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 <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
85 #include <linux/mm.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
97 #include <net/sock.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
104 #include <net/dst.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
124 #include <net/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130
131 #include "net-sysfs.h"
132
133 /* Instead of increasing this, you should create a hash table. */
134 #define MAX_GRO_SKBS 8
135
136 /* This should be increased if a protocol with a bigger head is added. */
137 #define GRO_MAX_HEAD (MAX_HEADER + 128)
138
139 /*
140 * The list of packet types we will receive (as opposed to discard)
141 * and the routines to invoke.
142 *
143 * Why 16. Because with 16 the only overlap we get on a hash of the
144 * low nibble of the protocol value is RARP/SNAP/X.25.
145 *
146 * NOTE: That is no longer true with the addition of VLAN tags. Not
147 * sure which should go first, but I bet it won't make much
148 * difference if we are running VLANs. The good news is that
149 * this protocol won't be in the list unless compiled in, so
150 * the average user (w/out VLANs) will not be adversely affected.
151 * --BLG
152 *
153 * 0800 IP
154 * 8100 802.1Q VLAN
155 * 0001 802.3
156 * 0002 AX.25
157 * 0004 802.2
158 * 8035 RARP
159 * 0005 SNAP
160 * 0805 X.25
161 * 0806 ARP
162 * 8137 IPX
163 * 0009 Localtalk
164 * 86DD IPv6
165 */
166
167 #define PTYPE_HASH_SIZE (16)
168 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
169
170 static DEFINE_SPINLOCK(ptype_lock);
171 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
172 static struct list_head ptype_all __read_mostly; /* Taps */
173
174 /*
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
176 * semaphore.
177 *
178 * Pure readers hold dev_base_lock for reading.
179 *
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
184 *
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
188 *
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
191 * semaphore held.
192 */
193 DEFINE_RWLOCK(dev_base_lock);
194
195 EXPORT_SYMBOL(dev_base_lock);
196
197 #define NETDEV_HASHBITS 8
198 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
199
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 {
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
204 }
205
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 {
208 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
209 }
210
211 /* Device list insertion */
212 static int list_netdevice(struct net_device *dev)
213 {
214 struct net *net = dev_net(dev);
215
216 ASSERT_RTNL();
217
218 write_lock_bh(&dev_base_lock);
219 list_add_tail(&dev->dev_list, &net->dev_base_head);
220 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
221 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
223 return 0;
224 }
225
226 /* Device list removal */
227 static void unlist_netdevice(struct net_device *dev)
228 {
229 ASSERT_RTNL();
230
231 /* Unlink dev from the device chain */
232 write_lock_bh(&dev_base_lock);
233 list_del(&dev->dev_list);
234 hlist_del(&dev->name_hlist);
235 hlist_del(&dev->index_hlist);
236 write_unlock_bh(&dev_base_lock);
237 }
238
239 /*
240 * Our notifier list
241 */
242
243 static RAW_NOTIFIER_HEAD(netdev_chain);
244
245 /*
246 * Device drivers call our routines to queue packets here. We empty the
247 * queue in the local softnet handler.
248 */
249
250 DEFINE_PER_CPU(struct softnet_data, softnet_data);
251
252 #ifdef CONFIG_LOCKDEP
253 /*
254 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
255 * according to dev->type
256 */
257 static const unsigned short netdev_lock_type[] =
258 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
259 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
260 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
261 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
262 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
263 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
264 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
265 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
266 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
267 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
268 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
269 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
270 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
271 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
272 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154, ARPHRD_IEEE802154_PHY,
273 ARPHRD_VOID, ARPHRD_NONE};
274
275 static const char *netdev_lock_name[] =
276 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
277 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
278 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
279 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
280 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
281 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
282 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
283 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
284 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
285 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
286 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
287 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
288 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
289 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
290 "_xmit_PHONET_PIPE", "_xmit_IEEE802154", "_xmit_IEEE802154_PHY",
291 "_xmit_VOID", "_xmit_NONE"};
292
293 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
294 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
295
296 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
297 {
298 int i;
299
300 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
301 if (netdev_lock_type[i] == dev_type)
302 return i;
303 /* the last key is used by default */
304 return ARRAY_SIZE(netdev_lock_type) - 1;
305 }
306
307 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
308 unsigned short dev_type)
309 {
310 int i;
311
312 i = netdev_lock_pos(dev_type);
313 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
314 netdev_lock_name[i]);
315 }
316
317 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
318 {
319 int i;
320
321 i = netdev_lock_pos(dev->type);
322 lockdep_set_class_and_name(&dev->addr_list_lock,
323 &netdev_addr_lock_key[i],
324 netdev_lock_name[i]);
325 }
326 #else
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
329 {
330 }
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 {
333 }
334 #endif
335
336 /*******************************************************************************
337
338 Protocol management and registration routines
339
340 *******************************************************************************/
341
342 /*
343 * Add a protocol ID to the list. Now that the input handler is
344 * smarter we can dispense with all the messy stuff that used to be
345 * here.
346 *
347 * BEWARE!!! Protocol handlers, mangling input packets,
348 * MUST BE last in hash buckets and checking protocol handlers
349 * MUST start from promiscuous ptype_all chain in net_bh.
350 * It is true now, do not change it.
351 * Explanation follows: if protocol handler, mangling packet, will
352 * be the first on list, it is not able to sense, that packet
353 * is cloned and should be copied-on-write, so that it will
354 * change it and subsequent readers will get broken packet.
355 * --ANK (980803)
356 */
357
358 /**
359 * dev_add_pack - add packet handler
360 * @pt: packet type declaration
361 *
362 * Add a protocol handler to the networking stack. The passed &packet_type
363 * is linked into kernel lists and may not be freed until it has been
364 * removed from the kernel lists.
365 *
366 * This call does not sleep therefore it can not
367 * guarantee all CPU's that are in middle of receiving packets
368 * will see the new packet type (until the next received packet).
369 */
370
371 void dev_add_pack(struct packet_type *pt)
372 {
373 int hash;
374
375 spin_lock_bh(&ptype_lock);
376 if (pt->type == htons(ETH_P_ALL))
377 list_add_rcu(&pt->list, &ptype_all);
378 else {
379 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
380 list_add_rcu(&pt->list, &ptype_base[hash]);
381 }
382 spin_unlock_bh(&ptype_lock);
383 }
384
385 /**
386 * __dev_remove_pack - remove packet handler
387 * @pt: packet type declaration
388 *
389 * Remove a protocol handler that was previously added to the kernel
390 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
391 * from the kernel lists and can be freed or reused once this function
392 * returns.
393 *
394 * The packet type might still be in use by receivers
395 * and must not be freed until after all the CPU's have gone
396 * through a quiescent state.
397 */
398 void __dev_remove_pack(struct packet_type *pt)
399 {
400 struct list_head *head;
401 struct packet_type *pt1;
402
403 spin_lock_bh(&ptype_lock);
404
405 if (pt->type == htons(ETH_P_ALL))
406 head = &ptype_all;
407 else
408 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
409
410 list_for_each_entry(pt1, head, list) {
411 if (pt == pt1) {
412 list_del_rcu(&pt->list);
413 goto out;
414 }
415 }
416
417 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
418 out:
419 spin_unlock_bh(&ptype_lock);
420 }
421 /**
422 * dev_remove_pack - remove packet handler
423 * @pt: packet type declaration
424 *
425 * Remove a protocol handler that was previously added to the kernel
426 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
427 * from the kernel lists and can be freed or reused once this function
428 * returns.
429 *
430 * This call sleeps to guarantee that no CPU is looking at the packet
431 * type after return.
432 */
433 void dev_remove_pack(struct packet_type *pt)
434 {
435 __dev_remove_pack(pt);
436
437 synchronize_net();
438 }
439
440 /******************************************************************************
441
442 Device Boot-time Settings Routines
443
444 *******************************************************************************/
445
446 /* Boot time configuration table */
447 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
448
449 /**
450 * netdev_boot_setup_add - add new setup entry
451 * @name: name of the device
452 * @map: configured settings for the device
453 *
454 * Adds new setup entry to the dev_boot_setup list. The function
455 * returns 0 on error and 1 on success. This is a generic routine to
456 * all netdevices.
457 */
458 static int netdev_boot_setup_add(char *name, struct ifmap *map)
459 {
460 struct netdev_boot_setup *s;
461 int i;
462
463 s = dev_boot_setup;
464 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
465 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
466 memset(s[i].name, 0, sizeof(s[i].name));
467 strlcpy(s[i].name, name, IFNAMSIZ);
468 memcpy(&s[i].map, map, sizeof(s[i].map));
469 break;
470 }
471 }
472
473 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
474 }
475
476 /**
477 * netdev_boot_setup_check - check boot time settings
478 * @dev: the netdevice
479 *
480 * Check boot time settings for the device.
481 * The found settings are set for the device to be used
482 * later in the device probing.
483 * Returns 0 if no settings found, 1 if they are.
484 */
485 int netdev_boot_setup_check(struct net_device *dev)
486 {
487 struct netdev_boot_setup *s = dev_boot_setup;
488 int i;
489
490 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
491 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
492 !strcmp(dev->name, s[i].name)) {
493 dev->irq = s[i].map.irq;
494 dev->base_addr = s[i].map.base_addr;
495 dev->mem_start = s[i].map.mem_start;
496 dev->mem_end = s[i].map.mem_end;
497 return 1;
498 }
499 }
500 return 0;
501 }
502
503
504 /**
505 * netdev_boot_base - get address from boot time settings
506 * @prefix: prefix for network device
507 * @unit: id for network device
508 *
509 * Check boot time settings for the base address of device.
510 * The found settings are set for the device to be used
511 * later in the device probing.
512 * Returns 0 if no settings found.
513 */
514 unsigned long netdev_boot_base(const char *prefix, int unit)
515 {
516 const struct netdev_boot_setup *s = dev_boot_setup;
517 char name[IFNAMSIZ];
518 int i;
519
520 sprintf(name, "%s%d", prefix, unit);
521
522 /*
523 * If device already registered then return base of 1
524 * to indicate not to probe for this interface
525 */
526 if (__dev_get_by_name(&init_net, name))
527 return 1;
528
529 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
530 if (!strcmp(name, s[i].name))
531 return s[i].map.base_addr;
532 return 0;
533 }
534
535 /*
536 * Saves at boot time configured settings for any netdevice.
537 */
538 int __init netdev_boot_setup(char *str)
539 {
540 int ints[5];
541 struct ifmap map;
542
543 str = get_options(str, ARRAY_SIZE(ints), ints);
544 if (!str || !*str)
545 return 0;
546
547 /* Save settings */
548 memset(&map, 0, sizeof(map));
549 if (ints[0] > 0)
550 map.irq = ints[1];
551 if (ints[0] > 1)
552 map.base_addr = ints[2];
553 if (ints[0] > 2)
554 map.mem_start = ints[3];
555 if (ints[0] > 3)
556 map.mem_end = ints[4];
557
558 /* Add new entry to the list */
559 return netdev_boot_setup_add(str, &map);
560 }
561
562 __setup("netdev=", netdev_boot_setup);
563
564 /*******************************************************************************
565
566 Device Interface Subroutines
567
568 *******************************************************************************/
569
570 /**
571 * __dev_get_by_name - find a device by its name
572 * @net: the applicable net namespace
573 * @name: name to find
574 *
575 * Find an interface by name. Must be called under RTNL semaphore
576 * or @dev_base_lock. If the name is found a pointer to the device
577 * is returned. If the name is not found then %NULL is returned. The
578 * reference counters are not incremented so the caller must be
579 * careful with locks.
580 */
581
582 struct net_device *__dev_get_by_name(struct net *net, const char *name)
583 {
584 struct hlist_node *p;
585
586 hlist_for_each(p, dev_name_hash(net, name)) {
587 struct net_device *dev
588 = hlist_entry(p, struct net_device, name_hlist);
589 if (!strncmp(dev->name, name, IFNAMSIZ))
590 return dev;
591 }
592 return NULL;
593 }
594
595 /**
596 * dev_get_by_name - find a device by its name
597 * @net: the applicable net namespace
598 * @name: name to find
599 *
600 * Find an interface by name. This can be called from any
601 * context and does its own locking. The returned handle has
602 * the usage count incremented and the caller must use dev_put() to
603 * release it when it is no longer needed. %NULL is returned if no
604 * matching device is found.
605 */
606
607 struct net_device *dev_get_by_name(struct net *net, const char *name)
608 {
609 struct net_device *dev;
610
611 read_lock(&dev_base_lock);
612 dev = __dev_get_by_name(net, name);
613 if (dev)
614 dev_hold(dev);
615 read_unlock(&dev_base_lock);
616 return dev;
617 }
618
619 /**
620 * __dev_get_by_index - find a device by its ifindex
621 * @net: the applicable net namespace
622 * @ifindex: index of device
623 *
624 * Search for an interface by index. Returns %NULL if the device
625 * is not found or a pointer to the device. The device has not
626 * had its reference counter increased so the caller must be careful
627 * about locking. The caller must hold either the RTNL semaphore
628 * or @dev_base_lock.
629 */
630
631 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
632 {
633 struct hlist_node *p;
634
635 hlist_for_each(p, dev_index_hash(net, ifindex)) {
636 struct net_device *dev
637 = hlist_entry(p, struct net_device, index_hlist);
638 if (dev->ifindex == ifindex)
639 return dev;
640 }
641 return NULL;
642 }
643
644
645 /**
646 * dev_get_by_index - find a device by its ifindex
647 * @net: the applicable net namespace
648 * @ifindex: index of device
649 *
650 * Search for an interface by index. Returns NULL if the device
651 * is not found or a pointer to the device. The device returned has
652 * had a reference added and the pointer is safe until the user calls
653 * dev_put to indicate they have finished with it.
654 */
655
656 struct net_device *dev_get_by_index(struct net *net, int ifindex)
657 {
658 struct net_device *dev;
659
660 read_lock(&dev_base_lock);
661 dev = __dev_get_by_index(net, ifindex);
662 if (dev)
663 dev_hold(dev);
664 read_unlock(&dev_base_lock);
665 return dev;
666 }
667
668 /**
669 * dev_getbyhwaddr - find a device by its hardware address
670 * @net: the applicable net namespace
671 * @type: media type of device
672 * @ha: hardware address
673 *
674 * Search for an interface by MAC address. Returns NULL if the device
675 * is not found or a pointer to the device. The caller must hold the
676 * rtnl semaphore. The returned device has not had its ref count increased
677 * and the caller must therefore be careful about locking
678 *
679 * BUGS:
680 * If the API was consistent this would be __dev_get_by_hwaddr
681 */
682
683 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
684 {
685 struct net_device *dev;
686
687 ASSERT_RTNL();
688
689 for_each_netdev(net, dev)
690 if (dev->type == type &&
691 !memcmp(dev->dev_addr, ha, dev->addr_len))
692 return dev;
693
694 return NULL;
695 }
696
697 EXPORT_SYMBOL(dev_getbyhwaddr);
698
699 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
700 {
701 struct net_device *dev;
702
703 ASSERT_RTNL();
704 for_each_netdev(net, dev)
705 if (dev->type == type)
706 return dev;
707
708 return NULL;
709 }
710
711 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
712
713 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
714 {
715 struct net_device *dev;
716
717 rtnl_lock();
718 dev = __dev_getfirstbyhwtype(net, type);
719 if (dev)
720 dev_hold(dev);
721 rtnl_unlock();
722 return dev;
723 }
724
725 EXPORT_SYMBOL(dev_getfirstbyhwtype);
726
727 /**
728 * dev_get_by_flags - find any device with given flags
729 * @net: the applicable net namespace
730 * @if_flags: IFF_* values
731 * @mask: bitmask of bits in if_flags to check
732 *
733 * Search for any interface with the given flags. Returns NULL if a device
734 * is not found or a pointer to the device. The device returned has
735 * had a reference added and the pointer is safe until the user calls
736 * dev_put to indicate they have finished with it.
737 */
738
739 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
740 {
741 struct net_device *dev, *ret;
742
743 ret = NULL;
744 read_lock(&dev_base_lock);
745 for_each_netdev(net, dev) {
746 if (((dev->flags ^ if_flags) & mask) == 0) {
747 dev_hold(dev);
748 ret = dev;
749 break;
750 }
751 }
752 read_unlock(&dev_base_lock);
753 return ret;
754 }
755
756 /**
757 * dev_valid_name - check if name is okay for network device
758 * @name: name string
759 *
760 * Network device names need to be valid file names to
761 * to allow sysfs to work. We also disallow any kind of
762 * whitespace.
763 */
764 int dev_valid_name(const char *name)
765 {
766 if (*name == '\0')
767 return 0;
768 if (strlen(name) >= IFNAMSIZ)
769 return 0;
770 if (!strcmp(name, ".") || !strcmp(name, ".."))
771 return 0;
772
773 while (*name) {
774 if (*name == '/' || isspace(*name))
775 return 0;
776 name++;
777 }
778 return 1;
779 }
780
781 /**
782 * __dev_alloc_name - allocate a name for a device
783 * @net: network namespace to allocate the device name in
784 * @name: name format string
785 * @buf: scratch buffer and result name string
786 *
787 * Passed a format string - eg "lt%d" it will try and find a suitable
788 * id. It scans list of devices to build up a free map, then chooses
789 * the first empty slot. The caller must hold the dev_base or rtnl lock
790 * while allocating the name and adding the device in order to avoid
791 * duplicates.
792 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
793 * Returns the number of the unit assigned or a negative errno code.
794 */
795
796 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
797 {
798 int i = 0;
799 const char *p;
800 const int max_netdevices = 8*PAGE_SIZE;
801 unsigned long *inuse;
802 struct net_device *d;
803
804 p = strnchr(name, IFNAMSIZ-1, '%');
805 if (p) {
806 /*
807 * Verify the string as this thing may have come from
808 * the user. There must be either one "%d" and no other "%"
809 * characters.
810 */
811 if (p[1] != 'd' || strchr(p + 2, '%'))
812 return -EINVAL;
813
814 /* Use one page as a bit array of possible slots */
815 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
816 if (!inuse)
817 return -ENOMEM;
818
819 for_each_netdev(net, d) {
820 if (!sscanf(d->name, name, &i))
821 continue;
822 if (i < 0 || i >= max_netdevices)
823 continue;
824
825 /* avoid cases where sscanf is not exact inverse of printf */
826 snprintf(buf, IFNAMSIZ, name, i);
827 if (!strncmp(buf, d->name, IFNAMSIZ))
828 set_bit(i, inuse);
829 }
830
831 i = find_first_zero_bit(inuse, max_netdevices);
832 free_page((unsigned long) inuse);
833 }
834
835 snprintf(buf, IFNAMSIZ, name, i);
836 if (!__dev_get_by_name(net, buf))
837 return i;
838
839 /* It is possible to run out of possible slots
840 * when the name is long and there isn't enough space left
841 * for the digits, or if all bits are used.
842 */
843 return -ENFILE;
844 }
845
846 /**
847 * dev_alloc_name - allocate a name for a device
848 * @dev: device
849 * @name: name format string
850 *
851 * Passed a format string - eg "lt%d" it will try and find a suitable
852 * id. It scans list of devices to build up a free map, then chooses
853 * the first empty slot. The caller must hold the dev_base or rtnl lock
854 * while allocating the name and adding the device in order to avoid
855 * duplicates.
856 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
857 * Returns the number of the unit assigned or a negative errno code.
858 */
859
860 int dev_alloc_name(struct net_device *dev, const char *name)
861 {
862 char buf[IFNAMSIZ];
863 struct net *net;
864 int ret;
865
866 BUG_ON(!dev_net(dev));
867 net = dev_net(dev);
868 ret = __dev_alloc_name(net, name, buf);
869 if (ret >= 0)
870 strlcpy(dev->name, buf, IFNAMSIZ);
871 return ret;
872 }
873
874
875 /**
876 * dev_change_name - change name of a device
877 * @dev: device
878 * @newname: name (or format string) must be at least IFNAMSIZ
879 *
880 * Change name of a device, can pass format strings "eth%d".
881 * for wildcarding.
882 */
883 int dev_change_name(struct net_device *dev, const char *newname)
884 {
885 char oldname[IFNAMSIZ];
886 int err = 0;
887 int ret;
888 struct net *net;
889
890 ASSERT_RTNL();
891 BUG_ON(!dev_net(dev));
892
893 net = dev_net(dev);
894 if (dev->flags & IFF_UP)
895 return -EBUSY;
896
897 if (!dev_valid_name(newname))
898 return -EINVAL;
899
900 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
901 return 0;
902
903 memcpy(oldname, dev->name, IFNAMSIZ);
904
905 if (strchr(newname, '%')) {
906 err = dev_alloc_name(dev, newname);
907 if (err < 0)
908 return err;
909 }
910 else if (__dev_get_by_name(net, newname))
911 return -EEXIST;
912 else
913 strlcpy(dev->name, newname, IFNAMSIZ);
914
915 rollback:
916 /* For now only devices in the initial network namespace
917 * are in sysfs.
918 */
919 if (net == &init_net) {
920 ret = device_rename(&dev->dev, dev->name);
921 if (ret) {
922 memcpy(dev->name, oldname, IFNAMSIZ);
923 return ret;
924 }
925 }
926
927 write_lock_bh(&dev_base_lock);
928 hlist_del(&dev->name_hlist);
929 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
930 write_unlock_bh(&dev_base_lock);
931
932 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
933 ret = notifier_to_errno(ret);
934
935 if (ret) {
936 /* err >= 0 after dev_alloc_name() or stores the first errno */
937 if (err >= 0) {
938 err = ret;
939 memcpy(dev->name, oldname, IFNAMSIZ);
940 goto rollback;
941 } else {
942 printk(KERN_ERR
943 "%s: name change rollback failed: %d.\n",
944 dev->name, ret);
945 }
946 }
947
948 return err;
949 }
950
951 /**
952 * dev_set_alias - change ifalias of a device
953 * @dev: device
954 * @alias: name up to IFALIASZ
955 * @len: limit of bytes to copy from info
956 *
957 * Set ifalias for a device,
958 */
959 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
960 {
961 ASSERT_RTNL();
962
963 if (len >= IFALIASZ)
964 return -EINVAL;
965
966 if (!len) {
967 if (dev->ifalias) {
968 kfree(dev->ifalias);
969 dev->ifalias = NULL;
970 }
971 return 0;
972 }
973
974 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
975 if (!dev->ifalias)
976 return -ENOMEM;
977
978 strlcpy(dev->ifalias, alias, len+1);
979 return len;
980 }
981
982
983 /**
984 * netdev_features_change - device changes features
985 * @dev: device to cause notification
986 *
987 * Called to indicate a device has changed features.
988 */
989 void netdev_features_change(struct net_device *dev)
990 {
991 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
992 }
993 EXPORT_SYMBOL(netdev_features_change);
994
995 /**
996 * netdev_state_change - device changes state
997 * @dev: device to cause notification
998 *
999 * Called to indicate a device has changed state. This function calls
1000 * the notifier chains for netdev_chain and sends a NEWLINK message
1001 * to the routing socket.
1002 */
1003 void netdev_state_change(struct net_device *dev)
1004 {
1005 if (dev->flags & IFF_UP) {
1006 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1007 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1008 }
1009 }
1010
1011 void netdev_bonding_change(struct net_device *dev)
1012 {
1013 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1014 }
1015 EXPORT_SYMBOL(netdev_bonding_change);
1016
1017 /**
1018 * dev_load - load a network module
1019 * @net: the applicable net namespace
1020 * @name: name of interface
1021 *
1022 * If a network interface is not present and the process has suitable
1023 * privileges this function loads the module. If module loading is not
1024 * available in this kernel then it becomes a nop.
1025 */
1026
1027 void dev_load(struct net *net, const char *name)
1028 {
1029 struct net_device *dev;
1030
1031 read_lock(&dev_base_lock);
1032 dev = __dev_get_by_name(net, name);
1033 read_unlock(&dev_base_lock);
1034
1035 if (!dev && capable(CAP_SYS_MODULE))
1036 request_module("%s", name);
1037 }
1038
1039 /**
1040 * dev_open - prepare an interface for use.
1041 * @dev: device to open
1042 *
1043 * Takes a device from down to up state. The device's private open
1044 * function is invoked and then the multicast lists are loaded. Finally
1045 * the device is moved into the up state and a %NETDEV_UP message is
1046 * sent to the netdev notifier chain.
1047 *
1048 * Calling this function on an active interface is a nop. On a failure
1049 * a negative errno code is returned.
1050 */
1051 int dev_open(struct net_device *dev)
1052 {
1053 const struct net_device_ops *ops = dev->netdev_ops;
1054 int ret;
1055
1056 ASSERT_RTNL();
1057
1058 /*
1059 * Is it already up?
1060 */
1061
1062 if (dev->flags & IFF_UP)
1063 return 0;
1064
1065 /*
1066 * Is it even present?
1067 */
1068 if (!netif_device_present(dev))
1069 return -ENODEV;
1070
1071 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1072 ret = notifier_to_errno(ret);
1073 if (ret)
1074 return ret;
1075
1076 /*
1077 * Call device private open method
1078 */
1079 set_bit(__LINK_STATE_START, &dev->state);
1080
1081 if (ops->ndo_validate_addr)
1082 ret = ops->ndo_validate_addr(dev);
1083
1084 if (!ret && ops->ndo_open)
1085 ret = ops->ndo_open(dev);
1086
1087 /*
1088 * If it went open OK then:
1089 */
1090
1091 if (ret)
1092 clear_bit(__LINK_STATE_START, &dev->state);
1093 else {
1094 /*
1095 * Set the flags.
1096 */
1097 dev->flags |= IFF_UP;
1098
1099 /*
1100 * Enable NET_DMA
1101 */
1102 net_dmaengine_get();
1103
1104 /*
1105 * Initialize multicasting status
1106 */
1107 dev_set_rx_mode(dev);
1108
1109 /*
1110 * Wakeup transmit queue engine
1111 */
1112 dev_activate(dev);
1113
1114 /*
1115 * ... and announce new interface.
1116 */
1117 call_netdevice_notifiers(NETDEV_UP, dev);
1118 }
1119
1120 return ret;
1121 }
1122
1123 /**
1124 * dev_close - shutdown an interface.
1125 * @dev: device to shutdown
1126 *
1127 * This function moves an active device into down state. A
1128 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1129 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1130 * chain.
1131 */
1132 int dev_close(struct net_device *dev)
1133 {
1134 const struct net_device_ops *ops = dev->netdev_ops;
1135 ASSERT_RTNL();
1136
1137 might_sleep();
1138
1139 if (!(dev->flags & IFF_UP))
1140 return 0;
1141
1142 /*
1143 * Tell people we are going down, so that they can
1144 * prepare to death, when device is still operating.
1145 */
1146 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1147
1148 clear_bit(__LINK_STATE_START, &dev->state);
1149
1150 /* Synchronize to scheduled poll. We cannot touch poll list,
1151 * it can be even on different cpu. So just clear netif_running().
1152 *
1153 * dev->stop() will invoke napi_disable() on all of it's
1154 * napi_struct instances on this device.
1155 */
1156 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1157
1158 dev_deactivate(dev);
1159
1160 /*
1161 * Call the device specific close. This cannot fail.
1162 * Only if device is UP
1163 *
1164 * We allow it to be called even after a DETACH hot-plug
1165 * event.
1166 */
1167 if (ops->ndo_stop)
1168 ops->ndo_stop(dev);
1169
1170 /*
1171 * Device is now down.
1172 */
1173
1174 dev->flags &= ~IFF_UP;
1175
1176 /*
1177 * Tell people we are down
1178 */
1179 call_netdevice_notifiers(NETDEV_DOWN, dev);
1180
1181 /*
1182 * Shutdown NET_DMA
1183 */
1184 net_dmaengine_put();
1185
1186 return 0;
1187 }
1188
1189
1190 /**
1191 * dev_disable_lro - disable Large Receive Offload on a device
1192 * @dev: device
1193 *
1194 * Disable Large Receive Offload (LRO) on a net device. Must be
1195 * called under RTNL. This is needed if received packets may be
1196 * forwarded to another interface.
1197 */
1198 void dev_disable_lro(struct net_device *dev)
1199 {
1200 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1201 dev->ethtool_ops->set_flags) {
1202 u32 flags = dev->ethtool_ops->get_flags(dev);
1203 if (flags & ETH_FLAG_LRO) {
1204 flags &= ~ETH_FLAG_LRO;
1205 dev->ethtool_ops->set_flags(dev, flags);
1206 }
1207 }
1208 WARN_ON(dev->features & NETIF_F_LRO);
1209 }
1210 EXPORT_SYMBOL(dev_disable_lro);
1211
1212
1213 static int dev_boot_phase = 1;
1214
1215 /*
1216 * Device change register/unregister. These are not inline or static
1217 * as we export them to the world.
1218 */
1219
1220 /**
1221 * register_netdevice_notifier - register a network notifier block
1222 * @nb: notifier
1223 *
1224 * Register a notifier to be called when network device events occur.
1225 * The notifier passed is linked into the kernel structures and must
1226 * not be reused until it has been unregistered. A negative errno code
1227 * is returned on a failure.
1228 *
1229 * When registered all registration and up events are replayed
1230 * to the new notifier to allow device to have a race free
1231 * view of the network device list.
1232 */
1233
1234 int register_netdevice_notifier(struct notifier_block *nb)
1235 {
1236 struct net_device *dev;
1237 struct net_device *last;
1238 struct net *net;
1239 int err;
1240
1241 rtnl_lock();
1242 err = raw_notifier_chain_register(&netdev_chain, nb);
1243 if (err)
1244 goto unlock;
1245 if (dev_boot_phase)
1246 goto unlock;
1247 for_each_net(net) {
1248 for_each_netdev(net, dev) {
1249 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1250 err = notifier_to_errno(err);
1251 if (err)
1252 goto rollback;
1253
1254 if (!(dev->flags & IFF_UP))
1255 continue;
1256
1257 nb->notifier_call(nb, NETDEV_UP, dev);
1258 }
1259 }
1260
1261 unlock:
1262 rtnl_unlock();
1263 return err;
1264
1265 rollback:
1266 last = dev;
1267 for_each_net(net) {
1268 for_each_netdev(net, dev) {
1269 if (dev == last)
1270 break;
1271
1272 if (dev->flags & IFF_UP) {
1273 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1274 nb->notifier_call(nb, NETDEV_DOWN, dev);
1275 }
1276 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1277 }
1278 }
1279
1280 raw_notifier_chain_unregister(&netdev_chain, nb);
1281 goto unlock;
1282 }
1283
1284 /**
1285 * unregister_netdevice_notifier - unregister a network notifier block
1286 * @nb: notifier
1287 *
1288 * Unregister a notifier previously registered by
1289 * register_netdevice_notifier(). The notifier is unlinked into the
1290 * kernel structures and may then be reused. A negative errno code
1291 * is returned on a failure.
1292 */
1293
1294 int unregister_netdevice_notifier(struct notifier_block *nb)
1295 {
1296 int err;
1297
1298 rtnl_lock();
1299 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1300 rtnl_unlock();
1301 return err;
1302 }
1303
1304 /**
1305 * call_netdevice_notifiers - call all network notifier blocks
1306 * @val: value passed unmodified to notifier function
1307 * @dev: net_device pointer passed unmodified to notifier function
1308 *
1309 * Call all network notifier blocks. Parameters and return value
1310 * are as for raw_notifier_call_chain().
1311 */
1312
1313 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1314 {
1315 return raw_notifier_call_chain(&netdev_chain, val, dev);
1316 }
1317
1318 /* When > 0 there are consumers of rx skb time stamps */
1319 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1320
1321 void net_enable_timestamp(void)
1322 {
1323 atomic_inc(&netstamp_needed);
1324 }
1325
1326 void net_disable_timestamp(void)
1327 {
1328 atomic_dec(&netstamp_needed);
1329 }
1330
1331 static inline void net_timestamp(struct sk_buff *skb)
1332 {
1333 if (atomic_read(&netstamp_needed))
1334 __net_timestamp(skb);
1335 else
1336 skb->tstamp.tv64 = 0;
1337 }
1338
1339 /*
1340 * Support routine. Sends outgoing frames to any network
1341 * taps currently in use.
1342 */
1343
1344 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1345 {
1346 struct packet_type *ptype;
1347
1348 #ifdef CONFIG_NET_CLS_ACT
1349 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1350 net_timestamp(skb);
1351 #else
1352 net_timestamp(skb);
1353 #endif
1354
1355 rcu_read_lock();
1356 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1357 /* Never send packets back to the socket
1358 * they originated from - MvS (miquels@drinkel.ow.org)
1359 */
1360 if ((ptype->dev == dev || !ptype->dev) &&
1361 (ptype->af_packet_priv == NULL ||
1362 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1363 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1364 if (!skb2)
1365 break;
1366
1367 /* skb->nh should be correctly
1368 set by sender, so that the second statement is
1369 just protection against buggy protocols.
1370 */
1371 skb_reset_mac_header(skb2);
1372
1373 if (skb_network_header(skb2) < skb2->data ||
1374 skb2->network_header > skb2->tail) {
1375 if (net_ratelimit())
1376 printk(KERN_CRIT "protocol %04x is "
1377 "buggy, dev %s\n",
1378 skb2->protocol, dev->name);
1379 skb_reset_network_header(skb2);
1380 }
1381
1382 skb2->transport_header = skb2->network_header;
1383 skb2->pkt_type = PACKET_OUTGOING;
1384 ptype->func(skb2, skb->dev, ptype, skb->dev);
1385 }
1386 }
1387 rcu_read_unlock();
1388 }
1389
1390
1391 static inline void __netif_reschedule(struct Qdisc *q)
1392 {
1393 struct softnet_data *sd;
1394 unsigned long flags;
1395
1396 local_irq_save(flags);
1397 sd = &__get_cpu_var(softnet_data);
1398 q->next_sched = sd->output_queue;
1399 sd->output_queue = q;
1400 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1401 local_irq_restore(flags);
1402 }
1403
1404 void __netif_schedule(struct Qdisc *q)
1405 {
1406 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1407 __netif_reschedule(q);
1408 }
1409 EXPORT_SYMBOL(__netif_schedule);
1410
1411 void dev_kfree_skb_irq(struct sk_buff *skb)
1412 {
1413 if (atomic_dec_and_test(&skb->users)) {
1414 struct softnet_data *sd;
1415 unsigned long flags;
1416
1417 local_irq_save(flags);
1418 sd = &__get_cpu_var(softnet_data);
1419 skb->next = sd->completion_queue;
1420 sd->completion_queue = skb;
1421 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1422 local_irq_restore(flags);
1423 }
1424 }
1425 EXPORT_SYMBOL(dev_kfree_skb_irq);
1426
1427 void dev_kfree_skb_any(struct sk_buff *skb)
1428 {
1429 if (in_irq() || irqs_disabled())
1430 dev_kfree_skb_irq(skb);
1431 else
1432 dev_kfree_skb(skb);
1433 }
1434 EXPORT_SYMBOL(dev_kfree_skb_any);
1435
1436
1437 /**
1438 * netif_device_detach - mark device as removed
1439 * @dev: network device
1440 *
1441 * Mark device as removed from system and therefore no longer available.
1442 */
1443 void netif_device_detach(struct net_device *dev)
1444 {
1445 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1446 netif_running(dev)) {
1447 netif_tx_stop_all_queues(dev);
1448 }
1449 }
1450 EXPORT_SYMBOL(netif_device_detach);
1451
1452 /**
1453 * netif_device_attach - mark device as attached
1454 * @dev: network device
1455 *
1456 * Mark device as attached from system and restart if needed.
1457 */
1458 void netif_device_attach(struct net_device *dev)
1459 {
1460 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1461 netif_running(dev)) {
1462 netif_tx_wake_all_queues(dev);
1463 __netdev_watchdog_up(dev);
1464 }
1465 }
1466 EXPORT_SYMBOL(netif_device_attach);
1467
1468 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1469 {
1470 return ((features & NETIF_F_GEN_CSUM) ||
1471 ((features & NETIF_F_IP_CSUM) &&
1472 protocol == htons(ETH_P_IP)) ||
1473 ((features & NETIF_F_IPV6_CSUM) &&
1474 protocol == htons(ETH_P_IPV6)) ||
1475 ((features & NETIF_F_FCOE_CRC) &&
1476 protocol == htons(ETH_P_FCOE)));
1477 }
1478
1479 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1480 {
1481 if (can_checksum_protocol(dev->features, skb->protocol))
1482 return true;
1483
1484 if (skb->protocol == htons(ETH_P_8021Q)) {
1485 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1486 if (can_checksum_protocol(dev->features & dev->vlan_features,
1487 veh->h_vlan_encapsulated_proto))
1488 return true;
1489 }
1490
1491 return false;
1492 }
1493
1494 /*
1495 * Invalidate hardware checksum when packet is to be mangled, and
1496 * complete checksum manually on outgoing path.
1497 */
1498 int skb_checksum_help(struct sk_buff *skb)
1499 {
1500 __wsum csum;
1501 int ret = 0, offset;
1502
1503 if (skb->ip_summed == CHECKSUM_COMPLETE)
1504 goto out_set_summed;
1505
1506 if (unlikely(skb_shinfo(skb)->gso_size)) {
1507 /* Let GSO fix up the checksum. */
1508 goto out_set_summed;
1509 }
1510
1511 offset = skb->csum_start - skb_headroom(skb);
1512 BUG_ON(offset >= skb_headlen(skb));
1513 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1514
1515 offset += skb->csum_offset;
1516 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1517
1518 if (skb_cloned(skb) &&
1519 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1520 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1521 if (ret)
1522 goto out;
1523 }
1524
1525 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1526 out_set_summed:
1527 skb->ip_summed = CHECKSUM_NONE;
1528 out:
1529 return ret;
1530 }
1531
1532 /**
1533 * skb_gso_segment - Perform segmentation on skb.
1534 * @skb: buffer to segment
1535 * @features: features for the output path (see dev->features)
1536 *
1537 * This function segments the given skb and returns a list of segments.
1538 *
1539 * It may return NULL if the skb requires no segmentation. This is
1540 * only possible when GSO is used for verifying header integrity.
1541 */
1542 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1543 {
1544 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1545 struct packet_type *ptype;
1546 __be16 type = skb->protocol;
1547 int err;
1548
1549 skb_reset_mac_header(skb);
1550 skb->mac_len = skb->network_header - skb->mac_header;
1551 __skb_pull(skb, skb->mac_len);
1552
1553 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1554 struct net_device *dev = skb->dev;
1555 struct ethtool_drvinfo info = {};
1556
1557 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1558 dev->ethtool_ops->get_drvinfo(dev, &info);
1559
1560 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1561 "ip_summed=%d",
1562 info.driver, dev ? dev->features : 0L,
1563 skb->sk ? skb->sk->sk_route_caps : 0L,
1564 skb->len, skb->data_len, skb->ip_summed);
1565
1566 if (skb_header_cloned(skb) &&
1567 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1568 return ERR_PTR(err);
1569 }
1570
1571 rcu_read_lock();
1572 list_for_each_entry_rcu(ptype,
1573 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1574 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1575 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1576 err = ptype->gso_send_check(skb);
1577 segs = ERR_PTR(err);
1578 if (err || skb_gso_ok(skb, features))
1579 break;
1580 __skb_push(skb, (skb->data -
1581 skb_network_header(skb)));
1582 }
1583 segs = ptype->gso_segment(skb, features);
1584 break;
1585 }
1586 }
1587 rcu_read_unlock();
1588
1589 __skb_push(skb, skb->data - skb_mac_header(skb));
1590
1591 return segs;
1592 }
1593
1594 EXPORT_SYMBOL(skb_gso_segment);
1595
1596 /* Take action when hardware reception checksum errors are detected. */
1597 #ifdef CONFIG_BUG
1598 void netdev_rx_csum_fault(struct net_device *dev)
1599 {
1600 if (net_ratelimit()) {
1601 printk(KERN_ERR "%s: hw csum failure.\n",
1602 dev ? dev->name : "<unknown>");
1603 dump_stack();
1604 }
1605 }
1606 EXPORT_SYMBOL(netdev_rx_csum_fault);
1607 #endif
1608
1609 /* Actually, we should eliminate this check as soon as we know, that:
1610 * 1. IOMMU is present and allows to map all the memory.
1611 * 2. No high memory really exists on this machine.
1612 */
1613
1614 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1615 {
1616 #ifdef CONFIG_HIGHMEM
1617 int i;
1618
1619 if (dev->features & NETIF_F_HIGHDMA)
1620 return 0;
1621
1622 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1623 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1624 return 1;
1625
1626 #endif
1627 return 0;
1628 }
1629
1630 struct dev_gso_cb {
1631 void (*destructor)(struct sk_buff *skb);
1632 };
1633
1634 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1635
1636 static void dev_gso_skb_destructor(struct sk_buff *skb)
1637 {
1638 struct dev_gso_cb *cb;
1639
1640 do {
1641 struct sk_buff *nskb = skb->next;
1642
1643 skb->next = nskb->next;
1644 nskb->next = NULL;
1645 kfree_skb(nskb);
1646 } while (skb->next);
1647
1648 cb = DEV_GSO_CB(skb);
1649 if (cb->destructor)
1650 cb->destructor(skb);
1651 }
1652
1653 /**
1654 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1655 * @skb: buffer to segment
1656 *
1657 * This function segments the given skb and stores the list of segments
1658 * in skb->next.
1659 */
1660 static int dev_gso_segment(struct sk_buff *skb)
1661 {
1662 struct net_device *dev = skb->dev;
1663 struct sk_buff *segs;
1664 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1665 NETIF_F_SG : 0);
1666
1667 segs = skb_gso_segment(skb, features);
1668
1669 /* Verifying header integrity only. */
1670 if (!segs)
1671 return 0;
1672
1673 if (IS_ERR(segs))
1674 return PTR_ERR(segs);
1675
1676 skb->next = segs;
1677 DEV_GSO_CB(skb)->destructor = skb->destructor;
1678 skb->destructor = dev_gso_skb_destructor;
1679
1680 return 0;
1681 }
1682
1683 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1684 struct netdev_queue *txq)
1685 {
1686 const struct net_device_ops *ops = dev->netdev_ops;
1687 int rc;
1688
1689 if (likely(!skb->next)) {
1690 if (!list_empty(&ptype_all))
1691 dev_queue_xmit_nit(skb, dev);
1692
1693 if (netif_needs_gso(dev, skb)) {
1694 if (unlikely(dev_gso_segment(skb)))
1695 goto out_kfree_skb;
1696 if (skb->next)
1697 goto gso;
1698 }
1699
1700 /*
1701 * If device doesnt need skb->dst, release it right now while
1702 * its hot in this cpu cache
1703 */
1704 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1705 skb_dst_drop(skb);
1706
1707 rc = ops->ndo_start_xmit(skb, dev);
1708 if (rc == 0)
1709 txq_trans_update(txq);
1710 /*
1711 * TODO: if skb_orphan() was called by
1712 * dev->hard_start_xmit() (for example, the unmodified
1713 * igb driver does that; bnx2 doesn't), then
1714 * skb_tx_software_timestamp() will be unable to send
1715 * back the time stamp.
1716 *
1717 * How can this be prevented? Always create another
1718 * reference to the socket before calling
1719 * dev->hard_start_xmit()? Prevent that skb_orphan()
1720 * does anything in dev->hard_start_xmit() by clearing
1721 * the skb destructor before the call and restoring it
1722 * afterwards, then doing the skb_orphan() ourselves?
1723 */
1724 return rc;
1725 }
1726
1727 gso:
1728 do {
1729 struct sk_buff *nskb = skb->next;
1730
1731 skb->next = nskb->next;
1732 nskb->next = NULL;
1733 rc = ops->ndo_start_xmit(nskb, dev);
1734 if (unlikely(rc)) {
1735 nskb->next = skb->next;
1736 skb->next = nskb;
1737 return rc;
1738 }
1739 txq_trans_update(txq);
1740 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1741 return NETDEV_TX_BUSY;
1742 } while (skb->next);
1743
1744 skb->destructor = DEV_GSO_CB(skb)->destructor;
1745
1746 out_kfree_skb:
1747 kfree_skb(skb);
1748 return 0;
1749 }
1750
1751 static u32 skb_tx_hashrnd;
1752
1753 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1754 {
1755 u32 hash;
1756
1757 if (skb_rx_queue_recorded(skb)) {
1758 hash = skb_get_rx_queue(skb);
1759 while (unlikely (hash >= dev->real_num_tx_queues))
1760 hash -= dev->real_num_tx_queues;
1761 return hash;
1762 }
1763
1764 if (skb->sk && skb->sk->sk_hash)
1765 hash = skb->sk->sk_hash;
1766 else
1767 hash = skb->protocol;
1768
1769 hash = jhash_1word(hash, skb_tx_hashrnd);
1770
1771 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1772 }
1773 EXPORT_SYMBOL(skb_tx_hash);
1774
1775 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1776 struct sk_buff *skb)
1777 {
1778 const struct net_device_ops *ops = dev->netdev_ops;
1779 u16 queue_index = 0;
1780
1781 if (ops->ndo_select_queue)
1782 queue_index = ops->ndo_select_queue(dev, skb);
1783 else if (dev->real_num_tx_queues > 1)
1784 queue_index = skb_tx_hash(dev, skb);
1785
1786 skb_set_queue_mapping(skb, queue_index);
1787 return netdev_get_tx_queue(dev, queue_index);
1788 }
1789
1790 /**
1791 * dev_queue_xmit - transmit a buffer
1792 * @skb: buffer to transmit
1793 *
1794 * Queue a buffer for transmission to a network device. The caller must
1795 * have set the device and priority and built the buffer before calling
1796 * this function. The function can be called from an interrupt.
1797 *
1798 * A negative errno code is returned on a failure. A success does not
1799 * guarantee the frame will be transmitted as it may be dropped due
1800 * to congestion or traffic shaping.
1801 *
1802 * -----------------------------------------------------------------------------------
1803 * I notice this method can also return errors from the queue disciplines,
1804 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1805 * be positive.
1806 *
1807 * Regardless of the return value, the skb is consumed, so it is currently
1808 * difficult to retry a send to this method. (You can bump the ref count
1809 * before sending to hold a reference for retry if you are careful.)
1810 *
1811 * When calling this method, interrupts MUST be enabled. This is because
1812 * the BH enable code must have IRQs enabled so that it will not deadlock.
1813 * --BLG
1814 */
1815 int dev_queue_xmit(struct sk_buff *skb)
1816 {
1817 struct net_device *dev = skb->dev;
1818 struct netdev_queue *txq;
1819 struct Qdisc *q;
1820 int rc = -ENOMEM;
1821
1822 /* GSO will handle the following emulations directly. */
1823 if (netif_needs_gso(dev, skb))
1824 goto gso;
1825
1826 if (skb_has_frags(skb) &&
1827 !(dev->features & NETIF_F_FRAGLIST) &&
1828 __skb_linearize(skb))
1829 goto out_kfree_skb;
1830
1831 /* Fragmented skb is linearized if device does not support SG,
1832 * or if at least one of fragments is in highmem and device
1833 * does not support DMA from it.
1834 */
1835 if (skb_shinfo(skb)->nr_frags &&
1836 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1837 __skb_linearize(skb))
1838 goto out_kfree_skb;
1839
1840 /* If packet is not checksummed and device does not support
1841 * checksumming for this protocol, complete checksumming here.
1842 */
1843 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1844 skb_set_transport_header(skb, skb->csum_start -
1845 skb_headroom(skb));
1846 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1847 goto out_kfree_skb;
1848 }
1849
1850 gso:
1851 /* Disable soft irqs for various locks below. Also
1852 * stops preemption for RCU.
1853 */
1854 rcu_read_lock_bh();
1855
1856 txq = dev_pick_tx(dev, skb);
1857 q = rcu_dereference(txq->qdisc);
1858
1859 #ifdef CONFIG_NET_CLS_ACT
1860 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1861 #endif
1862 if (q->enqueue) {
1863 spinlock_t *root_lock = qdisc_lock(q);
1864
1865 spin_lock(root_lock);
1866
1867 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1868 kfree_skb(skb);
1869 rc = NET_XMIT_DROP;
1870 } else {
1871 rc = qdisc_enqueue_root(skb, q);
1872 qdisc_run(q);
1873 }
1874 spin_unlock(root_lock);
1875
1876 goto out;
1877 }
1878
1879 /* The device has no queue. Common case for software devices:
1880 loopback, all the sorts of tunnels...
1881
1882 Really, it is unlikely that netif_tx_lock protection is necessary
1883 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1884 counters.)
1885 However, it is possible, that they rely on protection
1886 made by us here.
1887
1888 Check this and shot the lock. It is not prone from deadlocks.
1889 Either shot noqueue qdisc, it is even simpler 8)
1890 */
1891 if (dev->flags & IFF_UP) {
1892 int cpu = smp_processor_id(); /* ok because BHs are off */
1893
1894 if (txq->xmit_lock_owner != cpu) {
1895
1896 HARD_TX_LOCK(dev, txq, cpu);
1897
1898 if (!netif_tx_queue_stopped(txq)) {
1899 rc = 0;
1900 if (!dev_hard_start_xmit(skb, dev, txq)) {
1901 HARD_TX_UNLOCK(dev, txq);
1902 goto out;
1903 }
1904 }
1905 HARD_TX_UNLOCK(dev, txq);
1906 if (net_ratelimit())
1907 printk(KERN_CRIT "Virtual device %s asks to "
1908 "queue packet!\n", dev->name);
1909 } else {
1910 /* Recursion is detected! It is possible,
1911 * unfortunately */
1912 if (net_ratelimit())
1913 printk(KERN_CRIT "Dead loop on virtual device "
1914 "%s, fix it urgently!\n", dev->name);
1915 }
1916 }
1917
1918 rc = -ENETDOWN;
1919 rcu_read_unlock_bh();
1920
1921 out_kfree_skb:
1922 kfree_skb(skb);
1923 return rc;
1924 out:
1925 rcu_read_unlock_bh();
1926 return rc;
1927 }
1928
1929
1930 /*=======================================================================
1931 Receiver routines
1932 =======================================================================*/
1933
1934 int netdev_max_backlog __read_mostly = 1000;
1935 int netdev_budget __read_mostly = 300;
1936 int weight_p __read_mostly = 64; /* old backlog weight */
1937
1938 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1939
1940
1941 /**
1942 * netif_rx - post buffer to the network code
1943 * @skb: buffer to post
1944 *
1945 * This function receives a packet from a device driver and queues it for
1946 * the upper (protocol) levels to process. It always succeeds. The buffer
1947 * may be dropped during processing for congestion control or by the
1948 * protocol layers.
1949 *
1950 * return values:
1951 * NET_RX_SUCCESS (no congestion)
1952 * NET_RX_DROP (packet was dropped)
1953 *
1954 */
1955
1956 int netif_rx(struct sk_buff *skb)
1957 {
1958 struct softnet_data *queue;
1959 unsigned long flags;
1960
1961 /* if netpoll wants it, pretend we never saw it */
1962 if (netpoll_rx(skb))
1963 return NET_RX_DROP;
1964
1965 if (!skb->tstamp.tv64)
1966 net_timestamp(skb);
1967
1968 /*
1969 * The code is rearranged so that the path is the most
1970 * short when CPU is congested, but is still operating.
1971 */
1972 local_irq_save(flags);
1973 queue = &__get_cpu_var(softnet_data);
1974
1975 __get_cpu_var(netdev_rx_stat).total++;
1976 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1977 if (queue->input_pkt_queue.qlen) {
1978 enqueue:
1979 __skb_queue_tail(&queue->input_pkt_queue, skb);
1980 local_irq_restore(flags);
1981 return NET_RX_SUCCESS;
1982 }
1983
1984 napi_schedule(&queue->backlog);
1985 goto enqueue;
1986 }
1987
1988 __get_cpu_var(netdev_rx_stat).dropped++;
1989 local_irq_restore(flags);
1990
1991 kfree_skb(skb);
1992 return NET_RX_DROP;
1993 }
1994
1995 int netif_rx_ni(struct sk_buff *skb)
1996 {
1997 int err;
1998
1999 preempt_disable();
2000 err = netif_rx(skb);
2001 if (local_softirq_pending())
2002 do_softirq();
2003 preempt_enable();
2004
2005 return err;
2006 }
2007
2008 EXPORT_SYMBOL(netif_rx_ni);
2009
2010 static void net_tx_action(struct softirq_action *h)
2011 {
2012 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2013
2014 if (sd->completion_queue) {
2015 struct sk_buff *clist;
2016
2017 local_irq_disable();
2018 clist = sd->completion_queue;
2019 sd->completion_queue = NULL;
2020 local_irq_enable();
2021
2022 while (clist) {
2023 struct sk_buff *skb = clist;
2024 clist = clist->next;
2025
2026 WARN_ON(atomic_read(&skb->users));
2027 __kfree_skb(skb);
2028 }
2029 }
2030
2031 if (sd->output_queue) {
2032 struct Qdisc *head;
2033
2034 local_irq_disable();
2035 head = sd->output_queue;
2036 sd->output_queue = NULL;
2037 local_irq_enable();
2038
2039 while (head) {
2040 struct Qdisc *q = head;
2041 spinlock_t *root_lock;
2042
2043 head = head->next_sched;
2044
2045 root_lock = qdisc_lock(q);
2046 if (spin_trylock(root_lock)) {
2047 smp_mb__before_clear_bit();
2048 clear_bit(__QDISC_STATE_SCHED,
2049 &q->state);
2050 qdisc_run(q);
2051 spin_unlock(root_lock);
2052 } else {
2053 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2054 &q->state)) {
2055 __netif_reschedule(q);
2056 } else {
2057 smp_mb__before_clear_bit();
2058 clear_bit(__QDISC_STATE_SCHED,
2059 &q->state);
2060 }
2061 }
2062 }
2063 }
2064 }
2065
2066 static inline int deliver_skb(struct sk_buff *skb,
2067 struct packet_type *pt_prev,
2068 struct net_device *orig_dev)
2069 {
2070 atomic_inc(&skb->users);
2071 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2072 }
2073
2074 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2075
2076 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2077 /* This hook is defined here for ATM LANE */
2078 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2079 unsigned char *addr) __read_mostly;
2080 EXPORT_SYMBOL(br_fdb_test_addr_hook);
2081 #endif
2082
2083 /*
2084 * If bridge module is loaded call bridging hook.
2085 * returns NULL if packet was consumed.
2086 */
2087 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2088 struct sk_buff *skb) __read_mostly;
2089 EXPORT_SYMBOL(br_handle_frame_hook);
2090
2091 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2092 struct packet_type **pt_prev, int *ret,
2093 struct net_device *orig_dev)
2094 {
2095 struct net_bridge_port *port;
2096
2097 if (skb->pkt_type == PACKET_LOOPBACK ||
2098 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2099 return skb;
2100
2101 if (*pt_prev) {
2102 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2103 *pt_prev = NULL;
2104 }
2105
2106 return br_handle_frame_hook(port, skb);
2107 }
2108 #else
2109 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2110 #endif
2111
2112 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2113 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2114 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2115
2116 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2117 struct packet_type **pt_prev,
2118 int *ret,
2119 struct net_device *orig_dev)
2120 {
2121 if (skb->dev->macvlan_port == NULL)
2122 return skb;
2123
2124 if (*pt_prev) {
2125 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2126 *pt_prev = NULL;
2127 }
2128 return macvlan_handle_frame_hook(skb);
2129 }
2130 #else
2131 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2132 #endif
2133
2134 #ifdef CONFIG_NET_CLS_ACT
2135 /* TODO: Maybe we should just force sch_ingress to be compiled in
2136 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2137 * a compare and 2 stores extra right now if we dont have it on
2138 * but have CONFIG_NET_CLS_ACT
2139 * NOTE: This doesnt stop any functionality; if you dont have
2140 * the ingress scheduler, you just cant add policies on ingress.
2141 *
2142 */
2143 static int ing_filter(struct sk_buff *skb)
2144 {
2145 struct net_device *dev = skb->dev;
2146 u32 ttl = G_TC_RTTL(skb->tc_verd);
2147 struct netdev_queue *rxq;
2148 int result = TC_ACT_OK;
2149 struct Qdisc *q;
2150
2151 if (MAX_RED_LOOP < ttl++) {
2152 printk(KERN_WARNING
2153 "Redir loop detected Dropping packet (%d->%d)\n",
2154 skb->iif, dev->ifindex);
2155 return TC_ACT_SHOT;
2156 }
2157
2158 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2159 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2160
2161 rxq = &dev->rx_queue;
2162
2163 q = rxq->qdisc;
2164 if (q != &noop_qdisc) {
2165 spin_lock(qdisc_lock(q));
2166 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2167 result = qdisc_enqueue_root(skb, q);
2168 spin_unlock(qdisc_lock(q));
2169 }
2170
2171 return result;
2172 }
2173
2174 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2175 struct packet_type **pt_prev,
2176 int *ret, struct net_device *orig_dev)
2177 {
2178 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2179 goto out;
2180
2181 if (*pt_prev) {
2182 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2183 *pt_prev = NULL;
2184 } else {
2185 /* Huh? Why does turning on AF_PACKET affect this? */
2186 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2187 }
2188
2189 switch (ing_filter(skb)) {
2190 case TC_ACT_SHOT:
2191 case TC_ACT_STOLEN:
2192 kfree_skb(skb);
2193 return NULL;
2194 }
2195
2196 out:
2197 skb->tc_verd = 0;
2198 return skb;
2199 }
2200 #endif
2201
2202 /*
2203 * netif_nit_deliver - deliver received packets to network taps
2204 * @skb: buffer
2205 *
2206 * This function is used to deliver incoming packets to network
2207 * taps. It should be used when the normal netif_receive_skb path
2208 * is bypassed, for example because of VLAN acceleration.
2209 */
2210 void netif_nit_deliver(struct sk_buff *skb)
2211 {
2212 struct packet_type *ptype;
2213
2214 if (list_empty(&ptype_all))
2215 return;
2216
2217 skb_reset_network_header(skb);
2218 skb_reset_transport_header(skb);
2219 skb->mac_len = skb->network_header - skb->mac_header;
2220
2221 rcu_read_lock();
2222 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2223 if (!ptype->dev || ptype->dev == skb->dev)
2224 deliver_skb(skb, ptype, skb->dev);
2225 }
2226 rcu_read_unlock();
2227 }
2228
2229 /**
2230 * netif_receive_skb - process receive buffer from network
2231 * @skb: buffer to process
2232 *
2233 * netif_receive_skb() is the main receive data processing function.
2234 * It always succeeds. The buffer may be dropped during processing
2235 * for congestion control or by the protocol layers.
2236 *
2237 * This function may only be called from softirq context and interrupts
2238 * should be enabled.
2239 *
2240 * Return values (usually ignored):
2241 * NET_RX_SUCCESS: no congestion
2242 * NET_RX_DROP: packet was dropped
2243 */
2244 int netif_receive_skb(struct sk_buff *skb)
2245 {
2246 struct packet_type *ptype, *pt_prev;
2247 struct net_device *orig_dev;
2248 struct net_device *null_or_orig;
2249 int ret = NET_RX_DROP;
2250 __be16 type;
2251
2252 if (!skb->tstamp.tv64)
2253 net_timestamp(skb);
2254
2255 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2256 return NET_RX_SUCCESS;
2257
2258 /* if we've gotten here through NAPI, check netpoll */
2259 if (netpoll_receive_skb(skb))
2260 return NET_RX_DROP;
2261
2262 if (!skb->iif)
2263 skb->iif = skb->dev->ifindex;
2264
2265 null_or_orig = NULL;
2266 orig_dev = skb->dev;
2267 if (orig_dev->master) {
2268 if (skb_bond_should_drop(skb))
2269 null_or_orig = orig_dev; /* deliver only exact match */
2270 else
2271 skb->dev = orig_dev->master;
2272 }
2273
2274 __get_cpu_var(netdev_rx_stat).total++;
2275
2276 skb_reset_network_header(skb);
2277 skb_reset_transport_header(skb);
2278 skb->mac_len = skb->network_header - skb->mac_header;
2279
2280 pt_prev = NULL;
2281
2282 rcu_read_lock();
2283
2284 #ifdef CONFIG_NET_CLS_ACT
2285 if (skb->tc_verd & TC_NCLS) {
2286 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2287 goto ncls;
2288 }
2289 #endif
2290
2291 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2292 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2293 ptype->dev == orig_dev) {
2294 if (pt_prev)
2295 ret = deliver_skb(skb, pt_prev, orig_dev);
2296 pt_prev = ptype;
2297 }
2298 }
2299
2300 #ifdef CONFIG_NET_CLS_ACT
2301 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2302 if (!skb)
2303 goto out;
2304 ncls:
2305 #endif
2306
2307 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2308 if (!skb)
2309 goto out;
2310 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2311 if (!skb)
2312 goto out;
2313
2314 type = skb->protocol;
2315 list_for_each_entry_rcu(ptype,
2316 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2317 if (ptype->type == type &&
2318 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2319 ptype->dev == orig_dev)) {
2320 if (pt_prev)
2321 ret = deliver_skb(skb, pt_prev, orig_dev);
2322 pt_prev = ptype;
2323 }
2324 }
2325
2326 if (pt_prev) {
2327 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2328 } else {
2329 kfree_skb(skb);
2330 /* Jamal, now you will not able to escape explaining
2331 * me how you were going to use this. :-)
2332 */
2333 ret = NET_RX_DROP;
2334 }
2335
2336 out:
2337 rcu_read_unlock();
2338 return ret;
2339 }
2340
2341 /* Network device is going away, flush any packets still pending */
2342 static void flush_backlog(void *arg)
2343 {
2344 struct net_device *dev = arg;
2345 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2346 struct sk_buff *skb, *tmp;
2347
2348 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2349 if (skb->dev == dev) {
2350 __skb_unlink(skb, &queue->input_pkt_queue);
2351 kfree_skb(skb);
2352 }
2353 }
2354
2355 static int napi_gro_complete(struct sk_buff *skb)
2356 {
2357 struct packet_type *ptype;
2358 __be16 type = skb->protocol;
2359 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2360 int err = -ENOENT;
2361
2362 if (NAPI_GRO_CB(skb)->count == 1) {
2363 skb_shinfo(skb)->gso_size = 0;
2364 goto out;
2365 }
2366
2367 rcu_read_lock();
2368 list_for_each_entry_rcu(ptype, head, list) {
2369 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2370 continue;
2371
2372 err = ptype->gro_complete(skb);
2373 break;
2374 }
2375 rcu_read_unlock();
2376
2377 if (err) {
2378 WARN_ON(&ptype->list == head);
2379 kfree_skb(skb);
2380 return NET_RX_SUCCESS;
2381 }
2382
2383 out:
2384 return netif_receive_skb(skb);
2385 }
2386
2387 void napi_gro_flush(struct napi_struct *napi)
2388 {
2389 struct sk_buff *skb, *next;
2390
2391 for (skb = napi->gro_list; skb; skb = next) {
2392 next = skb->next;
2393 skb->next = NULL;
2394 napi_gro_complete(skb);
2395 }
2396
2397 napi->gro_count = 0;
2398 napi->gro_list = NULL;
2399 }
2400 EXPORT_SYMBOL(napi_gro_flush);
2401
2402 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2403 {
2404 struct sk_buff **pp = NULL;
2405 struct packet_type *ptype;
2406 __be16 type = skb->protocol;
2407 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2408 int same_flow;
2409 int mac_len;
2410 int ret;
2411
2412 if (!(skb->dev->features & NETIF_F_GRO))
2413 goto normal;
2414
2415 if (skb_is_gso(skb) || skb_has_frags(skb))
2416 goto normal;
2417
2418 rcu_read_lock();
2419 list_for_each_entry_rcu(ptype, head, list) {
2420 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2421 continue;
2422
2423 skb_set_network_header(skb, skb_gro_offset(skb));
2424 mac_len = skb->network_header - skb->mac_header;
2425 skb->mac_len = mac_len;
2426 NAPI_GRO_CB(skb)->same_flow = 0;
2427 NAPI_GRO_CB(skb)->flush = 0;
2428 NAPI_GRO_CB(skb)->free = 0;
2429
2430 pp = ptype->gro_receive(&napi->gro_list, skb);
2431 break;
2432 }
2433 rcu_read_unlock();
2434
2435 if (&ptype->list == head)
2436 goto normal;
2437
2438 same_flow = NAPI_GRO_CB(skb)->same_flow;
2439 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2440
2441 if (pp) {
2442 struct sk_buff *nskb = *pp;
2443
2444 *pp = nskb->next;
2445 nskb->next = NULL;
2446 napi_gro_complete(nskb);
2447 napi->gro_count--;
2448 }
2449
2450 if (same_flow)
2451 goto ok;
2452
2453 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2454 goto normal;
2455
2456 napi->gro_count++;
2457 NAPI_GRO_CB(skb)->count = 1;
2458 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2459 skb->next = napi->gro_list;
2460 napi->gro_list = skb;
2461 ret = GRO_HELD;
2462
2463 pull:
2464 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2465 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2466
2467 BUG_ON(skb->end - skb->tail < grow);
2468
2469 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2470
2471 skb->tail += grow;
2472 skb->data_len -= grow;
2473
2474 skb_shinfo(skb)->frags[0].page_offset += grow;
2475 skb_shinfo(skb)->frags[0].size -= grow;
2476
2477 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2478 put_page(skb_shinfo(skb)->frags[0].page);
2479 memmove(skb_shinfo(skb)->frags,
2480 skb_shinfo(skb)->frags + 1,
2481 --skb_shinfo(skb)->nr_frags);
2482 }
2483 }
2484
2485 ok:
2486 return ret;
2487
2488 normal:
2489 ret = GRO_NORMAL;
2490 goto pull;
2491 }
2492 EXPORT_SYMBOL(dev_gro_receive);
2493
2494 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2495 {
2496 struct sk_buff *p;
2497
2498 if (netpoll_rx_on(skb))
2499 return GRO_NORMAL;
2500
2501 for (p = napi->gro_list; p; p = p->next) {
2502 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2503 && !compare_ether_header(skb_mac_header(p),
2504 skb_gro_mac_header(skb));
2505 NAPI_GRO_CB(p)->flush = 0;
2506 }
2507
2508 return dev_gro_receive(napi, skb);
2509 }
2510
2511 int napi_skb_finish(int ret, struct sk_buff *skb)
2512 {
2513 int err = NET_RX_SUCCESS;
2514
2515 switch (ret) {
2516 case GRO_NORMAL:
2517 return netif_receive_skb(skb);
2518
2519 case GRO_DROP:
2520 err = NET_RX_DROP;
2521 /* fall through */
2522
2523 case GRO_MERGED_FREE:
2524 kfree_skb(skb);
2525 break;
2526 }
2527
2528 return err;
2529 }
2530 EXPORT_SYMBOL(napi_skb_finish);
2531
2532 void skb_gro_reset_offset(struct sk_buff *skb)
2533 {
2534 NAPI_GRO_CB(skb)->data_offset = 0;
2535 NAPI_GRO_CB(skb)->frag0 = NULL;
2536 NAPI_GRO_CB(skb)->frag0_len = 0;
2537
2538 if (skb->mac_header == skb->tail &&
2539 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2540 NAPI_GRO_CB(skb)->frag0 =
2541 page_address(skb_shinfo(skb)->frags[0].page) +
2542 skb_shinfo(skb)->frags[0].page_offset;
2543 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2544 }
2545 }
2546 EXPORT_SYMBOL(skb_gro_reset_offset);
2547
2548 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2549 {
2550 skb_gro_reset_offset(skb);
2551
2552 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2553 }
2554 EXPORT_SYMBOL(napi_gro_receive);
2555
2556 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2557 {
2558 __skb_pull(skb, skb_headlen(skb));
2559 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2560
2561 napi->skb = skb;
2562 }
2563 EXPORT_SYMBOL(napi_reuse_skb);
2564
2565 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2566 {
2567 struct net_device *dev = napi->dev;
2568 struct sk_buff *skb = napi->skb;
2569
2570 if (!skb) {
2571 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2572 if (!skb)
2573 goto out;
2574
2575 skb_reserve(skb, NET_IP_ALIGN);
2576
2577 napi->skb = skb;
2578 }
2579
2580 out:
2581 return skb;
2582 }
2583 EXPORT_SYMBOL(napi_get_frags);
2584
2585 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2586 {
2587 int err = NET_RX_SUCCESS;
2588
2589 switch (ret) {
2590 case GRO_NORMAL:
2591 case GRO_HELD:
2592 skb->protocol = eth_type_trans(skb, napi->dev);
2593
2594 if (ret == GRO_NORMAL)
2595 return netif_receive_skb(skb);
2596
2597 skb_gro_pull(skb, -ETH_HLEN);
2598 break;
2599
2600 case GRO_DROP:
2601 err = NET_RX_DROP;
2602 /* fall through */
2603
2604 case GRO_MERGED_FREE:
2605 napi_reuse_skb(napi, skb);
2606 break;
2607 }
2608
2609 return err;
2610 }
2611 EXPORT_SYMBOL(napi_frags_finish);
2612
2613 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2614 {
2615 struct sk_buff *skb = napi->skb;
2616 struct ethhdr *eth;
2617 unsigned int hlen;
2618 unsigned int off;
2619
2620 napi->skb = NULL;
2621
2622 skb_reset_mac_header(skb);
2623 skb_gro_reset_offset(skb);
2624
2625 off = skb_gro_offset(skb);
2626 hlen = off + sizeof(*eth);
2627 eth = skb_gro_header_fast(skb, off);
2628 if (skb_gro_header_hard(skb, hlen)) {
2629 eth = skb_gro_header_slow(skb, hlen, off);
2630 if (unlikely(!eth)) {
2631 napi_reuse_skb(napi, skb);
2632 skb = NULL;
2633 goto out;
2634 }
2635 }
2636
2637 skb_gro_pull(skb, sizeof(*eth));
2638
2639 /*
2640 * This works because the only protocols we care about don't require
2641 * special handling. We'll fix it up properly at the end.
2642 */
2643 skb->protocol = eth->h_proto;
2644
2645 out:
2646 return skb;
2647 }
2648 EXPORT_SYMBOL(napi_frags_skb);
2649
2650 int napi_gro_frags(struct napi_struct *napi)
2651 {
2652 struct sk_buff *skb = napi_frags_skb(napi);
2653
2654 if (!skb)
2655 return NET_RX_DROP;
2656
2657 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2658 }
2659 EXPORT_SYMBOL(napi_gro_frags);
2660
2661 static int process_backlog(struct napi_struct *napi, int quota)
2662 {
2663 int work = 0;
2664 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2665 unsigned long start_time = jiffies;
2666
2667 napi->weight = weight_p;
2668 do {
2669 struct sk_buff *skb;
2670
2671 local_irq_disable();
2672 skb = __skb_dequeue(&queue->input_pkt_queue);
2673 if (!skb) {
2674 __napi_complete(napi);
2675 local_irq_enable();
2676 break;
2677 }
2678 local_irq_enable();
2679
2680 netif_receive_skb(skb);
2681 } while (++work < quota && jiffies == start_time);
2682
2683 return work;
2684 }
2685
2686 /**
2687 * __napi_schedule - schedule for receive
2688 * @n: entry to schedule
2689 *
2690 * The entry's receive function will be scheduled to run
2691 */
2692 void __napi_schedule(struct napi_struct *n)
2693 {
2694 unsigned long flags;
2695
2696 local_irq_save(flags);
2697 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2698 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2699 local_irq_restore(flags);
2700 }
2701 EXPORT_SYMBOL(__napi_schedule);
2702
2703 void __napi_complete(struct napi_struct *n)
2704 {
2705 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2706 BUG_ON(n->gro_list);
2707
2708 list_del(&n->poll_list);
2709 smp_mb__before_clear_bit();
2710 clear_bit(NAPI_STATE_SCHED, &n->state);
2711 }
2712 EXPORT_SYMBOL(__napi_complete);
2713
2714 void napi_complete(struct napi_struct *n)
2715 {
2716 unsigned long flags;
2717
2718 /*
2719 * don't let napi dequeue from the cpu poll list
2720 * just in case its running on a different cpu
2721 */
2722 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2723 return;
2724
2725 napi_gro_flush(n);
2726 local_irq_save(flags);
2727 __napi_complete(n);
2728 local_irq_restore(flags);
2729 }
2730 EXPORT_SYMBOL(napi_complete);
2731
2732 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2733 int (*poll)(struct napi_struct *, int), int weight)
2734 {
2735 INIT_LIST_HEAD(&napi->poll_list);
2736 napi->gro_count = 0;
2737 napi->gro_list = NULL;
2738 napi->skb = NULL;
2739 napi->poll = poll;
2740 napi->weight = weight;
2741 list_add(&napi->dev_list, &dev->napi_list);
2742 napi->dev = dev;
2743 #ifdef CONFIG_NETPOLL
2744 spin_lock_init(&napi->poll_lock);
2745 napi->poll_owner = -1;
2746 #endif
2747 set_bit(NAPI_STATE_SCHED, &napi->state);
2748 }
2749 EXPORT_SYMBOL(netif_napi_add);
2750
2751 void netif_napi_del(struct napi_struct *napi)
2752 {
2753 struct sk_buff *skb, *next;
2754
2755 list_del_init(&napi->dev_list);
2756 napi_free_frags(napi);
2757
2758 for (skb = napi->gro_list; skb; skb = next) {
2759 next = skb->next;
2760 skb->next = NULL;
2761 kfree_skb(skb);
2762 }
2763
2764 napi->gro_list = NULL;
2765 napi->gro_count = 0;
2766 }
2767 EXPORT_SYMBOL(netif_napi_del);
2768
2769
2770 static void net_rx_action(struct softirq_action *h)
2771 {
2772 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2773 unsigned long time_limit = jiffies + 2;
2774 int budget = netdev_budget;
2775 void *have;
2776
2777 local_irq_disable();
2778
2779 while (!list_empty(list)) {
2780 struct napi_struct *n;
2781 int work, weight;
2782
2783 /* If softirq window is exhuasted then punt.
2784 * Allow this to run for 2 jiffies since which will allow
2785 * an average latency of 1.5/HZ.
2786 */
2787 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2788 goto softnet_break;
2789
2790 local_irq_enable();
2791
2792 /* Even though interrupts have been re-enabled, this
2793 * access is safe because interrupts can only add new
2794 * entries to the tail of this list, and only ->poll()
2795 * calls can remove this head entry from the list.
2796 */
2797 n = list_entry(list->next, struct napi_struct, poll_list);
2798
2799 have = netpoll_poll_lock(n);
2800
2801 weight = n->weight;
2802
2803 /* This NAPI_STATE_SCHED test is for avoiding a race
2804 * with netpoll's poll_napi(). Only the entity which
2805 * obtains the lock and sees NAPI_STATE_SCHED set will
2806 * actually make the ->poll() call. Therefore we avoid
2807 * accidently calling ->poll() when NAPI is not scheduled.
2808 */
2809 work = 0;
2810 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2811 work = n->poll(n, weight);
2812 trace_napi_poll(n);
2813 }
2814
2815 WARN_ON_ONCE(work > weight);
2816
2817 budget -= work;
2818
2819 local_irq_disable();
2820
2821 /* Drivers must not modify the NAPI state if they
2822 * consume the entire weight. In such cases this code
2823 * still "owns" the NAPI instance and therefore can
2824 * move the instance around on the list at-will.
2825 */
2826 if (unlikely(work == weight)) {
2827 if (unlikely(napi_disable_pending(n))) {
2828 local_irq_enable();
2829 napi_complete(n);
2830 local_irq_disable();
2831 } else
2832 list_move_tail(&n->poll_list, list);
2833 }
2834
2835 netpoll_poll_unlock(have);
2836 }
2837 out:
2838 local_irq_enable();
2839
2840 #ifdef CONFIG_NET_DMA
2841 /*
2842 * There may not be any more sk_buffs coming right now, so push
2843 * any pending DMA copies to hardware
2844 */
2845 dma_issue_pending_all();
2846 #endif
2847
2848 return;
2849
2850 softnet_break:
2851 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2852 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2853 goto out;
2854 }
2855
2856 static gifconf_func_t * gifconf_list [NPROTO];
2857
2858 /**
2859 * register_gifconf - register a SIOCGIF handler
2860 * @family: Address family
2861 * @gifconf: Function handler
2862 *
2863 * Register protocol dependent address dumping routines. The handler
2864 * that is passed must not be freed or reused until it has been replaced
2865 * by another handler.
2866 */
2867 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2868 {
2869 if (family >= NPROTO)
2870 return -EINVAL;
2871 gifconf_list[family] = gifconf;
2872 return 0;
2873 }
2874
2875
2876 /*
2877 * Map an interface index to its name (SIOCGIFNAME)
2878 */
2879
2880 /*
2881 * We need this ioctl for efficient implementation of the
2882 * if_indextoname() function required by the IPv6 API. Without
2883 * it, we would have to search all the interfaces to find a
2884 * match. --pb
2885 */
2886
2887 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2888 {
2889 struct net_device *dev;
2890 struct ifreq ifr;
2891
2892 /*
2893 * Fetch the caller's info block.
2894 */
2895
2896 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2897 return -EFAULT;
2898
2899 read_lock(&dev_base_lock);
2900 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2901 if (!dev) {
2902 read_unlock(&dev_base_lock);
2903 return -ENODEV;
2904 }
2905
2906 strcpy(ifr.ifr_name, dev->name);
2907 read_unlock(&dev_base_lock);
2908
2909 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2910 return -EFAULT;
2911 return 0;
2912 }
2913
2914 /*
2915 * Perform a SIOCGIFCONF call. This structure will change
2916 * size eventually, and there is nothing I can do about it.
2917 * Thus we will need a 'compatibility mode'.
2918 */
2919
2920 static int dev_ifconf(struct net *net, char __user *arg)
2921 {
2922 struct ifconf ifc;
2923 struct net_device *dev;
2924 char __user *pos;
2925 int len;
2926 int total;
2927 int i;
2928
2929 /*
2930 * Fetch the caller's info block.
2931 */
2932
2933 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2934 return -EFAULT;
2935
2936 pos = ifc.ifc_buf;
2937 len = ifc.ifc_len;
2938
2939 /*
2940 * Loop over the interfaces, and write an info block for each.
2941 */
2942
2943 total = 0;
2944 for_each_netdev(net, dev) {
2945 for (i = 0; i < NPROTO; i++) {
2946 if (gifconf_list[i]) {
2947 int done;
2948 if (!pos)
2949 done = gifconf_list[i](dev, NULL, 0);
2950 else
2951 done = gifconf_list[i](dev, pos + total,
2952 len - total);
2953 if (done < 0)
2954 return -EFAULT;
2955 total += done;
2956 }
2957 }
2958 }
2959
2960 /*
2961 * All done. Write the updated control block back to the caller.
2962 */
2963 ifc.ifc_len = total;
2964
2965 /*
2966 * Both BSD and Solaris return 0 here, so we do too.
2967 */
2968 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2969 }
2970
2971 #ifdef CONFIG_PROC_FS
2972 /*
2973 * This is invoked by the /proc filesystem handler to display a device
2974 * in detail.
2975 */
2976 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2977 __acquires(dev_base_lock)
2978 {
2979 struct net *net = seq_file_net(seq);
2980 loff_t off;
2981 struct net_device *dev;
2982
2983 read_lock(&dev_base_lock);
2984 if (!*pos)
2985 return SEQ_START_TOKEN;
2986
2987 off = 1;
2988 for_each_netdev(net, dev)
2989 if (off++ == *pos)
2990 return dev;
2991
2992 return NULL;
2993 }
2994
2995 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2996 {
2997 struct net *net = seq_file_net(seq);
2998 ++*pos;
2999 return v == SEQ_START_TOKEN ?
3000 first_net_device(net) : next_net_device((struct net_device *)v);
3001 }
3002
3003 void dev_seq_stop(struct seq_file *seq, void *v)
3004 __releases(dev_base_lock)
3005 {
3006 read_unlock(&dev_base_lock);
3007 }
3008
3009 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3010 {
3011 const struct net_device_stats *stats = dev_get_stats(dev);
3012
3013 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3014 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3015 dev->name, stats->rx_bytes, stats->rx_packets,
3016 stats->rx_errors,
3017 stats->rx_dropped + stats->rx_missed_errors,
3018 stats->rx_fifo_errors,
3019 stats->rx_length_errors + stats->rx_over_errors +
3020 stats->rx_crc_errors + stats->rx_frame_errors,
3021 stats->rx_compressed, stats->multicast,
3022 stats->tx_bytes, stats->tx_packets,
3023 stats->tx_errors, stats->tx_dropped,
3024 stats->tx_fifo_errors, stats->collisions,
3025 stats->tx_carrier_errors +
3026 stats->tx_aborted_errors +
3027 stats->tx_window_errors +
3028 stats->tx_heartbeat_errors,
3029 stats->tx_compressed);
3030 }
3031
3032 /*
3033 * Called from the PROCfs module. This now uses the new arbitrary sized
3034 * /proc/net interface to create /proc/net/dev
3035 */
3036 static int dev_seq_show(struct seq_file *seq, void *v)
3037 {
3038 if (v == SEQ_START_TOKEN)
3039 seq_puts(seq, "Inter-| Receive "
3040 " | Transmit\n"
3041 " face |bytes packets errs drop fifo frame "
3042 "compressed multicast|bytes packets errs "
3043 "drop fifo colls carrier compressed\n");
3044 else
3045 dev_seq_printf_stats(seq, v);
3046 return 0;
3047 }
3048
3049 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3050 {
3051 struct netif_rx_stats *rc = NULL;
3052
3053 while (*pos < nr_cpu_ids)
3054 if (cpu_online(*pos)) {
3055 rc = &per_cpu(netdev_rx_stat, *pos);
3056 break;
3057 } else
3058 ++*pos;
3059 return rc;
3060 }
3061
3062 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3063 {
3064 return softnet_get_online(pos);
3065 }
3066
3067 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3068 {
3069 ++*pos;
3070 return softnet_get_online(pos);
3071 }
3072
3073 static void softnet_seq_stop(struct seq_file *seq, void *v)
3074 {
3075 }
3076
3077 static int softnet_seq_show(struct seq_file *seq, void *v)
3078 {
3079 struct netif_rx_stats *s = v;
3080
3081 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3082 s->total, s->dropped, s->time_squeeze, 0,
3083 0, 0, 0, 0, /* was fastroute */
3084 s->cpu_collision );
3085 return 0;
3086 }
3087
3088 static const struct seq_operations dev_seq_ops = {
3089 .start = dev_seq_start,
3090 .next = dev_seq_next,
3091 .stop = dev_seq_stop,
3092 .show = dev_seq_show,
3093 };
3094
3095 static int dev_seq_open(struct inode *inode, struct file *file)
3096 {
3097 return seq_open_net(inode, file, &dev_seq_ops,
3098 sizeof(struct seq_net_private));
3099 }
3100
3101 static const struct file_operations dev_seq_fops = {
3102 .owner = THIS_MODULE,
3103 .open = dev_seq_open,
3104 .read = seq_read,
3105 .llseek = seq_lseek,
3106 .release = seq_release_net,
3107 };
3108
3109 static const struct seq_operations softnet_seq_ops = {
3110 .start = softnet_seq_start,
3111 .next = softnet_seq_next,
3112 .stop = softnet_seq_stop,
3113 .show = softnet_seq_show,
3114 };
3115
3116 static int softnet_seq_open(struct inode *inode, struct file *file)
3117 {
3118 return seq_open(file, &softnet_seq_ops);
3119 }
3120
3121 static const struct file_operations softnet_seq_fops = {
3122 .owner = THIS_MODULE,
3123 .open = softnet_seq_open,
3124 .read = seq_read,
3125 .llseek = seq_lseek,
3126 .release = seq_release,
3127 };
3128
3129 static void *ptype_get_idx(loff_t pos)
3130 {
3131 struct packet_type *pt = NULL;
3132 loff_t i = 0;
3133 int t;
3134
3135 list_for_each_entry_rcu(pt, &ptype_all, list) {
3136 if (i == pos)
3137 return pt;
3138 ++i;
3139 }
3140
3141 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3142 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3143 if (i == pos)
3144 return pt;
3145 ++i;
3146 }
3147 }
3148 return NULL;
3149 }
3150
3151 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3152 __acquires(RCU)
3153 {
3154 rcu_read_lock();
3155 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3156 }
3157
3158 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3159 {
3160 struct packet_type *pt;
3161 struct list_head *nxt;
3162 int hash;
3163
3164 ++*pos;
3165 if (v == SEQ_START_TOKEN)
3166 return ptype_get_idx(0);
3167
3168 pt = v;
3169 nxt = pt->list.next;
3170 if (pt->type == htons(ETH_P_ALL)) {
3171 if (nxt != &ptype_all)
3172 goto found;
3173 hash = 0;
3174 nxt = ptype_base[0].next;
3175 } else
3176 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3177
3178 while (nxt == &ptype_base[hash]) {
3179 if (++hash >= PTYPE_HASH_SIZE)
3180 return NULL;
3181 nxt = ptype_base[hash].next;
3182 }
3183 found:
3184 return list_entry(nxt, struct packet_type, list);
3185 }
3186
3187 static void ptype_seq_stop(struct seq_file *seq, void *v)
3188 __releases(RCU)
3189 {
3190 rcu_read_unlock();
3191 }
3192
3193 static int ptype_seq_show(struct seq_file *seq, void *v)
3194 {
3195 struct packet_type *pt = v;
3196
3197 if (v == SEQ_START_TOKEN)
3198 seq_puts(seq, "Type Device Function\n");
3199 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3200 if (pt->type == htons(ETH_P_ALL))
3201 seq_puts(seq, "ALL ");
3202 else
3203 seq_printf(seq, "%04x", ntohs(pt->type));
3204
3205 seq_printf(seq, " %-8s %pF\n",
3206 pt->dev ? pt->dev->name : "", pt->func);
3207 }
3208
3209 return 0;
3210 }
3211
3212 static const struct seq_operations ptype_seq_ops = {
3213 .start = ptype_seq_start,
3214 .next = ptype_seq_next,
3215 .stop = ptype_seq_stop,
3216 .show = ptype_seq_show,
3217 };
3218
3219 static int ptype_seq_open(struct inode *inode, struct file *file)
3220 {
3221 return seq_open_net(inode, file, &ptype_seq_ops,
3222 sizeof(struct seq_net_private));
3223 }
3224
3225 static const struct file_operations ptype_seq_fops = {
3226 .owner = THIS_MODULE,
3227 .open = ptype_seq_open,
3228 .read = seq_read,
3229 .llseek = seq_lseek,
3230 .release = seq_release_net,
3231 };
3232
3233
3234 static int __net_init dev_proc_net_init(struct net *net)
3235 {
3236 int rc = -ENOMEM;
3237
3238 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3239 goto out;
3240 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3241 goto out_dev;
3242 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3243 goto out_softnet;
3244
3245 if (wext_proc_init(net))
3246 goto out_ptype;
3247 rc = 0;
3248 out:
3249 return rc;
3250 out_ptype:
3251 proc_net_remove(net, "ptype");
3252 out_softnet:
3253 proc_net_remove(net, "softnet_stat");
3254 out_dev:
3255 proc_net_remove(net, "dev");
3256 goto out;
3257 }
3258
3259 static void __net_exit dev_proc_net_exit(struct net *net)
3260 {
3261 wext_proc_exit(net);
3262
3263 proc_net_remove(net, "ptype");
3264 proc_net_remove(net, "softnet_stat");
3265 proc_net_remove(net, "dev");
3266 }
3267
3268 static struct pernet_operations __net_initdata dev_proc_ops = {
3269 .init = dev_proc_net_init,
3270 .exit = dev_proc_net_exit,
3271 };
3272
3273 static int __init dev_proc_init(void)
3274 {
3275 return register_pernet_subsys(&dev_proc_ops);
3276 }
3277 #else
3278 #define dev_proc_init() 0
3279 #endif /* CONFIG_PROC_FS */
3280
3281
3282 /**
3283 * netdev_set_master - set up master/slave pair
3284 * @slave: slave device
3285 * @master: new master device
3286 *
3287 * Changes the master device of the slave. Pass %NULL to break the
3288 * bonding. The caller must hold the RTNL semaphore. On a failure
3289 * a negative errno code is returned. On success the reference counts
3290 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3291 * function returns zero.
3292 */
3293 int netdev_set_master(struct net_device *slave, struct net_device *master)
3294 {
3295 struct net_device *old = slave->master;
3296
3297 ASSERT_RTNL();
3298
3299 if (master) {
3300 if (old)
3301 return -EBUSY;
3302 dev_hold(master);
3303 }
3304
3305 slave->master = master;
3306
3307 synchronize_net();
3308
3309 if (old)
3310 dev_put(old);
3311
3312 if (master)
3313 slave->flags |= IFF_SLAVE;
3314 else
3315 slave->flags &= ~IFF_SLAVE;
3316
3317 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3318 return 0;
3319 }
3320
3321 static void dev_change_rx_flags(struct net_device *dev, int flags)
3322 {
3323 const struct net_device_ops *ops = dev->netdev_ops;
3324
3325 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3326 ops->ndo_change_rx_flags(dev, flags);
3327 }
3328
3329 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3330 {
3331 unsigned short old_flags = dev->flags;
3332 uid_t uid;
3333 gid_t gid;
3334
3335 ASSERT_RTNL();
3336
3337 dev->flags |= IFF_PROMISC;
3338 dev->promiscuity += inc;
3339 if (dev->promiscuity == 0) {
3340 /*
3341 * Avoid overflow.
3342 * If inc causes overflow, untouch promisc and return error.
3343 */
3344 if (inc < 0)
3345 dev->flags &= ~IFF_PROMISC;
3346 else {
3347 dev->promiscuity -= inc;
3348 printk(KERN_WARNING "%s: promiscuity touches roof, "
3349 "set promiscuity failed, promiscuity feature "
3350 "of device might be broken.\n", dev->name);
3351 return -EOVERFLOW;
3352 }
3353 }
3354 if (dev->flags != old_flags) {
3355 printk(KERN_INFO "device %s %s promiscuous mode\n",
3356 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3357 "left");
3358 if (audit_enabled) {
3359 current_uid_gid(&uid, &gid);
3360 audit_log(current->audit_context, GFP_ATOMIC,
3361 AUDIT_ANOM_PROMISCUOUS,
3362 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3363 dev->name, (dev->flags & IFF_PROMISC),
3364 (old_flags & IFF_PROMISC),
3365 audit_get_loginuid(current),
3366 uid, gid,
3367 audit_get_sessionid(current));
3368 }
3369
3370 dev_change_rx_flags(dev, IFF_PROMISC);
3371 }
3372 return 0;
3373 }
3374
3375 /**
3376 * dev_set_promiscuity - update promiscuity count on a device
3377 * @dev: device
3378 * @inc: modifier
3379 *
3380 * Add or remove promiscuity from a device. While the count in the device
3381 * remains above zero the interface remains promiscuous. Once it hits zero
3382 * the device reverts back to normal filtering operation. A negative inc
3383 * value is used to drop promiscuity on the device.
3384 * Return 0 if successful or a negative errno code on error.
3385 */
3386 int dev_set_promiscuity(struct net_device *dev, int inc)
3387 {
3388 unsigned short old_flags = dev->flags;
3389 int err;
3390
3391 err = __dev_set_promiscuity(dev, inc);
3392 if (err < 0)
3393 return err;
3394 if (dev->flags != old_flags)
3395 dev_set_rx_mode(dev);
3396 return err;
3397 }
3398
3399 /**
3400 * dev_set_allmulti - update allmulti count on a device
3401 * @dev: device
3402 * @inc: modifier
3403 *
3404 * Add or remove reception of all multicast frames to a device. While the
3405 * count in the device remains above zero the interface remains listening
3406 * to all interfaces. Once it hits zero the device reverts back to normal
3407 * filtering operation. A negative @inc value is used to drop the counter
3408 * when releasing a resource needing all multicasts.
3409 * Return 0 if successful or a negative errno code on error.
3410 */
3411
3412 int dev_set_allmulti(struct net_device *dev, int inc)
3413 {
3414 unsigned short old_flags = dev->flags;
3415
3416 ASSERT_RTNL();
3417
3418 dev->flags |= IFF_ALLMULTI;
3419 dev->allmulti += inc;
3420 if (dev->allmulti == 0) {
3421 /*
3422 * Avoid overflow.
3423 * If inc causes overflow, untouch allmulti and return error.
3424 */
3425 if (inc < 0)
3426 dev->flags &= ~IFF_ALLMULTI;
3427 else {
3428 dev->allmulti -= inc;
3429 printk(KERN_WARNING "%s: allmulti touches roof, "
3430 "set allmulti failed, allmulti feature of "
3431 "device might be broken.\n", dev->name);
3432 return -EOVERFLOW;
3433 }
3434 }
3435 if (dev->flags ^ old_flags) {
3436 dev_change_rx_flags(dev, IFF_ALLMULTI);
3437 dev_set_rx_mode(dev);
3438 }
3439 return 0;
3440 }
3441
3442 /*
3443 * Upload unicast and multicast address lists to device and
3444 * configure RX filtering. When the device doesn't support unicast
3445 * filtering it is put in promiscuous mode while unicast addresses
3446 * are present.
3447 */
3448 void __dev_set_rx_mode(struct net_device *dev)
3449 {
3450 const struct net_device_ops *ops = dev->netdev_ops;
3451
3452 /* dev_open will call this function so the list will stay sane. */
3453 if (!(dev->flags&IFF_UP))
3454 return;
3455
3456 if (!netif_device_present(dev))
3457 return;
3458
3459 if (ops->ndo_set_rx_mode)
3460 ops->ndo_set_rx_mode(dev);
3461 else {
3462 /* Unicast addresses changes may only happen under the rtnl,
3463 * therefore calling __dev_set_promiscuity here is safe.
3464 */
3465 if (dev->uc.count > 0 && !dev->uc_promisc) {
3466 __dev_set_promiscuity(dev, 1);
3467 dev->uc_promisc = 1;
3468 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3469 __dev_set_promiscuity(dev, -1);
3470 dev->uc_promisc = 0;
3471 }
3472
3473 if (ops->ndo_set_multicast_list)
3474 ops->ndo_set_multicast_list(dev);
3475 }
3476 }
3477
3478 void dev_set_rx_mode(struct net_device *dev)
3479 {
3480 netif_addr_lock_bh(dev);
3481 __dev_set_rx_mode(dev);
3482 netif_addr_unlock_bh(dev);
3483 }
3484
3485 /* hw addresses list handling functions */
3486
3487 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3488 int addr_len, unsigned char addr_type)
3489 {
3490 struct netdev_hw_addr *ha;
3491 int alloc_size;
3492
3493 if (addr_len > MAX_ADDR_LEN)
3494 return -EINVAL;
3495
3496 list_for_each_entry(ha, &list->list, list) {
3497 if (!memcmp(ha->addr, addr, addr_len) &&
3498 ha->type == addr_type) {
3499 ha->refcount++;
3500 return 0;
3501 }
3502 }
3503
3504
3505 alloc_size = sizeof(*ha);
3506 if (alloc_size < L1_CACHE_BYTES)
3507 alloc_size = L1_CACHE_BYTES;
3508 ha = kmalloc(alloc_size, GFP_ATOMIC);
3509 if (!ha)
3510 return -ENOMEM;
3511 memcpy(ha->addr, addr, addr_len);
3512 ha->type = addr_type;
3513 ha->refcount = 1;
3514 ha->synced = false;
3515 list_add_tail_rcu(&ha->list, &list->list);
3516 list->count++;
3517 return 0;
3518 }
3519
3520 static void ha_rcu_free(struct rcu_head *head)
3521 {
3522 struct netdev_hw_addr *ha;
3523
3524 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3525 kfree(ha);
3526 }
3527
3528 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3529 int addr_len, unsigned char addr_type)
3530 {
3531 struct netdev_hw_addr *ha;
3532
3533 list_for_each_entry(ha, &list->list, list) {
3534 if (!memcmp(ha->addr, addr, addr_len) &&
3535 (ha->type == addr_type || !addr_type)) {
3536 if (--ha->refcount)
3537 return 0;
3538 list_del_rcu(&ha->list);
3539 call_rcu(&ha->rcu_head, ha_rcu_free);
3540 list->count--;
3541 return 0;
3542 }
3543 }
3544 return -ENOENT;
3545 }
3546
3547 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3548 struct netdev_hw_addr_list *from_list,
3549 int addr_len,
3550 unsigned char addr_type)
3551 {
3552 int err;
3553 struct netdev_hw_addr *ha, *ha2;
3554 unsigned char type;
3555
3556 list_for_each_entry(ha, &from_list->list, list) {
3557 type = addr_type ? addr_type : ha->type;
3558 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3559 if (err)
3560 goto unroll;
3561 }
3562 return 0;
3563
3564 unroll:
3565 list_for_each_entry(ha2, &from_list->list, list) {
3566 if (ha2 == ha)
3567 break;
3568 type = addr_type ? addr_type : ha2->type;
3569 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3570 }
3571 return err;
3572 }
3573
3574 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3575 struct netdev_hw_addr_list *from_list,
3576 int addr_len,
3577 unsigned char addr_type)
3578 {
3579 struct netdev_hw_addr *ha;
3580 unsigned char type;
3581
3582 list_for_each_entry(ha, &from_list->list, list) {
3583 type = addr_type ? addr_type : ha->type;
3584 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3585 }
3586 }
3587
3588 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3589 struct netdev_hw_addr_list *from_list,
3590 int addr_len)
3591 {
3592 int err = 0;
3593 struct netdev_hw_addr *ha, *tmp;
3594
3595 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3596 if (!ha->synced) {
3597 err = __hw_addr_add(to_list, ha->addr,
3598 addr_len, ha->type);
3599 if (err)
3600 break;
3601 ha->synced = true;
3602 ha->refcount++;
3603 } else if (ha->refcount == 1) {
3604 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3605 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3606 }
3607 }
3608 return err;
3609 }
3610
3611 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3612 struct netdev_hw_addr_list *from_list,
3613 int addr_len)
3614 {
3615 struct netdev_hw_addr *ha, *tmp;
3616
3617 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3618 if (ha->synced) {
3619 __hw_addr_del(to_list, ha->addr,
3620 addr_len, ha->type);
3621 ha->synced = false;
3622 __hw_addr_del(from_list, ha->addr,
3623 addr_len, ha->type);
3624 }
3625 }
3626 }
3627
3628 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3629 {
3630 struct netdev_hw_addr *ha, *tmp;
3631
3632 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3633 list_del_rcu(&ha->list);
3634 call_rcu(&ha->rcu_head, ha_rcu_free);
3635 }
3636 list->count = 0;
3637 }
3638
3639 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3640 {
3641 INIT_LIST_HEAD(&list->list);
3642 list->count = 0;
3643 }
3644
3645 /* Device addresses handling functions */
3646
3647 static void dev_addr_flush(struct net_device *dev)
3648 {
3649 /* rtnl_mutex must be held here */
3650
3651 __hw_addr_flush(&dev->dev_addrs);
3652 dev->dev_addr = NULL;
3653 }
3654
3655 static int dev_addr_init(struct net_device *dev)
3656 {
3657 unsigned char addr[MAX_ADDR_LEN];
3658 struct netdev_hw_addr *ha;
3659 int err;
3660
3661 /* rtnl_mutex must be held here */
3662
3663 __hw_addr_init(&dev->dev_addrs);
3664 memset(addr, 0, sizeof(addr));
3665 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3666 NETDEV_HW_ADDR_T_LAN);
3667 if (!err) {
3668 /*
3669 * Get the first (previously created) address from the list
3670 * and set dev_addr pointer to this location.
3671 */
3672 ha = list_first_entry(&dev->dev_addrs.list,
3673 struct netdev_hw_addr, list);
3674 dev->dev_addr = ha->addr;
3675 }
3676 return err;
3677 }
3678
3679 /**
3680 * dev_addr_add - Add a device address
3681 * @dev: device
3682 * @addr: address to add
3683 * @addr_type: address type
3684 *
3685 * Add a device address to the device or increase the reference count if
3686 * it already exists.
3687 *
3688 * The caller must hold the rtnl_mutex.
3689 */
3690 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3691 unsigned char addr_type)
3692 {
3693 int err;
3694
3695 ASSERT_RTNL();
3696
3697 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3698 if (!err)
3699 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3700 return err;
3701 }
3702 EXPORT_SYMBOL(dev_addr_add);
3703
3704 /**
3705 * dev_addr_del - Release a device address.
3706 * @dev: device
3707 * @addr: address to delete
3708 * @addr_type: address type
3709 *
3710 * Release reference to a device address and remove it from the device
3711 * if the reference count drops to zero.
3712 *
3713 * The caller must hold the rtnl_mutex.
3714 */
3715 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3716 unsigned char addr_type)
3717 {
3718 int err;
3719 struct netdev_hw_addr *ha;
3720
3721 ASSERT_RTNL();
3722
3723 /*
3724 * We can not remove the first address from the list because
3725 * dev->dev_addr points to that.
3726 */
3727 ha = list_first_entry(&dev->dev_addrs.list,
3728 struct netdev_hw_addr, list);
3729 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3730 return -ENOENT;
3731
3732 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3733 addr_type);
3734 if (!err)
3735 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3736 return err;
3737 }
3738 EXPORT_SYMBOL(dev_addr_del);
3739
3740 /**
3741 * dev_addr_add_multiple - Add device addresses from another device
3742 * @to_dev: device to which addresses will be added
3743 * @from_dev: device from which addresses will be added
3744 * @addr_type: address type - 0 means type will be used from from_dev
3745 *
3746 * Add device addresses of the one device to another.
3747 **
3748 * The caller must hold the rtnl_mutex.
3749 */
3750 int dev_addr_add_multiple(struct net_device *to_dev,
3751 struct net_device *from_dev,
3752 unsigned char addr_type)
3753 {
3754 int err;
3755
3756 ASSERT_RTNL();
3757
3758 if (from_dev->addr_len != to_dev->addr_len)
3759 return -EINVAL;
3760 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3761 to_dev->addr_len, addr_type);
3762 if (!err)
3763 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3764 return err;
3765 }
3766 EXPORT_SYMBOL(dev_addr_add_multiple);
3767
3768 /**
3769 * dev_addr_del_multiple - Delete device addresses by another device
3770 * @to_dev: device where the addresses will be deleted
3771 * @from_dev: device by which addresses the addresses will be deleted
3772 * @addr_type: address type - 0 means type will used from from_dev
3773 *
3774 * Deletes addresses in to device by the list of addresses in from device.
3775 *
3776 * The caller must hold the rtnl_mutex.
3777 */
3778 int dev_addr_del_multiple(struct net_device *to_dev,
3779 struct net_device *from_dev,
3780 unsigned char addr_type)
3781 {
3782 ASSERT_RTNL();
3783
3784 if (from_dev->addr_len != to_dev->addr_len)
3785 return -EINVAL;
3786 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3787 to_dev->addr_len, addr_type);
3788 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3789 return 0;
3790 }
3791 EXPORT_SYMBOL(dev_addr_del_multiple);
3792
3793 /* multicast addresses handling functions */
3794
3795 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3796 void *addr, int alen, int glbl)
3797 {
3798 struct dev_addr_list *da;
3799
3800 for (; (da = *list) != NULL; list = &da->next) {
3801 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3802 alen == da->da_addrlen) {
3803 if (glbl) {
3804 int old_glbl = da->da_gusers;
3805 da->da_gusers = 0;
3806 if (old_glbl == 0)
3807 break;
3808 }
3809 if (--da->da_users)
3810 return 0;
3811
3812 *list = da->next;
3813 kfree(da);
3814 (*count)--;
3815 return 0;
3816 }
3817 }
3818 return -ENOENT;
3819 }
3820
3821 int __dev_addr_add(struct dev_addr_list **list, int *count,
3822 void *addr, int alen, int glbl)
3823 {
3824 struct dev_addr_list *da;
3825
3826 for (da = *list; da != NULL; da = da->next) {
3827 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3828 da->da_addrlen == alen) {
3829 if (glbl) {
3830 int old_glbl = da->da_gusers;
3831 da->da_gusers = 1;
3832 if (old_glbl)
3833 return 0;
3834 }
3835 da->da_users++;
3836 return 0;
3837 }
3838 }
3839
3840 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3841 if (da == NULL)
3842 return -ENOMEM;
3843 memcpy(da->da_addr, addr, alen);
3844 da->da_addrlen = alen;
3845 da->da_users = 1;
3846 da->da_gusers = glbl ? 1 : 0;
3847 da->next = *list;
3848 *list = da;
3849 (*count)++;
3850 return 0;
3851 }
3852
3853 /**
3854 * dev_unicast_delete - Release secondary unicast address.
3855 * @dev: device
3856 * @addr: address to delete
3857 *
3858 * Release reference to a secondary unicast address and remove it
3859 * from the device if the reference count drops to zero.
3860 *
3861 * The caller must hold the rtnl_mutex.
3862 */
3863 int dev_unicast_delete(struct net_device *dev, void *addr)
3864 {
3865 int err;
3866
3867 ASSERT_RTNL();
3868
3869 netif_addr_lock_bh(dev);
3870 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3871 NETDEV_HW_ADDR_T_UNICAST);
3872 if (!err)
3873 __dev_set_rx_mode(dev);
3874 netif_addr_unlock_bh(dev);
3875 return err;
3876 }
3877 EXPORT_SYMBOL(dev_unicast_delete);
3878
3879 /**
3880 * dev_unicast_add - add a secondary unicast address
3881 * @dev: device
3882 * @addr: address to add
3883 *
3884 * Add a secondary unicast address to the device or increase
3885 * the reference count if it already exists.
3886 *
3887 * The caller must hold the rtnl_mutex.
3888 */
3889 int dev_unicast_add(struct net_device *dev, void *addr)
3890 {
3891 int err;
3892
3893 ASSERT_RTNL();
3894
3895 netif_addr_lock_bh(dev);
3896 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3897 NETDEV_HW_ADDR_T_UNICAST);
3898 if (!err)
3899 __dev_set_rx_mode(dev);
3900 netif_addr_unlock_bh(dev);
3901 return err;
3902 }
3903 EXPORT_SYMBOL(dev_unicast_add);
3904
3905 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3906 struct dev_addr_list **from, int *from_count)
3907 {
3908 struct dev_addr_list *da, *next;
3909 int err = 0;
3910
3911 da = *from;
3912 while (da != NULL) {
3913 next = da->next;
3914 if (!da->da_synced) {
3915 err = __dev_addr_add(to, to_count,
3916 da->da_addr, da->da_addrlen, 0);
3917 if (err < 0)
3918 break;
3919 da->da_synced = 1;
3920 da->da_users++;
3921 } else if (da->da_users == 1) {
3922 __dev_addr_delete(to, to_count,
3923 da->da_addr, da->da_addrlen, 0);
3924 __dev_addr_delete(from, from_count,
3925 da->da_addr, da->da_addrlen, 0);
3926 }
3927 da = next;
3928 }
3929 return err;
3930 }
3931
3932 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3933 struct dev_addr_list **from, int *from_count)
3934 {
3935 struct dev_addr_list *da, *next;
3936
3937 da = *from;
3938 while (da != NULL) {
3939 next = da->next;
3940 if (da->da_synced) {
3941 __dev_addr_delete(to, to_count,
3942 da->da_addr, da->da_addrlen, 0);
3943 da->da_synced = 0;
3944 __dev_addr_delete(from, from_count,
3945 da->da_addr, da->da_addrlen, 0);
3946 }
3947 da = next;
3948 }
3949 }
3950
3951 /**
3952 * dev_unicast_sync - Synchronize device's unicast list to another device
3953 * @to: destination device
3954 * @from: source device
3955 *
3956 * Add newly added addresses to the destination device and release
3957 * addresses that have no users left. The source device must be
3958 * locked by netif_tx_lock_bh.
3959 *
3960 * This function is intended to be called from the dev->set_rx_mode
3961 * function of layered software devices.
3962 */
3963 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3964 {
3965 int err = 0;
3966
3967 if (to->addr_len != from->addr_len)
3968 return -EINVAL;
3969
3970 netif_addr_lock_bh(to);
3971 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
3972 if (!err)
3973 __dev_set_rx_mode(to);
3974 netif_addr_unlock_bh(to);
3975 return err;
3976 }
3977 EXPORT_SYMBOL(dev_unicast_sync);
3978
3979 /**
3980 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3981 * @to: destination device
3982 * @from: source device
3983 *
3984 * Remove all addresses that were added to the destination device by
3985 * dev_unicast_sync(). This function is intended to be called from the
3986 * dev->stop function of layered software devices.
3987 */
3988 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3989 {
3990 if (to->addr_len != from->addr_len)
3991 return;
3992
3993 netif_addr_lock_bh(from);
3994 netif_addr_lock(to);
3995 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
3996 __dev_set_rx_mode(to);
3997 netif_addr_unlock(to);
3998 netif_addr_unlock_bh(from);
3999 }
4000 EXPORT_SYMBOL(dev_unicast_unsync);
4001
4002 static void dev_unicast_flush(struct net_device *dev)
4003 {
4004 netif_addr_lock_bh(dev);
4005 __hw_addr_flush(&dev->uc);
4006 netif_addr_unlock_bh(dev);
4007 }
4008
4009 static void dev_unicast_init(struct net_device *dev)
4010 {
4011 __hw_addr_init(&dev->uc);
4012 }
4013
4014
4015 static void __dev_addr_discard(struct dev_addr_list **list)
4016 {
4017 struct dev_addr_list *tmp;
4018
4019 while (*list != NULL) {
4020 tmp = *list;
4021 *list = tmp->next;
4022 if (tmp->da_users > tmp->da_gusers)
4023 printk("__dev_addr_discard: address leakage! "
4024 "da_users=%d\n", tmp->da_users);
4025 kfree(tmp);
4026 }
4027 }
4028
4029 static void dev_addr_discard(struct net_device *dev)
4030 {
4031 netif_addr_lock_bh(dev);
4032
4033 __dev_addr_discard(&dev->mc_list);
4034 dev->mc_count = 0;
4035
4036 netif_addr_unlock_bh(dev);
4037 }
4038
4039 /**
4040 * dev_get_flags - get flags reported to userspace
4041 * @dev: device
4042 *
4043 * Get the combination of flag bits exported through APIs to userspace.
4044 */
4045 unsigned dev_get_flags(const struct net_device *dev)
4046 {
4047 unsigned flags;
4048
4049 flags = (dev->flags & ~(IFF_PROMISC |
4050 IFF_ALLMULTI |
4051 IFF_RUNNING |
4052 IFF_LOWER_UP |
4053 IFF_DORMANT)) |
4054 (dev->gflags & (IFF_PROMISC |
4055 IFF_ALLMULTI));
4056
4057 if (netif_running(dev)) {
4058 if (netif_oper_up(dev))
4059 flags |= IFF_RUNNING;
4060 if (netif_carrier_ok(dev))
4061 flags |= IFF_LOWER_UP;
4062 if (netif_dormant(dev))
4063 flags |= IFF_DORMANT;
4064 }
4065
4066 return flags;
4067 }
4068
4069 /**
4070 * dev_change_flags - change device settings
4071 * @dev: device
4072 * @flags: device state flags
4073 *
4074 * Change settings on device based state flags. The flags are
4075 * in the userspace exported format.
4076 */
4077 int dev_change_flags(struct net_device *dev, unsigned flags)
4078 {
4079 int ret, changes;
4080 int old_flags = dev->flags;
4081
4082 ASSERT_RTNL();
4083
4084 /*
4085 * Set the flags on our device.
4086 */
4087
4088 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4089 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4090 IFF_AUTOMEDIA)) |
4091 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4092 IFF_ALLMULTI));
4093
4094 /*
4095 * Load in the correct multicast list now the flags have changed.
4096 */
4097
4098 if ((old_flags ^ flags) & IFF_MULTICAST)
4099 dev_change_rx_flags(dev, IFF_MULTICAST);
4100
4101 dev_set_rx_mode(dev);
4102
4103 /*
4104 * Have we downed the interface. We handle IFF_UP ourselves
4105 * according to user attempts to set it, rather than blindly
4106 * setting it.
4107 */
4108
4109 ret = 0;
4110 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4111 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4112
4113 if (!ret)
4114 dev_set_rx_mode(dev);
4115 }
4116
4117 if (dev->flags & IFF_UP &&
4118 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4119 IFF_VOLATILE)))
4120 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4121
4122 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4123 int inc = (flags & IFF_PROMISC) ? +1 : -1;
4124 dev->gflags ^= IFF_PROMISC;
4125 dev_set_promiscuity(dev, inc);
4126 }
4127
4128 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4129 is important. Some (broken) drivers set IFF_PROMISC, when
4130 IFF_ALLMULTI is requested not asking us and not reporting.
4131 */
4132 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4133 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
4134 dev->gflags ^= IFF_ALLMULTI;
4135 dev_set_allmulti(dev, inc);
4136 }
4137
4138 /* Exclude state transition flags, already notified */
4139 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4140 if (changes)
4141 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4142
4143 return ret;
4144 }
4145
4146 /**
4147 * dev_set_mtu - Change maximum transfer unit
4148 * @dev: device
4149 * @new_mtu: new transfer unit
4150 *
4151 * Change the maximum transfer size of the network device.
4152 */
4153 int dev_set_mtu(struct net_device *dev, int new_mtu)
4154 {
4155 const struct net_device_ops *ops = dev->netdev_ops;
4156 int err;
4157
4158 if (new_mtu == dev->mtu)
4159 return 0;
4160
4161 /* MTU must be positive. */
4162 if (new_mtu < 0)
4163 return -EINVAL;
4164
4165 if (!netif_device_present(dev))
4166 return -ENODEV;
4167
4168 err = 0;
4169 if (ops->ndo_change_mtu)
4170 err = ops->ndo_change_mtu(dev, new_mtu);
4171 else
4172 dev->mtu = new_mtu;
4173
4174 if (!err && dev->flags & IFF_UP)
4175 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4176 return err;
4177 }
4178
4179 /**
4180 * dev_set_mac_address - Change Media Access Control Address
4181 * @dev: device
4182 * @sa: new address
4183 *
4184 * Change the hardware (MAC) address of the device
4185 */
4186 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4187 {
4188 const struct net_device_ops *ops = dev->netdev_ops;
4189 int err;
4190
4191 if (!ops->ndo_set_mac_address)
4192 return -EOPNOTSUPP;
4193 if (sa->sa_family != dev->type)
4194 return -EINVAL;
4195 if (!netif_device_present(dev))
4196 return -ENODEV;
4197 err = ops->ndo_set_mac_address(dev, sa);
4198 if (!err)
4199 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4200 return err;
4201 }
4202
4203 /*
4204 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4205 */
4206 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4207 {
4208 int err;
4209 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4210
4211 if (!dev)
4212 return -ENODEV;
4213
4214 switch (cmd) {
4215 case SIOCGIFFLAGS: /* Get interface flags */
4216 ifr->ifr_flags = (short) dev_get_flags(dev);
4217 return 0;
4218
4219 case SIOCGIFMETRIC: /* Get the metric on the interface
4220 (currently unused) */
4221 ifr->ifr_metric = 0;
4222 return 0;
4223
4224 case SIOCGIFMTU: /* Get the MTU of a device */
4225 ifr->ifr_mtu = dev->mtu;
4226 return 0;
4227
4228 case SIOCGIFHWADDR:
4229 if (!dev->addr_len)
4230 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4231 else
4232 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4233 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4234 ifr->ifr_hwaddr.sa_family = dev->type;
4235 return 0;
4236
4237 case SIOCGIFSLAVE:
4238 err = -EINVAL;
4239 break;
4240
4241 case SIOCGIFMAP:
4242 ifr->ifr_map.mem_start = dev->mem_start;
4243 ifr->ifr_map.mem_end = dev->mem_end;
4244 ifr->ifr_map.base_addr = dev->base_addr;
4245 ifr->ifr_map.irq = dev->irq;
4246 ifr->ifr_map.dma = dev->dma;
4247 ifr->ifr_map.port = dev->if_port;
4248 return 0;
4249
4250 case SIOCGIFINDEX:
4251 ifr->ifr_ifindex = dev->ifindex;
4252 return 0;
4253
4254 case SIOCGIFTXQLEN:
4255 ifr->ifr_qlen = dev->tx_queue_len;
4256 return 0;
4257
4258 default:
4259 /* dev_ioctl() should ensure this case
4260 * is never reached
4261 */
4262 WARN_ON(1);
4263 err = -EINVAL;
4264 break;
4265
4266 }
4267 return err;
4268 }
4269
4270 /*
4271 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4272 */
4273 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4274 {
4275 int err;
4276 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4277 const struct net_device_ops *ops;
4278
4279 if (!dev)
4280 return -ENODEV;
4281
4282 ops = dev->netdev_ops;
4283
4284 switch (cmd) {
4285 case SIOCSIFFLAGS: /* Set interface flags */
4286 return dev_change_flags(dev, ifr->ifr_flags);
4287
4288 case SIOCSIFMETRIC: /* Set the metric on the interface
4289 (currently unused) */
4290 return -EOPNOTSUPP;
4291
4292 case SIOCSIFMTU: /* Set the MTU of a device */
4293 return dev_set_mtu(dev, ifr->ifr_mtu);
4294
4295 case SIOCSIFHWADDR:
4296 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4297
4298 case SIOCSIFHWBROADCAST:
4299 if (ifr->ifr_hwaddr.sa_family != dev->type)
4300 return -EINVAL;
4301 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4302 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4303 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4304 return 0;
4305
4306 case SIOCSIFMAP:
4307 if (ops->ndo_set_config) {
4308 if (!netif_device_present(dev))
4309 return -ENODEV;
4310 return ops->ndo_set_config(dev, &ifr->ifr_map);
4311 }
4312 return -EOPNOTSUPP;
4313
4314 case SIOCADDMULTI:
4315 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4316 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4317 return -EINVAL;
4318 if (!netif_device_present(dev))
4319 return -ENODEV;
4320 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4321 dev->addr_len, 1);
4322
4323 case SIOCDELMULTI:
4324 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4325 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4326 return -EINVAL;
4327 if (!netif_device_present(dev))
4328 return -ENODEV;
4329 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4330 dev->addr_len, 1);
4331
4332 case SIOCSIFTXQLEN:
4333 if (ifr->ifr_qlen < 0)
4334 return -EINVAL;
4335 dev->tx_queue_len = ifr->ifr_qlen;
4336 return 0;
4337
4338 case SIOCSIFNAME:
4339 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4340 return dev_change_name(dev, ifr->ifr_newname);
4341
4342 /*
4343 * Unknown or private ioctl
4344 */
4345
4346 default:
4347 if ((cmd >= SIOCDEVPRIVATE &&
4348 cmd <= SIOCDEVPRIVATE + 15) ||
4349 cmd == SIOCBONDENSLAVE ||
4350 cmd == SIOCBONDRELEASE ||
4351 cmd == SIOCBONDSETHWADDR ||
4352 cmd == SIOCBONDSLAVEINFOQUERY ||
4353 cmd == SIOCBONDINFOQUERY ||
4354 cmd == SIOCBONDCHANGEACTIVE ||
4355 cmd == SIOCGMIIPHY ||
4356 cmd == SIOCGMIIREG ||
4357 cmd == SIOCSMIIREG ||
4358 cmd == SIOCBRADDIF ||
4359 cmd == SIOCBRDELIF ||
4360 cmd == SIOCSHWTSTAMP ||
4361 cmd == SIOCWANDEV) {
4362 err = -EOPNOTSUPP;
4363 if (ops->ndo_do_ioctl) {
4364 if (netif_device_present(dev))
4365 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4366 else
4367 err = -ENODEV;
4368 }
4369 } else
4370 err = -EINVAL;
4371
4372 }
4373 return err;
4374 }
4375
4376 /*
4377 * This function handles all "interface"-type I/O control requests. The actual
4378 * 'doing' part of this is dev_ifsioc above.
4379 */
4380
4381 /**
4382 * dev_ioctl - network device ioctl
4383 * @net: the applicable net namespace
4384 * @cmd: command to issue
4385 * @arg: pointer to a struct ifreq in user space
4386 *
4387 * Issue ioctl functions to devices. This is normally called by the
4388 * user space syscall interfaces but can sometimes be useful for
4389 * other purposes. The return value is the return from the syscall if
4390 * positive or a negative errno code on error.
4391 */
4392
4393 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4394 {
4395 struct ifreq ifr;
4396 int ret;
4397 char *colon;
4398
4399 /* One special case: SIOCGIFCONF takes ifconf argument
4400 and requires shared lock, because it sleeps writing
4401 to user space.
4402 */
4403
4404 if (cmd == SIOCGIFCONF) {
4405 rtnl_lock();
4406 ret = dev_ifconf(net, (char __user *) arg);
4407 rtnl_unlock();
4408 return ret;
4409 }
4410 if (cmd == SIOCGIFNAME)
4411 return dev_ifname(net, (struct ifreq __user *)arg);
4412
4413 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4414 return -EFAULT;
4415
4416 ifr.ifr_name[IFNAMSIZ-1] = 0;
4417
4418 colon = strchr(ifr.ifr_name, ':');
4419 if (colon)
4420 *colon = 0;
4421
4422 /*
4423 * See which interface the caller is talking about.
4424 */
4425
4426 switch (cmd) {
4427 /*
4428 * These ioctl calls:
4429 * - can be done by all.
4430 * - atomic and do not require locking.
4431 * - return a value
4432 */
4433 case SIOCGIFFLAGS:
4434 case SIOCGIFMETRIC:
4435 case SIOCGIFMTU:
4436 case SIOCGIFHWADDR:
4437 case SIOCGIFSLAVE:
4438 case SIOCGIFMAP:
4439 case SIOCGIFINDEX:
4440 case SIOCGIFTXQLEN:
4441 dev_load(net, ifr.ifr_name);
4442 read_lock(&dev_base_lock);
4443 ret = dev_ifsioc_locked(net, &ifr, cmd);
4444 read_unlock(&dev_base_lock);
4445 if (!ret) {
4446 if (colon)
4447 *colon = ':';
4448 if (copy_to_user(arg, &ifr,
4449 sizeof(struct ifreq)))
4450 ret = -EFAULT;
4451 }
4452 return ret;
4453
4454 case SIOCETHTOOL:
4455 dev_load(net, ifr.ifr_name);
4456 rtnl_lock();
4457 ret = dev_ethtool(net, &ifr);
4458 rtnl_unlock();
4459 if (!ret) {
4460 if (colon)
4461 *colon = ':';
4462 if (copy_to_user(arg, &ifr,
4463 sizeof(struct ifreq)))
4464 ret = -EFAULT;
4465 }
4466 return ret;
4467
4468 /*
4469 * These ioctl calls:
4470 * - require superuser power.
4471 * - require strict serialization.
4472 * - return a value
4473 */
4474 case SIOCGMIIPHY:
4475 case SIOCGMIIREG:
4476 case SIOCSIFNAME:
4477 if (!capable(CAP_NET_ADMIN))
4478 return -EPERM;
4479 dev_load(net, ifr.ifr_name);
4480 rtnl_lock();
4481 ret = dev_ifsioc(net, &ifr, cmd);
4482 rtnl_unlock();
4483 if (!ret) {
4484 if (colon)
4485 *colon = ':';
4486 if (copy_to_user(arg, &ifr,
4487 sizeof(struct ifreq)))
4488 ret = -EFAULT;
4489 }
4490 return ret;
4491
4492 /*
4493 * These ioctl calls:
4494 * - require superuser power.
4495 * - require strict serialization.
4496 * - do not return a value
4497 */
4498 case SIOCSIFFLAGS:
4499 case SIOCSIFMETRIC:
4500 case SIOCSIFMTU:
4501 case SIOCSIFMAP:
4502 case SIOCSIFHWADDR:
4503 case SIOCSIFSLAVE:
4504 case SIOCADDMULTI:
4505 case SIOCDELMULTI:
4506 case SIOCSIFHWBROADCAST:
4507 case SIOCSIFTXQLEN:
4508 case SIOCSMIIREG:
4509 case SIOCBONDENSLAVE:
4510 case SIOCBONDRELEASE:
4511 case SIOCBONDSETHWADDR:
4512 case SIOCBONDCHANGEACTIVE:
4513 case SIOCBRADDIF:
4514 case SIOCBRDELIF:
4515 case SIOCSHWTSTAMP:
4516 if (!capable(CAP_NET_ADMIN))
4517 return -EPERM;
4518 /* fall through */
4519 case SIOCBONDSLAVEINFOQUERY:
4520 case SIOCBONDINFOQUERY:
4521 dev_load(net, ifr.ifr_name);
4522 rtnl_lock();
4523 ret = dev_ifsioc(net, &ifr, cmd);
4524 rtnl_unlock();
4525 return ret;
4526
4527 case SIOCGIFMEM:
4528 /* Get the per device memory space. We can add this but
4529 * currently do not support it */
4530 case SIOCSIFMEM:
4531 /* Set the per device memory buffer space.
4532 * Not applicable in our case */
4533 case SIOCSIFLINK:
4534 return -EINVAL;
4535
4536 /*
4537 * Unknown or private ioctl.
4538 */
4539 default:
4540 if (cmd == SIOCWANDEV ||
4541 (cmd >= SIOCDEVPRIVATE &&
4542 cmd <= SIOCDEVPRIVATE + 15)) {
4543 dev_load(net, ifr.ifr_name);
4544 rtnl_lock();
4545 ret = dev_ifsioc(net, &ifr, cmd);
4546 rtnl_unlock();
4547 if (!ret && copy_to_user(arg, &ifr,
4548 sizeof(struct ifreq)))
4549 ret = -EFAULT;
4550 return ret;
4551 }
4552 /* Take care of Wireless Extensions */
4553 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4554 return wext_handle_ioctl(net, &ifr, cmd, arg);
4555 return -EINVAL;
4556 }
4557 }
4558
4559
4560 /**
4561 * dev_new_index - allocate an ifindex
4562 * @net: the applicable net namespace
4563 *
4564 * Returns a suitable unique value for a new device interface
4565 * number. The caller must hold the rtnl semaphore or the
4566 * dev_base_lock to be sure it remains unique.
4567 */
4568 static int dev_new_index(struct net *net)
4569 {
4570 static int ifindex;
4571 for (;;) {
4572 if (++ifindex <= 0)
4573 ifindex = 1;
4574 if (!__dev_get_by_index(net, ifindex))
4575 return ifindex;
4576 }
4577 }
4578
4579 /* Delayed registration/unregisteration */
4580 static LIST_HEAD(net_todo_list);
4581
4582 static void net_set_todo(struct net_device *dev)
4583 {
4584 list_add_tail(&dev->todo_list, &net_todo_list);
4585 }
4586
4587 static void rollback_registered(struct net_device *dev)
4588 {
4589 BUG_ON(dev_boot_phase);
4590 ASSERT_RTNL();
4591
4592 /* Some devices call without registering for initialization unwind. */
4593 if (dev->reg_state == NETREG_UNINITIALIZED) {
4594 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4595 "was registered\n", dev->name, dev);
4596
4597 WARN_ON(1);
4598 return;
4599 }
4600
4601 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4602
4603 /* If device is running, close it first. */
4604 dev_close(dev);
4605
4606 /* And unlink it from device chain. */
4607 unlist_netdevice(dev);
4608
4609 dev->reg_state = NETREG_UNREGISTERING;
4610
4611 synchronize_net();
4612
4613 /* Shutdown queueing discipline. */
4614 dev_shutdown(dev);
4615
4616
4617 /* Notify protocols, that we are about to destroy
4618 this device. They should clean all the things.
4619 */
4620 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4621
4622 /*
4623 * Flush the unicast and multicast chains
4624 */
4625 dev_unicast_flush(dev);
4626 dev_addr_discard(dev);
4627
4628 if (dev->netdev_ops->ndo_uninit)
4629 dev->netdev_ops->ndo_uninit(dev);
4630
4631 /* Notifier chain MUST detach us from master device. */
4632 WARN_ON(dev->master);
4633
4634 /* Remove entries from kobject tree */
4635 netdev_unregister_kobject(dev);
4636
4637 synchronize_net();
4638
4639 dev_put(dev);
4640 }
4641
4642 static void __netdev_init_queue_locks_one(struct net_device *dev,
4643 struct netdev_queue *dev_queue,
4644 void *_unused)
4645 {
4646 spin_lock_init(&dev_queue->_xmit_lock);
4647 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4648 dev_queue->xmit_lock_owner = -1;
4649 }
4650
4651 static void netdev_init_queue_locks(struct net_device *dev)
4652 {
4653 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4654 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4655 }
4656
4657 unsigned long netdev_fix_features(unsigned long features, const char *name)
4658 {
4659 /* Fix illegal SG+CSUM combinations. */
4660 if ((features & NETIF_F_SG) &&
4661 !(features & NETIF_F_ALL_CSUM)) {
4662 if (name)
4663 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4664 "checksum feature.\n", name);
4665 features &= ~NETIF_F_SG;
4666 }
4667
4668 /* TSO requires that SG is present as well. */
4669 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4670 if (name)
4671 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4672 "SG feature.\n", name);
4673 features &= ~NETIF_F_TSO;
4674 }
4675
4676 if (features & NETIF_F_UFO) {
4677 if (!(features & NETIF_F_GEN_CSUM)) {
4678 if (name)
4679 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4680 "since no NETIF_F_HW_CSUM feature.\n",
4681 name);
4682 features &= ~NETIF_F_UFO;
4683 }
4684
4685 if (!(features & NETIF_F_SG)) {
4686 if (name)
4687 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4688 "since no NETIF_F_SG feature.\n", name);
4689 features &= ~NETIF_F_UFO;
4690 }
4691 }
4692
4693 return features;
4694 }
4695 EXPORT_SYMBOL(netdev_fix_features);
4696
4697 /**
4698 * register_netdevice - register a network device
4699 * @dev: device to register
4700 *
4701 * Take a completed network device structure and add it to the kernel
4702 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4703 * chain. 0 is returned on success. A negative errno code is returned
4704 * on a failure to set up the device, or if the name is a duplicate.
4705 *
4706 * Callers must hold the rtnl semaphore. You may want
4707 * register_netdev() instead of this.
4708 *
4709 * BUGS:
4710 * The locking appears insufficient to guarantee two parallel registers
4711 * will not get the same name.
4712 */
4713
4714 int register_netdevice(struct net_device *dev)
4715 {
4716 struct hlist_head *head;
4717 struct hlist_node *p;
4718 int ret;
4719 struct net *net = dev_net(dev);
4720
4721 BUG_ON(dev_boot_phase);
4722 ASSERT_RTNL();
4723
4724 might_sleep();
4725
4726 /* When net_device's are persistent, this will be fatal. */
4727 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4728 BUG_ON(!net);
4729
4730 spin_lock_init(&dev->addr_list_lock);
4731 netdev_set_addr_lockdep_class(dev);
4732 netdev_init_queue_locks(dev);
4733
4734 dev->iflink = -1;
4735
4736 /* Init, if this function is available */
4737 if (dev->netdev_ops->ndo_init) {
4738 ret = dev->netdev_ops->ndo_init(dev);
4739 if (ret) {
4740 if (ret > 0)
4741 ret = -EIO;
4742 goto out;
4743 }
4744 }
4745
4746 if (!dev_valid_name(dev->name)) {
4747 ret = -EINVAL;
4748 goto err_uninit;
4749 }
4750
4751 dev->ifindex = dev_new_index(net);
4752 if (dev->iflink == -1)
4753 dev->iflink = dev->ifindex;
4754
4755 /* Check for existence of name */
4756 head = dev_name_hash(net, dev->name);
4757 hlist_for_each(p, head) {
4758 struct net_device *d
4759 = hlist_entry(p, struct net_device, name_hlist);
4760 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4761 ret = -EEXIST;
4762 goto err_uninit;
4763 }
4764 }
4765
4766 /* Fix illegal checksum combinations */
4767 if ((dev->features & NETIF_F_HW_CSUM) &&
4768 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4769 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4770 dev->name);
4771 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4772 }
4773
4774 if ((dev->features & NETIF_F_NO_CSUM) &&
4775 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4776 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4777 dev->name);
4778 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4779 }
4780
4781 dev->features = netdev_fix_features(dev->features, dev->name);
4782
4783 /* Enable software GSO if SG is supported. */
4784 if (dev->features & NETIF_F_SG)
4785 dev->features |= NETIF_F_GSO;
4786
4787 netdev_initialize_kobject(dev);
4788 ret = netdev_register_kobject(dev);
4789 if (ret)
4790 goto err_uninit;
4791 dev->reg_state = NETREG_REGISTERED;
4792
4793 /*
4794 * Default initial state at registry is that the
4795 * device is present.
4796 */
4797
4798 set_bit(__LINK_STATE_PRESENT, &dev->state);
4799
4800 dev_init_scheduler(dev);
4801 dev_hold(dev);
4802 list_netdevice(dev);
4803
4804 /* Notify protocols, that a new device appeared. */
4805 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4806 ret = notifier_to_errno(ret);
4807 if (ret) {
4808 rollback_registered(dev);
4809 dev->reg_state = NETREG_UNREGISTERED;
4810 }
4811 /*
4812 * Prevent userspace races by waiting until the network
4813 * device is fully setup before sending notifications.
4814 */
4815 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
4816
4817 out:
4818 return ret;
4819
4820 err_uninit:
4821 if (dev->netdev_ops->ndo_uninit)
4822 dev->netdev_ops->ndo_uninit(dev);
4823 goto out;
4824 }
4825
4826 /**
4827 * init_dummy_netdev - init a dummy network device for NAPI
4828 * @dev: device to init
4829 *
4830 * This takes a network device structure and initialize the minimum
4831 * amount of fields so it can be used to schedule NAPI polls without
4832 * registering a full blown interface. This is to be used by drivers
4833 * that need to tie several hardware interfaces to a single NAPI
4834 * poll scheduler due to HW limitations.
4835 */
4836 int init_dummy_netdev(struct net_device *dev)
4837 {
4838 /* Clear everything. Note we don't initialize spinlocks
4839 * are they aren't supposed to be taken by any of the
4840 * NAPI code and this dummy netdev is supposed to be
4841 * only ever used for NAPI polls
4842 */
4843 memset(dev, 0, sizeof(struct net_device));
4844
4845 /* make sure we BUG if trying to hit standard
4846 * register/unregister code path
4847 */
4848 dev->reg_state = NETREG_DUMMY;
4849
4850 /* initialize the ref count */
4851 atomic_set(&dev->refcnt, 1);
4852
4853 /* NAPI wants this */
4854 INIT_LIST_HEAD(&dev->napi_list);
4855
4856 /* a dummy interface is started by default */
4857 set_bit(__LINK_STATE_PRESENT, &dev->state);
4858 set_bit(__LINK_STATE_START, &dev->state);
4859
4860 return 0;
4861 }
4862 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4863
4864
4865 /**
4866 * register_netdev - register a network device
4867 * @dev: device to register
4868 *
4869 * Take a completed network device structure and add it to the kernel
4870 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4871 * chain. 0 is returned on success. A negative errno code is returned
4872 * on a failure to set up the device, or if the name is a duplicate.
4873 *
4874 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4875 * and expands the device name if you passed a format string to
4876 * alloc_netdev.
4877 */
4878 int register_netdev(struct net_device *dev)
4879 {
4880 int err;
4881
4882 rtnl_lock();
4883
4884 /*
4885 * If the name is a format string the caller wants us to do a
4886 * name allocation.
4887 */
4888 if (strchr(dev->name, '%')) {
4889 err = dev_alloc_name(dev, dev->name);
4890 if (err < 0)
4891 goto out;
4892 }
4893
4894 err = register_netdevice(dev);
4895 out:
4896 rtnl_unlock();
4897 return err;
4898 }
4899 EXPORT_SYMBOL(register_netdev);
4900
4901 /*
4902 * netdev_wait_allrefs - wait until all references are gone.
4903 *
4904 * This is called when unregistering network devices.
4905 *
4906 * Any protocol or device that holds a reference should register
4907 * for netdevice notification, and cleanup and put back the
4908 * reference if they receive an UNREGISTER event.
4909 * We can get stuck here if buggy protocols don't correctly
4910 * call dev_put.
4911 */
4912 static void netdev_wait_allrefs(struct net_device *dev)
4913 {
4914 unsigned long rebroadcast_time, warning_time;
4915
4916 rebroadcast_time = warning_time = jiffies;
4917 while (atomic_read(&dev->refcnt) != 0) {
4918 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4919 rtnl_lock();
4920
4921 /* Rebroadcast unregister notification */
4922 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4923
4924 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4925 &dev->state)) {
4926 /* We must not have linkwatch events
4927 * pending on unregister. If this
4928 * happens, we simply run the queue
4929 * unscheduled, resulting in a noop
4930 * for this device.
4931 */
4932 linkwatch_run_queue();
4933 }
4934
4935 __rtnl_unlock();
4936
4937 rebroadcast_time = jiffies;
4938 }
4939
4940 msleep(250);
4941
4942 if (time_after(jiffies, warning_time + 10 * HZ)) {
4943 printk(KERN_EMERG "unregister_netdevice: "
4944 "waiting for %s to become free. Usage "
4945 "count = %d\n",
4946 dev->name, atomic_read(&dev->refcnt));
4947 warning_time = jiffies;
4948 }
4949 }
4950 }
4951
4952 /* The sequence is:
4953 *
4954 * rtnl_lock();
4955 * ...
4956 * register_netdevice(x1);
4957 * register_netdevice(x2);
4958 * ...
4959 * unregister_netdevice(y1);
4960 * unregister_netdevice(y2);
4961 * ...
4962 * rtnl_unlock();
4963 * free_netdev(y1);
4964 * free_netdev(y2);
4965 *
4966 * We are invoked by rtnl_unlock().
4967 * This allows us to deal with problems:
4968 * 1) We can delete sysfs objects which invoke hotplug
4969 * without deadlocking with linkwatch via keventd.
4970 * 2) Since we run with the RTNL semaphore not held, we can sleep
4971 * safely in order to wait for the netdev refcnt to drop to zero.
4972 *
4973 * We must not return until all unregister events added during
4974 * the interval the lock was held have been completed.
4975 */
4976 void netdev_run_todo(void)
4977 {
4978 struct list_head list;
4979
4980 /* Snapshot list, allow later requests */
4981 list_replace_init(&net_todo_list, &list);
4982
4983 __rtnl_unlock();
4984
4985 while (!list_empty(&list)) {
4986 struct net_device *dev
4987 = list_entry(list.next, struct net_device, todo_list);
4988 list_del(&dev->todo_list);
4989
4990 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4991 printk(KERN_ERR "network todo '%s' but state %d\n",
4992 dev->name, dev->reg_state);
4993 dump_stack();
4994 continue;
4995 }
4996
4997 dev->reg_state = NETREG_UNREGISTERED;
4998
4999 on_each_cpu(flush_backlog, dev, 1);
5000
5001 netdev_wait_allrefs(dev);
5002
5003 /* paranoia */
5004 BUG_ON(atomic_read(&dev->refcnt));
5005 WARN_ON(dev->ip_ptr);
5006 WARN_ON(dev->ip6_ptr);
5007 WARN_ON(dev->dn_ptr);
5008
5009 if (dev->destructor)
5010 dev->destructor(dev);
5011
5012 /* Free network device */
5013 kobject_put(&dev->dev.kobj);
5014 }
5015 }
5016
5017 /**
5018 * dev_get_stats - get network device statistics
5019 * @dev: device to get statistics from
5020 *
5021 * Get network statistics from device. The device driver may provide
5022 * its own method by setting dev->netdev_ops->get_stats; otherwise
5023 * the internal statistics structure is used.
5024 */
5025 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5026 {
5027 const struct net_device_ops *ops = dev->netdev_ops;
5028
5029 if (ops->ndo_get_stats)
5030 return ops->ndo_get_stats(dev);
5031 else {
5032 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5033 struct net_device_stats *stats = &dev->stats;
5034 unsigned int i;
5035 struct netdev_queue *txq;
5036
5037 for (i = 0; i < dev->num_tx_queues; i++) {
5038 txq = netdev_get_tx_queue(dev, i);
5039 tx_bytes += txq->tx_bytes;
5040 tx_packets += txq->tx_packets;
5041 tx_dropped += txq->tx_dropped;
5042 }
5043 if (tx_bytes || tx_packets || tx_dropped) {
5044 stats->tx_bytes = tx_bytes;
5045 stats->tx_packets = tx_packets;
5046 stats->tx_dropped = tx_dropped;
5047 }
5048 return stats;
5049 }
5050 }
5051 EXPORT_SYMBOL(dev_get_stats);
5052
5053 static void netdev_init_one_queue(struct net_device *dev,
5054 struct netdev_queue *queue,
5055 void *_unused)
5056 {
5057 queue->dev = dev;
5058 }
5059
5060 static void netdev_init_queues(struct net_device *dev)
5061 {
5062 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5063 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5064 spin_lock_init(&dev->tx_global_lock);
5065 }
5066
5067 /**
5068 * alloc_netdev_mq - allocate network device
5069 * @sizeof_priv: size of private data to allocate space for
5070 * @name: device name format string
5071 * @setup: callback to initialize device
5072 * @queue_count: the number of subqueues to allocate
5073 *
5074 * Allocates a struct net_device with private data area for driver use
5075 * and performs basic initialization. Also allocates subquue structs
5076 * for each queue on the device at the end of the netdevice.
5077 */
5078 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5079 void (*setup)(struct net_device *), unsigned int queue_count)
5080 {
5081 struct netdev_queue *tx;
5082 struct net_device *dev;
5083 size_t alloc_size;
5084 struct net_device *p;
5085
5086 BUG_ON(strlen(name) >= sizeof(dev->name));
5087
5088 alloc_size = sizeof(struct net_device);
5089 if (sizeof_priv) {
5090 /* ensure 32-byte alignment of private area */
5091 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5092 alloc_size += sizeof_priv;
5093 }
5094 /* ensure 32-byte alignment of whole construct */
5095 alloc_size += NETDEV_ALIGN - 1;
5096
5097 p = kzalloc(alloc_size, GFP_KERNEL);
5098 if (!p) {
5099 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5100 return NULL;
5101 }
5102
5103 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5104 if (!tx) {
5105 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5106 "tx qdiscs.\n");
5107 goto free_p;
5108 }
5109
5110 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5111 dev->padded = (char *)dev - (char *)p;
5112
5113 if (dev_addr_init(dev))
5114 goto free_tx;
5115
5116 dev_unicast_init(dev);
5117
5118 dev_net_set(dev, &init_net);
5119
5120 dev->_tx = tx;
5121 dev->num_tx_queues = queue_count;
5122 dev->real_num_tx_queues = queue_count;
5123
5124 dev->gso_max_size = GSO_MAX_SIZE;
5125
5126 netdev_init_queues(dev);
5127
5128 INIT_LIST_HEAD(&dev->napi_list);
5129 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5130 setup(dev);
5131 strcpy(dev->name, name);
5132 return dev;
5133
5134 free_tx:
5135 kfree(tx);
5136
5137 free_p:
5138 kfree(p);
5139 return NULL;
5140 }
5141 EXPORT_SYMBOL(alloc_netdev_mq);
5142
5143 /**
5144 * free_netdev - free network device
5145 * @dev: device
5146 *
5147 * This function does the last stage of destroying an allocated device
5148 * interface. The reference to the device object is released.
5149 * If this is the last reference then it will be freed.
5150 */
5151 void free_netdev(struct net_device *dev)
5152 {
5153 struct napi_struct *p, *n;
5154
5155 release_net(dev_net(dev));
5156
5157 kfree(dev->_tx);
5158
5159 /* Flush device addresses */
5160 dev_addr_flush(dev);
5161
5162 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5163 netif_napi_del(p);
5164
5165 /* Compatibility with error handling in drivers */
5166 if (dev->reg_state == NETREG_UNINITIALIZED) {
5167 kfree((char *)dev - dev->padded);
5168 return;
5169 }
5170
5171 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5172 dev->reg_state = NETREG_RELEASED;
5173
5174 /* will free via device release */
5175 put_device(&dev->dev);
5176 }
5177
5178 /**
5179 * synchronize_net - Synchronize with packet receive processing
5180 *
5181 * Wait for packets currently being received to be done.
5182 * Does not block later packets from starting.
5183 */
5184 void synchronize_net(void)
5185 {
5186 might_sleep();
5187 synchronize_rcu();
5188 }
5189
5190 /**
5191 * unregister_netdevice - remove device from the kernel
5192 * @dev: device
5193 *
5194 * This function shuts down a device interface and removes it
5195 * from the kernel tables.
5196 *
5197 * Callers must hold the rtnl semaphore. You may want
5198 * unregister_netdev() instead of this.
5199 */
5200
5201 void unregister_netdevice(struct net_device *dev)
5202 {
5203 ASSERT_RTNL();
5204
5205 rollback_registered(dev);
5206 /* Finish processing unregister after unlock */
5207 net_set_todo(dev);
5208 }
5209
5210 /**
5211 * unregister_netdev - remove device from the kernel
5212 * @dev: device
5213 *
5214 * This function shuts down a device interface and removes it
5215 * from the kernel tables.
5216 *
5217 * This is just a wrapper for unregister_netdevice that takes
5218 * the rtnl semaphore. In general you want to use this and not
5219 * unregister_netdevice.
5220 */
5221 void unregister_netdev(struct net_device *dev)
5222 {
5223 rtnl_lock();
5224 unregister_netdevice(dev);
5225 rtnl_unlock();
5226 }
5227
5228 EXPORT_SYMBOL(unregister_netdev);
5229
5230 /**
5231 * dev_change_net_namespace - move device to different nethost namespace
5232 * @dev: device
5233 * @net: network namespace
5234 * @pat: If not NULL name pattern to try if the current device name
5235 * is already taken in the destination network namespace.
5236 *
5237 * This function shuts down a device interface and moves it
5238 * to a new network namespace. On success 0 is returned, on
5239 * a failure a netagive errno code is returned.
5240 *
5241 * Callers must hold the rtnl semaphore.
5242 */
5243
5244 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5245 {
5246 char buf[IFNAMSIZ];
5247 const char *destname;
5248 int err;
5249
5250 ASSERT_RTNL();
5251
5252 /* Don't allow namespace local devices to be moved. */
5253 err = -EINVAL;
5254 if (dev->features & NETIF_F_NETNS_LOCAL)
5255 goto out;
5256
5257 #ifdef CONFIG_SYSFS
5258 /* Don't allow real devices to be moved when sysfs
5259 * is enabled.
5260 */
5261 err = -EINVAL;
5262 if (dev->dev.parent)
5263 goto out;
5264 #endif
5265
5266 /* Ensure the device has been registrered */
5267 err = -EINVAL;
5268 if (dev->reg_state != NETREG_REGISTERED)
5269 goto out;
5270
5271 /* Get out if there is nothing todo */
5272 err = 0;
5273 if (net_eq(dev_net(dev), net))
5274 goto out;
5275
5276 /* Pick the destination device name, and ensure
5277 * we can use it in the destination network namespace.
5278 */
5279 err = -EEXIST;
5280 destname = dev->name;
5281 if (__dev_get_by_name(net, destname)) {
5282 /* We get here if we can't use the current device name */
5283 if (!pat)
5284 goto out;
5285 if (!dev_valid_name(pat))
5286 goto out;
5287 if (strchr(pat, '%')) {
5288 if (__dev_alloc_name(net, pat, buf) < 0)
5289 goto out;
5290 destname = buf;
5291 } else
5292 destname = pat;
5293 if (__dev_get_by_name(net, destname))
5294 goto out;
5295 }
5296
5297 /*
5298 * And now a mini version of register_netdevice unregister_netdevice.
5299 */
5300
5301 /* If device is running close it first. */
5302 dev_close(dev);
5303
5304 /* And unlink it from device chain */
5305 err = -ENODEV;
5306 unlist_netdevice(dev);
5307
5308 synchronize_net();
5309
5310 /* Shutdown queueing discipline. */
5311 dev_shutdown(dev);
5312
5313 /* Notify protocols, that we are about to destroy
5314 this device. They should clean all the things.
5315 */
5316 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5317
5318 /*
5319 * Flush the unicast and multicast chains
5320 */
5321 dev_unicast_flush(dev);
5322 dev_addr_discard(dev);
5323
5324 netdev_unregister_kobject(dev);
5325
5326 /* Actually switch the network namespace */
5327 dev_net_set(dev, net);
5328
5329 /* Assign the new device name */
5330 if (destname != dev->name)
5331 strcpy(dev->name, destname);
5332
5333 /* If there is an ifindex conflict assign a new one */
5334 if (__dev_get_by_index(net, dev->ifindex)) {
5335 int iflink = (dev->iflink == dev->ifindex);
5336 dev->ifindex = dev_new_index(net);
5337 if (iflink)
5338 dev->iflink = dev->ifindex;
5339 }
5340
5341 /* Fixup kobjects */
5342 err = netdev_register_kobject(dev);
5343 WARN_ON(err);
5344
5345 /* Add the device back in the hashes */
5346 list_netdevice(dev);
5347
5348 /* Notify protocols, that a new device appeared. */
5349 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5350
5351 /*
5352 * Prevent userspace races by waiting until the network
5353 * device is fully setup before sending notifications.
5354 */
5355 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5356
5357 synchronize_net();
5358 err = 0;
5359 out:
5360 return err;
5361 }
5362
5363 static int dev_cpu_callback(struct notifier_block *nfb,
5364 unsigned long action,
5365 void *ocpu)
5366 {
5367 struct sk_buff **list_skb;
5368 struct Qdisc **list_net;
5369 struct sk_buff *skb;
5370 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5371 struct softnet_data *sd, *oldsd;
5372
5373 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5374 return NOTIFY_OK;
5375
5376 local_irq_disable();
5377 cpu = smp_processor_id();
5378 sd = &per_cpu(softnet_data, cpu);
5379 oldsd = &per_cpu(softnet_data, oldcpu);
5380
5381 /* Find end of our completion_queue. */
5382 list_skb = &sd->completion_queue;
5383 while (*list_skb)
5384 list_skb = &(*list_skb)->next;
5385 /* Append completion queue from offline CPU. */
5386 *list_skb = oldsd->completion_queue;
5387 oldsd->completion_queue = NULL;
5388
5389 /* Find end of our output_queue. */
5390 list_net = &sd->output_queue;
5391 while (*list_net)
5392 list_net = &(*list_net)->next_sched;
5393 /* Append output queue from offline CPU. */
5394 *list_net = oldsd->output_queue;
5395 oldsd->output_queue = NULL;
5396
5397 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5398 local_irq_enable();
5399
5400 /* Process offline CPU's input_pkt_queue */
5401 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5402 netif_rx(skb);
5403
5404 return NOTIFY_OK;
5405 }
5406
5407
5408 /**
5409 * netdev_increment_features - increment feature set by one
5410 * @all: current feature set
5411 * @one: new feature set
5412 * @mask: mask feature set
5413 *
5414 * Computes a new feature set after adding a device with feature set
5415 * @one to the master device with current feature set @all. Will not
5416 * enable anything that is off in @mask. Returns the new feature set.
5417 */
5418 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5419 unsigned long mask)
5420 {
5421 /* If device needs checksumming, downgrade to it. */
5422 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5423 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5424 else if (mask & NETIF_F_ALL_CSUM) {
5425 /* If one device supports v4/v6 checksumming, set for all. */
5426 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5427 !(all & NETIF_F_GEN_CSUM)) {
5428 all &= ~NETIF_F_ALL_CSUM;
5429 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5430 }
5431
5432 /* If one device supports hw checksumming, set for all. */
5433 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5434 all &= ~NETIF_F_ALL_CSUM;
5435 all |= NETIF_F_HW_CSUM;
5436 }
5437 }
5438
5439 one |= NETIF_F_ALL_CSUM;
5440
5441 one |= all & NETIF_F_ONE_FOR_ALL;
5442 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5443 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5444
5445 return all;
5446 }
5447 EXPORT_SYMBOL(netdev_increment_features);
5448
5449 static struct hlist_head *netdev_create_hash(void)
5450 {
5451 int i;
5452 struct hlist_head *hash;
5453
5454 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5455 if (hash != NULL)
5456 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5457 INIT_HLIST_HEAD(&hash[i]);
5458
5459 return hash;
5460 }
5461
5462 /* Initialize per network namespace state */
5463 static int __net_init netdev_init(struct net *net)
5464 {
5465 INIT_LIST_HEAD(&net->dev_base_head);
5466
5467 net->dev_name_head = netdev_create_hash();
5468 if (net->dev_name_head == NULL)
5469 goto err_name;
5470
5471 net->dev_index_head = netdev_create_hash();
5472 if (net->dev_index_head == NULL)
5473 goto err_idx;
5474
5475 return 0;
5476
5477 err_idx:
5478 kfree(net->dev_name_head);
5479 err_name:
5480 return -ENOMEM;
5481 }
5482
5483 /**
5484 * netdev_drivername - network driver for the device
5485 * @dev: network device
5486 * @buffer: buffer for resulting name
5487 * @len: size of buffer
5488 *
5489 * Determine network driver for device.
5490 */
5491 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5492 {
5493 const struct device_driver *driver;
5494 const struct device *parent;
5495
5496 if (len <= 0 || !buffer)
5497 return buffer;
5498 buffer[0] = 0;
5499
5500 parent = dev->dev.parent;
5501
5502 if (!parent)
5503 return buffer;
5504
5505 driver = parent->driver;
5506 if (driver && driver->name)
5507 strlcpy(buffer, driver->name, len);
5508 return buffer;
5509 }
5510
5511 static void __net_exit netdev_exit(struct net *net)
5512 {
5513 kfree(net->dev_name_head);
5514 kfree(net->dev_index_head);
5515 }
5516
5517 static struct pernet_operations __net_initdata netdev_net_ops = {
5518 .init = netdev_init,
5519 .exit = netdev_exit,
5520 };
5521
5522 static void __net_exit default_device_exit(struct net *net)
5523 {
5524 struct net_device *dev;
5525 /*
5526 * Push all migratable of the network devices back to the
5527 * initial network namespace
5528 */
5529 rtnl_lock();
5530 restart:
5531 for_each_netdev(net, dev) {
5532 int err;
5533 char fb_name[IFNAMSIZ];
5534
5535 /* Ignore unmoveable devices (i.e. loopback) */
5536 if (dev->features & NETIF_F_NETNS_LOCAL)
5537 continue;
5538
5539 /* Delete virtual devices */
5540 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5541 dev->rtnl_link_ops->dellink(dev);
5542 goto restart;
5543 }
5544
5545 /* Push remaing network devices to init_net */
5546 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5547 err = dev_change_net_namespace(dev, &init_net, fb_name);
5548 if (err) {
5549 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5550 __func__, dev->name, err);
5551 BUG();
5552 }
5553 goto restart;
5554 }
5555 rtnl_unlock();
5556 }
5557
5558 static struct pernet_operations __net_initdata default_device_ops = {
5559 .exit = default_device_exit,
5560 };
5561
5562 /*
5563 * Initialize the DEV module. At boot time this walks the device list and
5564 * unhooks any devices that fail to initialise (normally hardware not
5565 * present) and leaves us with a valid list of present and active devices.
5566 *
5567 */
5568
5569 /*
5570 * This is called single threaded during boot, so no need
5571 * to take the rtnl semaphore.
5572 */
5573 static int __init net_dev_init(void)
5574 {
5575 int i, rc = -ENOMEM;
5576
5577 BUG_ON(!dev_boot_phase);
5578
5579 if (dev_proc_init())
5580 goto out;
5581
5582 if (netdev_kobject_init())
5583 goto out;
5584
5585 INIT_LIST_HEAD(&ptype_all);
5586 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5587 INIT_LIST_HEAD(&ptype_base[i]);
5588
5589 if (register_pernet_subsys(&netdev_net_ops))
5590 goto out;
5591
5592 /*
5593 * Initialise the packet receive queues.
5594 */
5595
5596 for_each_possible_cpu(i) {
5597 struct softnet_data *queue;
5598
5599 queue = &per_cpu(softnet_data, i);
5600 skb_queue_head_init(&queue->input_pkt_queue);
5601 queue->completion_queue = NULL;
5602 INIT_LIST_HEAD(&queue->poll_list);
5603
5604 queue->backlog.poll = process_backlog;
5605 queue->backlog.weight = weight_p;
5606 queue->backlog.gro_list = NULL;
5607 queue->backlog.gro_count = 0;
5608 }
5609
5610 dev_boot_phase = 0;
5611
5612 /* The loopback device is special if any other network devices
5613 * is present in a network namespace the loopback device must
5614 * be present. Since we now dynamically allocate and free the
5615 * loopback device ensure this invariant is maintained by
5616 * keeping the loopback device as the first device on the
5617 * list of network devices. Ensuring the loopback devices
5618 * is the first device that appears and the last network device
5619 * that disappears.
5620 */
5621 if (register_pernet_device(&loopback_net_ops))
5622 goto out;
5623
5624 if (register_pernet_device(&default_device_ops))
5625 goto out;
5626
5627 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5628 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5629
5630 hotcpu_notifier(dev_cpu_callback, 0);
5631 dst_init();
5632 dev_mcast_init();
5633 rc = 0;
5634 out:
5635 return rc;
5636 }
5637
5638 subsys_initcall(net_dev_init);
5639
5640 static int __init initialize_hashrnd(void)
5641 {
5642 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5643 return 0;
5644 }
5645
5646 late_initcall_sync(initialize_hashrnd);
5647
5648 EXPORT_SYMBOL(__dev_get_by_index);
5649 EXPORT_SYMBOL(__dev_get_by_name);
5650 EXPORT_SYMBOL(__dev_remove_pack);
5651 EXPORT_SYMBOL(dev_valid_name);
5652 EXPORT_SYMBOL(dev_add_pack);
5653 EXPORT_SYMBOL(dev_alloc_name);
5654 EXPORT_SYMBOL(dev_close);
5655 EXPORT_SYMBOL(dev_get_by_flags);
5656 EXPORT_SYMBOL(dev_get_by_index);
5657 EXPORT_SYMBOL(dev_get_by_name);
5658 EXPORT_SYMBOL(dev_open);
5659 EXPORT_SYMBOL(dev_queue_xmit);
5660 EXPORT_SYMBOL(dev_remove_pack);
5661 EXPORT_SYMBOL(dev_set_allmulti);
5662 EXPORT_SYMBOL(dev_set_promiscuity);
5663 EXPORT_SYMBOL(dev_change_flags);
5664 EXPORT_SYMBOL(dev_set_mtu);
5665 EXPORT_SYMBOL(dev_set_mac_address);
5666 EXPORT_SYMBOL(free_netdev);
5667 EXPORT_SYMBOL(netdev_boot_setup_check);
5668 EXPORT_SYMBOL(netdev_set_master);
5669 EXPORT_SYMBOL(netdev_state_change);
5670 EXPORT_SYMBOL(netif_receive_skb);
5671 EXPORT_SYMBOL(netif_rx);
5672 EXPORT_SYMBOL(register_gifconf);
5673 EXPORT_SYMBOL(register_netdevice);
5674 EXPORT_SYMBOL(register_netdevice_notifier);
5675 EXPORT_SYMBOL(skb_checksum_help);
5676 EXPORT_SYMBOL(synchronize_net);
5677 EXPORT_SYMBOL(unregister_netdevice);
5678 EXPORT_SYMBOL(unregister_netdevice_notifier);
5679 EXPORT_SYMBOL(net_enable_timestamp);
5680 EXPORT_SYMBOL(net_disable_timestamp);
5681 EXPORT_SYMBOL(dev_get_flags);
5682
5683 EXPORT_SYMBOL(dev_load);
5684
5685 EXPORT_PER_CPU_SYMBOL(softnet_data);
5686
|
This page was automatically generated by the
LXR engine.
|