_MAX)
2918 return -EINVAL;
2919
2920 dev->num_tc = -channel;
2921
2922 return 0;
2923 }
2924 EXPORT_SYMBOL(netdev_set_sb_channel);
2925
2926 /*
2927 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2928 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2929 */
netif_set_real_num_tx_queues(struct net_device * dev,unsigned int txq)2930 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2931 {
2932 bool disabling;
2933 int rc;
2934
2935 disabling = txq < dev->real_num_tx_queues;
2936
2937 if (txq < 1 || txq > dev->num_tx_queues)
2938 return -EINVAL;
2939
2940 if (dev->reg_state == NETREG_REGISTERED ||
2941 dev->reg_state == NETREG_UNREGISTERING) {
2942 ASSERT_RTNL();
2943
2944 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2945 txq);
2946 if (rc)
2947 return rc;
2948
2949 if (dev->num_tc)
2950 netif_setup_tc(dev, txq);
2951
2952 dev_qdisc_change_real_num_tx(dev, txq);
2953
2954 dev->real_num_tx_queues = txq;
2955
2956 if (disabling) {
2957 synchronize_net();
2958 qdisc_reset_all_tx_gt(dev, txq);
2959 #ifdef CONFIG_XPS
2960 netif_reset_xps_queues_gt(dev, txq);
2961 #endif
2962 }
2963 } else {
2964 dev->real_num_tx_queues = txq;
2965 }
2966
2967 return 0;
2968 }
2969 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2970
2971 #ifdef CONFIG_SYSFS
2972 /**
2973 * netif_set_real_num_rx_queues - set actual number of RX queues used
2974 * @dev: Network device
2975 * @rxq: Actual number of RX queues
2976 *
2977 * This must be called either with the rtnl_lock held or before
2978 * registration of the net device. Returns 0 on success, or a
2979 * negative error code. If called before registration, it always
2980 * succeeds.
2981 */
netif_set_real_num_rx_queues(struct net_device * dev,unsigned int rxq)2982 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2983 {
2984 int rc;
2985
2986 if (rxq < 1 || rxq > dev->num_rx_queues)
2987 return -EINVAL;
2988
2989 if (dev->reg_state == NETREG_REGISTERED) {
2990 ASSERT_RTNL();
2991
2992 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2993 rxq);
2994 if (rc)
2995 return rc;
2996 }
2997
2998 dev->real_num_rx_queues = rxq;
2999 return 0;
3000 }
3001 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3002 #endif
3003
3004 /**
3005 * netif_set_real_num_queues - set actual number of RX and TX queues used
3006 * @dev: Network device
3007 * @txq: Actual number of TX queues
3008 * @rxq: Actual number of RX queues
3009 *
3010 * Set the real number of both TX and RX queues.
3011 * Does nothing if the number of queues is already correct.
3012 */
netif_set_real_num_queues(struct net_device * dev,unsigned int txq,unsigned int rxq)3013 int netif_set_real_num_queues(struct net_device *dev,
3014 unsigned int txq, unsigned int rxq)
3015 {
3016 unsigned int old_rxq = dev->real_num_rx_queues;
3017 int err;
3018
3019 if (txq < 1 || txq > dev->num_tx_queues ||
3020 rxq < 1 || rxq > dev->num_rx_queues)
3021 return -EINVAL;
3022
3023 /* Start from increases, so the error path only does decreases -
3024 * decreases can't fail.
3025 */
3026 if (rxq > dev->real_num_rx_queues) {
3027 err = netif_set_real_num_rx_queues(dev, rxq);
3028 if (err)
3029 return err;
3030 }
3031 if (txq > dev->real_num_tx_queues) {
3032 err = netif_set_real_num_tx_queues(dev, txq);
3033 if (err)
3034 goto undo_rx;
3035 }
3036 if (rxq < dev->real_num_rx_queues)
3037 WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3038 if (txq < dev->real_num_tx_queues)
3039 WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3040
3041 return 0;
3042 undo_rx:
3043 WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3044 return err;
3045 }
3046 EXPORT_SYMBOL(netif_set_real_num_queues);
3047
3048 /**
3049 * netif_set_tso_max_size() - set the max size of TSO frames supported
3050 * @dev: netdev to update
3051 * @size: max skb->len of a TSO frame
3052 *
3053 * Set the limit on the size of TSO super-frames the device can handle.
3054 * Unless explicitly set the stack will assume the value of
3055 * %GSO_LEGACY_MAX_SIZE.
3056 */
netif_set_tso_max_size(struct net_device * dev,unsigned int size)3057 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3058 {
3059 dev->tso_max_size = min(GSO_MAX_SIZE, size);
3060 if (size < READ_ONCE(dev->gso_max_size))
3061 netif_set_gso_max_size(dev, size);
3062 if (size < READ_ONCE(dev->gso_ipv4_max_size))
3063 netif_set_gso_ipv4_max_size(dev, size);
3064 }
3065 EXPORT_SYMBOL(netif_set_tso_max_size);
3066
3067 /**
3068 * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3069 * @dev: netdev to update
3070 * @segs: max number of TCP segments
3071 *
3072 * Set the limit on the number of TCP segments the device can generate from
3073 * a single TSO super-frame.
3074 * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3075 */
netif_set_tso_max_segs(struct net_device * dev,unsigned int segs)3076 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3077 {
3078 dev->tso_max_segs = segs;
3079 if (segs < READ_ONCE(dev->gso_max_segs))
3080 netif_set_gso_max_segs(dev, segs);
3081 }
3082 EXPORT_SYMBOL(netif_set_tso_max_segs);
3083
3084 /**
3085 * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3086 * @to: netdev to update
3087 * @from: netdev from which to copy the limits
3088 */
netif_inherit_tso_max(struct net_device * to,const struct net_device * from)3089 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3090 {
3091 netif_set_tso_max_size(to, from->tso_max_size);
3092 netif_set_tso_max_segs(to, from->tso_max_segs);
3093 }
3094 EXPORT_SYMBOL(netif_inherit_tso_max);
3095
3096 /**
3097 * netif_get_num_default_rss_queues - default number of RSS queues
3098 *
3099 * Default value is the number of physical cores if there are only 1 or 2, or
3100 * divided by 2 if there are more.
3101 */
netif_get_num_default_rss_queues(void)3102 int netif_get_num_default_rss_queues(void)
3103 {
3104 cpumask_var_t cpus;
3105 int cpu, count = 0;
3106
3107 if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3108 return 1;
3109
3110 cpumask_copy(cpus, cpu_online_mask);
3111 for_each_cpu(cpu, cpus) {
3112 ++count;
3113 cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3114 }
3115 free_cpumask_var(cpus);
3116
3117 return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3118 }
3119 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3120
__netif_reschedule(struct Qdisc * q)3121 static void __netif_reschedule(struct Qdisc *q)
3122 {
3123 struct softnet_data *sd;
3124 unsigned long flags;
3125
3126 local_irq_save(flags);
3127 sd = this_cpu_ptr(&softnet_data);
3128 q->next_sched = NULL;
3129 *sd->output_queue_tailp = q;
3130 sd->output_queue_tailp = &q->next_sched;
3131 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3132 local_irq_restore(flags);
3133 }
3134
__netif_schedule(struct Qdisc * q)3135 void __netif_schedule(struct Qdisc *q)
3136 {
3137 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3138 __netif_reschedule(q);
3139 }
3140 EXPORT_SYMBOL(__netif_schedule);
3141
3142 struct dev_kfree_skb_cb {
3143 enum skb_drop_reason reason;
3144 };
3145
get_kfree_skb_cb(const struct sk_buff * skb)3146 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3147 {
3148 return (struct dev_kfree_skb_cb *)skb->cb;
3149 }
3150
netif_schedule_queue(struct netdev_queue * txq)3151 void netif_schedule_queue(struct netdev_queue *txq)
3152 {
3153 rcu_read_lock();
3154 if (!netif_xmit_stopped(txq)) {
3155 struct Qdisc *q = rcu_dereference(txq->qdisc);
3156
3157 __netif_schedule(q);
3158 }
3159 rcu_read_unlock();
3160 }
3161 EXPORT_SYMBOL(netif_schedule_queue);
3162
netif_tx_wake_queue(struct netdev_queue * dev_queue)3163 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3164 {
3165 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3166 struct Qdisc *q;
3167
3168 rcu_read_lock();
3169 q = rcu_dereference(dev_queue->qdisc);
3170 __netif_schedule(q);
3171 rcu_read_unlock();
3172 }
3173 }
3174 EXPORT_SYMBOL(netif_tx_wake_queue);
3175
dev_kfree_skb_irq_reason(struct sk_buff * skb,enum skb_drop_reason reason)3176 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3177 {
3178 unsigned long flags;
3179
3180 if (unlikely(!skb))
3181 return;
3182
3183 if (likely(refcount_read(&skb->users) == 1)) {
3184 smp_rmb();
3185 refcount_set(&skb->users, 0);
3186 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3187 return;
3188 }
3189 get_kfree_skb_cb(skb)->reason = reason;
3190 local_irq_save(flags);
3191 skb->next = __this_cpu_read(softnet_data.completion_queue);
3192 __this_cpu_write(softnet_data.completion_queue, skb);
3193 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3194 local_irq_restore(flags);
3195 }
3196 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3197
dev_kfree_skb_any_reason(struct sk_buff * skb,enum skb_drop_reason reason)3198 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3199 {
3200 if (in_hardirq() || irqs_disabled())
3201 dev_kfree_skb_irq_reason(skb, reason);
3202 else
3203 kfree_skb_reason(skb, reason);
3204 }
3205 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3206
3207
3208 /**
3209 * netif_device_detach - mark device as removed
3210 * @dev: network device
3211 *
3212 * Mark device as removed from system and therefore no longer available.
3213 */
netif_device_detach(struct net_device * dev)3214 void netif_device_detach(struct net_device *dev)
3215 {
3216 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3217 netif_running(dev)) {
3218 netif_tx_stop_all_queues(dev);
3219 }
3220 }
3221 EXPORT_SYMBOL(netif_device_detach);
3222
3223 /**
3224 * netif_device_attach - mark device as attached
3225 * @dev: network device
3226 *
3227 * Mark device as attached from system and restart if needed.
3228 */
netif_device_attach(struct net_device * dev)3229 void netif_device_attach(struct net_device *dev)
3230 {
3231 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3232 netif_running(dev)) {
3233 netif_tx_wake_all_queues(dev);
3234 __netdev_watchdog_up(dev);
3235 }
3236 }
3237 EXPORT_SYMBOL(netif_device_attach);
3238
3239 /*
3240 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3241 * to be used as a distribution range.
3242 */
skb_tx_hash(const struct net_device * dev,const struct net_device * sb_dev,struct sk_buff * skb)3243 static u16 skb_tx_hash(const struct net_device *dev,
3244 const struct net_device *sb_dev,
3245 struct sk_buff *skb)
3246 {
3247 u32 hash;
3248 u16 qoffset = 0;
3249 u16 qcount = dev->real_num_tx_queues;
3250
3251 if (dev->num_tc) {
3252 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3253
3254 qoffset = sb_dev->tc_to_txq[tc].offset;
3255 qcount = sb_dev->tc_to_txq[tc].count;
3256 if (unlikely(!qcount)) {
3257 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3258 sb_dev->name, qoffset, tc);
3259 qoffset = 0;
3260 qcount = dev->real_num_tx_queues;
3261 }
3262 }
3263
3264 if (skb_rx_queue_recorded(skb)) {
3265 DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3266 hash = skb_get_rx_queue(skb);
3267 if (hash >= qoffset)
3268 hash -= qoffset;
3269 while (unlikely(hash >= qcount))
3270 hash -= qcount;
3271 return hash + qoffset;
3272 }
3273
3274 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3275 }
3276
skb_warn_bad_offload(const struct sk_buff * skb)3277 void skb_warn_bad_offload(const struct sk_buff *skb)
3278 {
3279 static const netdev_features_t null_features;
3280 struct net_device *dev = skb->dev;
3281 const char *name = "";
3282
3283 if (!net_ratelimit())
3284 return;
3285
3286 if (dev) {
3287 if (dev->dev.parent)
3288 name = dev_driver_string(dev->dev.parent);
3289 else
3290 name = netdev_name(dev);
3291 }
3292 skb_dump(KERN_WARNING, skb, false);
3293 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3294 name, dev ? &dev->features : &null_features,
3295 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3296 }
3297
3298 /*
3299 * Invalidate hardware checksum when packet is to be mangled, and
3300 * complete checksum manually on outgoing path.
3301 */
skb_checksum_help(struct sk_buff * skb)3302 int skb_checksum_help(struct sk_buff *skb)
3303 {
3304 __wsum csum;
3305 int ret = 0, offset;
3306
3307 if (skb->ip_summed == CHECKSUM_COMPLETE)
3308 goto out_set_summed;
3309
3310 if (unlikely(skb_is_gso(skb))) {
3311 skb_warn_bad_offload(skb);
3312 return -EINVAL;
3313 }
3314
3315 if (!skb_frags_readable(skb)) {
3316 return -EFAULT;
3317 }
3318
3319 /* Before computing a checksum, we should make sure no frag could
3320 * be modified by an external entity : checksum could be wrong.
3321 */
3322 if (skb_has_shared_frag(skb)) {
3323 ret = __skb_linearize(skb);
3324 if (ret)
3325 goto out;
3326 }
3327
3328 offset = skb_checksum_start_offset(skb);
3329 ret = -EINVAL;
3330 if (unlikely(offset >= skb_headlen(skb))) {
3331 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3332 WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3333 offset, skb_headlen(skb));
3334 goto out;
3335 }
3336 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3337
3338 offset += skb->csum_offset;
3339 if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3340 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3341 WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3342 offset + sizeof(__sum16), skb_headlen(skb));
3343 goto out;
3344 }
3345 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3346 if (ret)
3347 goto out;
3348
3349 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3350 out_set_summed:
3351 skb->ip_summed = CHECKSUM_NONE;
3352 out:
3353 return ret;
3354 }
3355 EXPORT_SYMBOL(skb_checksum_help);
3356
skb_crc32c_csum_help(struct sk_buff * skb)3357 int skb_crc32c_csum_help(struct sk_buff *skb)
3358 {
3359 __le32 crc32c_csum;
3360 int ret = 0, offset, start;
3361
3362 if (skb->ip_summed != CHECKSUM_PARTIAL)
3363 goto out;
3364
3365 if (unlikely(skb_is_gso(skb)))
3366 goto out;
3367
3368 /* Before computing a checksum, we should make sure no frag could
3369 * be modified by an external entity : checksum could be wrong.
3370 */
3371 if (unlikely(skb_has_shared_frag(skb))) {
3372 ret = __skb_linearize(skb);
3373 if (ret)
3374 goto out;
3375 }
3376 start = skb_checksum_start_offset(skb);
3377 offset = start + offsetof(struct sctphdr, checksum);
3378 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3379 ret = -EINVAL;
3380 goto out;
3381 }
3382
3383 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3384 if (ret)
3385 goto out;
3386
3387 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3388 skb->len - start, ~(__u32)0,
3389 crc32c_csum_stub));
3390 *(__le32 *)(skb->data + offset) = crc32c_csum;
3391 skb_reset_csum_not_inet(skb);
3392 out:
3393 return ret;
3394 }
3395 EXPORT_SYMBOL(skb_crc32c_csum_help);
3396
skb_network_protocol(struct sk_buff * skb,int * depth)3397 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3398 {
3399 __be16 type = skb->protocol;
3400
3401 /* Tunnel gso handlers can set protocol to ethernet. */
3402 if (type == htons(ETH_P_TEB)) {
3403 struct ethhdr *eth;
3404
3405 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3406 return 0;
3407
3408 eth = (struct ethhdr *)skb->data;
3409 type = eth->h_proto;
3410 }
3411
3412 return vlan_get_protocol_and_depth(skb, type, depth);
3413 }
3414
3415
3416 /* Take action when hardware reception checksum errors are detected. */
3417 #ifdef CONFIG_BUG
do_netdev_rx_csum_fault(struct net_device * dev,struct sk_buff * skb)3418 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3419 {
3420 netdev_err(dev, "hw csum failure\n");
3421 skb_dump(KERN_ERR, skb, true);
3422 dump_stack();
3423 }
3424
netdev_rx_csum_fault(struct net_device * dev,struct sk_buff * skb)3425 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3426 {
3427 DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3428 }
3429 EXPORT_SYMBOL(netdev_rx_csum_fault);
3430 #endif
3431
3432 /* XXX: check that highmem exists at all on the given machine. */
illegal_highdma(struct net_device * dev,struct sk_buff * skb)3433 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3434 {
3435 #ifdef CONFIG_HIGHMEM
3436 int i;
3437
3438 if (!(dev->features & NETIF_F_HIGHDMA)) {
3439 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3440 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3441 struct page *page = skb_frag_page(frag);
3442
3443 if (page && PageHighMem(page))
3444 return 1;
3445 }
3446 }
3447 #endif
3448 return 0;
3449 }
3450
3451 /* If MPLS offload request, verify we are testing hardware MPLS features
3452 * instead of standard features for the netdev.
3453 */
3454 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
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 if (eth_p_mpls(type))
3460 features &= skb->dev->mpls_features;
3461
3462 return features;
3463 }
3464 #else
net_mpls_features(struct sk_buff * skb,netdev_features_t features,__be16 type)3465 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3466 netdev_features_t features,
3467 __be16 type)
3468 {
3469 return features;
3470 }
3471 #endif
3472
harmonize_features(struct sk_buff * skb,netdev_features_t features)3473 static netdev_features_t harmonize_features(struct sk_buff *skb,
3474 netdev_features_t features)
3475 {
3476 __be16 type;
3477
3478 type = skb_network_protocol(skb, NULL);
3479 features = net_mpls_features(skb, features, type);
3480
3481 if (skb->ip_summed != CHECKSUM_NONE &&
3482 !can_checksum_protocol(features, type)) {
3483 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3484 }
3485 if (illegal_highdma(skb->dev, skb))
3486 features &= ~NETIF_F_SG;
3487
3488 return features;
3489 }
3490
passthru_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3491 netdev_features_t passthru_features_check(struct sk_buff *skb,
3492 struct net_device *dev,
3493 netdev_features_t features)
3494 {
3495 return features;
3496 }
3497 EXPORT_SYMBOL(passthru_features_check);
3498
dflt_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3499 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3500 struct net_device *dev,
3501 netdev_features_t features)
3502 {
3503 return vlan_features_check(skb, features);
3504 }
3505
gso_features_check(const struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3506 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3507 struct net_device *dev,
3508 netdev_features_t features)
3509 {
3510 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3511
3512 if (gso_segs > READ_ONCE(dev->gso_max_segs))
3513 return features & ~NETIF_F_GSO_MASK;
3514
3515 if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size)))
3516 return features & ~NETIF_F_GSO_MASK;
3517
3518 if (!skb_shinfo(skb)->gso_type) {
3519 skb_warn_bad_offload(skb);
3520 return features & ~NETIF_F_GSO_MASK;
3521 }
3522
3523 /* Support for GSO partial features requires software
3524 * intervention before we can actually process the packets
3525 * so we need to strip support for any partial features now
3526 * and we can pull them back in after we have partially
3527 * segmented the frame.
3528 */
3529 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3530 features &= ~dev->gso_partial_features;
3531
3532 /* Make sure to clear the IPv4 ID mangling feature if the
3533 * IPv4 header has the potential to be fragmented.
3534 */
3535 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3536 struct iphdr *iph = skb->encapsulation ?
3537 inner_ip_hdr(skb) : ip_hdr(skb);
3538
3539 if (!(iph->frag_off & htons(IP_DF)))
3540 features &= ~NETIF_F_TSO_MANGLEID;
3541 }
3542
3543 return features;
3544 }
3545
netif_skb_features(struct sk_buff * skb)3546 netdev_features_t netif_skb_features(struct sk_buff *skb)
3547 {
3548 struct net_device *dev = skb->dev;
3549 netdev_features_t features = dev->features;
3550
3551 if (skb_is_gso(skb))
3552 features = gso_features_check(skb, dev, features);
3553
3554 /* If encapsulation offload request, verify we are testing
3555 * hardware encapsulation features instead of standard
3556 * features for the netdev
3557 */
3558 if (skb->encapsulation)
3559 features &= dev->hw_enc_features;
3560
3561 if (skb_vlan_tagged(skb))
3562 features = netdev_intersect_features(features,
3563 dev->vlan_features |
3564 NETIF_F_HW_VLAN_CTAG_TX |
3565 NETIF_F_HW_VLAN_STAG_TX);
3566
3567 if (dev->netdev_ops->ndo_features_check)
3568 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3569 features);
3570 else
3571 features &= dflt_features_check(skb, dev, features);
3572
3573 return harmonize_features(skb, features);
3574 }
3575 EXPORT_SYMBOL(netif_skb_features);
3576
xmit_one(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq,bool more)3577 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3578 struct netdev_queue *txq, bool more)
3579 {
3580 unsigned int len;
3581 int rc;
3582
3583 if (dev_nit_active(dev))
3584 dev_queue_xmit_nit(skb, dev);
3585
3586 len = skb->len;
3587 trace_net_dev_start_xmit(skb, dev);
3588 rc = netdev_start_xmit(skb, dev, txq, more);
3589 trace_net_dev_xmit(skb, rc, dev, len);
3590
3591 return rc;
3592 }
3593
dev_hard_start_xmit(struct sk_buff * first,struct net_device * dev,struct netdev_queue * txq,int * ret)3594 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3595 struct netdev_queue *txq, int *ret)
3596 {
3597 struct sk_buff *skb = first;
3598 int rc = NETDEV_TX_OK;
3599
3600 while (skb) {
3601 struct sk_buff *next = skb->next;
3602
3603 skb_mark_not_on_list(skb);
3604 rc = xmit_one(skb, dev, txq, next != NULL);
3605 if (unlikely(!dev_xmit_complete(rc))) {
3606 skb->next = next;
3607 goto out;
3608 }
3609
3610 skb = next;
3611 if (netif_tx_queue_stopped(txq) && skb) {
3612 rc = NETDEV_TX_BUSY;
3613 break;
3614 }
3615 }
3616
3617 out:
3618 *ret = rc;
3619 return skb;
3620 }
3621
validate_xmit_vlan(struct sk_buff * skb,netdev_features_t features)3622 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3623 netdev_features_t features)
3624 {
3625 if (skb_vlan_tag_present(skb) &&
3626 !vlan_hw_offload_capable(features, skb->vlan_proto))
3627 skb = __vlan_hwaccel_push_inside(skb);
3628 return skb;
3629 }
3630
skb_csum_hwoffload_help(struct sk_buff * skb,const netdev_features_t features)3631 int skb_csum_hwoffload_help(struct sk_buff *skb,
3632 const netdev_features_t features)
3633 {
3634 if (unlikely(skb_csum_is_sctp(skb)))
3635 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3636 skb_crc32c_csum_help(skb);
3637
3638 if (features & NETIF_F_HW_CSUM)
3639 return 0;
3640
3641 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3642 switch (skb->csum_offset) {
3643 case offsetof(struct tcphdr, check):
3644 case offsetof(struct udphdr, check):
3645 return 0;
3646 }
3647 }
3648
3649 return skb_checksum_help(skb);
3650 }
3651 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3652
validate_xmit_skb(struct sk_buff * skb,struct net_device * dev,bool * again)3653 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3654 {
3655 netdev_features_t features;
3656
3657 features = netif_skb_features(skb);
3658 skb = validate_xmit_vlan(skb, features);
3659 if (unlikely(!skb))
3660 goto out_null;
3661
3662 skb = sk_validate_xmit_skb(skb, dev);
3663 if (unlikely(!skb))
3664 goto out_null;
3665
3666 if (netif_needs_gso(skb, features)) {
3667 struct sk_buff *segs;
3668
3669 segs = skb_gso_segment(skb, features);
3670 if (IS_ERR(segs)) {
3671 goto out_kfree_skb;
3672 } else if (segs) {
3673 consume_skb(skb);
3674 skb = segs;
3675 }
3676 } else {
3677 if (skb_needs_linearize(skb, features) &&
3678 __skb_linearize(skb))
3679 goto out_kfree_skb;
3680
3681 /* If packet is not checksummed and device does not
3682 * support checksumming for this protocol, complete
3683 * checksumming here.
3684 */
3685 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3686 if (skb->encapsulation)
3687 skb_set_inner_transport_header(skb,
3688 skb_checksum_start_offset(skb));
3689 else
3690 skb_set_transport_header(skb,
3691 skb_checksum_start_offset(skb));
3692 if (skb_csum_hwoffload_help(skb, features))
3693 goto out_kfree_skb;
3694 }
3695 }
3696
3697 skb = validate_xmit_xfrm(skb, features, again);
3698
3699 return skb;
3700
3701 out_kfree_skb:
3702 kfree_skb(skb);
3703 out_null:
3704 dev_core_stats_tx_dropped_inc(dev);
3705 return NULL;
3706 }
3707
validate_xmit_skb_list(struct sk_buff * skb,struct net_device * dev,bool * again)3708 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3709 {
3710 struct sk_buff *next, *head = NULL, *tail;
3711
3712 for (; skb != NULL; skb = next) {
3713 next = skb->next;
3714 skb_mark_not_on_list(skb);
3715
3716 /* in case skb won't be segmented, point to itself */
3717 skb->prev = skb;
3718
3719 skb = validate_xmit_skb(skb, dev, again);
3720 if (!skb)
3721 continue;
3722
3723 if (!head)
3724 head = skb;
3725 else
3726 tail->next = skb;
3727 /* If skb was segmented, skb->prev points to
3728 * the last segment. If not, it still contains skb.
3729 */
3730 tail = skb->prev;
3731 }
3732 return head;
3733 }
3734 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3735
qdisc_pkt_len_init(struct sk_buff * skb)3736 static void qdisc_pkt_len_init(struct sk_buff *skb)
3737 {
3738 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3739
3740 qdisc_skb_cb(skb)->pkt_len = skb->len;
3741
3742 /* To get more precise estimation of bytes sent on wire,
3743 * we add to pkt_len the headers size of all segments
3744 */
3745 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3746 u16 gso_segs = shinfo->gso_segs;
3747 unsigned int hdr_len;
3748
3749 /* mac layer + network layer */
3750 hdr_len = skb_transport_offset(skb);
3751
3752 /* + transport layer */
3753 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3754 const struct tcphdr *th;
3755 struct tcphdr _tcphdr;
3756
3757 th = skb_header_pointer(skb, hdr_len,
3758 sizeof(_tcphdr), &_tcphdr);
3759 if (likely(th))
3760 hdr_len += __tcp_hdrlen(th);
3761 } else {
3762 struct udphdr _udphdr;
3763
3764 if (skb_header_pointer(skb, hdr_len,
3765 sizeof(_udphdr), &_udphdr))
3766 hdr_len += sizeof(struct udphdr);
3767 }
3768
3769 if (shinfo->gso_type & SKB_GSO_DODGY)
3770 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3771 shinfo->gso_size);
3772
3773 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3774 }
3775 }
3776
dev_qdisc_enqueue(struct sk_buff * skb,struct Qdisc * q,struct sk_buff ** to_free,struct netdev_queue * txq)3777 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3778 struct sk_buff **to_free,
3779 struct netdev_queue *txq)
3780 {
3781 int rc;
3782
3783 rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3784 if (rc == NET_XMIT_SUCCESS)
3785 trace_qdisc_enqueue(q, txq, skb);
3786 return rc;
3787 }
3788
__dev_xmit_skb(struct sk_buff * skb,struct Qdisc * q,struct net_device * dev,struct netdev_queue * txq)3789 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3790 struct net_device *dev,
3791 struct netdev_queue *txq)
3792 {
3793 spinlock_t *root_lock = qdisc_lock(q);
3794 struct sk_buff *to_free = NULL;
3795 bool contended;
3796 int rc;
3797
3798 qdisc_calculate_pkt_len(skb, q);
3799
3800 tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP);
3801
3802 if (q->flags & TCQ_F_NOLOCK) {
3803 if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3804 qdisc_run_begin(q)) {
3805 /* Retest nolock_qdisc_is_empty() within the protection
3806 * of q->seqlock to protect from racing with requeuing.
3807 */
3808 if (unlikely(!nolock_qdisc_is_empty(q))) {
3809 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3810 __qdisc_run(q);
3811 qdisc_run_end(q);
3812
3813 goto no_lock_out;
3814 }
3815
3816 qdisc_bstats_cpu_update(q, skb);
3817 if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3818 !nolock_qdisc_is_empty(q))
3819 __qdisc_run(q);
3820
3821 qdisc_run_end(q);
3822 return NET_XMIT_SUCCESS;
3823 }
3824
3825 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3826 qdisc_run(q);
3827
3828 no_lock_out:
3829 if (unlikely(to_free))
3830 kfree_skb_list_reason(to_free,
3831 tcf_get_drop_reason(to_free));
3832 return rc;
3833 }
3834
3835 if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) {
3836 kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
3837 return NET_XMIT_DROP;
3838 }
3839 /*
3840 * Heuristic to force contended enqueues to serialize on a
3841 * separate lock before trying to get qdisc main lock.
3842 * This permits qdisc->running owner to get the lock more
3843 * often and dequeue packets faster.
3844 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3845 * and then other tasks will only enqueue packets. The packets will be
3846 * sent after the qdisc owner is scheduled again. To prevent this
3847 * scenario the task always serialize on the lock.
3848 */
3849 contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3850 if (unlikely(contended))
3851 spin_lock(&q->busylock);
3852
3853 spin_lock(root_lock);
3854 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3855 __qdisc_drop(skb, &to_free);
3856 rc = NET_XMIT_DROP;
3857 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3858 qdisc_run_begin(q)) {
3859 /*
3860 * This is a work-conserving queue; there are no old skbs
3861 * waiting to be sent out; and the qdisc is not running -
3862 * xmit the skb directly.
3863 */
3864
3865 qdisc_bstats_update(q, skb);
3866
3867 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3868 if (unlikely(contended)) {
3869 spin_unlock(&q->busylock);
3870 contended = false;
3871 }
3872 __qdisc_run(q);
3873 }
3874
3875 qdisc_run_end(q);
3876 rc = NET_XMIT_SUCCESS;
3877 } else {
3878 WRITE_ONCE(q->owner, smp_processor_id());
3879 rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3880 WRITE_ONCE(q->owner, -1);
3881 if (qdisc_run_begin(q)) {
3882 if (unlikely(contended)) {
3883 spin_unlock(&q->busylock);
3884 contended = false;
3885 }
3886 __qdisc_run(q);
3887 qdisc_run_end(q);
3888 }
3889 }
3890 spin_unlock(root_lock);
3891 if (unlikely(to_free))
3892 kfree_skb_list_reason(to_free,
3893 tcf_get_drop_reason(to_free));
3894 if (unlikely(contended))
3895 spin_unlock(&q->busylock);
3896 return rc;
3897 }
3898
3899 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
skb_update_prio(struct sk_buff * skb)3900 static void skb_update_prio(struct sk_buff *skb)
3901 {
3902 const struct netprio_map *map;
3903 const struct sock *sk;
3904 unsigned int prioidx;
3905
3906 if (skb->priority)
3907 return;
3908 map = rcu_dereference_bh(skb->dev->priomap);
3909 if (!map)
3910 return;
3911 sk = skb_to_full_sk(skb);
3912 if (!sk)
3913 return;
3914
3915 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3916
3917 if (prioidx < map->priomap_len)
3918 skb->priority = map->priomap[prioidx];
3919 }
3920 #else
3921 #define skb_update_prio(skb)
3922 #endif
3923
3924 /**
3925 * dev_loopback_xmit - loop back @skb
3926 * @net: network namespace this loopback is happening in
3927 * @sk: sk needed to be a netfilter okfn
3928 * @skb: buffer to transmit
3929 */
dev_loopback_xmit(struct net * net,struct sock * sk,struct sk_buff * skb)3930 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3931 {
3932 skb_reset_mac_header(skb);
3933 __skb_pull(skb, skb_network_offset(skb));
3934 skb->pkt_type = PACKET_LOOPBACK;
3935 if (skb->ip_summed == CHECKSUM_NONE)
3936 skb->ip_summed = CHECKSUM_UNNECESSARY;
3937 DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3938 skb_dst_force(skb);
3939 netif_rx(skb);
3940 return 0;
3941 }
3942 EXPORT_SYMBOL(dev_loopback_xmit);
3943
3944 #ifdef CONFIG_NET_EGRESS
3945 static struct netdev_queue *
netdev_tx_queue_mapping(struct net_device * dev,struct sk_buff * skb)3946 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3947 {
3948 int qm = skb_get_queue_mapping(skb);
3949
3950 return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3951 }
3952
3953 #ifndef CONFIG_PREEMPT_RT
netdev_xmit_txqueue_skipped(void)3954 static bool netdev_xmit_txqueue_skipped(void)
3955 {
3956 return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3957 }
3958
netdev_xmit_skip_txqueue(bool skip)3959 void netdev_xmit_skip_txqueue(bool skip)
3960 {
3961 __this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3962 }
3963 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3964
3965 #else
netdev_xmit_txqueue_skipped(void)3966 static bool netdev_xmit_txqueue_skipped(void)
3967 {
3968 return current->net_xmit.skip_txqueue;
3969 }
3970
netdev_xmit_skip_txqueue(bool skip)3971 void netdev_xmit_skip_txqueue(bool skip)
3972 {
3973 current->net_xmit.skip_txqueue = skip;
3974 }
3975 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3976 #endif
3977 #endif /* CONFIG_NET_EGRESS */
3978
3979 #ifdef CONFIG_NET_XGRESS
tc_run(struct tcx_entry * entry,struct sk_buff * skb,enum skb_drop_reason * drop_reason)3980 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb,
3981 enum skb_drop_reason *drop_reason)
3982 {
3983 int ret = TC_ACT_UNSPEC;
3984 #ifdef CONFIG_NET_CLS_ACT
3985 struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
3986 struct tcf_result res;
3987
3988 if (!miniq)
3989 return ret;
3990
3991 if (static_branch_unlikely(&tcf_bypass_check_needed_key)) {
3992 if (tcf_block_bypass_sw(miniq->block))
3993 return ret;
3994 }
3995
3996 tc_skb_cb(skb)->mru = 0;
3997 tc_skb_cb(skb)->post_ct = false;
3998 tcf_set_drop_reason(skb, *drop_reason);
3999
4000 mini_qdisc_bstats_cpu_update(miniq, skb);
4001 ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
4002 /* Only tcf related quirks below. */
4003 switch (ret) {
4004 case TC_ACT_SHOT:
4005 *drop_reason = tcf_get_drop_reason(skb);
4006 mini_qdisc_qstats_cpu_drop(miniq);
4007 break;
4008 case TC_ACT_OK:
4009 case TC_ACT_RECLASSIFY:
4010 skb->tc_index = TC_H_MIN(res.classid);
4011 break;
4012 }
4013 #endif /* CONFIG_NET_CLS_ACT */
4014 return ret;
4015 }
4016
4017 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
4018
tcx_inc(void)4019 void tcx_inc(void)
4020 {
4021 static_branch_inc(&tcx_needed_key);
4022 }
4023
tcx_dec(void)4024 void tcx_dec(void)
4025 {
4026 static_branch_dec(&tcx_needed_key);
4027 }
4028
4029 static __always_inline enum tcx_action_base
tcx_run(const struct bpf_mprog_entry * entry,struct sk_buff * skb,const bool needs_mac)4030 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
4031 const bool needs_mac)
4032 {
4033 const struct bpf_mprog_fp *fp;
4034 const struct bpf_prog *prog;
4035 int ret = TCX_NEXT;
4036
4037 if (needs_mac)
4038 __skb_push(skb, skb->mac_len);
4039 bpf_mprog_foreach_prog(entry, fp, prog) {
4040 bpf_compute_data_pointers(skb);
4041 ret = bpf_prog_run(prog, skb);
4042 if (ret != TCX_NEXT)
4043 break;
4044 }
4045 if (needs_mac)
4046 __skb_pull(skb, skb->mac_len);
4047 return tcx_action_code(skb, ret);
4048 }
4049
4050 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)4051 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4052 struct net_device *orig_dev, bool *another)
4053 {
4054 struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
4055 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
4056 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4057 int sch_ret;
4058
4059 if (!entry)
4060 return skb;
4061
4062 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4063 if (*pt_prev) {
4064 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4065 *pt_prev = NULL;
4066 }
4067
4068 qdisc_skb_cb(skb)->pkt_len = skb->len;
4069 tcx_set_ingress(skb, true);
4070
4071 if (static_branch_unlikely(&tcx_needed_key)) {
4072 sch_ret = tcx_run(entry, skb, true);
4073 if (sch_ret != TC_ACT_UNSPEC)
4074 goto ingress_verdict;
4075 }
4076 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4077 ingress_verdict:
4078 switch (sch_ret) {
4079 case TC_ACT_REDIRECT:
4080 /* skb_mac_header check was done by BPF, so we can safely
4081 * push the L2 header back before redirecting to another
4082 * netdev.
4083 */
4084 __skb_push(skb, skb->mac_len);
4085 if (skb_do_redirect(skb) == -EAGAIN) {
4086 __skb_pull(skb, skb->mac_len);
4087 *another = true;
4088 break;
4089 }
4090 *ret = NET_RX_SUCCESS;
4091 bpf_net_ctx_clear(bpf_net_ctx);
4092 return NULL;
4093 case TC_ACT_SHOT:
4094 kfree_skb_reason(skb, drop_reason);
4095 *ret = NET_RX_DROP;
4096 bpf_net_ctx_clear(bpf_net_ctx);
4097 return NULL;
4098 /* used by tc_run */
4099 case TC_ACT_STOLEN:
4100 case TC_ACT_QUEUED:
4101 case TC_ACT_TRAP:
4102 consume_skb(skb);
4103 fallthrough;
4104 case TC_ACT_CONSUMED:
4105 *ret = NET_RX_SUCCESS;
4106 bpf_net_ctx_clear(bpf_net_ctx);
4107 return NULL;
4108 }
4109 bpf_net_ctx_clear(bpf_net_ctx);
4110
4111 return skb;
4112 }
4113
4114 static __always_inline struct sk_buff *
sch_handle_egress(struct sk_buff * skb,int * ret,struct net_device * dev)4115 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4116 {
4117 struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4118 enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
4119 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4120 int sch_ret;
4121
4122 if (!entry)
4123 return skb;
4124
4125 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4126
4127 /* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4128 * already set by the caller.
4129 */
4130 if (static_branch_unlikely(&tcx_needed_key)) {
4131 sch_ret = tcx_run(entry, skb, false);
4132 if (sch_ret != TC_ACT_UNSPEC)
4133 goto egress_verdict;
4134 }
4135 sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4136 egress_verdict:
4137 switch (sch_ret) {
4138 case TC_ACT_REDIRECT:
4139 /* No need to push/pop skb's mac_header here on egress! */
4140 skb_do_redirect(skb);
4141 *ret = NET_XMIT_SUCCESS;
4142 bpf_net_ctx_clear(bpf_net_ctx);
4143 return NULL;
4144 case TC_ACT_SHOT:
4145 kfree_skb_reason(skb, drop_reason);
4146 *ret = NET_XMIT_DROP;
4147 bpf_net_ctx_clear(bpf_net_ctx);
4148 return NULL;
4149 /* used by tc_run */
4150 case TC_ACT_STOLEN:
4151 case TC_ACT_QUEUED:
4152 case TC_ACT_TRAP:
4153 consume_skb(skb);
4154 fallthrough;
4155 case TC_ACT_CONSUMED:
4156 *ret = NET_XMIT_SUCCESS;
4157 bpf_net_ctx_clear(bpf_net_ctx);
4158 return NULL;
4159 }
4160 bpf_net_ctx_clear(bpf_net_ctx);
4161
4162 return skb;
4163 }
4164 #else
4165 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)4166 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4167 struct net_device *orig_dev, bool *another)
4168 {
4169 return skb;
4170 }
4171
4172 static __always_inline struct sk_buff *
sch_handle_egress(struct sk_buff * skb,int * ret,struct net_device * dev)4173 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4174 {
4175 return skb;
4176 }
4177 #endif /* CONFIG_NET_XGRESS */
4178
4179 #ifdef CONFIG_XPS
__get_xps_queue_idx(struct net_device * dev,struct sk_buff * skb,struct xps_dev_maps * dev_maps,unsigned int tci)4180 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4181 struct xps_dev_maps *dev_maps, unsigned int tci)
4182 {
4183 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4184 struct xps_map *map;
4185 int queue_index = -1;
4186
4187 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4188 return queue_index;
4189
4190 tci *= dev_maps->num_tc;
4191 tci += tc;
4192
4193 map = rcu_dereference(dev_maps->attr_map[tci]);
4194 if (map) {
4195 if (map->len == 1)
4196 queue_index = map->queues[0];
4197 else
4198 queue_index = map->queues[reciprocal_scale(
4199 skb_get_hash(skb), map->len)];
4200 if (unlikely(queue_index >= dev->real_num_tx_queues))
4201 queue_index = -1;
4202 }
4203 return queue_index;
4204 }
4205 #endif
4206
get_xps_queue(struct net_device * dev,struct net_device * sb_dev,struct sk_buff * skb)4207 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4208 struct sk_buff *skb)
4209 {
4210 #ifdef CONFIG_XPS
4211 struct xps_dev_maps *dev_maps;
4212 struct sock *sk = skb->sk;
4213 int queue_index = -1;
4214
4215 if (!static_key_false(&xps_needed))
4216 return -1;
4217
4218 rcu_read_lock();
4219 if (!static_key_false(&xps_rxqs_needed))
4220 goto get_cpus_map;
4221
4222 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4223 if (dev_maps) {
4224 int tci = sk_rx_queue_get(sk);
4225
4226 if (tci >= 0)
4227 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4228 tci);
4229 }
4230
4231 get_cpus_map:
4232 if (queue_index < 0) {
4233 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4234 if (dev_maps) {
4235 unsigned int tci = skb->sender_cpu - 1;
4236
4237 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4238 tci);
4239 }
4240 }
4241 rcu_read_unlock();
4242
4243 return queue_index;
4244 #else
4245 return -1;
4246 #endif
4247 }
4248
dev_pick_tx_zero(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4249 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4250 struct net_device *sb_dev)
4251 {
4252 return 0;
4253 }
4254 EXPORT_SYMBOL(dev_pick_tx_zero);
4255
netdev_pick_tx(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4256 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4257 struct net_device *sb_dev)
4258 {
4259 struct sock *sk = skb->sk;
4260 int queue_index = sk_tx_queue_get(sk);
4261
4262 sb_dev = sb_dev ? : dev;
4263
4264 if (queue_index < 0 || skb->ooo_okay ||
4265 queue_index >= dev->real_num_tx_queues) {
4266 int new_index = get_xps_queue(dev, sb_dev, skb);
4267
4268 if (new_index < 0)
4269 new_index = skb_tx_hash(dev, sb_dev, skb);
4270
4271 if (queue_index != new_index && sk &&
4272 sk_fullsock(sk) &&
4273 rcu_access_pointer(sk->sk_dst_cache))
4274 sk_tx_queue_set(sk, new_index);
4275
4276 queue_index = new_index;
4277 }
4278
4279 return queue_index;
4280 }
4281 EXPORT_SYMBOL(netdev_pick_tx);
4282
netdev_core_pick_tx(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)4283 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4284 struct sk_buff *skb,
4285 struct net_device *sb_dev)
4286 {
4287 int queue_index = 0;
4288
4289 #ifdef CONFIG_XPS
4290 u32 sender_cpu = skb->sender_cpu - 1;
4291
4292 if (sender_cpu >= (u32)NR_CPUS)
4293 skb->sender_cpu = raw_smp_processor_id() + 1;
4294 #endif
4295
4296 if (dev->real_num_tx_queues != 1) {
4297 const struct net_device_ops *ops = dev->netdev_ops;
4298
4299 if (ops->ndo_select_queue)
4300 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4301 else
4302 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4303
4304 queue_index = netdev_cap_txqueue(dev, queue_index);
4305 }
4306
4307 skb_set_queue_mapping(skb, queue_index);
4308 return netdev_get_tx_queue(dev, queue_index);
4309 }
4310
4311 /**
4312 * __dev_queue_xmit() - transmit a buffer
4313 * @skb: buffer to transmit
4314 * @sb_dev: suboordinate device used for L2 forwarding offload
4315 *
4316 * Queue a buffer for transmission to a network device. The caller must
4317 * have set the device and priority and built the buffer before calling
4318 * this function. The function can be called from an interrupt.
4319 *
4320 * When calling this method, interrupts MUST be enabled. This is because
4321 * the BH enable code must have IRQs enabled so that it will not deadlock.
4322 *
4323 * Regardless of the return value, the skb is consumed, so it is currently
4324 * difficult to retry a send to this method. (You can bump the ref count
4325 * before sending to hold a reference for retry if you are careful.)
4326 *
4327 * Return:
4328 * * 0 - buffer successfully transmitted
4329 * * positive qdisc return code - NET_XMIT_DROP etc.
4330 * * negative errno - other errors
4331 */
__dev_queue_xmit(struct sk_buff * skb,struct net_device * sb_dev)4332 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4333 {
4334 struct net_device *dev = skb->dev;
4335 struct netdev_queue *txq = NULL;
4336 struct Qdisc *q;
4337 int rc = -ENOMEM;
4338 bool again = false;
4339
4340 skb_reset_mac_header(skb);
4341 skb_assert_len(skb);
4342
4343 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4344 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4345
4346 /* Disable soft irqs for various locks below. Also
4347 * stops preemption for RCU.
4348 */
4349 rcu_read_lock_bh();
4350
4351 skb_update_prio(skb);
4352
4353 qdisc_pkt_len_init(skb);
4354 tcx_set_ingress(skb, false);
4355 #ifdef CONFIG_NET_EGRESS
4356 if (static_branch_unlikely(&egress_needed_key)) {
4357 if (nf_hook_egress_active()) {
4358 skb = nf_hook_egress(skb, &rc, dev);
4359 if (!skb)
4360 goto out;
4361 }
4362
4363 netdev_xmit_skip_txqueue(false);
4364
4365 nf_skip_egress(skb, true);
4366 skb = sch_handle_egress(skb, &rc, dev);
4367 if (!skb)
4368 goto out;
4369 nf_skip_egress(skb, false);
4370
4371 if (netdev_xmit_txqueue_skipped())
4372 txq = netdev_tx_queue_mapping(dev, skb);
4373 }
4374 #endif
4375 /* If device/qdisc don't need skb->dst, release it right now while
4376 * its hot in this cpu cache.
4377 */
4378 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4379 skb_dst_drop(skb);
4380 else
4381 skb_dst_force(skb);
4382
4383 if (!txq)
4384 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4385
4386 q = rcu_dereference_bh(txq->qdisc);
4387
4388 trace_net_dev_queue(skb);
4389 if (q->enqueue) {
4390 rc = __dev_xmit_skb(skb, q, dev, txq);
4391 goto out;
4392 }
4393
4394 /* The device has no queue. Common case for software devices:
4395 * loopback, all the sorts of tunnels...
4396
4397 * Really, it is unlikely that netif_tx_lock protection is necessary
4398 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4399 * counters.)
4400 * However, it is possible, that they rely on protection
4401 * made by us here.
4402
4403 * Check this and shot the lock. It is not prone from deadlocks.
4404 *Either shot noqueue qdisc, it is even simpler 8)
4405 */
4406 if (dev->flags & IFF_UP) {
4407 int cpu = smp_processor_id(); /* ok because BHs are off */
4408
4409 /* Other cpus might concurrently change txq->xmit_lock_owner
4410 * to -1 or to their cpu id, but not to our id.
4411 */
4412 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4413 if (dev_xmit_recursion())
4414 goto recursion_alert;
4415
4416 skb = validate_xmit_skb(skb, dev, &again);
4417 if (!skb)
4418 goto out;
4419
4420 HARD_TX_LOCK(dev, txq, cpu);
4421
4422 if (!netif_xmit_stopped(txq)) {
4423 dev_xmit_recursion_inc();
4424 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4425 dev_xmit_recursion_dec();
4426 if (dev_xmit_complete(rc)) {
4427 HARD_TX_UNLOCK(dev, txq);
4428 goto out;
4429 }
4430 }
4431 HARD_TX_UNLOCK(dev, txq);
4432 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4433 dev->name);
4434 } else {
4435 /* Recursion is detected! It is possible,
4436 * unfortunately
4437 */
4438 recursion_alert:
4439 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4440 dev->name);
4441 }
4442 }
4443
4444 rc = -ENETDOWN;
4445 rcu_read_unlock_bh();
4446
4447 dev_core_stats_tx_dropped_inc(dev);
4448 kfree_skb_list(skb);
4449 return rc;
4450 out:
4451 rcu_read_unlock_bh();
4452 return rc;
4453 }
4454 EXPORT_SYMBOL(__dev_queue_xmit);
4455
__dev_direct_xmit(struct sk_buff * skb,u16 queue_id)4456 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4457 {
4458 struct net_device *dev = skb->dev;
4459 struct sk_buff *orig_skb = skb;
4460 struct netdev_queue *txq;
4461 int ret = NETDEV_TX_BUSY;
4462 bool again = false;
4463
4464 if (unlikely(!netif_running(dev) ||
4465 !netif_carrier_ok(dev)))
4466 goto drop;
4467
4468 skb = validate_xmit_skb_list(skb, dev, &again);
4469 if (skb != orig_skb)
4470 goto drop;
4471
4472 skb_set_queue_mapping(skb, queue_id);
4473 txq = skb_get_tx_queue(dev, skb);
4474
4475 local_bh_disable();
4476
4477 dev_xmit_recursion_inc();
4478 HARD_TX_LOCK(dev, txq, smp_processor_id());
4479 if (!netif_xmit_frozen_or_drv_stopped(txq))
4480 ret = netdev_start_xmit(skb, dev, txq, false);
4481 HARD_TX_UNLOCK(dev, txq);
4482 dev_xmit_recursion_dec();
4483
4484 local_bh_enable();
4485 return ret;
4486 drop:
4487 dev_core_stats_tx_dropped_inc(dev);
4488 kfree_skb_list(skb);
4489 return NET_XMIT_DROP;
4490 }
4491 EXPORT_SYMBOL(__dev_direct_xmit);
4492
4493 /*************************************************************************
4494 * Receiver routines
4495 *************************************************************************/
4496 static DEFINE_PER_CPU(struct task_struct *, backlog_napi);
4497
4498 int weight_p __read_mostly = 64; /* old backlog weight */
4499 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4500 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4501
4502 /* Called with irq disabled */
____napi_schedule(struct softnet_data * sd,struct napi_struct * napi)4503 static inline void ____napi_schedule(struct softnet_data *sd,
4504 struct napi_struct *napi)
4505 {
4506 struct task_struct *thread;
4507
4508 lockdep_assert_irqs_disabled();
4509
4510 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4511 /* Paired with smp_mb__before_atomic() in
4512 * napi_enable()/dev_set_threaded().
4513 * Use READ_ONCE() to guarantee a complete
4514 * read on napi->thread. Only call
4515 * wake_up_process() when it's not NULL.
4516 */
4517 thread = READ_ONCE(napi->thread);
4518 if (thread) {
4519 if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi))
4520 goto use_local_napi;
4521
4522 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4523 wake_up_process(thread);
4524 return;
4525 }
4526 }
4527
4528 use_local_napi:
4529 list_add_tail(&napi->poll_list, &sd->poll_list);
4530 WRITE_ONCE(napi->list_owner, smp_processor_id());
4531 /* If not called from net_rx_action()
4532 * we have to raise NET_RX_SOFTIRQ.
4533 */
4534 if (!sd->in_net_rx_action)
4535 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4536 }
4537
4538 #ifdef CONFIG_RPS
4539
4540 struct static_key_false rps_needed __read_mostly;
4541 EXPORT_SYMBOL(rps_needed);
4542 struct static_key_false rfs_needed __read_mostly;
4543 EXPORT_SYMBOL(rfs_needed);
4544
4545 static struct rps_dev_flow *
set_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow * rflow,u16 next_cpu)4546 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4547 struct rps_dev_flow *rflow, u16 next_cpu)
4548 {
4549 if (next_cpu < nr_cpu_ids) {
4550 u32 head;
4551 #ifdef CONFIG_RFS_ACCEL
4552 struct netdev_rx_queue *rxqueue;
4553 struct rps_dev_flow_table *flow_table;
4554 struct rps_dev_flow *old_rflow;
4555 u16 rxq_index;
4556 u32 flow_id;
4557 int rc;
4558
4559 /* Should we steer this flow to a different hardware queue? */
4560 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4561 !(dev->features & NETIF_F_NTUPLE))
4562 goto out;
4563 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4564 if (rxq_index == skb_get_rx_queue(skb))
4565 goto out;
4566
4567 rxqueue = dev->_rx + rxq_index;
4568 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4569 if (!flow_table)
4570 goto out;
4571 flow_id = skb_get_hash(skb) & flow_table->mask;
4572 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4573 rxq_index, flow_id);
4574 if (rc < 0)
4575 goto out;
4576 old_rflow = rflow;
4577 rflow = &flow_table->flows[flow_id];
4578 WRITE_ONCE(rflow->filter, rc);
4579 if (old_rflow->filter == rc)
4580 WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
4581 out:
4582 #endif
4583 head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
4584 rps_input_queue_tail_save(&rflow->last_qtail, head);
4585 }
4586
4587 WRITE_ONCE(rflow->cpu, next_cpu);
4588 return rflow;
4589 }
4590
4591 /*
4592 * get_rps_cpu is called from netif_receive_skb and returns the target
4593 * CPU from the RPS map of the receiving queue for a given skb.
4594 * rcu_read_lock must be held on entry.
4595 */
get_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow ** rflowp)4596 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4597 struct rps_dev_flow **rflowp)
4598 {
4599 const struct rps_sock_flow_table *sock_flow_table;
4600 struct netdev_rx_queue *rxqueue = dev->_rx;
4601 struct rps_dev_flow_table *flow_table;
4602 struct rps_map *map;
4603 int cpu = -1;
4604 u32 tcpu;
4605 u32 hash;
4606
4607 if (skb_rx_queue_recorded(skb)) {
4608 u16 index = skb_get_rx_queue(skb);
4609
4610 if (unlikely(index >= dev->real_num_rx_queues)) {
4611 WARN_ONCE(dev->real_num_rx_queues > 1,
4612 "%s received packet on queue %u, but number "
4613 "of RX queues is %u\n",
4614 dev->name, index, dev->real_num_rx_queues);
4615 goto done;
4616 }
4617 rxqueue += index;
4618 }
4619
4620 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4621
4622 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4623 map = rcu_dereference(rxqueue->rps_map);
4624 if (!flow_table && !map)
4625 goto done;
4626
4627 skb_reset_network_header(skb);
4628 hash = skb_get_hash(skb);
4629 if (!hash)
4630 goto done;
4631
4632 sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
4633 if (flow_table && sock_flow_table) {
4634 struct rps_dev_flow *rflow;
4635 u32 next_cpu;
4636 u32 ident;
4637
4638 /* First check into global flow table if there is a match.
4639 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4640 */
4641 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4642 if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
4643 goto try_rps;
4644
4645 next_cpu = ident & net_hotdata.rps_cpu_mask;
4646
4647 /* OK, now we know there is a match,
4648 * we can look at the local (per receive queue) flow table
4649 */
4650 rflow = &flow_table->flows[hash & flow_table->mask];
4651 tcpu = rflow->cpu;
4652
4653 /*
4654 * If the desired CPU (where last recvmsg was done) is
4655 * different from current CPU (one in the rx-queue flow
4656 * table entry), switch if one of the following holds:
4657 * - Current CPU is unset (>= nr_cpu_ids).
4658 * - Current CPU is offline.
4659 * - The current CPU's queue tail has advanced beyond the
4660 * last packet that was enqueued using this table entry.
4661 * This guarantees that all previous packets for the flow
4662 * have been dequeued, thus preserving in order delivery.
4663 */
4664 if (unlikely(tcpu != next_cpu) &&
4665 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4666 ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
4667 rflow->last_qtail)) >= 0)) {
4668 tcpu = next_cpu;
4669 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4670 }
4671
4672 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4673 *rflowp = rflow;
4674 cpu = tcpu;
4675 goto done;
4676 }
4677 }
4678
4679 try_rps:
4680
4681 if (map) {
4682 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4683 if (cpu_online(tcpu)) {
4684 cpu = tcpu;
4685 goto done;
4686 }
4687 }
4688
4689 done:
4690 return cpu;
4691 }
4692
4693 #ifdef CONFIG_RFS_ACCEL
4694
4695 /**
4696 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4697 * @dev: Device on which the filter was set
4698 * @rxq_index: RX queue index
4699 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4700 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4701 *
4702 * Drivers that implement ndo_rx_flow_steer() should periodically call
4703 * this function for each installed filter and remove the filters for
4704 * which it returns %true.
4705 */
rps_may_expire_flow(struct net_device * dev,u16 rxq_index,u32 flow_id,u16 filter_id)4706 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4707 u32 flow_id, u16 filter_id)
4708 {
4709 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4710 struct rps_dev_flow_table *flow_table;
4711 struct rps_dev_flow *rflow;
4712 bool expire = true;
4713 unsigned int cpu;
4714
4715 rcu_read_lock();
4716 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4717 if (flow_table && flow_id <= flow_table->mask) {
4718 rflow = &flow_table->flows[flow_id];
4719 cpu = READ_ONCE(rflow->cpu);
4720 if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids &&
4721 ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) -
4722 READ_ONCE(rflow->last_qtail)) <
4723 (int)(10 * flow_table->mask)))
4724 expire = false;
4725 }
4726 rcu_read_unlock();
4727 return expire;
4728 }
4729 EXPORT_SYMBOL(rps_may_expire_flow);
4730
4731 #endif /* CONFIG_RFS_ACCEL */
4732
4733 /* Called from hardirq (IPI) context */
rps_trigger_softirq(void * data)4734 static void rps_trigger_softirq(void *data)
4735 {
4736 struct softnet_data *sd = data;
4737
4738 ____napi_schedule(sd, &sd->backlog);
4739 sd->received_rps++;
4740 }
4741
4742 #endif /* CONFIG_RPS */
4743
4744 /* Called from hardirq (IPI) context */
trigger_rx_softirq(void * data)4745 static void trigger_rx_softirq(void *data)
4746 {
4747 struct softnet_data *sd = data;
4748
4749 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4750 smp_store_release(&sd->defer_ipi_scheduled, 0);
4751 }
4752
4753 /*
4754 * After we queued a packet into sd->input_pkt_queue,
4755 * we need to make sure this queue is serviced soon.
4756 *
4757 * - If this is another cpu queue, link it to our rps_ipi_list,
4758 * and make sure we will process rps_ipi_list from net_rx_action().
4759 *
4760 * - If this is our own queue, NAPI schedule our backlog.
4761 * Note that this also raises NET_RX_SOFTIRQ.
4762 */
napi_schedule_rps(struct softnet_data * sd)4763 static void napi_schedule_rps(struct softnet_data *sd)
4764 {
4765 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4766
4767 #ifdef CONFIG_RPS
4768 if (sd != mysd) {
4769 if (use_backlog_threads()) {
4770 __napi_schedule_irqoff(&sd->backlog);
4771 return;
4772 }
4773
4774 sd->rps_ipi_next = mysd->rps_ipi_list;
4775 mysd->rps_ipi_list = sd;
4776
4777 /* If not called from net_rx_action() or napi_threaded_poll()
4778 * we have to raise NET_RX_SOFTIRQ.
4779 */
4780 if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4781 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4782 return;
4783 }
4784 #endif /* CONFIG_RPS */
4785 __napi_schedule_irqoff(&mysd->backlog);
4786 }
4787
kick_defer_list_purge(struct softnet_data * sd,unsigned int cpu)4788 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu)
4789 {
4790 unsigned long flags;
4791
4792 if (use_backlog_threads()) {
4793 backlog_lock_irq_save(sd, &flags);
4794
4795 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4796 __napi_schedule_irqoff(&sd->backlog);
4797
4798 backlog_unlock_irq_restore(sd, &flags);
4799
4800 } else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) {
4801 smp_call_function_single_async(cpu, &sd->defer_csd);
4802 }
4803 }
4804
4805 #ifdef CONFIG_NET_FLOW_LIMIT
4806 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4807 #endif
4808
skb_flow_limit(struct sk_buff * skb,unsigned int qlen)4809 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4810 {
4811 #ifdef CONFIG_NET_FLOW_LIMIT
4812 struct sd_flow_limit *fl;
4813 struct softnet_data *sd;
4814 unsigned int old_flow, new_flow;
4815
4816 if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
4817 return false;
4818
4819 sd = this_cpu_ptr(&softnet_data);
4820
4821 rcu_read_lock();
4822 fl = rcu_dereference(sd->flow_limit);
4823 if (fl) {
4824 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4825 old_flow = fl->history[fl->history_head];
4826 fl->history[fl->history_head] = new_flow;
4827
4828 fl->history_head++;
4829 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4830
4831 if (likely(fl->buckets[old_flow]))
4832 fl->buckets[old_flow]--;
4833
4834 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4835 fl->count++;
4836 rcu_read_unlock();
4837 return true;
4838 }
4839 }
4840 rcu_read_unlock();
4841 #endif
4842 return false;
4843 }
4844
4845 /*
4846 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4847 * queue (may be a remote CPU queue).
4848 */
enqueue_to_backlog(struct sk_buff * skb,int cpu,unsigned int * qtail)4849 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4850 unsigned int *qtail)
4851 {
4852 enum skb_drop_reason reason;
4853 struct softnet_data *sd;
4854 unsigned long flags;
4855 unsigned int qlen;
4856 int max_backlog;
4857 u32 tail;
4858
4859 reason = SKB_DROP_REASON_DEV_READY;
4860 if (!netif_running(skb->dev))
4861 goto bad_dev;
4862
4863 reason = SKB_DROP_REASON_CPU_BACKLOG;
4864 sd = &per_cpu(softnet_data, cpu);
4865
4866 qlen = skb_queue_len_lockless(&sd->input_pkt_queue);
4867 max_backlog = READ_ONCE(net_hotdata.max_backlog);
4868 if (unlikely(qlen > max_backlog))
4869 goto cpu_backlog_drop;
4870 backlog_lock_irq_save(sd, &flags);
4871 qlen = skb_queue_len(&sd->input_pkt_queue);
4872 if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) {
4873 if (!qlen) {
4874 /* Schedule NAPI for backlog device. We can use
4875 * non atomic operation as we own the queue lock.
4876 */
4877 if (!__test_and_set_bit(NAPI_STATE_SCHED,
4878 &sd->backlog.state))
4879 napi_schedule_rps(sd);
4880 }
4881 __skb_queue_tail(&sd->input_pkt_queue, skb);
4882 tail = rps_input_queue_tail_incr(sd);
4883 backlog_unlock_irq_restore(sd, &flags);
4884
4885 /* save the tail outside of the critical section */
4886 rps_input_queue_tail_save(qtail, tail);
4887 return NET_RX_SUCCESS;
4888 }
4889
4890 backlog_unlock_irq_restore(sd, &flags);
4891
4892 cpu_backlog_drop:
4893 atomic_inc(&sd->dropped);
4894 bad_dev:
4895 dev_core_stats_rx_dropped_inc(skb->dev);
4896 kfree_skb_reason(skb, reason);
4897 return NET_RX_DROP;
4898 }
4899
netif_get_rxqueue(struct sk_buff * skb)4900 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4901 {
4902 struct net_device *dev = skb->dev;
4903 struct netdev_rx_queue *rxqueue;
4904
4905 rxqueue = dev->_rx;
4906
4907 if (skb_rx_queue_recorded(skb)) {
4908 u16 index = skb_get_rx_queue(skb);
4909
4910 if (unlikely(index >= dev->real_num_rx_queues)) {
4911 WARN_ONCE(dev->real_num_rx_queues > 1,
4912 "%s received packet on queue %u, but number "
4913 "of RX queues is %u\n",
4914 dev->name, index, dev->real_num_rx_queues);
4915
4916 return rxqueue; /* Return first rxqueue */
4917 }
4918 rxqueue += index;
4919 }
4920 return rxqueue;
4921 }
4922
bpf_prog_run_generic_xdp(struct sk_buff * skb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)4923 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4924 struct bpf_prog *xdp_prog)
4925 {
4926 void *orig_data, *orig_data_end, *hard_start;
4927 struct netdev_rx_queue *rxqueue;
4928 bool orig_bcast, orig_host;
4929 u32 mac_len, frame_sz;
4930 __be16 orig_eth_type;
4931 struct ethhdr *eth;
4932 u32 metalen, act;
4933 int off;
4934
4935 /* The XDP program wants to see the packet starting at the MAC
4936 * header.
4937 */
4938 mac_len = skb->data - skb_mac_header(skb);
4939 hard_start = skb->data - skb_headroom(skb);
4940
4941 /* SKB "head" area always have tailroom for skb_shared_info */
4942 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4943 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4944
4945 rxqueue = netif_get_rxqueue(skb);
4946 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4947 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4948 skb_headlen(skb) + mac_len, true);
4949 if (skb_is_nonlinear(skb)) {
4950 skb_shinfo(skb)->xdp_frags_size = skb->data_len;
4951 xdp_buff_set_frags_flag(xdp);
4952 } else {
4953 xdp_buff_clear_frags_flag(xdp);
4954 }
4955
4956 orig_data_end = xdp->data_end;
4957 orig_data = xdp->data;
4958 eth = (struct ethhdr *)xdp->data;
4959 orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4960 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4961 orig_eth_type = eth->h_proto;
4962
4963 act = bpf_prog_run_xdp(xdp_prog, xdp);
4964
4965 /* check if bpf_xdp_adjust_head was used */
4966 off = xdp->data - orig_data;
4967 if (off) {
4968 if (off > 0)
4969 __skb_pull(skb, off);
4970 else if (off < 0)
4971 __skb_push(skb, -off);
4972
4973 skb->mac_header += off;
4974 skb_reset_network_header(skb);
4975 }
4976
4977 /* check if bpf_xdp_adjust_tail was used */
4978 off = xdp->data_end - orig_data_end;
4979 if (off != 0) {
4980 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4981 skb->len += off; /* positive on grow, negative on shrink */
4982 }
4983
4984 /* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
4985 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
4986 */
4987 if (xdp_buff_has_frags(xdp))
4988 skb->data_len = skb_shinfo(skb)->xdp_frags_size;
4989 else
4990 skb->data_len = 0;
4991
4992 /* check if XDP changed eth hdr such SKB needs update */
4993 eth = (struct ethhdr *)xdp->data;
4994 if ((orig_eth_type != eth->h_proto) ||
4995 (orig_host != ether_addr_equal_64bits(eth->h_dest,
4996 skb->dev->dev_addr)) ||
4997 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4998 __skb_push(skb, ETH_HLEN);
4999 skb->pkt_type = PACKET_HOST;
5000 skb->protocol = eth_type_trans(skb, skb->dev);
5001 }
5002
5003 /* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
5004 * before calling us again on redirect path. We do not call do_redirect
5005 * as we leave that up to the caller.
5006 *
5007 * Caller is responsible for managing lifetime of skb (i.e. calling
5008 * kfree_skb in response to actions it cannot handle/XDP_DROP).
5009 */
5010 switch (act) {
5011 case XDP_REDIRECT:
5012 case XDP_TX:
5013 __skb_push(skb, mac_len);
5014 break;
5015 case XDP_PASS:
5016 metalen = xdp->data - xdp->data_meta;
5017 if (metalen)
5018 skb_metadata_set(skb, metalen);
5019 break;
5020 }
5021
5022 return act;
5023 }
5024
5025 static int
netif_skb_check_for_xdp(struct sk_buff ** pskb,struct bpf_prog * prog)5026 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog)
5027 {
5028 struct sk_buff *skb = *pskb;
5029 int err, hroom, troom;
5030
5031 if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog))
5032 return 0;
5033
5034 /* In case we have to go down the path and also linearize,
5035 * then lets do the pskb_expand_head() work just once here.
5036 */
5037 hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
5038 troom = skb->tail + skb->data_len - skb->end;
5039 err = pskb_expand_head(skb,
5040 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
5041 troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
5042 if (err)
5043 return err;
5044
5045 return skb_linearize(skb);
5046 }
5047
netif_receive_generic_xdp(struct sk_buff ** pskb,struct xdp_buff * xdp,struct bpf_prog * xdp_prog)5048 static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
5049 struct xdp_buff *xdp,
5050 struct bpf_prog *xdp_prog)
5051 {
5052 struct sk_buff *skb = *pskb;
5053 u32 mac_len, act = XDP_DROP;
5054
5055 /* Reinjected packets coming from act_mirred or similar should
5056 * not get XDP generic processing.
5057 */
5058 if (skb_is_redirected(skb))
5059 return XDP_PASS;
5060
5061 /* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
5062 * bytes. This is the guarantee that also native XDP provides,
5063 * thus we need to do it here as well.
5064 */
5065 mac_len = skb->data - skb_mac_header(skb);
5066 __skb_push(skb, mac_len);
5067
5068 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
5069 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
5070 if (netif_skb_check_for_xdp(pskb, xdp_prog))
5071 goto do_drop;
5072 }
5073
5074 __skb_pull(*pskb, mac_len);
5075
5076 act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
5077 switch (act) {
5078 case XDP_REDIRECT:
5079 case XDP_TX:
5080 case XDP_PASS:
5081 break;
5082 default:
5083 bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
5084 fallthrough;
5085 case XDP_ABORTED:
5086 trace_xdp_exception((*pskb)->dev, xdp_prog, act);
5087 fallthrough;
5088 case XDP_DROP:
5089 do_drop:
5090 kfree_skb(*pskb);
5091 break;
5092 }
5093
5094 return act;
5095 }
5096
5097 /* When doing generic XDP we have to bypass the qdisc layer and the
5098 * network taps in order to match in-driver-XDP behavior. This also means
5099 * that XDP packets are able to starve other packets going through a qdisc,
5100 * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
5101 * queues, so they do not have this starvation issue.
5102 */
generic_xdp_tx(struct sk_buff * skb,struct bpf_prog * xdp_prog)5103 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
5104 {
5105 struct net_device *dev = skb->dev;
5106 struct netdev_queue *txq;
5107 bool free_skb = true;
5108 int cpu, rc;
5109
5110 txq = netdev_core_pick_tx(dev, skb, NULL);
5111 cpu = smp_processor_id();
5112 HARD_TX_LOCK(dev, txq, cpu);
5113 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
5114 rc = netdev_start_xmit(skb, dev, txq, 0);
5115 if (dev_xmit_complete(rc))
5116 free_skb = false;
5117 }
5118 HARD_TX_UNLOCK(dev, txq);
5119 if (free_skb) {
5120 trace_xdp_exception(dev, xdp_prog, XDP_TX);
5121 dev_core_stats_tx_dropped_inc(dev);
5122 kfree_skb(skb);
5123 }
5124 }
5125
5126 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
5127
do_xdp_generic(struct bpf_prog * xdp_prog,struct sk_buff ** pskb)5128 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb)
5129 {
5130 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
5131
5132 if (xdp_prog) {
5133 struct xdp_buff xdp;
5134 u32 act;
5135 int err;
5136
5137 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
5138 act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
5139 if (act != XDP_PASS) {
5140 switch (act) {
5141 case XDP_REDIRECT:
5142 err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
5143 &xdp, xdp_prog);
5144 if (err)
5145 goto out_redir;
5146 break;
5147 case XDP_TX:
5148 generic_xdp_tx(*pskb, xdp_prog);
5149 break;
5150 }
5151 bpf_net_ctx_clear(bpf_net_ctx);
5152 return XDP_DROP;
5153 }
5154 bpf_net_ctx_clear(bpf_net_ctx);
5155 }
5156 return XDP_PASS;
5157 out_redir:
5158 bpf_net_ctx_clear(bpf_net_ctx);
5159 kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
5160 return XDP_DROP;
5161 }
5162 EXPORT_SYMBOL_GPL(do_xdp_generic);
5163
netif_rx_internal(struct sk_buff * skb)5164 static int netif_rx_internal(struct sk_buff *skb)
5165 {
5166 int ret;
5167
5168 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5169
5170 trace_netif_rx(skb);
5171
5172 #ifdef CONFIG_RPS
5173 if (static_branch_unlikely(&rps_needed)) {
5174 struct rps_dev_flow voidflow, *rflow = &voidflow;
5175 int cpu;
5176
5177 rcu_read_lock();
5178
5179 cpu = get_rps_cpu(skb->dev, skb, &rflow);
5180 if (cpu < 0)
5181 cpu = smp_processor_id();
5182
5183 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5184
5185 rcu_read_unlock();
5186 } else
5187 #endif
5188 {
5189 unsigned int qtail;
5190
5191 ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5192 }
5193 return ret;
5194 }
5195
5196 /**
5197 * __netif_rx - Slightly optimized version of netif_rx
5198 * @skb: buffer to post
5199 *
5200 * This behaves as netif_rx except that it does not disable bottom halves.
5201 * As a result this function may only be invoked from the interrupt context
5202 * (either hard or soft interrupt).
5203 */
__netif_rx(struct sk_buff * skb)5204 int __netif_rx(struct sk_buff *skb)
5205 {
5206 int ret;
5207
5208 lockdep_assert_once(hardirq_count() | softirq_count());
5209
5210 trace_netif_rx_entry(skb);
5211 ret = netif_rx_internal(skb);
5212 trace_netif_rx_exit(ret);
5213 return ret;
5214 }
5215 EXPORT_SYMBOL(__netif_rx);
5216
5217 /**
5218 * netif_rx - post buffer to the network code
5219 * @skb: buffer to post
5220 *
5221 * This function receives a packet from a device driver and queues it for
5222 * the upper (protocol) levels to process via the backlog NAPI device. It
5223 * always succeeds. The buffer may be dropped during processing for
5224 * congestion control or by the protocol layers.
5225 * The network buffer is passed via the backlog NAPI device. Modern NIC
5226 * driver should use NAPI and GRO.
5227 * This function can used from interrupt and from process context. The
5228 * caller from process context must not disable interrupts before invoking
5229 * this function.
5230 *
5231 * return values:
5232 * NET_RX_SUCCESS (no congestion)
5233 * NET_RX_DROP (packet was dropped)
5234 *
5235 */
netif_rx(struct sk_buff * skb)5236 int netif_rx(struct sk_buff *skb)
5237 {
5238 bool need_bh_off = !(hardirq_count() | softirq_count());
5239 int ret;
5240
5241 if (need_bh_off)
5242 local_bh_disable();
5243 trace_netif_rx_entry(skb);
5244 ret = netif_rx_internal(skb);
5245 trace_netif_rx_exit(ret);
5246 if (need_bh_off)
5247 local_bh_enable();
5248 return ret;
5249 }
5250 EXPORT_SYMBOL(netif_rx);
5251
net_tx_action(void)5252 static __latent_entropy void net_tx_action(void)
5253 {
5254 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5255
5256 if (sd->completion_queue) {
5257 struct sk_buff *clist;
5258
5259 local_irq_disable();
5260 clist = sd->completion_queue;
5261 sd->completion_queue = NULL;
5262 local_irq_enable();
5263
5264 while (clist) {
5265 struct sk_buff *skb = clist;
5266
5267 clist = clist->next;
5268
5269 WARN_ON(refcount_read(&skb->users));
5270 if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5271 trace_consume_skb(skb, net_tx_action);
5272 else
5273 trace_kfree_skb(skb, net_tx_action,
5274 get_kfree_skb_cb(skb)->reason, NULL);
5275
5276 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5277 __kfree_skb(skb);
5278 else
5279 __napi_kfree_skb(skb,
5280 get_kfree_skb_cb(skb)->reason);
5281 }
5282 }
5283
5284 if (sd->output_queue) {
5285 struct Qdisc *head;
5286
5287 local_irq_disable();
5288 head = sd->output_queue;
5289 sd->output_queue = NULL;
5290 sd->output_queue_tailp = &sd->output_queue;
5291 local_irq_enable();
5292
5293 rcu_read_lock();
5294
5295 while (head) {
5296 struct Qdisc *q = head;
5297 spinlock_t *root_lock = NULL;
5298
5299 head = head->next_sched;
5300
5301 /* We need to make sure head->next_sched is read
5302 * before clearing __QDISC_STATE_SCHED
5303 */
5304 smp_mb__before_atomic();
5305
5306 if (!(q->flags & TCQ_F_NOLOCK)) {
5307 root_lock = qdisc_lock(q);
5308 spin_lock(root_lock);
5309 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5310 &q->state))) {
5311 /* There is a synchronize_net() between
5312 * STATE_DEACTIVATED flag being set and
5313 * qdisc_reset()/some_qdisc_is_busy() in
5314 * dev_deactivate(), so we can safely bail out
5315 * early here to avoid data race between
5316 * qdisc_deactivate() and some_qdisc_is_busy()
5317 * for lockless qdisc.
5318 */
5319 clear_bit(__QDISC_STATE_SCHED, &q->state);
5320 continue;
5321 }
5322
5323 clear_bit(__QDISC_STATE_SCHED, &q->state);
5324 qdisc_run(q);
5325 if (root_lock)
5326 spin_unlock(root_lock);
5327 }
5328
5329 rcu_read_unlock();
5330 }
5331
5332 xfrm_dev_backlog(sd);
5333 }
5334
5335 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5336 /* This hook is defined here for ATM LANE */
5337 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5338 unsigned char *addr) __read_mostly;
5339 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5340 #endif
5341
5342 /**
5343 * netdev_is_rx_handler_busy - check if receive handler is registered
5344 * @dev: device to check
5345 *
5346 * Check if a receive handler is already registered for a given device.
5347 * Return true if there one.
5348 *
5349 * The caller must hold the rtnl_mutex.
5350 */
netdev_is_rx_handler_busy(struct net_device * dev)5351 bool netdev_is_rx_handler_busy(struct net_device *dev)
5352 {
5353 ASSERT_RTNL();
5354 return dev && rtnl_dereference(dev->rx_handler);
5355 }
5356 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5357
5358 /**
5359 * netdev_rx_handler_register - register receive handler
5360 * @dev: device to register a handler for
5361 * @rx_handler: receive handler to register
5362 * @rx_handler_data: data pointer that is used by rx handler
5363 *
5364 * Register a receive handler for a device. This handler will then be
5365 * called from __netif_receive_skb. A negative errno code is returned
5366 * on a failure.
5367 *
5368 * The caller must hold the rtnl_mutex.
5369 *
5370 * For a general description of rx_handler, see enum rx_handler_result.
5371 */
netdev_rx_handler_register(struct net_device * dev,rx_handler_func_t * rx_handler,void * rx_handler_data)5372 int netdev_rx_handler_register(struct net_device *dev,
5373 rx_handler_func_t *rx_handler,
5374 void *rx_handler_data)
5375 {
5376 if (netdev_is_rx_handler_busy(dev))
5377 return -EBUSY;
5378
5379 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5380 return -EINVAL;
5381
5382 /* Note: rx_handler_data must be set before rx_handler */
5383 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5384 rcu_assign_pointer(dev->rx_handler, rx_handler);
5385
5386 return 0;
5387 }
5388 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5389
5390 /**
5391 * netdev_rx_handler_unregister - unregister receive handler
5392 * @dev: device to unregister a handler from
5393 *
5394 * Unregister a receive handler from a device.
5395 *
5396 * The caller must hold the rtnl_mutex.
5397 */
netdev_rx_handler_unregister(struct net_device * dev)5398 void netdev_rx_handler_unregister(struct net_device *dev)
5399 {
5400
5401 ASSERT_RTNL();
5402 RCU_INIT_POINTER(dev->rx_handler, NULL);
5403 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5404 * section has a guarantee to see a non NULL rx_handler_data
5405 * as well.
5406 */
5407 synchronize_net();
5408 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5409 }
5410 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5411
5412 /*
5413 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5414 * the special handling of PFMEMALLOC skbs.
5415 */
skb_pfmemalloc_protocol(struct sk_buff * skb)5416 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5417 {
5418 switch (skb->protocol) {
5419 case htons(ETH_P_ARP):
5420 case htons(ETH_P_IP):
5421 case htons(ETH_P_IPV6):
5422 case htons(ETH_P_8021Q):
5423 case htons(ETH_P_8021AD):
5424 return true;
5425 default:
5426 return false;
5427 }
5428 }
5429
nf_ingress(struct sk_buff * skb,struct packet_type ** pt_prev,int * ret,struct net_device * orig_dev)5430 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5431 int *ret, struct net_device *orig_dev)
5432 {
5433 if (nf_hook_ingress_active(skb)) {
5434 int ingress_retval;
5435
5436 if (*pt_prev) {
5437 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5438 *pt_prev = NULL;
5439 }
5440
5441 rcu_read_lock();
5442 ingress_retval = nf_hook_ingress(skb);
5443 rcu_read_unlock();
5444 return ingress_retval;
5445 }
5446 return 0;
5447 }
5448
__netif_receive_skb_core(struct sk_buff ** pskb,bool pfmemalloc,struct packet_type ** ppt_prev)5449 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5450 struct packet_type **ppt_prev)
5451 {
5452 struct packet_type *ptype, *pt_prev;
5453 rx_handler_func_t *rx_handler;
5454 struct sk_buff *skb = *pskb;
5455 struct net_device *orig_dev;
5456 bool deliver_exact = false;
5457 int ret = NET_RX_DROP;
5458 __be16 type;
5459
5460 net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5461
5462 trace_netif_receive_skb(skb);
5463
5464 orig_dev = skb->dev;
5465
5466 skb_reset_network_header(skb);
5467 if (!skb_transport_header_was_set(skb))
5468 skb_reset_transport_header(skb);
5469 skb_reset_mac_len(skb);
5470
5471 pt_prev = NULL;
5472
5473 another_round:
5474 skb->skb_iif = skb->dev->ifindex;
5475
5476 __this_cpu_inc(softnet_data.processed);
5477
5478 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5479 int ret2;
5480
5481 migrate_disable();
5482 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
5483 &skb);
5484 migrate_enable();
5485
5486 if (ret2 != XDP_PASS) {
5487 ret = NET_RX_DROP;
5488 goto out;
5489 }
5490 }
5491
5492 if (eth_type_vlan(skb->protocol)) {
5493 skb = skb_vlan_untag(skb);
5494 if (unlikely(!skb))
5495 goto out;
5496 }
5497
5498 if (skb_skip_tc_classify(skb))
5499 goto skip_classify;
5500
5501 if (pfmemalloc)
5502 goto skip_taps;
5503
5504 list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) {
5505 if (pt_prev)
5506 ret = deliver_skb(skb, pt_prev, orig_dev);
5507 pt_prev = ptype;
5508 }
5509
5510 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5511 if (pt_prev)
5512 ret = deliver_skb(skb, pt_prev, orig_dev);
5513 pt_prev = ptype;
5514 }
5515
5516 skip_taps:
5517 #ifdef CONFIG_NET_INGRESS
5518 if (static_branch_unlikely(&ingress_needed_key)) {
5519 bool another = false;
5520
5521 nf_skip_egress(skb, true);
5522 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5523 &another);
5524 if (another)
5525 goto another_round;
5526 if (!skb)
5527 goto out;
5528
5529 nf_skip_egress(skb, false);
5530 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5531 goto out;
5532 }
5533 #endif
5534 skb_reset_redirect(skb);
5535 skip_classify:
5536 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5537 goto drop;
5538
5539 if (skb_vlan_tag_present(skb)) {
5540 if (pt_prev) {
5541 ret = deliver_skb(skb, pt_prev, orig_dev);
5542 pt_prev = NULL;
5543 }
5544 if (vlan_do_receive(&skb))
5545 goto another_round;
5546 else if (unlikely(!skb))
5547 goto out;
5548 }
5549
5550 rx_handler = rcu_dereference(skb->dev->rx_handler);
5551 if (rx_handler) {
5552 if (pt_prev) {
5553 ret = deliver_skb(skb, pt_prev, orig_dev);
5554 pt_prev = NULL;
5555 }
5556 switch (rx_handler(&skb)) {
5557 case RX_HANDLER_CONSUMED:
5558 ret = NET_RX_SUCCESS;
5559 goto out;
5560 case RX_HANDLER_ANOTHER:
5561 goto another_round;
5562 case RX_HANDLER_EXACT:
5563 deliver_exact = true;
5564 break;
5565 case RX_HANDLER_PASS:
5566 break;
5567 default:
5568 BUG();
5569 }
5570 }
5571
5572 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5573 check_vlan_id:
5574 if (skb_vlan_tag_get_id(skb)) {
5575 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5576 * find vlan device.
5577 */
5578 skb->pkt_type = PACKET_OTHERHOST;
5579 } else if (eth_type_vlan(skb->protocol)) {
5580 /* Outer header is 802.1P with vlan 0, inner header is
5581 * 802.1Q or 802.1AD and vlan_do_receive() above could
5582 * not find vlan dev for vlan id 0.
5583 */
5584 __vlan_hwaccel_clear_tag(skb);
5585 skb = skb_vlan_untag(skb);
5586 if (unlikely(!skb))
5587 goto out;
5588 if (vlan_do_receive(&skb))
5589 /* After stripping off 802.1P header with vlan 0
5590 * vlan dev is found for inner header.
5591 */
5592 goto another_round;
5593 else if (unlikely(!skb))
5594 goto out;
5595 else
5596 /* We have stripped outer 802.1P vlan 0 header.
5597 * But could not find vlan dev.
5598 * check again for vlan id to set OTHERHOST.
5599 */
5600 goto check_vlan_id;
5601 }
5602 /* Note: we might in the future use prio bits
5603 * and set skb->priority like in vlan_do_receive()
5604 * For the time being, just ignore Priority Code Point
5605 */
5606 __vlan_hwaccel_clear_tag(skb);
5607 }
5608
5609 type = skb->protocol;
5610
5611 /* deliver only exact match when indicated */
5612 if (likely(!deliver_exact)) {
5613 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5614 &ptype_base[ntohs(type) &
5615 PTYPE_HASH_MASK]);
5616 }
5617
5618 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5619 &orig_dev->ptype_specific);
5620
5621 if (unlikely(skb->dev != orig_dev)) {
5622 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5623 &skb->dev->ptype_specific);
5624 }
5625
5626 if (pt_prev) {
5627 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5628 goto drop;
5629 *ppt_prev = pt_prev;
5630 } else {
5631 drop:
5632 if (!deliver_exact)
5633 dev_core_stats_rx_dropped_inc(skb->dev);
5634 else
5635 dev_core_stats_rx_nohandler_inc(skb->dev);
5636 kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5637 /* Jamal, now you will not able to escape explaining
5638 * me how you were going to use this. :-)
5639 */
5640 ret = NET_RX_DROP;
5641 }
5642
5643 out:
5644 /* The invariant here is that if *ppt_prev is not NULL
5645 * then skb should also be non-NULL.
5646 *
5647 * Apparently *ppt_prev assignment above holds this invariant due to
5648 * skb dereferencing near it.
5649 */
5650 *pskb = skb;
5651 return ret;
5652 }
5653
__netif_receive_skb_one_core(struct sk_buff * skb,bool pfmemalloc)5654 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5655 {
5656 struct net_device *orig_dev = skb->dev;
5657 struct packet_type *pt_prev = NULL;
5658 int ret;
5659
5660 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5661 if (pt_prev)
5662 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5663 skb->dev, pt_prev, orig_dev);
5664 return ret;
5665 }
5666
5667 /**
5668 * netif_receive_skb_core - special purpose version of netif_receive_skb
5669 * @skb: buffer to process
5670 *
5671 * More direct receive version of netif_receive_skb(). It should
5672 * only be used by callers that have a need to skip RPS and Generic XDP.
5673 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5674 *
5675 * This function may only be called from softirq context and interrupts
5676 * should be enabled.
5677 *
5678 * Return values (usually ignored):
5679 * NET_RX_SUCCESS: no congestion
5680 * NET_RX_DROP: packet was dropped
5681 */
netif_receive_skb_core(struct sk_buff * skb)5682 int netif_receive_skb_core(struct sk_buff *skb)
5683 {
5684 int ret;
5685
5686 rcu_read_lock();
5687 ret = __netif_receive_skb_one_core(skb, false);
5688 rcu_read_unlock();
5689
5690 return ret;
5691 }
5692 EXPORT_SYMBOL(netif_receive_skb_core);
5693
__netif_receive_skb_list_ptype(struct list_head * head,struct packet_type * pt_prev,struct net_device * orig_dev)5694 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5695 struct packet_type *pt_prev,
5696 struct net_device *orig_dev)
5697 {
5698 struct sk_buff *skb, *next;
5699
5700 if (!pt_prev)
5701 return;
5702 if (list_empty(head))
5703 return;
5704 if (pt_prev->list_func != NULL)
5705 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5706 ip_list_rcv, head, pt_prev, orig_dev);
5707 else
5708 list_for_each_entry_safe(skb, next, head, list) {
5709 skb_list_del_init(skb);
5710 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5711 }
5712 }
5713
__netif_receive_skb_list_core(struct list_head * head,bool pfmemalloc)5714 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5715 {
5716 /* Fast-path assumptions:
5717 * - There is no RX handler.
5718 * - Only one packet_type matches.
5719 * If either of these fails, we will end up doing some per-packet
5720 * processing in-line, then handling the 'last ptype' for the whole
5721 * sublist. This can't cause out-of-order delivery to any single ptype,
5722 * because the 'last ptype' must be constant across the sublist, and all
5723 * other ptypes are handled per-packet.
5724 */
5725 /* Current (common) ptype of sublist */
5726 struct packet_type *pt_curr = NULL;
5727 /* Current (common) orig_dev of sublist */
5728 struct net_device *od_curr = NULL;
5729 struct sk_buff *skb, *next;
5730 LIST_HEAD(sublist);
5731
5732 list_for_each_entry_safe(skb, next, head, list) {
5733 struct net_device *orig_dev = skb->dev;
5734 struct packet_type *pt_prev = NULL;
5735
5736 skb_list_del_init(skb);
5737 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5738 if (!pt_prev)
5739 continue;
5740 if (pt_curr != pt_prev || od_curr != orig_dev) {
5741 /* dispatch old sublist */
5742 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5743 /* start new sublist */
5744 INIT_LIST_HEAD(&sublist);
5745 pt_curr = pt_prev;
5746 od_curr = orig_dev;
5747 }
5748 list_add_tail(&skb->list, &sublist);
5749 }
5750
5751 /* dispatch final sublist */
5752 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5753 }
5754
__netif_receive_skb(struct sk_buff * skb)5755 static int __netif_receive_skb(struct sk_buff *skb)
5756 {
5757 int ret;
5758
5759 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5760 unsigned int noreclaim_flag;
5761
5762 /*
5763 * PFMEMALLOC skbs are special, they should
5764 * - be delivered to SOCK_MEMALLOC sockets only
5765 * - stay away from userspace
5766 * - have bounded memory usage
5767 *
5768 * Use PF_MEMALLOC as this saves us from propagating the allocation
5769 * context down to all allocation sites.
5770 */
5771 noreclaim_flag = memalloc_noreclaim_save();
5772 ret = __netif_receive_skb_one_core(skb, true);
5773 memalloc_noreclaim_restore(noreclaim_flag);
5774 } else
5775 ret = __netif_receive_skb_one_core(skb, false);
5776
5777 return ret;
5778 }
5779
__netif_receive_skb_list(struct list_head * head)5780 static void __netif_receive_skb_list(struct list_head *head)
5781 {
5782 unsigned long noreclaim_flag = 0;
5783 struct sk_buff *skb, *next;
5784 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5785
5786 list_for_each_entry_safe(skb, next, head, list) {
5787 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5788 struct list_head sublist;
5789
5790 /* Handle the previous sublist */
5791 list_cut_before(&sublist, head, &skb->list);
5792 if (!list_empty(&sublist))
5793 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5794 pfmemalloc = !pfmemalloc;
5795 /* See comments in __netif_receive_skb */
5796 if (pfmemalloc)
5797 noreclaim_flag = memalloc_noreclaim_save();
5798 else
5799 memalloc_noreclaim_restore(noreclaim_flag);
5800 }
5801 }
5802 /* Handle the remaining sublist */
5803 if (!list_empty(head))
5804 __netif_receive_skb_list_core(head, pfmemalloc);
5805 /* Restore pflags */
5806 if (pfmemalloc)
5807 memalloc_noreclaim_restore(noreclaim_flag);
5808 }
5809
generic_xdp_install(struct net_device * dev,struct netdev_bpf * xdp)5810 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5811 {
5812 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5813 struct bpf_prog *new = xdp->prog;
5814 int ret = 0;
5815
5816 switch (xdp->command) {
5817 case XDP_SETUP_PROG:
5818 rcu_assign_pointer(dev->xdp_prog, new);
5819 if (old)
5820 bpf_prog_put(old);
5821
5822 if (old && !new) {
5823 static_branch_dec(&generic_xdp_needed_key);
5824 } else if (new && !old) {
5825 static_branch_inc(&generic_xdp_needed_key);
5826 dev_disable_lro(dev);
5827 dev_disable_gro_hw(dev);
5828 }
5829 break;
5830
5831 default:
5832 ret = -EINVAL;
5833 break;
5834 }
5835
5836 return ret;
5837 }
5838
netif_receive_skb_internal(struct sk_buff * skb)5839 static int netif_receive_skb_internal(struct sk_buff *skb)
5840 {
5841 int ret;
5842
5843 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5844
5845 if (skb_defer_rx_timestamp(skb))
5846 return NET_RX_SUCCESS;
5847
5848 rcu_read_lock();
5849 #ifdef CONFIG_RPS
5850 if (static_branch_unlikely(&rps_needed)) {
5851 struct rps_dev_flow voidflow, *rflow = &voidflow;
5852 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5853
5854 if (cpu >= 0) {
5855 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5856 rcu_read_unlock();
5857 return ret;
5858 }
5859 }
5860 #endif
5861 ret = __netif_receive_skb(skb);
5862 rcu_read_unlock();
5863 return ret;
5864 }
5865
netif_receive_skb_list_internal(struct list_head * head)5866 void netif_receive_skb_list_internal(struct list_head *head)
5867 {
5868 struct sk_buff *skb, *next;
5869 LIST_HEAD(sublist);
5870
5871 list_for_each_entry_safe(skb, next, head, list) {
5872 net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
5873 skb);
5874 skb_list_del_init(skb);
5875 if (!skb_defer_rx_timestamp(skb))
5876 list_add_tail(&skb->list, &sublist);
5877 }
5878 list_splice_init(&sublist, head);
5879
5880 rcu_read_lock();
5881 #ifdef CONFIG_RPS
5882 if (static_branch_unlikely(&rps_needed)) {
5883 list_for_each_entry_safe(skb, next, head, list) {
5884 struct rps_dev_flow voidflow, *rflow = &voidflow;
5885 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5886
5887 if (cpu >= 0) {
5888 /* Will be handled, remove from list */
5889 skb_list_del_init(skb);
5890 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5891 }
5892 }
5893 }
5894 #endif
5895 __netif_receive_skb_list(head);
5896 rcu_read_unlock();
5897 }
5898
5899 /**
5900 * netif_receive_skb - process receive buffer from network
5901 * @skb: buffer to process
5902 *
5903 * netif_receive_skb() is the main receive data processing function.
5904 * It always succeeds. The buffer may be dropped during processing
5905 * for congestion control or by the protocol layers.
5906 *
5907 * This function may only be called from softirq context and interrupts
5908 * should be enabled.
5909 *
5910 * Return values (usually ignored):
5911 * NET_RX_SUCCESS: no congestion
5912 * NET_RX_DROP: packet was dropped
5913 */
netif_receive_skb(struct sk_buff * skb)5914 int netif_receive_skb(struct sk_buff *skb)
5915 {
5916 int ret;
5917
5918 trace_netif_receive_skb_entry(skb);
5919
5920 ret = netif_receive_skb_internal(skb);
5921 trace_netif_receive_skb_exit(ret);
5922
5923 return ret;
5924 }
5925 EXPORT_SYMBOL(netif_receive_skb);
5926
5927 /**
5928 * netif_receive_skb_list - process many receive buffers from network
5929 * @head: list of skbs to process.
5930 *
5931 * Since return value of netif_receive_skb() is normally ignored, and
5932 * wouldn't be meaningful for a list, this function returns void.
5933 *
5934 * This function may only be called from softirq context and interrupts
5935 * should be enabled.
5936 */
netif_receive_skb_list(struct list_head * head)5937 void netif_receive_skb_list(struct list_head *head)
5938 {
5939 struct sk_buff *skb;
5940
5941 if (list_empty(head))
5942 return;
5943 if (trace_netif_receive_skb_list_entry_enabled()) {
5944 list_for_each_entry(skb, head, list)
5945 trace_netif_receive_skb_list_entry(skb);
5946 }
5947 netif_receive_skb_list_internal(head);
5948 trace_netif_receive_skb_list_exit(0);
5949 }
5950 EXPORT_SYMBOL(netif_receive_skb_list);
5951
5952 static DEFINE_PER_CPU(struct work_struct, flush_works);
5953
5954 /* Network device is going away, flush any packets still pending */
flush_backlog(struct work_struct * work)5955 static void flush_backlog(struct work_struct *work)
5956 {
5957 struct sk_buff *skb, *tmp;
5958 struct softnet_data *sd;
5959
5960 local_bh_disable();
5961 sd = this_cpu_ptr(&softnet_data);
5962
5963 backlog_lock_irq_disable(sd);
5964 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5965 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5966 __skb_unlink(skb, &sd->input_pkt_queue);
5967 dev_kfree_skb_irq(skb);
5968 rps_input_queue_head_incr(sd);
5969 }
5970 }
5971 backlog_unlock_irq_enable(sd);
5972
5973 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
5974 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5975 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5976 __skb_unlink(skb, &sd->process_queue);
5977 kfree_skb(skb);
5978 rps_input_queue_head_incr(sd);
5979 }
5980 }
5981 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
5982 local_bh_enable();
5983 }
5984
flush_required(int cpu)5985 static bool flush_required(int cpu)
5986 {
5987 #if IS_ENABLED(CONFIG_RPS)
5988 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5989 bool do_flush;
5990
5991 backlog_lock_irq_disable(sd);
5992
5993 /* as insertion into process_queue happens with the rps lock held,
5994 * process_queue access may race only with dequeue
5995 */
5996 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5997 !skb_queue_empty_lockless(&sd->process_queue);
5998 backlog_unlock_irq_enable(sd);
5999
6000 return do_flush;
6001 #endif
6002 /* without RPS we can't safely check input_pkt_queue: during a
6003 * concurrent remote skb_queue_splice() we can detect as empty both
6004 * input_pkt_queue and process_queue even if the latter could end-up
6005 * containing a lot of packets.
6006 */
6007 return true;
6008 }
6009
flush_all_backlogs(void)6010 static void flush_all_backlogs(void)
6011 {
6012 static cpumask_t flush_cpus;
6013 unsigned int cpu;
6014
6015 /* since we are under rtnl lock protection we can use static data
6016 * for the cpumask and avoid allocating on stack the possibly
6017 * large mask
6018 */
6019 ASSERT_RTNL();
6020
6021 cpus_read_lock();
6022
6023 cpumask_clear(&flush_cpus);
6024 for_each_online_cpu(cpu) {
6025 if (flush_required(cpu)) {
6026 queue_work_on(cpu, system_highpri_wq,
6027 per_cpu_ptr(&flush_works, cpu));
6028 cpumask_set_cpu(cpu, &flush_cpus);
6029 }
6030 }
6031
6032 /* we can have in flight packet[s] on the cpus we are not flushing,
6033 * synchronize_net() in unregister_netdevice_many() will take care of
6034 * them
6035 */
6036 for_each_cpu(cpu, &flush_cpus)
6037 flush_work(per_cpu_ptr(&flush_works, cpu));
6038
6039 cpus_read_unlock();
6040 }
6041
net_rps_send_ipi(struct softnet_data * remsd)6042 static void net_rps_send_ipi(struct softnet_data *remsd)
6043 {
6044 #ifdef CONFIG_RPS
6045 while (remsd) {
6046 struct softnet_data *next = remsd->rps_ipi_next;
6047
6048 if (cpu_online(remsd->cpu))
6049 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6050 remsd = next;
6051 }
6052 #endif
6053 }
6054
6055 /*
6056 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6057 * Note: called with local irq disabled, but exits with local irq enabled.
6058 */
net_rps_action_and_irq_enable(struct softnet_data * sd)6059 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6060 {
6061 #ifdef CONFIG_RPS
6062 struct softnet_data *remsd = sd->rps_ipi_list;
6063
6064 if (!use_backlog_threads() && remsd) {
6065 sd->rps_ipi_list = NULL;
6066
6067 local_irq_enable();
6068
6069 /* Send pending IPI's to kick RPS processing on remote cpus. */
6070 net_rps_send_ipi(remsd);
6071 } else
6072 #endif
6073 local_irq_enable();
6074 }
6075
sd_has_rps_ipi_waiting(struct softnet_data * sd)6076 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6077 {
6078 #ifdef CONFIG_RPS
6079 return !use_backlog_threads() && sd->rps_ipi_list;
6080 #else
6081 return false;
6082 #endif
6083 }
6084
process_backlog(struct napi_struct * napi,int quota)6085 static int process_backlog(struct napi_struct *napi, int quota)
6086 {
6087 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6088 bool again = true;
6089 int work = 0;
6090
6091 /* Check if we have pending ipi, its better to send them now,
6092 * not waiting net_rx_action() end.
6093 */
6094 if (sd_has_rps_ipi_waiting(sd)) {
6095 local_irq_disable();
6096 net_rps_action_and_irq_enable(sd);
6097 }
6098
6099 napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
6100 while (again) {
6101 struct sk_buff *skb;
6102
6103 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6104 while ((skb = __skb_dequeue(&sd->process_queue))) {
6105 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6106 rcu_read_lock();
6107 __netif_receive_skb(skb);
6108 rcu_read_unlock();
6109 if (++work >= quota) {
6110 rps_input_queue_head_add(sd, work);
6111 return work;
6112 }
6113
6114 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6115 }
6116 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6117
6118 backlog_lock_irq_disable(sd);
6119 if (skb_queue_empty(&sd->input_pkt_queue)) {
6120 /*
6121 * Inline a custom version of __napi_complete().
6122 * only current cpu owns and manipulates this napi,
6123 * and NAPI_STATE_SCHED is the only possible flag set
6124 * on backlog.
6125 * We can use a plain write instead of clear_bit(),
6126 * and we dont need an smp_mb() memory barrier.
6127 */
6128 napi->state &= NAPIF_STATE_THREADED;
6129 again = false;
6130 } else {
6131 local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6132 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6133 &sd->process_queue);
6134 local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6135 }
6136 backlog_unlock_irq_enable(sd);
6137 }
6138
6139 if (work)
6140 rps_input_queue_head_add(sd, work);
6141 return work;
6142 }
6143
6144 /**
6145 * __napi_schedule - schedule for receive
6146 * @n: entry to schedule
6147 *
6148 * The entry's receive function will be scheduled to run.
6149 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6150 */
__napi_schedule(struct napi_struct * n)6151 void __napi_schedule(struct napi_struct *n)
6152 {
6153 unsigned long flags;
6154
6155 local_irq_save(flags);
6156 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6157 local_irq_restore(flags);
6158 }
6159 EXPORT_SYMBOL(__napi_schedule);
6160
6161 /**
6162 * napi_schedule_prep - check if napi can be scheduled
6163 * @n: napi context
6164 *
6165 * Test if NAPI routine is already running, and if not mark
6166 * it as running. This is used as a condition variable to
6167 * insure only one NAPI poll instance runs. We also make
6168 * sure there is no pending NAPI disable.
6169 */
napi_schedule_prep(struct napi_struct * n)6170 bool napi_schedule_prep(struct napi_struct *n)
6171 {
6172 unsigned long new, val = READ_ONCE(n->state);
6173
6174 do {
6175 if (unlikely(val & NAPIF_STATE_DISABLE))
6176 return false;
6177 new = val | NAPIF_STATE_SCHED;
6178
6179 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6180 * This was suggested by Alexander Duyck, as compiler
6181 * emits better code than :
6182 * if (val & NAPIF_STATE_SCHED)
6183 * new |= NAPIF_STATE_MISSED;
6184 */
6185 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6186 NAPIF_STATE_MISSED;
6187 } while (!try_cmpxchg(&n->state, &val, new));
6188
6189 return !(val & NAPIF_STATE_SCHED);
6190 }
6191 EXPORT_SYMBOL(napi_schedule_prep);
6192
6193 /**
6194 * __napi_schedule_irqoff - schedule for receive
6195 * @n: entry to schedule
6196 *
6197 * Variant of __napi_schedule() assuming hard irqs are masked.
6198 *
6199 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6200 * because the interrupt disabled assumption might not be true
6201 * due to force-threaded interrupts and spinlock substitution.
6202 */
__napi_schedule_irqoff(struct napi_struct * n)6203 void __napi_schedule_irqoff(struct napi_struct *n)
6204 {
6205 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6206 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6207 else
6208 __napi_schedule(n);
6209 }
6210 EXPORT_SYMBOL(__napi_schedule_irqoff);
6211
napi_complete_done(struct napi_struct * n,int work_done)6212 bool napi_complete_done(struct napi_struct *n, int work_done)
6213 {
6214 unsigned long flags, val, new, timeout = 0;
6215 bool ret = true;
6216
6217 /*
6218 * 1) Don't let napi dequeue from the cpu poll list
6219 * just in case its running on a different cpu.
6220 * 2) If we are busy polling, do nothing here, we have
6221 * the guarantee we will be called later.
6222 */
6223 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6224 NAPIF_STATE_IN_BUSY_POLL)))
6225 return false;
6226
6227 if (work_done) {
6228 if (n->gro_bitmask)
6229 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6230 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6231 }
6232 if (n->defer_hard_irqs_count > 0) {
6233 n->defer_hard_irqs_count--;
6234 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6235 if (timeout)
6236 ret = false;
6237 }
6238 if (n->gro_bitmask) {
6239 /* When the NAPI instance uses a timeout and keeps postponing
6240 * it, we need to bound somehow the time packets are kept in
6241 * the GRO layer
6242 */
6243 napi_gro_flush(n, !!timeout);
6244 }
6245
6246 gro_normal_list(n);
6247
6248 if (unlikely(!list_empty(&n->poll_list))) {
6249 /* If n->poll_list is not empty, we need to mask irqs */
6250 local_irq_save(flags);
6251 list_del_init(&n->poll_list);
6252 local_irq_restore(flags);
6253 }
6254 WRITE_ONCE(n->list_owner, -1);
6255
6256 val = READ_ONCE(n->state);
6257 do {
6258 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6259
6260 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6261 NAPIF_STATE_SCHED_THREADED |
6262 NAPIF_STATE_PREFER_BUSY_POLL);
6263
6264 /* If STATE_MISSED was set, leave STATE_SCHED set,
6265 * because we will call napi->poll() one more time.
6266 * This C code was suggested by Alexander Duyck to help gcc.
6267 */
6268 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6269 NAPIF_STATE_SCHED;
6270 } while (!try_cmpxchg(&n->state, &val, new));
6271
6272 if (unlikely(val & NAPIF_STATE_MISSED)) {
6273 __napi_schedule(n);
6274 return false;
6275 }
6276
6277 if (timeout)
6278 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6279 HRTIMER_MODE_REL_PINNED);
6280 return ret;
6281 }
6282 EXPORT_SYMBOL(napi_complete_done);
6283
6284 /* must be called under rcu_read_lock(), as we dont take a reference */
napi_by_id(unsigned int napi_id)6285 struct napi_struct *napi_by_id(unsigned int napi_id)
6286 {
6287 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6288 struct napi_struct *napi;
6289
6290 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6291 if (napi->napi_id == napi_id)
6292 return napi;
6293
6294 return NULL;
6295 }
6296
skb_defer_free_flush(struct softnet_data * sd)6297 static void skb_defer_free_flush(struct softnet_data *sd)
6298 {
6299 struct sk_buff *skb, *next;
6300
6301 /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6302 if (!READ_ONCE(sd->defer_list))
6303 return;
6304
6305 spin_lock(&sd->defer_lock);
6306 skb = sd->defer_list;
6307 sd->defer_list = NULL;
6308 sd->defer_count = 0;
6309 spin_unlock(&sd->defer_lock);
6310
6311 while (skb != NULL) {
6312 next = skb->next;
6313 napi_consume_skb(skb, 1);
6314 skb = next;
6315 }
6316 }
6317
6318 #if defined(CONFIG_NET_RX_BUSY_POLL)
6319
__busy_poll_stop(struct napi_struct * napi,bool skip_schedule)6320 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6321 {
6322 if (!skip_schedule) {
6323 gro_normal_list(napi);
6324 __napi_schedule(napi);
6325 return;
6326 }
6327
6328 if (napi->gro_bitmask) {
6329 /* flush too old packets
6330 * If HZ < 1000, flush all packets.
6331 */
6332 napi_gro_flush(napi, HZ >= 1000);
6333 }
6334
6335 gro_normal_list(napi);
6336 clear_bit(NAPI_STATE_SCHED, &napi->state);
6337 }
6338
6339 enum {
6340 NAPI_F_PREFER_BUSY_POLL = 1,
6341 NAPI_F_END_ON_RESCHED = 2,
6342 };
6343
busy_poll_stop(struct napi_struct * napi,void * have_poll_lock,unsigned flags,u16 budget)6344 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock,
6345 unsigned flags, u16 budget)
6346 {
6347 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6348 bool skip_schedule = false;
6349 unsigned long timeout;
6350 int rc;
6351
6352 /* Busy polling means there is a high chance device driver hard irq
6353 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6354 * set in napi_schedule_prep().
6355 * Since we are about to call napi->poll() once more, we can safely
6356 * clear NAPI_STATE_MISSED.
6357 *
6358 * Note: x86 could use a single "lock and ..." instruction
6359 * to perform these two clear_bit()
6360 */
6361 clear_bit(NAPI_STATE_MISSED, &napi->state);
6362 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6363
6364 local_bh_disable();
6365 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6366
6367 if (flags & NAPI_F_PREFER_BUSY_POLL) {
6368 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6369 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6370 if (napi->defer_hard_irqs_count && timeout) {
6371 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6372 skip_schedule = true;
6373 }
6374 }
6375
6376 /* All we really want here is to re-enable device interrupts.
6377 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6378 */
6379 rc = napi->poll(napi, budget);
6380 /* We can't gro_normal_list() here, because napi->poll() might have
6381 * rearmed the napi (napi_complete_done()) in which case it could
6382 * already be running on another CPU.
6383 */
6384 trace_napi_poll(napi, rc, budget);
6385 netpoll_poll_unlock(have_poll_lock);
6386 if (rc == budget)
6387 __busy_poll_stop(napi, skip_schedule);
6388 bpf_net_ctx_clear(bpf_net_ctx);
6389 local_bh_enable();
6390 }
6391
__napi_busy_loop(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,unsigned flags,u16 budget)6392 static void __napi_busy_loop(unsigned int napi_id,
6393 bool (*loop_end)(void *, unsigned long),
6394 void *loop_end_arg, unsigned flags, u16 budget)
6395 {
6396 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6397 int (*napi_poll)(struct napi_struct *napi, int budget);
6398 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6399 void *have_poll_lock = NULL;
6400 struct napi_struct *napi;
6401
6402 WARN_ON_ONCE(!rcu_read_lock_held());
6403
6404 restart:
6405 napi_poll = NULL;
6406
6407 napi = napi_by_id(napi_id);
6408 if (!napi)
6409 return;
6410
6411 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6412 preempt_disable();
6413 for (;;) {
6414 int work = 0;
6415
6416 local_bh_disable();
6417 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6418 if (!napi_poll) {
6419 unsigned long val = READ_ONCE(napi->state);
6420
6421 /* If multiple threads are competing for this napi,
6422 * we avoid dirtying napi->state as much as we can.
6423 */
6424 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6425 NAPIF_STATE_IN_BUSY_POLL)) {
6426 if (flags & NAPI_F_PREFER_BUSY_POLL)
6427 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6428 goto count;
6429 }
6430 if (cmpxchg(&napi->state, val,
6431 val | NAPIF_STATE_IN_BUSY_POLL |
6432 NAPIF_STATE_SCHED) != val) {
6433 if (flags & NAPI_F_PREFER_BUSY_POLL)
6434 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6435 goto count;
6436 }
6437 have_poll_lock = netpoll_poll_lock(napi);
6438 napi_poll = napi->poll;
6439 }
6440 work = napi_poll(napi, budget);
6441 trace_napi_poll(napi, work, budget);
6442 gro_normal_list(napi);
6443 count:
6444 if (work > 0)
6445 __NET_ADD_STATS(dev_net(napi->dev),
6446 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6447 skb_defer_free_flush(this_cpu_ptr(&softnet_data));
6448 bpf_net_ctx_clear(bpf_net_ctx);
6449 local_bh_enable();
6450
6451 if (!loop_end || loop_end(loop_end_arg, start_time))
6452 break;
6453
6454 if (unlikely(need_resched())) {
6455 if (flags & NAPI_F_END_ON_RESCHED)
6456 break;
6457 if (napi_poll)
6458 busy_poll_stop(napi, have_poll_lock, flags, budget);
6459 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6460 preempt_enable();
6461 rcu_read_unlock();
6462 cond_resched();
6463 rcu_read_lock();
6464 if (loop_end(loop_end_arg, start_time))
6465 return;
6466 goto restart;
6467 }
6468 cpu_relax();
6469 }
6470 if (napi_poll)
6471 busy_poll_stop(napi, have_poll_lock, flags, budget);
6472 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6473 preempt_enable();
6474 }
6475
napi_busy_loop_rcu(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,bool prefer_busy_poll,u16 budget)6476 void napi_busy_loop_rcu(unsigned int napi_id,
6477 bool (*loop_end)(void *, unsigned long),
6478 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6479 {
6480 unsigned flags = NAPI_F_END_ON_RESCHED;
6481
6482 if (prefer_busy_poll)
6483 flags |= NAPI_F_PREFER_BUSY_POLL;
6484
6485 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6486 }
6487
napi_busy_loop(unsigned int napi_id,bool (* loop_end)(void *,unsigned long),void * loop_end_arg,bool prefer_busy_poll,u16 budget)6488 void napi_busy_loop(unsigned int napi_id,
6489 bool (*loop_end)(void *, unsigned long),
6490 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6491 {
6492 unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0;
6493
6494 rcu_read_lock();
6495 __napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6496 rcu_read_unlock();
6497 }
6498 EXPORT_SYMBOL(napi_busy_loop);
6499
6500 #endif /* CONFIG_NET_RX_BUSY_POLL */
6501
napi_hash_add(struct napi_struct * napi)6502 static void napi_hash_add(struct napi_struct *napi)
6503 {
6504 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6505 return;
6506
6507 spin_lock(&napi_hash_lock);
6508
6509 /* 0..NR_CPUS range is reserved for sender_cpu use */
6510 do {
6511 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6512 napi_gen_id = MIN_NAPI_ID;
6513 } while (napi_by_id(napi_gen_id));
6514 napi->napi_id = napi_gen_id;
6515
6516 hlist_add_head_rcu(&napi->napi_hash_node,
6517 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6518
6519 spin_unlock(&napi_hash_lock);
6520 }
6521
6522 /* Warning : caller is responsible to make sure rcu grace period
6523 * is respected before freeing memory containing @napi
6524 */
napi_hash_del(struct napi_struct * napi)6525 static void napi_hash_del(struct napi_struct *napi)
6526 {
6527 spin_lock(&napi_hash_lock);
6528
6529 hlist_del_init_rcu(&napi->napi_hash_node);
6530
6531 spin_unlock(&napi_hash_lock);
6532 }
6533
napi_watchdog(struct hrtimer * timer)6534 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6535 {
6536 struct napi_struct *napi;
6537
6538 napi = container_of(timer, struct napi_struct, timer);
6539
6540 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6541 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6542 */
6543 if (!napi_disable_pending(napi) &&
6544 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6545 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6546 __napi_schedule_irqoff(napi);
6547 }
6548
6549 return HRTIMER_NORESTART;
6550 }
6551
init_gro_hash(struct napi_struct * napi)6552 static void init_gro_hash(struct napi_struct *napi)
6553 {
6554 int i;
6555
6556 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6557 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6558 napi->gro_hash[i].count = 0;
6559 }
6560 napi->gro_bitmask = 0;
6561 }
6562
dev_set_threaded(struct net_device * dev,bool threaded)6563 int dev_set_threaded(struct net_device *dev, bool threaded)
6564 {
6565 struct napi_struct *napi;
6566 int err = 0;
6567
6568 if (dev->threaded == threaded)
6569 return 0;
6570
6571 if (threaded) {
6572 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6573 if (!napi->thread) {
6574 err = napi_kthread_create(napi);
6575 if (err) {
6576 threaded = false;
6577 break;
6578 }
6579 }
6580 }
6581 }
6582
6583 WRITE_ONCE(dev->threaded, threaded);
6584
6585 /* Make sure kthread is created before THREADED bit
6586 * is set.
6587 */
6588 smp_mb__before_atomic();
6589
6590 /* Setting/unsetting threaded mode on a napi might not immediately
6591 * take effect, if the current napi instance is actively being
6592 * polled. In this case, the switch between threaded mode and
6593 * softirq mode will happen in the next round of napi_schedule().
6594 * This should not cause hiccups/stalls to the live traffic.
6595 */
6596 list_for_each_entry(napi, &dev->napi_list, dev_list)
6597 assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6598
6599 return err;
6600 }
6601 EXPORT_SYMBOL(dev_set_threaded);
6602
6603 /**
6604 * netif_queue_set_napi - Associate queue with the napi
6605 * @dev: device to which NAPI and queue belong
6606 * @queue_index: Index of queue
6607 * @type: queue type as RX or TX
6608 * @napi: NAPI context, pass NULL to clear previously set NAPI
6609 *
6610 * Set queue with its corresponding napi context. This should be done after
6611 * registering the NAPI handler for the queue-vector and the queues have been
6612 * mapped to the corresponding interrupt vector.
6613 */
netif_queue_set_napi(struct net_device * dev,unsigned int queue_index,enum netdev_queue_type type,struct napi_struct * napi)6614 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index,
6615 enum netdev_queue_type type, struct napi_struct *napi)
6616 {
6617 struct netdev_rx_queue *rxq;
6618 struct netdev_queue *txq;
6619
6620 if (WARN_ON_ONCE(napi && !napi->dev))
6621 return;
6622 if (dev->reg_state >= NETREG_REGISTERED)
6623 ASSERT_RTNL();
6624
6625 switch (type) {
6626 case NETDEV_QUEUE_TYPE_RX:
6627 rxq = __netif_get_rx_queue(dev, queue_index);
6628 rxq->napi = napi;
6629 return;
6630 case NETDEV_QUEUE_TYPE_TX:
6631 txq = netdev_get_tx_queue(dev, queue_index);
6632 txq->napi = napi;
6633 return;
6634 default:
6635 return;
6636 }
6637 }
6638 EXPORT_SYMBOL(netif_queue_set_napi);
6639
netif_napi_add_weight(struct net_device * dev,struct napi_struct * napi,int (* poll)(struct napi_struct *,int),int weight)6640 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6641 int (*poll)(struct napi_struct *, int), int weight)
6642 {
6643 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6644 return;
6645
6646 INIT_LIST_HEAD(&napi->poll_list);
6647 INIT_HLIST_NODE(&napi->napi_hash_node);
6648 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6649 napi->timer.function = napi_watchdog;
6650 init_gro_hash(napi);
6651 napi->skb = NULL;
6652 INIT_LIST_HEAD(&napi->rx_list);
6653 napi->rx_count = 0;
6654 napi->poll = poll;
6655 if (weight > NAPI_POLL_WEIGHT)
6656 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6657 weight);
6658 napi->weight = weight;
6659 napi->dev = dev;
6660 #ifdef CONFIG_NETPOLL
6661 napi->poll_owner = -1;
6662 #endif
6663 napi->list_owner = -1;
6664 set_bit(NAPI_STATE_SCHED, &napi->state);
6665 set_bit(NAPI_STATE_NPSVC, &napi->state);
6666 list_add_rcu(&napi->dev_list, &dev->napi_list);
6667 napi_hash_add(napi);
6668 napi_get_frags_check(napi);
6669 /* Create kthread for this napi if dev->threaded is set.
6670 * Clear dev->threaded if kthread creation failed so that
6671 * threaded mode will not be enabled in napi_enable().
6672 */
6673 if (dev->threaded && napi_kthread_create(napi))
6674 dev->threaded = false;
6675 netif_napi_set_irq(napi, -1);
6676 }
6677 EXPORT_SYMBOL(netif_napi_add_weight);
6678
napi_disable(struct napi_struct * n)6679 void napi_disable(struct napi_struct *n)
6680 {
6681 unsigned long val, new;
6682
6683 might_sleep();
6684 set_bit(NAPI_STATE_DISABLE, &n->state);
6685
6686 val = READ_ONCE(n->state);
6687 do {
6688 while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6689 usleep_range(20, 200);
6690 val = READ_ONCE(n->state);
6691 }
6692
6693 new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6694 new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6695 } while (!try_cmpxchg(&n->state, &val, new));
6696
6697 hrtimer_cancel(&n->timer);
6698
6699 clear_bit(NAPI_STATE_DISABLE, &n->state);
6700 }
6701 EXPORT_SYMBOL(napi_disable);
6702
6703 /**
6704 * napi_enable - enable NAPI scheduling
6705 * @n: NAPI context
6706 *
6707 * Resume NAPI from being scheduled on this context.
6708 * Must be paired with napi_disable.
6709 */
napi_enable(struct napi_struct * n)6710 void napi_enable(struct napi_struct *n)
6711 {
6712 unsigned long new, val = READ_ONCE(n->state);
6713
6714 do {
6715 BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6716
6717 new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6718 if (n->dev->threaded && n->thread)
6719 new |= NAPIF_STATE_THREADED;
6720 } while (!try_cmpxchg(&n->state, &val, new));
6721 }
6722 EXPORT_SYMBOL(napi_enable);
6723
flush_gro_hash(struct napi_struct * napi)6724 static void flush_gro_hash(struct napi_struct *napi)
6725 {
6726 int i;
6727
6728 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6729 struct sk_buff *skb, *n;
6730
6731 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6732 kfree_skb(skb);
6733 napi->gro_hash[i].count = 0;
6734 }
6735 }
6736
6737 /* Must be called in process context */
__netif_napi_del(struct napi_struct * napi)6738 void __netif_napi_del(struct napi_struct *napi)
6739 {
6740 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6741 return;
6742
6743 napi_hash_del(napi);
6744 list_del_rcu(&napi->dev_list);
6745 napi_free_frags(napi);
6746
6747 flush_gro_hash(napi);
6748 napi->gro_bitmask = 0;
6749
6750 if (napi->thread) {
6751 kthread_stop(napi->thread);
6752 napi->thread = NULL;
6753 }
6754 }
6755 EXPORT_SYMBOL(__netif_napi_del);
6756
__napi_poll(struct napi_struct * n,bool * repoll)6757 static int __napi_poll(struct napi_struct *n, bool *repoll)
6758 {
6759 int work, weight;
6760
6761 weight = n->weight;
6762
6763 /* This NAPI_STATE_SCHED test is for avoiding a race
6764 * with netpoll's poll_napi(). Only the entity which
6765 * obtains the lock and sees NAPI_STATE_SCHED set will
6766 * actually make the ->poll() call. Therefore we avoid
6767 * accidentally calling ->poll() when NAPI is not scheduled.
6768 */
6769 work = 0;
6770 if (napi_is_scheduled(n)) {
6771 work = n->poll(n, weight);
6772 trace_napi_poll(n, work, weight);
6773
6774 xdp_do_check_flushed(n);
6775 }
6776
6777 if (unlikely(work > weight))
6778 netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6779 n->poll, work, weight);
6780
6781 if (likely(work < weight))
6782 return work;
6783
6784 /* Drivers must not modify the NAPI state if they
6785 * consume the entire weight. In such cases this code
6786 * still "owns" the NAPI instance and therefore can
6787 * move the instance around on the list at-will.
6788 */
6789 if (unlikely(napi_disable_pending(n))) {
6790 napi_complete(n);
6791 return work;
6792 }
6793
6794 /* The NAPI context has more processing work, but busy-polling
6795 * is preferred. Exit early.
6796 */
6797 if (napi_prefer_busy_poll(n)) {
6798 if (napi_complete_done(n, work)) {
6799 /* If timeout is not set, we need to make sure
6800 * that the NAPI is re-scheduled.
6801 */
6802 napi_schedule(n);
6803 }
6804 return work;
6805 }
6806
6807 if (n->gro_bitmask) {
6808 /* flush too old packets
6809 * If HZ < 1000, flush all packets.
6810 */
6811 napi_gro_flush(n, HZ >= 1000);
6812 }
6813
6814 gro_normal_list(n);
6815
6816 /* Some drivers may have called napi_schedule
6817 * prior to exhausting their budget.
6818 */
6819 if (unlikely(!list_empty(&n->poll_list))) {
6820 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6821 n->dev ? n->dev->name : "backlog");
6822 return work;
6823 }
6824
6825 *repoll = true;
6826
6827 return work;
6828 }
6829
napi_poll(struct napi_struct * n,struct list_head * repoll)6830 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6831 {
6832 bool do_repoll = false;
6833 void *have;
6834 int work;
6835
6836 list_del_init(&n->poll_list);
6837
6838 have = netpoll_poll_lock(n);
6839
6840 work = __napi_poll(n, &do_repoll);
6841
6842 if (do_repoll)
6843 list_add_tail(&n->poll_list, repoll);
6844
6845 netpoll_poll_unlock(have);
6846
6847 return work;
6848 }
6849
napi_thread_wait(struct napi_struct * napi)6850 static int napi_thread_wait(struct napi_struct *napi)
6851 {
6852 set_current_state(TASK_INTERRUPTIBLE);
6853
6854 while (!kthread_should_stop()) {
6855 /* Testing SCHED_THREADED bit here to make sure the current
6856 * kthread owns this napi and could poll on this napi.
6857 * Testing SCHED bit is not enough because SCHED bit might be
6858 * set by some other busy poll thread or by napi_disable().
6859 */
6860 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) {
6861 WARN_ON(!list_empty(&napi->poll_list));
6862 __set_current_state(TASK_RUNNING);
6863 return 0;
6864 }
6865
6866 schedule();
6867 set_current_state(TASK_INTERRUPTIBLE);
6868 }
6869 __set_current_state(TASK_RUNNING);
6870
6871 return -1;
6872 }
6873
napi_threaded_poll_loop(struct napi_struct * napi)6874 static void napi_threaded_poll_loop(struct napi_struct *napi)
6875 {
6876 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6877 struct softnet_data *sd;
6878 unsigned long last_qs = jiffies;
6879
6880 for (;;) {
6881 bool repoll = false;
6882 void *have;
6883
6884 local_bh_disable();
6885 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6886
6887 sd = this_cpu_ptr(&softnet_data);
6888 sd->in_napi_threaded_poll = true;
6889
6890 have = netpoll_poll_lock(napi);
6891 __napi_poll(napi, &repoll);
6892 netpoll_poll_unlock(have);
6893
6894 sd->in_napi_threaded_poll = false;
6895 barrier();
6896
6897 if (sd_has_rps_ipi_waiting(sd)) {
6898 local_irq_disable();
6899 net_rps_action_and_irq_enable(sd);
6900 }
6901 skb_defer_free_flush(sd);
6902 bpf_net_ctx_clear(bpf_net_ctx);
6903 local_bh_enable();
6904
6905 if (!repoll)
6906 break;
6907
6908 rcu_softirq_qs_periodic(last_qs);
6909 cond_resched();
6910 }
6911 }
6912
napi_threaded_poll(void * data)6913 static int napi_threaded_poll(void *data)
6914 {
6915 struct napi_struct *napi = data;
6916
6917 while (!napi_thread_wait(napi))
6918 napi_threaded_poll_loop(napi);
6919
6920 return 0;
6921 }
6922
net_rx_action(void)6923 static __latent_entropy void net_rx_action(void)
6924 {
6925 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6926 unsigned long time_limit = jiffies +
6927 usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
6928 struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6929 int budget = READ_ONCE(net_hotdata.netdev_budget);
6930 LIST_HEAD(list);
6931 LIST_HEAD(repoll);
6932
6933 bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6934 start:
6935 sd->in_net_rx_action = true;
6936 local_irq_disable();
6937 list_splice_init(&sd->poll_list, &list);
6938 local_irq_enable();
6939
6940 for (;;) {
6941 struct napi_struct *n;
6942
6943 skb_defer_free_flush(sd);
6944
6945 if (list_empty(&list)) {
6946 if (list_empty(&repoll)) {
6947 sd->in_net_rx_action = false;
6948 barrier();
6949 /* We need to check if ____napi_schedule()
6950 * had refilled poll_list while
6951 * sd->in_net_rx_action was true.
6952 */
6953 if (!list_empty(&sd->poll_list))
6954 goto start;
6955 if (!sd_has_rps_ipi_waiting(sd))
6956 goto end;
6957 }
6958 break;
6959 }
6960
6961 n = list_first_entry(&list, struct napi_struct, poll_list);
6962 budget -= napi_poll(n, &repoll);
6963
6964 /* If softirq window is exhausted then punt.
6965 * Allow this to run for 2 jiffies since which will allow
6966 * an average latency of 1.5/HZ.
6967 */
6968 if (unlikely(budget <= 0 ||
6969 time_after_eq(jiffies, time_limit))) {
6970 sd->time_squeeze++;
6971 break;
6972 }
6973 }
6974
6975 local_irq_disable();
6976
6977 list_splice_tail_init(&sd->poll_list, &list);
6978 list_splice_tail(&repoll, &list);
6979 list_splice(&list, &sd->poll_list);
6980 if (!list_empty(&sd->poll_list))
6981 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6982 else
6983 sd->in_net_rx_action = false;
6984
6985 net_rps_action_and_irq_enable(sd);
6986 end:
6987 bpf_net_ctx_clear(bpf_net_ctx);
6988 }
6989
6990 struct netdev_adjacent {
6991 struct net_device *dev;
6992 netdevice_tracker dev_tracker;
6993
6994 /* upper master flag, there can only be one master device per list */
6995 bool master;
6996
6997 /* lookup ignore flag */
6998 bool ignore;
6999
7000 /* counter for the number of times this device was added to us */
7001 u16 ref_nr;
7002
7003 /* private field for the users */
7004 void *private;
7005
7006 struct list_head list;
7007 struct rcu_head rcu;
7008 };
7009
__netdev_find_adj(struct net_device * adj_dev,struct list_head * adj_list)7010 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7011 struct list_head *adj_list)
7012 {
7013 struct netdev_adjacent *adj;
7014
7015 list_for_each_entry(adj, adj_list, list) {
7016 if (adj->dev == adj_dev)
7017 return adj;
7018 }
7019 return NULL;
7020 }
7021
____netdev_has_upper_dev(struct net_device * upper_dev,struct netdev_nested_priv * priv)7022 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7023 struct netdev_nested_priv *priv)
7024 {
7025 struct net_device *dev = (struct net_device *)priv->data;
7026
7027 return upper_dev == dev;
7028 }
7029
7030 /**
7031 * netdev_has_upper_dev - Check if device is linked to an upper device
7032 * @dev: device
7033 * @upper_dev: upper device to check
7034 *
7035 * Find out if a device is linked to specified upper device and return true
7036 * in case it is. Note that this checks only immediate upper device,
7037 * not through a complete stack of devices. The caller must hold the RTNL lock.
7038 */
netdev_has_upper_dev(struct net_device * dev,struct net_device * upper_dev)7039 bool netdev_has_upper_dev(struct net_device *dev,
7040 struct net_device *upper_dev)
7041 {
7042 struct netdev_nested_priv priv = {
7043 .data = (void *)upper_dev,
7044 };
7045
7046 ASSERT_RTNL();
7047
7048 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7049 &priv);
7050 }
7051 EXPORT_SYMBOL(netdev_has_upper_dev);
7052
7053 /**
7054 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7055 * @dev: device
7056 * @upper_dev: upper device to check
7057 *
7058 * Find out if a device is linked to specified upper device and return true
7059 * in case it is. Note that this checks the entire upper device chain.
7060 * The caller must hold rcu lock.
7061 */
7062
netdev_has_upper_dev_all_rcu(struct net_device * dev,struct net_device * upper_dev)7063 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7064 struct net_device *upper_dev)
7065 {
7066 struct netdev_nested_priv priv = {
7067 .data = (void *)upper_dev,
7068 };
7069
7070 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7071 &priv);
7072 }
7073 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7074
7075 /**
7076 * netdev_has_any_upper_dev - Check if device is linked to some device
7077 * @dev: device
7078 *
7079 * Find out if a device is linked to an upper device and return true in case
7080 * it is. The caller must hold the RTNL lock.
7081 */
netdev_has_any_upper_dev(struct net_device * dev)7082 bool netdev_has_any_upper_dev(struct net_device *dev)
7083 {
7084 ASSERT_RTNL();
7085
7086 return !list_empty(&dev->adj_list.upper);
7087 }
7088 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7089
7090 /**
7091 * netdev_master_upper_dev_get - Get master upper device
7092 * @dev: device
7093 *
7094 * Find a master upper device and return pointer to it or NULL in case
7095 * it's not there. The caller must hold the RTNL lock.
7096 */
netdev_master_upper_dev_get(struct net_device * dev)7097 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7098 {
7099 struct netdev_adjacent *upper;
7100
7101 ASSERT_RTNL();
7102
7103 if (list_empty(&dev->adj_list.upper))
7104 return NULL;
7105
7106 upper = list_first_entry(&dev->adj_list.upper,
7107 struct netdev_adjacent, list);
7108 if (likely(upper->master))
7109 return upper->dev;
7110 return NULL;
7111 }
7112 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7113
__netdev_master_upper_dev_get(struct net_device * dev)7114 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7115 {
7116 struct netdev_adjacent *upper;
7117
7118 ASSERT_RTNL();
7119
7120 if (list_empty(&dev->adj_list.upper))
7121 return NULL;
7122
7123 upper = list_first_entry(&dev->adj_list.upper,
7124 struct netdev_adjacent, list);
7125 if (likely(upper->master) && !upper->ignore)
7126 return upper->dev;
7127 return NULL;
7128 }
7129
7130 /**
7131 * netdev_has_any_lower_dev - Check if device is linked to some device
7132 * @dev: device
7133 *
7134 * Find out if a device is linked to a lower device and return true in case
7135 * it is. The caller must hold the RTNL lock.
7136 */
netdev_has_any_lower_dev(struct net_device * dev)7137 static bool netdev_has_any_lower_dev(struct net_device *dev)
7138 {
7139 ASSERT_RTNL();
7140
7141 return !list_empty(&dev->adj_list.lower);
7142 }
7143
netdev_adjacent_get_private(struct list_head * adj_list)7144 void *netdev_adjacent_get_private(struct list_head *adj_list)
7145 {
7146 struct netdev_adjacent *adj;
7147
7148 adj = list_entry(adj_list, struct netdev_adjacent, list);
7149
7150 return adj->private;
7151 }
7152 EXPORT_SYMBOL(netdev_adjacent_get_private);
7153
7154 /**
7155 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7156 * @dev: device
7157 * @iter: list_head ** of the current position
7158 *
7159 * Gets the next device from the dev's upper list, starting from iter
7160 * position. The caller must hold RCU read lock.
7161 */
netdev_upper_get_next_dev_rcu(struct net_device * dev,struct list_head ** iter)7162 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7163 struct list_head **iter)
7164 {
7165 struct netdev_adjacent *upper;
7166
7167 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7168
7169 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7170
7171 if (&upper->list == &dev->adj_list.upper)
7172 return NULL;
7173
7174 *iter = &upper->list;
7175
7176 return upper->dev;
7177 }
7178 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7179
__netdev_next_upper_dev(struct net_device * dev,struct list_head ** iter,bool * ignore)7180 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7181 struct list_head **iter,
7182 bool *ignore)
7183 {
7184 struct netdev_adjacent *upper;
7185
7186 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7187
7188 if (&upper->list == &dev->adj_list.upper)
7189 return NULL;
7190
7191 *iter = &upper->list;
7192 *ignore = upper->ignore;
7193
7194 return upper->dev;
7195 }
7196
netdev_next_upper_dev_rcu(struct net_device * dev,struct list_head ** iter)7197 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7198 struct list_head **iter)
7199 {
7200 struct netdev_adjacent *upper;
7201
7202 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7203
7204 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7205
7206 if (&upper->list == &dev->adj_list.upper)
7207 return NULL;
7208
7209 *iter = &upper->list;
7210
7211 return upper->dev;
7212 }
7213
__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)7214 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7215 int (*fn)(struct net_device *dev,
7216 struct netdev_nested_priv *priv),
7217 struct netdev_nested_priv *priv)
7218 {
7219 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7220 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7221 int ret, cur = 0;
7222 bool ignore;
7223
7224 now = dev;
7225 iter = &dev->adj_list.upper;
7226
7227 while (1) {
7228 if (now != dev) {
7229 ret = fn(now, priv);
7230 if (ret)
7231 return ret;
7232 }
7233
7234 next = NULL;
7235 while (1) {
7236 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7237 if (!udev)
7238 break;
7239 if (ignore)
7240 continue;
7241
7242 next = udev;
7243 niter = &udev->adj_list.upper;
7244 dev_stack[cur] = now;
7245 iter_stack[cur++] = iter;
7246 break;
7247 }
7248
7249 if (!next) {
7250 if (!cur)
7251 return 0;
7252 next = dev_stack[--cur];
7253 niter = iter_stack[cur];
7254 }
7255
7256 now = next;
7257 iter = niter;
7258 }
7259
7260 return 0;
7261 }
7262
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)7263 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7264 int (*fn)(struct net_device *dev,
7265 struct netdev_nested_priv *priv),
7266 struct netdev_nested_priv *priv)
7267 {
7268 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7269 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7270 int ret, cur = 0;
7271
7272 now = dev;
7273 iter = &dev->adj_list.upper;
7274
7275 while (1) {
7276 if (now != dev) {
7277 ret = fn(now, priv);
7278 if (ret)
7279 return ret;
7280 }
7281
7282 next = NULL;
7283 while (1) {
7284 udev = netdev_next_upper_dev_rcu(now, &iter);
7285 if (!udev)
7286 break;
7287
7288 next = udev;
7289 niter = &udev->adj_list.upper;
7290 dev_stack[cur] = now;
7291 iter_stack[cur++] = iter;
7292 break;
7293 }
7294
7295 if (!next) {
7296 if (!cur)
7297 return 0;
7298 next = dev_stack[--cur];
7299 niter = iter_stack[cur];
7300 }
7301
7302 now = next;
7303 iter = niter;
7304 }
7305
7306 return 0;
7307 }
7308 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7309
__netdev_has_upper_dev(struct net_device * dev,struct net_device * upper_dev)7310 static bool __netdev_has_upper_dev(struct net_device *dev,
7311 struct net_device *upper_dev)
7312 {
7313 struct netdev_nested_priv priv = {
7314 .flags = 0,
7315 .data = (void *)upper_dev,
7316 };
7317
7318 ASSERT_RTNL();
7319
7320 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7321 &priv);
7322 }
7323
7324 /**
7325 * netdev_lower_get_next_private - Get the next ->private from the
7326 * lower neighbour list
7327 * @dev: device
7328 * @iter: list_head ** of the current position
7329 *
7330 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7331 * list, starting from iter position. The caller must hold either hold the
7332 * RTNL lock or its own locking that guarantees that the neighbour lower
7333 * list will remain unchanged.
7334 */
netdev_lower_get_next_private(struct net_device * dev,struct list_head ** iter)7335 void *netdev_lower_get_next_private(struct net_device *dev,
7336 struct list_head **iter)
7337 {
7338 struct netdev_adjacent *lower;
7339
7340 lower = list_entry(*iter, struct netdev_adjacent, list);
7341
7342 if (&lower->list == &dev->adj_list.lower)
7343 return NULL;
7344
7345 *iter = lower->list.next;
7346
7347 return lower->private;
7348 }
7349 EXPORT_SYMBOL(netdev_lower_get_next_private);
7350
7351 /**
7352 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7353 * lower neighbour list, RCU
7354 * variant
7355 * @dev: device
7356 * @iter: list_head ** of the current position
7357 *
7358 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7359 * list, starting from iter position. The caller must hold RCU read lock.
7360 */
netdev_lower_get_next_private_rcu(struct net_device * dev,struct list_head ** iter)7361 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7362 struct list_head **iter)
7363 {
7364 struct netdev_adjacent *lower;
7365
7366 WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7367
7368 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7369
7370 if (&lower->list == &dev->adj_list.lower)
7371 return NULL;
7372
7373 *iter = &lower->list;
7374
7375 return lower->private;
7376 }
7377 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7378
7379 /**
7380 * netdev_lower_get_next - Get the next device from the lower neighbour
7381 * list
7382 * @dev: device
7383 * @iter: list_head ** of the current position
7384 *
7385 * Gets the next netdev_adjacent from the dev's lower neighbour
7386 * list, starting from iter position. The caller must hold RTNL lock or
7387 * its own locking that guarantees that the neighbour lower
7388 * list will remain unchanged.
7389 */
netdev_lower_get_next(struct net_device * dev,struct list_head ** iter)7390 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7391 {
7392 struct netdev_adjacent *lower;
7393
7394 lower = list_entry(*iter, struct netdev_adjacent, list);
7395
7396 if (&lower->list == &dev->adj_list.lower)
7397 return NULL;
7398
7399 *iter = lower->list.next;
7400
7401 return lower->dev;
7402 }
7403 EXPORT_SYMBOL(netdev_lower_get_next);
7404
netdev_next_lower_dev(struct net_device * dev,struct list_head ** iter)7405 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7406 struct list_head **iter)
7407 {
7408 struct netdev_adjacent *lower;
7409
7410 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7411
7412 if (&lower->list == &dev->adj_list.lower)
7413 return NULL;
7414
7415 *iter = &lower->list;
7416
7417 return lower->dev;
7418 }
7419
__netdev_next_lower_dev(struct net_device * dev,struct list_head ** iter,bool * ignore)7420 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7421 struct list_head **iter,
7422 bool *ignore)
7423 {
7424 struct netdev_adjacent *lower;
7425
7426 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7427
7428 if (&lower->list == &dev->adj_list.lower)
7429 return NULL;
7430
7431 *iter = &lower->list;
7432 *ignore = lower->ignore;
7433
7434 return lower->dev;
7435 }
7436
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)7437 int netdev_walk_all_lower_dev(struct net_device *dev,
7438 int (*fn)(struct net_device *dev,
7439 struct netdev_nested_priv *priv),
7440 struct netdev_nested_priv *priv)
7441 {
7442 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7443 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7444 int ret, cur = 0;
7445
7446 now = dev;
7447 iter = &dev->adj_list.lower;
7448
7449 while (1) {
7450 if (now != dev) {
7451 ret = fn(now, priv);
7452 if (ret)
7453 return ret;
7454 }
7455
7456 next = NULL;
7457 while (1) {
7458 ldev = netdev_next_lower_dev(now, &iter);
7459 if (!ldev)
7460 break;
7461
7462 next = ldev;
7463 niter = &ldev->adj_list.lower;
7464 dev_stack[cur] = now;
7465 iter_stack[cur++] = iter;
7466 break;
7467 }
7468
7469 if (!next) {
7470 if (!cur)
7471 return 0;
7472 next = dev_stack[--cur];
7473 niter = iter_stack[cur];
7474 }
7475
7476 now = next;
7477 iter = niter;
7478 }
7479
7480 return 0;
7481 }
7482 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7483
__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)7484 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7485 int (*fn)(struct net_device *dev,
7486 struct netdev_nested_priv *priv),
7487 struct netdev_nested_priv *priv)
7488 {
7489 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7490 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7491 int ret, cur = 0;
7492 bool ignore;
7493
7494 now = dev;
7495 iter = &dev->adj_list.lower;
7496
7497 while (1) {
7498 if (now != dev) {
7499 ret = fn(now, priv);
7500 if (ret)
7501 return ret;
7502 }
7503
7504 next = NULL;
7505 while (1) {
7506 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7507 if (!ldev)
7508 break;
7509 if (ignore)
7510 continue;
7511
7512 next = ldev;
7513 niter = &ldev->adj_list.lower;
7514 dev_stack[cur] = now;
7515 iter_stack[cur++] = iter;
7516 break;
7517 }
7518
7519 if (!next) {
7520 if (!cur)
7521 return 0;
7522 next = dev_stack[--cur];
7523 niter = iter_stack[cur];
7524 }
7525
7526 now = next;
7527 iter = niter;
7528 }
7529
7530 return 0;
7531 }
7532
netdev_next_lower_dev_rcu(struct net_device * dev,struct list_head ** iter)7533 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7534 struct list_head **iter)
7535 {
7536 struct netdev_adjacent *lower;
7537
7538 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7539 if (&lower->list == &dev->adj_list.lower)
7540 return NULL;
7541
7542 *iter = &lower->list;
7543
7544 return lower->dev;
7545 }
7546 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7547
__netdev_upper_depth(struct net_device * dev)7548 static u8 __netdev_upper_depth(struct net_device *dev)
7549 {
7550 struct net_device *udev;
7551 struct list_head *iter;
7552 u8 max_depth = 0;
7553 bool ignore;
7554
7555 for (iter = &dev->adj_list.upper,
7556 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7557 udev;
7558 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7559 if (ignore)
7560 continue;
7561 if (max_depth < udev->upper_level)
7562 max_depth = udev->upper_level;
7563 }
7564
7565 return max_depth;
7566 }
7567
__netdev_lower_depth(struct net_device * dev)7568 static u8 __netdev_lower_depth(struct net_device *dev)
7569 {
7570 struct net_device *ldev;
7571 struct list_head *iter;
7572 u8 max_depth = 0;
7573 bool ignore;
7574
7575 for (iter = &dev->adj_list.lower,
7576 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7577 ldev;
7578 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7579 if (ignore)
7580 continue;
7581 if (max_depth < ldev->lower_level)
7582 max_depth = ldev->lower_level;
7583 }
7584
7585 return max_depth;
7586 }
7587
__netdev_update_upper_level(struct net_device * dev,struct netdev_nested_priv * __unused)7588 static int __netdev_update_upper_level(struct net_device *dev,
7589 struct netdev_nested_priv *__unused)
7590 {
7591 dev->upper_level = __netdev_upper_depth(dev) + 1;
7592 return 0;
7593 }
7594
7595 #ifdef CONFIG_LOCKDEP
7596 static LIST_HEAD(net_unlink_list);
7597
net_unlink_todo(struct net_device * dev)7598 static void net_unlink_todo(struct net_device *dev)
7599 {
7600 if (list_empty(&dev->unlink_list))
7601 list_add_tail(&dev->unlink_list, &net_unlink_list);
7602 }
7603 #endif
7604
__netdev_update_lower_level(struct net_device * dev,struct netdev_nested_priv * priv)7605 static int __netdev_update_lower_level(struct net_device *dev,
7606 struct netdev_nested_priv *priv)
7607 {
7608 dev->lower_level = __netdev_lower_depth(dev) + 1;
7609
7610 #ifdef CONFIG_LOCKDEP
7611 if (!priv)
7612 return 0;
7613
7614 if (priv->flags & NESTED_SYNC_IMM)
7615 dev->nested_level = dev->lower_level - 1;
7616 if (priv->flags & NESTED_SYNC_TODO)
7617 net_unlink_todo(dev);
7618 #endif
7619 return 0;
7620 }
7621
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)7622 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7623 int (*fn)(struct net_device *dev,
7624 struct netdev_nested_priv *priv),
7625 struct netdev_nested_priv *priv)
7626 {
7627 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7628 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7629 int ret, cur = 0;
7630
7631 now = dev;
7632 iter = &dev->adj_list.lower;
7633
7634 while (1) {
7635 if (now != dev) {
7636 ret = fn(now, priv);
7637 if (ret)
7638 return ret;
7639 }
7640
7641 next = NULL;
7642 while (1) {
7643 ldev = netdev_next_lower_dev_rcu(now, &iter);
7644 if (!ldev)
7645 break;
7646
7647 next = ldev;
7648 niter = &ldev->adj_list.lower;
7649 dev_stack[cur] = now;
7650 iter_stack[cur++] = iter;
7651 break;
7652 }
7653
7654 if (!next) {
7655 if (!cur)
7656 return 0;
7657 next = dev_stack[--cur];
7658 niter = iter_stack[cur];
7659 }
7660
7661 now = next;
7662 iter = niter;
7663 }
7664
7665 return 0;
7666 }
7667 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7668
7669 /**
7670 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7671 * lower neighbour list, RCU
7672 * variant
7673 * @dev: device
7674 *
7675 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7676 * list. The caller must hold RCU read lock.
7677 */
netdev_lower_get_first_private_rcu(struct net_device * dev)7678 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7679 {
7680 struct netdev_adjacent *lower;
7681
7682 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7683 struct netdev_adjacent, list);
7684 if (lower)
7685 return lower->private;
7686 return NULL;
7687 }
7688 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7689
7690 /**
7691 * netdev_master_upper_dev_get_rcu - Get master upper device
7692 * @dev: device
7693 *
7694 * Find a master upper device and return pointer to it or NULL in case
7695 * it's not there. The caller must hold the RCU read lock.
7696 */
netdev_master_upper_dev_get_rcu(struct net_device * dev)7697 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7698 {
7699 struct netdev_adjacent *upper;
7700
7701 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7702 struct netdev_adjacent, list);
7703 if (upper && likely(upper->master))
7704 return upper->dev;
7705 return NULL;
7706 }
7707 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7708
netdev_adjacent_sysfs_add(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)7709 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7710 struct net_device *adj_dev,
7711 struct list_head *dev_list)
7712 {
7713 char linkname[IFNAMSIZ+7];
7714
7715 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7716 "upper_%s" : "lower_%s", adj_dev->name);
7717 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7718 linkname);
7719 }
netdev_adjacent_sysfs_del(struct net_device * dev,char * name,struct list_head * dev_list)7720 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7721 char *name,
7722 struct list_head *dev_list)
7723 {
7724 char linkname[IFNAMSIZ+7];
7725
7726 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7727 "upper_%s" : "lower_%s", name);
7728 sysfs_remove_link(&(dev->dev.kobj), linkname);
7729 }
7730
netdev_adjacent_is_neigh_list(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)7731 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7732 struct net_device *adj_dev,
7733 struct list_head *dev_list)
7734 {
7735 return (dev_list == &dev->adj_list.upper ||
7736 dev_list == &dev->adj_list.lower) &&
7737 net_eq(dev_net(dev), dev_net(adj_dev));
7738 }
7739
__netdev_adjacent_dev_insert(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list,void * private,bool master)7740 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7741 struct net_device *adj_dev,
7742 struct list_head *dev_list,
7743 void *private, bool master)
7744 {
7745 struct netdev_adjacent *adj;
7746 int ret;
7747
7748 adj = __netdev_find_adj(adj_dev, dev_list);
7749
7750 if (adj) {
7751 adj->ref_nr += 1;
7752 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7753 dev->name, adj_dev->name, adj->ref_nr);
7754
7755 return 0;
7756 }
7757
7758 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7759 if (!adj)
7760 return -ENOMEM;
7761
7762 adj->dev = adj_dev;
7763 adj->master = master;
7764 adj->ref_nr = 1;
7765 adj->private = private;
7766 adj->ignore = false;
7767 netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7768
7769 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7770 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7771
7772 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7773 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7774 if (ret)
7775 goto free_adj;
7776 }
7777
7778 /* Ensure that master link is always the first item in list. */
7779 if (master) {
7780 ret = sysfs_create_link(&(dev->dev.kobj),
7781 &(adj_dev->dev.kobj), "master");
7782 if (ret)
7783 goto remove_symlinks;
7784
7785 list_add_rcu(&adj->list, dev_list);
7786 } else {
7787 list_add_tail_rcu(&adj->list, dev_list);
7788 }
7789
7790 return 0;
7791
7792 remove_symlinks:
7793 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7794 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7795 free_adj:
7796 netdev_put(adj_dev, &adj->dev_tracker);
7797 kfree(adj);
7798
7799 return ret;
7800 }
7801
__netdev_adjacent_dev_remove(struct net_device * dev,struct net_device * adj_dev,u16 ref_nr,struct list_head * dev_list)7802 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7803 struct net_device *adj_dev,
7804 u16 ref_nr,
7805 struct list_head *dev_list)
7806 {
7807 struct netdev_adjacent *adj;
7808
7809 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7810 dev->name, adj_dev->name, ref_nr);
7811
7812 adj = __netdev_find_adj(adj_dev, dev_list);
7813
7814 if (!adj) {
7815 pr_err("Adjacency does not exist for device %s from %s\n",
7816 dev->name, adj_dev->name);
7817 WARN_ON(1);
7818 return;
7819 }
7820
7821 if (adj->ref_nr > ref_nr) {
7822 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7823 dev->name, adj_dev->name, ref_nr,
7824 adj->ref_nr - ref_nr);
7825 adj->ref_nr -= ref_nr;
7826 return;
7827 }
7828
7829 if (adj->master)
7830 sysfs_remove_link(&(dev->dev.kobj), "master");
7831
7832 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7833 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7834
7835 list_del_rcu(&adj->list);
7836 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7837 adj_dev->name, dev->name, adj_dev->name);
7838 netdev_put(adj_dev, &adj->dev_tracker);
7839 kfree_rcu(adj, rcu);
7840 }
7841
__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)7842 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7843 struct net_device *upper_dev,
7844 struct list_head *up_list,
7845 struct list_head *down_list,
7846 void *private, bool master)
7847 {
7848 int ret;
7849
7850 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7851 private, master);
7852 if (ret)
7853 return ret;
7854
7855 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7856 private, false);
7857 if (ret) {
7858 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7859 return ret;
7860 }
7861
7862 return 0;
7863 }
7864
__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)7865 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7866 struct net_device *upper_dev,
7867 u16 ref_nr,
7868 struct list_head *up_list,
7869 struct list_head *down_list)
7870 {
7871 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7872 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7873 }
7874
__netdev_adjacent_dev_link_neighbour(struct net_device * dev,struct net_device * upper_dev,void * private,bool master)7875 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7876 struct net_device *upper_dev,
7877 void *private, bool master)
7878 {
7879 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7880 &dev->adj_list.upper,
7881 &upper_dev->adj_list.lower,
7882 private, master);
7883 }
7884
__netdev_adjacent_dev_unlink_neighbour(struct net_device * dev,struct net_device * upper_dev)7885 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7886 struct net_device *upper_dev)
7887 {
7888 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7889 &dev->adj_list.upper,
7890 &upper_dev->adj_list.lower);
7891 }
7892
__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)7893 static int __netdev_upper_dev_link(struct net_device *dev,
7894 struct net_device *upper_dev, bool master,
7895 void *upper_priv, void *upper_info,
7896 struct netdev_nested_priv *priv,
7897 struct netlink_ext_ack *extack)
7898 {
7899 struct netdev_notifier_changeupper_info changeupper_info = {
7900 .info = {
7901 .dev = dev,
7902 .extack = extack,
7903 },
7904 .upper_dev = upper_dev,
7905 .master = master,
7906 .linking = true,
7907 .upper_info = upper_info,
7908 };
7909 struct net_device *master_dev;
7910 int ret = 0;
7911
7912 ASSERT_RTNL();
7913
7914 if (dev == upper_dev)
7915 return -EBUSY;
7916
7917 /* To prevent loops, check if dev is not upper device to upper_dev. */
7918 if (__netdev_has_upper_dev(upper_dev, dev))
7919 return -EBUSY;
7920
7921 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7922 return -EMLINK;
7923
7924 if (!master) {
7925 if (__netdev_has_upper_dev(dev, upper_dev))
7926 return -EEXIST;
7927 } else {
7928 master_dev = __netdev_master_upper_dev_get(dev);
7929 if (master_dev)
7930 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7931 }
7932
7933 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7934 &changeupper_info.info);
7935 ret = notifier_to_errno(ret);
7936 if (ret)
7937 return ret;
7938
7939 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7940 master);
7941 if (ret)
7942 return ret;
7943
7944 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7945 &changeupper_info.info);
7946 ret = notifier_to_errno(ret);
7947 if (ret)
7948 goto rollback;
7949
7950 __netdev_update_upper_level(dev, NULL);
7951 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7952
7953 __netdev_update_lower_level(upper_dev, priv);
7954 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7955 priv);
7956
7957 return 0;
7958
7959 rollback:
7960 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7961
7962 return ret;
7963 }
7964
7965 /**
7966 * netdev_upper_dev_link - Add a link to the upper device
7967 * @dev: device
7968 * @upper_dev: new upper device
7969 * @extack: netlink extended ack
7970 *
7971 * Adds a link to device which is upper to this one. The caller must hold
7972 * the RTNL lock. On a failure a negative errno code is returned.
7973 * On success the reference counts are adjusted and the function
7974 * returns zero.
7975 */
netdev_upper_dev_link(struct net_device * dev,struct net_device * upper_dev,struct netlink_ext_ack * extack)7976 int netdev_upper_dev_link(struct net_device *dev,
7977 struct net_device *upper_dev,
7978 struct netlink_ext_ack *extack)
7979 {
7980 struct netdev_nested_priv priv = {
7981 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7982 .data = NULL,
7983 };
7984
7985 return __netdev_upper_dev_link(dev, upper_dev, false,
7986 NULL, NULL, &priv, extack);
7987 }
7988 EXPORT_SYMBOL(netdev_upper_dev_link);
7989
7990 /**
7991 * netdev_master_upper_dev_link - Add a master link to the upper device
7992 * @dev: device
7993 * @upper_dev: new upper device
7994 * @upper_priv: upper device private
7995 * @upper_info: upper info to be passed down via notifier
7996 * @extack: netlink extended ack
7997 *
7998 * Adds a link to device which is upper to this one. In this case, only
7999 * one master upper device can be linked, although other non-master devices
8000 * might be linked as well. The caller must hold the RTNL lock.
8001 * On a failure a negative errno code is returned. On success the reference
8002 * counts are adjusted and the function returns zero.
8003 */
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)8004 int netdev_master_upper_dev_link(struct net_device *dev,
8005 struct net_device *upper_dev,
8006 void *upper_priv, void *upper_info,
8007 struct netlink_ext_ack *extack)
8008 {
8009 struct netdev_nested_priv priv = {
8010 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8011 .data = NULL,
8012 };
8013
8014 return __netdev_upper_dev_link(dev, upper_dev, true,
8015 upper_priv, upper_info, &priv, extack);
8016 }
8017 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8018
__netdev_upper_dev_unlink(struct net_device * dev,struct net_device * upper_dev,struct netdev_nested_priv * priv)8019 static void __netdev_upper_dev_unlink(struct net_device *dev,
8020 struct net_device *upper_dev,
8021 struct netdev_nested_priv *priv)
8022 {
8023 struct netdev_notifier_changeupper_info changeupper_info = {
8024 .info = {
8025 .dev = dev,
8026 },
8027 .upper_dev = upper_dev,
8028 .linking = false,
8029 };
8030
8031 ASSERT_RTNL();
8032
8033 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8034
8035 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8036 &changeupper_info.info);
8037
8038 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8039
8040 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8041 &changeupper_info.info);
8042
8043 __netdev_update_upper_level(dev, NULL);
8044 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8045
8046 __netdev_update_lower_level(upper_dev, priv);
8047 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8048 priv);
8049 }
8050
8051 /**
8052 * netdev_upper_dev_unlink - Removes a link to upper device
8053 * @dev: device
8054 * @upper_dev: new upper device
8055 *
8056 * Removes a link to device which is upper to this one. The caller must hold
8057 * the RTNL lock.
8058 */
netdev_upper_dev_unlink(struct net_device * dev,struct net_device * upper_dev)8059 void netdev_upper_dev_unlink(struct net_device *dev,
8060 struct net_device *upper_dev)
8061 {
8062 struct netdev_nested_priv priv = {
8063 .flags = NESTED_SYNC_TODO,
8064 .data = NULL,
8065 };
8066
8067 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8068 }
8069 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8070
__netdev_adjacent_dev_set(struct net_device * upper_dev,struct net_device * lower_dev,bool val)8071 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8072 struct net_device *lower_dev,
8073 bool val)
8074 {
8075 struct netdev_adjacent *adj;
8076
8077 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8078 if (adj)
8079 adj->ignore = val;
8080
8081 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8082 if (adj)
8083 adj->ignore = val;
8084 }
8085
netdev_adjacent_dev_disable(struct net_device * upper_dev,struct net_device * lower_dev)8086 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8087 struct net_device *lower_dev)
8088 {
8089 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8090 }
8091
netdev_adjacent_dev_enable(struct net_device * upper_dev,struct net_device * lower_dev)8092 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8093 struct net_device *lower_dev)
8094 {
8095 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8096 }
8097
netdev_adjacent_change_prepare(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev,struct netlink_ext_ack * extack)8098 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8099 struct net_device *new_dev,
8100 struct net_device *dev,
8101 struct netlink_ext_ack *extack)
8102 {
8103 struct netdev_nested_priv priv = {
8104 .flags = 0,
8105 .data = NULL,
8106 };
8107 int err;
8108
8109 if (!new_dev)
8110 return 0;
8111
8112 if (old_dev && new_dev != old_dev)
8113 netdev_adjacent_dev_disable(dev, old_dev);
8114 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8115 extack);
8116 if (err) {
8117 if (old_dev && new_dev != old_dev)
8118 netdev_adjacent_dev_enable(dev, old_dev);
8119 return err;
8120 }
8121
8122 return 0;
8123 }
8124 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8125
netdev_adjacent_change_commit(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev)8126 void netdev_adjacent_change_commit(struct net_device *old_dev,
8127 struct net_device *new_dev,
8128 struct net_device *dev)
8129 {
8130 struct netdev_nested_priv priv = {
8131 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8132 .data = NULL,
8133 };
8134
8135 if (!new_dev || !old_dev)
8136 return;
8137
8138 if (new_dev == old_dev)
8139 return;
8140
8141 netdev_adjacent_dev_enable(dev, old_dev);
8142 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8143 }
8144 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8145
netdev_adjacent_change_abort(struct net_device * old_dev,struct net_device * new_dev,struct net_device * dev)8146 void netdev_adjacent_change_abort(struct net_device *old_dev,
8147 struct net_device *new_dev,
8148 struct net_device *dev)
8149 {
8150 struct netdev_nested_priv priv = {
8151 .flags = 0,
8152 .data = NULL,
8153 };
8154
8155 if (!new_dev)
8156 return;
8157
8158 if (old_dev && new_dev != old_dev)
8159 netdev_adjacent_dev_enable(dev, old_dev);
8160
8161 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8162 }
8163 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8164
8165 /**
8166 * netdev_bonding_info_change - Dispatch event about slave change
8167 * @dev: device
8168 * @bonding_info: info to dispatch
8169 *
8170 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8171 * The caller must hold the RTNL lock.
8172 */
netdev_bonding_info_change(struct net_device * dev,struct netdev_bonding_info * bonding_info)8173 void netdev_bonding_info_change(struct net_device *dev,
8174 struct netdev_bonding_info *bonding_info)
8175 {
8176 struct netdev_notifier_bonding_info info = {
8177 .info.dev = dev,
8178 };
8179
8180 memcpy(&info.bonding_info, bonding_info,
8181 sizeof(struct netdev_bonding_info));
8182 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8183 &info.info);
8184 }
8185 EXPORT_SYMBOL(netdev_bonding_info_change);
8186
netdev_offload_xstats_enable_l3(struct net_device * dev,struct netlink_ext_ack * extack)8187 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
8188 struct netlink_ext_ack *extack)
8189 {
8190 struct netdev_notifier_offload_xstats_info info = {
8191 .info.dev = dev,
8192 .info.extack = extack,
8193 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8194 };
8195 int err;
8196 int rc;
8197
8198 dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
8199 GFP_KERNEL);
8200 if (!dev->offload_xstats_l3)
8201 return -ENOMEM;
8202
8203 rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
8204 NETDEV_OFFLOAD_XSTATS_DISABLE,
8205 &info.info);
8206 err = notifier_to_errno(rc);
8207 if (err)
8208 goto free_stats;
8209
8210 return 0;
8211
8212 free_stats:
8213 kfree(dev->offload_xstats_l3);
8214 dev->offload_xstats_l3 = NULL;
8215 return err;
8216 }
8217
netdev_offload_xstats_enable(struct net_device * dev,enum netdev_offload_xstats_type type,struct netlink_ext_ack * extack)8218 int netdev_offload_xstats_enable(struct net_device *dev,
8219 enum netdev_offload_xstats_type type,
8220 struct netlink_ext_ack *extack)
8221 {
8222 ASSERT_RTNL();
8223
8224 if (netdev_offload_xstats_enabled(dev, type))
8225 return -EALREADY;
8226
8227 switch (type) {
8228 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8229 return netdev_offload_xstats_enable_l3(dev, extack);
8230 }
8231
8232 WARN_ON(1);
8233 return -EINVAL;
8234 }
8235 EXPORT_SYMBOL(netdev_offload_xstats_enable);
8236
netdev_offload_xstats_disable_l3(struct net_device * dev)8237 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
8238 {
8239 struct netdev_notifier_offload_xstats_info info = {
8240 .info.dev = dev,
8241 .type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8242 };
8243
8244 call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
8245 &info.info);
8246 kfree(dev->offload_xstats_l3);
8247 dev->offload_xstats_l3 = NULL;
8248 }
8249
netdev_offload_xstats_disable(struct net_device * dev,enum netdev_offload_xstats_type type)8250 int netdev_offload_xstats_disable(struct net_device *dev,
8251 enum netdev_offload_xstats_type type)
8252 {
8253 ASSERT_RTNL();
8254
8255 if (!netdev_offload_xstats_enabled(dev, type))
8256 return -EALREADY;
8257
8258 switch (type) {
8259 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8260 netdev_offload_xstats_disable_l3(dev);
8261 return 0;
8262 }
8263
8264 WARN_ON(1);
8265 return -EINVAL;
8266 }
8267 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8268
netdev_offload_xstats_disable_all(struct net_device * dev)8269 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8270 {
8271 netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8272 }
8273
8274 static struct rtnl_hw_stats64 *
netdev_offload_xstats_get_ptr(const struct net_device * dev,enum netdev_offload_xstats_type type)8275 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8276 enum netdev_offload_xstats_type type)
8277 {
8278 switch (type) {
8279 case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8280 return dev->offload_xstats_l3;
8281 }
8282
8283 WARN_ON(1);
8284 return NULL;
8285 }
8286
netdev_offload_xstats_enabled(const struct net_device * dev,enum netdev_offload_xstats_type type)8287 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8288 enum netdev_offload_xstats_type type)
8289 {
8290 ASSERT_RTNL();
8291
8292 return netdev_offload_xstats_get_ptr(dev, type);
8293 }
8294 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8295
8296 struct netdev_notifier_offload_xstats_ru {
8297 bool used;
8298 };
8299
8300 struct netdev_notifier_offload_xstats_rd {
8301 struct rtnl_hw_stats64 stats;
8302 bool used;
8303 };
8304
netdev_hw_stats64_add(struct rtnl_hw_stats64 * dest,const struct rtnl_hw_stats64 * src)8305 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8306 const struct rtnl_hw_stats64 *src)
8307 {
8308 dest->rx_packets += src->rx_packets;
8309 dest->tx_packets += src->tx_packets;
8310 dest->rx_bytes += src->rx_bytes;
8311 dest->tx_bytes += src->tx_bytes;
8312 dest->rx_errors += src->rx_errors;
8313 dest->tx_errors += src->tx_errors;
8314 dest->rx_dropped += src->rx_dropped;
8315 dest->tx_dropped += src->tx_dropped;
8316 dest->multicast += src->multicast;
8317 }
8318
netdev_offload_xstats_get_used(struct net_device * dev,enum netdev_offload_xstats_type type,bool * p_used,struct netlink_ext_ack * extack)8319 static int netdev_offload_xstats_get_used(struct net_device *dev,
8320 enum netdev_offload_xstats_type type,
8321 bool *p_used,
8322 struct netlink_ext_ack *extack)
8323 {
8324 struct netdev_notifier_offload_xstats_ru report_used = {};
8325 struct netdev_notifier_offload_xstats_info info = {
8326 .info.dev = dev,
8327 .info.extack = extack,
8328 .type = type,
8329 .report_used = &report_used,
8330 };
8331 int rc;
8332
8333 WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8334 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8335 &info.info);
8336 *p_used = report_used.used;
8337 return notifier_to_errno(rc);
8338 }
8339
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)8340 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8341 enum netdev_offload_xstats_type type,
8342 struct rtnl_hw_stats64 *p_stats,
8343 bool *p_used,
8344 struct netlink_ext_ack *extack)
8345 {
8346 struct netdev_notifier_offload_xstats_rd report_delta = {};
8347 struct netdev_notifier_offload_xstats_info info = {
8348 .info.dev = dev,
8349 .info.extack = extack,
8350 .type = type,
8351 .report_delta = &report_delta,
8352 };
8353 struct rtnl_hw_stats64 *stats;
8354 int rc;
8355
8356 stats = netdev_offload_xstats_get_ptr(dev, type);
8357 if (WARN_ON(!stats))
8358 return -EINVAL;
8359
8360 rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8361 &info.info);
8362
8363 /* Cache whatever we got, even if there was an error, otherwise the
8364 * successful stats retrievals would get lost.
8365 */
8366 netdev_hw_stats64_add(stats, &report_delta.stats);
8367
8368 if (p_stats)
8369 *p_stats = *stats;
8370 *p_used = report_delta.used;
8371
8372 return notifier_to_errno(rc);
8373 }
8374
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)8375 int netdev_offload_xstats_get(struct net_device *dev,
8376 enum netdev_offload_xstats_type type,
8377 struct rtnl_hw_stats64 *p_stats, bool *p_used,
8378 struct netlink_ext_ack *extack)
8379 {
8380 ASSERT_RTNL();
8381
8382 if (p_stats)
8383 return netdev_offload_xstats_get_stats(dev, type, p_stats,
8384 p_used, extack);
8385 else
8386 return netdev_offload_xstats_get_used(dev, type, p_used,
8387 extack);
8388 }
8389 EXPORT_SYMBOL(netdev_offload_xstats_get);
8390
8391 void
netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd * report_delta,const struct rtnl_hw_stats64 * stats)8392 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8393 const struct rtnl_hw_stats64 *stats)
8394 {
8395 report_delta->used = true;
8396 netdev_hw_stats64_add(&report_delta->stats, stats);
8397 }
8398 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8399
8400 void
netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru * report_used)8401 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8402 {
8403 report_used->used = true;
8404 }
8405 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8406
netdev_offload_xstats_push_delta(struct net_device * dev,enum netdev_offload_xstats_type type,const struct rtnl_hw_stats64 * p_stats)8407 void netdev_offload_xstats_push_delta(struct net_device *dev,
8408 enum netdev_offload_xstats_type type,
8409 const struct rtnl_hw_stats64 *p_stats)
8410 {
8411 struct rtnl_hw_stats64 *stats;
8412
8413 ASSERT_RTNL();
8414
8415 stats = netdev_offload_xstats_get_ptr(dev, type);
8416 if (WARN_ON(!stats))
8417 return;
8418
8419 netdev_hw_stats64_add(stats, p_stats);
8420 }
8421 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8422
8423 /**
8424 * netdev_get_xmit_slave - Get the xmit slave of master device
8425 * @dev: device
8426 * @skb: The packet
8427 * @all_slaves: assume all the slaves are active
8428 *
8429 * The reference counters are not incremented so the caller must be
8430 * careful with locks. The caller must hold RCU lock.
8431 * %NULL is returned if no slave is found.
8432 */
8433
netdev_get_xmit_slave(struct net_device * dev,struct sk_buff * skb,bool all_slaves)8434 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8435 struct sk_buff *skb,
8436 bool all_slaves)
8437 {
8438 const struct net_device_ops *ops = dev->netdev_ops;
8439
8440 if (!ops->ndo_get_xmit_slave)
8441 return NULL;
8442 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8443 }
8444 EXPORT_SYMBOL(netdev_get_xmit_slave);
8445
netdev_sk_get_lower_dev(struct net_device * dev,struct sock * sk)8446 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8447 struct sock *sk)
8448 {
8449 const struct net_device_ops *ops = dev->netdev_ops;
8450
8451 if (!ops->ndo_sk_get_lower_dev)
8452 return NULL;
8453 return ops->ndo_sk_get_lower_dev(dev, sk);
8454 }
8455
8456 /**
8457 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8458 * @dev: device
8459 * @sk: the socket
8460 *
8461 * %NULL is returned if no lower device is found.
8462 */
8463
netdev_sk_get_lowest_dev(struct net_device * dev,struct sock * sk)8464 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8465 struct sock *sk)
8466 {
8467 struct net_device *lower;
8468
8469 lower = netdev_sk_get_lower_dev(dev, sk);
8470 while (lower) {
8471 dev = lower;
8472 lower = netdev_sk_get_lower_dev(dev, sk);
8473 }
8474
8475 return dev;
8476 }
8477 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8478
netdev_adjacent_add_links(struct net_device * dev)8479 static void netdev_adjacent_add_links(struct net_device *dev)
8480 {
8481 struct netdev_adjacent *iter;
8482
8483 struct net *net = dev_net(dev);
8484
8485 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8486 if (!net_eq(net, dev_net(iter->dev)))
8487 continue;
8488 netdev_adjacent_sysfs_add(iter->dev, dev,
8489 &iter->dev->adj_list.lower);
8490 netdev_adjacent_sysfs_add(dev, iter->dev,
8491 &dev->adj_list.upper);
8492 }
8493
8494 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8495 if (!net_eq(net, dev_net(iter->dev)))
8496 continue;
8497 netdev_adjacent_sysfs_add(iter->dev, dev,
8498 &iter->dev->adj_list.upper);
8499 netdev_adjacent_sysfs_add(dev, iter->dev,
8500 &dev->adj_list.lower);
8501 }
8502 }
8503
netdev_adjacent_del_links(struct net_device * dev)8504 static void netdev_adjacent_del_links(struct net_device *dev)
8505 {
8506 struct netdev_adjacent *iter;
8507
8508 struct net *net = dev_net(dev);
8509
8510 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8511 if (!net_eq(net, dev_net(iter->dev)))
8512 continue;
8513 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8514 &iter->dev->adj_list.lower);
8515 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8516 &dev->adj_list.upper);
8517 }
8518
8519 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8520 if (!net_eq(net, dev_net(iter->dev)))
8521 continue;
8522 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8523 &iter->dev->adj_list.upper);
8524 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8525 &dev->adj_list.lower);
8526 }
8527 }
8528
netdev_adjacent_rename_links(struct net_device * dev,char * oldname)8529 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8530 {
8531 struct netdev_adjacent *iter;
8532
8533 struct net *net = dev_net(dev);
8534
8535 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8536 if (!net_eq(net, dev_net(iter->dev)))
8537 continue;
8538 netdev_adjacent_sysfs_del(iter->dev, oldname,
8539 &iter->dev->adj_list.lower);
8540 netdev_adjacent_sysfs_add(iter->dev, dev,
8541 &iter->dev->adj_list.lower);
8542 }
8543
8544 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8545 if (!net_eq(net, dev_net(iter->dev)))
8546 continue;
8547 netdev_adjacent_sysfs_del(iter->dev, oldname,
8548 &iter->dev->adj_list.upper);
8549 netdev_adjacent_sysfs_add(iter->dev, dev,
8550 &iter->dev->adj_list.upper);
8551 }
8552 }
8553
netdev_lower_dev_get_private(struct net_device * dev,struct net_device * lower_dev)8554 void *netdev_lower_dev_get_private(struct net_device *dev,
8555 struct net_device *lower_dev)
8556 {
8557 struct netdev_adjacent *lower;
8558
8559 if (!lower_dev)
8560 return NULL;
8561 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8562 if (!lower)
8563 return NULL;
8564
8565 return lower->private;
8566 }
8567 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8568
8569
8570 /**
8571 * netdev_lower_state_changed - Dispatch event about lower device state change
8572 * @lower_dev: device
8573 * @lower_state_info: state to dispatch
8574 *
8575 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8576 * The caller must hold the RTNL lock.
8577 */
netdev_lower_state_changed(struct net_device * lower_dev,void * lower_state_info)8578 void netdev_lower_state_changed(struct net_device *lower_dev,
8579 void *lower_state_info)
8580 {
8581 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8582 .info.dev = lower_dev,
8583 };
8584
8585 ASSERT_RTNL();
8586 changelowerstate_info.lower_state_info = lower_state_info;
8587 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8588 &changelowerstate_info.info);
8589 }
8590 EXPORT_SYMBOL(netdev_lower_state_changed);
8591
dev_change_rx_flags(struct net_device * dev,int flags)8592 static void dev_change_rx_flags(struct net_device *dev, int flags)
8593 {
8594 const struct net_device_ops *ops = dev->netdev_ops;
8595
8596 if (ops->ndo_change_rx_flags)
8597 ops->ndo_change_rx_flags(dev, flags);
8598 }
8599
__dev_set_promiscuity(struct net_device * dev,int inc,bool notify)8600 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8601 {
8602 unsigned int old_flags = dev->flags;
8603 unsigned int promiscuity, flags;
8604 kuid_t uid;
8605 kgid_t gid;
8606
8607 ASSERT_RTNL();
8608
8609 promiscuity = dev->promiscuity + inc;
8610 if (promiscuity == 0) {
8611 /*
8612 * Avoid overflow.
8613 * If inc causes overflow, untouch promisc and return error.
8614 */
8615 if (unlikely(inc > 0)) {
8616 netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8617 return -EOVERFLOW;
8618 }
8619 flags = old_flags & ~IFF_PROMISC;
8620 } else {
8621 flags = old_flags | IFF_PROMISC;
8622 }
8623 WRITE_ONCE(dev->promiscuity, promiscuity);
8624 if (flags != old_flags) {
8625 WRITE_ONCE(dev->flags, flags);
8626 netdev_info(dev, "%s promiscuous mode\n",
8627 dev->flags & IFF_PROMISC ? "entered" : "left");
8628 if (audit_enabled) {
8629 current_uid_gid(&uid, &gid);
8630 audit_log(audit_context(), GFP_ATOMIC,
8631 AUDIT_ANOM_PROMISCUOUS,
8632 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8633 dev->name, (dev->flags & IFF_PROMISC),
8634 (old_flags & IFF_PROMISC),
8635 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8636 from_kuid(&init_user_ns, uid),
8637 from_kgid(&init_user_ns, gid),
8638 audit_get_sessionid(current));
8639 }
8640
8641 dev_change_rx_flags(dev, IFF_PROMISC);
8642 }
8643 if (notify)
8644 __dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8645 return 0;
8646 }
8647
8648 /**
8649 * dev_set_promiscuity - update promiscuity count on a device
8650 * @dev: device
8651 * @inc: modifier
8652 *
8653 * Add or remove promiscuity from a device. While the count in the device
8654 * remains above zero the interface remains promiscuous. Once it hits zero
8655 * the device reverts back to normal filtering operation. A negative inc
8656 * value is used to drop promiscuity on the device.
8657 * Return 0 if successful or a negative errno code on error.
8658 */
dev_set_promiscuity(struct net_device * dev,int inc)8659 int dev_set_promiscuity(struct net_device *dev, int inc)
8660 {
8661 unsigned int old_flags = dev->flags;
8662 int err;
8663
8664 err = __dev_set_promiscuity(dev, inc, true);
8665 if (err < 0)
8666 return err;
8667 if (dev->flags != old_flags)
8668 dev_set_rx_mode(dev);
8669 return err;
8670 }
8671 EXPORT_SYMBOL(dev_set_promiscuity);
8672
__dev_set_allmulti(struct net_device * dev,int inc,bool notify)8673 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8674 {
8675 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8676 unsigned int allmulti, flags;
8677
8678 ASSERT_RTNL();
8679
8680 allmulti = dev->allmulti + inc;
8681 if (allmulti == 0) {
8682 /*
8683 * Avoid overflow.
8684 * If inc causes overflow, untouch allmulti and return error.
8685 */
8686 if (unlikely(inc > 0)) {
8687 netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8688 return -EOVERFLOW;
8689 }
8690 flags = old_flags & ~IFF_ALLMULTI;
8691 } else {
8692 flags = old_flags | IFF_ALLMULTI;
8693 }
8694 WRITE_ONCE(dev->allmulti, allmulti);
8695 if (flags != old_flags) {
8696 WRITE_ONCE(dev->flags, flags);
8697 netdev_info(dev, "%s allmulticast mode\n",
8698 dev->flags & IFF_ALLMULTI ? "entered" : "left");
8699 dev_change_rx_flags(dev, IFF_ALLMULTI);
8700 dev_set_rx_mode(dev);
8701 if (notify)
8702 __dev_notify_flags(dev, old_flags,
8703 dev->gflags ^ old_gflags, 0, NULL);
8704 }
8705 return 0;
8706 }
8707
8708 /**
8709 * dev_set_allmulti - update allmulti count on a device
8710 * @dev: device
8711 * @inc: modifier
8712 *
8713 * Add or remove reception of all multicast frames to a device. While the
8714 * count in the device remains above zero the interface remains listening
8715 * to all interfaces. Once it hits zero the device reverts back to normal
8716 * filtering operation. A negative @inc value is used to drop the counter
8717 * when releasing a resource needing all multicasts.
8718 * Return 0 if successful or a negative errno code on error.
8719 */
8720
dev_set_allmulti(struct net_device * dev,int inc)8721 int dev_set_allmulti(struct net_device *dev, int inc)
8722 {
8723 return __dev_set_allmulti(dev, inc, true);
8724 }
8725 EXPORT_SYMBOL(dev_set_allmulti);
8726
8727 /*
8728 * Upload unicast and multicast address lists to device and
8729 * configure RX filtering. When the device doesn't support unicast
8730 * filtering it is put in promiscuous mode while unicast addresses
8731 * are present.
8732 */
__dev_set_rx_mode(struct net_device * dev)8733 void __dev_set_rx_mode(struct net_device *dev)
8734 {
8735 const struct net_device_ops *ops = dev->netdev_ops;
8736
8737 /* dev_open will call this function so the list will stay sane. */
8738 if (!(dev->flags&IFF_UP))
8739 return;
8740
8741 if (!netif_device_present(dev))
8742 return;
8743
8744 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8745 /* Unicast addresses changes may only happen under the rtnl,
8746 * therefore calling __dev_set_promiscuity here is safe.
8747 */
8748 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8749 __dev_set_promiscuity(dev, 1, false);
8750 dev->uc_promisc = true;
8751 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8752 __dev_set_promiscuity(dev, -1, false);
8753 dev->uc_promisc = false;
8754 }
8755 }
8756
8757 if (ops->ndo_set_rx_mode)
8758 ops->ndo_set_rx_mode(dev);
8759 }
8760
dev_set_rx_mode(struct net_device * dev)8761 void dev_set_rx_mode(struct net_device *dev)
8762 {
8763 netif_addr_lock_bh(dev);
8764 __dev_set_rx_mode(dev);
8765 netif_addr_unlock_bh(dev);
8766 }
8767
8768 /**
8769 * dev_get_flags - get flags reported to userspace
8770 * @dev: device
8771 *
8772 * Get the combination of flag bits exported through APIs to userspace.
8773 */
dev_get_flags(const struct net_device * dev)8774 unsigned int dev_get_flags(const struct net_device *dev)
8775 {
8776 unsigned int flags;
8777
8778 flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC |
8779 IFF_ALLMULTI |
8780 IFF_RUNNING |
8781 IFF_LOWER_UP |
8782 IFF_DORMANT)) |
8783 (READ_ONCE(dev->gflags) & (IFF_PROMISC |
8784 IFF_ALLMULTI));
8785
8786 if (netif_running(dev)) {
8787 if (netif_oper_up(dev))
8788 flags |= IFF_RUNNING;
8789 if (netif_carrier_ok(dev))
8790 flags |= IFF_LOWER_UP;
8791 if (netif_dormant(dev))
8792 flags |= IFF_DORMANT;
8793 }
8794
8795 return flags;
8796 }
8797 EXPORT_SYMBOL(dev_get_flags);
8798
__dev_change_flags(struct net_device * dev,unsigned int flags,struct netlink_ext_ack * extack)8799 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8800 struct netlink_ext_ack *extack)
8801 {
8802 unsigned int old_flags = dev->flags;
8803 int ret;
8804
8805 ASSERT_RTNL();
8806
8807 /*
8808 * Set the flags on our device.
8809 */
8810
8811 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8812 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8813 IFF_AUTOMEDIA)) |
8814 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8815 IFF_ALLMULTI));
8816
8817 /*
8818 * Load in the correct multicast list now the flags have changed.
8819 */
8820
8821 if ((old_flags ^ flags) & IFF_MULTICAST)
8822 dev_change_rx_flags(dev, IFF_MULTICAST);
8823
8824 dev_set_rx_mode(dev);
8825
8826 /*
8827 * Have we downed the interface. We handle IFF_UP ourselves
8828 * according to user attempts to set it, rather than blindly
8829 * setting it.
8830 */
8831
8832 ret = 0;
8833 if ((old_flags ^ flags) & IFF_UP) {
8834 if (old_flags & IFF_UP)
8835 __dev_close(dev);
8836 else
8837 ret = __dev_open(dev, extack);
8838 }
8839
8840 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8841 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8842 unsigned int old_flags = dev->flags;
8843
8844 dev->gflags ^= IFF_PROMISC;
8845
8846 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8847 if (dev->flags != old_flags)
8848 dev_set_rx_mode(dev);
8849 }
8850
8851 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8852 * is important. Some (broken) drivers set IFF_PROMISC, when
8853 * IFF_ALLMULTI is requested not asking us and not reporting.
8854 */
8855 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8856 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8857
8858 dev->gflags ^= IFF_ALLMULTI;
8859 __dev_set_allmulti(dev, inc, false);
8860 }
8861
8862 return ret;
8863 }
8864
__dev_notify_flags(struct net_device * dev,unsigned int old_flags,unsigned int gchanges,u32 portid,const struct nlmsghdr * nlh)8865 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8866 unsigned int gchanges, u32 portid,
8867 const struct nlmsghdr *nlh)
8868 {
8869 unsigned int changes = dev->flags ^ old_flags;
8870
8871 if (gchanges)
8872 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8873
8874 if (changes & IFF_UP) {
8875 if (dev->flags & IFF_UP)
8876 call_netdevice_notifiers(NETDEV_UP, dev);
8877 else
8878 call_netdevice_notifiers(NETDEV_DOWN, dev);
8879 }
8880
8881 if (dev->flags & IFF_UP &&
8882 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8883 struct netdev_notifier_change_info change_info = {
8884 .info = {
8885 .dev = dev,
8886 },
8887 .flags_changed = changes,
8888 };
8889
8890 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8891 }
8892 }
8893
8894 /**
8895 * dev_change_flags - change device settings
8896 * @dev: device
8897 * @flags: device state flags
8898 * @extack: netlink extended ack
8899 *
8900 * Change settings on device based state flags. The flags are
8901 * in the userspace exported format.
8902 */
dev_change_flags(struct net_device * dev,unsigned int flags,struct netlink_ext_ack * extack)8903 int dev_change_flags(struct net_device *dev, unsigned int flags,
8904 struct netlink_ext_ack *extack)
8905 {
8906 int ret;
8907 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8908
8909 ret = __dev_change_flags(dev, flags, extack);
8910 if (ret < 0)
8911 return ret;
8912
8913 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8914 __dev_notify_flags(dev, old_flags, changes, 0, NULL);
8915 return ret;
8916 }
8917 EXPORT_SYMBOL(dev_change_flags);
8918
__dev_set_mtu(struct net_device * dev,int new_mtu)8919 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8920 {
8921 const struct net_device_ops *ops = dev->netdev_ops;
8922
8923 if (ops->ndo_change_mtu)
8924 return ops->ndo_change_mtu(dev, new_mtu);
8925
8926 /* Pairs with all the lockless reads of dev->mtu in the stack */
8927 WRITE_ONCE(dev->mtu, new_mtu);
8928 return 0;
8929 }
8930 EXPORT_SYMBOL(__dev_set_mtu);
8931
dev_validate_mtu(struct net_device * dev,int new_mtu,struct netlink_ext_ack * extack)8932 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8933 struct netlink_ext_ack *extack)
8934 {
8935 /* MTU must be positive, and in range */
8936 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8937 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8938 return -EINVAL;
8939 }
8940
8941 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8942 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8943 return -EINVAL;
8944 }
8945 return 0;
8946 }
8947
8948 /**
8949 * dev_set_mtu_ext - Change maximum transfer unit
8950 * @dev: device
8951 * @new_mtu: new transfer unit
8952 * @extack: netlink extended ack
8953 *
8954 * Change the maximum transfer size of the network device.
8955 */
dev_set_mtu_ext(struct net_device * dev,int new_mtu,struct netlink_ext_ack * extack)8956 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8957 struct netlink_ext_ack *extack)
8958 {
8959 int err, orig_mtu;
8960
8961 if (new_mtu == dev->mtu)
8962 return 0;
8963
8964 err = dev_validate_mtu(dev, new_mtu, extack);
8965 if (err)
8966 return err;
8967
8968 if (!netif_device_present(dev))
8969 return -ENODEV;
8970
8971 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8972 err = notifier_to_errno(err);
8973 if (err)
8974 return err;
8975
8976 orig_mtu = dev->mtu;
8977 err = __dev_set_mtu(dev, new_mtu);
8978
8979 if (!err) {
8980 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8981 orig_mtu);
8982 err = notifier_to_errno(err);
8983 if (err) {
8984 /* setting mtu back and notifying everyone again,
8985 * so that they have a chance to revert changes.
8986 */
8987 __dev_set_mtu(dev, orig_mtu);
8988 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8989 new_mtu);
8990 }
8991 }
8992 return err;
8993 }
8994
dev_set_mtu(struct net_device * dev,int new_mtu)8995 int dev_set_mtu(struct net_device *dev, int new_mtu)
8996 {
8997 struct netlink_ext_ack extack;
8998 int err;
8999
9000 memset(&extack, 0, sizeof(extack));
9001 err = dev_set_mtu_ext(dev, new_mtu, &extack);
9002 if (err && extack._msg)
9003 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
9004 return err;
9005 }
9006 EXPORT_SYMBOL(dev_set_mtu);
9007
9008 /**
9009 * dev_change_tx_queue_len - Change TX queue length of a netdevice
9010 * @dev: device
9011 * @new_len: new tx queue length
9012 */
dev_change_tx_queue_len(struct net_device * dev,unsigned long new_len)9013 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9014 {
9015 unsigned int orig_len = dev->tx_queue_len;
9016 int res;
9017
9018 if (new_len != (unsigned int)new_len)
9019 return -ERANGE;
9020
9021 if (new_len != orig_len) {
9022 WRITE_ONCE(dev->tx_queue_len, new_len);
9023 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9024 res = notifier_to_errno(res);
9025 if (res)
9026 goto err_rollback;
9027 res = dev_qdisc_change_tx_queue_len(dev);
9028 if (res)
9029 goto err_rollback;
9030 }
9031
9032 return 0;
9033
9034 err_rollback:
9035 netdev_err(dev, "refused to change device tx_queue_len\n");
9036 WRITE_ONCE(dev->tx_queue_len, orig_len);
9037 return res;
9038 }
9039
9040 /**
9041 * dev_set_group - Change group this device belongs to
9042 * @dev: device
9043 * @new_group: group this device should belong to
9044 */
dev_set_group(struct net_device * dev,int new_group)9045 void dev_set_group(struct net_device *dev, int new_group)
9046 {
9047 dev->group = new_group;
9048 }
9049
9050 /**
9051 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9052 * @dev: device
9053 * @addr: new address
9054 * @extack: netlink extended ack
9055 */
dev_pre_changeaddr_notify(struct net_device * dev,const char * addr,struct netlink_ext_ack * extack)9056 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9057 struct netlink_ext_ack *extack)
9058 {
9059 struct netdev_notifier_pre_changeaddr_info info = {
9060 .info.dev = dev,
9061 .info.extack = extack,
9062 .dev_addr = addr,
9063 };
9064 int rc;
9065
9066 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9067 return notifier_to_errno(rc);
9068 }
9069 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9070
9071 /**
9072 * dev_set_mac_address - Change Media Access Control Address
9073 * @dev: device
9074 * @sa: new address
9075 * @extack: netlink extended ack
9076 *
9077 * Change the hardware (MAC) address of the device
9078 */
dev_set_mac_address(struct net_device * dev,struct sockaddr * sa,struct netlink_ext_ack * extack)9079 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9080 struct netlink_ext_ack *extack)
9081 {
9082 const struct net_device_ops *ops = dev->netdev_ops;
9083 int err;
9084
9085 if (!ops->ndo_set_mac_address)
9086 return -EOPNOTSUPP;
9087 if (sa->sa_family != dev->type)
9088 return -EINVAL;
9089 if (!netif_device_present(dev))
9090 return -ENODEV;
9091 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9092 if (err)
9093 return err;
9094 if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) {
9095 err = ops->ndo_set_mac_address(dev, sa);
9096 if (err)
9097 return err;
9098 }
9099 dev->addr_assign_type = NET_ADDR_SET;
9100 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9101 add_device_randomness(dev->dev_addr, dev->addr_len);
9102 return 0;
9103 }
9104 EXPORT_SYMBOL(dev_set_mac_address);
9105
9106 DECLARE_RWSEM(dev_addr_sem);
9107
dev_set_mac_address_user(struct net_device * dev,struct sockaddr * sa,struct netlink_ext_ack * extack)9108 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9109 struct netlink_ext_ack *extack)
9110 {
9111 int ret;
9112
9113 down_write(&dev_addr_sem);
9114 ret = dev_set_mac_address(dev, sa, extack);
9115 up_write(&dev_addr_sem);
9116 return ret;
9117 }
9118 EXPORT_SYMBOL(dev_set_mac_address_user);
9119
dev_get_mac_address(struct sockaddr * sa,struct net * net,char * dev_name)9120 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9121 {
9122 size_t size = sizeof(sa->sa_data_min);
9123 struct net_device *dev;
9124 int ret = 0;
9125
9126 down_read(&dev_addr_sem);
9127 rcu_read_lock();
9128
9129 dev = dev_get_by_name_rcu(net, dev_name);
9130 if (!dev) {
9131 ret = -ENODEV;
9132 goto unlock;
9133 }
9134 if (!dev->addr_len)
9135 memset(sa->sa_data, 0, size);
9136 else
9137 memcpy(sa->sa_data, dev->dev_addr,
9138 min_t(size_t, size, dev->addr_len));
9139 sa->sa_family = dev->type;
9140
9141 unlock:
9142 rcu_read_unlock();
9143 up_read(&dev_addr_sem);
9144 return ret;
9145 }
9146 EXPORT_SYMBOL(dev_get_mac_address);
9147
9148 /**
9149 * dev_change_carrier - Change device carrier
9150 * @dev: device
9151 * @new_carrier: new value
9152 *
9153 * Change device carrier
9154 */
dev_change_carrier(struct net_device * dev,bool new_carrier)9155 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9156 {
9157 const struct net_device_ops *ops = dev->netdev_ops;
9158
9159 if (!ops->ndo_change_carrier)
9160 return -EOPNOTSUPP;
9161 if (!netif_device_present(dev))
9162 return -ENODEV;
9163 return ops->ndo_change_carrier(dev, new_carrier);
9164 }
9165
9166 /**
9167 * dev_get_phys_port_id - Get device physical port ID
9168 * @dev: device
9169 * @ppid: port ID
9170 *
9171 * Get device physical port ID
9172 */
dev_get_phys_port_id(struct net_device * dev,struct netdev_phys_item_id * ppid)9173 int dev_get_phys_port_id(struct net_device *dev,
9174 struct netdev_phys_item_id *ppid)
9175 {
9176 const struct net_device_ops *ops = dev->netdev_ops;
9177
9178 if (!ops->ndo_get_phys_port_id)
9179 return -EOPNOTSUPP;
9180 return ops->ndo_get_phys_port_id(dev, ppid);
9181 }
9182
9183 /**
9184 * dev_get_phys_port_name - Get device physical port name
9185 * @dev: device
9186 * @name: port name
9187 * @len: limit of bytes to copy to name
9188 *
9189 * Get device physical port name
9190 */
dev_get_phys_port_name(struct net_device * dev,char * name,size_t len)9191 int dev_get_phys_port_name(struct net_device *dev,
9192 char *name, size_t len)
9193 {
9194 const struct net_device_ops *ops = dev->netdev_ops;
9195 int err;
9196
9197 if (ops->ndo_get_phys_port_name) {
9198 err = ops->ndo_get_phys_port_name(dev, name, len);
9199 if (err != -EOPNOTSUPP)
9200 return err;
9201 }
9202 return devlink_compat_phys_port_name_get(dev, name, len);
9203 }
9204
9205 /**
9206 * dev_get_port_parent_id - Get the device's port parent identifier
9207 * @dev: network device
9208 * @ppid: pointer to a storage for the port's parent identifier
9209 * @recurse: allow/disallow recursion to lower devices
9210 *
9211 * Get the devices's port parent identifier
9212 */
dev_get_port_parent_id(struct net_device * dev,struct netdev_phys_item_id * ppid,bool recurse)9213 int dev_get_port_parent_id(struct net_device *dev,
9214 struct netdev_phys_item_id *ppid,
9215 bool recurse)
9216 {
9217 const struct net_device_ops *ops = dev->netdev_ops;
9218 struct netdev_phys_item_id first = { };
9219 struct net_device *lower_dev;
9220 struct list_head *iter;
9221 int err;
9222
9223 if (ops->ndo_get_port_parent_id) {
9224 err = ops->ndo_get_port_parent_id(dev, ppid);
9225 if (err != -EOPNOTSUPP)
9226 return err;
9227 }
9228
9229 err = devlink_compat_switch_id_get(dev, ppid);
9230 if (!recurse || err != -EOPNOTSUPP)
9231 return err;
9232
9233 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9234 err = dev_get_port_parent_id(lower_dev, ppid, true);
9235 if (err)
9236 break;
9237 if (!first.id_len)
9238 first = *ppid;
9239 else if (memcmp(&first, ppid, sizeof(*ppid)))
9240 return -EOPNOTSUPP;
9241 }
9242
9243 return err;
9244 }
9245 EXPORT_SYMBOL(dev_get_port_parent_id);
9246
9247 /**
9248 * netdev_port_same_parent_id - Indicate if two network devices have
9249 * the same port parent identifier
9250 * @a: first network device
9251 * @b: second network device
9252 */
netdev_port_same_parent_id(struct net_device * a,struct net_device * b)9253 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9254 {
9255 struct netdev_phys_item_id a_id = { };
9256 struct netdev_phys_item_id b_id = { };
9257
9258 if (dev_get_port_parent_id(a, &a_id, true) ||
9259 dev_get_port_parent_id(b, &b_id, true))
9260 return false;
9261
9262 return netdev_phys_item_id_same(&a_id, &b_id);
9263 }
9264 EXPORT_SYMBOL(netdev_port_same_parent_id);
9265
9266 /**
9267 * dev_change_proto_down - set carrier according to proto_down.
9268 *
9269 * @dev: device
9270 * @proto_down: new value
9271 */
dev_change_proto_down(struct net_device * dev,bool proto_down)9272 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9273 {
9274 if (!dev->change_proto_down)
9275 return -EOPNOTSUPP;
9276 if (!netif_device_present(dev))
9277 return -ENODEV;
9278 if (proto_down)
9279 netif_carrier_off(dev);
9280 else
9281 netif_carrier_on(dev);
9282 WRITE_ONCE(dev->proto_down, proto_down);
9283 return 0;
9284 }
9285
9286 /**
9287 * dev_change_proto_down_reason - proto down reason
9288 *
9289 * @dev: device
9290 * @mask: proto down mask
9291 * @value: proto down value
9292 */
dev_change_proto_down_reason(struct net_device * dev,unsigned long mask,u32 value)9293 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9294 u32 value)
9295 {
9296 u32 proto_down_reason;
9297 int b;
9298
9299 if (!mask) {
9300 proto_down_reason = value;
9301 } else {
9302 proto_down_reason = dev->proto_down_reason;
9303 for_each_set_bit(b, &mask, 32) {
9304 if (value & (1 << b))
9305 proto_down_reason |= BIT(b);
9306 else
9307 proto_down_reason &= ~BIT(b);
9308 }
9309 }
9310 WRITE_ONCE(dev->proto_down_reason, proto_down_reason);
9311 }
9312
9313 struct bpf_xdp_link {
9314 struct bpf_link link;
9315 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9316 int flags;
9317 };
9318
dev_xdp_mode(struct net_device * dev,u32 flags)9319 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9320 {
9321 if (flags & XDP_FLAGS_HW_MODE)
9322 return XDP_MODE_HW;
9323 if (flags & XDP_FLAGS_DRV_MODE)
9324 return XDP_MODE_DRV;
9325 if (flags & XDP_FLAGS_SKB_MODE)
9326 return XDP_MODE_SKB;
9327 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9328 }
9329
dev_xdp_bpf_op(struct net_device * dev,enum bpf_xdp_mode mode)9330 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9331 {
9332 switch (mode) {
9333 case XDP_MODE_SKB:
9334 return generic_xdp_install;
9335 case XDP_MODE_DRV:
9336 case XDP_MODE_HW:
9337 return dev->netdev_ops->ndo_bpf;
9338 default:
9339 return NULL;
9340 }
9341 }
9342
dev_xdp_link(struct net_device * dev,enum bpf_xdp_mode mode)9343 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9344 enum bpf_xdp_mode mode)
9345 {
9346 return dev->xdp_state[mode].link;
9347 }
9348
dev_xdp_prog(struct net_device * dev,enum bpf_xdp_mode mode)9349 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9350 enum bpf_xdp_mode mode)
9351 {
9352 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9353
9354 if (link)
9355 return link->link.prog;
9356 return dev->xdp_state[mode].prog;
9357 }
9358
dev_xdp_prog_count(struct net_device * dev)9359 u8 dev_xdp_prog_count(struct net_device *dev)
9360 {
9361 u8 count = 0;
9362 int i;
9363
9364 for (i = 0; i < __MAX_XDP_MODE; i++)
9365 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9366 count++;
9367 return count;
9368 }
9369 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9370
dev_xdp_propagate(struct net_device * dev,struct netdev_bpf * bpf)9371 int dev_xdp_propagate(struct net_device *dev, struct netdev_bpf *bpf)
9372 {
9373 if (!dev->netdev_ops->ndo_bpf)
9374 return -EOPNOTSUPP;
9375
9376 if (dev_get_min_mp_channel_count(dev)) {
9377 NL_SET_ERR_MSG(bpf->extack, "unable to propagate XDP to device using memory provider");
9378 return -EBUSY;
9379 }
9380
9381 return dev->netdev_ops->ndo_bpf(dev, bpf);
9382 }
9383 EXPORT_SYMBOL_GPL(dev_xdp_propagate);
9384
dev_xdp_prog_id(struct net_device * dev,enum bpf_xdp_mode mode)9385 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9386 {
9387 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9388
9389 return prog ? prog->aux->id : 0;
9390 }
9391
dev_xdp_set_link(struct net_device * dev,enum bpf_xdp_mode mode,struct bpf_xdp_link * link)9392 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9393 struct bpf_xdp_link *link)
9394 {
9395 dev->xdp_state[mode].link = link;
9396 dev->xdp_state[mode].prog = NULL;
9397 }
9398
dev_xdp_set_prog(struct net_device * dev,enum bpf_xdp_mode mode,struct bpf_prog * prog)9399 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9400 struct bpf_prog *prog)
9401 {
9402 dev->xdp_state[mode].link = NULL;
9403 dev->xdp_state[mode].prog = prog;
9404 }
9405
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)9406 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9407 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9408 u32 flags, struct bpf_prog *prog)
9409 {
9410 struct netdev_bpf xdp;
9411 int err;
9412
9413 if (dev_get_min_mp_channel_count(dev)) {
9414 NL_SET_ERR_MSG(extack, "unable to install XDP to device using memory provider");
9415 return -EBUSY;
9416 }
9417
9418 memset(&xdp, 0, sizeof(xdp));
9419 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9420 xdp.extack = extack;
9421 xdp.flags = flags;
9422 xdp.prog = prog;
9423
9424 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9425 * "moved" into driver), so they don't increment it on their own, but
9426 * they do decrement refcnt when program is detached or replaced.
9427 * Given net_device also owns link/prog, we need to bump refcnt here
9428 * to prevent drivers from underflowing it.
9429 */
9430 if (prog)
9431 bpf_prog_inc(prog);
9432 err = bpf_op(dev, &xdp);
9433 if (err) {
9434 if (prog)
9435 bpf_prog_put(prog);
9436 return err;
9437 }
9438
9439 if (mode != XDP_MODE_HW)
9440 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9441
9442 return 0;
9443 }
9444
dev_xdp_uninstall(struct net_device * dev)9445 static void dev_xdp_uninstall(struct net_device *dev)
9446 {
9447 struct bpf_xdp_link *link;
9448 struct bpf_prog *prog;
9449 enum bpf_xdp_mode mode;
9450 bpf_op_t bpf_op;
9451
9452 ASSERT_RTNL();
9453
9454 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9455 prog = dev_xdp_prog(dev, mode);
9456 if (!prog)
9457 continue;
9458
9459 bpf_op = dev_xdp_bpf_op(dev, mode);
9460 if (!bpf_op)
9461 continue;
9462
9463 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9464
9465 /* auto-detach link from net device */
9466 link = dev_xdp_link(dev, mode);
9467 if (link)
9468 link->dev = NULL;
9469 else
9470 bpf_prog_put(prog);
9471
9472 dev_xdp_set_link(dev, mode, NULL);
9473 }
9474 }
9475
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)9476 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9477 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9478 struct bpf_prog *old_prog, u32 flags)
9479 {
9480 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9481 struct bpf_prog *cur_prog;
9482 struct net_device *upper;
9483 struct list_head *iter;
9484 enum bpf_xdp_mode mode;
9485 bpf_op_t bpf_op;
9486 int err;
9487
9488 ASSERT_RTNL();
9489
9490 /* either link or prog attachment, never both */
9491 if (link && (new_prog || old_prog))
9492 return -EINVAL;
9493 /* link supports only XDP mode flags */
9494 if (link && (flags & ~XDP_FLAGS_MODES)) {
9495 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9496 return -EINVAL;
9497 }
9498 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9499 if (num_modes > 1) {
9500 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9501 return -EINVAL;
9502 }
9503 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9504 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9505 NL_SET_ERR_MSG(extack,
9506 "More than one program loaded, unset mode is ambiguous");
9507 return -EINVAL;
9508 }
9509 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9510 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9511 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9512 return -EINVAL;
9513 }
9514
9515 mode = dev_xdp_mode(dev, flags);
9516 /* can't replace attached link */
9517 if (dev_xdp_link(dev, mode)) {
9518 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9519 return -EBUSY;
9520 }
9521
9522 /* don't allow if an upper device already has a program */
9523 netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9524 if (dev_xdp_prog_count(upper) > 0) {
9525 NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9526 return -EEXIST;
9527 }
9528 }
9529
9530 cur_prog = dev_xdp_prog(dev, mode);
9531 /* can't replace attached prog with link */
9532 if (link && cur_prog) {
9533 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9534 return -EBUSY;
9535 }
9536 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9537 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9538 return -EEXIST;
9539 }
9540
9541 /* put effective new program into new_prog */
9542 if (link)
9543 new_prog = link->link.prog;
9544
9545 if (new_prog) {
9546 bool offload = mode == XDP_MODE_HW;
9547 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9548 ? XDP_MODE_DRV : XDP_MODE_SKB;
9549
9550 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9551 NL_SET_ERR_MSG(extack, "XDP program already attached");
9552 return -EBUSY;
9553 }
9554 if (!offload && dev_xdp_prog(dev, other_mode)) {
9555 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9556 return -EEXIST;
9557 }
9558 if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9559 NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9560 return -EINVAL;
9561 }
9562 if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9563 NL_SET_ERR_MSG(extack, "Program bound to different device");
9564 return -EINVAL;
9565 }
9566 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9567 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9568 return -EINVAL;
9569 }
9570 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9571 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9572 return -EINVAL;
9573 }
9574 }
9575
9576 /* don't call drivers if the effective program didn't change */
9577 if (new_prog != cur_prog) {
9578 bpf_op = dev_xdp_bpf_op(dev, mode);
9579 if (!bpf_op) {
9580 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9581 return -EOPNOTSUPP;
9582 }
9583
9584 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9585 if (err)
9586 return err;
9587 }
9588
9589 if (link)
9590 dev_xdp_set_link(dev, mode, link);
9591 else
9592 dev_xdp_set_prog(dev, mode, new_prog);
9593 if (cur_prog)
9594 bpf_prog_put(cur_prog);
9595
9596 return 0;
9597 }
9598
dev_xdp_attach_link(struct net_device * dev,struct netlink_ext_ack * extack,struct bpf_xdp_link * link)9599 static int dev_xdp_attach_link(struct net_device *dev,
9600 struct netlink_ext_ack *extack,
9601 struct bpf_xdp_link *link)
9602 {
9603 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9604 }
9605
dev_xdp_detach_link(struct net_device * dev,struct netlink_ext_ack * extack,struct bpf_xdp_link * link)9606 static int dev_xdp_detach_link(struct net_device *dev,
9607 struct netlink_ext_ack *extack,
9608 struct bpf_xdp_link *link)
9609 {
9610 enum bpf_xdp_mode mode;
9611 bpf_op_t bpf_op;
9612
9613 ASSERT_RTNL();
9614
9615 mode = dev_xdp_mode(dev, link->flags);
9616 if (dev_xdp_link(dev, mode) != link)
9617 return -EINVAL;
9618
9619 bpf_op = dev_xdp_bpf_op(dev, mode);
9620 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9621 dev_xdp_set_link(dev, mode, NULL);
9622 return 0;
9623 }
9624
bpf_xdp_link_release(struct bpf_link * link)9625 static void bpf_xdp_link_release(struct bpf_link *link)
9626 {
9627 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9628
9629 rtnl_lock();
9630
9631 /* if racing with net_device's tear down, xdp_link->dev might be
9632 * already NULL, in which case link was already auto-detached
9633 */
9634 if (xdp_link->dev) {
9635 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9636 xdp_link->dev = NULL;
9637 }
9638
9639 rtnl_unlock();
9640 }
9641
bpf_xdp_link_detach(struct bpf_link * link)9642 static int bpf_xdp_link_detach(struct bpf_link *link)
9643 {
9644 bpf_xdp_link_release(link);
9645 return 0;
9646 }
9647
bpf_xdp_link_dealloc(struct bpf_link * link)9648 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9649 {
9650 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9651
9652 kfree(xdp_link);
9653 }
9654
bpf_xdp_link_show_fdinfo(const struct bpf_link * link,struct seq_file * seq)9655 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9656 struct seq_file *seq)
9657 {
9658 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9659 u32 ifindex = 0;
9660
9661 rtnl_lock();
9662 if (xdp_link->dev)
9663 ifindex = xdp_link->dev->ifindex;
9664 rtnl_unlock();
9665
9666 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9667 }
9668
bpf_xdp_link_fill_link_info(const struct bpf_link * link,struct bpf_link_info * info)9669 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9670 struct bpf_link_info *info)
9671 {
9672 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9673 u32 ifindex = 0;
9674
9675 rtnl_lock();
9676 if (xdp_link->dev)
9677 ifindex = xdp_link->dev->ifindex;
9678 rtnl_unlock();
9679
9680 info->xdp.ifindex = ifindex;
9681 return 0;
9682 }
9683
bpf_xdp_link_update(struct bpf_link * link,struct bpf_prog * new_prog,struct bpf_prog * old_prog)9684 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9685 struct bpf_prog *old_prog)
9686 {
9687 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9688 enum bpf_xdp_mode mode;
9689 bpf_op_t bpf_op;
9690 int err = 0;
9691
9692 rtnl_lock();
9693
9694 /* link might have been auto-released already, so fail */
9695 if (!xdp_link->dev) {
9696 err = -ENOLINK;
9697 goto out_unlock;
9698 }
9699
9700 if (old_prog && link->prog != old_prog) {
9701 err = -EPERM;
9702 goto out_unlock;
9703 }
9704 old_prog = link->prog;
9705 if (old_prog->type != new_prog->type ||
9706 old_prog->expected_attach_type != new_prog->expected_attach_type) {
9707 err = -EINVAL;
9708 goto out_unlock;
9709 }
9710
9711 if (old_prog == new_prog) {
9712 /* no-op, don't disturb drivers */
9713 bpf_prog_put(new_prog);
9714 goto out_unlock;
9715 }
9716
9717 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9718 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9719 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9720 xdp_link->flags, new_prog);
9721 if (err)
9722 goto out_unlock;
9723
9724 old_prog = xchg(&link->prog, new_prog);
9725 bpf_prog_put(old_prog);
9726
9727 out_unlock:
9728 rtnl_unlock();
9729 return err;
9730 }
9731
9732 static const struct bpf_link_ops bpf_xdp_link_lops = {
9733 .release = bpf_xdp_link_release,
9734 .dealloc = bpf_xdp_link_dealloc,
9735 .detach = bpf_xdp_link_detach,
9736 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9737 .fill_link_info = bpf_xdp_link_fill_link_info,
9738 .update_prog = bpf_xdp_link_update,
9739 };
9740
bpf_xdp_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)9741 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9742 {
9743 struct net *net = current->nsproxy->net_ns;
9744 struct bpf_link_primer link_primer;
9745 struct netlink_ext_ack extack = {};
9746 struct bpf_xdp_link *link;
9747 struct net_device *dev;
9748 int err, fd;
9749
9750 rtnl_lock();
9751 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9752 if (!dev) {
9753 rtnl_unlock();
9754 return -EINVAL;
9755 }
9756
9757 link = kzalloc(sizeof(*link), GFP_USER);
9758 if (!link) {
9759 err = -ENOMEM;
9760 goto unlock;
9761 }
9762
9763 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9764 link->dev = dev;
9765 link->flags = attr->link_create.flags;
9766
9767 err = bpf_link_prime(&link->link, &link_primer);
9768 if (err) {
9769 kfree(link);
9770 goto unlock;
9771 }
9772
9773 err = dev_xdp_attach_link(dev, &extack, link);
9774 rtnl_unlock();
9775
9776 if (err) {
9777 link->dev = NULL;
9778 bpf_link_cleanup(&link_primer);
9779 trace_bpf_xdp_link_attach_failed(extack._msg);
9780 goto out_put_dev;
9781 }
9782
9783 fd = bpf_link_settle(&link_primer);
9784 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9785 dev_put(dev);
9786 return fd;
9787
9788 unlock:
9789 rtnl_unlock();
9790
9791 out_put_dev:
9792 dev_put(dev);
9793 return err;
9794 }
9795
9796 /**
9797 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9798 * @dev: device
9799 * @extack: netlink extended ack
9800 * @fd: new program fd or negative value to clear
9801 * @expected_fd: old program fd that userspace expects to replace or clear
9802 * @flags: xdp-related flags
9803 *
9804 * Set or clear a bpf program for a device
9805 */
dev_change_xdp_fd(struct net_device * dev,struct netlink_ext_ack * extack,int fd,int expected_fd,u32 flags)9806 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9807 int fd, int expected_fd, u32 flags)
9808 {
9809 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9810 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9811 int err;
9812
9813 ASSERT_RTNL();
9814
9815 if (fd >= 0) {
9816 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9817 mode != XDP_MODE_SKB);
9818 if (IS_ERR(new_prog))
9819 return PTR_ERR(new_prog);
9820 }
9821
9822 if (expected_fd >= 0) {
9823 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9824 mode != XDP_MODE_SKB);
9825 if (IS_ERR(old_prog)) {
9826 err = PTR_ERR(old_prog);
9827 old_prog = NULL;
9828 goto err_out;
9829 }
9830 }
9831
9832 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9833
9834 err_out:
9835 if (err && new_prog)
9836 bpf_prog_put(new_prog);
9837 if (old_prog)
9838 bpf_prog_put(old_prog);
9839 return err;
9840 }
9841
dev_get_min_mp_channel_count(const struct net_device * dev)9842 u32 dev_get_min_mp_channel_count(const struct net_device *dev)
9843 {
9844 int i;
9845
9846 ASSERT_RTNL();
9847
9848 for (i = dev->real_num_rx_queues - 1; i >= 0; i--)
9849 if (dev->_rx[i].mp_params.mp_priv)
9850 /* The channel count is the idx plus 1. */
9851 return i + 1;
9852
9853 return 0;
9854 }
9855
9856 /**
9857 * dev_index_reserve() - allocate an ifindex in a namespace
9858 * @net: the applicable net namespace
9859 * @ifindex: requested ifindex, pass %0 to get one allocated
9860 *
9861 * Allocate a ifindex for a new device. Caller must either use the ifindex
9862 * to store the device (via list_netdevice()) or call dev_index_release()
9863 * to give the index up.
9864 *
9865 * Return: a suitable unique value for a new device interface number or -errno.
9866 */
dev_index_reserve(struct net * net,u32 ifindex)9867 static int dev_index_reserve(struct net *net, u32 ifindex)
9868 {
9869 int err;
9870
9871 if (ifindex > INT_MAX) {
9872 DEBUG_NET_WARN_ON_ONCE(1);
9873 return -EINVAL;
9874 }
9875
9876 if (!ifindex)
9877 err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL,
9878 xa_limit_31b, &net->ifindex, GFP_KERNEL);
9879 else
9880 err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL);
9881 if (err < 0)
9882 return err;
9883
9884 return ifindex;
9885 }
9886
dev_index_release(struct net * net,int ifindex)9887 static void dev_index_release(struct net *net, int ifindex)
9888 {
9889 /* Expect only unused indexes, unlist_netdevice() removes the used */
9890 WARN_ON(xa_erase(&net->dev_by_index, ifindex));
9891 }
9892
9893 /* Delayed registration/unregisteration */
9894 LIST_HEAD(net_todo_list);
9895 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9896 atomic_t dev_unreg_count = ATOMIC_INIT(0);
9897
net_set_todo(struct net_device * dev)9898 static void net_set_todo(struct net_device *dev)
9899 {
9900 list_add_tail(&dev->todo_list, &net_todo_list);
9901 }
9902
netdev_sync_upper_features(struct net_device * lower,struct net_device * upper,netdev_features_t features)9903 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9904 struct net_device *upper, netdev_features_t features)
9905 {
9906 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9907 netdev_features_t feature;
9908 int feature_bit;
9909
9910 for_each_netdev_feature(upper_disables, feature_bit) {
9911 feature = __NETIF_F_BIT(feature_bit);
9912 if (!(upper->wanted_features & feature)
9913 && (features & feature)) {
9914 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9915 &feature, upper->name);
9916 features &= ~feature;
9917 }
9918 }
9919
9920 return features;
9921 }
9922
netdev_sync_lower_features(struct net_device * upper,struct net_device * lower,netdev_features_t features)9923 static void netdev_sync_lower_features(struct net_device *upper,
9924 struct net_device *lower, netdev_features_t features)
9925 {
9926 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9927 netdev_features_t feature;
9928 int feature_bit;
9929
9930 for_each_netdev_feature(upper_disables, feature_bit) {
9931 feature = __NETIF_F_BIT(feature_bit);
9932 if (!(features & feature) && (lower->features & feature)) {
9933 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9934 &feature, lower->name);
9935 lower->wanted_features &= ~feature;
9936 __netdev_update_features(lower);
9937
9938 if (unlikely(lower->features & feature))
9939 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9940 &feature, lower->name);
9941 else
9942 netdev_features_change(lower);
9943 }
9944 }
9945 }
9946
netdev_has_ip_or_hw_csum(netdev_features_t features)9947 static bool netdev_has_ip_or_hw_csum(netdev_features_t features)
9948 {
9949 netdev_features_t ip_csum_mask = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
9950 bool ip_csum = (features & ip_csum_mask) == ip_csum_mask;
9951 bool hw_csum = features & NETIF_F_HW_CSUM;
9952
9953 return ip_csum || hw_csum;
9954 }
9955
netdev_fix_features(struct net_device * dev,netdev_features_t features)9956 static netdev_features_t netdev_fix_features(struct net_device *dev,
9957 netdev_features_t features)
9958 {
9959 /* Fix illegal checksum combinations */
9960 if ((features & NETIF_F_HW_CSUM) &&
9961 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9962 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9963 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9964 }
9965
9966 /* TSO requires that SG is present as well. */
9967 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9968 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9969 features &= ~NETIF_F_ALL_TSO;
9970 }
9971
9972 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9973 !(features & NETIF_F_IP_CSUM)) {
9974 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9975 features &= ~NETIF_F_TSO;
9976 features &= ~NETIF_F_TSO_ECN;
9977 }
9978
9979 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9980 !(features & NETIF_F_IPV6_CSUM)) {
9981 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9982 features &= ~NETIF_F_TSO6;
9983 }
9984
9985 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9986 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9987 features &= ~NETIF_F_TSO_MANGLEID;
9988
9989 /* TSO ECN requires that TSO is present as well. */
9990 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9991 features &= ~NETIF_F_TSO_ECN;
9992
9993 /* Software GSO depends on SG. */
9994 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9995 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9996 features &= ~NETIF_F_GSO;
9997 }
9998
9999 /* GSO partial features require GSO partial be set */
10000 if ((features & dev->gso_partial_features) &&
10001 !(features & NETIF_F_GSO_PARTIAL)) {
10002 netdev_dbg(dev,
10003 "Dropping partially supported GSO features since no GSO partial.\n");
10004 features &= ~dev->gso_partial_features;
10005 }
10006
10007 if (!(features & NETIF_F_RXCSUM)) {
10008 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
10009 * successfully merged by hardware must also have the
10010 * checksum verified by hardware. If the user does not
10011 * want to enable RXCSUM, logically, we should disable GRO_HW.
10012 */
10013 if (features & NETIF_F_GRO_HW) {
10014 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
10015 features &= ~NETIF_F_GRO_HW;
10016 }
10017 }
10018
10019 /* LRO/HW-GRO features cannot be combined with RX-FCS */
10020 if (features & NETIF_F_RXFCS) {
10021 if (features & NETIF_F_LRO) {
10022 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
10023 features &= ~NETIF_F_LRO;
10024 }
10025
10026 if (features & NETIF_F_GRO_HW) {
10027 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
10028 features &= ~NETIF_F_GRO_HW;
10029 }
10030 }
10031
10032 if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
10033 netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
10034 features &= ~NETIF_F_LRO;
10035 }
10036
10037 if ((features & NETIF_F_HW_TLS_TX) && !netdev_has_ip_or_hw_csum(features)) {
10038 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
10039 features &= ~NETIF_F_HW_TLS_TX;
10040 }
10041
10042 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
10043 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
10044 features &= ~NETIF_F_HW_TLS_RX;
10045 }
10046
10047 if ((features & NETIF_F_GSO_UDP_L4) && !netdev_has_ip_or_hw_csum(features)) {
10048 netdev_dbg(dev, "Dropping USO feature since no CSUM feature.\n");
10049 features &= ~NETIF_F_GSO_UDP_L4;
10050 }
10051
10052 return features;
10053 }
10054
__netdev_update_features(struct net_device * dev)10055 int __netdev_update_features(struct net_device *dev)
10056 {
10057 struct net_device *upper, *lower;
10058 netdev_features_t features;
10059 struct list_head *iter;
10060 int err = -1;
10061
10062 ASSERT_RTNL();
10063
10064 features = netdev_get_wanted_features(dev);
10065
10066 if (dev->netdev_ops->ndo_fix_features)
10067 features = dev->netdev_ops->ndo_fix_features(dev, features);
10068
10069 /* driver might be less strict about feature dependencies */
10070 features = netdev_fix_features(dev, features);
10071
10072 /* some features can't be enabled if they're off on an upper device */
10073 netdev_for_each_upper_dev_rcu(dev, upper, iter)
10074 features = netdev_sync_upper_features(dev, upper, features);
10075
10076 if (dev->features == features)
10077 goto sync_lower;
10078
10079 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10080 &dev->features, &features);
10081
10082 if (dev->netdev_ops->ndo_set_features)
10083 err = dev->netdev_ops->ndo_set_features(dev, features);
10084 else
10085 err = 0;
10086
10087 if (unlikely(err < 0)) {
10088 netdev_err(dev,
10089 "set_features() failed (%d); wanted %pNF, left %pNF\n",
10090 err, &features, &dev->features);
10091 /* return non-0 since some features might have changed and
10092 * it's better to fire a spurious notification than miss it
10093 */
10094 return -1;
10095 }
10096
10097 sync_lower:
10098 /* some features must be disabled on lower devices when disabled
10099 * on an upper device (think: bonding master or bridge)
10100 */
10101 netdev_for_each_lower_dev(dev, lower, iter)
10102 netdev_sync_lower_features(dev, lower, features);
10103
10104 if (!err) {
10105 netdev_features_t diff = features ^ dev->features;
10106
10107 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10108 /* udp_tunnel_{get,drop}_rx_info both need
10109 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10110 * device, or they won't do anything.
10111 * Thus we need to update dev->features
10112 * *before* calling udp_tunnel_get_rx_info,
10113 * but *after* calling udp_tunnel_drop_rx_info.
10114 */
10115 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10116 dev->features = features;
10117 udp_tunnel_get_rx_info(dev);
10118 } else {
10119 udp_tunnel_drop_rx_info(dev);
10120 }
10121 }
10122
10123 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10124 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10125 dev->features = features;
10126 err |= vlan_get_rx_ctag_filter_info(dev);
10127 } else {
10128 vlan_drop_rx_ctag_filter_info(dev);
10129 }
10130 }
10131
10132 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10133 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10134 dev->features = features;
10135 err |= vlan_get_rx_stag_filter_info(dev);
10136 } else {
10137 vlan_drop_rx_stag_filter_info(dev);
10138 }
10139 }
10140
10141 dev->features = features;
10142 }
10143
10144 return err < 0 ? 0 : 1;
10145 }
10146
10147 /**
10148 * netdev_update_features - recalculate device features
10149 * @dev: the device to check
10150 *
10151 * Recalculate dev->features set and send notifications if it
10152 * has changed. Should be called after driver or hardware dependent
10153 * conditions might have changed that influence the features.
10154 */
netdev_update_features(struct net_device * dev)10155 void netdev_update_features(struct net_device *dev)
10156 {
10157 if (__netdev_update_features(dev))
10158 netdev_features_change(dev);
10159 }
10160 EXPORT_SYMBOL(netdev_update_features);
10161
10162 /**
10163 * netdev_change_features - recalculate device features
10164 * @dev: the device to check
10165 *
10166 * Recalculate dev->features set and send notifications even
10167 * if they have not changed. Should be called instead of
10168 * netdev_update_features() if also dev->vlan_features might
10169 * have changed to allow the changes to be propagated to stacked
10170 * VLAN devices.
10171 */
netdev_change_features(struct net_device * dev)10172 void netdev_change_features(struct net_device *dev)
10173 {
10174 __netdev_update_features(dev);
10175 netdev_features_change(dev);
10176 }
10177 EXPORT_SYMBOL(netdev_change_features);
10178
10179 /**
10180 * netif_stacked_transfer_operstate - transfer operstate
10181 * @rootdev: the root or lower level device to transfer state from
10182 * @dev: the device to transfer operstate to
10183 *
10184 * Transfer operational state from root to device. This is normally
10185 * called when a stacking relationship exists between the root
10186 * device and the device(a leaf device).
10187 */
netif_stacked_transfer_operstate(const struct net_device * rootdev,struct net_device * dev)10188 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10189 struct net_device *dev)
10190 {
10191 if (rootdev->operstate == IF_OPER_DORMANT)
10192 netif_dormant_on(dev);
10193 else
10194 netif_dormant_off(dev);
10195
10196 if (rootdev->operstate == IF_OPER_TESTING)
10197 netif_testing_on(dev);
10198 else
10199 netif_testing_off(dev);
10200
10201 if (netif_carrier_ok(rootdev))
10202 netif_carrier_on(dev);
10203 else
10204 netif_carrier_off(dev);
10205 }
10206 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10207
netif_alloc_rx_queues(struct net_device * dev)10208 static int netif_alloc_rx_queues(struct net_device *dev)
10209 {
10210 unsigned int i, count = dev->num_rx_queues;
10211 struct netdev_rx_queue *rx;
10212 size_t sz = count * sizeof(*rx);
10213 int err = 0;
10214
10215 BUG_ON(count < 1);
10216
10217 rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10218 if (!rx)
10219 return -ENOMEM;
10220
10221 dev->_rx = rx;
10222
10223 for (i = 0; i < count; i++) {
10224 rx[i].dev = dev;
10225
10226 /* XDP RX-queue setup */
10227 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10228 if (err < 0)
10229 goto err_rxq_info;
10230 }
10231 return 0;
10232
10233 err_rxq_info:
10234 /* Rollback successful reg's and free other resources */
10235 while (i--)
10236 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10237 kvfree(dev->_rx);
10238 dev->_rx = NULL;
10239 return err;
10240 }
10241
netif_free_rx_queues(struct net_device * dev)10242 static void netif_free_rx_queues(struct net_device *dev)
10243 {
10244 unsigned int i, count = dev->num_rx_queues;
10245
10246 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10247 if (!dev->_rx)
10248 return;
10249
10250 for (i = 0; i < count; i++)
10251 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10252
10253 kvfree(dev->_rx);
10254 }
10255
netdev_init_one_queue(struct net_device * dev,struct netdev_queue * queue,void * _unused)10256 static void netdev_init_one_queue(struct net_device *dev,
10257 struct netdev_queue *queue, void *_unused)
10258 {
10259 /* Initialize queue lock */
10260 spin_lock_init(&queue->_xmit_lock);
10261 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10262 queue->xmit_lock_owner = -1;
10263 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10264 queue->dev = dev;
10265 #ifdef CONFIG_BQL
10266 dql_init(&queue->dql, HZ);
10267 #endif
10268 }
10269
netif_free_tx_queues(struct net_device * dev)10270 static void netif_free_tx_queues(struct net_device *dev)
10271 {
10272 kvfree(dev->_tx);
10273 }
10274
netif_alloc_netdev_queues(struct net_device * dev)10275 static int netif_alloc_netdev_queues(struct net_device *dev)
10276 {
10277 unsigned int count = dev->num_tx_queues;
10278 struct netdev_queue *tx;
10279 size_t sz = count * sizeof(*tx);
10280
10281 if (count < 1 || count > 0xffff)
10282 return -EINVAL;
10283
10284 tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10285 if (!tx)
10286 return -ENOMEM;
10287
10288 dev->_tx = tx;
10289
10290 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10291 spin_lock_init(&dev->tx_global_lock);
10292
10293 return 0;
10294 }
10295
netif_tx_stop_all_queues(struct net_device * dev)10296 void netif_tx_stop_all_queues(struct net_device *dev)
10297 {
10298 unsigned int i;
10299
10300 for (i = 0; i < dev->num_tx_queues; i++) {
10301 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10302
10303 netif_tx_stop_queue(txq);
10304 }
10305 }
10306 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10307
netdev_do_alloc_pcpu_stats(struct net_device * dev)10308 static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
10309 {
10310 void __percpu *v;
10311
10312 /* Drivers implementing ndo_get_peer_dev must support tstat
10313 * accounting, so that skb_do_redirect() can bump the dev's
10314 * RX stats upon network namespace switch.
10315 */
10316 if (dev->netdev_ops->ndo_get_peer_dev &&
10317 dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
10318 return -EOPNOTSUPP;
10319
10320 switch (dev->pcpu_stat_type) {
10321 case NETDEV_PCPU_STAT_NONE:
10322 return 0;
10323 case NETDEV_PCPU_STAT_LSTATS:
10324 v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
10325 break;
10326 case NETDEV_PCPU_STAT_TSTATS:
10327 v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
10328 break;
10329 case NETDEV_PCPU_STAT_DSTATS:
10330 v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
10331 break;
10332 default:
10333 return -EINVAL;
10334 }
10335
10336 return v ? 0 : -ENOMEM;
10337 }
10338
netdev_do_free_pcpu_stats(struct net_device * dev)10339 static void netdev_do_free_pcpu_stats(struct net_device *dev)
10340 {
10341 switch (dev->pcpu_stat_type) {
10342 case NETDEV_PCPU_STAT_NONE:
10343 return;
10344 case NETDEV_PCPU_STAT_LSTATS:
10345 free_percpu(dev->lstats);
10346 break;
10347 case NETDEV_PCPU_STAT_TSTATS:
10348 free_percpu(dev->tstats);
10349 break;
10350 case NETDEV_PCPU_STAT_DSTATS:
10351 free_percpu(dev->dstats);
10352 break;
10353 }
10354 }
10355
netdev_free_phy_link_topology(struct net_device * dev)10356 static void netdev_free_phy_link_topology(struct net_device *dev)
10357 {
10358 struct phy_link_topology *topo = dev->link_topo;
10359
10360 if (IS_ENABLED(CONFIG_PHYLIB) && topo) {
10361 xa_destroy(&topo->phys);
10362 kfree(topo);
10363 dev->link_topo = NULL;
10364 }
10365 }
10366
10367 /**
10368 * register_netdevice() - register a network device
10369 * @dev: device to register
10370 *
10371 * Take a prepared network device structure and make it externally accessible.
10372 * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
10373 * Callers must hold the rtnl lock - you may want register_netdev()
10374 * instead of this.
10375 */
register_netdevice(struct net_device * dev)10376 int register_netdevice(struct net_device *dev)
10377 {
10378 int ret;
10379 struct net *net = dev_net(dev);
10380
10381 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10382 NETDEV_FEATURE_COUNT);
10383 BUG_ON(dev_boot_phase);
10384 ASSERT_RTNL();
10385
10386 might_sleep();
10387
10388 /* When net_device's are persistent, this will be fatal. */
10389 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10390 BUG_ON(!net);
10391
10392 ret = ethtool_check_ops(dev->ethtool_ops);
10393 if (ret)
10394 return ret;
10395
10396 /* rss ctx ID 0 is reserved for the default context, start from 1 */
10397 xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1);
10398 mutex_init(&dev->ethtool->rss_lock);
10399
10400 spin_lock_init(&dev->addr_list_lock);
10401 netdev_set_addr_lockdep_class(dev);
10402
10403 ret = dev_get_valid_name(net, dev, dev->name);
10404 if (ret < 0)
10405 goto out;
10406
10407 ret = -ENOMEM;
10408 dev->name_node = netdev_name_node_head_alloc(dev);
10409 if (!dev->name_node)
10410 goto out;
10411
10412 /* Init, if this function is available */
10413 if (dev->netdev_ops->ndo_init) {
10414 ret = dev->netdev_ops->ndo_init(dev);
10415 if (ret) {
10416 if (ret > 0)
10417 ret = -EIO;
10418 goto err_free_name;
10419 }
10420 }
10421
10422 if (((dev->hw_features | dev->features) &
10423 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10424 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10425 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10426 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10427 ret = -EINVAL;
10428 goto err_uninit;
10429 }
10430
10431 ret = netdev_do_alloc_pcpu_stats(dev);
10432 if (ret)
10433 goto err_uninit;
10434
10435 ret = dev_index_reserve(net, dev->ifindex);
10436 if (ret < 0)
10437 goto err_free_pcpu;
10438 dev->ifindex = ret;
10439
10440 /* Transfer changeable features to wanted_features and enable
10441 * software offloads (GSO and GRO).
10442 */
10443 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10444 dev->features |= NETIF_F_SOFT_FEATURES;
10445
10446 if (dev->udp_tunnel_nic_info) {
10447 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10448 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10449 }
10450
10451 dev->wanted_features = dev->features & dev->hw_features;
10452
10453 if (!(dev->flags & IFF_LOOPBACK))
10454 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10455
10456 /* If IPv4 TCP segmentation offload is supported we should also
10457 * allow the device to enable segmenting the frame with the option
10458 * of ignoring a static IP ID value. This doesn't enable the
10459 * feature itself but allows the user to enable it later.
10460 */
10461 if (dev->hw_features & NETIF_F_TSO)
10462 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10463 if (dev->vlan_features & NETIF_F_TSO)
10464 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10465 if (dev->mpls_features & NETIF_F_TSO)
10466 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10467 if (dev->hw_enc_features & NETIF_F_TSO)
10468 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10469
10470 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10471 */
10472 dev->vlan_features |= NETIF_F_HIGHDMA;
10473
10474 /* Make NETIF_F_SG inheritable to tunnel devices.
10475 */
10476 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10477
10478 /* Make NETIF_F_SG inheritable to MPLS.
10479 */
10480 dev->mpls_features |= NETIF_F_SG;
10481
10482 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10483 ret = notifier_to_errno(ret);
10484 if (ret)
10485 goto err_ifindex_release;
10486
10487 ret = netdev_register_kobject(dev);
10488
10489 WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED);
10490
10491 if (ret)
10492 goto err_uninit_notify;
10493
10494 __netdev_update_features(dev);
10495
10496 /*
10497 * Default initial state at registry is that the
10498 * device is present.
10499 */
10500
10501 set_bit(__LINK_STATE_PRESENT, &dev->state);
10502
10503 linkwatch_init_dev(dev);
10504
10505 dev_init_scheduler(dev);
10506
10507 netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10508 list_netdevice(dev);
10509
10510 add_device_randomness(dev->dev_addr, dev->addr_len);
10511
10512 /* If the device has permanent device address, driver should
10513 * set dev_addr and also addr_assign_type should be set to
10514 * NET_ADDR_PERM (default value).
10515 */
10516 if (dev->addr_assign_type == NET_ADDR_PERM)
10517 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10518
10519 /* Notify protocols, that a new device appeared. */
10520 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10521 ret = notifier_to_errno(ret);
10522 if (ret) {
10523 /* Expect explicit free_netdev() on failure */
10524 dev->needs_free_netdev = false;
10525 unregister_netdevice_queue(dev, NULL);
10526 goto out;
10527 }
10528 /*
10529 * Prevent userspace races by waiting until the network
10530 * device is fully setup before sending notifications.
10531 */
10532 if (!dev->rtnl_link_ops ||
10533 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10534 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10535
10536 out:
10537 return ret;
10538
10539 err_uninit_notify:
10540 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10541 err_ifindex_release:
10542 dev_index_release(net, dev->ifindex);
10543 err_free_pcpu:
10544 netdev_do_free_pcpu_stats(dev);
10545 err_uninit:
10546 if (dev->netdev_ops->ndo_uninit)
10547 dev->netdev_ops->ndo_uninit(dev);
10548 if (dev->priv_destructor)
10549 dev->priv_destructor(dev);
10550 err_free_name:
10551 netdev_name_node_free(dev->name_node);
10552 goto out;
10553 }
10554 EXPORT_SYMBOL(register_netdevice);
10555
10556 /* Initialize the core of a dummy net device.
10557 * This is useful if you are calling this function after alloc_netdev(),
10558 * since it does not memset the net_device fields.
10559 */
init_dummy_netdev_core(struct net_device * dev)10560 static void init_dummy_netdev_core(struct net_device *dev)
10561 {
10562 /* make sure we BUG if trying to hit standard
10563 * register/unregister code path
10564 */
10565 dev->reg_state = NETREG_DUMMY;
10566
10567 /* NAPI wants this */
10568 INIT_LIST_HEAD(&dev->napi_list);
10569
10570 /* a dummy interface is started by default */
10571 set_bit(__LINK_STATE_PRESENT, &dev->state);
10572 set_bit(__LINK_STATE_START, &dev->state);
10573
10574 /* napi_busy_loop stats accounting wants this */
10575 dev_net_set(dev, &init_net);
10576
10577 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10578 * because users of this 'device' dont need to change
10579 * its refcount.
10580 */
10581 }
10582
10583 /**
10584 * init_dummy_netdev - init a dummy network device for NAPI
10585 * @dev: device to init
10586 *
10587 * This takes a network device structure and initializes the minimum
10588 * amount of fields so it can be used to schedule NAPI polls without
10589 * registering a full blown interface. This is to be used by drivers
10590 * that need to tie several hardware interfaces to a single NAPI
10591 * poll scheduler due to HW limitations.
10592 */
init_dummy_netdev(struct net_device * dev)10593 void init_dummy_netdev(struct net_device *dev)
10594 {
10595 /* Clear everything. Note we don't initialize spinlocks
10596 * as they aren't supposed to be taken by any of the
10597 * NAPI code and this dummy netdev is supposed to be
10598 * only ever used for NAPI polls
10599 */
10600 memset(dev, 0, sizeof(struct net_device));
10601 init_dummy_netdev_core(dev);
10602 }
10603 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10604
10605 /**
10606 * register_netdev - register a network device
10607 * @dev: device to register
10608 *
10609 * Take a completed network device structure and add it to the kernel
10610 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10611 * chain. 0 is returned on success. A negative errno code is returned
10612 * on a failure to set up the device, or if the name is a duplicate.
10613 *
10614 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10615 * and expands the device name if you passed a format string to
10616 * alloc_netdev.
10617 */
register_netdev(struct net_device * dev)10618 int register_netdev(struct net_device *dev)
10619 {
10620 int err;
10621
10622 if (rtnl_lock_killable())
10623 return -EINTR;
10624 err = register_netdevice(dev);
10625 rtnl_unlock();
10626 return err;
10627 }
10628 EXPORT_SYMBOL(register_netdev);
10629
netdev_refcnt_read(const struct net_device * dev)10630 int netdev_refcnt_read(const struct net_device *dev)
10631 {
10632 #ifdef CONFIG_PCPU_DEV_REFCNT
10633 int i, refcnt = 0;
10634
10635 for_each_possible_cpu(i)
10636 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10637 return refcnt;
10638 #else
10639 return refcount_read(&dev->dev_refcnt);
10640 #endif
10641 }
10642 EXPORT_SYMBOL(netdev_refcnt_read);
10643
10644 int netdev_unregister_timeout_secs __read_mostly = 10;
10645
10646 #define WAIT_REFS_MIN_MSECS 1
10647 #define WAIT_REFS_MAX_MSECS 250
10648 /**
10649 * netdev_wait_allrefs_any - wait until all references are gone.
10650 * @list: list of net_devices to wait on
10651 *
10652 * This is called when unregistering network devices.
10653 *
10654 * Any protocol or device that holds a reference should register
10655 * for netdevice notification, and cleanup and put back the
10656 * reference if they receive an UNREGISTER event.
10657 * We can get stuck here if buggy protocols don't correctly
10658 * call dev_put.
10659 */
netdev_wait_allrefs_any(struct list_head * list)10660 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10661 {
10662 unsigned long rebroadcast_time, warning_time;
10663 struct net_device *dev;
10664 int wait = 0;
10665
10666 rebroadcast_time = warning_time = jiffies;
10667
10668 list_for_each_entry(dev, list, todo_list)
10669 if (netdev_refcnt_read(dev) == 1)
10670 return dev;
10671
10672 while (true) {
10673 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10674 rtnl_lock();
10675
10676 /* Rebroadcast unregister notification */
10677 list_for_each_entry(dev, list, todo_list)
10678 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10679
10680 __rtnl_unlock();
10681 rcu_barrier();
10682 rtnl_lock();
10683
10684 list_for_each_entry(dev, list, todo_list)
10685 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10686 &dev->state)) {
10687 /* We must not have linkwatch events
10688 * pending on unregister. If this
10689 * happens, we simply run the queue
10690 * unscheduled, resulting in a noop
10691 * for this device.
10692 */
10693 linkwatch_run_queue();
10694 break;
10695 }
10696
10697 __rtnl_unlock();
10698
10699 rebroadcast_time = jiffies;
10700 }
10701
10702 rcu_barrier();
10703
10704 if (!wait) {
10705 wait = WAIT_REFS_MIN_MSECS;
10706 } else {
10707 msleep(wait);
10708 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10709 }
10710
10711 list_for_each_entry(dev, list, todo_list)
10712 if (netdev_refcnt_read(dev) == 1)
10713 return dev;
10714
10715 if (time_after(jiffies, warning_time +
10716 READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10717 list_for_each_entry(dev, list, todo_list) {
10718 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10719 dev->name, netdev_refcnt_read(dev));
10720 ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10721 }
10722
10723 warning_time = jiffies;
10724 }
10725 }
10726 }
10727
10728 /* The sequence is:
10729 *
10730 * rtnl_lock();
10731 * ...
10732 * register_netdevice(x1);
10733 * register_netdevice(x2);
10734 * ...
10735 * unregister_netdevice(y1);
10736 * unregister_netdevice(y2);
10737 * ...
10738 * rtnl_unlock();
10739 * free_netdev(y1);
10740 * free_netdev(y2);
10741 *
10742 * We are invoked by rtnl_unlock().
10743 * This allows us to deal with problems:
10744 * 1) We can delete sysfs objects which invoke hotplug
10745 * without deadlocking with linkwatch via keventd.
10746 * 2) Since we run with the RTNL semaphore not held, we can sleep
10747 * safely in order to wait for the netdev refcnt to drop to zero.
10748 *
10749 * We must not return until all unregister events added during
10750 * the interval the lock was held have been completed.
10751 */
netdev_run_todo(void)10752 void netdev_run_todo(void)
10753 {
10754 struct net_device *dev, *tmp;
10755 struct list_head list;
10756 int cnt;
10757 #ifdef CONFIG_LOCKDEP
10758 struct list_head unlink_list;
10759
10760 list_replace_init(&net_unlink_list, &unlink_list);
10761
10762 while (!list_empty(&unlink_list)) {
10763 struct net_device *dev = list_first_entry(&unlink_list,
10764 struct net_device,
10765 unlink_list);
10766 list_del_init(&dev->unlink_list);
10767 dev->nested_level = dev->lower_level - 1;
10768 }
10769 #endif
10770
10771 /* Snapshot list, allow later requests */
10772 list_replace_init(&net_todo_list, &list);
10773
10774 __rtnl_unlock();
10775
10776 /* Wait for rcu callbacks to finish before next phase */
10777 if (!list_empty(&list))
10778 rcu_barrier();
10779
10780 list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10781 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10782 netdev_WARN(dev, "run_todo but not unregistering\n");
10783 list_del(&dev->todo_list);
10784 continue;
10785 }
10786
10787 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED);
10788 linkwatch_sync_dev(dev);
10789 }
10790
10791 cnt = 0;
10792 while (!list_empty(&list)) {
10793 dev = netdev_wait_allrefs_any(&list);
10794 list_del(&dev->todo_list);
10795
10796 /* paranoia */
10797 BUG_ON(netdev_refcnt_read(dev) != 1);
10798 BUG_ON(!list_empty(&dev->ptype_all));
10799 BUG_ON(!list_empty(&dev->ptype_specific));
10800 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10801 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10802
10803 netdev_do_free_pcpu_stats(dev);
10804 if (dev->priv_destructor)
10805 dev->priv_destructor(dev);
10806 if (dev->needs_free_netdev)
10807 free_netdev(dev);
10808
10809 cnt++;
10810
10811 /* Free network device */
10812 kobject_put(&dev->dev.kobj);
10813 }
10814 if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count))
10815 wake_up(&netdev_unregistering_wq);
10816 }
10817
10818 /* Collate per-cpu network dstats statistics
10819 *
10820 * Read per-cpu network statistics from dev->dstats and populate the related
10821 * fields in @s.
10822 */
dev_fetch_dstats(struct rtnl_link_stats64 * s,const struct pcpu_dstats __percpu * dstats)10823 static void dev_fetch_dstats(struct rtnl_link_stats64 *s,
10824 const struct pcpu_dstats __percpu *dstats)
10825 {
10826 int cpu;
10827
10828 for_each_possible_cpu(cpu) {
10829 u64 rx_packets, rx_bytes, rx_drops;
10830 u64 tx_packets, tx_bytes, tx_drops;
10831 const struct pcpu_dstats *stats;
10832 unsigned int start;
10833
10834 stats = per_cpu_ptr(dstats, cpu);
10835 do {
10836 start = u64_stats_fetch_begin(&stats->syncp);
10837 rx_packets = u64_stats_read(&stats->rx_packets);
10838 rx_bytes = u64_stats_read(&stats->rx_bytes);
10839 rx_drops = u64_stats_read(&stats->rx_drops);
10840 tx_packets = u64_stats_read(&stats->tx_packets);
10841 tx_bytes = u64_stats_read(&stats->tx_bytes);
10842 tx_drops = u64_stats_read(&stats->tx_drops);
10843 } while (u64_stats_fetch_retry(&stats->syncp, start));
10844
10845 s->rx_packets += rx_packets;
10846 s->rx_bytes += rx_bytes;
10847 s->rx_dropped += rx_drops;
10848 s->tx_packets += tx_packets;
10849 s->tx_bytes += tx_bytes;
10850 s->tx_dropped += tx_drops;
10851 }
10852 }
10853
10854 /* ndo_get_stats64 implementation for dtstats-based accounting.
10855 *
10856 * Populate @s from dev->stats and dev->dstats. This is used internally by the
10857 * core for NETDEV_PCPU_STAT_DSTAT-type stats collection.
10858 */
dev_get_dstats64(const struct net_device * dev,struct rtnl_link_stats64 * s)10859 static void dev_get_dstats64(const struct net_device *dev,
10860 struct rtnl_link_stats64 *s)
10861 {
10862 netdev_stats_to_stats64(s, &dev->stats);
10863 dev_fetch_dstats(s, dev->dstats);
10864 }
10865
10866 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10867 * all the same fields in the same order as net_device_stats, with only
10868 * the type differing, but rtnl_link_stats64 may have additional fields
10869 * at the end for newer counters.
10870 */
netdev_stats_to_stats64(struct rtnl_link_stats64 * stats64,const struct net_device_stats * netdev_stats)10871 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10872 const struct net_device_stats *netdev_stats)
10873 {
10874 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10875 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10876 u64 *dst = (u64 *)stats64;
10877
10878 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10879 for (i = 0; i < n; i++)
10880 dst[i] = (unsigned long)atomic_long_read(&src[i]);
10881 /* zero out counters that only exist in rtnl_link_stats64 */
10882 memset((char *)stats64 + n * sizeof(u64), 0,
10883 sizeof(*stats64) - n * sizeof(u64));
10884 }
10885 EXPORT_SYMBOL(netdev_stats_to_stats64);
10886
netdev_core_stats_alloc(struct net_device * dev)10887 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc(
10888 struct net_device *dev)
10889 {
10890 struct net_device_core_stats __percpu *p;
10891
10892 p = alloc_percpu_gfp(struct net_device_core_stats,
10893 GFP_ATOMIC | __GFP_NOWARN);
10894
10895 if (p && cmpxchg(&dev->core_stats, NULL, p))
10896 free_percpu(p);
10897
10898 /* This READ_ONCE() pairs with the cmpxchg() above */
10899 return READ_ONCE(dev->core_stats);
10900 }
10901
netdev_core_stats_inc(struct net_device * dev,u32 offset)10902 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset)
10903 {
10904 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10905 struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
10906 unsigned long __percpu *field;
10907
10908 if (unlikely(!p)) {
10909 p = netdev_core_stats_alloc(dev);
10910 if (!p)
10911 return;
10912 }
10913
10914 field = (unsigned long __percpu *)((void __percpu *)p + offset);
10915 this_cpu_inc(*field);
10916 }
10917 EXPORT_SYMBOL_GPL(netdev_core_stats_inc);
10918
10919 /**
10920 * dev_get_stats - get network device statistics
10921 * @dev: device to get statistics from
10922 * @storage: place to store stats
10923 *
10924 * Get network statistics from device. Return @storage.
10925 * The device driver may provide its own method by setting
10926 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10927 * otherwise the internal statistics structure is used.
10928 */
dev_get_stats(struct net_device * dev,struct rtnl_link_stats64 * storage)10929 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10930 struct rtnl_link_stats64 *storage)
10931 {
10932 const struct net_device_ops *ops = dev->netdev_ops;
10933 const struct net_device_core_stats __percpu *p;
10934
10935 if (ops->ndo_get_stats64) {
10936 memset(storage, 0, sizeof(*storage));
10937 ops->ndo_get_stats64(dev, storage);
10938 } else if (ops->ndo_get_stats) {
10939 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10940 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) {
10941 dev_get_tstats64(dev, storage);
10942 } else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) {
10943 dev_get_dstats64(dev, storage);
10944 } else {
10945 netdev_stats_to_stats64(storage, &dev->stats);
10946 }
10947
10948 /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10949 p = READ_ONCE(dev->core_stats);
10950 if (p) {
10951 const struct net_device_core_stats *core_stats;
10952 int i;
10953
10954 for_each_possible_cpu(i) {
10955 core_stats = per_cpu_ptr(p, i);
10956 storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10957 storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10958 storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10959 storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10960 }
10961 }
10962 return storage;
10963 }
10964 EXPORT_SYMBOL(dev_get_stats);
10965
10966 /**
10967 * dev_fetch_sw_netstats - get per-cpu network device statistics
10968 * @s: place to store stats
10969 * @netstats: per-cpu network stats to read from
10970 *
10971 * Read per-cpu network statistics and populate the related fields in @s.
10972 */
dev_fetch_sw_netstats(struct rtnl_link_stats64 * s,const struct pcpu_sw_netstats __percpu * netstats)10973 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10974 const struct pcpu_sw_netstats __percpu *netstats)
10975 {
10976 int cpu;
10977
10978 for_each_possible_cpu(cpu) {
10979 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10980 const struct pcpu_sw_netstats *stats;
10981 unsigned int start;
10982
10983 stats = per_cpu_ptr(netstats, cpu);
10984 do {
10985 start = u64_stats_fetch_begin(&stats->syncp);
10986 rx_packets = u64_stats_read(&stats->rx_packets);
10987 rx_bytes = u64_stats_read(&stats->rx_bytes);
10988 tx_packets = u64_stats_read(&stats->tx_packets);
10989 tx_bytes = u64_stats_read(&stats->tx_bytes);
10990 } while (u64_stats_fetch_retry(&stats->syncp, start));
10991
10992 s->rx_packets += rx_packets;
10993 s->rx_bytes += rx_bytes;
10994 s->tx_packets += tx_packets;
10995 s->tx_bytes += tx_bytes;
10996 }
10997 }
10998 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10999
11000 /**
11001 * dev_get_tstats64 - ndo_get_stats64 implementation
11002 * @dev: device to get statistics from
11003 * @s: place to store stats
11004 *
11005 * Populate @s from dev->stats and dev->tstats. Can be used as
11006 * ndo_get_stats64() callback.
11007 */
dev_get_tstats64(struct net_device * dev,struct rtnl_link_stats64 * s)11008 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
11009 {
11010 netdev_stats_to_stats64(s, &dev->stats);
11011 dev_fetch_sw_netstats(s, dev->tstats);
11012 }
11013 EXPORT_SYMBOL_GPL(dev_get_tstats64);
11014
dev_ingress_queue_create(struct net_device * dev)11015 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
11016 {
11017 struct netdev_queue *queue = dev_ingress_queue(dev);
11018
11019 #ifdef CONFIG_NET_CLS_ACT
11020 if (queue)
11021 return queue;
11022 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
11023 if (!queue)
11024 return NULL;
11025 netdev_init_one_queue(dev, queue, NULL);
11026 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
11027 RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
11028 rcu_assign_pointer(dev->ingress_queue, queue);
11029 #endif
11030 return queue;
11031 }
11032
11033 static const struct ethtool_ops default_ethtool_ops;
11034
netdev_set_default_ethtool_ops(struct net_device * dev,const struct ethtool_ops * ops)11035 void netdev_set_default_ethtool_ops(struct net_device *dev,
11036 const struct ethtool_ops *ops)
11037 {
11038 if (dev->ethtool_ops == &default_ethtool_ops)
11039 dev->ethtool_ops = ops;
11040 }
11041 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
11042
11043 /**
11044 * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
11045 * @dev: netdev to enable the IRQ coalescing on
11046 *
11047 * Sets a conservative default for SW IRQ coalescing. Users can use
11048 * sysfs attributes to override the default values.
11049 */
netdev_sw_irq_coalesce_default_on(struct net_device * dev)11050 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
11051 {
11052 WARN_ON(dev->reg_state == NETREG_REGISTERED);
11053
11054 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
11055 dev->gro_flush_timeout = 20000;
11056 dev->napi_defer_hard_irqs = 1;
11057 }
11058 }
11059 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
11060
11061 /**
11062 * alloc_netdev_mqs - allocate network device
11063 * @sizeof_priv: size of private data to allocate space for
11064 * @name: device name format string
11065 * @name_assign_type: origin of device name
11066 * @setup: callback to initialize device
11067 * @txqs: the number of TX subqueues to allocate
11068 * @rxqs: the number of RX subqueues to allocate
11069 *
11070 * Allocates a struct net_device with private data area for driver use
11071 * and performs basic initialization. Also allocates subqueue structs
11072 * for each queue on the device.
11073 */
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)11074 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
11075 unsigned char name_assign_type,
11076 void (*setup)(struct net_device *),
11077 unsigned int txqs, unsigned int rxqs)
11078 {
11079 struct net_device *dev;
11080
11081 BUG_ON(strlen(name) >= sizeof(dev->name));
11082
11083 if (txqs < 1) {
11084 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
11085 return NULL;
11086 }
11087
11088 if (rxqs < 1) {
11089 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
11090 return NULL;
11091 }
11092
11093 dev = kvzalloc(struct_size(dev, priv, sizeof_priv),
11094 GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11095 if (!dev)
11096 return NULL;
11097
11098 dev->priv_len = sizeof_priv;
11099
11100 ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
11101 #ifdef CONFIG_PCPU_DEV_REFCNT
11102 dev->pcpu_refcnt = alloc_percpu(int);
11103 if (!dev->pcpu_refcnt)
11104 goto free_dev;
11105 __dev_hold(dev);
11106 #else
11107 refcount_set(&dev->dev_refcnt, 1);
11108 #endif
11109
11110 if (dev_addr_init(dev))
11111 goto free_pcpu;
11112
11113 dev_mc_init(dev);
11114 dev_uc_init(dev);
11115
11116 dev_net_set(dev, &init_net);
11117
11118 dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
11119 dev->xdp_zc_max_segs = 1;
11120 dev->gso_max_segs = GSO_MAX_SEGS;
11121 dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
11122 dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
11123 dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
11124 dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
11125 dev->tso_max_segs = TSO_MAX_SEGS;
11126 dev->upper_level = 1;
11127 dev->lower_level = 1;
11128 #ifdef CONFIG_LOCKDEP
11129 dev->nested_level = 0;
11130 INIT_LIST_HEAD(&dev->unlink_list);
11131 #endif
11132
11133 INIT_LIST_HEAD(&dev->napi_list);
11134 INIT_LIST_HEAD(&dev->unreg_list);
11135 INIT_LIST_HEAD(&dev->close_list);
11136 INIT_LIST_HEAD(&dev->link_watch_list);
11137 INIT_LIST_HEAD(&dev->adj_list.upper);
11138 INIT_LIST_HEAD(&dev->adj_list.lower);
11139 INIT_LIST_HEAD(&dev->ptype_all);
11140 INIT_LIST_HEAD(&dev->ptype_specific);
11141 INIT_LIST_HEAD(&dev->net_notifier_list);
11142 #ifdef CONFIG_NET_SCHED
11143 hash_init(dev->qdisc_hash);
11144 #endif
11145
11146 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
11147 setup(dev);
11148
11149 if (!dev->tx_queue_len) {
11150 dev->priv_flags |= IFF_NO_QUEUE;
11151 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
11152 }
11153
11154 dev->num_tx_queues = txqs;
11155 dev->real_num_tx_queues = txqs;
11156 if (netif_alloc_netdev_queues(dev))
11157 goto free_all;
11158
11159 dev->num_rx_queues = rxqs;
11160 dev->real_num_rx_queues = rxqs;
11161 if (netif_alloc_rx_queues(dev))
11162 goto free_all;
11163 dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT);
11164 if (!dev->ethtool)
11165 goto free_all;
11166
11167 strscpy(dev->name, name);
11168 dev->name_assign_type = name_assign_type;
11169 dev->group = INIT_NETDEV_GROUP;
11170 if (!dev->ethtool_ops)
11171 dev->ethtool_ops = &default_ethtool_ops;
11172
11173 nf_hook_netdev_init(dev);
11174
11175 return dev;
11176
11177 free_all:
11178 free_netdev(dev);
11179 return NULL;
11180
11181 free_pcpu:
11182 #ifdef CONFIG_PCPU_DEV_REFCNT
11183 free_percpu(dev->pcpu_refcnt);
11184 free_dev:
11185 #endif
11186 kvfree(dev);
11187 return NULL;
11188 }
11189 EXPORT_SYMBOL(alloc_netdev_mqs);
11190
11191 /**
11192 * free_netdev - free network device
11193 * @dev: device
11194 *
11195 * This function does the last stage of destroying an allocated device
11196 * interface. The reference to the device object is released. If this
11197 * is the last reference then it will be freed.Must be called in process
11198 * context.
11199 */
free_netdev(struct net_device * dev)11200 void free_netdev(struct net_device *dev)
11201 {
11202 struct napi_struct *p, *n;
11203
11204 might_sleep();
11205
11206 /* When called immediately after register_netdevice() failed the unwind
11207 * handling may still be dismantling the device. Handle that case by
11208 * deferring the free.
11209 */
11210 if (dev->reg_state == NETREG_UNREGISTERING) {
11211 ASSERT_RTNL();
11212 dev->needs_free_netdev = true;
11213 return;
11214 }
11215
11216 kfree(dev->ethtool);
11217 netif_free_tx_queues(dev);
11218 netif_free_rx_queues(dev);
11219
11220 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
11221
11222 /* Flush device addresses */
11223 dev_addr_flush(dev);
11224
11225 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
11226 netif_napi_del(p);
11227
11228 ref_tracker_dir_exit(&dev->refcnt_tracker);
11229 #ifdef CONFIG_PCPU_DEV_REFCNT
11230 free_percpu(dev->pcpu_refcnt);
11231 dev->pcpu_refcnt = NULL;
11232 #endif
11233 free_percpu(dev->core_stats);
11234 dev->core_stats = NULL;
11235 free_percpu(dev->xdp_bulkq);
11236 dev->xdp_bulkq = NULL;
11237
11238 netdev_free_phy_link_topology(dev);
11239
11240 /* Compatibility with error handling in drivers */
11241 if (dev->reg_state == NETREG_UNINITIALIZED ||
11242 dev->reg_state == NETREG_DUMMY) {
11243 kvfree(dev);
11244 return;
11245 }
11246
11247 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
11248 WRITE_ONCE(dev->reg_state, NETREG_RELEASED);
11249
11250 /* will free via device release */
11251 put_device(&dev->dev);
11252 }
11253 EXPORT_SYMBOL(free_netdev);
11254
11255 /**
11256 * alloc_netdev_dummy - Allocate and initialize a dummy net device.
11257 * @sizeof_priv: size of private data to allocate space for
11258 *
11259 * Return: the allocated net_device on success, NULL otherwise
11260 */
alloc_netdev_dummy(int sizeof_priv)11261 struct net_device *alloc_netdev_dummy(int sizeof_priv)
11262 {
11263 return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN,
11264 init_dummy_netdev_core);
11265 }
11266 EXPORT_SYMBOL_GPL(alloc_netdev_dummy);
11267
11268 /**
11269 * synchronize_net - Synchronize with packet receive processing
11270 *
11271 * Wait for packets currently being received to be done.
11272 * Does not block later packets from starting.
11273 */
synchronize_net(void)11274 void synchronize_net(void)
11275 {
11276 might_sleep();
11277 if (rtnl_is_locked())
11278 synchronize_rcu_expedited();
11279 else
11280 synchronize_rcu();
11281 }
11282 EXPORT_SYMBOL(synchronize_net);
11283
netdev_rss_contexts_free(struct net_device * dev)11284 static void netdev_rss_contexts_free(struct net_device *dev)
11285 {
11286 struct ethtool_rxfh_context *ctx;
11287 unsigned long context;
11288
11289 mutex_lock(&dev->ethtool->rss_lock);
11290 xa_for_each(&dev->ethtool->rss_ctx, context, ctx) {
11291 struct ethtool_rxfh_param rxfh;
11292
11293 rxfh.indir = ethtool_rxfh_context_indir(ctx);
11294 rxfh.key = ethtool_rxfh_context_key(ctx);
11295 rxfh.hfunc = ctx->hfunc;
11296 rxfh.input_xfrm = ctx->input_xfrm;
11297 rxfh.rss_context = context;
11298 rxfh.rss_delete = true;
11299
11300 xa_erase(&dev->ethtool->rss_ctx, context);
11301 if (dev->ethtool_ops->create_rxfh_context)
11302 dev->ethtool_ops->remove_rxfh_context(dev, ctx,
11303 context, NULL);
11304 else
11305 dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL);
11306 kfree(ctx);
11307 }
11308 xa_destroy(&dev->ethtool->rss_ctx);
11309 mutex_unlock(&dev->ethtool->rss_lock);
11310 }
11311
11312 /**
11313 * unregister_netdevice_queue - remove device from the kernel
11314 * @dev: device
11315 * @head: list
11316 *
11317 * This function shuts down a device interface and removes it
11318 * from the kernel tables.
11319 * If head not NULL, device is queued to be unregistered later.
11320 *
11321 * Callers must hold the rtnl semaphore. You may want
11322 * unregister_netdev() instead of this.
11323 */
11324
unregister_netdevice_queue(struct net_device * dev,struct list_head * head)11325 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11326 {
11327 ASSERT_RTNL();
11328
11329 if (head) {
11330 list_move_tail(&dev->unreg_list, head);
11331 } else {
11332 LIST_HEAD(single);
11333
11334 list_add(&dev->unreg_list, &single);
11335 unregister_netdevice_many(&single);
11336 }
11337 }
11338 EXPORT_SYMBOL(unregister_netdevice_queue);
11339
unregister_netdevice_many_notify(struct list_head * head,u32 portid,const struct nlmsghdr * nlh)11340 void unregister_netdevice_many_notify(struct list_head *head,
11341 u32 portid, const struct nlmsghdr *nlh)
11342 {
11343 struct net_device *dev, *tmp;
11344 LIST_HEAD(close_head);
11345 int cnt = 0;
11346
11347 BUG_ON(dev_boot_phase);
11348 ASSERT_RTNL();
11349
11350 if (list_empty(head))
11351 return;
11352
11353 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11354 /* Some devices call without registering
11355 * for initialization unwind. Remove those
11356 * devices and proceed with the remaining.
11357 */
11358 if (dev->reg_state == NETREG_UNINITIALIZED) {
11359 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11360 dev->name, dev);
11361
11362 WARN_ON(1);
11363 list_del(&dev->unreg_list);
11364 continue;
11365 }
11366 dev->dismantle = true;
11367 BUG_ON(dev->reg_state != NETREG_REGISTERED);
11368 }
11369
11370 /* If device is running, close it first. */
11371 list_for_each_entry(dev, head, unreg_list)
11372 list_add_tail(&dev->close_list, &close_head);
11373 dev_close_many(&close_head, true);
11374
11375 list_for_each_entry(dev, head, unreg_list) {
11376 /* And unlink it from device chain. */
11377 unlist_netdevice(dev);
11378 WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING);
11379 }
11380 flush_all_backlogs();
11381
11382 synchronize_net();
11383
11384 list_for_each_entry(dev, head, unreg_list) {
11385 struct sk_buff *skb = NULL;
11386
11387 /* Shutdown queueing discipline. */
11388 dev_shutdown(dev);
11389 dev_tcx_uninstall(dev);
11390 dev_xdp_uninstall(dev);
11391 bpf_dev_bound_netdev_unregister(dev);
11392 dev_dmabuf_uninstall(dev);
11393
11394 netdev_offload_xstats_disable_all(dev);
11395
11396 /* Notify protocols, that we are about to destroy
11397 * this device. They should clean all the things.
11398 */
11399 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11400
11401 if (!dev->rtnl_link_ops ||
11402 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11403 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11404 GFP_KERNEL, NULL, 0,
11405 portid, nlh);
11406
11407 /*
11408 * Flush the unicast and multicast chains
11409 */
11410 dev_uc_flush(dev);
11411 dev_mc_flush(dev);
11412
11413 netdev_name_node_alt_flush(dev);
11414 netdev_name_node_free(dev->name_node);
11415
11416 netdev_rss_contexts_free(dev);
11417
11418 call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
11419
11420 if (dev->netdev_ops->ndo_uninit)
11421 dev->netdev_ops->ndo_uninit(dev);
11422
11423 mutex_destroy(&dev->ethtool->rss_lock);
11424
11425 if (skb)
11426 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
11427
11428 /* Notifier chain MUST detach us all upper devices. */
11429 WARN_ON(netdev_has_any_upper_dev(dev));
11430 WARN_ON(netdev_has_any_lower_dev(dev));
11431
11432 /* Remove entries from kobject tree */
11433 netdev_unregister_kobject(dev);
11434 #ifdef CONFIG_XPS
11435 /* Remove XPS queueing entries */
11436 netif_reset_xps_queues_gt(dev, 0);
11437 #endif
11438 }
11439
11440 synchronize_net();
11441
11442 list_for_each_entry(dev, head, unreg_list) {
11443 netdev_put(dev, &dev->dev_registered_tracker);
11444 net_set_todo(dev);
11445 cnt++;
11446 }
11447 atomic_add(cnt, &dev_unreg_count);
11448
11449 list_del(head);
11450 }
11451
11452 /**
11453 * unregister_netdevice_many - unregister many devices
11454 * @head: list of devices
11455 *
11456 * Note: As most callers use a stack allocated list_head,
11457 * we force a list_del() to make sure stack won't be corrupted later.
11458 */
unregister_netdevice_many(struct list_head * head)11459 void unregister_netdevice_many(struct list_head *head)
11460 {
11461 unregister_netdevice_many_notify(head, 0, NULL);
11462 }
11463 EXPORT_SYMBOL(unregister_netdevice_many);
11464
11465 /**
11466 * unregister_netdev - remove device from the kernel
11467 * @dev: device
11468 *
11469 * This function shuts down a device interface and removes it
11470 * from the kernel tables.
11471 *
11472 * This is just a wrapper for unregister_netdevice that takes
11473 * the rtnl semaphore. In general you want to use this and not
11474 * unregister_netdevice.
11475 */
unregister_netdev(struct net_device * dev)11476 void unregister_netdev(struct net_device *dev)
11477 {
11478 rtnl_lock();
11479 unregister_netdevice(dev);
11480 rtnl_unlock();
11481 }
11482 EXPORT_SYMBOL(unregister_netdev);
11483
11484 /**
11485 * __dev_change_net_namespace - move device to different nethost namespace
11486 * @dev: device
11487 * @net: network namespace
11488 * @pat: If not NULL name pattern to try if the current device name
11489 * is already taken in the destination network namespace.
11490 * @new_ifindex: If not zero, specifies device index in the target
11491 * namespace.
11492 *
11493 * This function shuts down a device interface and moves it
11494 * to a new network namespace. On success 0 is returned, on
11495 * a failure a netagive errno code is returned.
11496 *
11497 * Callers must hold the rtnl semaphore.
11498 */
11499
__dev_change_net_namespace(struct net_device * dev,struct net * net,const char * pat,int new_ifindex)11500 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11501 const char *pat, int new_ifindex)
11502 {
11503 struct netdev_name_node *name_node;
11504 struct net *net_old = dev_net(dev);
11505 char new_name[IFNAMSIZ] = {};
11506 int err, new_nsid;
11507
11508 ASSERT_RTNL();
11509
11510 /* Don't allow namespace local devices to be moved. */
11511 err = -EINVAL;
11512 if (dev->netns_local)
11513 goto out;
11514
11515 /* Ensure the device has been registered */
11516 if (dev->reg_state != NETREG_REGISTERED)
11517 goto out;
11518
11519 /* Get out if there is nothing todo */
11520 err = 0;
11521 if (net_eq(net_old, net))
11522 goto out;
11523
11524 /* Pick the destination device name, and ensure
11525 * we can use it in the destination network namespace.
11526 */
11527 err = -EEXIST;
11528 if (netdev_name_in_use(net, dev->name)) {
11529 /* We get here if we can't use the current device name */
11530 if (!pat)
11531 goto out;
11532 err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST);
11533 if (err < 0)
11534 goto out;
11535 }
11536 /* Check that none of the altnames conflicts. */
11537 err = -EEXIST;
11538 netdev_for_each_altname(dev, name_node)
11539 if (netdev_name_in_use(net, name_node->name))
11540 goto out;
11541
11542 /* Check that new_ifindex isn't used yet. */
11543 if (new_ifindex) {
11544 err = dev_index_reserve(net, new_ifindex);
11545 if (err < 0)
11546 goto out;
11547 } else {
11548 /* If there is an ifindex conflict assign a new one */
11549 err = dev_index_reserve(net, dev->ifindex);
11550 if (err == -EBUSY)
11551 err = dev_index_reserve(net, 0);
11552 if (err < 0)
11553 goto out;
11554 new_ifindex = err;
11555 }
11556
11557 /*
11558 * And now a mini version of register_netdevice unregister_netdevice.
11559 */
11560
11561 /* If device is running close it first. */
11562 dev_close(dev);
11563
11564 /* And unlink it from device chain */
11565 unlist_netdevice(dev);
11566
11567 synchronize_net();
11568
11569 /* Shutdown queueing discipline. */
11570 dev_shutdown(dev);
11571
11572 /* Notify protocols, that we are about to destroy
11573 * this device. They should clean all the things.
11574 *
11575 * Note that dev->reg_state stays at NETREG_REGISTERED.
11576 * This is wanted because this way 8021q and macvlan know
11577 * the device is just moving and can keep their slaves up.
11578 */
11579 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11580 rcu_barrier();
11581
11582 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11583
11584 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11585 new_ifindex);
11586
11587 /*
11588 * Flush the unicast and multicast chains
11589 */
11590 dev_uc_flush(dev);
11591 dev_mc_flush(dev);
11592
11593 /* Send a netdev-removed uevent to the old namespace */
11594 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11595 netdev_adjacent_del_links(dev);
11596
11597 /* Move per-net netdevice notifiers that are following the netdevice */
11598 move_netdevice_notifiers_dev_net(dev, net);
11599
11600 /* Actually switch the network namespace */
11601 dev_net_set(dev, net);
11602 dev->ifindex = new_ifindex;
11603
11604 if (new_name[0]) {
11605 /* Rename the netdev to prepared name */
11606 write_seqlock_bh(&netdev_rename_lock);
11607 strscpy(dev->name, new_name, IFNAMSIZ);
11608 write_sequnlock_bh(&netdev_rename_lock);
11609 }
11610
11611 /* Fixup kobjects */
11612 dev_set_uevent_suppress(&dev->dev, 1);
11613 err = device_rename(&dev->dev, dev->name);
11614 dev_set_uevent_suppress(&dev->dev, 0);
11615 WARN_ON(err);
11616
11617 /* Send a netdev-add uevent to the new namespace */
11618 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11619 netdev_adjacent_add_links(dev);
11620
11621 /* Adapt owner in case owning user namespace of target network
11622 * namespace is different from the original one.
11623 */
11624 err = netdev_change_owner(dev, net_old, net);
11625 WARN_ON(err);
11626
11627 /* Add the device back in the hashes */
11628 list_netdevice(dev);
11629
11630 /* Notify protocols, that a new device appeared. */
11631 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11632
11633 /*
11634 * Prevent userspace races by waiting until the network
11635 * device is fully setup before sending notifications.
11636 */
11637 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11638
11639 synchronize_net();
11640 err = 0;
11641 out:
11642 return err;
11643 }
11644 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11645
dev_cpu_dead(unsigned int oldcpu)11646 static int dev_cpu_dead(unsigned int oldcpu)
11647 {
11648 struct sk_buff **list_skb;
11649 struct sk_buff *skb;
11650 unsigned int cpu;
11651 struct softnet_data *sd, *oldsd, *remsd = NULL;
11652
11653 local_irq_disable();
11654 cpu = smp_processor_id();
11655 sd = &per_cpu(softnet_data, cpu);
11656 oldsd = &per_cpu(softnet_data, oldcpu);
11657
11658 /* Find end of our completion_queue. */
11659 list_skb = &sd->completion_queue;
11660 while (*list_skb)
11661 list_skb = &(*list_skb)->next;
11662 /* Append completion queue from offline CPU. */
11663 *list_skb = oldsd->completion_queue;
11664 oldsd->completion_queue = NULL;
11665
11666 /* Append output queue from offline CPU. */
11667 if (oldsd->output_queue) {
11668 *sd->output_queue_tailp = oldsd->output_queue;
11669 sd->output_queue_tailp = oldsd->output_queue_tailp;
11670 oldsd->output_queue = NULL;
11671 oldsd->output_queue_tailp = &oldsd->output_queue;
11672 }
11673 /* Append NAPI poll list from offline CPU, with one exception :
11674 * process_backlog() must be called by cpu owning percpu backlog.
11675 * We properly handle process_queue & input_pkt_queue later.
11676 */
11677 while (!list_empty(&oldsd->poll_list)) {
11678 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11679 struct napi_struct,
11680 poll_list);
11681
11682 list_del_init(&napi->poll_list);
11683 if (napi->poll == process_backlog)
11684 napi->state &= NAPIF_STATE_THREADED;
11685 else
11686 ____napi_schedule(sd, napi);
11687 }
11688
11689 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11690 local_irq_enable();
11691
11692 if (!use_backlog_threads()) {
11693 #ifdef CONFIG_RPS
11694 remsd = oldsd->rps_ipi_list;
11695 oldsd->rps_ipi_list = NULL;
11696 #endif
11697 /* send out pending IPI's on offline CPU */
11698 net_rps_send_ipi(remsd);
11699 }
11700
11701 /* Process offline CPU's input_pkt_queue */
11702 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11703 netif_rx(skb);
11704 rps_input_queue_head_incr(oldsd);
11705 }
11706 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11707 netif_rx(skb);
11708 rps_input_queue_head_incr(oldsd);
11709 }
11710
11711 return 0;
11712 }
11713
11714 /**
11715 * netdev_increment_features - increment feature set by one
11716 * @all: current feature set
11717 * @one: new feature set
11718 * @mask: mask feature set
11719 *
11720 * Computes a new feature set after adding a device with feature set
11721 * @one to the master device with current feature set @all. Will not
11722 * enable anything that is off in @mask. Returns the new feature set.
11723 */
netdev_increment_features(netdev_features_t all,netdev_features_t one,netdev_features_t mask)11724 netdev_features_t netdev_increment_features(netdev_features_t all,
11725 netdev_features_t one, netdev_features_t mask)
11726 {
11727 if (mask & NETIF_F_HW_CSUM)
11728 mask |= NETIF_F_CSUM_MASK;
11729 mask |= NETIF_F_VLAN_CHALLENGED;
11730
11731 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11732 all &= one | ~NETIF_F_ALL_FOR_ALL;
11733
11734 /* If one device supports hw checksumming, set for all. */
11735 if (all & NETIF_F_HW_CSUM)
11736 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11737
11738 return all;
11739 }
11740 EXPORT_SYMBOL(netdev_increment_features);
11741
netdev_create_hash(void)11742 static struct hlist_head * __net_init netdev_create_hash(void)
11743 {
11744 int i;
11745 struct hlist_head *hash;
11746
11747 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11748 if (hash != NULL)
11749 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11750 INIT_HLIST_HEAD(&hash[i]);
11751
11752 return hash;
11753 }
11754
11755 /* Initialize per network namespace state */
netdev_init(struct net * net)11756 static int __net_init netdev_init(struct net *net)
11757 {
11758 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11759 8 * sizeof_field(struct napi_struct, gro_bitmask));
11760
11761 INIT_LIST_HEAD(&net->dev_base_head);
11762
11763 net->dev_name_head = netdev_create_hash();
11764 if (net->dev_name_head == NULL)
11765 goto err_name;
11766
11767 net->dev_index_head = netdev_create_hash();
11768 if (net->dev_index_head == NULL)
11769 goto err_idx;
11770
11771 xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1);
11772
11773 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11774
11775 return 0;
11776
11777 err_idx:
11778 kfree(net->dev_name_head);
11779 err_name:
11780 return -ENOMEM;
11781 }
11782
11783 /**
11784 * netdev_drivername - network driver for the device
11785 * @dev: network device
11786 *
11787 * Determine network driver for device.
11788 */
netdev_drivername(const struct net_device * dev)11789 const char *netdev_drivername(const struct net_device *dev)
11790 {
11791 const struct device_driver *driver;
11792 const struct device *parent;
11793 const char *empty = "";
11794
11795 parent = dev->dev.parent;
11796 if (!parent)
11797 return empty;
11798
11799 driver = parent->driver;
11800 if (driver && driver->name)
11801 return driver->name;
11802 return empty;
11803 }
11804
__netdev_printk(const char * level,const struct net_device * dev,struct va_format * vaf)11805 static void __netdev_printk(const char *level, const struct net_device *dev,
11806 struct va_format *vaf)
11807 {
11808 if (dev && dev->dev.parent) {
11809 dev_printk_emit(level[1] - '0',
11810 dev->dev.parent,
11811 "%s %s %s%s: %pV",
11812 dev_driver_string(dev->dev.parent),
11813 dev_name(dev->dev.parent),
11814 netdev_name(dev), netdev_reg_state(dev),
11815 vaf);
11816 } else if (dev) {
11817 printk("%s%s%s: %pV",
11818 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11819 } else {
11820 printk("%s(NULL net_device): %pV", level, vaf);
11821 }
11822 }
11823
netdev_printk(const char * level,const struct net_device * dev,const char * format,...)11824 void netdev_printk(const char *level, const struct net_device *dev,
11825 const char *format, ...)
11826 {
11827 struct va_format vaf;
11828 va_list args;
11829
11830 va_start(args, format);
11831
11832 vaf.fmt = format;
11833 vaf.va = &args;
11834
11835 __netdev_printk(level, dev, &vaf);
11836
11837 va_end(args);
11838 }
11839 EXPORT_SYMBOL(netdev_printk);
11840
11841 #define define_netdev_printk_level(func, level) \
11842 void func(const struct net_device *dev, const char *fmt, ...) \
11843 { \
11844 struct va_format vaf; \
11845 va_list args; \
11846 \
11847 va_start(args, fmt); \
11848 \
11849 vaf.fmt = fmt; \
11850 vaf.va = &args; \
11851 \
11852 __netdev_printk(level, dev, &vaf); \
11853 \
11854 va_end(args); \
11855 } \
11856 EXPORT_SYMBOL(func);
11857
11858 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11859 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11860 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11861 define_netdev_printk_level(netdev_err, KERN_ERR);
11862 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11863 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11864 define_netdev_printk_level(netdev_info, KERN_INFO);
11865
netdev_exit(struct net * net)11866 static void __net_exit netdev_exit(struct net *net)
11867 {
11868 kfree(net->dev_name_head);
11869 kfree(net->dev_index_head);
11870 xa_destroy(&net->dev_by_index);
11871 if (net != &init_net)
11872 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11873 }
11874
11875 static struct pernet_operations __net_initdata netdev_net_ops = {
11876 .init = netdev_init,
11877 .exit = netdev_exit,
11878 };
11879
default_device_exit_net(struct net * net)11880 static void __net_exit default_device_exit_net(struct net *net)
11881 {
11882 struct netdev_name_node *name_node, *tmp;
11883 struct net_device *dev, *aux;
11884 /*
11885 * Push all migratable network devices back to the
11886 * initial network namespace
11887 */
11888 ASSERT_RTNL();
11889 for_each_netdev_safe(net, dev, aux) {
11890 int err;
11891 char fb_name[IFNAMSIZ];
11892
11893 /* Ignore unmoveable devices (i.e. loopback) */
11894 if (dev->netns_local)
11895 continue;
11896
11897 /* Leave virtual devices for the generic cleanup */
11898 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11899 continue;
11900
11901 /* Push remaining network devices to init_net */
11902 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11903 if (netdev_name_in_use(&init_net, fb_name))
11904 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11905
11906 netdev_for_each_altname_safe(dev, name_node, tmp)
11907 if (netdev_name_in_use(&init_net, name_node->name))
11908 __netdev_name_node_alt_destroy(name_node);
11909
11910 err = dev_change_net_namespace(dev, &init_net, fb_name);
11911 if (err) {
11912 pr_emerg("%s: failed to move %s to init_net: %d\n",
11913 __func__, dev->name, err);
11914 BUG();
11915 }
11916 }
11917 }
11918
default_device_exit_batch(struct list_head * net_list)11919 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11920 {
11921 /* At exit all network devices most be removed from a network
11922 * namespace. Do this in the reverse order of registration.
11923 * Do this across as many network namespaces as possible to
11924 * improve batching efficiency.
11925 */
11926 struct net_device *dev;
11927 struct net *net;
11928 LIST_HEAD(dev_kill_list);
11929
11930 rtnl_lock();
11931 list_for_each_entry(net, net_list, exit_list) {
11932 default_device_exit_net(net);
11933 cond_resched();
11934 }
11935
11936 list_for_each_entry(net, net_list, exit_list) {
11937 for_each_netdev_reverse(net, dev) {
11938 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11939 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11940 else
11941 unregister_netdevice_queue(dev, &dev_kill_list);
11942 }
11943 }
11944 unregister_netdevice_many(&dev_kill_list);
11945 rtnl_unlock();
11946 }
11947
11948 static struct pernet_operations __net_initdata default_device_ops = {
11949 .exit_batch = default_device_exit_batch,
11950 };
11951
net_dev_struct_check(void)11952 static void __init net_dev_struct_check(void)
11953 {
11954 /* TX read-mostly hotpath */
11955 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags_fast);
11956 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops);
11957 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops);
11958 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx);
11959 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues);
11960 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size);
11961 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size);
11962 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs);
11963 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features);
11964 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc);
11965 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu);
11966 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom);
11967 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq);
11968 #ifdef CONFIG_XPS
11969 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps);
11970 #endif
11971 #ifdef CONFIG_NETFILTER_EGRESS
11972 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress);
11973 #endif
11974 #ifdef CONFIG_NET_XGRESS
11975 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress);
11976 #endif
11977 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160);
11978
11979 /* TXRX read-mostly hotpath */
11980 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats);
11981 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state);
11982 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags);
11983 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len);
11984 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features);
11985 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr);
11986 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46);
11987
11988 /* RX read-mostly hotpath */
11989 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific);
11990 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex);
11991 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues);
11992 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx);
11993 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout);
11994 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs);
11995 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size);
11996 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size);
11997 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler);
11998 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data);
11999 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net);
12000 #ifdef CONFIG_NETPOLL
12001 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo);
12002 #endif
12003 #ifdef CONFIG_NET_XGRESS
12004 CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress);
12005 #endif
12006 CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104);
12007 }
12008
12009 /*
12010 * Initialize the DEV module. At boot time this walks the device list and
12011 * unhooks any devices that fail to initialise (normally hardware not
12012 * present) and leaves us with a valid list of present and active devices.
12013 *
12014 */
12015
12016 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
12017 #define SYSTEM_PERCPU_PAGE_POOL_SIZE ((1 << 20) / PAGE_SIZE)
12018
net_page_pool_create(int cpuid)12019 static int net_page_pool_create(int cpuid)
12020 {
12021 #if IS_ENABLED(CONFIG_PAGE_POOL)
12022 struct page_pool_params page_pool_params = {
12023 .pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
12024 .flags = PP_FLAG_SYSTEM_POOL,
12025 .nid = cpu_to_mem(cpuid),
12026 };
12027 struct page_pool *pp_ptr;
12028
12029 pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
12030 if (IS_ERR(pp_ptr))
12031 return -ENOMEM;
12032
12033 per_cpu(system_page_pool, cpuid) = pp_ptr;
12034 #endif
12035 return 0;
12036 }
12037
backlog_napi_should_run(unsigned int cpu)12038 static int backlog_napi_should_run(unsigned int cpu)
12039 {
12040 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12041 struct napi_struct *napi = &sd->backlog;
12042
12043 return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
12044 }
12045
run_backlog_napi(unsigned int cpu)12046 static void run_backlog_napi(unsigned int cpu)
12047 {
12048 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12049
12050 napi_threaded_poll_loop(&sd->backlog);
12051 }
12052
backlog_napi_setup(unsigned int cpu)12053 static void backlog_napi_setup(unsigned int cpu)
12054 {
12055 struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12056 struct napi_struct *napi = &sd->backlog;
12057
12058 napi->thread = this_cpu_read(backlog_napi);
12059 set_bit(NAPI_STATE_THREADED, &napi->state);
12060 }
12061
12062 static struct smp_hotplug_thread backlog_threads = {
12063 .store = &backlog_napi,
12064 .thread_should_run = backlog_napi_should_run,
12065 .thread_fn = run_backlog_napi,
12066 .thread_comm = "backlog_napi/%u",
12067 .setup = backlog_napi_setup,
12068 };
12069
12070 /*
12071 * This is called single threaded during boot, so no need
12072 * to take the rtnl semaphore.
12073 */
net_dev_init(void)12074 static int __init net_dev_init(void)
12075 {
12076 int i, rc = -ENOMEM;
12077
12078 BUG_ON(!dev_boot_phase);
12079
12080 net_dev_struct_check();
12081
12082 if (dev_proc_init())
12083 goto out;
12084
12085 if (netdev_kobject_init())
12086 goto out;
12087
12088 for (i = 0; i < PTYPE_HASH_SIZE; i++)
12089 INIT_LIST_HEAD(&ptype_base[i]);
12090
12091 if (register_pernet_subsys(&netdev_net_ops))
12092 goto out;
12093
12094 /*
12095 * Initialise the packet receive queues.
12096 */
12097
12098 for_each_possible_cpu(i) {
12099 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
12100 struct softnet_data *sd = &per_cpu(softnet_data, i);
12101
12102 INIT_WORK(flush, flush_backlog);
12103
12104 skb_queue_head_init(&sd->input_pkt_queue);
12105 skb_queue_head_init(&sd->process_queue);
12106 #ifdef CONFIG_XFRM_OFFLOAD
12107 skb_queue_head_init(&sd->xfrm_backlog);
12108 #endif
12109 INIT_LIST_HEAD(&sd->poll_list);
12110 sd->output_queue_tailp = &sd->output_queue;
12111 #ifdef CONFIG_RPS
12112 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
12113 sd->cpu = i;
12114 #endif
12115 INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
12116 spin_lock_init(&sd->defer_lock);
12117
12118 init_gro_hash(&sd->backlog);
12119 sd->backlog.poll = process_backlog;
12120 sd->backlog.weight = weight_p;
12121 INIT_LIST_HEAD(&sd->backlog.poll_list);
12122
12123 if (net_page_pool_create(i))
12124 goto out;
12125 }
12126 if (use_backlog_threads())
12127 smpboot_register_percpu_thread(&backlog_threads);
12128
12129 dev_boot_phase = 0;
12130
12131 /* The loopback device is special if any other network devices
12132 * is present in a network namespace the loopback device must
12133 * be present. Since we now dynamically allocate and free the
12134 * loopback device ensure this invariant is maintained by
12135 * keeping the loopback device as the first device on the
12136 * list of network devices. Ensuring the loopback devices
12137 * is the first device that appears and the last network device
12138 * that disappears.
12139 */
12140 if (register_pernet_device(&loopback_net_ops))
12141 goto out;
12142
12143 if (register_pernet_device(&default_device_ops))
12144 goto out;
12145
12146 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
12147 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
12148
12149 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
12150 NULL, dev_cpu_dead);
12151 WARN_ON(rc < 0);
12152 rc = 0;
12153
12154 /* avoid static key IPIs to isolated CPUs */
12155 if (housekeeping_enabled(HK_TYPE_MISC))
12156 net_enable_timestamp();
12157 out:
12158 if (rc < 0) {
12159 for_each_possible_cpu(i) {
12160 struct page_pool *pp_ptr;
12161
12162 pp_ptr = per_cpu(system_page_pool, i);
12163 if (!pp_ptr)
12164 continue;
12165
12166 page_pool_destroy(pp_ptr);
12167 per_cpu(system_page_pool, i) = NULL;
12168 }
12169 }
12170
12171 return rc;
12172 }
12173
12174 subsys_initcall(net_dev_init);
12175