1 /*
2 * NET3 Protocol independent device support routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Derived from the non IP parts of dev.c 1.0.19
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 *
14 * Additional Authors:
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
21 *
22 * Changes:
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
34 * drivers
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
44 * call a packet.
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
58 * 1 device.
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
66 * the backlog queue.
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
73 */
74
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.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/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
98 #include <net/sock.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
101 #include <net/dst.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
120 #include <net/ip.h>
121 #include <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138
139 #include "net-sysfs.h"
140
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
143
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
146
147 static DEFINE_SPINLOCK(ptype_lock);
148 static DEFINE_SPINLOCK(offload_lock);
149 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
150 struct list_head ptype_all __read_mostly; /* Taps */
151 static struct list_head offload_base __read_mostly;
152
153 static int netif_rx_internal(struct sk_buff *skb);
154 static int call_netdevice_notifiers_info(unsigned long val,
155 struct net_device *dev,
156 struct netdev_notifier_info *info);
157
158 /*
159 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * semaphore.
161 *
162 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
163 *
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base_head list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
168 *
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
172 *
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
175 * semaphore held.
176 */
177 DEFINE_RWLOCK(dev_base_lock);
178 EXPORT_SYMBOL(dev_base_lock);
179
180 /* protects napi_hash addition/deletion and napi_gen_id */
181 static DEFINE_SPINLOCK(napi_hash_lock);
182
183 static unsigned int napi_gen_id;
184 static DEFINE_HASHTABLE(napi_hash, 8);
185
186 static seqcount_t devnet_rename_seq;
187
dev_base_seq_inc(struct net * net)188 static inline void dev_base_seq_inc(struct net *net)
189 {
190 while (++net->dev_base_seq == 0);
191 }
192
dev_name_hash(struct net * net,const char * name)193 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
194 {
195 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
196
197 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
198 }
199
dev_index_hash(struct net * net,int ifindex)200 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
201 {
202 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
203 }
204
rps_lock(struct softnet_data * sd)205 static inline void rps_lock(struct softnet_data *sd)
206 {
207 #ifdef CONFIG_RPS
208 spin_lock(&sd->input_pkt_queue.lock);
209 #endif
210 }
211
rps_unlock(struct softnet_data * sd)212 static inline void rps_unlock(struct softnet_data *sd)
213 {
214 #ifdef CONFIG_RPS
215 spin_unlock(&sd->input_pkt_queue.lock);
216 #endif
217 }
218
219 /* Device list insertion */
list_netdevice(struct net_device * dev)220 static void list_netdevice(struct net_device *dev)
221 {
222 struct net *net = dev_net(dev);
223
224 ASSERT_RTNL();
225
226 write_lock_bh(&dev_base_lock);
227 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
228 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
229 hlist_add_head_rcu(&dev->index_hlist,
230 dev_index_hash(net, dev->ifindex));
231 write_unlock_bh(&dev_base_lock);
232
233 dev_base_seq_inc(net);
234 }
235
236 /* Device list removal
237 * caller must respect a RCU grace period before freeing/reusing dev
238 */
unlist_netdevice(struct net_device * dev)239 static void unlist_netdevice(struct net_device *dev)
240 {
241 ASSERT_RTNL();
242
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del_rcu(&dev->dev_list);
246 hlist_del_rcu(&dev->name_hlist);
247 hlist_del_rcu(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
249
250 dev_base_seq_inc(dev_net(dev));
251 }
252
253 /*
254 * Our notifier list
255 */
256
257 static RAW_NOTIFIER_HEAD(netdev_chain);
258
259 /*
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
262 */
263
264 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
265 EXPORT_PER_CPU_SYMBOL(softnet_data);
266
267 #ifdef CONFIG_LOCKDEP
268 /*
269 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
270 * according to dev->type
271 */
272 static const unsigned short netdev_lock_type[] =
273 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
274 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
275 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
276 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
277 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
278 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
279 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
280 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
281 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
282 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
283 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
284 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
285 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
286 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
287 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
288
289 static const char *const netdev_lock_name[] =
290 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
291 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
292 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
293 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
294 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
295 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
296 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
297 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
298 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
299 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
300 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
301 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
302 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
303 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
304 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
305
306 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
308
netdev_lock_pos(unsigned short dev_type)309 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
310 {
311 int i;
312
313 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
314 if (netdev_lock_type[i] == dev_type)
315 return i;
316 /* the last key is used by default */
317 return ARRAY_SIZE(netdev_lock_type) - 1;
318 }
319
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)320 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
321 unsigned short dev_type)
322 {
323 int i;
324
325 i = netdev_lock_pos(dev_type);
326 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
327 netdev_lock_name[i]);
328 }
329
netdev_set_addr_lockdep_class(struct net_device * dev)330 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
331 {
332 int i;
333
334 i = netdev_lock_pos(dev->type);
335 lockdep_set_class_and_name(&dev->addr_list_lock,
336 &netdev_addr_lock_key[i],
337 netdev_lock_name[i]);
338 }
339 #else
netdev_set_xmit_lockdep_class(spinlock_t * lock,unsigned short dev_type)340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
342 {
343 }
netdev_set_addr_lockdep_class(struct net_device * dev)344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
345 {
346 }
347 #endif
348
349 /*******************************************************************************
350
351 Protocol management and registration routines
352
353 *******************************************************************************/
354
355 /*
356 * Add a protocol ID to the list. Now that the input handler is
357 * smarter we can dispense with all the messy stuff that used to be
358 * here.
359 *
360 * BEWARE!!! Protocol handlers, mangling input packets,
361 * MUST BE last in hash buckets and checking protocol handlers
362 * MUST start from promiscuous ptype_all chain in net_bh.
363 * It is true now, do not change it.
364 * Explanation follows: if protocol handler, mangling packet, will
365 * be the first on list, it is not able to sense, that packet
366 * is cloned and should be copied-on-write, so that it will
367 * change it and subsequent readers will get broken packet.
368 * --ANK (980803)
369 */
370
ptype_head(const struct packet_type * pt)371 static inline struct list_head *ptype_head(const struct packet_type *pt)
372 {
373 if (pt->type == htons(ETH_P_ALL))
374 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
375 else
376 return pt->dev ? &pt->dev->ptype_specific :
377 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
378 }
379
380 /**
381 * dev_add_pack - add packet handler
382 * @pt: packet type declaration
383 *
384 * Add a protocol handler to the networking stack. The passed &packet_type
385 * is linked into kernel lists and may not be freed until it has been
386 * removed from the kernel lists.
387 *
388 * This call does not sleep therefore it can not
389 * guarantee all CPU's that are in middle of receiving packets
390 * will see the new packet type (until the next received packet).
391 */
392
dev_add_pack(struct packet_type * pt)393 void dev_add_pack(struct packet_type *pt)
394 {
395 struct list_head *head = ptype_head(pt);
396
397 spin_lock(&ptype_lock);
398 list_add_rcu(&pt->list, head);
399 spin_unlock(&ptype_lock);
400 }
401 EXPORT_SYMBOL(dev_add_pack);
402
403 /**
404 * __dev_remove_pack - remove packet handler
405 * @pt: packet type declaration
406 *
407 * Remove a protocol handler that was previously added to the kernel
408 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
409 * from the kernel lists and can be freed or reused once this function
410 * returns.
411 *
412 * The packet type might still be in use by receivers
413 * and must not be freed until after all the CPU's have gone
414 * through a quiescent state.
415 */
__dev_remove_pack(struct packet_type * pt)416 void __dev_remove_pack(struct packet_type *pt)
417 {
418 struct list_head *head = ptype_head(pt);
419 struct packet_type *pt1;
420
421 spin_lock(&ptype_lock);
422
423 list_for_each_entry(pt1, head, list) {
424 if (pt == pt1) {
425 list_del_rcu(&pt->list);
426 goto out;
427 }
428 }
429
430 pr_warn("dev_remove_pack: %p not found\n", pt);
431 out:
432 spin_unlock(&ptype_lock);
433 }
434 EXPORT_SYMBOL(__dev_remove_pack);
435
436 /**
437 * dev_remove_pack - remove packet handler
438 * @pt: packet type declaration
439 *
440 * Remove a protocol handler that was previously added to the kernel
441 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
442 * from the kernel lists and can be freed or reused once this function
443 * returns.
444 *
445 * This call sleeps to guarantee that no CPU is looking at the packet
446 * type after return.
447 */
dev_remove_pack(struct packet_type * pt)448 void dev_remove_pack(struct packet_type *pt)
449 {
450 __dev_remove_pack(pt);
451
452 synchronize_net();
453 }
454 EXPORT_SYMBOL(dev_remove_pack);
455
456
457 /**
458 * dev_add_offload - register offload handlers
459 * @po: protocol offload declaration
460 *
461 * Add protocol offload handlers to the networking stack. The passed
462 * &proto_offload is linked into kernel lists and may not be freed until
463 * it has been removed from the kernel lists.
464 *
465 * This call does not sleep therefore it can not
466 * guarantee all CPU's that are in middle of receiving packets
467 * will see the new offload handlers (until the next received packet).
468 */
dev_add_offload(struct packet_offload * po)469 void dev_add_offload(struct packet_offload *po)
470 {
471 struct list_head *head = &offload_base;
472
473 spin_lock(&offload_lock);
474 list_add_rcu(&po->list, head);
475 spin_unlock(&offload_lock);
476 }
477 EXPORT_SYMBOL(dev_add_offload);
478
479 /**
480 * __dev_remove_offload - remove offload handler
481 * @po: packet offload declaration
482 *
483 * Remove a protocol offload handler that was previously added to the
484 * kernel offload handlers by dev_add_offload(). The passed &offload_type
485 * is removed from the kernel lists and can be freed or reused once this
486 * function returns.
487 *
488 * The packet type might still be in use by receivers
489 * and must not be freed until after all the CPU's have gone
490 * through a quiescent state.
491 */
__dev_remove_offload(struct packet_offload * po)492 static void __dev_remove_offload(struct packet_offload *po)
493 {
494 struct list_head *head = &offload_base;
495 struct packet_offload *po1;
496
497 spin_lock(&offload_lock);
498
499 list_for_each_entry(po1, head, list) {
500 if (po == po1) {
501 list_del_rcu(&po->list);
502 goto out;
503 }
504 }
505
506 pr_warn("dev_remove_offload: %p not found\n", po);
507 out:
508 spin_unlock(&offload_lock);
509 }
510
511 /**
512 * dev_remove_offload - remove packet offload handler
513 * @po: packet offload declaration
514 *
515 * Remove a packet offload handler that was previously added to the kernel
516 * offload handlers by dev_add_offload(). The passed &offload_type is
517 * removed from the kernel lists and can be freed or reused once this
518 * function returns.
519 *
520 * This call sleeps to guarantee that no CPU is looking at the packet
521 * type after return.
522 */
dev_remove_offload(struct packet_offload * po)523 void dev_remove_offload(struct packet_offload *po)
524 {
525 __dev_remove_offload(po);
526
527 synchronize_net();
528 }
529 EXPORT_SYMBOL(dev_remove_offload);
530
531 /******************************************************************************
532
533 Device Boot-time Settings Routines
534
535 *******************************************************************************/
536
537 /* Boot time configuration table */
538 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
539
540 /**
541 * netdev_boot_setup_add - add new setup entry
542 * @name: name of the device
543 * @map: configured settings for the device
544 *
545 * Adds new setup entry to the dev_boot_setup list. The function
546 * returns 0 on error and 1 on success. This is a generic routine to
547 * all netdevices.
548 */
netdev_boot_setup_add(char * name,struct ifmap * map)549 static int netdev_boot_setup_add(char *name, struct ifmap *map)
550 {
551 struct netdev_boot_setup *s;
552 int i;
553
554 s = dev_boot_setup;
555 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
556 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
557 memset(s[i].name, 0, sizeof(s[i].name));
558 strlcpy(s[i].name, name, IFNAMSIZ);
559 memcpy(&s[i].map, map, sizeof(s[i].map));
560 break;
561 }
562 }
563
564 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
565 }
566
567 /**
568 * netdev_boot_setup_check - check boot time settings
569 * @dev: the netdevice
570 *
571 * Check boot time settings for the device.
572 * The found settings are set for the device to be used
573 * later in the device probing.
574 * Returns 0 if no settings found, 1 if they are.
575 */
netdev_boot_setup_check(struct net_device * dev)576 int netdev_boot_setup_check(struct net_device *dev)
577 {
578 struct netdev_boot_setup *s = dev_boot_setup;
579 int i;
580
581 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
582 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
583 !strcmp(dev->name, s[i].name)) {
584 dev->irq = s[i].map.irq;
585 dev->base_addr = s[i].map.base_addr;
586 dev->mem_start = s[i].map.mem_start;
587 dev->mem_end = s[i].map.mem_end;
588 return 1;
589 }
590 }
591 return 0;
592 }
593 EXPORT_SYMBOL(netdev_boot_setup_check);
594
595
596 /**
597 * netdev_boot_base - get address from boot time settings
598 * @prefix: prefix for network device
599 * @unit: id for network device
600 *
601 * Check boot time settings for the base address of device.
602 * The found settings are set for the device to be used
603 * later in the device probing.
604 * Returns 0 if no settings found.
605 */
netdev_boot_base(const char * prefix,int unit)606 unsigned long netdev_boot_base(const char *prefix, int unit)
607 {
608 const struct netdev_boot_setup *s = dev_boot_setup;
609 char name[IFNAMSIZ];
610 int i;
611
612 sprintf(name, "%s%d", prefix, unit);
613
614 /*
615 * If device already registered then return base of 1
616 * to indicate not to probe for this interface
617 */
618 if (__dev_get_by_name(&init_net, name))
619 return 1;
620
621 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
622 if (!strcmp(name, s[i].name))
623 return s[i].map.base_addr;
624 return 0;
625 }
626
627 /*
628 * Saves at boot time configured settings for any netdevice.
629 */
netdev_boot_setup(char * str)630 int __init netdev_boot_setup(char *str)
631 {
632 int ints[5];
633 struct ifmap map;
634
635 str = get_options(str, ARRAY_SIZE(ints), ints);
636 if (!str || !*str)
637 return 0;
638
639 /* Save settings */
640 memset(&map, 0, sizeof(map));
641 if (ints[0] > 0)
642 map.irq = ints[1];
643 if (ints[0] > 1)
644 map.base_addr = ints[2];
645 if (ints[0] > 2)
646 map.mem_start = ints[3];
647 if (ints[0] > 3)
648 map.mem_end = ints[4];
649
650 /* Add new entry to the list */
651 return netdev_boot_setup_add(str, &map);
652 }
653
654 __setup("netdev=", netdev_boot_setup);
655
656 /*******************************************************************************
657
658 Device Interface Subroutines
659
660 *******************************************************************************/
661
662 /**
663 * dev_get_iflink - get 'iflink' value of a interface
664 * @dev: targeted interface
665 *
666 * Indicates the ifindex the interface is linked to.
667 * Physical interfaces have the same 'ifindex' and 'iflink' values.
668 */
669
dev_get_iflink(const struct net_device * dev)670 int dev_get_iflink(const struct net_device *dev)
671 {
672 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
673 return dev->netdev_ops->ndo_get_iflink(dev);
674
675 return dev->ifindex;
676 }
677 EXPORT_SYMBOL(dev_get_iflink);
678
679 /**
680 * __dev_get_by_name - find a device by its name
681 * @net: the applicable net namespace
682 * @name: name to find
683 *
684 * Find an interface by name. Must be called under RTNL semaphore
685 * or @dev_base_lock. If the name is found a pointer to the device
686 * is returned. If the name is not found then %NULL is returned. The
687 * reference counters are not incremented so the caller must be
688 * careful with locks.
689 */
690
__dev_get_by_name(struct net * net,const char * name)691 struct net_device *__dev_get_by_name(struct net *net, const char *name)
692 {
693 struct net_device *dev;
694 struct hlist_head *head = dev_name_hash(net, name);
695
696 hlist_for_each_entry(dev, head, name_hlist)
697 if (!strncmp(dev->name, name, IFNAMSIZ))
698 return dev;
699
700 return NULL;
701 }
702 EXPORT_SYMBOL(__dev_get_by_name);
703
704 /**
705 * dev_get_by_name_rcu - find a device by its name
706 * @net: the applicable net namespace
707 * @name: name to find
708 *
709 * Find an interface by name.
710 * If the name is found a pointer to the device is returned.
711 * If the name is not found then %NULL is returned.
712 * The reference counters are not incremented so the caller must be
713 * careful with locks. The caller must hold RCU lock.
714 */
715
dev_get_by_name_rcu(struct net * net,const char * name)716 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
717 {
718 struct net_device *dev;
719 struct hlist_head *head = dev_name_hash(net, name);
720
721 hlist_for_each_entry_rcu(dev, head, name_hlist)
722 if (!strncmp(dev->name, name, IFNAMSIZ))
723 return dev;
724
725 return NULL;
726 }
727 EXPORT_SYMBOL(dev_get_by_name_rcu);
728
729 /**
730 * dev_get_by_name - find a device by its name
731 * @net: the applicable net namespace
732 * @name: name to find
733 *
734 * Find an interface by name. This can be called from any
735 * context and does its own locking. The returned handle has
736 * the usage count incremented and the caller must use dev_put() to
737 * release it when it is no longer needed. %NULL is returned if no
738 * matching device is found.
739 */
740
dev_get_by_name(struct net * net,const char * name)741 struct net_device *dev_get_by_name(struct net *net, const char *name)
742 {
743 struct net_device *dev;
744
745 rcu_read_lock();
746 dev = dev_get_by_name_rcu(net, name);
747 if (dev)
748 dev_hold(dev);
749 rcu_read_unlock();
750 return dev;
751 }
752 EXPORT_SYMBOL(dev_get_by_name);
753
754 /**
755 * __dev_get_by_index - find a device by its ifindex
756 * @net: the applicable net namespace
757 * @ifindex: index of device
758 *
759 * Search for an interface by index. Returns %NULL if the device
760 * is not found or a pointer to the device. The device has not
761 * had its reference counter increased so the caller must be careful
762 * about locking. The caller must hold either the RTNL semaphore
763 * or @dev_base_lock.
764 */
765
__dev_get_by_index(struct net * net,int ifindex)766 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
767 {
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
770
771 hlist_for_each_entry(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
773 return dev;
774
775 return NULL;
776 }
777 EXPORT_SYMBOL(__dev_get_by_index);
778
779 /**
780 * dev_get_by_index_rcu - find a device by its ifindex
781 * @net: the applicable net namespace
782 * @ifindex: index of device
783 *
784 * Search for an interface by index. Returns %NULL if the device
785 * is not found or a pointer to the device. The device has not
786 * had its reference counter increased so the caller must be careful
787 * about locking. The caller must hold RCU lock.
788 */
789
dev_get_by_index_rcu(struct net * net,int ifindex)790 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
791 {
792 struct net_device *dev;
793 struct hlist_head *head = dev_index_hash(net, ifindex);
794
795 hlist_for_each_entry_rcu(dev, head, index_hlist)
796 if (dev->ifindex == ifindex)
797 return dev;
798
799 return NULL;
800 }
801 EXPORT_SYMBOL(dev_get_by_index_rcu);
802
803
804 /**
805 * dev_get_by_index - find a device by its ifindex
806 * @net: the applicable net namespace
807 * @ifindex: index of device
808 *
809 * Search for an interface by index. Returns NULL if the device
810 * is not found or a pointer to the device. The device returned has
811 * had a reference added and the pointer is safe until the user calls
812 * dev_put to indicate they have finished with it.
813 */
814
dev_get_by_index(struct net * net,int ifindex)815 struct net_device *dev_get_by_index(struct net *net, int ifindex)
816 {
817 struct net_device *dev;
818
819 rcu_read_lock();
820 dev = dev_get_by_index_rcu(net, ifindex);
821 if (dev)
822 dev_hold(dev);
823 rcu_read_unlock();
824 return dev;
825 }
826 EXPORT_SYMBOL(dev_get_by_index);
827
828 /**
829 * netdev_get_name - get a netdevice name, knowing its ifindex.
830 * @net: network namespace
831 * @name: a pointer to the buffer where the name will be stored.
832 * @ifindex: the ifindex of the interface to get the name from.
833 *
834 * The use of raw_seqcount_begin() and cond_resched() before
835 * retrying is required as we want to give the writers a chance
836 * to complete when CONFIG_PREEMPT is not set.
837 */
netdev_get_name(struct net * net,char * name,int ifindex)838 int netdev_get_name(struct net *net, char *name, int ifindex)
839 {
840 struct net_device *dev;
841 unsigned int seq;
842
843 retry:
844 seq = raw_seqcount_begin(&devnet_rename_seq);
845 rcu_read_lock();
846 dev = dev_get_by_index_rcu(net, ifindex);
847 if (!dev) {
848 rcu_read_unlock();
849 return -ENODEV;
850 }
851
852 strcpy(name, dev->name);
853 rcu_read_unlock();
854 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
855 cond_resched();
856 goto retry;
857 }
858
859 return 0;
860 }
861
862 /**
863 * dev_getbyhwaddr_rcu - find a device by its hardware address
864 * @net: the applicable net namespace
865 * @type: media type of device
866 * @ha: hardware address
867 *
868 * Search for an interface by MAC address. Returns NULL if the device
869 * is not found or a pointer to the device.
870 * The caller must hold RCU or RTNL.
871 * The returned device has not had its ref count increased
872 * and the caller must therefore be careful about locking
873 *
874 */
875
dev_getbyhwaddr_rcu(struct net * net,unsigned short type,const char * ha)876 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
877 const char *ha)
878 {
879 struct net_device *dev;
880
881 for_each_netdev_rcu(net, dev)
882 if (dev->type == type &&
883 !memcmp(dev->dev_addr, ha, dev->addr_len))
884 return dev;
885
886 return NULL;
887 }
888 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
889
__dev_getfirstbyhwtype(struct net * net,unsigned short type)890 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
891 {
892 struct net_device *dev;
893
894 ASSERT_RTNL();
895 for_each_netdev(net, dev)
896 if (dev->type == type)
897 return dev;
898
899 return NULL;
900 }
901 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
902
dev_getfirstbyhwtype(struct net * net,unsigned short type)903 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
904 {
905 struct net_device *dev, *ret = NULL;
906
907 rcu_read_lock();
908 for_each_netdev_rcu(net, dev)
909 if (dev->type == type) {
910 dev_hold(dev);
911 ret = dev;
912 break;
913 }
914 rcu_read_unlock();
915 return ret;
916 }
917 EXPORT_SYMBOL(dev_getfirstbyhwtype);
918
919 /**
920 * __dev_get_by_flags - find any device with given flags
921 * @net: the applicable net namespace
922 * @if_flags: IFF_* values
923 * @mask: bitmask of bits in if_flags to check
924 *
925 * Search for any interface with the given flags. Returns NULL if a device
926 * is not found or a pointer to the device. Must be called inside
927 * rtnl_lock(), and result refcount is unchanged.
928 */
929
__dev_get_by_flags(struct net * net,unsigned short if_flags,unsigned short mask)930 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
931 unsigned short mask)
932 {
933 struct net_device *dev, *ret;
934
935 ASSERT_RTNL();
936
937 ret = NULL;
938 for_each_netdev(net, dev) {
939 if (((dev->flags ^ if_flags) & mask) == 0) {
940 ret = dev;
941 break;
942 }
943 }
944 return ret;
945 }
946 EXPORT_SYMBOL(__dev_get_by_flags);
947
948 /**
949 * dev_valid_name - check if name is okay for network device
950 * @name: name string
951 *
952 * Network device names need to be valid file names to
953 * to allow sysfs to work. We also disallow any kind of
954 * whitespace.
955 */
dev_valid_name(const char * name)956 bool dev_valid_name(const char *name)
957 {
958 if (*name == '\0')
959 return false;
960 if (strlen(name) >= IFNAMSIZ)
961 return false;
962 if (!strcmp(name, ".") || !strcmp(name, ".."))
963 return false;
964
965 while (*name) {
966 if (*name == '/' || *name == ':' || isspace(*name))
967 return false;
968 name++;
969 }
970 return true;
971 }
972 EXPORT_SYMBOL(dev_valid_name);
973
974 /**
975 * __dev_alloc_name - allocate a name for a device
976 * @net: network namespace to allocate the device name in
977 * @name: name format string
978 * @buf: scratch buffer and result name string
979 *
980 * Passed a format string - eg "lt%d" it will try and find a suitable
981 * id. It scans list of devices to build up a free map, then chooses
982 * the first empty slot. The caller must hold the dev_base or rtnl lock
983 * while allocating the name and adding the device in order to avoid
984 * duplicates.
985 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
986 * Returns the number of the unit assigned or a negative errno code.
987 */
988
__dev_alloc_name(struct net * net,const char * name,char * buf)989 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
990 {
991 int i = 0;
992 const char *p;
993 const int max_netdevices = 8*PAGE_SIZE;
994 unsigned long *inuse;
995 struct net_device *d;
996
997 p = strnchr(name, IFNAMSIZ-1, '%');
998 if (p) {
999 /*
1000 * Verify the string as this thing may have come from
1001 * the user. There must be either one "%d" and no other "%"
1002 * characters.
1003 */
1004 if (p[1] != 'd' || strchr(p + 2, '%'))
1005 return -EINVAL;
1006
1007 /* Use one page as a bit array of possible slots */
1008 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1009 if (!inuse)
1010 return -ENOMEM;
1011
1012 for_each_netdev(net, d) {
1013 if (!sscanf(d->name, name, &i))
1014 continue;
1015 if (i < 0 || i >= max_netdevices)
1016 continue;
1017
1018 /* avoid cases where sscanf is not exact inverse of printf */
1019 snprintf(buf, IFNAMSIZ, name, i);
1020 if (!strncmp(buf, d->name, IFNAMSIZ))
1021 set_bit(i, inuse);
1022 }
1023
1024 i = find_first_zero_bit(inuse, max_netdevices);
1025 free_page((unsigned long) inuse);
1026 }
1027
1028 if (buf != name)
1029 snprintf(buf, IFNAMSIZ, name, i);
1030 if (!__dev_get_by_name(net, buf))
1031 return i;
1032
1033 /* It is possible to run out of possible slots
1034 * when the name is long and there isn't enough space left
1035 * for the digits, or if all bits are used.
1036 */
1037 return -ENFILE;
1038 }
1039
1040 /**
1041 * dev_alloc_name - allocate a name for a device
1042 * @dev: device
1043 * @name: name format string
1044 *
1045 * Passed a format string - eg "lt%d" it will try and find a suitable
1046 * id. It scans list of devices to build up a free map, then chooses
1047 * the first empty slot. The caller must hold the dev_base or rtnl lock
1048 * while allocating the name and adding the device in order to avoid
1049 * duplicates.
1050 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1051 * Returns the number of the unit assigned or a negative errno code.
1052 */
1053
dev_alloc_name(struct net_device * dev,const char * name)1054 int dev_alloc_name(struct net_device *dev, const char *name)
1055 {
1056 char buf[IFNAMSIZ];
1057 struct net *net;
1058 int ret;
1059
1060 BUG_ON(!dev_net(dev));
1061 net = dev_net(dev);
1062 ret = __dev_alloc_name(net, name, buf);
1063 if (ret >= 0)
1064 strlcpy(dev->name, buf, IFNAMSIZ);
1065 return ret;
1066 }
1067 EXPORT_SYMBOL(dev_alloc_name);
1068
dev_alloc_name_ns(struct net * net,struct net_device * dev,const char * name)1069 static int dev_alloc_name_ns(struct net *net,
1070 struct net_device *dev,
1071 const char *name)
1072 {
1073 char buf[IFNAMSIZ];
1074 int ret;
1075
1076 ret = __dev_alloc_name(net, name, buf);
1077 if (ret >= 0)
1078 strlcpy(dev->name, buf, IFNAMSIZ);
1079 return ret;
1080 }
1081
dev_get_valid_name(struct net * net,struct net_device * dev,const char * name)1082 static int dev_get_valid_name(struct net *net,
1083 struct net_device *dev,
1084 const char *name)
1085 {
1086 BUG_ON(!net);
1087
1088 if (!dev_valid_name(name))
1089 return -EINVAL;
1090
1091 if (strchr(name, '%'))
1092 return dev_alloc_name_ns(net, dev, name);
1093 else if (__dev_get_by_name(net, name))
1094 return -EEXIST;
1095 else if (dev->name != name)
1096 strlcpy(dev->name, name, IFNAMSIZ);
1097
1098 return 0;
1099 }
1100
1101 /**
1102 * dev_change_name - change name of a device
1103 * @dev: device
1104 * @newname: name (or format string) must be at least IFNAMSIZ
1105 *
1106 * Change name of a device, can pass format strings "eth%d".
1107 * for wildcarding.
1108 */
dev_change_name(struct net_device * dev,const char * newname)1109 int dev_change_name(struct net_device *dev, const char *newname)
1110 {
1111 unsigned char old_assign_type;
1112 char oldname[IFNAMSIZ];
1113 int err = 0;
1114 int ret;
1115 struct net *net;
1116
1117 ASSERT_RTNL();
1118 BUG_ON(!dev_net(dev));
1119
1120 net = dev_net(dev);
1121 if (dev->flags & IFF_UP)
1122 return -EBUSY;
1123
1124 write_seqcount_begin(&devnet_rename_seq);
1125
1126 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1127 write_seqcount_end(&devnet_rename_seq);
1128 return 0;
1129 }
1130
1131 memcpy(oldname, dev->name, IFNAMSIZ);
1132
1133 err = dev_get_valid_name(net, dev, newname);
1134 if (err < 0) {
1135 write_seqcount_end(&devnet_rename_seq);
1136 return err;
1137 }
1138
1139 if (oldname[0] && !strchr(oldname, '%'))
1140 netdev_info(dev, "renamed from %s\n", oldname);
1141
1142 old_assign_type = dev->name_assign_type;
1143 dev->name_assign_type = NET_NAME_RENAMED;
1144
1145 rollback:
1146 ret = device_rename(&dev->dev, dev->name);
1147 if (ret) {
1148 memcpy(dev->name, oldname, IFNAMSIZ);
1149 dev->name_assign_type = old_assign_type;
1150 write_seqcount_end(&devnet_rename_seq);
1151 return ret;
1152 }
1153
1154 write_seqcount_end(&devnet_rename_seq);
1155
1156 netdev_adjacent_rename_links(dev, oldname);
1157
1158 write_lock_bh(&dev_base_lock);
1159 hlist_del_rcu(&dev->name_hlist);
1160 write_unlock_bh(&dev_base_lock);
1161
1162 synchronize_rcu();
1163
1164 write_lock_bh(&dev_base_lock);
1165 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1166 write_unlock_bh(&dev_base_lock);
1167
1168 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1169 ret = notifier_to_errno(ret);
1170
1171 if (ret) {
1172 /* err >= 0 after dev_alloc_name() or stores the first errno */
1173 if (err >= 0) {
1174 err = ret;
1175 write_seqcount_begin(&devnet_rename_seq);
1176 memcpy(dev->name, oldname, IFNAMSIZ);
1177 memcpy(oldname, newname, IFNAMSIZ);
1178 dev->name_assign_type = old_assign_type;
1179 old_assign_type = NET_NAME_RENAMED;
1180 goto rollback;
1181 } else {
1182 pr_err("%s: name change rollback failed: %d\n",
1183 dev->name, ret);
1184 }
1185 }
1186
1187 return err;
1188 }
1189
1190 /**
1191 * dev_set_alias - change ifalias of a device
1192 * @dev: device
1193 * @alias: name up to IFALIASZ
1194 * @len: limit of bytes to copy from info
1195 *
1196 * Set ifalias for a device,
1197 */
dev_set_alias(struct net_device * dev,const char * alias,size_t len)1198 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1199 {
1200 char *new_ifalias;
1201
1202 ASSERT_RTNL();
1203
1204 if (len >= IFALIASZ)
1205 return -EINVAL;
1206
1207 if (!len) {
1208 kfree(dev->ifalias);
1209 dev->ifalias = NULL;
1210 return 0;
1211 }
1212
1213 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1214 if (!new_ifalias)
1215 return -ENOMEM;
1216 dev->ifalias = new_ifalias;
1217
1218 strlcpy(dev->ifalias, alias, len+1);
1219 return len;
1220 }
1221
1222
1223 /**
1224 * netdev_features_change - device changes features
1225 * @dev: device to cause notification
1226 *
1227 * Called to indicate a device has changed features.
1228 */
netdev_features_change(struct net_device * dev)1229 void netdev_features_change(struct net_device *dev)
1230 {
1231 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1232 }
1233 EXPORT_SYMBOL(netdev_features_change);
1234
1235 /**
1236 * netdev_state_change - device changes state
1237 * @dev: device to cause notification
1238 *
1239 * Called to indicate a device has changed state. This function calls
1240 * the notifier chains for netdev_chain and sends a NEWLINK message
1241 * to the routing socket.
1242 */
netdev_state_change(struct net_device * dev)1243 void netdev_state_change(struct net_device *dev)
1244 {
1245 if (dev->flags & IFF_UP) {
1246 struct netdev_notifier_change_info change_info;
1247
1248 change_info.flags_changed = 0;
1249 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1250 &change_info.info);
1251 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1252 }
1253 }
1254 EXPORT_SYMBOL(netdev_state_change);
1255
1256 /**
1257 * netdev_notify_peers - notify network peers about existence of @dev
1258 * @dev: network device
1259 *
1260 * Generate traffic such that interested network peers are aware of
1261 * @dev, such as by generating a gratuitous ARP. This may be used when
1262 * a device wants to inform the rest of the network about some sort of
1263 * reconfiguration such as a failover event or virtual machine
1264 * migration.
1265 */
netdev_notify_peers(struct net_device * dev)1266 void netdev_notify_peers(struct net_device *dev)
1267 {
1268 rtnl_lock();
1269 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1270 rtnl_unlock();
1271 }
1272 EXPORT_SYMBOL(netdev_notify_peers);
1273
__dev_open(struct net_device * dev)1274 static int __dev_open(struct net_device *dev)
1275 {
1276 const struct net_device_ops *ops = dev->netdev_ops;
1277 int ret;
1278
1279 ASSERT_RTNL();
1280
1281 if (!netif_device_present(dev))
1282 return -ENODEV;
1283
1284 /* Block netpoll from trying to do any rx path servicing.
1285 * If we don't do this there is a chance ndo_poll_controller
1286 * or ndo_poll may be running while we open the device
1287 */
1288 netpoll_poll_disable(dev);
1289
1290 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1291 ret = notifier_to_errno(ret);
1292 if (ret)
1293 return ret;
1294
1295 set_bit(__LINK_STATE_START, &dev->state);
1296
1297 if (ops->ndo_validate_addr)
1298 ret = ops->ndo_validate_addr(dev);
1299
1300 if (!ret && ops->ndo_open)
1301 ret = ops->ndo_open(dev);
1302
1303 netpoll_poll_enable(dev);
1304
1305 if (ret)
1306 clear_bit(__LINK_STATE_START, &dev->state);
1307 else {
1308 dev->flags |= IFF_UP;
1309 dev_set_rx_mode(dev);
1310 dev_activate(dev);
1311 add_device_randomness(dev->dev_addr, dev->addr_len);
1312 }
1313
1314 return ret;
1315 }
1316
1317 /**
1318 * dev_open - prepare an interface for use.
1319 * @dev: device to open
1320 *
1321 * Takes a device from down to up state. The device's private open
1322 * function is invoked and then the multicast lists are loaded. Finally
1323 * the device is moved into the up state and a %NETDEV_UP message is
1324 * sent to the netdev notifier chain.
1325 *
1326 * Calling this function on an active interface is a nop. On a failure
1327 * a negative errno code is returned.
1328 */
dev_open(struct net_device * dev)1329 int dev_open(struct net_device *dev)
1330 {
1331 int ret;
1332
1333 if (dev->flags & IFF_UP)
1334 return 0;
1335
1336 ret = __dev_open(dev);
1337 if (ret < 0)
1338 return ret;
1339
1340 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1341 call_netdevice_notifiers(NETDEV_UP, dev);
1342
1343 return ret;
1344 }
1345 EXPORT_SYMBOL(dev_open);
1346
__dev_close_many(struct list_head * head)1347 static int __dev_close_many(struct list_head *head)
1348 {
1349 struct net_device *dev;
1350
1351 ASSERT_RTNL();
1352 might_sleep();
1353
1354 list_for_each_entry(dev, head, close_list) {
1355 /* Temporarily disable netpoll until the interface is down */
1356 netpoll_poll_disable(dev);
1357
1358 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1359
1360 clear_bit(__LINK_STATE_START, &dev->state);
1361
1362 /* Synchronize to scheduled poll. We cannot touch poll list, it
1363 * can be even on different cpu. So just clear netif_running().
1364 *
1365 * dev->stop() will invoke napi_disable() on all of it's
1366 * napi_struct instances on this device.
1367 */
1368 smp_mb__after_atomic(); /* Commit netif_running(). */
1369 }
1370
1371 dev_deactivate_many(head);
1372
1373 list_for_each_entry(dev, head, close_list) {
1374 const struct net_device_ops *ops = dev->netdev_ops;
1375
1376 /*
1377 * Call the device specific close. This cannot fail.
1378 * Only if device is UP
1379 *
1380 * We allow it to be called even after a DETACH hot-plug
1381 * event.
1382 */
1383 if (ops->ndo_stop)
1384 ops->ndo_stop(dev);
1385
1386 dev->flags &= ~IFF_UP;
1387 netpoll_poll_enable(dev);
1388 }
1389
1390 return 0;
1391 }
1392
__dev_close(struct net_device * dev)1393 static int __dev_close(struct net_device *dev)
1394 {
1395 int retval;
1396 LIST_HEAD(single);
1397
1398 list_add(&dev->close_list, &single);
1399 retval = __dev_close_many(&single);
1400 list_del(&single);
1401
1402 return retval;
1403 }
1404
dev_close_many(struct list_head * head,bool unlink)1405 int dev_close_many(struct list_head *head, bool unlink)
1406 {
1407 struct net_device *dev, *tmp;
1408
1409 /* Remove the devices that don't need to be closed */
1410 list_for_each_entry_safe(dev, tmp, head, close_list)
1411 if (!(dev->flags & IFF_UP))
1412 list_del_init(&dev->close_list);
1413
1414 __dev_close_many(head);
1415
1416 list_for_each_entry_safe(dev, tmp, head, close_list) {
1417 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1418 call_netdevice_notifiers(NETDEV_DOWN, dev);
1419 if (unlink)
1420 list_del_init(&dev->close_list);
1421 }
1422
1423 return 0;
1424 }
1425 EXPORT_SYMBOL(dev_close_many);
1426
1427 /**
1428 * dev_close - shutdown an interface.
1429 * @dev: device to shutdown
1430 *
1431 * This function moves an active device into down state. A
1432 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1433 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1434 * chain.
1435 */
dev_close(struct net_device * dev)1436 int dev_close(struct net_device *dev)
1437 {
1438 if (dev->flags & IFF_UP) {
1439 LIST_HEAD(single);
1440
1441 list_add(&dev->close_list, &single);
1442 dev_close_many(&single, true);
1443 list_del(&single);
1444 }
1445 return 0;
1446 }
1447 EXPORT_SYMBOL(dev_close);
1448
1449
1450 /**
1451 * dev_disable_lro - disable Large Receive Offload on a device
1452 * @dev: device
1453 *
1454 * Disable Large Receive Offload (LRO) on a net device. Must be
1455 * called under RTNL. This is needed if received packets may be
1456 * forwarded to another interface.
1457 */
dev_disable_lro(struct net_device * dev)1458 void dev_disable_lro(struct net_device *dev)
1459 {
1460 struct net_device *lower_dev;
1461 struct list_head *iter;
1462
1463 dev->wanted_features &= ~NETIF_F_LRO;
1464 netdev_update_features(dev);
1465
1466 if (unlikely(dev->features & NETIF_F_LRO))
1467 netdev_WARN(dev, "failed to disable LRO!\n");
1468
1469 netdev_for_each_lower_dev(dev, lower_dev, iter)
1470 dev_disable_lro(lower_dev);
1471 }
1472 EXPORT_SYMBOL(dev_disable_lro);
1473
call_netdevice_notifier(struct notifier_block * nb,unsigned long val,struct net_device * dev)1474 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1475 struct net_device *dev)
1476 {
1477 struct netdev_notifier_info info;
1478
1479 netdev_notifier_info_init(&info, dev);
1480 return nb->notifier_call(nb, val, &info);
1481 }
1482
1483 static int dev_boot_phase = 1;
1484
1485 /**
1486 * register_netdevice_notifier - register a network notifier block
1487 * @nb: notifier
1488 *
1489 * Register a notifier to be called when network device events occur.
1490 * The notifier passed is linked into the kernel structures and must
1491 * not be reused until it has been unregistered. A negative errno code
1492 * is returned on a failure.
1493 *
1494 * When registered all registration and up events are replayed
1495 * to the new notifier to allow device to have a race free
1496 * view of the network device list.
1497 */
1498
register_netdevice_notifier(struct notifier_block * nb)1499 int register_netdevice_notifier(struct notifier_block *nb)
1500 {
1501 struct net_device *dev;
1502 struct net_device *last;
1503 struct net *net;
1504 int err;
1505
1506 rtnl_lock();
1507 err = raw_notifier_chain_register(&netdev_chain, nb);
1508 if (err)
1509 goto unlock;
1510 if (dev_boot_phase)
1511 goto unlock;
1512 for_each_net(net) {
1513 for_each_netdev(net, dev) {
1514 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1515 err = notifier_to_errno(err);
1516 if (err)
1517 goto rollback;
1518
1519 if (!(dev->flags & IFF_UP))
1520 continue;
1521
1522 call_netdevice_notifier(nb, NETDEV_UP, dev);
1523 }
1524 }
1525
1526 unlock:
1527 rtnl_unlock();
1528 return err;
1529
1530 rollback:
1531 last = dev;
1532 for_each_net(net) {
1533 for_each_netdev(net, dev) {
1534 if (dev == last)
1535 goto outroll;
1536
1537 if (dev->flags & IFF_UP) {
1538 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1539 dev);
1540 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1541 }
1542 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1543 }
1544 }
1545
1546 outroll:
1547 raw_notifier_chain_unregister(&netdev_chain, nb);
1548 goto unlock;
1549 }
1550 EXPORT_SYMBOL(register_netdevice_notifier);
1551
1552 /**
1553 * unregister_netdevice_notifier - unregister a network notifier block
1554 * @nb: notifier
1555 *
1556 * Unregister a notifier previously registered by
1557 * register_netdevice_notifier(). The notifier is unlinked into the
1558 * kernel structures and may then be reused. A negative errno code
1559 * is returned on a failure.
1560 *
1561 * After unregistering unregister and down device events are synthesized
1562 * for all devices on the device list to the removed notifier to remove
1563 * the need for special case cleanup code.
1564 */
1565
unregister_netdevice_notifier(struct notifier_block * nb)1566 int unregister_netdevice_notifier(struct notifier_block *nb)
1567 {
1568 struct net_device *dev;
1569 struct net *net;
1570 int err;
1571
1572 rtnl_lock();
1573 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1574 if (err)
1575 goto unlock;
1576
1577 for_each_net(net) {
1578 for_each_netdev(net, dev) {
1579 if (dev->flags & IFF_UP) {
1580 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1581 dev);
1582 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1583 }
1584 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1585 }
1586 }
1587 unlock:
1588 rtnl_unlock();
1589 return err;
1590 }
1591 EXPORT_SYMBOL(unregister_netdevice_notifier);
1592
1593 /**
1594 * call_netdevice_notifiers_info - call all network notifier blocks
1595 * @val: value passed unmodified to notifier function
1596 * @dev: net_device pointer passed unmodified to notifier function
1597 * @info: notifier information data
1598 *
1599 * Call all network notifier blocks. Parameters and return value
1600 * are as for raw_notifier_call_chain().
1601 */
1602
call_netdevice_notifiers_info(unsigned long val,struct net_device * dev,struct netdev_notifier_info * info)1603 static int call_netdevice_notifiers_info(unsigned long val,
1604 struct net_device *dev,
1605 struct netdev_notifier_info *info)
1606 {
1607 ASSERT_RTNL();
1608 netdev_notifier_info_init(info, dev);
1609 return raw_notifier_call_chain(&netdev_chain, val, info);
1610 }
1611
1612 /**
1613 * call_netdevice_notifiers - call all network notifier blocks
1614 * @val: value passed unmodified to notifier function
1615 * @dev: net_device pointer passed unmodified to notifier function
1616 *
1617 * Call all network notifier blocks. Parameters and return value
1618 * are as for raw_notifier_call_chain().
1619 */
1620
call_netdevice_notifiers(unsigned long val,struct net_device * dev)1621 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1622 {
1623 struct netdev_notifier_info info;
1624
1625 return call_netdevice_notifiers_info(val, dev, &info);
1626 }
1627 EXPORT_SYMBOL(call_netdevice_notifiers);
1628
1629 #ifdef CONFIG_NET_CLS_ACT
1630 static struct static_key ingress_needed __read_mostly;
1631
net_inc_ingress_queue(void)1632 void net_inc_ingress_queue(void)
1633 {
1634 static_key_slow_inc(&ingress_needed);
1635 }
1636 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1637
net_dec_ingress_queue(void)1638 void net_dec_ingress_queue(void)
1639 {
1640 static_key_slow_dec(&ingress_needed);
1641 }
1642 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1643 #endif
1644
1645 static struct static_key netstamp_needed __read_mostly;
1646 #ifdef HAVE_JUMP_LABEL
1647 /* We are not allowed to call static_key_slow_dec() from irq context
1648 * If net_disable_timestamp() is called from irq context, defer the
1649 * static_key_slow_dec() calls.
1650 */
1651 static atomic_t netstamp_needed_deferred;
1652 #endif
1653
net_enable_timestamp(void)1654 void net_enable_timestamp(void)
1655 {
1656 #ifdef HAVE_JUMP_LABEL
1657 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1658
1659 if (deferred) {
1660 while (--deferred)
1661 static_key_slow_dec(&netstamp_needed);
1662 return;
1663 }
1664 #endif
1665 static_key_slow_inc(&netstamp_needed);
1666 }
1667 EXPORT_SYMBOL(net_enable_timestamp);
1668
net_disable_timestamp(void)1669 void net_disable_timestamp(void)
1670 {
1671 #ifdef HAVE_JUMP_LABEL
1672 if (in_interrupt()) {
1673 atomic_inc(&netstamp_needed_deferred);
1674 return;
1675 }
1676 #endif
1677 static_key_slow_dec(&netstamp_needed);
1678 }
1679 EXPORT_SYMBOL(net_disable_timestamp);
1680
net_timestamp_set(struct sk_buff * skb)1681 static inline void net_timestamp_set(struct sk_buff *skb)
1682 {
1683 skb->tstamp.tv64 = 0;
1684 if (static_key_false(&netstamp_needed))
1685 __net_timestamp(skb);
1686 }
1687
1688 #define net_timestamp_check(COND, SKB) \
1689 if (static_key_false(&netstamp_needed)) { \
1690 if ((COND) && !(SKB)->tstamp.tv64) \
1691 __net_timestamp(SKB); \
1692 } \
1693
is_skb_forwardable(struct net_device * dev,struct sk_buff * skb)1694 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1695 {
1696 unsigned int len;
1697
1698 if (!(dev->flags & IFF_UP))
1699 return false;
1700
1701 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1702 if (skb->len <= len)
1703 return true;
1704
1705 /* if TSO is enabled, we don't care about the length as the packet
1706 * could be forwarded without being segmented before
1707 */
1708 if (skb_is_gso(skb))
1709 return true;
1710
1711 return false;
1712 }
1713 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1714
__dev_forward_skb(struct net_device * dev,struct sk_buff * skb)1715 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1716 {
1717 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1718 unlikely(!is_skb_forwardable(dev, skb))) {
1719 atomic_long_inc(&dev->rx_dropped);
1720 kfree_skb(skb);
1721 return NET_RX_DROP;
1722 }
1723
1724 skb_scrub_packet(skb, true);
1725 skb->priority = 0;
1726 skb->protocol = eth_type_trans(skb, dev);
1727 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1728
1729 return 0;
1730 }
1731 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1732
1733 /**
1734 * dev_forward_skb - loopback an skb to another netif
1735 *
1736 * @dev: destination network device
1737 * @skb: buffer to forward
1738 *
1739 * return values:
1740 * NET_RX_SUCCESS (no congestion)
1741 * NET_RX_DROP (packet was dropped, but freed)
1742 *
1743 * dev_forward_skb can be used for injecting an skb from the
1744 * start_xmit function of one device into the receive queue
1745 * of another device.
1746 *
1747 * The receiving device may be in another namespace, so
1748 * we have to clear all information in the skb that could
1749 * impact namespace isolation.
1750 */
dev_forward_skb(struct net_device * dev,struct sk_buff * skb)1751 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1752 {
1753 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1754 }
1755 EXPORT_SYMBOL_GPL(dev_forward_skb);
1756
deliver_skb(struct sk_buff * skb,struct packet_type * pt_prev,struct net_device * orig_dev)1757 static inline int deliver_skb(struct sk_buff *skb,
1758 struct packet_type *pt_prev,
1759 struct net_device *orig_dev)
1760 {
1761 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1762 return -ENOMEM;
1763 atomic_inc(&skb->users);
1764 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1765 }
1766
deliver_ptype_list_skb(struct sk_buff * skb,struct packet_type ** pt,struct net_device * orig_dev,__be16 type,struct list_head * ptype_list)1767 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1768 struct packet_type **pt,
1769 struct net_device *orig_dev,
1770 __be16 type,
1771 struct list_head *ptype_list)
1772 {
1773 struct packet_type *ptype, *pt_prev = *pt;
1774
1775 list_for_each_entry_rcu(ptype, ptype_list, list) {
1776 if (ptype->type != type)
1777 continue;
1778 if (pt_prev)
1779 deliver_skb(skb, pt_prev, orig_dev);
1780 pt_prev = ptype;
1781 }
1782 *pt = pt_prev;
1783 }
1784
skb_loop_sk(struct packet_type * ptype,struct sk_buff * skb)1785 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1786 {
1787 if (!ptype->af_packet_priv || !skb->sk)
1788 return false;
1789
1790 if (ptype->id_match)
1791 return ptype->id_match(ptype, skb->sk);
1792 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1793 return true;
1794
1795 return false;
1796 }
1797
1798 /*
1799 * Support routine. Sends outgoing frames to any network
1800 * taps currently in use.
1801 */
1802
dev_queue_xmit_nit(struct sk_buff * skb,struct net_device * dev)1803 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1804 {
1805 struct packet_type *ptype;
1806 struct sk_buff *skb2 = NULL;
1807 struct packet_type *pt_prev = NULL;
1808 struct list_head *ptype_list = &ptype_all;
1809
1810 rcu_read_lock();
1811 again:
1812 list_for_each_entry_rcu(ptype, ptype_list, list) {
1813 /* Never send packets back to the socket
1814 * they originated from - MvS (miquels@drinkel.ow.org)
1815 */
1816 if (skb_loop_sk(ptype, skb))
1817 continue;
1818
1819 if (pt_prev) {
1820 deliver_skb(skb2, pt_prev, skb->dev);
1821 pt_prev = ptype;
1822 continue;
1823 }
1824
1825 /* need to clone skb, done only once */
1826 skb2 = skb_clone(skb, GFP_ATOMIC);
1827 if (!skb2)
1828 goto out_unlock;
1829
1830 net_timestamp_set(skb2);
1831
1832 /* skb->nh should be correctly
1833 * set by sender, so that the second statement is
1834 * just protection against buggy protocols.
1835 */
1836 skb_reset_mac_header(skb2);
1837
1838 if (skb_network_header(skb2) < skb2->data ||
1839 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1840 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1841 ntohs(skb2->protocol),
1842 dev->name);
1843 skb_reset_network_header(skb2);
1844 }
1845
1846 skb2->transport_header = skb2->network_header;
1847 skb2->pkt_type = PACKET_OUTGOING;
1848 pt_prev = ptype;
1849 }
1850
1851 if (ptype_list == &ptype_all) {
1852 ptype_list = &dev->ptype_all;
1853 goto again;
1854 }
1855 out_unlock:
1856 if (pt_prev)
1857 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1858 rcu_read_unlock();
1859 }
1860
1861 /**
1862 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1863 * @dev: Network device
1864 * @txq: number of queues available
1865 *
1866 * If real_num_tx_queues is changed the tc mappings may no longer be
1867 * valid. To resolve this verify the tc mapping remains valid and if
1868 * not NULL the mapping. With no priorities mapping to this
1869 * offset/count pair it will no longer be used. In the worst case TC0
1870 * is invalid nothing can be done so disable priority mappings. If is
1871 * expected that drivers will fix this mapping if they can before
1872 * calling netif_set_real_num_tx_queues.
1873 */
netif_setup_tc(struct net_device * dev,unsigned int txq)1874 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1875 {
1876 int i;
1877 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1878
1879 /* If TC0 is invalidated disable TC mapping */
1880 if (tc->offset + tc->count > txq) {
1881 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1882 dev->num_tc = 0;
1883 return;
1884 }
1885
1886 /* Invalidated prio to tc mappings set to TC0 */
1887 for (i = 1; i < TC_BITMASK + 1; i++) {
1888 int q = netdev_get_prio_tc_map(dev, i);
1889
1890 tc = &dev->tc_to_txq[q];
1891 if (tc->offset + tc->count > txq) {
1892 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1893 i, q);
1894 netdev_set_prio_tc_map(dev, i, 0);
1895 }
1896 }
1897 }
1898
1899 #ifdef CONFIG_XPS
1900 static DEFINE_MUTEX(xps_map_mutex);
1901 #define xmap_dereference(P) \
1902 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1903
remove_xps_queue(struct xps_dev_maps * dev_maps,int cpu,u16 index)1904 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1905 int cpu, u16 index)
1906 {
1907 struct xps_map *map = NULL;
1908 int pos;
1909
1910 if (dev_maps)
1911 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1912
1913 for (pos = 0; map && pos < map->len; pos++) {
1914 if (map->queues[pos] == index) {
1915 if (map->len > 1) {
1916 map->queues[pos] = map->queues[--map->len];
1917 } else {
1918 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1919 kfree_rcu(map, rcu);
1920 map = NULL;
1921 }
1922 break;
1923 }
1924 }
1925
1926 return map;
1927 }
1928
netif_reset_xps_queues_gt(struct net_device * dev,u16 index)1929 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1930 {
1931 struct xps_dev_maps *dev_maps;
1932 int cpu, i;
1933 bool active = false;
1934
1935 mutex_lock(&xps_map_mutex);
1936 dev_maps = xmap_dereference(dev->xps_maps);
1937
1938 if (!dev_maps)
1939 goto out_no_maps;
1940
1941 for_each_possible_cpu(cpu) {
1942 for (i = index; i < dev->num_tx_queues; i++) {
1943 if (!remove_xps_queue(dev_maps, cpu, i))
1944 break;
1945 }
1946 if (i == dev->num_tx_queues)
1947 active = true;
1948 }
1949
1950 if (!active) {
1951 RCU_INIT_POINTER(dev->xps_maps, NULL);
1952 kfree_rcu(dev_maps, rcu);
1953 }
1954
1955 for (i = index; i < dev->num_tx_queues; i++)
1956 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1957 NUMA_NO_NODE);
1958
1959 out_no_maps:
1960 mutex_unlock(&xps_map_mutex);
1961 }
1962
expand_xps_map(struct xps_map * map,int cpu,u16 index)1963 static struct xps_map *expand_xps_map(struct xps_map *map,
1964 int cpu, u16 index)
1965 {
1966 struct xps_map *new_map;
1967 int alloc_len = XPS_MIN_MAP_ALLOC;
1968 int i, pos;
1969
1970 for (pos = 0; map && pos < map->len; pos++) {
1971 if (map->queues[pos] != index)
1972 continue;
1973 return map;
1974 }
1975
1976 /* Need to add queue to this CPU's existing map */
1977 if (map) {
1978 if (pos < map->alloc_len)
1979 return map;
1980
1981 alloc_len = map->alloc_len * 2;
1982 }
1983
1984 /* Need to allocate new map to store queue on this CPU's map */
1985 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1986 cpu_to_node(cpu));
1987 if (!new_map)
1988 return NULL;
1989
1990 for (i = 0; i < pos; i++)
1991 new_map->queues[i] = map->queues[i];
1992 new_map->alloc_len = alloc_len;
1993 new_map->len = pos;
1994
1995 return new_map;
1996 }
1997
netif_set_xps_queue(struct net_device * dev,const struct cpumask * mask,u16 index)1998 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1999 u16 index)
2000 {
2001 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2002 struct xps_map *map, *new_map;
2003 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2004 int cpu, numa_node_id = -2;
2005 bool active = false;
2006
2007 mutex_lock(&xps_map_mutex);
2008
2009 dev_maps = xmap_dereference(dev->xps_maps);
2010
2011 /* allocate memory for queue storage */
2012 for_each_online_cpu(cpu) {
2013 if (!cpumask_test_cpu(cpu, mask))
2014 continue;
2015
2016 if (!new_dev_maps)
2017 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2018 if (!new_dev_maps) {
2019 mutex_unlock(&xps_map_mutex);
2020 return -ENOMEM;
2021 }
2022
2023 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2024 NULL;
2025
2026 map = expand_xps_map(map, cpu, index);
2027 if (!map)
2028 goto error;
2029
2030 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2031 }
2032
2033 if (!new_dev_maps)
2034 goto out_no_new_maps;
2035
2036 for_each_possible_cpu(cpu) {
2037 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2038 /* add queue to CPU maps */
2039 int pos = 0;
2040
2041 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2042 while ((pos < map->len) && (map->queues[pos] != index))
2043 pos++;
2044
2045 if (pos == map->len)
2046 map->queues[map->len++] = index;
2047 #ifdef CONFIG_NUMA
2048 if (numa_node_id == -2)
2049 numa_node_id = cpu_to_node(cpu);
2050 else if (numa_node_id != cpu_to_node(cpu))
2051 numa_node_id = -1;
2052 #endif
2053 } else if (dev_maps) {
2054 /* fill in the new device map from the old device map */
2055 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2056 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2057 }
2058
2059 }
2060
2061 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2062
2063 /* Cleanup old maps */
2064 if (dev_maps) {
2065 for_each_possible_cpu(cpu) {
2066 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2067 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2068 if (map && map != new_map)
2069 kfree_rcu(map, rcu);
2070 }
2071
2072 kfree_rcu(dev_maps, rcu);
2073 }
2074
2075 dev_maps = new_dev_maps;
2076 active = true;
2077
2078 out_no_new_maps:
2079 /* update Tx queue numa node */
2080 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2081 (numa_node_id >= 0) ? numa_node_id :
2082 NUMA_NO_NODE);
2083
2084 if (!dev_maps)
2085 goto out_no_maps;
2086
2087 /* removes queue from unused CPUs */
2088 for_each_possible_cpu(cpu) {
2089 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2090 continue;
2091
2092 if (remove_xps_queue(dev_maps, cpu, index))
2093 active = true;
2094 }
2095
2096 /* free map if not active */
2097 if (!active) {
2098 RCU_INIT_POINTER(dev->xps_maps, NULL);
2099 kfree_rcu(dev_maps, rcu);
2100 }
2101
2102 out_no_maps:
2103 mutex_unlock(&xps_map_mutex);
2104
2105 return 0;
2106 error:
2107 /* remove any maps that we added */
2108 for_each_possible_cpu(cpu) {
2109 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2110 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2111 NULL;
2112 if (new_map && new_map != map)
2113 kfree(new_map);
2114 }
2115
2116 mutex_unlock(&xps_map_mutex);
2117
2118 kfree(new_dev_maps);
2119 return -ENOMEM;
2120 }
2121 EXPORT_SYMBOL(netif_set_xps_queue);
2122
2123 #endif
2124 /*
2125 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2126 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2127 */
netif_set_real_num_tx_queues(struct net_device * dev,unsigned int txq)2128 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2129 {
2130 int rc;
2131
2132 if (txq < 1 || txq > dev->num_tx_queues)
2133 return -EINVAL;
2134
2135 if (dev->reg_state == NETREG_REGISTERED ||
2136 dev->reg_state == NETREG_UNREGISTERING) {
2137 ASSERT_RTNL();
2138
2139 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2140 txq);
2141 if (rc)
2142 return rc;
2143
2144 if (dev->num_tc)
2145 netif_setup_tc(dev, txq);
2146
2147 if (txq < dev->real_num_tx_queues) {
2148 qdisc_reset_all_tx_gt(dev, txq);
2149 #ifdef CONFIG_XPS
2150 netif_reset_xps_queues_gt(dev, txq);
2151 #endif
2152 }
2153 }
2154
2155 dev->real_num_tx_queues = txq;
2156 return 0;
2157 }
2158 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2159
2160 #ifdef CONFIG_SYSFS
2161 /**
2162 * netif_set_real_num_rx_queues - set actual number of RX queues used
2163 * @dev: Network device
2164 * @rxq: Actual number of RX queues
2165 *
2166 * This must be called either with the rtnl_lock held or before
2167 * registration of the net device. Returns 0 on success, or a
2168 * negative error code. If called before registration, it always
2169 * succeeds.
2170 */
netif_set_real_num_rx_queues(struct net_device * dev,unsigned int rxq)2171 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2172 {
2173 int rc;
2174
2175 if (rxq < 1 || rxq > dev->num_rx_queues)
2176 return -EINVAL;
2177
2178 if (dev->reg_state == NETREG_REGISTERED) {
2179 ASSERT_RTNL();
2180
2181 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2182 rxq);
2183 if (rc)
2184 return rc;
2185 }
2186
2187 dev->real_num_rx_queues = rxq;
2188 return 0;
2189 }
2190 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2191 #endif
2192
2193 /**
2194 * netif_get_num_default_rss_queues - default number of RSS queues
2195 *
2196 * This routine should set an upper limit on the number of RSS queues
2197 * used by default by multiqueue devices.
2198 */
netif_get_num_default_rss_queues(void)2199 int netif_get_num_default_rss_queues(void)
2200 {
2201 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2202 }
2203 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2204
__netif_reschedule(struct Qdisc * q)2205 static inline void __netif_reschedule(struct Qdisc *q)
2206 {
2207 struct softnet_data *sd;
2208 unsigned long flags;
2209
2210 local_irq_save(flags);
2211 sd = this_cpu_ptr(&softnet_data);
2212 q->next_sched = NULL;
2213 *sd->output_queue_tailp = q;
2214 sd->output_queue_tailp = &q->next_sched;
2215 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2216 local_irq_restore(flags);
2217 }
2218
__netif_schedule(struct Qdisc * q)2219 void __netif_schedule(struct Qdisc *q)
2220 {
2221 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2222 __netif_reschedule(q);
2223 }
2224 EXPORT_SYMBOL(__netif_schedule);
2225
2226 struct dev_kfree_skb_cb {
2227 enum skb_free_reason reason;
2228 };
2229
get_kfree_skb_cb(const struct sk_buff * skb)2230 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2231 {
2232 return (struct dev_kfree_skb_cb *)skb->cb;
2233 }
2234
netif_schedule_queue(struct netdev_queue * txq)2235 void netif_schedule_queue(struct netdev_queue *txq)
2236 {
2237 rcu_read_lock();
2238 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2239 struct Qdisc *q = rcu_dereference(txq->qdisc);
2240
2241 __netif_schedule(q);
2242 }
2243 rcu_read_unlock();
2244 }
2245 EXPORT_SYMBOL(netif_schedule_queue);
2246
2247 /**
2248 * netif_wake_subqueue - allow sending packets on subqueue
2249 * @dev: network device
2250 * @queue_index: sub queue index
2251 *
2252 * Resume individual transmit queue of a device with multiple transmit queues.
2253 */
netif_wake_subqueue(struct net_device * dev,u16 queue_index)2254 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2255 {
2256 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2257
2258 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2259 struct Qdisc *q;
2260
2261 rcu_read_lock();
2262 q = rcu_dereference(txq->qdisc);
2263 __netif_schedule(q);
2264 rcu_read_unlock();
2265 }
2266 }
2267 EXPORT_SYMBOL(netif_wake_subqueue);
2268
netif_tx_wake_queue(struct netdev_queue * dev_queue)2269 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2270 {
2271 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2272 struct Qdisc *q;
2273
2274 rcu_read_lock();
2275 q = rcu_dereference(dev_queue->qdisc);
2276 __netif_schedule(q);
2277 rcu_read_unlock();
2278 }
2279 }
2280 EXPORT_SYMBOL(netif_tx_wake_queue);
2281
__dev_kfree_skb_irq(struct sk_buff * skb,enum skb_free_reason reason)2282 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2283 {
2284 unsigned long flags;
2285
2286 if (likely(atomic_read(&skb->users) == 1)) {
2287 smp_rmb();
2288 atomic_set(&skb->users, 0);
2289 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2290 return;
2291 }
2292 get_kfree_skb_cb(skb)->reason = reason;
2293 local_irq_save(flags);
2294 skb->next = __this_cpu_read(softnet_data.completion_queue);
2295 __this_cpu_write(softnet_data.completion_queue, skb);
2296 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2297 local_irq_restore(flags);
2298 }
2299 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2300
__dev_kfree_skb_any(struct sk_buff * skb,enum skb_free_reason reason)2301 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2302 {
2303 if (in_irq() || irqs_disabled())
2304 __dev_kfree_skb_irq(skb, reason);
2305 else
2306 dev_kfree_skb(skb);
2307 }
2308 EXPORT_SYMBOL(__dev_kfree_skb_any);
2309
2310
2311 /**
2312 * netif_device_detach - mark device as removed
2313 * @dev: network device
2314 *
2315 * Mark device as removed from system and therefore no longer available.
2316 */
netif_device_detach(struct net_device * dev)2317 void netif_device_detach(struct net_device *dev)
2318 {
2319 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2320 netif_running(dev)) {
2321 netif_tx_stop_all_queues(dev);
2322 }
2323 }
2324 EXPORT_SYMBOL(netif_device_detach);
2325
2326 /**
2327 * netif_device_attach - mark device as attached
2328 * @dev: network device
2329 *
2330 * Mark device as attached from system and restart if needed.
2331 */
netif_device_attach(struct net_device * dev)2332 void netif_device_attach(struct net_device *dev)
2333 {
2334 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2335 netif_running(dev)) {
2336 netif_tx_wake_all_queues(dev);
2337 __netdev_watchdog_up(dev);
2338 }
2339 }
2340 EXPORT_SYMBOL(netif_device_attach);
2341
skb_warn_bad_offload(const struct sk_buff * skb)2342 static void skb_warn_bad_offload(const struct sk_buff *skb)
2343 {
2344 static const netdev_features_t null_features = 0;
2345 struct net_device *dev = skb->dev;
2346 const char *driver = "";
2347
2348 if (!net_ratelimit())
2349 return;
2350
2351 if (dev && dev->dev.parent)
2352 driver = dev_driver_string(dev->dev.parent);
2353
2354 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2355 "gso_type=%d ip_summed=%d\n",
2356 driver, dev ? &dev->features : &null_features,
2357 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2358 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2359 skb_shinfo(skb)->gso_type, skb->ip_summed);
2360 }
2361
2362 /*
2363 * Invalidate hardware checksum when packet is to be mangled, and
2364 * complete checksum manually on outgoing path.
2365 */
skb_checksum_help(struct sk_buff * skb)2366 int skb_checksum_help(struct sk_buff *skb)
2367 {
2368 __wsum csum;
2369 int ret = 0, offset;
2370
2371 if (skb->ip_summed == CHECKSUM_COMPLETE)
2372 goto out_set_summed;
2373
2374 if (unlikely(skb_shinfo(skb)->gso_size)) {
2375 skb_warn_bad_offload(skb);
2376 return -EINVAL;
2377 }
2378
2379 /* Before computing a checksum, we should make sure no frag could
2380 * be modified by an external entity : checksum could be wrong.
2381 */
2382 if (skb_has_shared_frag(skb)) {
2383 ret = __skb_linearize(skb);
2384 if (ret)
2385 goto out;
2386 }
2387
2388 offset = skb_checksum_start_offset(skb);
2389 BUG_ON(offset >= skb_headlen(skb));
2390 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2391
2392 offset += skb->csum_offset;
2393 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2394
2395 if (skb_cloned(skb) &&
2396 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2397 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2398 if (ret)
2399 goto out;
2400 }
2401
2402 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2403 out_set_summed:
2404 skb->ip_summed = CHECKSUM_NONE;
2405 out:
2406 return ret;
2407 }
2408 EXPORT_SYMBOL(skb_checksum_help);
2409
skb_network_protocol(struct sk_buff * skb,int * depth)2410 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2411 {
2412 __be16 type = skb->protocol;
2413
2414 /* Tunnel gso handlers can set protocol to ethernet. */
2415 if (type == htons(ETH_P_TEB)) {
2416 struct ethhdr *eth;
2417
2418 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2419 return 0;
2420
2421 eth = (struct ethhdr *)skb_mac_header(skb);
2422 type = eth->h_proto;
2423 }
2424
2425 return __vlan_get_protocol(skb, type, depth);
2426 }
2427
2428 /**
2429 * skb_mac_gso_segment - mac layer segmentation handler.
2430 * @skb: buffer to segment
2431 * @features: features for the output path (see dev->features)
2432 */
skb_mac_gso_segment(struct sk_buff * skb,netdev_features_t features)2433 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2434 netdev_features_t features)
2435 {
2436 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2437 struct packet_offload *ptype;
2438 int vlan_depth = skb->mac_len;
2439 __be16 type = skb_network_protocol(skb, &vlan_depth);
2440
2441 if (unlikely(!type))
2442 return ERR_PTR(-EINVAL);
2443
2444 __skb_pull(skb, vlan_depth);
2445
2446 rcu_read_lock();
2447 list_for_each_entry_rcu(ptype, &offload_base, list) {
2448 if (ptype->type == type && ptype->callbacks.gso_segment) {
2449 segs = ptype->callbacks.gso_segment(skb, features);
2450 break;
2451 }
2452 }
2453 rcu_read_unlock();
2454
2455 __skb_push(skb, skb->data - skb_mac_header(skb));
2456
2457 return segs;
2458 }
2459 EXPORT_SYMBOL(skb_mac_gso_segment);
2460
2461
2462 /* openvswitch calls this on rx path, so we need a different check.
2463 */
skb_needs_check(struct sk_buff * skb,bool tx_path)2464 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2465 {
2466 if (tx_path)
2467 return skb->ip_summed != CHECKSUM_PARTIAL;
2468 else
2469 return skb->ip_summed == CHECKSUM_NONE;
2470 }
2471
2472 /**
2473 * __skb_gso_segment - Perform segmentation on skb.
2474 * @skb: buffer to segment
2475 * @features: features for the output path (see dev->features)
2476 * @tx_path: whether it is called in TX path
2477 *
2478 * This function segments the given skb and returns a list of segments.
2479 *
2480 * It may return NULL if the skb requires no segmentation. This is
2481 * only possible when GSO is used for verifying header integrity.
2482 *
2483 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2484 */
__skb_gso_segment(struct sk_buff * skb,netdev_features_t features,bool tx_path)2485 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2486 netdev_features_t features, bool tx_path)
2487 {
2488 if (unlikely(skb_needs_check(skb, tx_path))) {
2489 int err;
2490
2491 skb_warn_bad_offload(skb);
2492
2493 err = skb_cow_head(skb, 0);
2494 if (err < 0)
2495 return ERR_PTR(err);
2496 }
2497
2498 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2499 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2500
2501 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2502 SKB_GSO_CB(skb)->encap_level = 0;
2503
2504 skb_reset_mac_header(skb);
2505 skb_reset_mac_len(skb);
2506
2507 return skb_mac_gso_segment(skb, features);
2508 }
2509 EXPORT_SYMBOL(__skb_gso_segment);
2510
2511 /* Take action when hardware reception checksum errors are detected. */
2512 #ifdef CONFIG_BUG
netdev_rx_csum_fault(struct net_device * dev)2513 void netdev_rx_csum_fault(struct net_device *dev)
2514 {
2515 if (net_ratelimit()) {
2516 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2517 dump_stack();
2518 }
2519 }
2520 EXPORT_SYMBOL(netdev_rx_csum_fault);
2521 #endif
2522
2523 /* Actually, we should eliminate this check as soon as we know, that:
2524 * 1. IOMMU is present and allows to map all the memory.
2525 * 2. No high memory really exists on this machine.
2526 */
2527
illegal_highdma(struct net_device * dev,struct sk_buff * skb)2528 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2529 {
2530 #ifdef CONFIG_HIGHMEM
2531 int i;
2532 if (!(dev->features & NETIF_F_HIGHDMA)) {
2533 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2534 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2535 if (PageHighMem(skb_frag_page(frag)))
2536 return 1;
2537 }
2538 }
2539
2540 if (PCI_DMA_BUS_IS_PHYS) {
2541 struct device *pdev = dev->dev.parent;
2542
2543 if (!pdev)
2544 return 0;
2545 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2546 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2547 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2548 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2549 return 1;
2550 }
2551 }
2552 #endif
2553 return 0;
2554 }
2555
2556 /* If MPLS offload request, verify we are testing hardware MPLS features
2557 * instead of standard features for the netdev.
2558 */
2559 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
net_mpls_features(struct sk_buff * skb,netdev_features_t features,__be16 type)2560 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2561 netdev_features_t features,
2562 __be16 type)
2563 {
2564 if (eth_p_mpls(type))
2565 features &= skb->dev->mpls_features;
2566
2567 return features;
2568 }
2569 #else
net_mpls_features(struct sk_buff * skb,netdev_features_t features,__be16 type)2570 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2571 netdev_features_t features,
2572 __be16 type)
2573 {
2574 return features;
2575 }
2576 #endif
2577
harmonize_features(struct sk_buff * skb,netdev_features_t features)2578 static netdev_features_t harmonize_features(struct sk_buff *skb,
2579 netdev_features_t features)
2580 {
2581 int tmp;
2582 __be16 type;
2583
2584 type = skb_network_protocol(skb, &tmp);
2585 features = net_mpls_features(skb, features, type);
2586
2587 if (skb->ip_summed != CHECKSUM_NONE &&
2588 !can_checksum_protocol(features, type)) {
2589 features &= ~NETIF_F_ALL_CSUM;
2590 } else if (illegal_highdma(skb->dev, skb)) {
2591 features &= ~NETIF_F_SG;
2592 }
2593
2594 return features;
2595 }
2596
passthru_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)2597 netdev_features_t passthru_features_check(struct sk_buff *skb,
2598 struct net_device *dev,
2599 netdev_features_t features)
2600 {
2601 return features;
2602 }
2603 EXPORT_SYMBOL(passthru_features_check);
2604
dflt_features_check(const struct sk_buff * skb,struct net_device * dev,netdev_features_t features)2605 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2606 struct net_device *dev,
2607 netdev_features_t features)
2608 {
2609 return vlan_features_check(skb, features);
2610 }
2611
netif_skb_features(struct sk_buff * skb)2612 netdev_features_t netif_skb_features(struct sk_buff *skb)
2613 {
2614 struct net_device *dev = skb->dev;
2615 netdev_features_t features = dev->features;
2616 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2617
2618 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2619 features &= ~NETIF_F_GSO_MASK;
2620
2621 /* If encapsulation offload request, verify we are testing
2622 * hardware encapsulation features instead of standard
2623 * features for the netdev
2624 */
2625 if (skb->encapsulation)
2626 features &= dev->hw_enc_features;
2627
2628 if (skb_vlan_tagged(skb))
2629 features = netdev_intersect_features(features,
2630 dev->vlan_features |
2631 NETIF_F_HW_VLAN_CTAG_TX |
2632 NETIF_F_HW_VLAN_STAG_TX);
2633
2634 if (dev->netdev_ops->ndo_features_check)
2635 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2636 features);
2637 else
2638 features &= dflt_features_check(skb, dev, features);
2639
2640 return harmonize_features(skb, features);
2641 }
2642 EXPORT_SYMBOL(netif_skb_features);
2643
xmit_one(struct sk_buff * skb,struct net_device * dev,struct netdev_queue * txq,bool more)2644 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2645 struct netdev_queue *txq, bool more)
2646 {
2647 unsigned int len;
2648 int rc;
2649
2650 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2651 dev_queue_xmit_nit(skb, dev);
2652
2653 len = skb->len;
2654 trace_net_dev_start_xmit(skb, dev);
2655 rc = netdev_start_xmit(skb, dev, txq, more);
2656 trace_net_dev_xmit(skb, rc, dev, len);
2657
2658 return rc;
2659 }
2660
dev_hard_start_xmit(struct sk_buff * first,struct net_device * dev,struct netdev_queue * txq,int * ret)2661 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2662 struct netdev_queue *txq, int *ret)
2663 {
2664 struct sk_buff *skb = first;
2665 int rc = NETDEV_TX_OK;
2666
2667 while (skb) {
2668 struct sk_buff *next = skb->next;
2669
2670 skb->next = NULL;
2671 rc = xmit_one(skb, dev, txq, next != NULL);
2672 if (unlikely(!dev_xmit_complete(rc))) {
2673 skb->next = next;
2674 goto out;
2675 }
2676
2677 skb = next;
2678 if (netif_xmit_stopped(txq) && skb) {
2679 rc = NETDEV_TX_BUSY;
2680 break;
2681 }
2682 }
2683
2684 out:
2685 *ret = rc;
2686 return skb;
2687 }
2688
validate_xmit_vlan(struct sk_buff * skb,netdev_features_t features)2689 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2690 netdev_features_t features)
2691 {
2692 if (skb_vlan_tag_present(skb) &&
2693 !vlan_hw_offload_capable(features, skb->vlan_proto))
2694 skb = __vlan_hwaccel_push_inside(skb);
2695 return skb;
2696 }
2697
validate_xmit_skb(struct sk_buff * skb,struct net_device * dev)2698 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2699 {
2700 netdev_features_t features;
2701
2702 if (skb->next)
2703 return skb;
2704
2705 features = netif_skb_features(skb);
2706 skb = validate_xmit_vlan(skb, features);
2707 if (unlikely(!skb))
2708 goto out_null;
2709
2710 if (netif_needs_gso(skb, features)) {
2711 struct sk_buff *segs;
2712
2713 segs = skb_gso_segment(skb, features);
2714 if (IS_ERR(segs)) {
2715 goto out_kfree_skb;
2716 } else if (segs) {
2717 consume_skb(skb);
2718 skb = segs;
2719 }
2720 } else {
2721 if (skb_needs_linearize(skb, features) &&
2722 __skb_linearize(skb))
2723 goto out_kfree_skb;
2724
2725 /* If packet is not checksummed and device does not
2726 * support checksumming for this protocol, complete
2727 * checksumming here.
2728 */
2729 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2730 if (skb->encapsulation)
2731 skb_set_inner_transport_header(skb,
2732 skb_checksum_start_offset(skb));
2733 else
2734 skb_set_transport_header(skb,
2735 skb_checksum_start_offset(skb));
2736 if (!(features & NETIF_F_ALL_CSUM) &&
2737 skb_checksum_help(skb))
2738 goto out_kfree_skb;
2739 }
2740 }
2741
2742 return skb;
2743
2744 out_kfree_skb:
2745 kfree_skb(skb);
2746 out_null:
2747 return NULL;
2748 }
2749
validate_xmit_skb_list(struct sk_buff * skb,struct net_device * dev)2750 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2751 {
2752 struct sk_buff *next, *head = NULL, *tail;
2753
2754 for (; skb != NULL; skb = next) {
2755 next = skb->next;
2756 skb->next = NULL;
2757
2758 /* in case skb wont be segmented, point to itself */
2759 skb->prev = skb;
2760
2761 skb = validate_xmit_skb(skb, dev);
2762 if (!skb)
2763 continue;
2764
2765 if (!head)
2766 head = skb;
2767 else
2768 tail->next = skb;
2769 /* If skb was segmented, skb->prev points to
2770 * the last segment. If not, it still contains skb.
2771 */
2772 tail = skb->prev;
2773 }
2774 return head;
2775 }
2776
qdisc_pkt_len_init(struct sk_buff * skb)2777 static void qdisc_pkt_len_init(struct sk_buff *skb)
2778 {
2779 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2780
2781 qdisc_skb_cb(skb)->pkt_len = skb->len;
2782
2783 /* To get more precise estimation of bytes sent on wire,
2784 * we add to pkt_len the headers size of all segments
2785 */
2786 if (shinfo->gso_size) {
2787 unsigned int hdr_len;
2788 u16 gso_segs = shinfo->gso_segs;
2789
2790 /* mac layer + network layer */
2791 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2792
2793 /* + transport layer */
2794 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2795 hdr_len += tcp_hdrlen(skb);
2796 else
2797 hdr_len += sizeof(struct udphdr);
2798
2799 if (shinfo->gso_type & SKB_GSO_DODGY)
2800 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2801 shinfo->gso_size);
2802
2803 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2804 }
2805 }
2806
__dev_xmit_skb(struct sk_buff * skb,struct Qdisc * q,struct net_device * dev,struct netdev_queue * txq)2807 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2808 struct net_device *dev,
2809 struct netdev_queue *txq)
2810 {
2811 spinlock_t *root_lock = qdisc_lock(q);
2812 bool contended;
2813 int rc;
2814
2815 qdisc_pkt_len_init(skb);
2816 qdisc_calculate_pkt_len(skb, q);
2817 /*
2818 * Heuristic to force contended enqueues to serialize on a
2819 * separate lock before trying to get qdisc main lock.
2820 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2821 * often and dequeue packets faster.
2822 */
2823 contended = qdisc_is_running(q);
2824 if (unlikely(contended))
2825 spin_lock(&q->busylock);
2826
2827 spin_lock(root_lock);
2828 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2829 kfree_skb(skb);
2830 rc = NET_XMIT_DROP;
2831 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2832 qdisc_run_begin(q)) {
2833 /*
2834 * This is a work-conserving queue; there are no old skbs
2835 * waiting to be sent out; and the qdisc is not running -
2836 * xmit the skb directly.
2837 */
2838
2839 qdisc_bstats_update(q, skb);
2840
2841 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2842 if (unlikely(contended)) {
2843 spin_unlock(&q->busylock);
2844 contended = false;
2845 }
2846 __qdisc_run(q);
2847 } else
2848 qdisc_run_end(q);
2849
2850 rc = NET_XMIT_SUCCESS;
2851 } else {
2852 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2853 if (qdisc_run_begin(q)) {
2854 if (unlikely(contended)) {
2855 spin_unlock(&q->busylock);
2856 contended = false;
2857 }
2858 __qdisc_run(q);
2859 }
2860 }
2861 spin_unlock(root_lock);
2862 if (unlikely(contended))
2863 spin_unlock(&q->busylock);
2864 return rc;
2865 }
2866
2867 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
skb_update_prio(struct sk_buff * skb)2868 static void skb_update_prio(struct sk_buff *skb)
2869 {
2870 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2871
2872 if (!skb->priority && skb->sk && map) {
2873 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2874
2875 if (prioidx < map->priomap_len)
2876 skb->priority = map->priomap[prioidx];
2877 }
2878 }
2879 #else
2880 #define skb_update_prio(skb)
2881 #endif
2882
2883 DEFINE_PER_CPU(int, xmit_recursion);
2884 EXPORT_SYMBOL(xmit_recursion);
2885
2886 #define RECURSION_LIMIT 10
2887
2888 /**
2889 * dev_loopback_xmit - loop back @skb
2890 * @skb: buffer to transmit
2891 */
dev_loopback_xmit(struct sock * sk,struct sk_buff * skb)2892 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb)
2893 {
2894 skb_reset_mac_header(skb);
2895 __skb_pull(skb, skb_network_offset(skb));
2896 skb->pkt_type = PACKET_LOOPBACK;
2897 skb->ip_summed = CHECKSUM_UNNECESSARY;
2898 WARN_ON(!skb_dst(skb));
2899 skb_dst_force(skb);
2900 netif_rx_ni(skb);
2901 return 0;
2902 }
2903 EXPORT_SYMBOL(dev_loopback_xmit);
2904
2905 /**
2906 * __dev_queue_xmit - transmit a buffer
2907 * @skb: buffer to transmit
2908 * @accel_priv: private data used for L2 forwarding offload
2909 *
2910 * Queue a buffer for transmission to a network device. The caller must
2911 * have set the device and priority and built the buffer before calling
2912 * this function. The function can be called from an interrupt.
2913 *
2914 * A negative errno code is returned on a failure. A success does not
2915 * guarantee the frame will be transmitted as it may be dropped due
2916 * to congestion or traffic shaping.
2917 *
2918 * -----------------------------------------------------------------------------------
2919 * I notice this method can also return errors from the queue disciplines,
2920 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2921 * be positive.
2922 *
2923 * Regardless of the return value, the skb is consumed, so it is currently
2924 * difficult to retry a send to this method. (You can bump the ref count
2925 * before sending to hold a reference for retry if you are careful.)
2926 *
2927 * When calling this method, interrupts MUST be enabled. This is because
2928 * the BH enable code must have IRQs enabled so that it will not deadlock.
2929 * --BLG
2930 */
__dev_queue_xmit(struct sk_buff * skb,void * accel_priv)2931 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2932 {
2933 struct net_device *dev = skb->dev;
2934 struct netdev_queue *txq;
2935 struct Qdisc *q;
2936 int rc = -ENOMEM;
2937
2938 skb_reset_mac_header(skb);
2939
2940 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2941 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2942
2943 /* Disable soft irqs for various locks below. Also
2944 * stops preemption for RCU.
2945 */
2946 rcu_read_lock_bh();
2947
2948 skb_update_prio(skb);
2949
2950 /* If device/qdisc don't need skb->dst, release it right now while
2951 * its hot in this cpu cache.
2952 */
2953 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2954 skb_dst_drop(skb);
2955 else
2956 skb_dst_force(skb);
2957
2958 txq = netdev_pick_tx(dev, skb, accel_priv);
2959 q = rcu_dereference_bh(txq->qdisc);
2960
2961 #ifdef CONFIG_NET_CLS_ACT
2962 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2963 #endif
2964 trace_net_dev_queue(skb);
2965 if (q->enqueue) {
2966 rc = __dev_xmit_skb(skb, q, dev, txq);
2967 goto out;
2968 }
2969
2970 /* The device has no queue. Common case for software devices:
2971 loopback, all the sorts of tunnels...
2972
2973 Really, it is unlikely that netif_tx_lock protection is necessary
2974 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2975 counters.)
2976 However, it is possible, that they rely on protection
2977 made by us here.
2978
2979 Check this and shot the lock. It is not prone from deadlocks.
2980 Either shot noqueue qdisc, it is even simpler 8)
2981 */
2982 if (dev->flags & IFF_UP) {
2983 int cpu = smp_processor_id(); /* ok because BHs are off */
2984
2985 if (txq->xmit_lock_owner != cpu) {
2986
2987 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2988 goto recursion_alert;
2989
2990 skb = validate_xmit_skb(skb, dev);
2991 if (!skb)
2992 goto drop;
2993
2994 HARD_TX_LOCK(dev, txq, cpu);
2995
2996 if (!netif_xmit_stopped(txq)) {
2997 __this_cpu_inc(xmit_recursion);
2998 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2999 __this_cpu_dec(xmit_recursion);
3000 if (dev_xmit_complete(rc)) {
3001 HARD_TX_UNLOCK(dev, txq);
3002 goto out;
3003 }
3004 }
3005 HARD_TX_UNLOCK(dev, txq);
3006 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3007 dev->name);
3008 } else {
3009 /* Recursion is detected! It is possible,
3010 * unfortunately
3011 */
3012 recursion_alert:
3013 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3014 dev->name);
3015 }
3016 }
3017
3018 rc = -ENETDOWN;
3019 drop:
3020 rcu_read_unlock_bh();
3021
3022 atomic_long_inc(&dev->tx_dropped);
3023 kfree_skb_list(skb);
3024 return rc;
3025 out:
3026 rcu_read_unlock_bh();
3027 return rc;
3028 }
3029
dev_queue_xmit_sk(struct sock * sk,struct sk_buff * skb)3030 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb)
3031 {
3032 return __dev_queue_xmit(skb, NULL);
3033 }
3034 EXPORT_SYMBOL(dev_queue_xmit_sk);
3035
dev_queue_xmit_accel(struct sk_buff * skb,void * accel_priv)3036 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3037 {
3038 return __dev_queue_xmit(skb, accel_priv);
3039 }
3040 EXPORT_SYMBOL(dev_queue_xmit_accel);
3041
3042
3043 /*=======================================================================
3044 Receiver routines
3045 =======================================================================*/
3046
3047 int netdev_max_backlog __read_mostly = 1000;
3048 EXPORT_SYMBOL(netdev_max_backlog);
3049
3050 int netdev_tstamp_prequeue __read_mostly = 1;
3051 int netdev_budget __read_mostly = 300;
3052 int weight_p __read_mostly = 64; /* old backlog weight */
3053
3054 /* Called with irq disabled */
____napi_schedule(struct softnet_data * sd,struct napi_struct * napi)3055 static inline void ____napi_schedule(struct softnet_data *sd,
3056 struct napi_struct *napi)
3057 {
3058 list_add_tail(&napi->poll_list, &sd->poll_list);
3059 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3060 }
3061
3062 #ifdef CONFIG_RPS
3063
3064 /* One global table that all flow-based protocols share. */
3065 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3066 EXPORT_SYMBOL(rps_sock_flow_table);
3067 u32 rps_cpu_mask __read_mostly;
3068 EXPORT_SYMBOL(rps_cpu_mask);
3069
3070 struct static_key rps_needed __read_mostly;
3071
3072 static struct rps_dev_flow *
set_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow * rflow,u16 next_cpu)3073 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3074 struct rps_dev_flow *rflow, u16 next_cpu)
3075 {
3076 if (next_cpu < nr_cpu_ids) {
3077 #ifdef CONFIG_RFS_ACCEL
3078 struct netdev_rx_queue *rxqueue;
3079 struct rps_dev_flow_table *flow_table;
3080 struct rps_dev_flow *old_rflow;
3081 u32 flow_id;
3082 u16 rxq_index;
3083 int rc;
3084
3085 /* Should we steer this flow to a different hardware queue? */
3086 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3087 !(dev->features & NETIF_F_NTUPLE))
3088 goto out;
3089 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3090 if (rxq_index == skb_get_rx_queue(skb))
3091 goto out;
3092
3093 rxqueue = dev->_rx + rxq_index;
3094 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3095 if (!flow_table)
3096 goto out;
3097 flow_id = skb_get_hash(skb) & flow_table->mask;
3098 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3099 rxq_index, flow_id);
3100 if (rc < 0)
3101 goto out;
3102 old_rflow = rflow;
3103 rflow = &flow_table->flows[flow_id];
3104 rflow->filter = rc;
3105 if (old_rflow->filter == rflow->filter)
3106 old_rflow->filter = RPS_NO_FILTER;
3107 out:
3108 #endif
3109 rflow->last_qtail =
3110 per_cpu(softnet_data, next_cpu).input_queue_head;
3111 }
3112
3113 rflow->cpu = next_cpu;
3114 return rflow;
3115 }
3116
3117 /*
3118 * get_rps_cpu is called from netif_receive_skb and returns the target
3119 * CPU from the RPS map of the receiving queue for a given skb.
3120 * rcu_read_lock must be held on entry.
3121 */
get_rps_cpu(struct net_device * dev,struct sk_buff * skb,struct rps_dev_flow ** rflowp)3122 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3123 struct rps_dev_flow **rflowp)
3124 {
3125 const struct rps_sock_flow_table *sock_flow_table;
3126 struct netdev_rx_queue *rxqueue = dev->_rx;
3127 struct rps_dev_flow_table *flow_table;
3128 struct rps_map *map;
3129 int cpu = -1;
3130 u32 tcpu;
3131 u32 hash;
3132
3133 if (skb_rx_queue_recorded(skb)) {
3134 u16 index = skb_get_rx_queue(skb);
3135
3136 if (unlikely(index >= dev->real_num_rx_queues)) {
3137 WARN_ONCE(dev->real_num_rx_queues > 1,
3138 "%s received packet on queue %u, but number "
3139 "of RX queues is %u\n",
3140 dev->name, index, dev->real_num_rx_queues);
3141 goto done;
3142 }
3143 rxqueue += index;
3144 }
3145
3146 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3147
3148 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3149 map = rcu_dereference(rxqueue->rps_map);
3150 if (!flow_table && !map)
3151 goto done;
3152
3153 skb_reset_network_header(skb);
3154 hash = skb_get_hash(skb);
3155 if (!hash)
3156 goto done;
3157
3158 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3159 if (flow_table && sock_flow_table) {
3160 struct rps_dev_flow *rflow;
3161 u32 next_cpu;
3162 u32 ident;
3163
3164 /* First check into global flow table if there is a match */
3165 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3166 if ((ident ^ hash) & ~rps_cpu_mask)
3167 goto try_rps;
3168
3169 next_cpu = ident & rps_cpu_mask;
3170
3171 /* OK, now we know there is a match,
3172 * we can look at the local (per receive queue) flow table
3173 */
3174 rflow = &flow_table->flows[hash & flow_table->mask];
3175 tcpu = rflow->cpu;
3176
3177 /*
3178 * If the desired CPU (where last recvmsg was done) is
3179 * different from current CPU (one in the rx-queue flow
3180 * table entry), switch if one of the following holds:
3181 * - Current CPU is unset (>= nr_cpu_ids).
3182 * - Current CPU is offline.
3183 * - The current CPU's queue tail has advanced beyond the
3184 * last packet that was enqueued using this table entry.
3185 * This guarantees that all previous packets for the flow
3186 * have been dequeued, thus preserving in order delivery.
3187 */
3188 if (unlikely(tcpu != next_cpu) &&
3189 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3190 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3191 rflow->last_qtail)) >= 0)) {
3192 tcpu = next_cpu;
3193 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3194 }
3195
3196 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3197 *rflowp = rflow;
3198 cpu = tcpu;
3199 goto done;
3200 }
3201 }
3202
3203 try_rps:
3204
3205 if (map) {
3206 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3207 if (cpu_online(tcpu)) {
3208 cpu = tcpu;
3209 goto done;
3210 }
3211 }
3212
3213 done:
3214 return cpu;
3215 }
3216
3217 #ifdef CONFIG_RFS_ACCEL
3218
3219 /**
3220 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3221 * @dev: Device on which the filter was set
3222 * @rxq_index: RX queue index
3223 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3224 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3225 *
3226 * Drivers that implement ndo_rx_flow_steer() should periodically call
3227 * this function for each installed filter and remove the filters for
3228 * which it returns %true.
3229 */
rps_may_expire_flow(struct net_device * dev,u16 rxq_index,u32 flow_id,u16 filter_id)3230 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3231 u32 flow_id, u16 filter_id)
3232 {
3233 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3234 struct rps_dev_flow_table *flow_table;
3235 struct rps_dev_flow *rflow;
3236 bool expire = true;
3237 unsigned int cpu;
3238
3239 rcu_read_lock();
3240 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3241 if (flow_table && flow_id <= flow_table->mask) {
3242 rflow = &flow_table->flows[flow_id];
3243 cpu = ACCESS_ONCE(rflow->cpu);
3244 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3245 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3246 rflow->last_qtail) <
3247 (int)(10 * flow_table->mask)))
3248 expire = false;
3249 }
3250 rcu_read_unlock();
3251 return expire;
3252 }
3253 EXPORT_SYMBOL(rps_may_expire_flow);
3254
3255 #endif /* CONFIG_RFS_ACCEL */
3256
3257 /* Called from hardirq (IPI) context */
rps_trigger_softirq(void * data)3258 static void rps_trigger_softirq(void *data)
3259 {
3260 struct softnet_data *sd = data;
3261
3262 ____napi_schedule(sd, &sd->backlog);
3263 sd->received_rps++;
3264 }
3265
3266 #endif /* CONFIG_RPS */
3267
3268 /*
3269 * Check if this softnet_data structure is another cpu one
3270 * If yes, queue it to our IPI list and return 1
3271 * If no, return 0
3272 */
rps_ipi_queued(struct softnet_data * sd)3273 static int rps_ipi_queued(struct softnet_data *sd)
3274 {
3275 #ifdef CONFIG_RPS
3276 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3277
3278 if (sd != mysd) {
3279 sd->rps_ipi_next = mysd->rps_ipi_list;
3280 mysd->rps_ipi_list = sd;
3281
3282 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3283 return 1;
3284 }
3285 #endif /* CONFIG_RPS */
3286 return 0;
3287 }
3288
3289 #ifdef CONFIG_NET_FLOW_LIMIT
3290 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3291 #endif
3292
skb_flow_limit(struct sk_buff * skb,unsigned int qlen)3293 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3294 {
3295 #ifdef CONFIG_NET_FLOW_LIMIT
3296 struct sd_flow_limit *fl;
3297 struct softnet_data *sd;
3298 unsigned int old_flow, new_flow;
3299
3300 if (qlen < (netdev_max_backlog >> 1))
3301 return false;
3302
3303 sd = this_cpu_ptr(&softnet_data);
3304
3305 rcu_read_lock();
3306 fl = rcu_dereference(sd->flow_limit);
3307 if (fl) {
3308 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3309 old_flow = fl->history[fl->history_head];
3310 fl->history[fl->history_head] = new_flow;
3311
3312 fl->history_head++;
3313 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3314
3315 if (likely(fl->buckets[old_flow]))
3316 fl->buckets[old_flow]--;
3317
3318 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3319 fl->count++;
3320 rcu_read_unlock();
3321 return true;
3322 }
3323 }
3324 rcu_read_unlock();
3325 #endif
3326 return false;
3327 }
3328
3329 /*
3330 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3331 * queue (may be a remote CPU queue).
3332 */
enqueue_to_backlog(struct sk_buff * skb,int cpu,unsigned int * qtail)3333 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3334 unsigned int *qtail)
3335 {
3336 struct softnet_data *sd;
3337 unsigned long flags;
3338 unsigned int qlen;
3339
3340 sd = &per_cpu(softnet_data, cpu);
3341
3342 local_irq_save(flags);
3343
3344 rps_lock(sd);
3345 if (!netif_running(skb->dev))
3346 goto drop;
3347 qlen = skb_queue_len(&sd->input_pkt_queue);
3348 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3349 if (qlen) {
3350 enqueue:
3351 __skb_queue_tail(&sd->input_pkt_queue, skb);
3352 input_queue_tail_incr_save(sd, qtail);
3353 rps_unlock(sd);
3354 local_irq_restore(flags);
3355 return NET_RX_SUCCESS;
3356 }
3357
3358 /* Schedule NAPI for backlog device
3359 * We can use non atomic operation since we own the queue lock
3360 */
3361 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3362 if (!rps_ipi_queued(sd))
3363 ____napi_schedule(sd, &sd->backlog);
3364 }
3365 goto enqueue;
3366 }
3367
3368 drop:
3369 sd->dropped++;
3370 rps_unlock(sd);
3371
3372 local_irq_restore(flags);
3373
3374 atomic_long_inc(&skb->dev->rx_dropped);
3375 kfree_skb(skb);
3376 return NET_RX_DROP;
3377 }
3378
netif_rx_internal(struct sk_buff * skb)3379 static int netif_rx_internal(struct sk_buff *skb)
3380 {
3381 int ret;
3382
3383 net_timestamp_check(netdev_tstamp_prequeue, skb);
3384
3385 trace_netif_rx(skb);
3386 #ifdef CONFIG_RPS
3387 if (static_key_false(&rps_needed)) {
3388 struct rps_dev_flow voidflow, *rflow = &voidflow;
3389 int cpu;
3390
3391 preempt_disable();
3392 rcu_read_lock();
3393
3394 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3395 if (cpu < 0)
3396 cpu = smp_processor_id();
3397
3398 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3399
3400 rcu_read_unlock();
3401 preempt_enable();
3402 } else
3403 #endif
3404 {
3405 unsigned int qtail;
3406 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3407 put_cpu();
3408 }
3409 return ret;
3410 }
3411
3412 /**
3413 * netif_rx - post buffer to the network code
3414 * @skb: buffer to post
3415 *
3416 * This function receives a packet from a device driver and queues it for
3417 * the upper (protocol) levels to process. It always succeeds. The buffer
3418 * may be dropped during processing for congestion control or by the
3419 * protocol layers.
3420 *
3421 * return values:
3422 * NET_RX_SUCCESS (no congestion)
3423 * NET_RX_DROP (packet was dropped)
3424 *
3425 */
3426
netif_rx(struct sk_buff * skb)3427 int netif_rx(struct sk_buff *skb)
3428 {
3429 trace_netif_rx_entry(skb);
3430
3431 return netif_rx_internal(skb);
3432 }
3433 EXPORT_SYMBOL(netif_rx);
3434
netif_rx_ni(struct sk_buff * skb)3435 int netif_rx_ni(struct sk_buff *skb)
3436 {
3437 int err;
3438
3439 trace_netif_rx_ni_entry(skb);
3440
3441 preempt_disable();
3442 err = netif_rx_internal(skb);
3443 if (local_softirq_pending())
3444 do_softirq();
3445 preempt_enable();
3446
3447 return err;
3448 }
3449 EXPORT_SYMBOL(netif_rx_ni);
3450
net_tx_action(struct softirq_action * h)3451 static void net_tx_action(struct softirq_action *h)
3452 {
3453 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3454
3455 if (sd->completion_queue) {
3456 struct sk_buff *clist;
3457
3458 local_irq_disable();
3459 clist = sd->completion_queue;
3460 sd->completion_queue = NULL;
3461 local_irq_enable();
3462
3463 while (clist) {
3464 struct sk_buff *skb = clist;
3465 clist = clist->next;
3466
3467 WARN_ON(atomic_read(&skb->users));
3468 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3469 trace_consume_skb(skb);
3470 else
3471 trace_kfree_skb(skb, net_tx_action);
3472 __kfree_skb(skb);
3473 }
3474 }
3475
3476 if (sd->output_queue) {
3477 struct Qdisc *head;
3478
3479 local_irq_disable();
3480 head = sd->output_queue;
3481 sd->output_queue = NULL;
3482 sd->output_queue_tailp = &sd->output_queue;
3483 local_irq_enable();
3484
3485 while (head) {
3486 struct Qdisc *q = head;
3487 spinlock_t *root_lock;
3488
3489 head = head->next_sched;
3490
3491 root_lock = qdisc_lock(q);
3492 if (spin_trylock(root_lock)) {
3493 smp_mb__before_atomic();
3494 clear_bit(__QDISC_STATE_SCHED,
3495 &q->state);
3496 qdisc_run(q);
3497 spin_unlock(root_lock);
3498 } else {
3499 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3500 &q->state)) {
3501 __netif_reschedule(q);
3502 } else {
3503 smp_mb__before_atomic();
3504 clear_bit(__QDISC_STATE_SCHED,
3505 &q->state);
3506 }
3507 }
3508 }
3509 }
3510 }
3511
3512 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3513 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3514 /* This hook is defined here for ATM LANE */
3515 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3516 unsigned char *addr) __read_mostly;
3517 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3518 #endif
3519
3520 #ifdef CONFIG_NET_CLS_ACT
3521 /* TODO: Maybe we should just force sch_ingress to be compiled in
3522 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3523 * a compare and 2 stores extra right now if we dont have it on
3524 * but have CONFIG_NET_CLS_ACT
3525 * NOTE: This doesn't stop any functionality; if you dont have
3526 * the ingress scheduler, you just can't add policies on ingress.
3527 *
3528 */
ing_filter(struct sk_buff * skb,struct netdev_queue * rxq)3529 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3530 {
3531 struct net_device *dev = skb->dev;
3532 u32 ttl = G_TC_RTTL(skb->tc_verd);
3533 int result = TC_ACT_OK;
3534 struct Qdisc *q;
3535
3536 if (unlikely(MAX_RED_LOOP < ttl++)) {
3537 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3538 skb->skb_iif, dev->ifindex);
3539 return TC_ACT_SHOT;
3540 }
3541
3542 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3543 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3544
3545 q = rcu_dereference(rxq->qdisc);
3546 if (q != &noop_qdisc) {
3547 spin_lock(qdisc_lock(q));
3548 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3549 result = qdisc_enqueue_root(skb, q);
3550 spin_unlock(qdisc_lock(q));
3551 }
3552
3553 return result;
3554 }
3555
handle_ing(struct sk_buff * skb,struct packet_type ** pt_prev,int * ret,struct net_device * orig_dev)3556 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3557 struct packet_type **pt_prev,
3558 int *ret, struct net_device *orig_dev)
3559 {
3560 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3561
3562 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3563 return skb;
3564
3565 if (*pt_prev) {
3566 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3567 *pt_prev = NULL;
3568 }
3569
3570 switch (ing_filter(skb, rxq)) {
3571 case TC_ACT_SHOT:
3572 case TC_ACT_STOLEN:
3573 kfree_skb(skb);
3574 return NULL;
3575 }
3576
3577 return skb;
3578 }
3579 #endif
3580
3581 /**
3582 * netdev_rx_handler_register - register receive handler
3583 * @dev: device to register a handler for
3584 * @rx_handler: receive handler to register
3585 * @rx_handler_data: data pointer that is used by rx handler
3586 *
3587 * Register a receive handler for a device. This handler will then be
3588 * called from __netif_receive_skb. A negative errno code is returned
3589 * on a failure.
3590 *
3591 * The caller must hold the rtnl_mutex.
3592 *
3593 * For a general description of rx_handler, see enum rx_handler_result.
3594 */
netdev_rx_handler_register(struct net_device * dev,rx_handler_func_t * rx_handler,void * rx_handler_data)3595 int netdev_rx_handler_register(struct net_device *dev,
3596 rx_handler_func_t *rx_handler,
3597 void *rx_handler_data)
3598 {
3599 ASSERT_RTNL();
3600
3601 if (dev->rx_handler)
3602 return -EBUSY;
3603
3604 /* Note: rx_handler_data must be set before rx_handler */
3605 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3606 rcu_assign_pointer(dev->rx_handler, rx_handler);
3607
3608 return 0;
3609 }
3610 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3611
3612 /**
3613 * netdev_rx_handler_unregister - unregister receive handler
3614 * @dev: device to unregister a handler from
3615 *
3616 * Unregister a receive handler from a device.
3617 *
3618 * The caller must hold the rtnl_mutex.
3619 */
netdev_rx_handler_unregister(struct net_device * dev)3620 void netdev_rx_handler_unregister(struct net_device *dev)
3621 {
3622
3623 ASSERT_RTNL();
3624 RCU_INIT_POINTER(dev->rx_handler, NULL);
3625 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3626 * section has a guarantee to see a non NULL rx_handler_data
3627 * as well.
3628 */
3629 synchronize_net();
3630 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3631 }
3632 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3633
3634 /*
3635 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3636 * the special handling of PFMEMALLOC skbs.
3637 */
skb_pfmemalloc_protocol(struct sk_buff * skb)3638 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3639 {
3640 switch (skb->protocol) {
3641 case htons(ETH_P_ARP):
3642 case htons(ETH_P_IP):
3643 case htons(ETH_P_IPV6):
3644 case htons(ETH_P_8021Q):
3645 case htons(ETH_P_8021AD):
3646 return true;
3647 default:
3648 return false;
3649 }
3650 }
3651
__netif_receive_skb_core(struct sk_buff * skb,bool pfmemalloc)3652 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3653 {
3654 struct packet_type *ptype, *pt_prev;
3655 rx_handler_func_t *rx_handler;
3656 struct net_device *orig_dev;
3657 bool deliver_exact = false;
3658 int ret = NET_RX_DROP;
3659 __be16 type;
3660
3661 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3662
3663 trace_netif_receive_skb(skb);
3664
3665 orig_dev = skb->dev;
3666
3667 skb_reset_network_header(skb);
3668 if (!skb_transport_header_was_set(skb))
3669 skb_reset_transport_header(skb);
3670 skb_reset_mac_len(skb);
3671
3672 pt_prev = NULL;
3673
3674 another_round:
3675 skb->skb_iif = skb->dev->ifindex;
3676
3677 __this_cpu_inc(softnet_data.processed);
3678
3679 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3680 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3681 skb = skb_vlan_untag(skb);
3682 if (unlikely(!skb))
3683 goto out;
3684 }
3685
3686 #ifdef CONFIG_NET_CLS_ACT
3687 if (skb->tc_verd & TC_NCLS) {
3688 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3689 goto ncls;
3690 }
3691 #endif
3692
3693 if (pfmemalloc)
3694 goto skip_taps;
3695
3696 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3697 if (pt_prev)
3698 ret = deliver_skb(skb, pt_prev, orig_dev);
3699 pt_prev = ptype;
3700 }
3701
3702 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3703 if (pt_prev)
3704 ret = deliver_skb(skb, pt_prev, orig_dev);
3705 pt_prev = ptype;
3706 }
3707
3708 skip_taps:
3709 #ifdef CONFIG_NET_CLS_ACT
3710 if (static_key_false(&ingress_needed)) {
3711 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3712 if (!skb)
3713 goto out;
3714 }
3715
3716 skb->tc_verd = 0;
3717 ncls:
3718 #endif
3719 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3720 goto drop;
3721
3722 if (skb_vlan_tag_present(skb)) {
3723 if (pt_prev) {
3724 ret = deliver_skb(skb, pt_prev, orig_dev);
3725 pt_prev = NULL;
3726 }
3727 if (vlan_do_receive(&skb))
3728 goto another_round;
3729 else if (unlikely(!skb))
3730 goto out;
3731 }
3732
3733 rx_handler = rcu_dereference(skb->dev->rx_handler);
3734 if (rx_handler) {
3735 if (pt_prev) {
3736 ret = deliver_skb(skb, pt_prev, orig_dev);
3737 pt_prev = NULL;
3738 }
3739 switch (rx_handler(&skb)) {
3740 case RX_HANDLER_CONSUMED:
3741 ret = NET_RX_SUCCESS;
3742 goto out;
3743 case RX_HANDLER_ANOTHER:
3744 goto another_round;
3745 case RX_HANDLER_EXACT:
3746 deliver_exact = true;
3747 case RX_HANDLER_PASS:
3748 break;
3749 default:
3750 BUG();
3751 }
3752 }
3753
3754 if (unlikely(skb_vlan_tag_present(skb))) {
3755 if (skb_vlan_tag_get_id(skb))
3756 skb->pkt_type = PACKET_OTHERHOST;
3757 /* Note: we might in the future use prio bits
3758 * and set skb->priority like in vlan_do_receive()
3759 * For the time being, just ignore Priority Code Point
3760 */
3761 skb->vlan_tci = 0;
3762 }
3763
3764 type = skb->protocol;
3765
3766 /* deliver only exact match when indicated */
3767 if (likely(!deliver_exact)) {
3768 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3769 &ptype_base[ntohs(type) &
3770 PTYPE_HASH_MASK]);
3771 }
3772
3773 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3774 &orig_dev->ptype_specific);
3775
3776 if (unlikely(skb->dev != orig_dev)) {
3777 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3778 &skb->dev->ptype_specific);
3779 }
3780
3781 if (pt_prev) {
3782 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3783 goto drop;
3784 else
3785 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3786 } else {
3787 drop:
3788 atomic_long_inc(&skb->dev->rx_dropped);
3789 kfree_skb(skb);
3790 /* Jamal, now you will not able to escape explaining
3791 * me how you were going to use this. :-)
3792 */
3793 ret = NET_RX_DROP;
3794 }
3795
3796 out:
3797 return ret;
3798 }
3799
__netif_receive_skb(struct sk_buff * skb)3800 static int __netif_receive_skb(struct sk_buff *skb)
3801 {
3802 int ret;
3803
3804 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3805 unsigned long pflags = current->flags;
3806
3807 /*
3808 * PFMEMALLOC skbs are special, they should
3809 * - be delivered to SOCK_MEMALLOC sockets only
3810 * - stay away from userspace
3811 * - have bounded memory usage
3812 *
3813 * Use PF_MEMALLOC as this saves us from propagating the allocation
3814 * context down to all allocation sites.
3815 */
3816 current->flags |= PF_MEMALLOC;
3817 ret = __netif_receive_skb_core(skb, true);
3818 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3819 } else
3820 ret = __netif_receive_skb_core(skb, false);
3821
3822 return ret;
3823 }
3824
netif_receive_skb_internal(struct sk_buff * skb)3825 static int netif_receive_skb_internal(struct sk_buff *skb)
3826 {
3827 int ret;
3828
3829 net_timestamp_check(netdev_tstamp_prequeue, skb);
3830
3831 if (skb_defer_rx_timestamp(skb))
3832 return NET_RX_SUCCESS;
3833
3834 rcu_read_lock();
3835
3836 #ifdef CONFIG_RPS
3837 if (static_key_false(&rps_needed)) {
3838 struct rps_dev_flow voidflow, *rflow = &voidflow;
3839 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
3840
3841 if (cpu >= 0) {
3842 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3843 rcu_read_unlock();
3844 return ret;
3845 }
3846 }
3847 #endif
3848 ret = __netif_receive_skb(skb);
3849 rcu_read_unlock();
3850 return ret;
3851 }
3852
3853 /**
3854 * netif_receive_skb - process receive buffer from network
3855 * @skb: buffer to process
3856 *
3857 * netif_receive_skb() is the main receive data processing function.
3858 * It always succeeds. The buffer may be dropped during processing
3859 * for congestion control or by the protocol layers.
3860 *
3861 * This function may only be called from softirq context and interrupts
3862 * should be enabled.
3863 *
3864 * Return values (usually ignored):
3865 * NET_RX_SUCCESS: no congestion
3866 * NET_RX_DROP: packet was dropped
3867 */
netif_receive_skb_sk(struct sock * sk,struct sk_buff * skb)3868 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb)
3869 {
3870 trace_netif_receive_skb_entry(skb);
3871
3872 return netif_receive_skb_internal(skb);
3873 }
3874 EXPORT_SYMBOL(netif_receive_skb_sk);
3875
3876 /* Network device is going away, flush any packets still pending
3877 * Called with irqs disabled.
3878 */
flush_backlog(void * arg)3879 static void flush_backlog(void *arg)
3880 {
3881 struct net_device *dev = arg;
3882 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3883 struct sk_buff *skb, *tmp;
3884
3885 rps_lock(sd);
3886 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3887 if (skb->dev == dev) {
3888 __skb_unlink(skb, &sd->input_pkt_queue);
3889 kfree_skb(skb);
3890 input_queue_head_incr(sd);
3891 }
3892 }
3893 rps_unlock(sd);
3894
3895 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3896 if (skb->dev == dev) {
3897 __skb_unlink(skb, &sd->process_queue);
3898 kfree_skb(skb);
3899 input_queue_head_incr(sd);
3900 }
3901 }
3902 }
3903
napi_gro_complete(struct sk_buff * skb)3904 static int napi_gro_complete(struct sk_buff *skb)
3905 {
3906 struct packet_offload *ptype;
3907 __be16 type = skb->protocol;
3908 struct list_head *head = &offload_base;
3909 int err = -ENOENT;
3910
3911 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3912
3913 if (NAPI_GRO_CB(skb)->count == 1) {
3914 skb_shinfo(skb)->gso_size = 0;
3915 goto out;
3916 }
3917
3918 rcu_read_lock();
3919 list_for_each_entry_rcu(ptype, head, list) {
3920 if (ptype->type != type || !ptype->callbacks.gro_complete)
3921 continue;
3922
3923 err = ptype->callbacks.gro_complete(skb, 0);
3924 break;
3925 }
3926 rcu_read_unlock();
3927
3928 if (err) {
3929 WARN_ON(&ptype->list == head);
3930 kfree_skb(skb);
3931 return NET_RX_SUCCESS;
3932 }
3933
3934 out:
3935 return netif_receive_skb_internal(skb);
3936 }
3937
3938 /* napi->gro_list contains packets ordered by age.
3939 * youngest packets at the head of it.
3940 * Complete skbs in reverse order to reduce latencies.
3941 */
napi_gro_flush(struct napi_struct * napi,bool flush_old)3942 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3943 {
3944 struct sk_buff *skb, *prev = NULL;
3945
3946 /* scan list and build reverse chain */
3947 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3948 skb->prev = prev;
3949 prev = skb;
3950 }
3951
3952 for (skb = prev; skb; skb = prev) {
3953 skb->next = NULL;
3954
3955 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3956 return;
3957
3958 prev = skb->prev;
3959 napi_gro_complete(skb);
3960 napi->gro_count--;
3961 }
3962
3963 napi->gro_list = NULL;
3964 }
3965 EXPORT_SYMBOL(napi_gro_flush);
3966
gro_list_prepare(struct napi_struct * napi,struct sk_buff * skb)3967 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3968 {
3969 struct sk_buff *p;
3970 unsigned int maclen = skb->dev->hard_header_len;
3971 u32 hash = skb_get_hash_raw(skb);
3972
3973 for (p = napi->gro_list; p; p = p->next) {
3974 unsigned long diffs;
3975
3976 NAPI_GRO_CB(p)->flush = 0;
3977
3978 if (hash != skb_get_hash_raw(p)) {
3979 NAPI_GRO_CB(p)->same_flow = 0;
3980 continue;
3981 }
3982
3983 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3984 diffs |= p->vlan_tci ^ skb->vlan_tci;
3985 if (maclen == ETH_HLEN)
3986 diffs |= compare_ether_header(skb_mac_header(p),
3987 skb_mac_header(skb));
3988 else if (!diffs)
3989 diffs = memcmp(skb_mac_header(p),
3990 skb_mac_header(skb),
3991 maclen);
3992 NAPI_GRO_CB(p)->same_flow = !diffs;
3993 }
3994 }
3995
skb_gro_reset_offset(struct sk_buff * skb)3996 static void skb_gro_reset_offset(struct sk_buff *skb)
3997 {
3998 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3999 const skb_frag_t *frag0 = &pinfo->frags[0];
4000
4001 NAPI_GRO_CB(skb)->data_offset = 0;
4002 NAPI_GRO_CB(skb)->frag0 = NULL;
4003 NAPI_GRO_CB(skb)->frag0_len = 0;
4004
4005 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4006 pinfo->nr_frags &&
4007 !PageHighMem(skb_frag_page(frag0))) {
4008 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4009 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4010 }
4011 }
4012
gro_pull_from_frag0(struct sk_buff * skb,int grow)4013 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4014 {
4015 struct skb_shared_info *pinfo = skb_shinfo(skb);
4016
4017 BUG_ON(skb->end - skb->tail < grow);
4018
4019 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4020
4021 skb->data_len -= grow;
4022 skb->tail += grow;
4023
4024 pinfo->frags[0].page_offset += grow;
4025 skb_frag_size_sub(&pinfo->frags[0], grow);
4026
4027 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4028 skb_frag_unref(skb, 0);
4029 memmove(pinfo->frags, pinfo->frags + 1,
4030 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4031 }
4032 }
4033
dev_gro_receive(struct napi_struct * napi,struct sk_buff * skb)4034 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4035 {
4036 struct sk_buff **pp = NULL;
4037 struct packet_offload *ptype;
4038 __be16 type = skb->protocol;
4039 struct list_head *head = &offload_base;
4040 int same_flow;
4041 enum gro_result ret;
4042 int grow;
4043
4044 if (!(skb->dev->features & NETIF_F_GRO))
4045 goto normal;
4046
4047 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4048 goto normal;
4049
4050 gro_list_prepare(napi, skb);
4051
4052 rcu_read_lock();
4053 list_for_each_entry_rcu(ptype, head, list) {
4054 if (ptype->type != type || !ptype->callbacks.gro_receive)
4055 continue;
4056
4057 skb_set_network_header(skb, skb_gro_offset(skb));
4058 skb_reset_mac_len(skb);
4059 NAPI_GRO_CB(skb)->same_flow = 0;
4060 NAPI_GRO_CB(skb)->flush = 0;
4061 NAPI_GRO_CB(skb)->free = 0;
4062 NAPI_GRO_CB(skb)->udp_mark = 0;
4063 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4064
4065 /* Setup for GRO checksum validation */
4066 switch (skb->ip_summed) {
4067 case CHECKSUM_COMPLETE:
4068 NAPI_GRO_CB(skb)->csum = skb->csum;
4069 NAPI_GRO_CB(skb)->csum_valid = 1;
4070 NAPI_GRO_CB(skb)->csum_cnt = 0;
4071 break;
4072 case CHECKSUM_UNNECESSARY:
4073 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4074 NAPI_GRO_CB(skb)->csum_valid = 0;
4075 break;
4076 default:
4077 NAPI_GRO_CB(skb)->csum_cnt = 0;
4078 NAPI_GRO_CB(skb)->csum_valid = 0;
4079 }
4080
4081 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4082 break;
4083 }
4084 rcu_read_unlock();
4085
4086 if (&ptype->list == head)
4087 goto normal;
4088
4089 same_flow = NAPI_GRO_CB(skb)->same_flow;
4090 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4091
4092 if (pp) {
4093 struct sk_buff *nskb = *pp;
4094
4095 *pp = nskb->next;
4096 nskb->next = NULL;
4097 napi_gro_complete(nskb);
4098 napi->gro_count--;
4099 }
4100
4101 if (same_flow)
4102 goto ok;
4103
4104 if (NAPI_GRO_CB(skb)->flush)
4105 goto normal;
4106
4107 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4108 struct sk_buff *nskb = napi->gro_list;
4109
4110 /* locate the end of the list to select the 'oldest' flow */
4111 while (nskb->next) {
4112 pp = &nskb->next;
4113 nskb = *pp;
4114 }
4115 *pp = NULL;
4116 nskb->next = NULL;
4117 napi_gro_complete(nskb);
4118 } else {
4119 napi->gro_count++;
4120 }
4121 NAPI_GRO_CB(skb)->count = 1;
4122 NAPI_GRO_CB(skb)->age = jiffies;
4123 NAPI_GRO_CB(skb)->last = skb;
4124 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4125 skb->next = napi->gro_list;
4126 napi->gro_list = skb;
4127 ret = GRO_HELD;
4128
4129 pull:
4130 grow = skb_gro_offset(skb) - skb_headlen(skb);
4131 if (grow > 0)
4132 gro_pull_from_frag0(skb, grow);
4133 ok:
4134 return ret;
4135
4136 normal:
4137 ret = GRO_NORMAL;
4138 goto pull;
4139 }
4140
gro_find_receive_by_type(__be16 type)4141 struct packet_offload *gro_find_receive_by_type(__be16 type)
4142 {
4143 struct list_head *offload_head = &offload_base;
4144 struct packet_offload *ptype;
4145
4146 list_for_each_entry_rcu(ptype, offload_head, list) {
4147 if (ptype->type != type || !ptype->callbacks.gro_receive)
4148 continue;
4149 return ptype;
4150 }
4151 return NULL;
4152 }
4153 EXPORT_SYMBOL(gro_find_receive_by_type);
4154
gro_find_complete_by_type(__be16 type)4155 struct packet_offload *gro_find_complete_by_type(__be16 type)
4156 {
4157 struct list_head *offload_head = &offload_base;
4158 struct packet_offload *ptype;
4159
4160 list_for_each_entry_rcu(ptype, offload_head, list) {
4161 if (ptype->type != type || !ptype->callbacks.gro_complete)
4162 continue;
4163 return ptype;
4164 }
4165 return NULL;
4166 }
4167 EXPORT_SYMBOL(gro_find_complete_by_type);
4168
napi_skb_finish(gro_result_t ret,struct sk_buff * skb)4169 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4170 {
4171 switch (ret) {
4172 case GRO_NORMAL:
4173 if (netif_receive_skb_internal(skb))
4174 ret = GRO_DROP;
4175 break;
4176
4177 case GRO_DROP:
4178 kfree_skb(skb);
4179 break;
4180
4181 case GRO_MERGED_FREE:
4182 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4183 kmem_cache_free(skbuff_head_cache, skb);
4184 else
4185 __kfree_skb(skb);
4186 break;
4187
4188 case GRO_HELD:
4189 case GRO_MERGED:
4190 break;
4191 }
4192
4193 return ret;
4194 }
4195
napi_gro_receive(struct napi_struct * napi,struct sk_buff * skb)4196 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4197 {
4198 trace_napi_gro_receive_entry(skb);
4199
4200 skb_gro_reset_offset(skb);
4201
4202 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4203 }
4204 EXPORT_SYMBOL(napi_gro_receive);
4205
napi_reuse_skb(struct napi_struct * napi,struct sk_buff * skb)4206 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4207 {
4208 if (unlikely(skb->pfmemalloc)) {
4209 consume_skb(skb);
4210 return;
4211 }
4212 __skb_pull(skb, skb_headlen(skb));
4213 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4214 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4215 skb->vlan_tci = 0;
4216 skb->dev = napi->dev;
4217 skb->skb_iif = 0;
4218 skb->encapsulation = 0;
4219 skb_shinfo(skb)->gso_type = 0;
4220 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4221
4222 napi->skb = skb;
4223 }
4224
napi_get_frags(struct napi_struct * napi)4225 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4226 {
4227 struct sk_buff *skb = napi->skb;
4228
4229 if (!skb) {
4230 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4231 napi->skb = skb;
4232 }
4233 return skb;
4234 }
4235 EXPORT_SYMBOL(napi_get_frags);
4236
napi_frags_finish(struct napi_struct * napi,struct sk_buff * skb,gro_result_t ret)4237 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4238 struct sk_buff *skb,
4239 gro_result_t ret)
4240 {
4241 switch (ret) {
4242 case GRO_NORMAL:
4243 case GRO_HELD:
4244 __skb_push(skb, ETH_HLEN);
4245 skb->protocol = eth_type_trans(skb, skb->dev);
4246 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4247 ret = GRO_DROP;
4248 break;
4249
4250 case GRO_DROP:
4251 case GRO_MERGED_FREE:
4252 napi_reuse_skb(napi, skb);
4253 break;
4254
4255 case GRO_MERGED:
4256 break;
4257 }
4258
4259 return ret;
4260 }
4261
4262 /* Upper GRO stack assumes network header starts at gro_offset=0
4263 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4264 * We copy ethernet header into skb->data to have a common layout.
4265 */
napi_frags_skb(struct napi_struct * napi)4266 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4267 {
4268 struct sk_buff *skb = napi->skb;
4269 const struct ethhdr *eth;
4270 unsigned int hlen = sizeof(*eth);
4271
4272 napi->skb = NULL;
4273
4274 skb_reset_mac_header(skb);
4275 skb_gro_reset_offset(skb);
4276
4277 eth = skb_gro_header_fast(skb, 0);
4278 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4279 eth = skb_gro_header_slow(skb, hlen, 0);
4280 if (unlikely(!eth)) {
4281 napi_reuse_skb(napi, skb);
4282 return NULL;
4283 }
4284 } else {
4285 gro_pull_from_frag0(skb, hlen);
4286 NAPI_GRO_CB(skb)->frag0 += hlen;
4287 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4288 }
4289 __skb_pull(skb, hlen);
4290
4291 /*
4292 * This works because the only protocols we care about don't require
4293 * special handling.
4294 * We'll fix it up properly in napi_frags_finish()
4295 */
4296 skb->protocol = eth->h_proto;
4297
4298 return skb;
4299 }
4300
napi_gro_frags(struct napi_struct * napi)4301 gro_result_t napi_gro_frags(struct napi_struct *napi)
4302 {
4303 struct sk_buff *skb = napi_frags_skb(napi);
4304
4305 if (!skb)
4306 return GRO_DROP;
4307
4308 trace_napi_gro_frags_entry(skb);
4309
4310 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4311 }
4312 EXPORT_SYMBOL(napi_gro_frags);
4313
4314 /* Compute the checksum from gro_offset and return the folded value
4315 * after adding in any pseudo checksum.
4316 */
__skb_gro_checksum_complete(struct sk_buff * skb)4317 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4318 {
4319 __wsum wsum;
4320 __sum16 sum;
4321
4322 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4323
4324 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4325 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4326 if (likely(!sum)) {
4327 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4328 !skb->csum_complete_sw)
4329 netdev_rx_csum_fault(skb->dev);
4330 }
4331
4332 NAPI_GRO_CB(skb)->csum = wsum;
4333 NAPI_GRO_CB(skb)->csum_valid = 1;
4334
4335 return sum;
4336 }
4337 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4338
4339 /*
4340 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4341 * Note: called with local irq disabled, but exits with local irq enabled.
4342 */
net_rps_action_and_irq_enable(struct softnet_data * sd)4343 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4344 {
4345 #ifdef CONFIG_RPS
4346 struct softnet_data *remsd = sd->rps_ipi_list;
4347
4348 if (remsd) {
4349 sd->rps_ipi_list = NULL;
4350
4351 local_irq_enable();
4352
4353 /* Send pending IPI's to kick RPS processing on remote cpus. */
4354 while (remsd) {
4355 struct softnet_data *next = remsd->rps_ipi_next;
4356
4357 if (cpu_online(remsd->cpu))
4358 smp_call_function_single_async(remsd->cpu,
4359 &remsd->csd);
4360 remsd = next;
4361 }
4362 } else
4363 #endif
4364 local_irq_enable();
4365 }
4366
sd_has_rps_ipi_waiting(struct softnet_data * sd)4367 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4368 {
4369 #ifdef CONFIG_RPS
4370 return sd->rps_ipi_list != NULL;
4371 #else
4372 return false;
4373 #endif
4374 }
4375
process_backlog(struct napi_struct * napi,int quota)4376 static int process_backlog(struct napi_struct *napi, int quota)
4377 {
4378 int work = 0;
4379 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4380
4381 /* Check if we have pending ipi, its better to send them now,
4382 * not waiting net_rx_action() end.
4383 */
4384 if (sd_has_rps_ipi_waiting(sd)) {
4385 local_irq_disable();
4386 net_rps_action_and_irq_enable(sd);
4387 }
4388
4389 napi->weight = weight_p;
4390 local_irq_disable();
4391 while (1) {
4392 struct sk_buff *skb;
4393
4394 while ((skb = __skb_dequeue(&sd->process_queue))) {
4395 rcu_read_lock();
4396 local_irq_enable();
4397 __netif_receive_skb(skb);
4398 rcu_read_unlock();
4399 local_irq_disable();
4400 input_queue_head_incr(sd);
4401 if (++work >= quota) {
4402 local_irq_enable();
4403 return work;
4404 }
4405 }
4406
4407 rps_lock(sd);
4408 if (skb_queue_empty(&sd->input_pkt_queue)) {
4409 /*
4410 * Inline a custom version of __napi_complete().
4411 * only current cpu owns and manipulates this napi,
4412 * and NAPI_STATE_SCHED is the only possible flag set
4413 * on backlog.
4414 * We can use a plain write instead of clear_bit(),
4415 * and we dont need an smp_mb() memory barrier.
4416 */
4417 napi->state = 0;
4418 rps_unlock(sd);
4419
4420 break;
4421 }
4422
4423 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4424 &sd->process_queue);
4425 rps_unlock(sd);
4426 }
4427 local_irq_enable();
4428
4429 return work;
4430 }
4431
4432 /**
4433 * __napi_schedule - schedule for receive
4434 * @n: entry to schedule
4435 *
4436 * The entry's receive function will be scheduled to run.
4437 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4438 */
__napi_schedule(struct napi_struct * n)4439 void __napi_schedule(struct napi_struct *n)
4440 {
4441 unsigned long flags;
4442
4443 local_irq_save(flags);
4444 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4445 local_irq_restore(flags);
4446 }
4447 EXPORT_SYMBOL(__napi_schedule);
4448
4449 /**
4450 * __napi_schedule_irqoff - schedule for receive
4451 * @n: entry to schedule
4452 *
4453 * Variant of __napi_schedule() assuming hard irqs are masked
4454 */
__napi_schedule_irqoff(struct napi_struct * n)4455 void __napi_schedule_irqoff(struct napi_struct *n)
4456 {
4457 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4458 }
4459 EXPORT_SYMBOL(__napi_schedule_irqoff);
4460
__napi_complete(struct napi_struct * n)4461 void __napi_complete(struct napi_struct *n)
4462 {
4463 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4464
4465 list_del_init(&n->poll_list);
4466 smp_mb__before_atomic();
4467 clear_bit(NAPI_STATE_SCHED, &n->state);
4468 }
4469 EXPORT_SYMBOL(__napi_complete);
4470
napi_complete_done(struct napi_struct * n,int work_done)4471 void napi_complete_done(struct napi_struct *n, int work_done)
4472 {
4473 unsigned long flags;
4474
4475 /*
4476 * don't let napi dequeue from the cpu poll list
4477 * just in case its running on a different cpu
4478 */
4479 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4480 return;
4481
4482 if (n->gro_list) {
4483 unsigned long timeout = 0;
4484
4485 if (work_done)
4486 timeout = n->dev->gro_flush_timeout;
4487
4488 if (timeout)
4489 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4490 HRTIMER_MODE_REL_PINNED);
4491 else
4492 napi_gro_flush(n, false);
4493 }
4494 if (likely(list_empty(&n->poll_list))) {
4495 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4496 } else {
4497 /* If n->poll_list is not empty, we need to mask irqs */
4498 local_irq_save(flags);
4499 __napi_complete(n);
4500 local_irq_restore(flags);
4501 }
4502 }
4503 EXPORT_SYMBOL(napi_complete_done);
4504
4505 /* must be called under rcu_read_lock(), as we dont take a reference */
napi_by_id(unsigned int napi_id)4506 struct napi_struct *napi_by_id(unsigned int napi_id)
4507 {
4508 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4509 struct napi_struct *napi;
4510
4511 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4512 if (napi->napi_id == napi_id)
4513 return napi;
4514
4515 return NULL;
4516 }
4517 EXPORT_SYMBOL_GPL(napi_by_id);
4518
napi_hash_add(struct napi_struct * napi)4519 void napi_hash_add(struct napi_struct *napi)
4520 {
4521 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4522
4523 spin_lock(&napi_hash_lock);
4524
4525 /* 0 is not a valid id, we also skip an id that is taken
4526 * we expect both events to be extremely rare
4527 */
4528 napi->napi_id = 0;
4529 while (!napi->napi_id) {
4530 napi->napi_id = ++napi_gen_id;
4531 if (napi_by_id(napi->napi_id))
4532 napi->napi_id = 0;
4533 }
4534
4535 hlist_add_head_rcu(&napi->napi_hash_node,
4536 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4537
4538 spin_unlock(&napi_hash_lock);
4539 }
4540 }
4541 EXPORT_SYMBOL_GPL(napi_hash_add);
4542
4543 /* Warning : caller is responsible to make sure rcu grace period
4544 * is respected before freeing memory containing @napi
4545 */
napi_hash_del(struct napi_struct * napi)4546 void napi_hash_del(struct napi_struct *napi)
4547 {
4548 spin_lock(&napi_hash_lock);
4549
4550 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4551 hlist_del_rcu(&napi->napi_hash_node);
4552
4553 spin_unlock(&napi_hash_lock);
4554 }
4555 EXPORT_SYMBOL_GPL(napi_hash_del);
4556
napi_watchdog(struct hrtimer * timer)4557 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4558 {
4559 struct napi_struct *napi;
4560
4561 napi = container_of(timer, struct napi_struct, timer);
4562 if (napi->gro_list)
4563 napi_schedule(napi);
4564
4565 return HRTIMER_NORESTART;
4566 }
4567
netif_napi_add(struct net_device * dev,struct napi_struct * napi,int (* poll)(struct napi_struct *,int),int weight)4568 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4569 int (*poll)(struct napi_struct *, int), int weight)
4570 {
4571 INIT_LIST_HEAD(&napi->poll_list);
4572 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4573 napi->timer.function = napi_watchdog;
4574 napi->gro_count = 0;
4575 napi->gro_list = NULL;
4576 napi->skb = NULL;
4577 napi->poll = poll;
4578 if (weight > NAPI_POLL_WEIGHT)
4579 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4580 weight, dev->name);
4581 napi->weight = weight;
4582 list_add(&napi->dev_list, &dev->napi_list);
4583 napi->dev = dev;
4584 #ifdef CONFIG_NETPOLL
4585 spin_lock_init(&napi->poll_lock);
4586 napi->poll_owner = -1;
4587 #endif
4588 set_bit(NAPI_STATE_SCHED, &napi->state);
4589 }
4590 EXPORT_SYMBOL(netif_napi_add);
4591
napi_disable(struct napi_struct * n)4592 void napi_disable(struct napi_struct *n)
4593 {
4594 might_sleep();
4595 set_bit(NAPI_STATE_DISABLE, &n->state);
4596
4597 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4598 msleep(1);
4599
4600 hrtimer_cancel(&n->timer);
4601
4602 clear_bit(NAPI_STATE_DISABLE, &n->state);
4603 }
4604 EXPORT_SYMBOL(napi_disable);
4605
netif_napi_del(struct napi_struct * napi)4606 void netif_napi_del(struct napi_struct *napi)
4607 {
4608 list_del_init(&napi->dev_list);
4609 napi_free_frags(napi);
4610
4611 kfree_skb_list(napi->gro_list);
4612 napi->gro_list = NULL;
4613 napi->gro_count = 0;
4614 }
4615 EXPORT_SYMBOL(netif_napi_del);
4616
napi_poll(struct napi_struct * n,struct list_head * repoll)4617 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4618 {
4619 void *have;
4620 int work, weight;
4621
4622 list_del_init(&n->poll_list);
4623
4624 have = netpoll_poll_lock(n);
4625
4626 weight = n->weight;
4627
4628 /* This NAPI_STATE_SCHED test is for avoiding a race
4629 * with netpoll's poll_napi(). Only the entity which
4630 * obtains the lock and sees NAPI_STATE_SCHED set will
4631 * actually make the ->poll() call. Therefore we avoid
4632 * accidentally calling ->poll() when NAPI is not scheduled.
4633 */
4634 work = 0;
4635 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4636 work = n->poll(n, weight);
4637 trace_napi_poll(n);
4638 }
4639
4640 WARN_ON_ONCE(work > weight);
4641
4642 if (likely(work < weight))
4643 goto out_unlock;
4644
4645 /* Drivers must not modify the NAPI state if they
4646 * consume the entire weight. In such cases this code
4647 * still "owns" the NAPI instance and therefore can
4648 * move the instance around on the list at-will.
4649 */
4650 if (unlikely(napi_disable_pending(n))) {
4651 napi_complete(n);
4652 goto out_unlock;
4653 }
4654
4655 if (n->gro_list) {
4656 /* flush too old packets
4657 * If HZ < 1000, flush all packets.
4658 */
4659 napi_gro_flush(n, HZ >= 1000);
4660 }
4661
4662 /* Some drivers may have called napi_schedule
4663 * prior to exhausting their budget.
4664 */
4665 if (unlikely(!list_empty(&n->poll_list))) {
4666 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4667 n->dev ? n->dev->name : "backlog");
4668 goto out_unlock;
4669 }
4670
4671 list_add_tail(&n->poll_list, repoll);
4672
4673 out_unlock:
4674 netpoll_poll_unlock(have);
4675
4676 return work;
4677 }
4678
net_rx_action(struct softirq_action * h)4679 static void net_rx_action(struct softirq_action *h)
4680 {
4681 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4682 unsigned long time_limit = jiffies + 2;
4683 int budget = netdev_budget;
4684 LIST_HEAD(list);
4685 LIST_HEAD(repoll);
4686
4687 local_irq_disable();
4688 list_splice_init(&sd->poll_list, &list);
4689 local_irq_enable();
4690
4691 for (;;) {
4692 struct napi_struct *n;
4693
4694 if (list_empty(&list)) {
4695 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4696 return;
4697 break;
4698 }
4699
4700 n = list_first_entry(&list, struct napi_struct, poll_list);
4701 budget -= napi_poll(n, &repoll);
4702
4703 /* If softirq window is exhausted then punt.
4704 * Allow this to run for 2 jiffies since which will allow
4705 * an average latency of 1.5/HZ.
4706 */
4707 if (unlikely(budget <= 0 ||
4708 time_after_eq(jiffies, time_limit))) {
4709 sd->time_squeeze++;
4710 break;
4711 }
4712 }
4713
4714 local_irq_disable();
4715
4716 list_splice_tail_init(&sd->poll_list, &list);
4717 list_splice_tail(&repoll, &list);
4718 list_splice(&list, &sd->poll_list);
4719 if (!list_empty(&sd->poll_list))
4720 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4721
4722 net_rps_action_and_irq_enable(sd);
4723 }
4724
4725 struct netdev_adjacent {
4726 struct net_device *dev;
4727
4728 /* upper master flag, there can only be one master device per list */
4729 bool master;
4730
4731 /* counter for the number of times this device was added to us */
4732 u16 ref_nr;
4733
4734 /* private field for the users */
4735 void *private;
4736
4737 struct list_head list;
4738 struct rcu_head rcu;
4739 };
4740
__netdev_find_adj(struct net_device * dev,struct net_device * adj_dev,struct list_head * adj_list)4741 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4742 struct net_device *adj_dev,
4743 struct list_head *adj_list)
4744 {
4745 struct netdev_adjacent *adj;
4746
4747 list_for_each_entry(adj, adj_list, list) {
4748 if (adj->dev == adj_dev)
4749 return adj;
4750 }
4751 return NULL;
4752 }
4753
4754 /**
4755 * netdev_has_upper_dev - Check if device is linked to an upper device
4756 * @dev: device
4757 * @upper_dev: upper device to check
4758 *
4759 * Find out if a device is linked to specified upper device and return true
4760 * in case it is. Note that this checks only immediate upper device,
4761 * not through a complete stack of devices. The caller must hold the RTNL lock.
4762 */
netdev_has_upper_dev(struct net_device * dev,struct net_device * upper_dev)4763 bool netdev_has_upper_dev(struct net_device *dev,
4764 struct net_device *upper_dev)
4765 {
4766 ASSERT_RTNL();
4767
4768 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4769 }
4770 EXPORT_SYMBOL(netdev_has_upper_dev);
4771
4772 /**
4773 * netdev_has_any_upper_dev - Check if device is linked to some device
4774 * @dev: device
4775 *
4776 * Find out if a device is linked to an upper device and return true in case
4777 * it is. The caller must hold the RTNL lock.
4778 */
netdev_has_any_upper_dev(struct net_device * dev)4779 static bool netdev_has_any_upper_dev(struct net_device *dev)
4780 {
4781 ASSERT_RTNL();
4782
4783 return !list_empty(&dev->all_adj_list.upper);
4784 }
4785
4786 /**
4787 * netdev_master_upper_dev_get - Get master upper device
4788 * @dev: device
4789 *
4790 * Find a master upper device and return pointer to it or NULL in case
4791 * it's not there. The caller must hold the RTNL lock.
4792 */
netdev_master_upper_dev_get(struct net_device * dev)4793 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4794 {
4795 struct netdev_adjacent *upper;
4796
4797 ASSERT_RTNL();
4798
4799 if (list_empty(&dev->adj_list.upper))
4800 return NULL;
4801
4802 upper = list_first_entry(&dev->adj_list.upper,
4803 struct netdev_adjacent, list);
4804 if (likely(upper->master))
4805 return upper->dev;
4806 return NULL;
4807 }
4808 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4809
netdev_adjacent_get_private(struct list_head * adj_list)4810 void *netdev_adjacent_get_private(struct list_head *adj_list)
4811 {
4812 struct netdev_adjacent *adj;
4813
4814 adj = list_entry(adj_list, struct netdev_adjacent, list);
4815
4816 return adj->private;
4817 }
4818 EXPORT_SYMBOL(netdev_adjacent_get_private);
4819
4820 /**
4821 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4822 * @dev: device
4823 * @iter: list_head ** of the current position
4824 *
4825 * Gets the next device from the dev's upper list, starting from iter
4826 * position. The caller must hold RCU read lock.
4827 */
netdev_upper_get_next_dev_rcu(struct net_device * dev,struct list_head ** iter)4828 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4829 struct list_head **iter)
4830 {
4831 struct netdev_adjacent *upper;
4832
4833 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4834
4835 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4836
4837 if (&upper->list == &dev->adj_list.upper)
4838 return NULL;
4839
4840 *iter = &upper->list;
4841
4842 return upper->dev;
4843 }
4844 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4845
4846 /**
4847 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4848 * @dev: device
4849 * @iter: list_head ** of the current position
4850 *
4851 * Gets the next device from the dev's upper list, starting from iter
4852 * position. The caller must hold RCU read lock.
4853 */
netdev_all_upper_get_next_dev_rcu(struct net_device * dev,struct list_head ** iter)4854 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4855 struct list_head **iter)
4856 {
4857 struct netdev_adjacent *upper;
4858
4859 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4860
4861 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4862
4863 if (&upper->list == &dev->all_adj_list.upper)
4864 return NULL;
4865
4866 *iter = &upper->list;
4867
4868 return upper->dev;
4869 }
4870 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4871
4872 /**
4873 * netdev_lower_get_next_private - Get the next ->private from the
4874 * lower neighbour list
4875 * @dev: device
4876 * @iter: list_head ** of the current position
4877 *
4878 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4879 * list, starting from iter position. The caller must hold either hold the
4880 * RTNL lock or its own locking that guarantees that the neighbour lower
4881 * list will remain unchainged.
4882 */
netdev_lower_get_next_private(struct net_device * dev,struct list_head ** iter)4883 void *netdev_lower_get_next_private(struct net_device *dev,
4884 struct list_head **iter)
4885 {
4886 struct netdev_adjacent *lower;
4887
4888 lower = list_entry(*iter, struct netdev_adjacent, list);
4889
4890 if (&lower->list == &dev->adj_list.lower)
4891 return NULL;
4892
4893 *iter = lower->list.next;
4894
4895 return lower->private;
4896 }
4897 EXPORT_SYMBOL(netdev_lower_get_next_private);
4898
4899 /**
4900 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4901 * lower neighbour list, RCU
4902 * variant
4903 * @dev: device
4904 * @iter: list_head ** of the current position
4905 *
4906 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4907 * list, starting from iter position. The caller must hold RCU read lock.
4908 */
netdev_lower_get_next_private_rcu(struct net_device * dev,struct list_head ** iter)4909 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4910 struct list_head **iter)
4911 {
4912 struct netdev_adjacent *lower;
4913
4914 WARN_ON_ONCE(!rcu_read_lock_held());
4915
4916 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4917
4918 if (&lower->list == &dev->adj_list.lower)
4919 return NULL;
4920
4921 *iter = &lower->list;
4922
4923 return lower->private;
4924 }
4925 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4926
4927 /**
4928 * netdev_lower_get_next - Get the next device from the lower neighbour
4929 * list
4930 * @dev: device
4931 * @iter: list_head ** of the current position
4932 *
4933 * Gets the next netdev_adjacent from the dev's lower neighbour
4934 * list, starting from iter position. The caller must hold RTNL lock or
4935 * its own locking that guarantees that the neighbour lower
4936 * list will remain unchainged.
4937 */
netdev_lower_get_next(struct net_device * dev,struct list_head ** iter)4938 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4939 {
4940 struct netdev_adjacent *lower;
4941
4942 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4943
4944 if (&lower->list == &dev->adj_list.lower)
4945 return NULL;
4946
4947 *iter = &lower->list;
4948
4949 return lower->dev;
4950 }
4951 EXPORT_SYMBOL(netdev_lower_get_next);
4952
4953 /**
4954 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4955 * lower neighbour list, RCU
4956 * variant
4957 * @dev: device
4958 *
4959 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4960 * list. The caller must hold RCU read lock.
4961 */
netdev_lower_get_first_private_rcu(struct net_device * dev)4962 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4963 {
4964 struct netdev_adjacent *lower;
4965
4966 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4967 struct netdev_adjacent, list);
4968 if (lower)
4969 return lower->private;
4970 return NULL;
4971 }
4972 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4973
4974 /**
4975 * netdev_master_upper_dev_get_rcu - Get master upper device
4976 * @dev: device
4977 *
4978 * Find a master upper device and return pointer to it or NULL in case
4979 * it's not there. The caller must hold the RCU read lock.
4980 */
netdev_master_upper_dev_get_rcu(struct net_device * dev)4981 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4982 {
4983 struct netdev_adjacent *upper;
4984
4985 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4986 struct netdev_adjacent, list);
4987 if (upper && likely(upper->master))
4988 return upper->dev;
4989 return NULL;
4990 }
4991 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4992
netdev_adjacent_sysfs_add(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)4993 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4994 struct net_device *adj_dev,
4995 struct list_head *dev_list)
4996 {
4997 char linkname[IFNAMSIZ+7];
4998 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4999 "upper_%s" : "lower_%s", adj_dev->name);
5000 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5001 linkname);
5002 }
netdev_adjacent_sysfs_del(struct net_device * dev,char * name,struct list_head * dev_list)5003 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5004 char *name,
5005 struct list_head *dev_list)
5006 {
5007 char linkname[IFNAMSIZ+7];
5008 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5009 "upper_%s" : "lower_%s", name);
5010 sysfs_remove_link(&(dev->dev.kobj), linkname);
5011 }
5012
netdev_adjacent_is_neigh_list(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)5013 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5014 struct net_device *adj_dev,
5015 struct list_head *dev_list)
5016 {
5017 return (dev_list == &dev->adj_list.upper ||
5018 dev_list == &dev->adj_list.lower) &&
5019 net_eq(dev_net(dev), dev_net(adj_dev));
5020 }
5021
__netdev_adjacent_dev_insert(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list,void * private,bool master)5022 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5023 struct net_device *adj_dev,
5024 struct list_head *dev_list,
5025 void *private, bool master)
5026 {
5027 struct netdev_adjacent *adj;
5028 int ret;
5029
5030 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5031
5032 if (adj) {
5033 adj->ref_nr++;
5034 return 0;
5035 }
5036
5037 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5038 if (!adj)
5039 return -ENOMEM;
5040
5041 adj->dev = adj_dev;
5042 adj->master = master;
5043 adj->ref_nr = 1;
5044 adj->private = private;
5045 dev_hold(adj_dev);
5046
5047 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5048 adj_dev->name, dev->name, adj_dev->name);
5049
5050 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5051 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5052 if (ret)
5053 goto free_adj;
5054 }
5055
5056 /* Ensure that master link is always the first item in list. */
5057 if (master) {
5058 ret = sysfs_create_link(&(dev->dev.kobj),
5059 &(adj_dev->dev.kobj), "master");
5060 if (ret)
5061 goto remove_symlinks;
5062
5063 list_add_rcu(&adj->list, dev_list);
5064 } else {
5065 list_add_tail_rcu(&adj->list, dev_list);
5066 }
5067
5068 return 0;
5069
5070 remove_symlinks:
5071 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5072 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5073 free_adj:
5074 kfree(adj);
5075 dev_put(adj_dev);
5076
5077 return ret;
5078 }
5079
__netdev_adjacent_dev_remove(struct net_device * dev,struct net_device * adj_dev,struct list_head * dev_list)5080 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5081 struct net_device *adj_dev,
5082 struct list_head *dev_list)
5083 {
5084 struct netdev_adjacent *adj;
5085
5086 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5087
5088 if (!adj) {
5089 pr_err("tried to remove device %s from %s\n",
5090 dev->name, adj_dev->name);
5091 BUG();
5092 }
5093
5094 if (adj->ref_nr > 1) {
5095 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5096 adj->ref_nr-1);
5097 adj->ref_nr--;
5098 return;
5099 }
5100
5101 if (adj->master)
5102 sysfs_remove_link(&(dev->dev.kobj), "master");
5103
5104 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5105 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5106
5107 list_del_rcu(&adj->list);
5108 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5109 adj_dev->name, dev->name, adj_dev->name);
5110 dev_put(adj_dev);
5111 kfree_rcu(adj, rcu);
5112 }
5113
__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)5114 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5115 struct net_device *upper_dev,
5116 struct list_head *up_list,
5117 struct list_head *down_list,
5118 void *private, bool master)
5119 {
5120 int ret;
5121
5122 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5123 master);
5124 if (ret)
5125 return ret;
5126
5127 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5128 false);
5129 if (ret) {
5130 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5131 return ret;
5132 }
5133
5134 return 0;
5135 }
5136
__netdev_adjacent_dev_link(struct net_device * dev,struct net_device * upper_dev)5137 static int __netdev_adjacent_dev_link(struct net_device *dev,
5138 struct net_device *upper_dev)
5139 {
5140 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5141 &dev->all_adj_list.upper,
5142 &upper_dev->all_adj_list.lower,
5143 NULL, false);
5144 }
5145
__netdev_adjacent_dev_unlink_lists(struct net_device * dev,struct net_device * upper_dev,struct list_head * up_list,struct list_head * down_list)5146 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5147 struct net_device *upper_dev,
5148 struct list_head *up_list,
5149 struct list_head *down_list)
5150 {
5151 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5152 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5153 }
5154
__netdev_adjacent_dev_unlink(struct net_device * dev,struct net_device * upper_dev)5155 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5156 struct net_device *upper_dev)
5157 {
5158 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5159 &dev->all_adj_list.upper,
5160 &upper_dev->all_adj_list.lower);
5161 }
5162
__netdev_adjacent_dev_link_neighbour(struct net_device * dev,struct net_device * upper_dev,void * private,bool master)5163 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5164 struct net_device *upper_dev,
5165 void *private, bool master)
5166 {
5167 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5168
5169 if (ret)
5170 return ret;
5171
5172 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5173 &dev->adj_list.upper,
5174 &upper_dev->adj_list.lower,
5175 private, master);
5176 if (ret) {
5177 __netdev_adjacent_dev_unlink(dev, upper_dev);
5178 return ret;
5179 }
5180
5181 return 0;
5182 }
5183
__netdev_adjacent_dev_unlink_neighbour(struct net_device * dev,struct net_device * upper_dev)5184 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5185 struct net_device *upper_dev)
5186 {
5187 __netdev_adjacent_dev_unlink(dev, upper_dev);
5188 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5189 &dev->adj_list.upper,
5190 &upper_dev->adj_list.lower);
5191 }
5192
__netdev_upper_dev_link(struct net_device * dev,struct net_device * upper_dev,bool master,void * private)5193 static int __netdev_upper_dev_link(struct net_device *dev,
5194 struct net_device *upper_dev, bool master,
5195 void *private)
5196 {
5197 struct netdev_adjacent *i, *j, *to_i, *to_j;
5198 int ret = 0;
5199
5200 ASSERT_RTNL();
5201
5202 if (dev == upper_dev)
5203 return -EBUSY;
5204
5205 /* To prevent loops, check if dev is not upper device to upper_dev. */
5206 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5207 return -EBUSY;
5208
5209 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper))
5210 return -EEXIST;
5211
5212 if (master && netdev_master_upper_dev_get(dev))
5213 return -EBUSY;
5214
5215 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5216 master);
5217 if (ret)
5218 return ret;
5219
5220 /* Now that we linked these devs, make all the upper_dev's
5221 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5222 * versa, and don't forget the devices itself. All of these
5223 * links are non-neighbours.
5224 */
5225 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5226 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5227 pr_debug("Interlinking %s with %s, non-neighbour\n",
5228 i->dev->name, j->dev->name);
5229 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5230 if (ret)
5231 goto rollback_mesh;
5232 }
5233 }
5234
5235 /* add dev to every upper_dev's upper device */
5236 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5237 pr_debug("linking %s's upper device %s with %s\n",
5238 upper_dev->name, i->dev->name, dev->name);
5239 ret = __netdev_adjacent_dev_link(dev, i->dev);
5240 if (ret)
5241 goto rollback_upper_mesh;
5242 }
5243
5244 /* add upper_dev to every dev's lower device */
5245 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5246 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5247 i->dev->name, upper_dev->name);
5248 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5249 if (ret)
5250 goto rollback_lower_mesh;
5251 }
5252
5253 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5254 return 0;
5255
5256 rollback_lower_mesh:
5257 to_i = i;
5258 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5259 if (i == to_i)
5260 break;
5261 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5262 }
5263
5264 i = NULL;
5265
5266 rollback_upper_mesh:
5267 to_i = i;
5268 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5269 if (i == to_i)
5270 break;
5271 __netdev_adjacent_dev_unlink(dev, i->dev);
5272 }
5273
5274 i = j = NULL;
5275
5276 rollback_mesh:
5277 to_i = i;
5278 to_j = j;
5279 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5280 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5281 if (i == to_i && j == to_j)
5282 break;
5283 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5284 }
5285 if (i == to_i)
5286 break;
5287 }
5288
5289 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5290
5291 return ret;
5292 }
5293
5294 /**
5295 * netdev_upper_dev_link - Add a link to the upper device
5296 * @dev: device
5297 * @upper_dev: new upper device
5298 *
5299 * Adds a link to device which is upper to this one. The caller must hold
5300 * the RTNL lock. On a failure a negative errno code is returned.
5301 * On success the reference counts are adjusted and the function
5302 * returns zero.
5303 */
netdev_upper_dev_link(struct net_device * dev,struct net_device * upper_dev)5304 int netdev_upper_dev_link(struct net_device *dev,
5305 struct net_device *upper_dev)
5306 {
5307 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5308 }
5309 EXPORT_SYMBOL(netdev_upper_dev_link);
5310
5311 /**
5312 * netdev_master_upper_dev_link - Add a master link to the upper device
5313 * @dev: device
5314 * @upper_dev: new upper device
5315 *
5316 * Adds a link to device which is upper to this one. In this case, only
5317 * one master upper device can be linked, although other non-master devices
5318 * might be linked as well. The caller must hold the RTNL lock.
5319 * On a failure a negative errno code is returned. On success the reference
5320 * counts are adjusted and the function returns zero.
5321 */
netdev_master_upper_dev_link(struct net_device * dev,struct net_device * upper_dev)5322 int netdev_master_upper_dev_link(struct net_device *dev,
5323 struct net_device *upper_dev)
5324 {
5325 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5326 }
5327 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5328
netdev_master_upper_dev_link_private(struct net_device * dev,struct net_device * upper_dev,void * private)5329 int netdev_master_upper_dev_link_private(struct net_device *dev,
5330 struct net_device *upper_dev,
5331 void *private)
5332 {
5333 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5334 }
5335 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5336
5337 /**
5338 * netdev_upper_dev_unlink - Removes a link to upper device
5339 * @dev: device
5340 * @upper_dev: new upper device
5341 *
5342 * Removes a link to device which is upper to this one. The caller must hold
5343 * the RTNL lock.
5344 */
netdev_upper_dev_unlink(struct net_device * dev,struct net_device * upper_dev)5345 void netdev_upper_dev_unlink(struct net_device *dev,
5346 struct net_device *upper_dev)
5347 {
5348 struct netdev_adjacent *i, *j;
5349 ASSERT_RTNL();
5350
5351 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5352
5353 /* Here is the tricky part. We must remove all dev's lower
5354 * devices from all upper_dev's upper devices and vice
5355 * versa, to maintain the graph relationship.
5356 */
5357 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5358 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5359 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5360
5361 /* remove also the devices itself from lower/upper device
5362 * list
5363 */
5364 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5365 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5366
5367 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5368 __netdev_adjacent_dev_unlink(dev, i->dev);
5369
5370 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5371 }
5372 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5373
5374 /**
5375 * netdev_bonding_info_change - Dispatch event about slave change
5376 * @dev: device
5377 * @bonding_info: info to dispatch
5378 *
5379 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5380 * The caller must hold the RTNL lock.
5381 */
netdev_bonding_info_change(struct net_device * dev,struct netdev_bonding_info * bonding_info)5382 void netdev_bonding_info_change(struct net_device *dev,
5383 struct netdev_bonding_info *bonding_info)
5384 {
5385 struct netdev_notifier_bonding_info info;
5386
5387 memcpy(&info.bonding_info, bonding_info,
5388 sizeof(struct netdev_bonding_info));
5389 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5390 &info.info);
5391 }
5392 EXPORT_SYMBOL(netdev_bonding_info_change);
5393
netdev_adjacent_add_links(struct net_device * dev)5394 static void netdev_adjacent_add_links(struct net_device *dev)
5395 {
5396 struct netdev_adjacent *iter;
5397
5398 struct net *net = dev_net(dev);
5399
5400 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5401 if (!net_eq(net,dev_net(iter->dev)))
5402 continue;
5403 netdev_adjacent_sysfs_add(iter->dev, dev,
5404 &iter->dev->adj_list.lower);
5405 netdev_adjacent_sysfs_add(dev, iter->dev,
5406 &dev->adj_list.upper);
5407 }
5408
5409 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5410 if (!net_eq(net,dev_net(iter->dev)))
5411 continue;
5412 netdev_adjacent_sysfs_add(iter->dev, dev,
5413 &iter->dev->adj_list.upper);
5414 netdev_adjacent_sysfs_add(dev, iter->dev,
5415 &dev->adj_list.lower);
5416 }
5417 }
5418
netdev_adjacent_del_links(struct net_device * dev)5419 static void netdev_adjacent_del_links(struct net_device *dev)
5420 {
5421 struct netdev_adjacent *iter;
5422
5423 struct net *net = dev_net(dev);
5424
5425 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5426 if (!net_eq(net,dev_net(iter->dev)))
5427 continue;
5428 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5429 &iter->dev->adj_list.lower);
5430 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5431 &dev->adj_list.upper);
5432 }
5433
5434 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5435 if (!net_eq(net,dev_net(iter->dev)))
5436 continue;
5437 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5438 &iter->dev->adj_list.upper);
5439 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5440 &dev->adj_list.lower);
5441 }
5442 }
5443
netdev_adjacent_rename_links(struct net_device * dev,char * oldname)5444 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5445 {
5446 struct netdev_adjacent *iter;
5447
5448 struct net *net = dev_net(dev);
5449
5450 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5451 if (!net_eq(net,dev_net(iter->dev)))
5452 continue;
5453 netdev_adjacent_sysfs_del(iter->dev, oldname,
5454 &iter->dev->adj_list.lower);
5455 netdev_adjacent_sysfs_add(iter->dev, dev,
5456 &iter->dev->adj_list.lower);
5457 }
5458
5459 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5460 if (!net_eq(net,dev_net(iter->dev)))
5461 continue;
5462 netdev_adjacent_sysfs_del(iter->dev, oldname,
5463 &iter->dev->adj_list.upper);
5464 netdev_adjacent_sysfs_add(iter->dev, dev,
5465 &iter->dev->adj_list.upper);
5466 }
5467 }
5468
netdev_lower_dev_get_private(struct net_device * dev,struct net_device * lower_dev)5469 void *netdev_lower_dev_get_private(struct net_device *dev,
5470 struct net_device *lower_dev)
5471 {
5472 struct netdev_adjacent *lower;
5473
5474 if (!lower_dev)
5475 return NULL;
5476 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5477 if (!lower)
5478 return NULL;
5479
5480 return lower->private;
5481 }
5482 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5483
5484
dev_get_nest_level(struct net_device * dev,bool (* type_check)(struct net_device * dev))5485 int dev_get_nest_level(struct net_device *dev,
5486 bool (*type_check)(struct net_device *dev))
5487 {
5488 struct net_device *lower = NULL;
5489 struct list_head *iter;
5490 int max_nest = -1;
5491 int nest;
5492
5493 ASSERT_RTNL();
5494
5495 netdev_for_each_lower_dev(dev, lower, iter) {
5496 nest = dev_get_nest_level(lower, type_check);
5497 if (max_nest < nest)
5498 max_nest = nest;
5499 }
5500
5501 if (type_check(dev))
5502 max_nest++;
5503
5504 return max_nest;
5505 }
5506 EXPORT_SYMBOL(dev_get_nest_level);
5507
dev_change_rx_flags(struct net_device * dev,int flags)5508 static void dev_change_rx_flags(struct net_device *dev, int flags)
5509 {
5510 const struct net_device_ops *ops = dev->netdev_ops;
5511
5512 if (ops->ndo_change_rx_flags)
5513 ops->ndo_change_rx_flags(dev, flags);
5514 }
5515
__dev_set_promiscuity(struct net_device * dev,int inc,bool notify)5516 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5517 {
5518 unsigned int old_flags = dev->flags;
5519 kuid_t uid;
5520 kgid_t gid;
5521
5522 ASSERT_RTNL();
5523
5524 dev->flags |= IFF_PROMISC;
5525 dev->promiscuity += inc;
5526 if (dev->promiscuity == 0) {
5527 /*
5528 * Avoid overflow.
5529 * If inc causes overflow, untouch promisc and return error.
5530 */
5531 if (inc < 0)
5532 dev->flags &= ~IFF_PROMISC;
5533 else {
5534 dev->promiscuity -= inc;
5535 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5536 dev->name);
5537 return -EOVERFLOW;
5538 }
5539 }
5540 if (dev->flags != old_flags) {
5541 pr_info("device %s %s promiscuous mode\n",
5542 dev->name,
5543 dev->flags & IFF_PROMISC ? "entered" : "left");
5544 if (audit_enabled) {
5545 current_uid_gid(&uid, &gid);
5546 audit_log(current->audit_context, GFP_ATOMIC,
5547 AUDIT_ANOM_PROMISCUOUS,
5548 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5549 dev->name, (dev->flags & IFF_PROMISC),
5550 (old_flags & IFF_PROMISC),
5551 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5552 from_kuid(&init_user_ns, uid),
5553 from_kgid(&init_user_ns, gid),
5554 audit_get_sessionid(current));
5555 }
5556
5557 dev_change_rx_flags(dev, IFF_PROMISC);
5558 }
5559 if (notify)
5560 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5561 return 0;
5562 }
5563
5564 /**
5565 * dev_set_promiscuity - update promiscuity count on a device
5566 * @dev: device
5567 * @inc: modifier
5568 *
5569 * Add or remove promiscuity from a device. While the count in the device
5570 * remains above zero the interface remains promiscuous. Once it hits zero
5571 * the device reverts back to normal filtering operation. A negative inc
5572 * value is used to drop promiscuity on the device.
5573 * Return 0 if successful or a negative errno code on error.
5574 */
dev_set_promiscuity(struct net_device * dev,int inc)5575 int dev_set_promiscuity(struct net_device *dev, int inc)
5576 {
5577 unsigned int old_flags = dev->flags;
5578 int err;
5579
5580 err = __dev_set_promiscuity(dev, inc, true);
5581 if (err < 0)
5582 return err;
5583 if (dev->flags != old_flags)
5584 dev_set_rx_mode(dev);
5585 return err;
5586 }
5587 EXPORT_SYMBOL(dev_set_promiscuity);
5588
__dev_set_allmulti(struct net_device * dev,int inc,bool notify)5589 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5590 {
5591 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5592
5593 ASSERT_RTNL();
5594
5595 dev->flags |= IFF_ALLMULTI;
5596 dev->allmulti += inc;
5597 if (dev->allmulti == 0) {
5598 /*
5599 * Avoid overflow.
5600 * If inc causes overflow, untouch allmulti and return error.
5601 */
5602 if (inc < 0)
5603 dev->flags &= ~IFF_ALLMULTI;
5604 else {
5605 dev->allmulti -= inc;
5606 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5607 dev->name);
5608 return -EOVERFLOW;
5609 }
5610 }
5611 if (dev->flags ^ old_flags) {
5612 dev_change_rx_flags(dev, IFF_ALLMULTI);
5613 dev_set_rx_mode(dev);
5614 if (notify)
5615 __dev_notify_flags(dev, old_flags,
5616 dev->gflags ^ old_gflags);
5617 }
5618 return 0;
5619 }
5620
5621 /**
5622 * dev_set_allmulti - update allmulti count on a device
5623 * @dev: device
5624 * @inc: modifier
5625 *
5626 * Add or remove reception of all multicast frames to a device. While the
5627 * count in the device remains above zero the interface remains listening
5628 * to all interfaces. Once it hits zero the device reverts back to normal
5629 * filtering operation. A negative @inc value is used to drop the counter
5630 * when releasing a resource needing all multicasts.
5631 * Return 0 if successful or a negative errno code on error.
5632 */
5633
dev_set_allmulti(struct net_device * dev,int inc)5634 int dev_set_allmulti(struct net_device *dev, int inc)
5635 {
5636 return __dev_set_allmulti(dev, inc, true);
5637 }
5638 EXPORT_SYMBOL(dev_set_allmulti);
5639
5640 /*
5641 * Upload unicast and multicast address lists to device and
5642 * configure RX filtering. When the device doesn't support unicast
5643 * filtering it is put in promiscuous mode while unicast addresses
5644 * are present.
5645 */
__dev_set_rx_mode(struct net_device * dev)5646 void __dev_set_rx_mode(struct net_device *dev)
5647 {
5648 const struct net_device_ops *ops = dev->netdev_ops;
5649
5650 /* dev_open will call this function so the list will stay sane. */
5651 if (!(dev->flags&IFF_UP))
5652 return;
5653
5654 if (!netif_device_present(dev))
5655 return;
5656
5657 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5658 /* Unicast addresses changes may only happen under the rtnl,
5659 * therefore calling __dev_set_promiscuity here is safe.
5660 */
5661 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5662 __dev_set_promiscuity(dev, 1, false);
5663 dev->uc_promisc = true;
5664 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5665 __dev_set_promiscuity(dev, -1, false);
5666 dev->uc_promisc = false;
5667 }
5668 }
5669
5670 if (ops->ndo_set_rx_mode)
5671 ops->ndo_set_rx_mode(dev);
5672 }
5673
dev_set_rx_mode(struct net_device * dev)5674 void dev_set_rx_mode(struct net_device *dev)
5675 {
5676 netif_addr_lock_bh(dev);
5677 __dev_set_rx_mode(dev);
5678 netif_addr_unlock_bh(dev);
5679 }
5680
5681 /**
5682 * dev_get_flags - get flags reported to userspace
5683 * @dev: device
5684 *
5685 * Get the combination of flag bits exported through APIs to userspace.
5686 */
dev_get_flags(const struct net_device * dev)5687 unsigned int dev_get_flags(const struct net_device *dev)
5688 {
5689 unsigned int flags;
5690
5691 flags = (dev->flags & ~(IFF_PROMISC |
5692 IFF_ALLMULTI |
5693 IFF_RUNNING |
5694 IFF_LOWER_UP |
5695 IFF_DORMANT)) |
5696 (dev->gflags & (IFF_PROMISC |
5697 IFF_ALLMULTI));
5698
5699 if (netif_running(dev)) {
5700 if (netif_oper_up(dev))
5701 flags |= IFF_RUNNING;
5702 if (netif_carrier_ok(dev))
5703 flags |= IFF_LOWER_UP;
5704 if (netif_dormant(dev))
5705 flags |= IFF_DORMANT;
5706 }
5707
5708 return flags;
5709 }
5710 EXPORT_SYMBOL(dev_get_flags);
5711
__dev_change_flags(struct net_device * dev,unsigned int flags)5712 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5713 {
5714 unsigned int old_flags = dev->flags;
5715 int ret;
5716
5717 ASSERT_RTNL();
5718
5719 /*
5720 * Set the flags on our device.
5721 */
5722
5723 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5724 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5725 IFF_AUTOMEDIA)) |
5726 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5727 IFF_ALLMULTI));
5728
5729 /*
5730 * Load in the correct multicast list now the flags have changed.
5731 */
5732
5733 if ((old_flags ^ flags) & IFF_MULTICAST)
5734 dev_change_rx_flags(dev, IFF_MULTICAST);
5735
5736 dev_set_rx_mode(dev);
5737
5738 /*
5739 * Have we downed the interface. We handle IFF_UP ourselves
5740 * according to user attempts to set it, rather than blindly
5741 * setting it.
5742 */
5743
5744 ret = 0;
5745 if ((old_flags ^ flags) & IFF_UP)
5746 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5747
5748 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5749 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5750 unsigned int old_flags = dev->flags;
5751
5752 dev->gflags ^= IFF_PROMISC;
5753
5754 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5755 if (dev->flags != old_flags)
5756 dev_set_rx_mode(dev);
5757 }
5758
5759 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5760 is important. Some (broken) drivers set IFF_PROMISC, when
5761 IFF_ALLMULTI is requested not asking us and not reporting.
5762 */
5763 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5764 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5765
5766 dev->gflags ^= IFF_ALLMULTI;
5767 __dev_set_allmulti(dev, inc, false);
5768 }
5769
5770 return ret;
5771 }
5772
__dev_notify_flags(struct net_device * dev,unsigned int old_flags,unsigned int gchanges)5773 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5774 unsigned int gchanges)
5775 {
5776 unsigned int changes = dev->flags ^ old_flags;
5777
5778 if (gchanges)
5779 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5780
5781 if (changes & IFF_UP) {
5782 if (dev->flags & IFF_UP)
5783 call_netdevice_notifiers(NETDEV_UP, dev);
5784 else
5785 call_netdevice_notifiers(NETDEV_DOWN, dev);
5786 }
5787
5788 if (dev->flags & IFF_UP &&
5789 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5790 struct netdev_notifier_change_info change_info;
5791
5792 change_info.flags_changed = changes;
5793 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5794 &change_info.info);
5795 }
5796 }
5797
5798 /**
5799 * dev_change_flags - change device settings
5800 * @dev: device
5801 * @flags: device state flags
5802 *
5803 * Change settings on device based state flags. The flags are
5804 * in the userspace exported format.
5805 */
dev_change_flags(struct net_device * dev,unsigned int flags)5806 int dev_change_flags(struct net_device *dev, unsigned int flags)
5807 {
5808 int ret;
5809 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5810
5811 ret = __dev_change_flags(dev, flags);
5812 if (ret < 0)
5813 return ret;
5814
5815 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5816 __dev_notify_flags(dev, old_flags, changes);
5817 return ret;
5818 }
5819 EXPORT_SYMBOL(dev_change_flags);
5820
__dev_set_mtu(struct net_device * dev,int new_mtu)5821 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5822 {
5823 const struct net_device_ops *ops = dev->netdev_ops;
5824
5825 if (ops->ndo_change_mtu)
5826 return ops->ndo_change_mtu(dev, new_mtu);
5827
5828 dev->mtu = new_mtu;
5829 return 0;
5830 }
5831
5832 /**
5833 * dev_set_mtu - Change maximum transfer unit
5834 * @dev: device
5835 * @new_mtu: new transfer unit
5836 *
5837 * Change the maximum transfer size of the network device.
5838 */
dev_set_mtu(struct net_device * dev,int new_mtu)5839 int dev_set_mtu(struct net_device *dev, int new_mtu)
5840 {
5841 int err, orig_mtu;
5842
5843 if (new_mtu == dev->mtu)
5844 return 0;
5845
5846 /* MTU must be positive. */
5847 if (new_mtu < 0)
5848 return -EINVAL;
5849
5850 if (!netif_device_present(dev))
5851 return -ENODEV;
5852
5853 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5854 err = notifier_to_errno(err);
5855 if (err)
5856 return err;
5857
5858 orig_mtu = dev->mtu;
5859 err = __dev_set_mtu(dev, new_mtu);
5860
5861 if (!err) {
5862 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5863 err = notifier_to_errno(err);
5864 if (err) {
5865 /* setting mtu back and notifying everyone again,
5866 * so that they have a chance to revert changes.
5867 */
5868 __dev_set_mtu(dev, orig_mtu);
5869 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5870 }
5871 }
5872 return err;
5873 }
5874 EXPORT_SYMBOL(dev_set_mtu);
5875
5876 /**
5877 * dev_set_group - Change group this device belongs to
5878 * @dev: device
5879 * @new_group: group this device should belong to
5880 */
dev_set_group(struct net_device * dev,int new_group)5881 void dev_set_group(struct net_device *dev, int new_group)
5882 {
5883 dev->group = new_group;
5884 }
5885 EXPORT_SYMBOL(dev_set_group);
5886
5887 /**
5888 * dev_set_mac_address - Change Media Access Control Address
5889 * @dev: device
5890 * @sa: new address
5891 *
5892 * Change the hardware (MAC) address of the device
5893 */
dev_set_mac_address(struct net_device * dev,struct sockaddr * sa)5894 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5895 {
5896 const struct net_device_ops *ops = dev->netdev_ops;
5897 int err;
5898
5899 if (!ops->ndo_set_mac_address)
5900 return -EOPNOTSUPP;
5901 if (sa->sa_family != dev->type)
5902 return -EINVAL;
5903 if (!netif_device_present(dev))
5904 return -ENODEV;
5905 err = ops->ndo_set_mac_address(dev, sa);
5906 if (err)
5907 return err;
5908 dev->addr_assign_type = NET_ADDR_SET;
5909 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5910 add_device_randomness(dev->dev_addr, dev->addr_len);
5911 return 0;
5912 }
5913 EXPORT_SYMBOL(dev_set_mac_address);
5914
5915 /**
5916 * dev_change_carrier - Change device carrier
5917 * @dev: device
5918 * @new_carrier: new value
5919 *
5920 * Change device carrier
5921 */
dev_change_carrier(struct net_device * dev,bool new_carrier)5922 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5923 {
5924 const struct net_device_ops *ops = dev->netdev_ops;
5925
5926 if (!ops->ndo_change_carrier)
5927 return -EOPNOTSUPP;
5928 if (!netif_device_present(dev))
5929 return -ENODEV;
5930 return ops->ndo_change_carrier(dev, new_carrier);
5931 }
5932 EXPORT_SYMBOL(dev_change_carrier);
5933
5934 /**
5935 * dev_get_phys_port_id - Get device physical port ID
5936 * @dev: device
5937 * @ppid: port ID
5938 *
5939 * Get device physical port ID
5940 */
dev_get_phys_port_id(struct net_device * dev,struct netdev_phys_item_id * ppid)5941 int dev_get_phys_port_id(struct net_device *dev,
5942 struct netdev_phys_item_id *ppid)
5943 {
5944 const struct net_device_ops *ops = dev->netdev_ops;
5945
5946 if (!ops->ndo_get_phys_port_id)
5947 return -EOPNOTSUPP;
5948 return ops->ndo_get_phys_port_id(dev, ppid);
5949 }
5950 EXPORT_SYMBOL(dev_get_phys_port_id);
5951
5952 /**
5953 * dev_get_phys_port_name - Get device physical port name
5954 * @dev: device
5955 * @name: port name
5956 *
5957 * Get device physical port name
5958 */
dev_get_phys_port_name(struct net_device * dev,char * name,size_t len)5959 int dev_get_phys_port_name(struct net_device *dev,
5960 char *name, size_t len)
5961 {
5962 const struct net_device_ops *ops = dev->netdev_ops;
5963
5964 if (!ops->ndo_get_phys_port_name)
5965 return -EOPNOTSUPP;
5966 return ops->ndo_get_phys_port_name(dev, name, len);
5967 }
5968 EXPORT_SYMBOL(dev_get_phys_port_name);
5969
5970 /**
5971 * dev_new_index - allocate an ifindex
5972 * @net: the applicable net namespace
5973 *
5974 * Returns a suitable unique value for a new device interface
5975 * number. The caller must hold the rtnl semaphore or the
5976 * dev_base_lock to be sure it remains unique.
5977 */
dev_new_index(struct net * net)5978 static int dev_new_index(struct net *net)
5979 {
5980 int ifindex = net->ifindex;
5981 for (;;) {
5982 if (++ifindex <= 0)
5983 ifindex = 1;
5984 if (!__dev_get_by_index(net, ifindex))
5985 return net->ifindex = ifindex;
5986 }
5987 }
5988
5989 /* Delayed registration/unregisteration */
5990 static LIST_HEAD(net_todo_list);
5991 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5992
net_set_todo(struct net_device * dev)5993 static void net_set_todo(struct net_device *dev)
5994 {
5995 list_add_tail(&dev->todo_list, &net_todo_list);
5996 dev_net(dev)->dev_unreg_count++;
5997 }
5998
rollback_registered_many(struct list_head * head)5999 static void rollback_registered_many(struct list_head *head)
6000 {
6001 struct net_device *dev, *tmp;
6002 LIST_HEAD(close_head);
6003
6004 BUG_ON(dev_boot_phase);
6005 ASSERT_RTNL();
6006
6007 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6008 /* Some devices call without registering
6009 * for initialization unwind. Remove those
6010 * devices and proceed with the remaining.
6011 */
6012 if (dev->reg_state == NETREG_UNINITIALIZED) {
6013 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6014 dev->name, dev);
6015
6016 WARN_ON(1);
6017 list_del(&dev->unreg_list);
6018 continue;
6019 }
6020 dev->dismantle = true;
6021 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6022 }
6023
6024 /* If device is running, close it first. */
6025 list_for_each_entry(dev, head, unreg_list)
6026 list_add_tail(&dev->close_list, &close_head);
6027 dev_close_many(&close_head, true);
6028
6029 list_for_each_entry(dev, head, unreg_list) {
6030 /* And unlink it from device chain. */
6031 unlist_netdevice(dev);
6032
6033 dev->reg_state = NETREG_UNREGISTERING;
6034 on_each_cpu(flush_backlog, dev, 1);
6035 }
6036
6037 synchronize_net();
6038
6039 list_for_each_entry(dev, head, unreg_list) {
6040 struct sk_buff *skb = NULL;
6041
6042 /* Shutdown queueing discipline. */
6043 dev_shutdown(dev);
6044
6045
6046 /* Notify protocols, that we are about to destroy
6047 this device. They should clean all the things.
6048 */
6049 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6050
6051 if (!dev->rtnl_link_ops ||
6052 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6053 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6054 GFP_KERNEL);
6055
6056 /*
6057 * Flush the unicast and multicast chains
6058 */
6059 dev_uc_flush(dev);
6060 dev_mc_flush(dev);
6061
6062 if (dev->netdev_ops->ndo_uninit)
6063 dev->netdev_ops->ndo_uninit(dev);
6064
6065 if (skb)
6066 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6067
6068 /* Notifier chain MUST detach us all upper devices. */
6069 WARN_ON(netdev_has_any_upper_dev(dev));
6070
6071 /* Remove entries from kobject tree */
6072 netdev_unregister_kobject(dev);
6073 #ifdef CONFIG_XPS
6074 /* Remove XPS queueing entries */
6075 netif_reset_xps_queues_gt(dev, 0);
6076 #endif
6077 }
6078
6079 synchronize_net();
6080
6081 list_for_each_entry(dev, head, unreg_list)
6082 dev_put(dev);
6083 }
6084
rollback_registered(struct net_device * dev)6085 static void rollback_registered(struct net_device *dev)
6086 {
6087 LIST_HEAD(single);
6088
6089 list_add(&dev->unreg_list, &single);
6090 rollback_registered_many(&single);
6091 list_del(&single);
6092 }
6093
netdev_fix_features(struct net_device * dev,netdev_features_t features)6094 static netdev_features_t netdev_fix_features(struct net_device *dev,
6095 netdev_features_t features)
6096 {
6097 /* Fix illegal checksum combinations */
6098 if ((features & NETIF_F_HW_CSUM) &&
6099 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6100 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6101 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6102 }
6103
6104 /* TSO requires that SG is present as well. */
6105 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6106 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6107 features &= ~NETIF_F_ALL_TSO;
6108 }
6109
6110 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6111 !(features & NETIF_F_IP_CSUM)) {
6112 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6113 features &= ~NETIF_F_TSO;
6114 features &= ~NETIF_F_TSO_ECN;
6115 }
6116
6117 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6118 !(features & NETIF_F_IPV6_CSUM)) {
6119 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6120 features &= ~NETIF_F_TSO6;
6121 }
6122
6123 /* TSO ECN requires that TSO is present as well. */
6124 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6125 features &= ~NETIF_F_TSO_ECN;
6126
6127 /* Software GSO depends on SG. */
6128 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6129 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6130 features &= ~NETIF_F_GSO;
6131 }
6132
6133 /* UFO needs SG and checksumming */
6134 if (features & NETIF_F_UFO) {
6135 /* maybe split UFO into V4 and V6? */
6136 if (!((features & NETIF_F_GEN_CSUM) ||
6137 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6138 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6139 netdev_dbg(dev,
6140 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6141 features &= ~NETIF_F_UFO;
6142 }
6143
6144 if (!(features & NETIF_F_SG)) {
6145 netdev_dbg(dev,
6146 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6147 features &= ~NETIF_F_UFO;
6148 }
6149 }
6150
6151 #ifdef CONFIG_NET_RX_BUSY_POLL
6152 if (dev->netdev_ops->ndo_busy_poll)
6153 features |= NETIF_F_BUSY_POLL;
6154 else
6155 #endif
6156 features &= ~NETIF_F_BUSY_POLL;
6157
6158 return features;
6159 }
6160
__netdev_update_features(struct net_device * dev)6161 int __netdev_update_features(struct net_device *dev)
6162 {
6163 netdev_features_t features;
6164 int err = 0;
6165
6166 ASSERT_RTNL();
6167
6168 features = netdev_get_wanted_features(dev);
6169
6170 if (dev->netdev_ops->ndo_fix_features)
6171 features = dev->netdev_ops->ndo_fix_features(dev, features);
6172
6173 /* driver might be less strict about feature dependencies */
6174 features = netdev_fix_features(dev, features);
6175
6176 if (dev->features == features)
6177 return 0;
6178
6179 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6180 &dev->features, &features);
6181
6182 if (dev->netdev_ops->ndo_set_features)
6183 err = dev->netdev_ops->ndo_set_features(dev, features);
6184
6185 if (unlikely(err < 0)) {
6186 netdev_err(dev,
6187 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6188 err, &features, &dev->features);
6189 return -1;
6190 }
6191
6192 if (!err)
6193 dev->features = features;
6194
6195 return 1;
6196 }
6197
6198 /**
6199 * netdev_update_features - recalculate device features
6200 * @dev: the device to check
6201 *
6202 * Recalculate dev->features set and send notifications if it
6203 * has changed. Should be called after driver or hardware dependent
6204 * conditions might have changed that influence the features.
6205 */
netdev_update_features(struct net_device * dev)6206 void netdev_update_features(struct net_device *dev)
6207 {
6208 if (__netdev_update_features(dev))
6209 netdev_features_change(dev);
6210 }
6211 EXPORT_SYMBOL(netdev_update_features);
6212
6213 /**
6214 * netdev_change_features - recalculate device features
6215 * @dev: the device to check
6216 *
6217 * Recalculate dev->features set and send notifications even
6218 * if they have not changed. Should be called instead of
6219 * netdev_update_features() if also dev->vlan_features might
6220 * have changed to allow the changes to be propagated to stacked
6221 * VLAN devices.
6222 */
netdev_change_features(struct net_device * dev)6223 void netdev_change_features(struct net_device *dev)
6224 {
6225 __netdev_update_features(dev);
6226 netdev_features_change(dev);
6227 }
6228 EXPORT_SYMBOL(netdev_change_features);
6229
6230 /**
6231 * netif_stacked_transfer_operstate - transfer operstate
6232 * @rootdev: the root or lower level device to transfer state from
6233 * @dev: the device to transfer operstate to
6234 *
6235 * Transfer operational state from root to device. This is normally
6236 * called when a stacking relationship exists between the root
6237 * device and the device(a leaf device).
6238 */
netif_stacked_transfer_operstate(const struct net_device * rootdev,struct net_device * dev)6239 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6240 struct net_device *dev)
6241 {
6242 if (rootdev->operstate == IF_OPER_DORMANT)
6243 netif_dormant_on(dev);
6244 else
6245 netif_dormant_off(dev);
6246
6247 if (netif_carrier_ok(rootdev)) {
6248 if (!netif_carrier_ok(dev))
6249 netif_carrier_on(dev);
6250 } else {
6251 if (netif_carrier_ok(dev))
6252 netif_carrier_off(dev);
6253 }
6254 }
6255 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6256
6257 #ifdef CONFIG_SYSFS
netif_alloc_rx_queues(struct net_device * dev)6258 static int netif_alloc_rx_queues(struct net_device *dev)
6259 {
6260 unsigned int i, count = dev->num_rx_queues;
6261 struct netdev_rx_queue *rx;
6262 size_t sz = count * sizeof(*rx);
6263
6264 BUG_ON(count < 1);
6265
6266 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6267 if (!rx) {
6268 rx = vzalloc(sz);
6269 if (!rx)
6270 return -ENOMEM;
6271 }
6272 dev->_rx = rx;
6273
6274 for (i = 0; i < count; i++)
6275 rx[i].dev = dev;
6276 return 0;
6277 }
6278 #endif
6279
netdev_init_one_queue(struct net_device * dev,struct netdev_queue * queue,void * _unused)6280 static void netdev_init_one_queue(struct net_device *dev,
6281 struct netdev_queue *queue, void *_unused)
6282 {
6283 /* Initialize queue lock */
6284 spin_lock_init(&queue->_xmit_lock);
6285 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6286 queue->xmit_lock_owner = -1;
6287 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6288 queue->dev = dev;
6289 #ifdef CONFIG_BQL
6290 dql_init(&queue->dql, HZ);
6291 #endif
6292 }
6293
netif_free_tx_queues(struct net_device * dev)6294 static void netif_free_tx_queues(struct net_device *dev)
6295 {
6296 kvfree(dev->_tx);
6297 }
6298
netif_alloc_netdev_queues(struct net_device * dev)6299 static int netif_alloc_netdev_queues(struct net_device *dev)
6300 {
6301 unsigned int count = dev->num_tx_queues;
6302 struct netdev_queue *tx;
6303 size_t sz = count * sizeof(*tx);
6304
6305 if (count < 1 || count > 0xffff)
6306 return -EINVAL;
6307
6308 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6309 if (!tx) {
6310 tx = vzalloc(sz);
6311 if (!tx)
6312 return -ENOMEM;
6313 }
6314 dev->_tx = tx;
6315
6316 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6317 spin_lock_init(&dev->tx_global_lock);
6318
6319 return 0;
6320 }
6321
6322 /**
6323 * register_netdevice - register a network device
6324 * @dev: device to register
6325 *
6326 * Take a completed network device structure and add it to the kernel
6327 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6328 * chain. 0 is returned on success. A negative errno code is returned
6329 * on a failure to set up the device, or if the name is a duplicate.
6330 *
6331 * Callers must hold the rtnl semaphore. You may want
6332 * register_netdev() instead of this.
6333 *
6334 * BUGS:
6335 * The locking appears insufficient to guarantee two parallel registers
6336 * will not get the same name.
6337 */
6338
register_netdevice(struct net_device * dev)6339 int register_netdevice(struct net_device *dev)
6340 {
6341 int ret;
6342 struct net *net = dev_net(dev);
6343
6344 BUG_ON(dev_boot_phase);
6345 ASSERT_RTNL();
6346
6347 might_sleep();
6348
6349 /* When net_device's are persistent, this will be fatal. */
6350 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6351 BUG_ON(!net);
6352
6353 spin_lock_init(&dev->addr_list_lock);
6354 netdev_set_addr_lockdep_class(dev);
6355
6356 ret = dev_get_valid_name(net, dev, dev->name);
6357 if (ret < 0)
6358 goto out;
6359
6360 /* Init, if this function is available */
6361 if (dev->netdev_ops->ndo_init) {
6362 ret = dev->netdev_ops->ndo_init(dev);
6363 if (ret) {
6364 if (ret > 0)
6365 ret = -EIO;
6366 goto out;
6367 }
6368 }
6369
6370 if (((dev->hw_features | dev->features) &
6371 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6372 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6373 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6374 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6375 ret = -EINVAL;
6376 goto err_uninit;
6377 }
6378
6379 ret = -EBUSY;
6380 if (!dev->ifindex)
6381 dev->ifindex = dev_new_index(net);
6382 else if (__dev_get_by_index(net, dev->ifindex))
6383 goto err_uninit;
6384
6385 /* Transfer changeable features to wanted_features and enable
6386 * software offloads (GSO and GRO).
6387 */
6388 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6389 dev->features |= NETIF_F_SOFT_FEATURES;
6390 dev->wanted_features = dev->features & dev->hw_features;
6391
6392 if (!(dev->flags & IFF_LOOPBACK)) {
6393 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6394 }
6395
6396 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6397 */
6398 dev->vlan_features |= NETIF_F_HIGHDMA;
6399
6400 /* Make NETIF_F_SG inheritable to tunnel devices.
6401 */
6402 dev->hw_enc_features |= NETIF_F_SG;
6403
6404 /* Make NETIF_F_SG inheritable to MPLS.
6405 */
6406 dev->mpls_features |= NETIF_F_SG;
6407
6408 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6409 ret = notifier_to_errno(ret);
6410 if (ret)
6411 goto err_uninit;
6412
6413 ret = netdev_register_kobject(dev);
6414 if (ret)
6415 goto err_uninit;
6416 dev->reg_state = NETREG_REGISTERED;
6417
6418 __netdev_update_features(dev);
6419
6420 /*
6421 * Default initial state at registry is that the
6422 * device is present.
6423 */
6424
6425 set_bit(__LINK_STATE_PRESENT, &dev->state);
6426
6427 linkwatch_init_dev(dev);
6428
6429 dev_init_scheduler(dev);
6430 dev_hold(dev);
6431 list_netdevice(dev);
6432 add_device_randomness(dev->dev_addr, dev->addr_len);
6433
6434 /* If the device has permanent device address, driver should
6435 * set dev_addr and also addr_assign_type should be set to
6436 * NET_ADDR_PERM (default value).
6437 */
6438 if (dev->addr_assign_type == NET_ADDR_PERM)
6439 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6440
6441 /* Notify protocols, that a new device appeared. */
6442 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6443 ret = notifier_to_errno(ret);
6444 if (ret) {
6445 rollback_registered(dev);
6446 dev->reg_state = NETREG_UNREGISTERED;
6447 }
6448 /*
6449 * Prevent userspace races by waiting until the network
6450 * device is fully setup before sending notifications.
6451 */
6452 if (!dev->rtnl_link_ops ||
6453 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6454 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6455
6456 out:
6457 return ret;
6458
6459 err_uninit:
6460 if (dev->netdev_ops->ndo_uninit)
6461 dev->netdev_ops->ndo_uninit(dev);
6462 goto out;
6463 }
6464 EXPORT_SYMBOL(register_netdevice);
6465
6466 /**
6467 * init_dummy_netdev - init a dummy network device for NAPI
6468 * @dev: device to init
6469 *
6470 * This takes a network device structure and initialize the minimum
6471 * amount of fields so it can be used to schedule NAPI polls without
6472 * registering a full blown interface. This is to be used by drivers
6473 * that need to tie several hardware interfaces to a single NAPI
6474 * poll scheduler due to HW limitations.
6475 */
init_dummy_netdev(struct net_device * dev)6476 int init_dummy_netdev(struct net_device *dev)
6477 {
6478 /* Clear everything. Note we don't initialize spinlocks
6479 * are they aren't supposed to be taken by any of the
6480 * NAPI code and this dummy netdev is supposed to be
6481 * only ever used for NAPI polls
6482 */
6483 memset(dev, 0, sizeof(struct net_device));
6484
6485 /* make sure we BUG if trying to hit standard
6486 * register/unregister code path
6487 */
6488 dev->reg_state = NETREG_DUMMY;
6489
6490 /* NAPI wants this */
6491 INIT_LIST_HEAD(&dev->napi_list);
6492
6493 /* a dummy interface is started by default */
6494 set_bit(__LINK_STATE_PRESENT, &dev->state);
6495 set_bit(__LINK_STATE_START, &dev->state);
6496
6497 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6498 * because users of this 'device' dont need to change
6499 * its refcount.
6500 */
6501
6502 return 0;
6503 }
6504 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6505
6506
6507 /**
6508 * register_netdev - register a network device
6509 * @dev: device to register
6510 *
6511 * Take a completed network device structure and add it to the kernel
6512 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6513 * chain. 0 is returned on success. A negative errno code is returned
6514 * on a failure to set up the device, or if the name is a duplicate.
6515 *
6516 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6517 * and expands the device name if you passed a format string to
6518 * alloc_netdev.
6519 */
register_netdev(struct net_device * dev)6520 int register_netdev(struct net_device *dev)
6521 {
6522 int err;
6523
6524 rtnl_lock();
6525 err = register_netdevice(dev);
6526 rtnl_unlock();
6527 return err;
6528 }
6529 EXPORT_SYMBOL(register_netdev);
6530
netdev_refcnt_read(const struct net_device * dev)6531 int netdev_refcnt_read(const struct net_device *dev)
6532 {
6533 int i, refcnt = 0;
6534
6535 for_each_possible_cpu(i)
6536 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6537 return refcnt;
6538 }
6539 EXPORT_SYMBOL(netdev_refcnt_read);
6540
6541 /**
6542 * netdev_wait_allrefs - wait until all references are gone.
6543 * @dev: target net_device
6544 *
6545 * This is called when unregistering network devices.
6546 *
6547 * Any protocol or device that holds a reference should register
6548 * for netdevice notification, and cleanup and put back the
6549 * reference if they receive an UNREGISTER event.
6550 * We can get stuck here if buggy protocols don't correctly
6551 * call dev_put.
6552 */
netdev_wait_allrefs(struct net_device * dev)6553 static void netdev_wait_allrefs(struct net_device *dev)
6554 {
6555 unsigned long rebroadcast_time, warning_time;
6556 int refcnt;
6557
6558 linkwatch_forget_dev(dev);
6559
6560 rebroadcast_time = warning_time = jiffies;
6561 refcnt = netdev_refcnt_read(dev);
6562
6563 while (refcnt != 0) {
6564 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6565 rtnl_lock();
6566
6567 /* Rebroadcast unregister notification */
6568 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6569
6570 __rtnl_unlock();
6571 rcu_barrier();
6572 rtnl_lock();
6573
6574 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6575 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6576 &dev->state)) {
6577 /* We must not have linkwatch events
6578 * pending on unregister. If this
6579 * happens, we simply run the queue
6580 * unscheduled, resulting in a noop
6581 * for this device.
6582 */
6583 linkwatch_run_queue();
6584 }
6585
6586 __rtnl_unlock();
6587
6588 rebroadcast_time = jiffies;
6589 }
6590
6591 msleep(250);
6592
6593 refcnt = netdev_refcnt_read(dev);
6594
6595 if (time_after(jiffies, warning_time + 10 * HZ)) {
6596 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6597 dev->name, refcnt);
6598 warning_time = jiffies;
6599 }
6600 }
6601 }
6602
6603 /* The sequence is:
6604 *
6605 * rtnl_lock();
6606 * ...
6607 * register_netdevice(x1);
6608 * register_netdevice(x2);
6609 * ...
6610 * unregister_netdevice(y1);
6611 * unregister_netdevice(y2);
6612 * ...
6613 * rtnl_unlock();
6614 * free_netdev(y1);
6615 * free_netdev(y2);
6616 *
6617 * We are invoked by rtnl_unlock().
6618 * This allows us to deal with problems:
6619 * 1) We can delete sysfs objects which invoke hotplug
6620 * without deadlocking with linkwatch via keventd.
6621 * 2) Since we run with the RTNL semaphore not held, we can sleep
6622 * safely in order to wait for the netdev refcnt to drop to zero.
6623 *
6624 * We must not return until all unregister events added during
6625 * the interval the lock was held have been completed.
6626 */
netdev_run_todo(void)6627 void netdev_run_todo(void)
6628 {
6629 struct list_head list;
6630
6631 /* Snapshot list, allow later requests */
6632 list_replace_init(&net_todo_list, &list);
6633
6634 __rtnl_unlock();
6635
6636
6637 /* Wait for rcu callbacks to finish before next phase */
6638 if (!list_empty(&list))
6639 rcu_barrier();
6640
6641 while (!list_empty(&list)) {
6642 struct net_device *dev
6643 = list_first_entry(&list, struct net_device, todo_list);
6644 list_del(&dev->todo_list);
6645
6646 rtnl_lock();
6647 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6648 __rtnl_unlock();
6649
6650 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6651 pr_err("network todo '%s' but state %d\n",
6652 dev->name, dev->reg_state);
6653 dump_stack();
6654 continue;
6655 }
6656
6657 dev->reg_state = NETREG_UNREGISTERED;
6658
6659 netdev_wait_allrefs(dev);
6660
6661 /* paranoia */
6662 BUG_ON(netdev_refcnt_read(dev));
6663 BUG_ON(!list_empty(&dev->ptype_all));
6664 BUG_ON(!list_empty(&dev->ptype_specific));
6665 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6666 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6667 WARN_ON(dev->dn_ptr);
6668
6669 if (dev->destructor)
6670 dev->destructor(dev);
6671
6672 /* Report a network device has been unregistered */
6673 rtnl_lock();
6674 dev_net(dev)->dev_unreg_count--;
6675 __rtnl_unlock();
6676 wake_up(&netdev_unregistering_wq);
6677
6678 /* Free network device */
6679 kobject_put(&dev->dev.kobj);
6680 }
6681 }
6682
6683 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6684 * fields in the same order, with only the type differing.
6685 */
netdev_stats_to_stats64(struct rtnl_link_stats64 * stats64,const struct net_device_stats * netdev_stats)6686 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6687 const struct net_device_stats *netdev_stats)
6688 {
6689 #if BITS_PER_LONG == 64
6690 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6691 memcpy(stats64, netdev_stats, sizeof(*stats64));
6692 #else
6693 size_t i, n = sizeof(*stats64) / sizeof(u64);
6694 const unsigned long *src = (const unsigned long *)netdev_stats;
6695 u64 *dst = (u64 *)stats64;
6696
6697 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6698 sizeof(*stats64) / sizeof(u64));
6699 for (i = 0; i < n; i++)
6700 dst[i] = src[i];
6701 #endif
6702 }
6703 EXPORT_SYMBOL(netdev_stats_to_stats64);
6704
6705 /**
6706 * dev_get_stats - get network device statistics
6707 * @dev: device to get statistics from
6708 * @storage: place to store stats
6709 *
6710 * Get network statistics from device. Return @storage.
6711 * The device driver may provide its own method by setting
6712 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6713 * otherwise the internal statistics structure is used.
6714 */
dev_get_stats(struct net_device * dev,struct rtnl_link_stats64 * storage)6715 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6716 struct rtnl_link_stats64 *storage)
6717 {
6718 const struct net_device_ops *ops = dev->netdev_ops;
6719
6720 if (ops->ndo_get_stats64) {
6721 memset(storage, 0, sizeof(*storage));
6722 ops->ndo_get_stats64(dev, storage);
6723 } else if (ops->ndo_get_stats) {
6724 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6725 } else {
6726 netdev_stats_to_stats64(storage, &dev->stats);
6727 }
6728 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6729 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6730 return storage;
6731 }
6732 EXPORT_SYMBOL(dev_get_stats);
6733
dev_ingress_queue_create(struct net_device * dev)6734 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6735 {
6736 struct netdev_queue *queue = dev_ingress_queue(dev);
6737
6738 #ifdef CONFIG_NET_CLS_ACT
6739 if (queue)
6740 return queue;
6741 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6742 if (!queue)
6743 return NULL;
6744 netdev_init_one_queue(dev, queue, NULL);
6745 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6746 queue->qdisc_sleeping = &noop_qdisc;
6747 rcu_assign_pointer(dev->ingress_queue, queue);
6748 #endif
6749 return queue;
6750 }
6751
6752 static const struct ethtool_ops default_ethtool_ops;
6753
netdev_set_default_ethtool_ops(struct net_device * dev,const struct ethtool_ops * ops)6754 void netdev_set_default_ethtool_ops(struct net_device *dev,
6755 const struct ethtool_ops *ops)
6756 {
6757 if (dev->ethtool_ops == &default_ethtool_ops)
6758 dev->ethtool_ops = ops;
6759 }
6760 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6761
netdev_freemem(struct net_device * dev)6762 void netdev_freemem(struct net_device *dev)
6763 {
6764 char *addr = (char *)dev - dev->padded;
6765
6766 kvfree(addr);
6767 }
6768
6769 /**
6770 * alloc_netdev_mqs - allocate network device
6771 * @sizeof_priv: size of private data to allocate space for
6772 * @name: device name format string
6773 * @name_assign_type: origin of device name
6774 * @setup: callback to initialize device
6775 * @txqs: the number of TX subqueues to allocate
6776 * @rxqs: the number of RX subqueues to allocate
6777 *
6778 * Allocates a struct net_device with private data area for driver use
6779 * and performs basic initialization. Also allocates subqueue structs
6780 * for each queue on the device.
6781 */
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)6782 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6783 unsigned char name_assign_type,
6784 void (*setup)(struct net_device *),
6785 unsigned int txqs, unsigned int rxqs)
6786 {
6787 struct net_device *dev;
6788 size_t alloc_size;
6789 struct net_device *p;
6790
6791 BUG_ON(strlen(name) >= sizeof(dev->name));
6792
6793 if (txqs < 1) {
6794 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6795 return NULL;
6796 }
6797
6798 #ifdef CONFIG_SYSFS
6799 if (rxqs < 1) {
6800 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6801 return NULL;
6802 }
6803 #endif
6804
6805 alloc_size = sizeof(struct net_device);
6806 if (sizeof_priv) {
6807 /* ensure 32-byte alignment of private area */
6808 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6809 alloc_size += sizeof_priv;
6810 }
6811 /* ensure 32-byte alignment of whole construct */
6812 alloc_size += NETDEV_ALIGN - 1;
6813
6814 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6815 if (!p)
6816 p = vzalloc(alloc_size);
6817 if (!p)
6818 return NULL;
6819
6820 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6821 dev->padded = (char *)dev - (char *)p;
6822
6823 dev->pcpu_refcnt = alloc_percpu(int);
6824 if (!dev->pcpu_refcnt)
6825 goto free_dev;
6826
6827 if (dev_addr_init(dev))
6828 goto free_pcpu;
6829
6830 dev_mc_init(dev);
6831 dev_uc_init(dev);
6832
6833 dev_net_set(dev, &init_net);
6834
6835 dev->gso_max_size = GSO_MAX_SIZE;
6836 dev->gso_max_segs = GSO_MAX_SEGS;
6837 dev->gso_min_segs = 0;
6838
6839 INIT_LIST_HEAD(&dev->napi_list);
6840 INIT_LIST_HEAD(&dev->unreg_list);
6841 INIT_LIST_HEAD(&dev->close_list);
6842 INIT_LIST_HEAD(&dev->link_watch_list);
6843 INIT_LIST_HEAD(&dev->adj_list.upper);
6844 INIT_LIST_HEAD(&dev->adj_list.lower);
6845 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6846 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6847 INIT_LIST_HEAD(&dev->ptype_all);
6848 INIT_LIST_HEAD(&dev->ptype_specific);
6849 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6850 setup(dev);
6851
6852 dev->num_tx_queues = txqs;
6853 dev->real_num_tx_queues = txqs;
6854 if (netif_alloc_netdev_queues(dev))
6855 goto free_all;
6856
6857 #ifdef CONFIG_SYSFS
6858 dev->num_rx_queues = rxqs;
6859 dev->real_num_rx_queues = rxqs;
6860 if (netif_alloc_rx_queues(dev))
6861 goto free_all;
6862 #endif
6863
6864 strcpy(dev->name, name);
6865 dev->name_assign_type = name_assign_type;
6866 dev->group = INIT_NETDEV_GROUP;
6867 if (!dev->ethtool_ops)
6868 dev->ethtool_ops = &default_ethtool_ops;
6869 return dev;
6870
6871 free_all:
6872 free_netdev(dev);
6873 return NULL;
6874
6875 free_pcpu:
6876 free_percpu(dev->pcpu_refcnt);
6877 free_dev:
6878 netdev_freemem(dev);
6879 return NULL;
6880 }
6881 EXPORT_SYMBOL(alloc_netdev_mqs);
6882
6883 /**
6884 * free_netdev - free network device
6885 * @dev: device
6886 *
6887 * This function does the last stage of destroying an allocated device
6888 * interface. The reference to the device object is released.
6889 * If this is the last reference then it will be freed.
6890 */
free_netdev(struct net_device * dev)6891 void free_netdev(struct net_device *dev)
6892 {
6893 struct napi_struct *p, *n;
6894
6895 netif_free_tx_queues(dev);
6896 #ifdef CONFIG_SYSFS
6897 kvfree(dev->_rx);
6898 #endif
6899
6900 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6901
6902 /* Flush device addresses */
6903 dev_addr_flush(dev);
6904
6905 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6906 netif_napi_del(p);
6907
6908 free_percpu(dev->pcpu_refcnt);
6909 dev->pcpu_refcnt = NULL;
6910
6911 /* Compatibility with error handling in drivers */
6912 if (dev->reg_state == NETREG_UNINITIALIZED) {
6913 netdev_freemem(dev);
6914 return;
6915 }
6916
6917 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6918 dev->reg_state = NETREG_RELEASED;
6919
6920 /* will free via device release */
6921 put_device(&dev->dev);
6922 }
6923 EXPORT_SYMBOL(free_netdev);
6924
6925 /**
6926 * synchronize_net - Synchronize with packet receive processing
6927 *
6928 * Wait for packets currently being received to be done.
6929 * Does not block later packets from starting.
6930 */
synchronize_net(void)6931 void synchronize_net(void)
6932 {
6933 might_sleep();
6934 if (rtnl_is_locked())
6935 synchronize_rcu_expedited();
6936 else
6937 synchronize_rcu();
6938 }
6939 EXPORT_SYMBOL(synchronize_net);
6940
6941 /**
6942 * unregister_netdevice_queue - remove device from the kernel
6943 * @dev: device
6944 * @head: list
6945 *
6946 * This function shuts down a device interface and removes it
6947 * from the kernel tables.
6948 * If head not NULL, device is queued to be unregistered later.
6949 *
6950 * Callers must hold the rtnl semaphore. You may want
6951 * unregister_netdev() instead of this.
6952 */
6953
unregister_netdevice_queue(struct net_device * dev,struct list_head * head)6954 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6955 {
6956 ASSERT_RTNL();
6957
6958 if (head) {
6959 list_move_tail(&dev->unreg_list, head);
6960 } else {
6961 rollback_registered(dev);
6962 /* Finish processing unregister after unlock */
6963 net_set_todo(dev);
6964 }
6965 }
6966 EXPORT_SYMBOL(unregister_netdevice_queue);
6967
6968 /**
6969 * unregister_netdevice_many - unregister many devices
6970 * @head: list of devices
6971 *
6972 * Note: As most callers use a stack allocated list_head,
6973 * we force a list_del() to make sure stack wont be corrupted later.
6974 */
unregister_netdevice_many(struct list_head * head)6975 void unregister_netdevice_many(struct list_head *head)
6976 {
6977 struct net_device *dev;
6978
6979 if (!list_empty(head)) {
6980 rollback_registered_many(head);
6981 list_for_each_entry(dev, head, unreg_list)
6982 net_set_todo(dev);
6983 list_del(head);
6984 }
6985 }
6986 EXPORT_SYMBOL(unregister_netdevice_many);
6987
6988 /**
6989 * unregister_netdev - remove device from the kernel
6990 * @dev: device
6991 *
6992 * This function shuts down a device interface and removes it
6993 * from the kernel tables.
6994 *
6995 * This is just a wrapper for unregister_netdevice that takes
6996 * the rtnl semaphore. In general you want to use this and not
6997 * unregister_netdevice.
6998 */
unregister_netdev(struct net_device * dev)6999 void unregister_netdev(struct net_device *dev)
7000 {
7001 rtnl_lock();
7002 unregister_netdevice(dev);
7003 rtnl_unlock();
7004 }
7005 EXPORT_SYMBOL(unregister_netdev);
7006
7007 /**
7008 * dev_change_net_namespace - move device to different nethost namespace
7009 * @dev: device
7010 * @net: network namespace
7011 * @pat: If not NULL name pattern to try if the current device name
7012 * is already taken in the destination network namespace.
7013 *
7014 * This function shuts down a device interface and moves it
7015 * to a new network namespace. On success 0 is returned, on
7016 * a failure a netagive errno code is returned.
7017 *
7018 * Callers must hold the rtnl semaphore.
7019 */
7020
dev_change_net_namespace(struct net_device * dev,struct net * net,const char * pat)7021 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7022 {
7023 int err;
7024
7025 ASSERT_RTNL();
7026
7027 /* Don't allow namespace local devices to be moved. */
7028 err = -EINVAL;
7029 if (dev->features & NETIF_F_NETNS_LOCAL)
7030 goto out;
7031
7032 /* Ensure the device has been registrered */
7033 if (dev->reg_state != NETREG_REGISTERED)
7034 goto out;
7035
7036 /* Get out if there is nothing todo */
7037 err = 0;
7038 if (net_eq(dev_net(dev), net))
7039 goto out;
7040
7041 /* Pick the destination device name, and ensure
7042 * we can use it in the destination network namespace.
7043 */
7044 err = -EEXIST;
7045 if (__dev_get_by_name(net, dev->name)) {
7046 /* We get here if we can't use the current device name */
7047 if (!pat)
7048 goto out;
7049 if (dev_get_valid_name(net, dev, pat) < 0)
7050 goto out;
7051 }
7052
7053 /*
7054 * And now a mini version of register_netdevice unregister_netdevice.
7055 */
7056
7057 /* If device is running close it first. */
7058 dev_close(dev);
7059
7060 /* And unlink it from device chain */
7061 err = -ENODEV;
7062 unlist_netdevice(dev);
7063
7064 synchronize_net();
7065
7066 /* Shutdown queueing discipline. */
7067 dev_shutdown(dev);
7068
7069 /* Notify protocols, that we are about to destroy
7070 this device. They should clean all the things.
7071
7072 Note that dev->reg_state stays at NETREG_REGISTERED.
7073 This is wanted because this way 8021q and macvlan know
7074 the device is just moving and can keep their slaves up.
7075 */
7076 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7077 rcu_barrier();
7078 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7079 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7080
7081 /*
7082 * Flush the unicast and multicast chains
7083 */
7084 dev_uc_flush(dev);
7085 dev_mc_flush(dev);
7086
7087 /* Send a netdev-removed uevent to the old namespace */
7088 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7089 netdev_adjacent_del_links(dev);
7090
7091 /* Actually switch the network namespace */
7092 dev_net_set(dev, net);
7093
7094 /* If there is an ifindex conflict assign a new one */
7095 if (__dev_get_by_index(net, dev->ifindex))
7096 dev->ifindex = dev_new_index(net);
7097
7098 /* Send a netdev-add uevent to the new namespace */
7099 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7100 netdev_adjacent_add_links(dev);
7101
7102 /* Fixup kobjects */
7103 err = device_rename(&dev->dev, dev->name);
7104 WARN_ON(err);
7105
7106 /* Add the device back in the hashes */
7107 list_netdevice(dev);
7108
7109 /* Notify protocols, that a new device appeared. */
7110 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7111
7112 /*
7113 * Prevent userspace races by waiting until the network
7114 * device is fully setup before sending notifications.
7115 */
7116 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7117
7118 synchronize_net();
7119 err = 0;
7120 out:
7121 return err;
7122 }
7123 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7124
dev_cpu_callback(struct notifier_block * nfb,unsigned long action,void * ocpu)7125 static int dev_cpu_callback(struct notifier_block *nfb,
7126 unsigned long action,
7127 void *ocpu)
7128 {
7129 struct sk_buff **list_skb;
7130 struct sk_buff *skb;
7131 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7132 struct softnet_data *sd, *oldsd;
7133
7134 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7135 return NOTIFY_OK;
7136
7137 local_irq_disable();
7138 cpu = smp_processor_id();
7139 sd = &per_cpu(softnet_data, cpu);
7140 oldsd = &per_cpu(softnet_data, oldcpu);
7141
7142 /* Find end of our completion_queue. */
7143 list_skb = &sd->completion_queue;
7144 while (*list_skb)
7145 list_skb = &(*list_skb)->next;
7146 /* Append completion queue from offline CPU. */
7147 *list_skb = oldsd->completion_queue;
7148 oldsd->completion_queue = NULL;
7149
7150 /* Append output queue from offline CPU. */
7151 if (oldsd->output_queue) {
7152 *sd->output_queue_tailp = oldsd->output_queue;
7153 sd->output_queue_tailp = oldsd->output_queue_tailp;
7154 oldsd->output_queue = NULL;
7155 oldsd->output_queue_tailp = &oldsd->output_queue;
7156 }
7157 /* Append NAPI poll list from offline CPU, with one exception :
7158 * process_backlog() must be called by cpu owning percpu backlog.
7159 * We properly handle process_queue & input_pkt_queue later.
7160 */
7161 while (!list_empty(&oldsd->poll_list)) {
7162 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7163 struct napi_struct,
7164 poll_list);
7165
7166 list_del_init(&napi->poll_list);
7167 if (napi->poll == process_backlog)
7168 napi->state = 0;
7169 else
7170 ____napi_schedule(sd, napi);
7171 }
7172
7173 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7174 local_irq_enable();
7175
7176 /* Process offline CPU's input_pkt_queue */
7177 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7178 netif_rx_ni(skb);
7179 input_queue_head_incr(oldsd);
7180 }
7181 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7182 netif_rx_ni(skb);
7183 input_queue_head_incr(oldsd);
7184 }
7185
7186 return NOTIFY_OK;
7187 }
7188
7189
7190 /**
7191 * netdev_increment_features - increment feature set by one
7192 * @all: current feature set
7193 * @one: new feature set
7194 * @mask: mask feature set
7195 *
7196 * Computes a new feature set after adding a device with feature set
7197 * @one to the master device with current feature set @all. Will not
7198 * enable anything that is off in @mask. Returns the new feature set.
7199 */
netdev_increment_features(netdev_features_t all,netdev_features_t one,netdev_features_t mask)7200 netdev_features_t netdev_increment_features(netdev_features_t all,
7201 netdev_features_t one, netdev_features_t mask)
7202 {
7203 if (mask & NETIF_F_GEN_CSUM)
7204 mask |= NETIF_F_ALL_CSUM;
7205 mask |= NETIF_F_VLAN_CHALLENGED;
7206
7207 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7208 all &= one | ~NETIF_F_ALL_FOR_ALL;
7209
7210 /* If one device supports hw checksumming, set for all. */
7211 if (all & NETIF_F_GEN_CSUM)
7212 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7213
7214 return all;
7215 }
7216 EXPORT_SYMBOL(netdev_increment_features);
7217
netdev_create_hash(void)7218 static struct hlist_head * __net_init netdev_create_hash(void)
7219 {
7220 int i;
7221 struct hlist_head *hash;
7222
7223 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7224 if (hash != NULL)
7225 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7226 INIT_HLIST_HEAD(&hash[i]);
7227
7228 return hash;
7229 }
7230
7231 /* Initialize per network namespace state */
netdev_init(struct net * net)7232 static int __net_init netdev_init(struct net *net)
7233 {
7234 if (net != &init_net)
7235 INIT_LIST_HEAD(&net->dev_base_head);
7236
7237 net->dev_name_head = netdev_create_hash();
7238 if (net->dev_name_head == NULL)
7239 goto err_name;
7240
7241 net->dev_index_head = netdev_create_hash();
7242 if (net->dev_index_head == NULL)
7243 goto err_idx;
7244
7245 return 0;
7246
7247 err_idx:
7248 kfree(net->dev_name_head);
7249 err_name:
7250 return -ENOMEM;
7251 }
7252
7253 /**
7254 * netdev_drivername - network driver for the device
7255 * @dev: network device
7256 *
7257 * Determine network driver for device.
7258 */
netdev_drivername(const struct net_device * dev)7259 const char *netdev_drivername(const struct net_device *dev)
7260 {
7261 const struct device_driver *driver;
7262 const struct device *parent;
7263 const char *empty = "";
7264
7265 parent = dev->dev.parent;
7266 if (!parent)
7267 return empty;
7268
7269 driver = parent->driver;
7270 if (driver && driver->name)
7271 return driver->name;
7272 return empty;
7273 }
7274
__netdev_printk(const char * level,const struct net_device * dev,struct va_format * vaf)7275 static void __netdev_printk(const char *level, const struct net_device *dev,
7276 struct va_format *vaf)
7277 {
7278 if (dev && dev->dev.parent) {
7279 dev_printk_emit(level[1] - '0',
7280 dev->dev.parent,
7281 "%s %s %s%s: %pV",
7282 dev_driver_string(dev->dev.parent),
7283 dev_name(dev->dev.parent),
7284 netdev_name(dev), netdev_reg_state(dev),
7285 vaf);
7286 } else if (dev) {
7287 printk("%s%s%s: %pV",
7288 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7289 } else {
7290 printk("%s(NULL net_device): %pV", level, vaf);
7291 }
7292 }
7293
netdev_printk(const char * level,const struct net_device * dev,const char * format,...)7294 void netdev_printk(const char *level, const struct net_device *dev,
7295 const char *format, ...)
7296 {
7297 struct va_format vaf;
7298 va_list args;
7299
7300 va_start(args, format);
7301
7302 vaf.fmt = format;
7303 vaf.va = &args;
7304
7305 __netdev_printk(level, dev, &vaf);
7306
7307 va_end(args);
7308 }
7309 EXPORT_SYMBOL(netdev_printk);
7310
7311 #define define_netdev_printk_level(func, level) \
7312 void func(const struct net_device *dev, const char *fmt, ...) \
7313 { \
7314 struct va_format vaf; \
7315 va_list args; \
7316 \
7317 va_start(args, fmt); \
7318 \
7319 vaf.fmt = fmt; \
7320 vaf.va = &args; \
7321 \
7322 __netdev_printk(level, dev, &vaf); \
7323 \
7324 va_end(args); \
7325 } \
7326 EXPORT_SYMBOL(func);
7327
7328 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7329 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7330 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7331 define_netdev_printk_level(netdev_err, KERN_ERR);
7332 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7333 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7334 define_netdev_printk_level(netdev_info, KERN_INFO);
7335
netdev_exit(struct net * net)7336 static void __net_exit netdev_exit(struct net *net)
7337 {
7338 kfree(net->dev_name_head);
7339 kfree(net->dev_index_head);
7340 }
7341
7342 static struct pernet_operations __net_initdata netdev_net_ops = {
7343 .init = netdev_init,
7344 .exit = netdev_exit,
7345 };
7346
default_device_exit(struct net * net)7347 static void __net_exit default_device_exit(struct net *net)
7348 {
7349 struct net_device *dev, *aux;
7350 /*
7351 * Push all migratable network devices back to the
7352 * initial network namespace
7353 */
7354 rtnl_lock();
7355 for_each_netdev_safe(net, dev, aux) {
7356 int err;
7357 char fb_name[IFNAMSIZ];
7358
7359 /* Ignore unmoveable devices (i.e. loopback) */
7360 if (dev->features & NETIF_F_NETNS_LOCAL)
7361 continue;
7362
7363 /* Leave virtual devices for the generic cleanup */
7364 if (dev->rtnl_link_ops)
7365 continue;
7366
7367 /* Push remaining network devices to init_net */
7368 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7369 err = dev_change_net_namespace(dev, &init_net, fb_name);
7370 if (err) {
7371 pr_emerg("%s: failed to move %s to init_net: %d\n",
7372 __func__, dev->name, err);
7373 BUG();
7374 }
7375 }
7376 rtnl_unlock();
7377 }
7378
rtnl_lock_unregistering(struct list_head * net_list)7379 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7380 {
7381 /* Return with the rtnl_lock held when there are no network
7382 * devices unregistering in any network namespace in net_list.
7383 */
7384 struct net *net;
7385 bool unregistering;
7386 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7387
7388 add_wait_queue(&netdev_unregistering_wq, &wait);
7389 for (;;) {
7390 unregistering = false;
7391 rtnl_lock();
7392 list_for_each_entry(net, net_list, exit_list) {
7393 if (net->dev_unreg_count > 0) {
7394 unregistering = true;
7395 break;
7396 }
7397 }
7398 if (!unregistering)
7399 break;
7400 __rtnl_unlock();
7401
7402 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7403 }
7404 remove_wait_queue(&netdev_unregistering_wq, &wait);
7405 }
7406
default_device_exit_batch(struct list_head * net_list)7407 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7408 {
7409 /* At exit all network devices most be removed from a network
7410 * namespace. Do this in the reverse order of registration.
7411 * Do this across as many network namespaces as possible to
7412 * improve batching efficiency.
7413 */
7414 struct net_device *dev;
7415 struct net *net;
7416 LIST_HEAD(dev_kill_list);
7417
7418 /* To prevent network device cleanup code from dereferencing
7419 * loopback devices or network devices that have been freed
7420 * wait here for all pending unregistrations to complete,
7421 * before unregistring the loopback device and allowing the
7422 * network namespace be freed.
7423 *
7424 * The netdev todo list containing all network devices
7425 * unregistrations that happen in default_device_exit_batch
7426 * will run in the rtnl_unlock() at the end of
7427 * default_device_exit_batch.
7428 */
7429 rtnl_lock_unregistering(net_list);
7430 list_for_each_entry(net, net_list, exit_list) {
7431 for_each_netdev_reverse(net, dev) {
7432 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7433 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7434 else
7435 unregister_netdevice_queue(dev, &dev_kill_list);
7436 }
7437 }
7438 unregister_netdevice_many(&dev_kill_list);
7439 rtnl_unlock();
7440 }
7441
7442 static struct pernet_operations __net_initdata default_device_ops = {
7443 .exit = default_device_exit,
7444 .exit_batch = default_device_exit_batch,
7445 };
7446
7447 /*
7448 * Initialize the DEV module. At boot time this walks the device list and
7449 * unhooks any devices that fail to initialise (normally hardware not
7450 * present) and leaves us with a valid list of present and active devices.
7451 *
7452 */
7453
7454 /*
7455 * This is called single threaded during boot, so no need
7456 * to take the rtnl semaphore.
7457 */
net_dev_init(void)7458 static int __init net_dev_init(void)
7459 {
7460 int i, rc = -ENOMEM;
7461
7462 BUG_ON(!dev_boot_phase);
7463
7464 if (dev_proc_init())
7465 goto out;
7466
7467 if (netdev_kobject_init())
7468 goto out;
7469
7470 INIT_LIST_HEAD(&ptype_all);
7471 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7472 INIT_LIST_HEAD(&ptype_base[i]);
7473
7474 INIT_LIST_HEAD(&offload_base);
7475
7476 if (register_pernet_subsys(&netdev_net_ops))
7477 goto out;
7478
7479 /*
7480 * Initialise the packet receive queues.
7481 */
7482
7483 for_each_possible_cpu(i) {
7484 struct softnet_data *sd = &per_cpu(softnet_data, i);
7485
7486 skb_queue_head_init(&sd->input_pkt_queue);
7487 skb_queue_head_init(&sd->process_queue);
7488 INIT_LIST_HEAD(&sd->poll_list);
7489 sd->output_queue_tailp = &sd->output_queue;
7490 #ifdef CONFIG_RPS
7491 sd->csd.func = rps_trigger_softirq;
7492 sd->csd.info = sd;
7493 sd->cpu = i;
7494 #endif
7495
7496 sd->backlog.poll = process_backlog;
7497 sd->backlog.weight = weight_p;
7498 }
7499
7500 dev_boot_phase = 0;
7501
7502 /* The loopback device is special if any other network devices
7503 * is present in a network namespace the loopback device must
7504 * be present. Since we now dynamically allocate and free the
7505 * loopback device ensure this invariant is maintained by
7506 * keeping the loopback device as the first device on the
7507 * list of network devices. Ensuring the loopback devices
7508 * is the first device that appears and the last network device
7509 * that disappears.
7510 */
7511 if (register_pernet_device(&loopback_net_ops))
7512 goto out;
7513
7514 if (register_pernet_device(&default_device_ops))
7515 goto out;
7516
7517 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7518 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7519
7520 hotcpu_notifier(dev_cpu_callback, 0);
7521 dst_init();
7522 rc = 0;
7523 out:
7524 return rc;
7525 }
7526
7527 subsys_initcall(net_dev_init);
7528